Unnamed: 0
int64 0
354
| Company
stringclasses 13
values | Fichier
stringlengths 7
123
| Content
stringlengths 1.77k
32.8k
|
|---|---|---|---|
109 | bcg | BCG-Executive-Perspectives-GenAI-in-Data-and-Digital-Platforms-EP6-14Nov2024.pdf | Executive
Perspectives
The Future of the AI-Driven
Tech and People Stack
Data and Digital Platforms
November 2024
In this BCG
Introduction
Executive Perspective,
We meet often with CEOs to discuss AI---a topic that is both captivating and rapidly
we articulate the vision
changing. After working with over 1,000 clients in the past year, we are sharing our
and value of the future
most recent learning in a new series designed to help CEOs navigate AI. With
AI at an inflection point, the focus in 2024 is on turning AI’s potential into of the AI-driven tech
real profit.
and people stack
In this edition, we discuss the future of data and digital platforms as well as the role
AI will play in fundamentally transforming the tech function and wider
organization. We address key questions on the minds of leaders:
• How do I set up my organization to maximize value from data and enable AI
capabilities at scale?
.d
e
• How do I evolve my tech stack and AI platforms to support building AI v re
se
applications?
r
sth
g
• How can I build an organization that is fast and composable?
ir
llA
.p
u
o
rG
• How do I get started…and how do I get this right? g
n
itlu
s
n
o
C
n
o
This document is a guide for CEOs and technology leaders to cut through ts o
B
y
b
the hype around the future of the AI-driven tech and people stack and 4
2
0
2
understand what creates value now and in the future. ©
th
g
iry
p
1 o
C
Executive summary | Data and digital platforms—new technology,
ways of working, and skills required to succeed in the AI Age
AI spending will continue to increase rapidly, with a CAGR of ~30% in the next 3 years. To unlock
Evolving the
value, organizations are building new AI platforms, improving data capabilities, and
stack is key to
rewiring processes (e.g., 20% marketing savings via GenAI automation, 25% lower drug discovery
unlock AI value…
time via GenAI)
For effective and responsible AI development and scaling, a so-called “horizontal stack” with a
dedicated AI layer is essential. This typically requires updates to existing layers to integrate with
... through a
LLM models
horizontal stack
and organization A platform-based operating model, while not specific to AI, is instrumental in maximizing value
.d
e v re
se
from a horizontal stack. It fosters improved cross-functional collaboration and integration
r
sth
g
ir
llA
.p
u
o
rG
g
Rewiring your Success hinges on robust tech capabilities, including establishing model selection frameworks, n itlu
s
n
architecture and centralizing platforms, implementing data orchestration, and setting up evaluation methods o
C
n
o
IT operational ts
o
B
y
setup is critical Complementing these technical capabilities are key operating model drivers, revolving b
4
2
0
2
for success around AI roles and governance, planning AI talent, and championing AI leadership and culture ©
th
g
iry
p
2 o
C
Are our data and digital platforms set up to enable
AI capabilities and capture value from AI?
Do we have a clear strategy around which models,
platforms, data sets, and evaluation criteria we
need to set up?
Key questions you
2 should be asking Are our op model and current setup lined up to
efficiently deliver on our evolving tech stack/needs?
your CTO
.d
e
v
re
Are our IT leadership, talent, and skill set ready se
r
sth
g
for an AI transformation?
ir
llA
.p
u
o
rG
g
n
itlu
What are the expected tech and people costs s n
o
C
n
o
needed to enable AI value capture? ts
o
B
y
b
4
2
0
2
©
th
g
iry
p
3 o
C
Are our data and digital platforms set up to enable
AI capabilities and capture value from AI?
Do we have a clear strategy around which models,
platforms, data sets, and evaluation criteria we
need to set up?
Key questions you
2 should be asking Are our op model and current setup lined up to
efficiently deliver on our evolving tech stack/needs?
your CTO
.d
e
v
re
Are our IT leadership, talent, and skill set ready se
r
sth
g
for an AI transformation?
ir
llA
.p
u
o
rG
g
n
itlu
What are the expected tech and people costs s n
o
C
n
o
needed to enable AI value capture? ts
o
B
y
b
4
2
0
2
©
th
g
iry
p
4 o
C
Companies that are ahead in their tech modernization
journey find it easier to deploy AI applications and capabilities
Seamless
Ability to support AI High
Medium
applications and
Low
capabilities
PAST TODAY EMERGING FUTURE
How
Waterfall Agile DevOps and platform op model NoOps and platform op model
Microservices, API, Serverless services
N-tiers and data- and AI/data-enabled and AI/data-enabled
Monolithic apps
enabled apps apps apps
What .d
e
v
re
se
Siloed architecture
H aro cr hiz ito en ct ta ul rl ey layered C Aro cm hip teo cs ta ub rl ee C aro cm hip teo cs ta ub rl ee r sth
g
ir
llA
.p
u
o
rG
g
n
itlu
s
n
o
C
Where n
Containers Serverless computing o ts
o
B
y
b
4
2
0
2
©
Physical Private virtualization Hybrid cloud Public cloud th
g
iry
p
5 o
C
Looking closer at the stack, an incremental AI layer, coupled with
improvements to existing layers, will lay the foundation for enabling AI applications
Today's data and digital platforms GenAI reshape: dedicated AI layer
Smart Business Layer Smart Business Layer
Conversational apps AI apps
Chat Search Copilot Expert systems
Omnichannel
App builders Language Library Low/No code
Business services
AI services Omnichannel
Speech Text Image Biz. services
Data Layer AI Layer
Guardrails
Data products
Orchestration
Ops and
Repository and storage Operational data services E2E app vendor Model garden .d
monitoring e
v
Ingestion and distribution
F Mou on dd ea
l
t pio lan ta fol rm models re se
r
sth
g
ir
llA
Core Transaction Layer Data Layer .p u
o
Data products
rG
g
n
ERP Other systems Repository and storage Operational data services itlu
s
Distribution and integration n o
C
n
o
ts
Infrastructure and Cloud Core Transaction Layer o B
y
b
4
2
Infrastructure and Cloud 0
2
On-prem Cloud Hybrid ©
On-prem Cloud Hybrid TPU/GPU th
g
iry
p
New layer Existing layer capability upgrade 6 o C
By modernizing their stack, leading companies across industries are driving
innovation and growth and reaping material benefits from GenAI
Note: Select examples only; for more specifics about value capture, please refer to BCG’s Executive Perspectives on respective topics
GenAI led task automation, +20%
GenAI assisted tender
+50%
Multinational Luxury automobile
content localization, and
document creation and
savings in health CPG manufacturer
savings on research time
efficiency gain offer analysis
marketing costs
GenAI sales assistant +25%
2X Multinational Leverage LLM to prioritize
deployment vs. traditional Large telecom
cosmetics call deflection
touchpoints (search, retail increase in ROI manufacturer company BU cost reduction opportunities
media)
.d
e
v
re
se
+80%
Alcoholic $1Bn
Increase in customer r sth
g
GenAI assisted marketing
beverages
Large retail value perception ir
llA
content development external agency company chain through GenAI-based .p u o
additional sales rG
cost savings pricing and markdowns g
n
itlu
s
n
o
C
n
Cycle time reduction in o
25% ts
+35% Consumer o B
GenAI leveraged to draft Pharmaceutical drug discovery through y
biopharma b 4
clinical study reports
time efficiency gain
company company drug discovery time GenAI-based lead 2 0 2
©
reduction optimization th
g
iry
p
7 o
Source: BCG case experience C
Are our data and digital platforms set up to enable
AI capabilities and capture value from AI?
Do we have a clear strategy around which models,
platforms, data sets, and evaluation criteria we
need to set up?
Key questions you
2 should be asking Are our op model and current setup lined up to
efficiently deliver on our evolving tech stack/needs?
your CTO
.d
e
v
re
Are our IT leadership, talent, and skill set ready se
r
sth
g
for an AI transformation?
ir
llA
.p
u
o
rG
g
n
itlu
What are the expected tech and people costs s n
o
C
n
o
needed to enable AI value capture? ts
o
B
y
b
4
2
0
2
©
th
g
iry
p
8 o
C
Modernizing four key areas of the current stack can appropriately
enable and maximize value realization from AI
Smart Business Layer (systems of engagement)
Deep dive to follow
…
MODELS AI copilots Conversational apps AI assistants
Diversify beyond a single model, including
1
competitive open-source options as model AI Layer 4
adaptation becomes key for competitive advantage
Guardrails
PLATFORMS 1 Orchestration Ops and
E2E app monitoring
2 Adopt AI platforms to centralize model hosting, Model garden Foundation/other small models
vendors
streamline builds, and enhance scalability,
n
effectively bridging talent gaps 2 Model platform o
y i t
t a
i r u r g .d
DATA
Data Layer c e S e t n
I
e v re se
r sth
3 Leverage orchestration and model adaptation,
3 Data products
Operational
g ir
llA
bolstered by robust data capabilities and access to .p
Repository and storage data u o
diverse data sets, to maintain competitive advantage rG
g
services n
Distribution and integration
itlu
s
n
o
C
Deep dive to follow n o
EVALUATION ts
Core Transaction Layer o
B
Build capabilities to measure success in response y b
4 4
2
quality, technical performance, responsible AI, 0
Infra and Cloud Layer Public cloud Private cloud Specialized hardware (GPU & TPU) 2 ©
security, and cost th
g
iry
p
9 o
C
As large language models continue to proliferate, combining the right model
selection criteria with the required platform capabilities is critical to scaling AI efforts
BCG’s selection criteria framework helps companies …and support those models with the right
find the right models at the right cost… AI-platform built-in services
Model types
Proprietary Open-source Third-party In-house
1 Model output 6 Data sensitivity
models models models models
Compatibility with
2 Model size 7 Platform Capabilities
platform/model provider
Model Management
3 Model capabilities 8 Economics MLOps tools for model deployment, monitoring, and management
Integration with Data and Knowledge
..dd
ee
vv
4 Performance 9 RAI1 and regulatory Integration with data (e.g., RAG), applications, cloud services, and infra rree
ssee
Model Governance
rr
sstthh
gg
5 Flexibility/Fine-tuning 10 Optimization consideration Product-specific documentation and audit trails for regulatory compliance
iirr
llllAA
..pp
uu
oo
rrGG
gg
nn
iittlluu
ss
nn
Partnerships can oo
Infrastructure Infrastructure Platforms Applications CC
nn
oo
accelerate outcomes by
ttss
oo
BB
yy
bb
simplifying development, ` 44 22 00
22
©©
testing, and deployment tthh
gg
iirryy
pp
10 oo
1. Responsible AI CC
Expanding five critical capabilities enables organizations
to measure ongoing success
Area Evaluation criteria Illustrative
Content Relevance Specificity Truthfulness …
Response
quality
Verbiage Emotional connection Language style …
The extent to which responses from AI applications meet expectations
Technical Resource
Latency Throughput Availability …
performance consumption
The technical KPIs that AI solutions must achieve to fulfill business need
Responsible .d
Fairness Interpretability Compliance Accountability … e v
re
AI se
r
sth
g
The level of trust and ethical integration of AI solutions ir
llA
.p
u
Business Customer o rG
ROI Inference cost Cost reduction … g
impact satisfaction n itlu
s
n
o
C
The amount of business value enabled by AI use cases – function of above criteria as well as adoption, operating model, and change management
n
o
ts
o
B
Security and Model denial of Prompt/Response y b
Data privacy Access control … 4 2
privacy service (DDoS) monitoring 0 2
©
th
The extent to which the AI system is secure from threats and vulnerabilities New or enhanced evaluation dimensions g iry
p
11 o
C
Are our data and digital platforms set up to enable
AI capabilities and capture value from AI?
Do we have a clear strategy around which models,
platforms, data sets, and evaluation criteria we
need to set up?
Key questions you
2 should be asking Are our op model and current setup lined up to
efficiently deliver on our evolving tech stack/needs?
your CTO
.d
e
v
re
Are our IT leadership, talent, and skill set ready se
r
sth
g
for an AI transformation?
ir
llA
.p
u
o
rG
g
n
itlu
What are the expected tech and people costs s n
o
C
n
o
needed to enable AI value capture? ts
o
B
y
b
4
2
0
2
©
th
g
iry
p
12 o
C
A platform-led organization that is fast and composable should implement a
reshaped tech stack and benefit from the speed and agility it enables
Platforms deliver shared products/services …maximizing value from
that multiple BUs combine and consume… the evolved AI stack
Deploy AI solutions once with
1
enterprise-wide impact
Eliminate duplicated capabilities
Avoidbespoke AI solutions
+30% Cost reduction by removingduplication
Traditional organization Matrix organization Platform operating model
• Business units and functions • Business units and functions • Business units driving missions 2 Create reusable AI platforms for .d
e
• Silos and bespoke processes • Some shared capabilities • Accelerated by shared platforms
whole enterprise v re
se
• Hierarchical prioritization • Hierarchical prioritization • Dynamically aligned priorities
Set up acentralized AIcapability
r
sth
g
ProvidestandardizedAI services
ir
llA
.p
u
o
+25% Productivity inc. by combining capabilities rG
Common platforms with composable, reusable services to cut costs and boost efficiency
g
n
itlu
s
n
o
3 Unlock efficiency and agility C
n
Teams focused on products and shared, reusable services based on self-service principles o ts
o
Bring data, people, and tech together B
y
b
4
2
Enable quicker scaling of AI solutions 0 2
©
Teams E2E responsible and enabled by cross-functional ways of working
+50% Fastertime tomarket
th
g
iry
p
13 o
Source: BCG experience C
Top companies lean into the 10-20-70 principle to drive toward a platform-led
organization; they see AI as mainly a people transformation vs. tech-only
Non-exhaustive
Deep dive
People, organization, and processes Structure and roles to follow
Effective processes supported by talent
and change management practices
• Reskilled teams to tackle evolving responsibilities/roles
• AI capabilities integrated in cross-functional teams
Governance, business outcomes,
70% 70%
and ways of working
• AI portfolio governance fully integrated in POM portfolio
of the effort • Iterative development and deployment of new solutions
Focus of
and services and business outcome-based steering
of digital/AI transformation
.d
transformations Sourcing of e v
re
talent and tech se
is on people
10%
r sth
g
and processes
20%
• AI capabilities/workforce planning centrally
ir
llA
orchestrated and built up within platforms .p
u
o
• Increased AI collaboration (insourcing vs. outsourcing) rG
g
n
itlu
s
Algorithms Culture and behavior n o
C
n
Data science capabilities o
Technology ts
o
to develop and B
Scalable and modernized • Leaders engaging on AI ambition and championing RAI y b
implement algorithms 4
stack to support business • Change management strategies to help the workforce 2 0 2
navigate and adapt to changes brought by AI ©
needs th
g
iry
p
14 o
C
To adapt to AI, organizations need to adjust responsibilities
and activities performed by employees…
Non-exhaustive
AI is driving demand for … and also fundamentally evolving the
new digital skills … day-to-day activities across roles
• AI algorithms and methods
• Foundational models training
• Model fine-tuning
• Prompt engineering Integration Reduced Increased Enhanced Expedited
with intelligent communication cross-functional evaluative decision
• …
machines tasks teaming thinking making
Number of job postings requiring
Employees will Automation will With the Employees will be Automation will
GenAI-related skills (US)
.d
e
v
increasingly need help cross- increasing required bring more re
se
# Unique job postings to integrate their functional teams complexity of to critically visibility into
r
sth
g
10,000 90X
9,624
workflow with the communicate work, teams evaluate the timelines and
ir
llA
.p
u
capabilities of seamlessly will need to output of AI tools bottlenecks to o
6,055 rG
g
intelligent across multiple become more and identify enable faster n
5,000
itlu
s
machines and channels by cross-functional to potential biases decision n o
1,804 C
n
seamlessly breaking down achieve business or errors making and o ts
36 95 107 o B
0 interact together information silos objectives accountability y b
4
2021 2022 2022 2023 2023 2024 2
0
2
(H2) (H1) (H2) (H1) (H2) (H1) ©
th
g
iry
p
Source: Job postings data from Lightcast; BCG analysis 15 o
C
… while introducing new IT roles, reskilling product roles,
and developing new skills
New tech and digital … to enable teams that build … supported by other roles
roles needed … AI products and skills… that will evolve too
Required roles New requirements AI team roles Responsibilities Existing roles2 New requirements
Chief AI officer Oversee and implement AI Product manager Manage everyday Engineer Have prompt
initiatives team activities and better engineeringskills
understand AI capabilities
AI ethics and Develop/implement UI/UX designer Design AI interfaces
compliance officer ethics and compliance Business analyst Help capture the
policies businesslogic
AI/ML governance Establish governance Data analyst Work with high volumes
Data engineer Prepare data in a .d
specialist framework for responsible ofdata e
usable form v re
use of data and AI se
Cybersecurity roles Address new risks
r
sth
LLM1 ops engineer S prim ompl pif ty
p
a rn od
ce
a su
s
t ao nm da mte
o del
Data scientist D me os dig en
ls
AI analytical
to consider
g ir
llA
.p
u
fine-tuning o rG
Software developer Develop scalable code for g n
Prompt engineer Improve prompt results to AI applications
itlu
s
n
create consistent outputs o C
n
across use cases o
ts
o
B
y
LLM engineer Design LLM b
4
2
model pipelines 0 2
©
th
g
iry
p
1.Large language model 2. Non-exhaustive 16 o
C
Are our data and digital platforms set up to enable
AI capabilities and capture value from AI?
Do we have a clear strategy around which models,
platforms, data sets, and evaluation criteria we
need to set up?
Key questions you
2 should be asking Are our op model and current setup lined up to
efficiently deliver on our evolving tech stack/needs?
your CTO
.d
e
v
re
Are our IT leadership, talent, and skill set ready se
r
sth
g
for an AI transformation?
ir
llA
.p
u
o
rG
g
n
itlu
What are the expected tech and people costs s n
o
C
n
o
needed to enable AI value capture? ts
o
B
y
b
4
2
0
2
©
th
g
iry
p
17 o
C
With AI spending expected to rise with a CAGR of ~30% (and ~85% for GenAI),
companies are prioritizing different levers to control AI-related IT costs
AI is predicted to drive technology costs for Different levers are being adopted by
businesses with a CAGR of ~30% until 2027 companies to control AI-related costs
Forecasted tech cost of business demand for AI and GenAI1 Q: What cost reduction measure(s) are you planning to prioritize in the
CAGR coming year to control AI/GenAI-related IT costs?2 [Multiple choice question]
'22-27
391
Non-exhaustive
Values normalized 9% GenAI infrastructure 71%
to 100 in 2022
11% GenAI platform and app software 100% Value-based prioritization of GenAI use cases 38%
307
9% 9% GenAI IT and business services 94%
In-house resource/shadow spending optimization 30%
10% Overall GenAI ~85%
232
8% .d
9% Consolidation of vendors/systems 28% e v re
7%
se
177 6%
r
sth
Simplification of current tech stack 28% g
135
8 5%
%
71% Other AI (excluding GenAI) 24%
ir
llA
.p
u
6% IT project portfolio optimization 26% o rG
100 73% g
n
78%
itlu
s
83% Cloud spending optimization/migration back to on-prem 15% n o
C
89%
n
95% o
ts
o
Optimization of cybersecurity costs 12% B
y
b
4
2
0
2022 2023 2024 2025 2026 2027 Overall CAGR ~30% 2
©
th
g
1. IDC AI Implementation Market Outlook: Worldwide Core IT Spending for AI Forecast, December 2023. 2. Build for the Future Research 2024 (n=1,000 respondents) iry
p
Note: Core IT spending includes infrastructure hardware, software, public cloud services, and IT/business services (devices and telecommunications services excluded) 18 o C
Source: BCG case experience
To maximize value, it’s key to adopt an approach that reimagines functional
processes, builds AI platforms, and establishes guardrails to control cost and risk
Transformation approach delivers 3-4X ROI Key ingredients
Illustrative
Proof of value
30-45% ROI
E2E reimaging of a functional
End-to-end AI functional process—not isolated use cases –
A
transformation with concrete evaluations proving
impact
Difference in
Proof of scale
ROI and impact
Ability to deliver AI platforms, backed .d
e
v
by scaled data products, serving
re
se
B Discrete use case many functions, and with platform r sth
implementation
economics
g ir
llA
8-15% .p
u
ROI o rG
Proof of control g n
itlu
s
n
Meeting the business where it is; o C
n
o
controlling costs (people + tech), ts
o
B
y
Months after ethical and bias risks, cyber risk, b
4
2
project initiation and data risk (scale without control 0 2
©
is chaos)
th
g
iry
p
Source: Build for the Future Research 2022/23 19 o
C
How to get
.d
e
v
re
se
r
sth
g
started
ir
llA
.p
u
o
rG
g
n
itlu
s
n
o
C
n
o
ts
o
B
y
b
4
2
0
2
©
th
g
iry
p
20 o
C
Getting started | Practical next steps for CTOs
to get ready for their AI transformation
Set aspiration ❑ Develop an AI aspiration aligned with strategic business priorities
and objectives ❑ Establish key objectives and measurable success criteria for tracking AI initiatives
Understand capabilities ❑ Identify key gaps and the largest opportunity areas to strengthen AI tech stack
of your current stack ❑ Benchmark existing tech stack capabilities and maturity versus peers
❑ Develop robust data layer to ensure data quality, reliability, and accessibility
Invest in foundational
❑ Implement scalable infrastructure to support advanced AI models
capabilities
.d
e
❑ Identify and build key technical skills required (e.g., AI programming) v re
se
r
sth
Prioritize building
❑ Build key GenAI layers including models, platforms, guardrails, and orchestration g ir
llA
.p
❑ Establish capabilities to evaluate and measure success across performance, cost, and RAI u
o
capabilities for AI layer rG
g
n
❑ Identify and drive changes in operating model and ways of working to maximize value itlu
s
n
o
C
n
❑ Further widen the platform operating model to maximize value from AI tech stack o ts
o
B
Scale broadly via a y
b
❑ Implement capabilities through AI-focused structures (e.g., CoEs for RAI) 4
2
platform-based org 0 2
©
❑ Drive continuous development and change management for new AI solutions/services th
g
iry
p
21 o
C
NAMR
BCG experts |
Vladimir David Julie
Lukic Martin Bedard
Key contacts
for data and digital
Matthew Beth Djon
platform Kropp Viner Kleine
transformation
EMESA
Nicolas Marc Marcus Julien
de Bellefonds Schuuring Wittig Marx
Remco Dan Adrien Tom
.d
Mol Sack Duthoit Martin e v
re
se
r
sth
g
APAC
ir
llA
.p
u
o
rG
g
n
Jeff Julian Aparna
itlu
s
n
o
Walters King Kapoor C
n
o
ts
o
B
y
b
4
2
0
2
©
Romain de Akira Nipun th
g
Laubier Abe Kalra
iry
p
22 o
C |
110 | bcg | BCG-Executive-Perspectives-Unlocking-Impact-from-AI-Customer-Service-Ops-EP3-28August2024.pdf | Executive
Perspectives
Unlocking Impact from AI
Customer Service Operations
August 2024
1
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
In this BCG
Introduction
Executive Perspective,
We meet often with CEOs to discuss AI—a topic that is both captivating and rapidly we articulate the vision
changing. After working with over 1,000 clients in the past year, we are sharing our
and value of the future
most recent learnings in a new series designed to help CEOs navigate AI.
of customer service
With AI at an inflection point, the focus in 2024 is on turning AI’s potential into real
profit.
with AI
In this edition, we discuss the future of customer service and the role AI will play in
turbocharging growth. We address key questions on the minds of service leaders:
• How will the economics of customer service change with AI?
• How will the customer experience evolve?
• What will my future customer service team need to look like?
• How do I get started…and where should I focus?
This document is a guide for CEOs and customer service leaders to cut
through the hype around AI in service operations and understand what
unlocks value now and in the future.
1
2
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
Summary | Unlocking impact from AI in customer service operations
As customers spend 14B hours/year contacting customer service, leading players set long-term ambitions to
Executives improve service productivity by up to 60% while enhancing customer experience with positive impacts on
retention and additional sales
must act on
AI now While most companies focus on support response, transforming upstream is imperative to maximize value
– including deflection, self-healing, and prevention to create a competitive edge
Unlocking the full potential of AI in customer service requires an end-to-end reshaping of the entire operation
from prevention and self-service to service delivery, with focus dimensions being:
AI impacts all
elements of • Team skilling and structures: Increase productivity, which will lead to fewer but multi-skilled frontline teams with
redesigned agent journeys, focusing on data generation for upstream prevention
customer service
operations • AI Ops capabilities: Establish new roles and skills that build, shape, and govern AI
• Tech ecosystem: Build modular components on layered stack
Develop a value-focused AI strategy with clear and visionary roadmap to realize impact that balances
short-term benefits and unlocking of long-term value
Execute
successfully
Broad, “at-scale” enablement of service agents and overall change management (including customer-
facing communication) is critical to drive adoption – starting with leader enablement
3
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
Leading players set ambitions to improve customer service
productivity by up to 60%
The baseline The future Value enabled by
On average, Leading companies
from AI
~10%
customers spend set ambition of up to
models
14B 60%
from tech/IT
~20%
solutions
Hours per year contacting Productivity improvement
customer service for customer service
from people
~70%
and process
transformation
Source: BCG research 3
4
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
Deep dive: ambition | Examples of leading players
with high ambitions across industries
Select case examples
Global Global Global
tech company high street bank financial company
~60% reduction in resolution time ~30% productivity improvement ~50% cost savings over the
~50% in volume reduction by 2030 over the next three years next five years
Global tech company with Global bank operating in Private label and co-brand
contact centers serving ~20 countries, with ~15k credit card issuer with over
120 countries FTE in customer service 100 brands signed in the US
Diagnosed current state,
Embedded AI agent assist Designed target model
designed a 5-year AI vision
tool into ways of working for for omni-channel customer
and implementation
support agents service, with AI at the heart
roadmap
New target operating First wave of sprints Defined service vision and
model for AI-led support and (tech, people, process strategy with focus on the
interim states across three changes) to realize early target customer and agent AI
time horizons value and fund the journey experience
5
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
Short-term customer service P&L impact ranges between ~10% and 20%
Productivity uplift on Short-term Long-term
individual use cases today P&L value impact P&L ambition
~30-40% ~10-20% ≥ 60%
Individual use cases Short-term, realizable Leading companies with
proving uplift on productivity P&L impact across the function ambition to realize up to 60%
of ~30-40% already today of ~10-20% productivity increase long-term
Source: BCG research
6
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
Deep dive: short-term impact | Initial productivity uplift
for tech company already achieved today
BCG case example
What impact do they
Where did they start? What are they doing? already see today?
Large tech company with initial AI
Defined North Star vision to inform
deployment for their support agents Reduction in average
ambition and direction of the customer ~80%
focused on auto-case summarization time agents spend on
support organization
and knowledge mgmt. for cases
case summaries
Stood up an AI-focused program office
No unified vision for the desired
to ensure priority activities are executed
outcomes of AI
against
Inconsistent user experience and Developed a detailed impact model to Decrease in need
siloed, uncoordinated efforts to inform savings potential and help 10-15% of expert help to
implement AI use cases across teams prioritize use cases
solve cases
Limited enablement plan and Launched a change management
change management efforts leading program to ensure feedback is received
to poor adoption and key messages are disseminated
Increase in volume
Unclear use case prioritization Designed and launched continuous
~10%
approach for informing the engineering improvement process of AI-assistant- of cases handled
roadmap deployed use cases by agents
6
In addition to productivity, AI can radically enhance Customer experience…
To-be vision
AI-augmented agent
As-is Dynamic process based on multiple parallel
predictions improving average service quality
Human agent
Fixed, linear processes,
“Customer need” Is customer on the
“one size fits all” thinking right plan?
response, high variance
End-to-end process Has customer called
Customer in quality depending on Customer thinking before?
individual agent
Has customer been
Risk-based thinking
scammed?
Optimized
How much detail
Solve the 1st Personalization conversation
Explain the bill is needed?
order problem based on
What is customer’s
Situational thinking real-time
context?
“Why is my “Why is my context
Solve the 1st order
bill so high?” bill so high?” Explain the bill
problem
Commercial Make a
thinking cross-sell offer
Commercial thinking Make a relevant offer
Improve welcome
AI augmentation goes beyond Root-cause thinking
communications
just automating this process
7
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
8
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
… with benefits on both customer satisfaction
and commercial performance
Better customer experience Stronger commercial excellence
Pre-empt calls and foster self-help Foster sales excellence
• Prevent issues and requests from arising in the first place • Identify real-time sales opportunities
• Generate proactive actions resolving issues before need for call • Offer successful selling arguments and provide personalized
• Steer customers to self-service (voice/digital self-help) sales pitch to agents
• Use virtual avatar that proactively engages with customers in a
sales funnel to close the sale
Make interactions more seamless
• Use AI-powered assistants (chat, voice, or virtual [avatar]) to
Increase cross-/upsell performance
offer quick, accurate, round-the-clock support across service and
• Identify most appropriate by-product and provide sale details
sales processes
• Enrich recommendations with personalized sales arguments • Generate more personalized answers
• Enable more engaging conversational interactions (e.g.,
Reduce churn
generative self-service IVR)
• Spot customers at risk with predictive analysis
• Identify root causes of customer dissatisfaction
• Guide agents with ladders and personalized scripts
+10-20 NPS +20-35% CLTV1
1. Customer Lifetime Value
9
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
To realize full potential, transformation of the entire customer service
value chain required, with support response being the typical starting point
Pre-empt Self-heal Self-help Support response
Leveraging AI to Using AI to fix Empowering Enabling support
prevent issues issues before the customers with teams and agents to
and requests from customer notices AI-based tools and resolve customer
arising in the them and without information to issues in the most
first place customer effort self-solve their issues efficient manner
Typical starting point
10
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
Deep dive | AI transforms the entire customer journey,
including upstream prevention
Problem Pre-empt Self-heal Self-help Support response
Fixed without Automation Attended by
customer realization without humans humans
Prevent Fix error
Phone
problem from before customer AI voice bot Human agent
ever existing notices call
Web
Human agent
AI chat bot
chat
Product/ Customer Contacts
service identifies customer
problem problem service
Human agent
AI Email form
Email email identifies
problem
Branch Self-service Human agent
kiosk meeting
AI Human
11
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
Majority of value is unlocked by upstream prevention and realized
by transforming both technology and people and processes
20% (-30%)
15%
100%
40%
70%
30%
5%
20%
30%
tifeneb
ytivitcudorp
%
Illustrative
Pre-empt Self-heal Self-help Support response
Upstream prevention
~70%
People and
process
transformation
~30%
Standalone tech
Pre-empt Self-heal Digital Voice Handling time Multi-skilling Inefficiencies Productivity
self-help self-help reduction and smart from scaling benefit
routing
Note: As visualized, productivity benefit per use case often driven by combination of people and process transformation plus standalone tech; Source: BCG research
12
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
Many have started implementing AI in customer service – we identified
five common pitfalls that prevent value from being realized
Common pitfalls Three key success factors
Technology- Limited focus on people, process and change
driven management leading to low adoption
Focus on end-to-end
transformation
Individually built use cases that don’t re-use
Fragmentation
common components
Use case- Improving status quo w/o leveraging transformative Set an ambitious
centric power to change the whole service function top-down target
Implementation w/o pathway to scale and realize
POC-focused
business value
Measure P&L
Striving for AI to be perfect vs. providing a better
impact from “day 1”
Perfectionism customer experience than the average agent (while
of course still ensuring factually correct responses)
12
13
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
AI-driven transformation impacts all elements
of customer service operations
Support response
Pre-empt Self-heal Self-help
Inhouse | Outsourced
Demand management and orchestration Team skilling and structures
Workforce management and scheduling
Recruitment and retention
Learning and development (L&D)
Outsourcer management
Quality assurance (QA)
Knowledge management
AI Ops (new capability)
Tech (partner) ecosystem
Leadership ways of working
Change management
Deep dives
14
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
Deep dive: team skilling and structures | Three key shifts
Illustrative – not to scale
Today Future Key changes
Service Service
Ops Management Ops Mgmt.
Delivery Delivery
Move from people-heavy teams to
Supervisors Supervisors 20-40% smaller frontline teams in 2-3
years and 60-70% smaller frontline
Team Second Line/Back Office Experts
teams in the long-term
size
First Line First Line
Service Service
Delivery Delivery
AI OPS to be included as central
OPS element of the Service Delivery (with
Operations AI Core functions Core functions Sup-
Support1 Ops limited role of Operations Support in
port1
Team
the future)
composition
Service Service
Ops Mgmt. Ops Mgmt.
Delivery Delivery Shift from high share of time spent
Supervisors Supervisors OPS
Operations AI on run activities to continuous
Sup-
Support1 Ops
Second Line/Back Office Experts port1 improvement and delivery of
Team
First Line First Line long-term change
focus
Customer Service Technology Customer Service Technology and Data
Time spent on activities: Run Continuous improvement and delivery of long-term change
1. Functions including : WFM, knowledge management, partner management, training, QA, etc. 14
15
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
Deep dive: team skilling and structures |
Shift to multi-skilled second- and third-level support
Illustrative
Mass Market
Japanese language Japanese language
and Premium B2B
Produc St panish language ProducStpanish language
A A
Product Mass Market Premium B2B Product Mass Market Prem. B2B
ProduAct ProduAct Product
ProBduct B A,B,C
AProduct Product
Product
ProduBct ProduBct
A&B
ProCduct C
PBroduct Product
C C Customer Customer
Product Product
profile I, profile II,
C C
issue X issue Y
Starting point: many discrete Next 2-3 years: Target state:
teams in contact center, split by:
• Lower volumes handled • AI-augmented agents handle broad
• Product • Mix of contacts is more complex set of requests, customers,
• Contact reason • Agents assisted by AI co-pilot languages, etc.
• Customer segment • Teams become more multi-skilled • Routing based on individual best
• Language and fungible match, not job title
• Etc.
= single-skilled teams = multi-skilled team
Deep dive: AI Ops capability | New capabilities and roles
build, shape, and govern AI
Roles that Roles that Roles that
BUILD AI SHAPE AI GOVERN AI
Technology specialists who Business and functional experts Professionals who monitor AI
build and monitor AI models who collaborate with customer- outputs to ensure software is
and support technology platforms, facing agents to articulate business driving returns while verifying
leveraging advanced technical needs and integrate models into technology is being used safely
capabilities business processes and ethically
10%
Illustrative CS
40% CS
split of
80% CS
responsibilities:
90% Tech
60% Tech
20% Tech
16
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
17
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
Deep dive: change management | Focus on increasing leadership,
customer, and agent engagement
Enablement AI adoption AI adoption Agents’ skillset
of leaders by customers by agents and capability mix
Build case for change Nudge customers Establish training Drive change in
and narrative, incl. to use AI-enabled programs, change workforce capabilities
benefits and metrics self-service through networks and through agent
to provide leaders suggestions and feedback loops to upskilling, new hiring,
with tactics for leading adjustments to drive adoption and cultural change
the change and their options
driving adoption
18
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
Long-term ambition of value realization requires time horizon
of over 24 months with first benefits possible after 3 months
Boosting impact of Unlocking AI impact Reimagining service
early use cases through new use cases experience with AI
(at ~0-12 months) (at ~3-24 months) (at ~6-24+ months)
Increased adoption and Initial unlocks of full AI Fundamental shift of entire
success of early use cases driving potential via new high-impact service function that
measurable and marketable and moderate-effort use cases transforms the experience for
performance impact customers and employees
Incremental benefits within Step-function change in Watershed leap: Radical change
current service model experience and efficiency in experience and efficiency
19
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
Deploying and scaling AI effectively demands an optimized
technology stack tailored to support and expand AI use cases
Key considerations
Smart Business Layer
Case
Conversational App Copilot Text Enterprise System Other example
Guardrails and Monitoring
1
AI Layer Capabilities to ensure correct behavior of AI (e.g., RLHF,
Guardrails and Transparency and red-teaming, constitutional AI)
1 Content moderation Observability/feedback mgmt.
monitoring ethical guidelines
Orchestration
2
2 Orchestration Prompt flow Agents Chains Ops monitoring
New capabilities expected to coordinate different models and
PromptOps calls to internal and external APIs
First party - Foundational Other small Open-source Other small
4 Model models and pre-trained APIs models models models 3 LLMOps Ops and Monitoring
garden 3
Embeddings
MLOps New capabilities to ensure correct operation of AI use cases
(including models, pipelines, and data)
5 Model platform Featurestore Modelhosting/activation FMOps
Model Garden
4
Integration Layer Model capabilities required for use cases may impact near-
6 term platform selection. Open source for build use cases.
(expect multiple)
Data Layer Model Platform 5
Accounts
Call Client Internal Support multiple models, privacy controls, performance
and Products Other
recordings profile policies The layer will (1-2 preferred platform(s) in short-run)
contracts
facilitate
Core Transaction Layer access to the
Integration
6
client system
ERP C360 CRM Others Integrate AI use cases to client enterprise systems to leverage
for retrieving
client knowledge base, CRM system, ERP systems, etc.
relevant data
Infra and Cloud Layer and other
7 knowledge 7 Infra and Cloud
Public Cloud Private Cloud GPU TPU
resources Ensure AI choices align with overall hosting strategy
(multi/hybrid cloud); plan for higher infra consumption
Developed and implemented by BCG
19
20
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
Reusable modules are key to achieve scale at pace
Reusable MODULES … … that can be integrated into SOLUTIONS
Product modules rely on reusable sub-components that combine to enable Product modules integrate into Deep.ai solutions,
desired functionalities combined with proven UX/UI
Reusable code assets are bundled into reusable modules …
… enabling sophisticated
P1 Silence batch slicing M1 Call transcription customer service-
P2 Whisper on-prem integration M2 Transcription post-processing relevant solutions
P3 Transcription cleaning M3 Call summarization
Virtual
PX Speaker parser M4 Parameters extraction
Assistants
… that constitute larger tech capabilities. Combinations of
multiple modules augment each other's capabilities
Enterprise
Knowledge Content
Base Mgmt. Creation
Batch Real-time
AI - ML Real-time commercial Example
models Insights scripting Insights
Providing features Input for real-time Search Cognitive
and feedback and update of batch Reporting Engines
for future interactions
Way forward to a value-focused AI transformation of customer service
AI vision PoC and value Transformation and
and roadmap potential testing change management
• Baseline starting point and • Build and launch PoCs and capture • Drive and manage tech rollout
organizational challenges today learnings
• Execute operational
(including baselining of as-is,
• Evaluate technical architecture, data transformation at scale, e.g.,
narrative, tech foundation, etc.)
options and build tech readiness integrate AI into key processes,
• Define North Star vision/AI plan establish op. model
ambition and align key stakeholders
• Create detailed impact assessment • Drive change management and
• Prioritize and design use cases and test future value potential communications plan, e.g., change
agent and customer adoption, etc.
• Conduct high-level impact • Build out operational
assessment transformation plans for further • Select and onboard further tech
rollout and scale-up partners, as needed
• Set up roadmap for transformation
and tech rollout • Capture learnings and benefits
While some companies currently focus on PoCs, the full potential will be unlocked
through a strong, tailored vision and a successful transformation and change management
21
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
22
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
BCG experts | Key contacts for customer service
operations AI transformation
EMESA
Marcus Alfonso Ignacio Yasmine Juan Martin
Wittig Abella Hafner Hamri Maglione
Hrvoje Anne Alexander Nicholas
Jenkač Kleppe Noßmann Clark
NAMR
Simon Kirti Varun
Bamberger Choudhary Khurana |
111 | bcg | BCG-Executive-Perspectives-Unlocking-Impact-from-AI-Finance-EP5-24Oct2024.pdf | Executive
Perspectives
Unlocking Potential from AI and
GenAI
Finance
October 2024
Introduction In this BCG
Executive Perspective,
AI a key topic top for CFOs today. After working with numerous clients
we articulate our
in the past year, we are sharing our most recent learning in a new series
vision for the future
designed to help CFOs navigate AI. With AI at an inflection point, the focus
in 2024 is on turning AI’s potential into real effectiveness and efficiency gains. of finance with AI
In this edition, we discuss the future of finance and the role AI will play
in unlocking the finance function’s full potential. We address key questions on
the minds of finance leaders:
• Where are the opportunities for AI in finance?
• How will the finance processes evolve as a result?
.d
e
• What impact does AI have on the finance operating model and talent? vre
s
e
r
s
th
• How do I get started… and how do I get this right? g
ir
llA
.p
u
o
r
G
g
n
This document is a guide for CFOs and finance leaders to cut through
itlu
s
n
o
the hype around AI in finance and understand what creates value C
n
o
ts
now and in the future. o B
y
b
4
2
0
2
©
th
g
iry
p
1 o
C
Executive summary | Unlocking impact from AI in the finance function
While finance functions have taken a bit longer to get started, CFOs are now starting to
The time to act explore GenAI and how it can help finance become proactive "value drivers" for business
on AI is now
~15% of companies1 are actively piloting or implementing GenAI in the finance function
AI is a key unlock for finance to deliver new capabilities (e.g., AI-based forecasting) and
higher-quality business support (e.g., GenAI-powered FP&A insight generation)
AI unlocks many
sources of value AI delivers both efficiency and effectiveness benefits for the finance function, e.g.:
• Efficiency: 20-40% capacity unlock, enabling more focus on business partnerships
.d
e
• Effectiveness: 50%+ increase in forecast accuracy, 2x faster insight for decision making vre
s
e
r
s
th
g
ir
llA
CFOs start their journey by building high-value use cases and demonstrating early impact
.p
u
o
r
G
g
n
itlu
Executing Yet to fully unlock value from AI, finance functions should move beyond use case exploration to s n
o
C
n
successfully reinventing end-to-end (E2E) processes, by: o ts
o
B
• Adapting the operating model, talent, and ways of working to support new processes y b
4
2
0
• Modernizing the tech stack and rewiring data hierarchies to enable further value gains 2
©
th
g
iry
p
1: BCG CFO Excellence Panel survey 2024, N = 204 2 o
C
Next-gen finance functions break the compromise between efficiency and
effectiveness, to become proactive value drivers for the business
Next-gen finance functions actively …by delivering on BOTH
integrate strategy and drive value… effectiveness and cost
“Excessive” “Truly
excellent”
Increasing contribution to
company’s strategic
decision making Value
driver
s
s
Strategic Integrates e
n
partner corporate and e
financial v
i t .d
Business Evaluates strategy; c e e vre
partner performance partners with f f E s e r s
and partners re: BUs in th
Platform
Supports and
strategy with optimization g ir llA
manager reports BU BUs; actively & manages .p u
Traditional
Shares best
operations; d ari nve ds
r
ep sla on un rcin eg value driver o r G
g n
function practices and over ss eee tts in t garget allocation performance itlu s n
maximizes BU “Bureaucratic” “Minimalistic” o C
Serves as n
passive
synergies o ts
o
B
“bookkeeper" Efficiency y b
4
2
Finance functions Top performers 0 2
Facilitator Partner Initiator ©
th
g
iry
p
Source: BCG CFO Excellence Panel, N = 767 3 o
C
Inspiration | What next-gen finance functions look like
Common starting points Digitally mature functions AI-enabled functions
Planning, Fragmented data, heavy Excel use, Integrated cross-functional
Predictive AI models for planning,
budgeting, & manually intensive and siloed planning, modernized processes &
forecasting, and scenario analysis
forecasting budgets / forecasts tools with single-source data
Reporting Static tools / reports with limited Automated dashboards with Self-serve dashboards with GenAI-
and business visualization, built on manual standardized metrics, using based commentary, supported by
intelligence processing of fragmented data harmonized data & definitions conversational querying and search
Fragmented ERP; Excel-heavy, Lean GL, automated journal entries, Predictive close, AI reconciliations,
General
time-consuming reconciliations; system-triggered controls & GenAI entry recommendations, AI .d
accounting
manual statutory reporting exception monitoring pattern / error recognition
e vre
s
e
r
s
th
g
ir
llA
Lack of end-to-end systems, leading Integrated sub-ledgers, automated AI-based data extraction (e.g., OCR) .p
Finance u
o
to off-tool calculations, tracking / invoice generation / processing, & validation, predictive collections, r G
operations g
n
monitoring via Excel auto-matching and data validation supplier analytics, AI controls itlu
s
n
o
C
n
o
ts
Integrated solutions for transaction AI-based cash forecasting, FX o B
Expert Reliance on Excel & manual inputs y
b
processing and global monitoring, hedging models, working capital 4
2
functions limits visibility and future planning 0 2
with standard insights / alerts optimization, risk monitoring ©
th
g
iry
p
4 o
C
A combination of levers is required to achieve finance excellence, and AI is
an important lever in the digital modernization journey
Levers to achieve finance excellence:
Digital modernization AI tools bring cognitive and predictive capabilities to augment
including AI existing tech solutions (e.g., ERP, specialized tools, RPA)
End-to-end process
n
o
reimagination
i
t
a
e
r
• Reinvent the finance offering with c
Generative AI
e
•
n
A
dee
d
lw
o ivp
eA
t
r
I
A
- ep
I
x
o
a
pw
c
oe
r
nor ee
s
nsd
t
Ec iaa
2
lEp va apb
lr
ui oli ect eie ss
ses to
u l
a v
r o
f
R pro ob co et si sc
A
i
a
ln
enr att dei rf nli mlc
ii
ngi aa
e
gcl
n hc ie
n e
.d e vre
s e r
s
th g ir
llA
l a Best-of-breed automation .p u
i t applications o r G
n g
New operating model
e
t o
L ae ccg oa ucy
n
tE inR gP / (e.g., treasury) n itlu
s n o
and talent profiles P systems C n
o
ts
o
B
y
b
• Changes in roles, accountabilities, and 4
2
0
2
skills/expectations driving need for Recency of application ©
th
g
new capabilities and ways of working iry
p
5 o
C
While GenAI is still nascent, companies are actively exploring its potential, in
addition to expanding the use of traditional AI
GenAI in finance
still nascent, in GenAI in finance 10% 64% 11% 7% 8%
exploration stage
AI/AA-based forecasting 4% 12% 23% 61%
A
Traditional AI
&
P
already prevalent
F .d
AI-based report generation 6% 21% 35% 38% e
or being planned
vre
s
e
r
s
across multiple th
g
areas of finance g
n
AI-ba rese cod
n
a cc ic lio au tin ot
n
6% 19% 47% 28%
ir
llA
.p u
i t n o r G
g
u n
o
itlu
c c AI-based invoice recognition 4% 31% 42% 23% s n o C
A n
o
ts
o
B
y
b
4
2
0
2
©
Not explored Plan to implement Being implemented th
g
iry
Source: Select extracts from BCG CFO Excellence Panel survey 2024, N = 204 Exploring Pilot started 6 p o
C
Early estimates – potential to change as tech evolves
AI combined with other levers will be critical to unlock transformative
value across the finance excellence journey
Transformation of FP&A Transformation of Accounting & FinOps How AI accelerates
(~20-30% efficiency gains applying all levers) (~25-35% efficiency gains applying all levers) this transformation:
• Reduction of manual work,,
From To From To
e.g., data extraction, rule /
pattern-based validation,
Compliance tracking;
Business data support
calculation
system controls
Customer/supplier
Scenario models,
Preparing plans,
contracting and comms. • Task reinvention through
market analysis, AI- Policy writing, workflow
budgets, & forecasts
based optimization improvement, etc. AI-created first drafts, smart
Close & consol., .d
reconciliation and Predictive risk mgmt. reviews/alerts, analytics, e vre
s
Data analysis and Self-service and statutory reporting and controls; AI pattern etc. e r s
mac nre aa gt eio mn
e
o nf
t
s rt ea pti oc
r ts
proactive insight / error recognition
th
g ir
llA
Transaction tracking, GenAI-enabled • New finance capabilities .p
u
o
Stress-testing AI-built review and approval negotiation & analytics and insights enabled by AI r G
g
n
budgets & forecasts AI-enabled close, itlu
s
recons, & statutory n o
C
Data collection, reporting n o
processing, & validation Insight generation from Manual transaction ts o
dashboards; ad hoc processing
oE fx Ac Ie -p geti no en r- ab ta es de d
in
r ve ov icie ew
s,
Capacity unlock for more
B
y b
4
analytics 2 0
journal entries, etc. strategic advisory and 2
©
th
business partnering g
iry
p
Note: Estimated 3- to 5-year impact 7 o
C
Size = Illustrative share of workflow; pace of change will vary greatly by the starting point of your function
Use cases | Key opportunities exist for AI across finance processes
Planning,
Business strategy: Analysis AI-drafted plans and budgets, GenAI-created variance AI-based forecast and
budgeting, & of market/demand, leveraging KPI driver trees and investigation and scenario modeling, AI
forecasting competitive landscape, etc. automated data feeds commentary generation refinement of driver trees
Reporting
Standard dashboards/reports: GenAI- Ad hoc reporting: GenAI-based data Proactive monitoring (e.g., overspend,
and business drafted commentary, performance search and visualization, performance project delays) and optimization (e.g.,
intelligence analysis, creation of leadership decks analysis and insight generation working capital)
Subledger close: AI-recommended Consolidation & filings: AI-based Compliance & policies: AI-based controls
General
journal entries, proactive/predictive close balance sheet reconciliations, preparation and error detection, GenAI policy and
accounting
monitoring of statutory reports guidebook writing
.d
e
Finance Procure to pay: AI/OCR invoice processing and Order to cash: AI-based prediction models for credit Others
vre
s e
operations
matching, payment terms monitoring, supplier scoring, early warnings for DSO/aging, predictive (payroll, fixed
r
s th
g
spend/risk analytics + optimization suggestions collections with AI, GenAI-tailored customer comms. assets, etc.) ir
llA
.p
u
o
r
G
Expert T fore rea cs au sr tiy n: gA
,
I F-b Xa hs ee dd
g
c inas gh
,
caT lca ux l:
a
A tiI o-b na
,
s pe rod
a
p cr to ivv eis fio lan gi sn g
fo r
Inv ce as lt lo Qr &r Ael pa rt eio pn
,
is n:
v
E ea sr ton rin gs Ri ws ak
r
m nina gn sa ug se inm ge pn at t:
t
eE ra nr ly g n itlu
s
functions n
o
balance sheet optimization deferred tax impacts sentiment analysis recognition; fraud detection C
n
o
ts
o
B
y
Cross-finance opportunities: Chatbots, guided workflows, co-pilot support, etc. b
4
2
0
2
©
th
Analysis of opportunities based on additional value g
High Medium Low iry
that AI can unlock (beyond other digital tools): 8 p o
C
Example 1
AI in action (I/III) | Using AI-based financial forecasting to drive impact
Context Solution overview Impact
Large manufacturing client AI-based demand forecasting engine leveraging
internal and external data:
Struggled to forecast market 50%+
accurately (17% increase in forecast • Trend-sensing engine to identify early shifts in market Improved
forecasting
error)... sentiment, production technology, etc.
accuracy
..driving business challenges: • ML-based modeling to adjust for business trends such as
seasonality, competition, etc.
• Inability to anticipate and adjust to
market situation Driver tree models linking executive metrics 80%+
to operational variables (e.g., production units):
Reduction ..dd
• Increased costs of labor & overtime, ee
in forecast
vvrree
ss
transport, and increased inventory • More granular, data-derived forecasts for root cause variance ee rr ss
tthh
• Customer dissatisfaction from
analysis and sensitivity assessment and bias gg iirr
llllAA
..pp
uu
decreased service levels and fewer • Ability to model various demand / production scenarios oo
rr
GG
on-time, in-full deliveries and enhance decision making gg nn
iittlluu
ss
Forecasts nn
oo
AI/ML center of excellence created, to enhance models
and scenarios
CC
nn
oo
and develop new AI offerings generated
ttss
oo BB
yy
bb
rapidly 44
22
00
22
©©
tthh
gg
iirryy
pp
9 oo
CC
Example 1
Deep dive | Structured and leading-edge AI modeling provides
step-change in operational driver forecasting capabilities
Illustrative
Operational driver for financial model
Monthly demand (in #)
Endogenous time-series forecast
Baseline
Features: seasonality, long-term category trend,…
Machine learning time-series correction
Operational
Features: market data, competition, economy,
drivers
.d
e
COVID stringency,… vre
s
e
r
s
th
g
ir
llA
.p
Planner override (micro / macro) u o
One-off r G
Features: extreme weather occurrences, g n
events
itlu
regulatory changes, supply shortages,… s n
o
C
n
o
ts
o
B
y
b
Impact: Variability better explained & anticipated, forecast error cut, bias reduced 4 2
0
2
©
th
g
iry
p
10 o
C
Example 1
Deep dive | Driver trees link operational drivers to financial results,
underpinned by AI algorithms
Illustrative
Level 2 metrics Operational drivers Data inputs
OTR shipping
Automatic inputs
Cost per mile
OTR supplies
Intra-network Automatic data feed of actuals/
costs
forecasts directly from data systems
Miles
Miles per run
AI model outputs
Transportation
Runs
Outputs from relevant AI models
.d
e
costs
vre
s
(e.g., demand forecasting model) e
r
s
th
Units that serve as inputs into driver trees g ir
llA
.p
u
o
r
Pallets G
Units per pallet g
n
Outbound
Manual inputs
itlu
s
transportation n o
C
costs Cost per pallet Assumptions and manual adjustments n o
ts
(CPP) Historical CPP o B
y
b
4
2
0
2
©
Fuel price Δ
th
g
iry
p
11 o
C
Example 2
AI in action (II/III) | Using GenAI-based business intelligence to drive
impact
Context Solution overview Impact
Large US retailer struggles to get GenAI chatbot for conversational queries of data and
deep insights on performance, due generation of dynamic visualization:
to rapidly changing market / 2-4x
• Provides 2nd and 3rd order performance insight, with
demand environment, complexity of Faster report
suggested actions for improvement
generation
product portfolio / categories, and
fluctuations across cost areas • Creates live charts that can be interrogated, adapted, and
exported into leadership decks
FP&A analyst faces several
challenges while investigating Driver tree engine enables FP&A analyst to build and
variances to advise business live-edit dynamic relationship models between key <1 day
..dd
partners on the drivers behind operational and financial variables: Turnaround ee vvrree
performance:
• Driver model enables deep investigation into root causes
f ao nr aa ld
y
th ico sc ss ee rr
ss tthh
gg
• Poor view of drivers’ under/ behind variances, to enable timely corrective action
iirr
llllAA
..pp
overperformance uu oo
Scalable infrastructure that can be connected to any rr GG
gg
nn
• Significant time needed to analyze data source or system, providing visibility across BUs, iittlluu
ss
nn
oo
variances and identify root causes categories, and processes More CC
nn
oo
data--driven
ttss
oo
• Slow turnaround for ad hoc data BB
yy
decision making bb
44
investigation requests 22
00
22
©©
tthh
gg
iirryy
pp
12 oo
CC
Example 2
Deep dive | GenAI-based chatbot interface and driver tree relationships
allow for quick drilldown into root causes of performance
Dynamic charts generated
Conversation between analyst and tool throughout the conversation
FP&A analyst GenAI chatbot
What is driving the forecasted sales miss for [category A]
for November and December of 2024?
The forecasted sales miss is due primarily to a drop in sales velocity
(-25%), with a slight decrease in expected average price (-5%).
What are the drivers of the drop in sales velocity?
The forecasted sales velocity decline was primarily due to a large decrease
(-20%) in purchases per order and small decrease in order rate (-6%).
.d
e
What were the drivers of decreased purchases per order?
vre
s
e
r
s
th
g
Purchases per order forecast model attributes 50% of the decline to ir
llA
anticipated increase in inflation (+1%), to predicted CDP reduction (-2%), and .p
u
the remaining 35% due to a drop in expected competitor prices for similar o r
G
g
systems ([CLIENT] 15% higher than the expected competitor average price). n itlu
s
n
o
What actions can we take to increase purchases per order? C
n
o
ts
o
B
The economic outlook cannot be adjusted by any levers. However, planning y
b
4
for a steeper discount rate on [category A] heading into Q4 2024 will help 2 0
2
offset some of the expected losses and improve the outlook for CY 2024. ©
th
g
iry
p
13 o
C
Example 3
AI in action (III/III) | Using GenAI-based annual report creation to drive
impact
Context Solution overview Impact
Controllership team at a large GenAI-based creation of draft 10K/annual reports
company looking to reduce the time based on past filings and latest internal data
40-60%
spent on generating investor reports
• Tool auto-refreshes data, shifting human focus from Automation
by having Gen AI write the text-
data collection to review / refinement of statutory
heavy section:
reports
GenAI-drafted MD&A commentary, using external
• Manual data aggregation and
data to synthesize market trends and implications
consolidation for 10K-style reporting
• Tool adapts comments to highlight key business
• Significant time spent on analyzing
conditions, e.g., demand, industry fluctuations, 2K hours
peers and market trends to write ..dd
MD&A commentary
economic landscape cS oa nv te rod
l
li en
r
sth he
ip
ee vvrree
ss ee
Performance benchmarking across peers, based on organization
rr
ss tthh
• More time spent by consolidation
publicly reported financial metrics, strategic
gg iirr
llllAA
team on drafting reports and less ..pp
announcements, and other news uu oo
rr
time on review / insights GG
gg
nn
• GenAI provides quick answers on competitive
40+
iittlluu
ss
nn
oo
performance and peer outlook, replacing time- CC
Benchmark nn
oo
intensive manual analysis ttss
companies oo
BB
yy
compared via bb
44
22
GenAI tool 00 22
©©
tthh
gg
iirryy
pp
14 oo
CC
Example 3
Deep dive | GenAI-enabled tool supports finance function
in generating the annual report and inquiring about peer companies
Annual report generation First-draft commentary generated by tool leveraging past
• Creation of draft of annual report annual reports, news articles, competitor publications, etc.
commentary based on prior reports
• Benchmarks and sentiment-based
text editing, including translation
and proofreading
• Conversational interactions to refine
and improve on GenAI-created drafts
..dd
ee
vvrree
ss
ee
Peer reports and
rr
ss tthh
gg
press release inquiries
iirr
llllAA
..pp
uu
oo
• Query peer reports and press releases rr GG
gg
nn
to refine commentary around market
iittlluu
ss
nn
oo
CC
sentiment, industry outlook, etc. nn
oo
ttss
oo
BB
• Generate answers on peer financial yy
bb
44
22
performance based on public data 00 22
©©
tthh
gg
iirryy
pp
15 oo
CC
AI implications: processes | AI enables finance team members to shift from
manual data tasks to strategic insights and business collaboration
Data collection and Performance Insights and Management Illustrative time
variance calculation analysis recommendations reporting
allocation1
80%
FP&A analyst accesses Analyst looks into high Analyst writes Variance calculations
multiple systems to variance items by variance comments and comments are
pull and validate data reviewing source explaining findings and consolidated into
Current
data, cross-referencing flagging key cost items PPT report and sent
Excel used to calculate 20%
process
operational metrics, to be resolved and to leadership
variance vs. budget and
Example of and sending remaining budget
past actuals
variance analysis
Report Analysis
questions to business available
generation & insight
teams .d
e
vre
s
FP&A analyst uploads GenAI queries run to Analyst builds options GenAI used to build
70% e r
s th
g
prompt to GenAI tool, investigate driver trees to optimize cost and leadership report
ir
llA
.p
to quickly collate data and dimensions (BU, simulates P&L with recommendation u o
r
30% G
and run calculations region, GL item, etc.) impact of options with and P&L forecast g n
Future
itlu
ML tool s n
process GenAI builds charts, GenAI provides Analyst reviews and o C
n
o
comments for review reasons for the higher- ML tool enables live socializes report with ts
With GenAI
than-trend cost review and refinement stakeholders
Data Insight & o B
y b
4
collection & decision 2
0
variances with business team
calculation support
2
©
th
g
iry
1Based on client experiences of typical breakdown of FP&A time spent 16 p o
C
• Shift of CFO role toward "chief performance officer"
AI implications: driving strategic direction and decision making
New mandates
operating model | • Increased focus on data stewardship, to build new
for the CFO
insights and analytics for evolving business needs
AI drives changes organization
• Custodian of value, providing investment funding and
in roles, mandates,
monitoring benefits realization from org-wide GenAI efforts
and ways of working
across finance
• Evolution of finance service catalogue, with new
New offerings for the business (e.g., AI optimization engines)
engagement • Increased push toward AI-powered self-service,
models with driving leaner "finance business partner" teams
the business • Greater cross-functional collaboration, to fully leverage
internal financial, sales and operational data for insight
.d
e
vre
s
e
• New roles and profiles (e.g., solution architects) r s
th
g
to identify AI opportunities and build use cases ir
llA
.p
• Need for enhanced digital skill sets within finance, u o
Reinvention of r G
such as data analytics and AI capabilities g n
finance talent
itlu
(e.g., ability to create scripts) s n
o
and skill sets C
n
o
• Increased need for strong business acumen within ts
o
B
finance, resulting from new AI-based offerings and y b
4
2
reduction of transactional work 0 2
©
th
g
iry
p
17 o
C
AI implications: technology | Three types of vendors
for AI in finance are suitable for different needs and use cases
Nonexhaustive
Enterprise tech solutions Point solutions Foundation builders
Augments their existing ERP / EPM Leverages AI to offer specific solutions Provides infrastructure and out-of-box
offerings with AI capabilities tailored for use cases, with focus on models to support a broad set of use cases
building new analytics (including finance)
Example offerings:
Example offerings: Clients can leverage a mix of open- and
• GenAI invoice creation and AR
Approach
closed-source models / engines to create
management • AI / ML budgeting & forecasting
and use solutions tailored to their use case
• Controls / transaction review • GenAI management reporting & business
cases
intelligence
• GenAI customer / vendor comms
• Predictive collections
• Fraud / risk detection
.d
e
vre
s
e
r
s
Sample ERP, Ariba, Tableau
th
g
Concur GPT AI Assist
ir
llA
vendors .p
u
o
r
G
Ability to customize g n
Standard, scalable offerings itlu
s
n
o
C
n
o
Suitable to augment existing tools Suitable for companies looking for Suitable for companies looking to ts
o
B
with limited build effort off-the-shelf models with flexible build highly tailored models with y b
4
2
customization ability in-house resources 0 2
©
th
g
iry
p
Source: Expert interviews 18 o
C
AI implications: technology | Target state tech landscape for finance will
significantly evolve, driven by AI requirements
Illustrative platform design incorporating AI Immediate priorities for AI execution
• Define stakeholder needs: AI use cases need to be defined
Smart business layer
based on current needs, vision for finance capabilities
Interrogable dashboards/ Text generation
Chatbot/copilot/ Conversation-based
models/interfaces for (document creation, • Create roadmap: Companies are increasingly building
knowledge search code building
decision support emails, etc.)
quick AI pilots to prove effectiveness/efficiency impact
• Build models: AI tools can be developed today on top of
AI layer
existing stack, prioritizing areas with higher data fidelity
Model gardens for Machine learning / Knowledge graphs /
RAI guardrails • Set RAI guardrails: Standards are defined for data /
LLMs/generative text predictive models relationship models
models, based on responsible AI frameworks
.d
e
vre
s
e
Data layer
r
s
th
• Augment using AI: While AI models are evolving, ML g
Data products
ir
llA
capabilities can be used to accelerate data clean-up .p
Repository & storage Operational data services u
o
and improve quality / governance r G
Distribution & integration g n
itlu
s
n
o
C
Core transaction layer • Explore out-of-box capabilities: AI solutions are n
o
ts
o
increasingly being embedded into transactional B
Infrastructure and cloud
y
b
On-prem Cloud Hybrid TPU/GPU solutions (e.g., AI controls within ERP) 4 2 0
2
©
th
g
iry
New layer Transformed layers 19 p o
C
Getting started | 6 critical success factors for CFOs driving AI
1 Systematic Use AI as a catalyst to accelerate end-to-end finance transformation, including processes and
transformation operating model
2 Value-focused Act as the value guardian, driving the highest-impact use cases and monitoring early benefits
build realization
3 Technology Leverage off-the-shelf tools when possible and selectively build use cases
flexibility in-house when existing offerings do not fully address the requirement
.d
e
4 Data Be the "chief data officer," continuously exploring opportunities to better leverage big data vre
s
e
foundations for finance and business
r
s th
g
ir
llA
.p
u
o
5 Quality Establish safeguards against hallucinations and ensure reliability / security of results (e.g., r G
g
n
governance human-in-the-loop review, GenAI testing and evaluation)
itlu
s
n
o
C
n
o
ts
o
6 Leadership Get finance leaders and key business stakeholders onboard; drive change management/ B y
b
4
buy-in culture toward supporting AI efforts 2 0
2
©
th
g
iry
p
Source: Learning from BCG case experiences 20 o
C
NAMR CFO EXCELLENCE (CFOx)
BCG experts |
Michael Hardik
Key contacts James Tucker Jody Foldesy
Demyttenaere Sheth
for AI in finance
Laurin
Aissa Matt Malavika
transformation Henderson
Boudadi Harris Vishwanath
Menton
EMESA CFOx APAC CFOx
Alexander Marc Sebastian Anand
Roos Rodt Stange Veeraraghavan
Anne
Anna Hendrik .d
e
Oberauer
Ruellan
Schnelle
vre
s
e
du Créhu r s
th
g
ir
llA
.p
Andreas Norbert u o
Patrick Weber r G
Toth Wünsche g
n
itlu
s
n
o
C
n
BCG X o ts
o
B
y
b
4
2
0
2
Shervin Aaron ©
Mike Beyer Nick Tanaka th
Khodabandeh Arnoldsen g iry
p
21 o
C |
112 | bcg | BCG-Executive-Perspectives-Future-of-Sales-with-AI-EP2-5Aug2024.pdf | Executive
Perspectives
The Future of Sales with AI
B2B Sales
August 2024
1
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
Introduction In this BCG
Executive Perspective,
We meet often with CEOs to discuss AI---a topic that is both captivating and rapidly
we articulate the vision
changing. After working with over 1,000 clients in the past year, we are sharing our
and value of the future
most recent learnings in a new series designed to help CEOs navigate AI.
With AI at an inflection point, the focus in 2024 is on turning AI’s potential into real of sales with AI
profit.
In this edition, we discuss the future of B2B sales, and the role AI will play in
turbocharging growth. We address key questions on the minds of sales leaders:
• What will my sales team look like? Will I need a different team?
• How will the economics of sales change?
• How will the customer experience evolve as a result?
• How do I get started…and how do I get this right?
This document is a guide for CEOs and sales leaders to cut through the hype
around AI in B2B sales and understand what creates value now and in the
future.
2
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
Executive summary | The Future of Sales with AI
Market conditions and economics of B2B sales are rapidly changing: increased competition, evolving buyer
expectations, and economic uncertainty create a burning platform to reshape B2B sales
The time to act
GenAI is a breakthrough technology that, combined with PredAI, enables a step-change from traditional sales to
on AI in sales
augmented, assisted, and autonomous selling
is now
There is an opportunity to drive 1.8x margin impact through revenue growth and increased efficiency
Leading players are starting to scale, so companies need to mobilize to stay competitive
AI will reshape Reshape B2B sales teams and roles with massive seller productivity gains, augmented by AI team members
B2B sales teams and autonomous agents – with specific roles and scale of impact differing by industry
and customer Reshape customer experience by breaking down functional siloes between sales, marketing, and service, and
experience enabling new buying experiences
To successfully deploy AI in B2B sales and drive outcomes @ scale, organizations need to take a portfolio and
Executing transformational mindset, combine GenAI and PredAI within the tech stack to enable AI team members,
successfully and rewire the op model with a 90% focus on people and process change
requires a Sales leaders play a critical role in driving this change, breaking down siloes between teams, and making bold
transformational investments in tech and upskilling
mindset To get started, define your objectives and North Star, prioritize use cases, and start with proof-of-concepts that
demonstrate value, and scale up successive waves of capabilities while enabling the sales team
3
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
Why now | 6 key trends shaping market … fueling the burning
conditions and economics for B2B sales … platform to transform
Commoditization and a surge of new entrants are driving companies to invest in
Increased
retaining customers via post-sales support, customer success programs, relationship
AI can shape companies’
competition
management, and value-added services
response to these dynamics…
• Personalize offers and
B2B buyers expect a consumer-like buying experience with ease of access to
Shifts in buyer
information and quick response times, pushing sellers to offer intuitive and user- experiences
expectations
friendly buying processes
• Predict churn and trigger actions
• Automate routine tasks and
Longer sales Decision makers are getting more complex (e.g., buying groups) and taking more services
cycles time to evaluate options, due to increased scrutiny on ROI and cost-effectiveness
...while unlocking more growth
Sales teams are becoming more specialized, requiring more comprehensive and with higher returns
More sellers involved
integrated sales strategies to address complex buyer needs with cross-functional
in sales process • Drive more effective acquisition
teams
• Unlock better cross-sell/up-sell
• Reduce cost to serve
More complex The ecosystem of partners and marketplaces has grown in scale and complexity
channels yet remains a critical channel to drive scale and efficient cost to sell
Leading players are starting to
scale, so companies need to
Rising cost of goods, economic fluctuations, and uncertainty are leading to tighter
Uncertainty and mobilize to stay competitive
budgets and higher scrutiny on spending from buyers while increasing cost
budget constraints
pressure on own P&Ls
3
4
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
The future of sales | Turbocharging scalable growth with AI at its core
Current AI horizon Next (Gen)AI horizons
Age-old selling Augmented selling Assisted selling Autonomous selling
Entirely seller-driven Insight and productivity Real-time assisted selling Digital sales avatar
Subjective sales motion reliant Sellers armed with AI-powered Real-time support and assistance to Agents enabling auto-prospect,
on seller initiative next-best action, talk tracks, and sellers during customer engagements, nurture demand, 24x7 engagement,
basic workflow automation reshaping workflows and teams involving humans as needed
5
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
The Future of Sales Imagine a world where…
Autonomous sales agents … autonomous sales agents own sales motions from acquisition through support, for B2C
and long-tail B2B customers, enabling personalized selling at scale
GenAI-powered virtual … sales teams are augmented by GenAI team members, like intelligent sales assistants,
team members providing personalized scripts and customer insights or virtual solution engineers, navigating
complex portfolios and customizations
Divergence of strategic vs. … virtual sales assistants take over more transactional tasks, reducing the need for human
transactional sales intervention in standard transactions and freeing up time to focus on strategic and relationship
selling
“Smart selling” through real-time … predictive selling becomes the norm, with automated, real-time analytics and coaching fully
analytics and coaching integrated into sales tools, enabling agents to sell to the right customer, in the right moment with
the right offer, price, and message
AI-powered hyper-personalization … personalized offers, promos, sales pitches, and sell-in materials based on real-time buyer
behavior and data analytics are produced at 10x the speed, breaking down traditional silos
between marketing, sales, and pricing
Highly autonomous sales operations … fully automated AI systems manage much of sales operations, including targeting, lead
scoring and nurturing, and forecasting, reducing errors and increasing efficiency
Revolutionized sales enablement … AI-powered coaching and scenario-based learning based on real-world insights from
everyday sales interactions unlock step-changes in seller performance, reduced ramp-up time, and
dissemination of best practices into everyday action
5
6
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
Value | 1.8x margin impact through CLV growth and GTM efficiency
Seller Joe, with a sales quota of $10K EBITDA per week ($4K loaded per week + 40% indirect sales costs )
Realization of these types of
$1.5 EBITDA per $2.7 EBITDA per
$ GTM Cost $ GTM Cost impacts requires investment
… 50% better … 75% better 40% increase 50-60% of tasks ~50% decrease AI capabilities at scale,
in digital addressed by in back-office
acquisition1 cross-sell2 embedded in the flow
channels3 GenAI4 costs5
of work
$1.8K
Upskilling and new ways
$1.4K
$1.3K
of working for existing
$1.2K
sellers
$2.3K … 5%
expansion
… 25% $18K
in margin
lower New talent and agile
churn6
$10K operating model to
continually innovate
User engagement
More Higher value More digital Increased Optimized
customers per customer engagement seller back office throughout the journey
productivity
to enhance adoption
1.Assuming 30% new vs. recurring business; 2. Assuming 10% cross-sell of full deal value; 3. Assuming 20% digital value;
4.30 – 40% conversion of time to revenue; 5. Assuming 50% reduction in contract management, issue resolution, and data 6
management; 6. Assuming 10% churn rate. Source: BCG experience
7
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
Value | We are moving clients toward the future state and unlocking value
through cutting-edge AI solutions across the sales life cycle
Cross-industry, not exhaustive
Discover Learn Try Buy Use
Product-Need Identifier Sales Buddy Autonomous Chatbot
Sales Avatar (Sophia)
GenAI to identify product needs based on Offer recommendation visualization Suggest next-best answer to
Replace human touchpoint with
websites, PDFs, 3rd-party databases and call-flow guidance agent (based on customer
an avatar (video+voice)
• 10%seller productivity boost • 5%revenue uplift profile and product catalogue)
• 54%uplift in sales
• 26%conversion uptake • 33ppimproved offer accuracy • 54%uplift in sales
Real-Time Negotiation Support • 26%conversion uptake
Content Generator Provide agents with real time transcripts, • 2x increase in breadth of
Generate pictures/text for emails or sms summaries, and recommendation on next-best products sold
topicduring customer calls
• 5hrs/week time savings for seller
• 2xcross-sell and up-sell lift
Virtual Assistant
Reinvent the customer
Sales Info Assistant Post-Call Email Generator Customer Service Chatbot
experience with product reco
Support sellers with necessary info, on Generate emails based on conversation Customer-facing GenAI-
and trial
product description and application or client (GenAI content and outcome powered chatbot to handle queries,
• 2x higher ROI
RAG: from PDFs and websites) basic transactions
• New customer experience
• 30-40new cross-sell leads per rep • 20% reduction in customer
Relationship Co-Pilot
• 3-5%EBITDA increase in pilot regions service cost
Support account and relationship managers
to prepare for customer-centric conversations
RFP Responder Engineer Co-Pilot
GenAI agent to independently Support solution engineers by taking in
create and measure winning RFPs customer and technical needs from various Sales Coaching
• 50%faster creation input sources and develop specifications Train agents systematically and at scale based on call records, leveraging the right
meeting marketrequirements argumentation and coaching skills
• 35%quicker comparison
• 15%improved seller performance
Autonomous Agent Sales Assistant Solution Engineer Sales Coach
8
.devreser
sthgir
llA .puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
Example use case | Sales assistant provides next-best action,
improves quality of customer conversations, and increases efficiency
High impact during 1st experiments
Five key elements for AI buddy solution
1 Centralized Customer info, including interactions,
customer info in single, unified interface
2 Product Ranked product recommendations recommendations based on customer data and triggers
3 Comprehensive Descriptions, applications, and pitch
product information ideas, to support sales efforts
4 Real-time feedback Collection of feedback for continuous
and selling tips improvement and selling tips
5 Direct access to Products and SKUs filtered by
product details and availability, direct access to detailed
sheets product sheets provided for customer
presentations
8
.devreser
sthgir
llA .puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
Example: Industrial Goods
1
One-stop-shop of customer Customer Information Customer Interaction History Actioned Recommendations History Pin to Top Collapsible
Account Information
inform[aSatniiotizned,] including history 4
C -orporate Branch S Ce 1g 5ments S Inu db us se tg rm iae l Cnt os atings ImmC Ao uu stnert ar ly iadiate feC Pohela yi MndergebSoluationcs Inkc collection 30-40 New leads on cross-sell per
for continuous improvement
[Sanitized] External Data sales rep every month
Annual Revenue Number Of Employees
123414 124,235
[Sanitized] Status
Active
5
[Sanitized] I Nnternal Company Info on product category: 3-5% EBITDA increase in pilot
description, application,
regions expected in 20241
RECOMMENDATIONS Product Category CDM feedback: Accept Reject HOW DO WE ADaD VnALUdE pitch ideas
Share of Wallet •When offering this product category, also discuss with client our
Synthtic Waxes 5.0 Synthetic Waxes Send Mail blending services
Ethylacetate 4.0 98% Sales rep satisfaction and
Xylene 3.0 Description: DATA DRIVEN TIPS
Propanol 2.0 dS eyn sit gh ne eti dc tw o a px re os v ia dr ee sa p g er co ifu icp po rf o i pn ed ru tis et sr i sa ol tc h the em ci oca al ts in u gs se td h ein y t ah re e p ur so ed du ic nt ,io sn u co hf i an sd gu ls ot sr sia ,l h c ao ra dt nin eg ss s. , aT nh des se c rw aa tcx hes r ea sre is tc ar nea ceted through chemical processes and are •T toh ge e tc hu es rt o wm ite hr Sb yu ny t h3 e p tir co Wdu ac xt es s, .that are often used in end products high adoption rates across2
Dentonites 1.0 Applications: •The purchase of Synthetic Waxes has been rising in your region
Additives in the formulation of industrial coatings
Combination Product Trigger •Improving the durability and appearance of coatings
•Providing protection against environmental factors such as UV radiation and moisture
Other Inorganic Flame Retardants
Pitch ideas:
Seas No in tra il tT er sigger C • •o C Cs u ot nse t sf of ie smc tet izi nv ee td qs w uyn aa lt x ih e tye s t atic o n dW m sa e ux e pe t p s s lpw ye i ft c oh i rf is c uu ncp o ine ar tti eio n rr rg up pre e tr eqfo dur i pm re ra omn de uc ne ct ts ion [Sanitized] 6 Additional selling tips,
2 Gap FillingTrigger Alte Nr an ta ut ri ave l s w: axes = Silicone-base additives – Polyethylene waxes including reasons why
RecommeOrgnanicd Pigmaenttsions selected by AI
by AI andPurc htaser Reimgindegrsers,
Methylethylketone (MEK) 5.0 Product Search: Search Product Name Search
with ranking
Expe Oc tt hed e rO Gr ld ye cr os lethers 1.0 Polyolefin Wax 7 Products and SKU details,
Description: All Product Categories pS ry on vth ide et sic p w eca ix fie cs p a rr oe p a e rg tr ieo su sp o o t f t hin ed cu os atr tii nal g c sh te hm eyi c aa rl es uu ss ee dd ii nn , t sh ue c p hr ao sd gu lc ot sio sn , h o a[f r i dn Snd eu ass st n,r aia nil dtc o isa czrti aen tg cdhs. r]T eh se iss te a nw ca exes are created through chemical processes and are designed to filtered by availability Search by Product Category Search
Product information:
Precursor: Biocide, Chemical weapons, Explosive, Marine Pollutant
3 Format: Liquid Search option to get
Available SKUs:
information on other [Sanitized] 8 Direct access to
products during call/visit
Documentation: PDS_Polyolefin Wax.[pSdf a SnDiSt_iPzoelyodle]fin Wax.pdf product sheets
[Sanitized]
[Sanitized]
1.Based on initial indicators during project/pilot
2.98% of responses of sales-= rep are tool/logic "meets" or "exceeds" expectation in a fully anonymous survey
Source: BCG
9
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
How to get it right | Our perspective on winning with AI in sales
• Leading with a bold vision for the future of sales
Reshaping to
Reshaping B2B sales • Restructuring B2B sales teams to change the composition and
drive teams and the customer introduce AI team members to augment sellers
experience with AI
outcomes • Redesigning the customer experience by breaking functional siloes
between marketing, sales, and service
• Unlocking value with GenAI as the “next layer” to activate PredAI
Combining PredAI with
Unlocking decisions and precision
GenAI to maximize value
data and tech • Accelerating scalable solutions by helping engineer the target state
creation
architecture, leveraging the right ecosystem of partnerships
• Shaping the future sales roles and op model, and scoping the skills
Transforming people and
Rewiring the and change needed
operating model for
op model • Building an AI experiment and scale muscle through build-operate-
competitive advantage
transfer
10
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
Reshaping B2B sales teams | Investment in AI will enable
a change in team composition
Today Future Key changes
70%
redeployed for Brand Marketing • Marketing Ops – automated targeting and fast
Marketing Ops
brand marketing journey tailoring, higher focus on brand marketing
Marketing Ops
• KAM – more accounts per team as non-client-
facing work dramatically reduced
40%
• ISR – shift of interactions to remote settings;
redeployed for
many efficiencies from tech augmentation
growth
• FSR – many interactions moved to remote and
AR-enabled (including demos, order taking)
30%
customer service re-
deployed toward
customer success
Note: Illustrative – only selected sales roles shown for directional impact
1.Original function expected to reduce headcount, and other function (e.g., brand marketing, sales re-skilling, data and tech CoE) expected to increase headcount
MAK Acct Field
ISR
Mgrs Sales
eCommerce Specialists
Distributor Mgmt
MAK
Illustrative – roles and impacts
will vary by industry
1
Acct Field
ISR
Mgrs Sales
Customer Success • Customer Success – differentiator, requires
Customer Success human intervention for complex solutions
• Customer Service – heavy automation through
Customer Service
Customer Service self-serve and autonomous agents
• Sales Training and Enablement – AI to
Sales 1
80% Sales Re-skilling automate training, but critical change mgmt.
Sales Training
Training
sales ops and re-skilling task will be differentiator
redeployed to • Data and Tech – many tasks automated, but
Sales 1
Sales Operations data and tech CoE Data and Tech CoE new data/tech roles required for model training,
Ops
prompt engineering, product owner
11
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
Reshaping B2B sales teams | Value levers will differ across
sales models and industries
High
Precision matching Deep customer insights for
of product and customer complex relationships
(e.g., personalized product (e.g., deal and commercial coaching,
recommendations, product configuration) solution optimization and SOW generation)
Customer retention and
Higher service level and
direct digital engagement
hyper-personalization
(e.g., churn prediction, virtual
(e.g., offer/pitch personalization,
autonomous seller, automated call
pricing optimization and approval)
guidance and talking points)
Low
Customer concentration
ytisrevid
tcudorP
syenruoj
remotsuc
detamotuA
dna
tnempoleved
pihsnoitaleR
nalp
tnuocca
Deep technical and product information
Personalized, value-driven messaging
Low High
12
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
Reshaping B2B sales teams | New AI team members
will amplify the impact of sales teams
AI sales team supporting
the seller
Intelligent Sales Assistant
Seller
Owns administrative tasks and helps
sellers engage the right customers with
the right offers, qualify, convert, and
Customer close deals
Sales Planning and
Solution Engineer Operations
Develops proposals and solution Executes sophisticated
configurations, provides technical input planning to optimize
as a part of the sales cycle coverage, territory design,
and goal setting. Advanced
automation for deal desk,
Sales Coach
approvals, performance
Provides sellers with real-time management functions
recommendations, enables effective
Virtual Seller practice, guides managers on where to
spend time
Engages directly with customers from customer
identification through closure in an entirely
AI-powered channel
13
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
Reshaping customer experience | AI will break functional siloes
between marketing, sales, and service to better orient around the customer
Discover Learn Try Buy Use
Develop the content, With the help of the
In the time to typically do customer research, quickly
campaigns, and journeys to sales assistant, sales
tailor automated journeys and content that align to
enable sellers to tailor for teams can tailor
buyer needs and leverage seller insights
individual customer journeys campaigns, engage and
nurture customers
Sellers are more effective and efficient in
Time re-deployed to their core sales activities: relationship, Time re-deployed to
support retention pipeline, and deal management build upper funnel
With the help of the
sales assistant, sales
Set up relevant notifications around key issues Consolidate product, service insights;
teams can stay tuned
and adoption, tailor adoption journeys aligned create standard adoption plans
into adoption and
to buyer's value proposition and QBR templates
usage and plan the
right expansion plays
14
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
Unlocking data and tech | Combining PredAI + GenAI
to maximize value creation
(Traditional) Predictive AI/ML GenAI for augmentation
for decision making and automation
Precision – eliminating the guess work Productivity and performance improvement
Which client or person Insights about sales performance
Human augmentation
Which action
Sales prep
• Sales coach
• Sales collaterals
Which product
In-flight selling
• RT sales advisor
Future of • Product advisor
Which offer or price Sales Sales workflows
• Digital content generation
• RFP generation
Which channel
Automation
Autonomous sales agents
When
15
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
Unlocking data and tech | New reference architecture as tech stack
evolves to support integrated delivery of PredAI and GenAI at scale
Sellers and Customers
New/upgraded elements
ytiruceS
noitargetnI
Key evolutions
1 Engagement Channels Existing and new engagement channels 1 Engagement Channels become tightly
integrated through the smart business
Conversational apps (Gen)AI apps Omnichannel layer, enabling sellers and customers to
2 Smart Business Layer interact with each other and GenAI team
(Gen)AI services App builders Business services
members, seamlessly flowing between
channels of choice
Guardrails
2 Smart Business Layer adds (Gen)AI
Orchestration
E2E app Ops and
3 (Gen)AI Layer applications and supporting
vendors Model garden Foundation models ML models monitoring
development tools powered by GenAI,
also enabling the integrated delivery of
Model platform
PredAI and GenAI. Hosts the GenAI
team members
Data products
Operational 3 New (Gen)AI Layer supports secure
4 Data Layer Repository and storage
data services
access to and use of both internally
Distribution and integration
and externally hosted foundation
models, together with any existing ML
Core Transaction
CRM ERP models
Layer
Infrastructure 4 Data Layer includes new data sources
On-premise Cloud Hybrid TPU/GPU
and Cloud (typically unstructured and of new
modalities) and the means to ingest
and use them in (Gen)AI applications
15
Note: The engagement channels are typically represented as part of the respective components of the smart business layer, but
we have explicitly represented them here given their importance in the sales processes
16
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
Unlocking data and tech | Integration across all communications modes
made possible through smart business layer and GenAI
Phone
Text
Email
Collaboration and IM
Doc Creation
Mobile Experience
CRM
Integrated Apps
Stand-alone apps
reyaL
ssenisuB
tramS
Phone
Text
Email
Collaboration and IM
Doc Creation
Mobile Experience
CRM
Integrated Apps
ataD
lanoitcasnarT
dna
setadpU
Current State: Proliferation of mostly separate tools Future State: Integration across all channels, with
and channels, large effort to coordinate across them seamless seller interactions flowing between them
reyaL
ssenisuB
tramS
Seller Seller
• Attempts to create "SuperApp" as singular channel providing • GenAI enables integration of all channels, across all modes
all capabilities have largely failed (text, voice, image, video, etc.) via a smart business layer allowing
• Integration between channels is largely point-to-point for easy addition of new channels
• Manual work is expected to translate and capture interactions • Conversations with customers and GenAI team members seamlessly
apart from those in text (e.g., submitting call reports from visits) transition between channels of choice and need
Sales engagement density 16
17
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
Rewiring the op model | Sales AI transformation is 90% change management
Focus on people and process rewiring while building tech, data, and AI capabilities
Compared with typical data-driven transformation, the success of sales
AI relies even more on change management across a sales organization
Typical digital
transformation:
90% 10%
10%
AI
Focus on sales Focus on technology,
change management data, and AI
20% • Leadership activation: drive • Deploy sales technology
Data and technology
enthusiasm and clear sales vision to the frontline
• Sales team engagement: co-create • Utilize sales-specific ML
and iterate with sales reps models, traditional AI,
70% Business process and GenAI • Executional excellence: redefine
change management
sales processes and roles • Integrate sales systems
and automate E2E • Culture and effectiveness: adapt
sales strategies and KPIs
• Training and enablement: upskill
teams and build capabilities
18
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
Rewiring the op model | Five pillars of sales change management
to ensure sustainable impact from AI transformations
Leadership People Executional Culture and Enablement and
Activation Engagement Excellence Effectiveness Training
• Activate leadership to • Iteratively co-create • Refine sales • Implement new • Implement rapid tool
create role models for tooling and tech with the organizations and roles communication and training
upcoming change frontline to ensure robust collaboration tools
• Ensure responsibility for • Develop training plan
management technology from the get-go
global sales tool strategy • Introduce gamification (e.g., new role-play training
• Equip leaders with tailored • Continuously refine and roadmap features to drive peer leveraging KPIs/alerts)
messaging and tools to based on recurrent and competition/recognition
• Adapt key processes • Build up (Gen) AI
effectively frequent feedback
(e.g., shorter, more • Refine KPIs to reflect champions (black belt
communicate the change sessions
dynamic, cross-functional) productivity gains logic) to act as multipliers
vision and benefits
• Share progress with and drive change
• Review omnichannel • Ensure (short-term)
• Create excitement in frontline to foster trust in organically
collaboration including incentives drive
leaders and end users by the transformation and
capacity of team adoption and cross- • Activate leaders and
integrating comms outcomes of every sprint
members and priority functional/team/regional champions in sales teams
approach into existing
shifts collaboration (e.g., via train-the-trainer
sales, global, and geo
initiatives)
forums across channels • Reflect required • Implement user-level
governance changes monitoring
(e.g., decision input, tech
participants)
19
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
How to get started | Our perspective on the road to
unlock the value of AI in sales
Set a clear Define business objective and value levers for your AI transformation (e.g., productivity gains,
objective cost reduction, revenue growth), including upfront success metrics to measure impact
Define Define how you will leverage AI to reshape your sales approach and map required rewiring of
North Star op model (people, process, ways of working, etc.)
Assess Identify maturity of tech stack, create roadmap to required target, and invest in foundational capabilities
tech stack to sustain transformation
Prioritize Select use cases, starting with highest-value ones to fund the journey and defining detailed action plans
use cases to seize them; start small but build to scale
Build proof Develop proof of concept to validate value, test, and capture implications considering principles of
of concept responsible AI
Enable Create excitement, enable team participation, and protect learning capacity for quick upskilling and
your team early adoption through personalized change management plans
Develop a Rapidly develop a comprehensive workforce plan to identify and close talent gaps, ensuring the
workforce plan necessary skills and support are in place for the broader transformation
20
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
Our AI offer | Helping clients reshape the future of sales
Reimagination and transformation of client's sales org
As a partner
to fully profit from the new tech capabilities
• Higher effectiveness and efficiency of sales reps (inside, field, KAM, …) • Co-creation from
• Decision support for sales leadership and sales operations 1st day of project
Fast
• New tech-enabled routes to market
realization • Enablement to
advance the
of P&L
by
AI/GenAI journey
impact
• Integrated
at scale
approach,
Modular tech assets to accelerate time-to-market, including change
augmenting tech stack toward target state architecture management and
skills development
PredAI+GenAI Data Enterprise
Integrations …
assets platforms applications
21
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
NAMR
BCG Experts |
Phillip Bryan Japjit Audrey
Andersen Gauch Ghai Hawks
Key Contacts
for the Future of
Matt Justin John Marina
Kalmus McBride Merchant Nekrasova
Sales with AI
Ben Jit Matt
Quirt Tan Ward
EMESA
Alfonso Lena Roberto David
Abella David De Angelis Galley
Ignacio Juan Martin Patrick Basir
Hafner Roda Maglione Müller Mustaghni
Jatin Guillaume
Srivastava Triclot
21 |
113 | bcg | BCG-Executive-Perspectives-CEOs-Roadmap-on-Generative-AI.pdf | EExxeeccuuttiivvee
PPeerrssppeeccttiivveess
The CEO’s Roadmap on
Generative AI
March 2023
0
Introduction to this document
The release of ChatGPT in late 2022 is analogous to Mosaic’s launch
three decades prior. In 1993, it was clear that the internet would bring
a major revolution across all businesses in less than a decade.
The most focused of business models, and the strongest of brands can be In this BCG Executive
blown to bits by new information technology
Perspectives edition, we
-- Philip Evans, in his book "Blown to Bits"
explore how CEOs can take
Similarly, it is clear that Generative AI will bring another major
full advantage of the
revolution across all businesses. Today companies are focused on
productivity gains and technical limitations, but CEOs need to move coming revolution with
.d
e
the focus to business model innovation. v
Generative AI
re
se
r
sth
g
This is no small task, and CEOs—who are likely several steps
ir
llA
removed from the technology itself—may feel they are at a
.p
u
o
rG
crossroads. But from our perspective, the priority for CEOs is not to be g n
itlu
fully immersed in the technology. It is to understand how Generative sn
o
C
n
AI will impact their organization and their industry, and what strategic o
tso
B
choices will enable them to exploit opportunities and manage y
b
3
2
challenges. 0 2
©
th
g
iry
1 p
1 o
C
Human-AI augmentation of the future
Focus on standardization
• Use cases focus around automation
Prior to
and routinization to reduce
• Humans as passive recipients of technology tools
Traditional ML costs and replace human
• Humans as operators of processes
effort
Focus on augmenting
• Use cases around making decisions with data
With
decision making to create
• Humans actively using technology with data
.d
e
v
Traditional ML most efficient systems and re
se
• Humans as operators of processes processes r sth
g
ir
llA
.p
u
o
rG
With • Use cases around augmenting human creativity Focus on enabling greater g n itlu
sn
• Humans supervising AI on first drafts o
Generative productivity and creativity, C
n
o
AI/Foundation • Humans as designers of content and auditors of AI to solve unsolved problems
tso
B
y
b
/ Might augment decision 3
Models • Making decisions based on statistics and 2 0 2
making in some cases ©
sequencing th
g
iry
p
22 o
C
CEOs don't need to understand the technology behind Generative AI to create
business model innovation; instead, they need to understand its key features
No Code / Low Code "Infinite Memory" Lack of Truth Function
With a convenient chatbot-like Generative AI, trained on vast As a probabilistic model,
interface, Generative AI amounts of data, offers users Generative AI generates the most
democratizes access for all access to an automated system likely output to a query. This can
including those not well versed in that provides seemingly infinite sometimes create hallucinations .d
e
v
re
tech. "English is the hottest new memory and acts as a i.e., outputs completely separated se
programming language" according knowledgeable personal aide2 from objective truth
r
sth
g
ir
llA
to Andrej Karpathy1 .p
u
o
rG
g
n
itlu
sn
o
C
n
o
Defining features that will drive Business Model Innovation
tso
B
y
b
3
2
0
2
©
th
g
iry
NOTE: 1. Andrej Karpathy is a premier computer scientist and one of the founding members of OpenAI. 2. While not technically infinite, GPT-3 was trained on ~500 billion tokens, which gives users an 3 p o
C
impression of an "infinite memory" database
Executive Summary | CEOs must make choices across three key pillars
Discover your strategic advantage through experimentation
POTENTIAL a. Generative AI is accelerating across every industry, it is time to act now or be left behind
1 b. Use cases that rely on existing large language model (LLM) applications will be important to stay
Which use cases will
competitive, but they won't offer differentiation – CEOs need to discover the company's golden
differentiate your
use case
organization?
c. When use cases are identified plan the right implementation approach: fine-tune or train
d. Plan for long-term advantage through investment in talent and infrastructure
Prepare your workforce with strategic workforce planning and transforming op models
PEOPLE
a. CEOs will need to address key org questions for change management, talent and operating models
How should CEOs adapt
2
b. Generative AI will redefine roles and responsibilities across the organization
org structures and .d
prepare employees for
c. As AI adoption accelerates, CEOs need to develop a strategic workforce plan e v re
se
deployment? d. CEOs will need to consider new operating models,however we expect thatagile (or bionic) r sth
g
modelswill remain the most effective and scalable in the long term ir
llA
.p
u
o
rG
Protect your business with clear policies that address the limitations of Generative AI g n
POLICIES itlu
sn
o
a. Generative AI presents critical risks for which companies will need to be prepared C
How will the company 3 n o
ensure ethical guardrails b. Prepare for risk through clear policies and training that define roles and responsibilities on how to
tso
B
y
and legal protections are use Generative AI with a measure of confidence b 3
2
0
2
in place? c. CEOs should ensure the organization adapts responsible AI norms for long term risk mitigation ©
th
g
iry
p
4 o
C
BCG Executive
Perspectives
AGENDA Potential: Discover your strategic
advantage
People: Prepare your workforce
.d
e
v
re
se
r
sth
Policies: Protect your business g ir
llA
.p
u
o
rG
g
n
itlu
sn
o
C
n
o
tso
B
y
b
3
2
0
2
©
th
g
iry
p
5 o
C
Interest in Generative AI is exploding, fueled by the launch of ChatGPT
1a It is time to act now
Interest in Generative AI has grown This is driven by the release of ChatGPT,
exponentially since Q4 2022 which has taken the world by storm
Google Search Interest (100 = max interest)
Wall Street
100
Journal
90 Interest in
80 ChatGPT
70
60
.d
e
Even at it's peak, interest in v
45 00 t ch oe
m
m pe at ra ev te or s ie
n
td eo ree ss
t
n io nt 87 100 Fortune
re
se r
sth
g
30
ChatGPT today 59
ir
llA
.p
u
o
20 rG
g
n
10
14 11 11 13 9 5 3
itlu
sn
0 1 o C
TechCrunch n
o
09/21 10/21 11/21 12/21 01/22 02/22 03/22 11/22 12/22 01/23 02/23
tso
…and many more B
y
b
3
2
0
2
ChatGPT Metaverse ©
th
g
iry
p
6 o
C
Companies are already seeing a transformative effect from
using Generative AI
1a It is time to act now
Technology Consumer Biopharma
Automated on-model fashion image Generative AI Identified a novel drug
~88%
generation resulted in candidate for the treatment of Idiopathic
Pulmonary Fibrosis in
1.5X
Of software developers 21 days
reported higher productivity when using a
generative AI code assistant1 Increase in retailer conversion rate2
(vs. years with traditional methods)3
.d
e
v
re
Financial Institutions Entertainment Insurance
se
r
sth
g
ir
llA
Synthetic GAN-enhance training set for Generate novel animated motions from a InsureTech platforms leveraging .p
u
fraud detection achieved a single training motion sequence with generative AI to reduce up to o rG
g
n
~98% ~97.2% ~30% itlu
sn
o
C
n
o
tso
accuracy rate quality score on natural movements of customer service costs6 B
y
b
(vs. 97% with unprocessed original data)4 (vs. 84.6% with traditional methods)5 3
2
0
2
©
1. Quantifying GitHub Copilot’s impact on developer productivity and happiness 2. Vue.ai helps fashion retailers create high-quality on-model product photos 3.Deep learning
th
g
enables rapid identification of potent DDR1 kinase inhibitors 4. Using generative adversarial networks for improving classification effectiveness in credit card fraud detection
iry
p
7 o
5. GANimator: Neural Motion Synthesis from a Single Sequence 6. Insurtech COVU Leverages OpenAI to Streamline Insurance Agency Operations C
Total addressable market is expected to reach ~$120B by 2027
1a It is time to act now
Generative AI TAM ($B)
CAGR
121
2022-2025
32 BFSI2 75%
88
+66%
21 Consumer 64%
23
61
15 22 Healthcare 85%
15
35
11
15
16 Media 59%
.d
e v
re
2 2
9
1
122
3
3
18
4
3
23
6
65
5
48
19
0
9
7
11
9
42 11 92 OPu thb eli rc
1
sector 5 62 1%
%
se r sth
g ir
llA
.p u
o
rG
g
2022 2023 2024 2025 2026 2027 n
itlu
sn
o
C
131% 94% 91% 74% 44% 38% YoY growth of Gen. AI market n
o
tso
B
y
b
3
2
0
2
©
NOTE: 1. Other includes Industrial Goods, Energy, and Telecom markets; 2. BFSI includes Insurance (~$2B 2025) and Financial Institutions (~$13B 2025) including retail and
th
g
wholesale banking, asset and wealth management, and private equity
iry
p
8 o
Source: AI TAM research; Expert interviews; BCG analysis C
Focus on Building
Long-Term Advantage
1b Discover the company's golden use case
Strengthen competitive positioning with truly unique use
For the CEO, the
cases that both drive value and are challenging to adopt
key is to identify CORE OR (i.e., have a barrier to entry for competitors)
the company’s "GOLDEN" • For example, in pharmaceuticals companies,
Use Cases Generative AI can drive core R&D to produce new
so-called golden
drugs/molecules at record pace
use cases that
drive competitive
advantage
..dd
ee
vv
Non-core use rree
There is low barrier to adopting use cases that rely on ssee
cases are table
NON-CORE
e kexi es pti n pg
a
cL eL M
wi
ta hp op tl hic ea rt i oo rn gs a,
n
b iu zat tt ih oe ny
s
will be important to
rr
sstthh
gg iirr
llllAA
..pp
stakes, everyone uu
Use Cases oo rrGG
• For example,purchasing Generative AI tools that create gg
nn
will adopt them
automatic summaries of meeting notes
iittlluu
ssnn
oo
CC
nn
oo
ttssoo
BB
yy
bb
33
22
00
22
Table-Stakes to use cases ©©
tthh
gg
will improve efficiency
iirryy
pp
9 oo
CC
Golden use cases will add to a company's unique competitive advantage
in the marketplace, while non-core use cases are readily adopted by all
1b Discover the company's golden use case
Productivity Gains Efficiency Gains Innovation
First drafts Predictive maintenance Building novel proteins
with Jasper AI with an Equipment Manufacturer with ProFluent
What is Jasper Doing? What is the Equipment Manufacturer Doing? What is ProFluent Doing?
Web-based application for businesses powered by Building proof-of-concept for global end-to-end Creating novel proteins that do not exist in nature,
Generative AI that helps teams create tailored predictive maintenance of fleet with IoT sensors aimed at advancing drug treatment. Proof-of-
content up to 10x faster powered by Generative AI concept shown with creation of novel proteins
with anti-microbial properties
How is the Equipment Manufacturer Doing it?
How is ProFluent Doing it?
How is Jasper Doing it?
IoT sensors constantly monitor key indications of
Using "inverse design", i.e., working backwards
B tuu nil et
d
a
o
m
n
o 5d 0e +l uo sn
e
t -co ap
s
o ef
s
O sup ce hn A aI s' s
w
G riP tiT n- g3 ,, fine-
performance through signals from parts, and relay from desired properties to create proteins.
.d
e v re
that information back to a Generative AI powered Gartner believes that by 2030, 30% of new drugs se
copyediting, advertising, and content creation
back-end software
will be discovered using this method
r
sth
g
ir
llA
Why is generative AI better vs traditional ML? Why is generative AI better vs traditional ML? Why is generative AI better vs traditional ML? .p
u
o
Traditional ML incapable for such a task. It does Identification of anomalies in sensor data is Similar to Jasper, traditional ML does not have rG
g
not have any "generative" capabilities for new text difficult since failure data is rare in real-world. "generative" capabilities and thus is not great at n
itlu
adapted to use-case Generative AI can generate synthetic data, and creating never before seen protein structures by sn
o
C
better predict failures before occurrence self-learning from training dataset n
o
tso
B
y
Non-core use case Golden use cases b
3
2
0
Productivity improvement will be table stakes For the equipment manufacturer, high quality of maintenance is a core part of their business 2
©
since all businesses will adopt model. Similarly for ProFluent, protein synthesis is at the heart of their business. Generative
th
g
AI strengthens competitive positioning for both companies in their core business activities
iry
p
10 o
C
While foundation models today are used for generative use cases, this may
expand to include discriminative use cases as well in the future
1b Discover the company's golden use case
Not exhaustive
Any writing task
Algorithmic
(e.g., meeting notes,
trading
editing)
New material
Discriminative Dynamic pricing Generative
synthesis through
Recommendation
engines
uses of AI engine inverse design uses of AI
.d
e
Curr Te rn at dly
it
ii on
n
d ao
l
m Ma Lin of forD ece am sta in nd
g
Fraud
D
( de
ee
.
sgs
i.
g,i g nan
sr )c
d hir ta ef ct ts
ure
Cu Fr or ue nn dtl ay ti in
o
nd o mm oa di en
ls
of
v
re
se r
sth
g ir
llA
.p
u.d
e
v re s
e r
s th
g
ir
llA
.p
Ad spend
detection
Customer facing
o rG
g
nu o
rG
g
optimization chatbots
itlu snn
itlu
s
on
Co
n oC
n
tsoo
ts
Bo
y bB
y
Foundation models are currently being used f do ir
s
g ce rin me ir na ati tv ive
e
u use
se
c a case ses;
s
h ao sw we ev le lr, this may expand in the future to cover cert 1a 1in
3 2 0 2
©
th
g iry
p o
Cb 2 2 0 2
©
th
g iry
p o
C
Once use cases are selected, CEOs should make strategic choices
about whether to fine-tune existing LLMs or to train a custom model
1c Fine tune or train
Decision Tree for Foundation Model Choice
Open-source likely cheaper to
Cheaper
modify/operate, but requires more talent Consider open-source model
Consider data privacy implications of
Consider several tradeoffs – storing it in the cloud vs. on-premise Fast to implement
performance, budget, in-house Finetune
NO talent to modify/maintain, Models Limited flexibility
safety guardrails etc. Buy access to foundation
model provider Dependent on core-
Models from providers more expensive, but
Buy from provider with service agreement model
likely better along other dimensions
.d
e
v
re
se
D
i On
Re ds eir ve ed
l
oF pu en dc t mio on da eli lt sy does not already exist
NO
C Pao rtn nes ri d we ithr ma
o
p dea
l
r pt rn ove idr es
r
h wi hp
o supports
F pl oe tx ei nb til ii at ly fa on
r
d
r
sth
g ir
llA
.p
u
training a new proprietary model differentiation o
rG
Use case necessitates creation of proprietary Train g n
model to win competitive advantage Expensive and time-
itlu
Models sn
o
consuming C
Consider designing a n
o
YES proprietary model
tso
B
Technical know-how y
Large budget for Build/Train model in-house b 3
YES compute and In-house needed 2 0 2
©
Likely limited to select companies due to th
talent available for build g
costs, compute, and talent requirements iry
p
& maintenance of model? 12 o C
Training a custom LLM will offer greater flexibility, but that comes with
high costs and capability requirements
1c Fine tune or train
Develop New, Enhance Fine-tune
1 2 3
Cutting-edge Existing Existing
foundation model foundation model foundation model
Create a new foundation model Partner with LLM provider to significantly Fine-tune existing foundation model for
in-house from scratch. Costs scale enhance existing model (e.g., feeding related tasks (e.g., fine-tuning ChatGPT for
with model complexity complex company-proprietary data) legal memo writing)
$50 - $90M+ $1 - $10M $10 - $100k+
Estimated cost for complex models Estimated cost Estimated cost .d
e
v
re
se
Main drivers of cost: Main drivers of cost: Main drivers of cost:
r
sth
g
• Hardware (i.e.GPUs or TPUs): $30M1 • Training runs: $1M -$5M3 • Data gathering and labelling: $10k+4
ir
llA
• Training runs: $10M+2 • Partnership costs: variable • Computational costs: minimal
.p
u
o
rG
• People and R&D costs: variable g
n
itlu
sn
o
C
Usage Costs – $7M to $15M yearly (costs 30x to 50x lower if not using the most advanced model) n o
tso
B
y
b
3
GPT4 costs $0.06 for ~750 words. 5k to 10k employees each using the technology 100 times a day costs ~$7M to $15M 2
0
2
©
th
g
1. Meta's LLaMA used 2048 A100 GPUs for training, each of which can cost ~$20k. See https://wandb.ai/vincenttu/blog_posts/reports/Meta-AI-Released-LLaMA--
iry
p
13 o
VmlldzozNjM5MTAz?galleryTag=ml-news. 2. A single training run for GPT-3 is projected to cost $12M. See https://venturebeat.com/ai/ai-machine-learning-openai-gpt-3-size-isnt- C
everything/. 3. Training runs here likely less intensive than full-scale model training, leading to lower costs.
Carefully assess the timing of Generative AI investments considering tech and
talent; move too soon and risk wasting money, too slow and risk falling behind
1d Plan for long-term advantage
It could take 5+ years for low error tolerance use cases to be feasible1 … …but research is becoming
proprietary
Error tolerance
Key metric to evaluate readiness of Generative AI is the error tolerance of chosen use-case
Open-source:
OpenAI's GPT-2
Research is also
Adoption Illustrative moving very quickly:
Near-term | HIGH
1
error tolerance Meta's LLaMA
released 2/24/23,
Use cases where errors are OK outperforming GPT-3
• e.g., drug development since on many tasks
.d
e
Me tod (0li e -u 2rm a yrn se .c )r eror
L
t
(oo
3l
-w
e 5r
+
e
a
yr
n
rr sco
.e
)r
Long remaining
s
m
sc aoi fe eln
e
tyct i
u
as lt nes
d
sr e
u
ev
fg
fi
g
ie cew
as
c
te
e
yv de r by
y AI for Op
MeP enr tAo aIp
''
ssr Li Ge LPt aa
T
Mr -3y A;:
G 3/P 1T 4- /4
2
r 3eleased on
v re
se
r sth
g ir llA
.p
High error runway for generative u
o
tolerance AI adoption Longer-term | LOW rG
3 Waiting too long to invest into Generative AI g n
(0-1 yrs.)
3
error tolerance
today may mean that businesses risk falling
itlu
sn
o
behind. Research into high-performing C
Use cases with low room for error n
o
• e.g., doctors using chatbots to
foundation models is increasingly proprietary tso
B
2 retrieve and query a patient's and guarded as a source of competitive y b
3
1 medical history for easy access advantage. 2 0 2
Market maturity ©
th
g
1. Sequoia expects first drafts produced by Generative AI in certain domains to be better than human professionals by 2030
iry
p
14 o
See https://www.sequoiacap.com/article/generative-ai-a-creative-new-world/ C
BCG Executive
Perspectives
AGENDA Potential: Discover your strategic advantage
People: Prepare your workforce
.d
e
v
Policies: Protect your business re
se
r
sth
g
ir
llA
.p
u
o
rG
g
n
itlu
sn
o
C
n
o
tso
B
y
b
3
2
0
2
©
th
g
iry
p
15 o
C
To achieve the Human-AI augmentation of the future, CEOs should answer
questions for change management, workforce planning, and op model design
2a Address key organizational questions
Key considerations to craft a Generative AI adoption plan
Managing Culture and Strategic Organization and
Change in Company Workforce Plan Operating Model Design
Overarching Cultivate a culture that embraces Build a workforce that will be Create an efficient operating model
Goal AI like another coworker competitive 10 years from now that balances scale and agility
Key • How can professional identity concerns be • What new skills and talent will be crucial for • What existing roles and responsibilities will
.d
e
Questions managed to encourage AI adoption? long-term advantage? change because of Generative AI? v
re
Addressed:
• How can a culture of human and AI • What new competencies will managers need • How should I organize my departments for
se
r
sth
g
collaboration be fostered? to lead an AI-augmented workforce? efficient collaboration with AI ir
llA
• How can management communication • How should training/recruiting be adjusted • Where should LLMs and data scientists sit .p
u
o
create positive momentum to build a high-performing workforce? within the organization?
rG
g
n
itlu
sn
o
C
n
o
tso
B
y
A successful Generative AI adoption plan is customized to each organization, driven by the b
3
2
0
2
industry the company operates in, its current AI readiness, and the golden use cases it selects ©
th
g
iry
p
1166 o
C
While traditional AI has augmented the capabilities of managers and decision
makers, Generative AI will augment the capabilities of individual contributors
2b Redefine roles and responsibilities
Traditional AI/ML Generative AI creates
empowers individuals to first draft content,
make decisions, changing changing the role of
the role of managers individual contributors
This is the first time that a
Traditional AI and ML algorithms Generative AI augments technology developed in
augments decision making content creation
Silicon Valley benefits the
Lower-level individuals can now make data- Individuals will spend less time creating
lives of everyday people so .d
e
d
s
Tur hi
p
iv spe con
hr
d
t
ae nc gi esi so tn hs
e
w ri ot lh eo ou ft tm hea n ma ag ne am ge en
r
t
f rom
VS.
f
s
Ti ur hs
p
it se- d
r
cvr ha
i
asf
i
nt ns gga
e
A
sn Id
j
og
m
be
no taer se
ra
k
t sti
e
m
od
fe
c
i
nore dnv
it
vi es
in
din
t
ug
a
o
l
r
quickly a -n
-
d
S
as tyo
a
t Na an dg ei lb lal ,y
v re
se r
sth
g ir
llA
.p
u.d e
v re s
e r
s th
g
ir
llA
.p
decision maker to a manager of teaming contributors to include auditor or CEO of Microsoft o rG
g
nu o rG
g
and relationship dynamics supervisor of Generative AI itlu snn itlu
s
• For e.g., at ExxonMobil, geoscientists • For e.g., Andrej Karpathy, a founding o Cn o
n oC
n
use ML algorithms to decide where and member of OpenAI, said "Copilot has tsoo
ts
Bo
how to extract oil at maximum dramatically accelerated my coding… y bB y
e mff aic ni ae gn ec ry
s
with limited guidance of I
p
rd oo mn' pt te &ve en
d
r ite "ally code [anymore], I
1177
3 2 0 2
©
th
g iry
p o
Cb 2 2 0 2
©
th
g iry
p o
C
While traditional AI has augmented the capabilities of managers and decision
2b Redefine roles and responsibilities
makers, Generative AI will augment the capabilities of individual contributors
FROM: Key roles today TO: New roles tomorrow
A role centered around creating marketing A supervisor role with AI on content, with
content and executing campaigns increased time devoted to strategic thinking
• Creating marketing content and ideas • Supervising AI for first drafts of creative
from scratch briefs and brand guidelines and overall
better and faster marketing content
• Managing social media accounts,
scheduling and uploading posts • Building deeper relationships with
customers, suppliers, and brand
• Writing creative briefs to interface with
ambassadors
advertisement agencies
.d
• Tracking ad campaign performance metrics • Increased focus on brand strategy, e v re
se
• Creating brand guidelines to drive
positioning, and target audience r
sth
alignment across all stakeholders
identification g ir
llA
• Increased focus on personalized marketing .p
u
o
campaigns using Generative AI-powered
rG
g
n
tools
itlu
sn
o
C
n
o
tso
B
y
b
3
2
0
2
©
Core role changes for a marketer
th
g
iry
p
18 o
C
2b Redefine roles and responsibilities
Generative AI will redefine roles across the organization
Carefully consider the professional identity of your employees when making changes to role definitions
Tasks today that Generative AI Future tasks
Sample roles can provide first drafts for (in addition to verifying first drafts)
Social Media Creating social media content, scheduling
Building relationships with customers and followers
Specialist and uploading posts
Marketing
Advertisers Developing creative material (e.g., videos) Exploring new advertising channels and opportunities
Preparing and maintaining financial Identifying and implementing new accounting policies and
Accountant
accounts programs
Finance
Ensuring compliance with labor laws and regulations,
Payroll Specialist Processing employee payroll and taxes
providing guidance and support to employees .d
e
v
re
se
r
sth
Software Engineers
L tro aw n- sv la al tu ioe
n
coding and debugging, code R (ee .gv .i ,e bw ein ttg
e
rc o red ce
o
s ma mfet ey n, dd ae ts ii og nn i en ng
g
n ine ew
s
)complex algorithms g ir
llA
IT
.p
u
o
rG
Resolving system-wide problems, supporting complex
g
Help Desk Support Troubleshooting common issues n
technical issues
itlu
sn
o
C
n
o
Lead generation, follow-ups, logging Build relationships with customers, understand their needs
tso
B
Sales Rep y
b
customer interactions in CRM systems and pain-points 3
2
Sales 0
2
©
Develop complex pricing models, customized deals for th
Deals Desk Support Log quotes, and request sales approvals g
customers
iry
p
19 o
C
While traditional AI has augmented the capabilities of managers and decision
2b Redefine roles and responsibilities
makers, Generative AI will augment the capabilities of individual contributors
Employees are expressing concern about To successfully adopt Generative AI,
the impact to their professional identity CEOs must alleviate these concerns
Work with HR to understand how roles will
evolve and regularly pulse check employee
sentiment as their AI initiatives roll out
The Atlantic
Develop a transparent change
management initiative that will both help
employees embrace their new AI coworkers and
TIME ensure employees retain autonomy .d e
Magazine
v
re
se
r
sth
g ir
llA
.p
u.d
e
v re
s
e r
s th
g
ir
llA
.p
While some roles will be adversely impacted
o rG
g
nu o
rG
g
by Generative AI, overall Humans aren’t going
itlu snn
itlu
s
on
Fortune
anywhere — and in fact are needed to
C n
o
tsoo C
n o
ts
Bo
deploy AI effectively and ethically
2200
y
b 3 2 0 2
©
th
g iry
p o
CB
y b 2 2 0 2
©
th
g iry
p o
C
As Generative AI adoption accelerates, CEOs need to use their learnings
to develop a strategic workforce plan
2c Develop a strategic workforce plan
ANTICIPATE
Understand talent and skills needed
DEVELOP
to deliver on business strategy
Upskill and reskill talent at speed with
high reach and high richness • What workforce changes are needed
as the company steadily adopts Generative
• What key skills will be needed to work AI?
effectively with Generative AI?
• What are the risks associated with
• What training programs can upskill the workforce changes, and how to mitigate
DEVELOP ANTICIPATE
workforce at speed? them?
.d
e
v
re
ENGAGE ATTRACT se
ENGAGE ATTRACT r
sth
g
Deliver unmatched talent value Source creatively securing ir
llA
proposition and experience best-in-class candidate experience .p
u
o
rG
g
n
• How to create a culture of continuous • How should the interviewing process itlu
learning and development that encourages change to surface the talents needed in a
sn
o
C
n
employees to use Generative AI? Generative AI dominated world? o tso
B
• What is the company's value proposition • How should the sourcing process change y b
3
2
to employees in a Generative AI world? to ensure candidates with new skillsets are 0 2
©
attracted to the company? th
g
iry
p
21 o
C
Consider centralizing the IT/R&D function supplying LLMs and data engineers
2d Consider new operating models
Senior C-suite role (e.g., Chief AI Officer)
Elevates the importance of Generative AI to the C-suite
and signals importance across the organization
Finance
Each functional department
interfaces with the Central IT/
R&D to:
• Supply all collected data for
model training
Central IT/R&D HR .d
• Embed data scientists e v
re
Sales
aC no dl l se uct ps
p
d lia et sa d, atr ta ai n es
n
gL iL nM ees r,
s
w buit ih ldin
f
uth ne ctir
io
d ne ap la er xtm pee rn tit ss
e
to se r
sth
g
• Request data engineers to
ir
llA
.p
u
fine-tune LLMs for specific o
rG
use-cases g n
itlu
sn
o
C
n
Accounting
Operations
T wh iti hs c ar ce ea nte trs
a
a
l
as uca thla ob rl ie
ty
m foo rd de al
ta
o tso
B
y b
ownership and model control 3 2
0
2
©
th
g
iry
p
2222 o
C
We expect that agile (or platform) models will remain the most
effective and scalable in the long term
2d Consider new operating models
Decentralized
Sample model for a platform organization
Front end teams have autonomy to
serve customers
Division 1 Division 2 Division 3
(e.g., B2B business) (e.g., B2C business) (e.g., Mfg. business) Scalable
Processes are identified and scaled to
BD team BD team BD team
serve front end teams and to learn
Operation team Operation team Operation team
Front-end
Product team Product team … Product team Flexible
… … …
.d
e
Business EP Mer cc eh na tn erdise cU ense ter
r
cO er nd te er
r
P ca ey nm tee rnt …
T
a
Innec
d
th
e
ln
o
go
c
ral ao lig tzy
ea
da til olo nw
,
ts
o
f o cr
r
ep ae tr es o thn ea l piz ua lltion v re
se
r
sth
g ir
llA
.p
u.d e
v re
s
e r
s th g
ir
llA
.p
One source of all data and
o rG
g
nu o
rG
g
Data EP Customer Data Transaction data …
information
itlu snn
itlu
s
on
Co
n oC
n
tsoo
ts
Responsive B y bo B
y
Technology EP Safety ET Brain Cloud Computing … Modular technology available to
2
3all 3 2 0 2
©
th
g iry
p o
Cb 2 2 0 2
©
th
g iry
p o
C
BCG Executive
Perspectives
AGENDA Potential: Discover your strategic advantage
People: Prepare your workforce
.d
e
v
Policies: Protect your business re
se
r
sth
g
ir
llA
.p
u
o
rG
g
n
itlu
sn
o
C
n
o
tso
B
y
b
3
2
0
2
©
th
g
iry
p
24 o
C
Risks associated with Generative AI are showing up in the real world rapidly
.d
e
v
re
se
r
sth
g
ir
llA
.p
u
o
rG
g
n
itlu
sn
o
C
n
o
tso
B
y
b
3
2
0
2
©
th
g
iry
p
25 o
C
GenAI creates fundamental shifts impacting the Responsible AI (RAI) approach
Ease of use is much higher now: Shadow AI will be on steroids:
• Anybody (even non-technical staff) can • Capability overhang can emerge in
use these capabilities with very few unexpected corners of the organization
technical resources (e.g., data, (e.g., compared to only technical divisions
compute, expertise) before)
• Smaller need for large teams and • Time to detect, resolve, and mitigate
budgets limiting visibility for incidents is much higher violating the
Democratization managers and governance mechanisms principle of surprise aversion in risk
management
.d
e
v
re
se
Buying / renting Latent and opaque risks
r
sth
g
from 3PP: outside of in-house scope:
ir
llA
.p
u
o
rG
• Foundation models require a lot of • Limited visibility on data lineage (e.g., g
n
compute, data, and expertise and are copyright infringement) and model training
itlu
sn
o
C
overwhelmingly procured rather than (e.g., using confidential information to n
o
built in-house upgrade models)
tso
B
y
b
3
• Small set of entities can provide • Limited control on functionality changes 2
0
3PP Reliance 2
©
these foundation models on the technical roadmap
th
g
iry
p
26 o
C
Companies must be wary of critical risks of Generative AI today before
adopting the technology
3a Generative AI presents critical risks Not Exhaustive
Energy use and Capability Biased Copyright
environmental harm Overhang Outputs Infringement
Generative AI uses more energy on Due to its probabilistic nature, Real world data is often biased. Generative AI is trained on publicly
compute, both during model training Generative AI can sometimes show Without oversight, the Generative AI available data, much of which is
and usage than traditional ML. While unexpected capabilities upon models trained on this data also carry copyright protected. This can lead to
more efficient computation techniques deployment (e.g., several users tricked bias. Mitigation techniques include lawsuits by IP holders. Mitigation
are being developed, mitigation today is ChatGPT and bypassed its security to Reinforcement Learning with Human strategies rely heavily on foundation
limited to usage of more access kernel model). This risk is Feedback (RLHF) where the model is model providers to obey copyright laws,
environmentally sustainable energy difficult to fully mitigate, but extensive taught to be unbiased, yet this method and for governments to create new laws
sources pre-launch testing will help is not perfect for Generative AI
.d
e
v
re
se
Lack of Sophisticated Leaks of r
sth
g
Truth Function Phishing and Fraud proprietary data Shadow AI ir
llA
.p
Generative AI can sometimes produce Generative AI makes cybercrime easier When training Generative AI models in Employee application of external u o
rG
factually incorrect responses presented – generating convincing phishing the cloud, companies transmit generative AI tools without adequate g
n
in a very convincing manner. To emails or deepfakes instantly. To proprietary data which the data may be guidance or supervision creating risk
itlu
sn
o
mitigate risks from using incorrect mitigate this risk, companies must leaked in a security breach. To mitigate and causing harm. To mitigate this risk, C
n
o
information, companies must mandate strengthen cybersecurity protocols, this risk, companies can instead choose companies must create detailed and tso
B
double checking all Generative AI train employees on new safety risks, to train models on-prem vs. cloud, clear Generative AI use guidelines and y
b
3
outputs, and limiting its use to non- and consider deploying Generative AI although this necessitates other policies 2 0
2
critical tasks today themselves to to catch fraud tradeoffs ©
th
g
iry
p
27 o
C
Not all risks are created equal, with some posing a higher bus |
114 | bcg | a-genai-roadmap-for-fis.pdf | A Generative AI Roadmap for Financial
Institutions
NOVEMBER 13, 2023
By Stiene Riemer, Michael Strauß, Ella Rabener, Jeanne Kwong Bickford, Pim Hilbers, Nipun
Kalra, Aparna Kapoor, Julian King, Silvio Palumbo, Neil Pardasani, Marc Pauly, Kirsten Rulf, and Michael
Widowitz
READING TIME: 15 MIN
In just a decade, artificial intelligence (AI) has transformed from a promising research topic into an
accessible and pivotal technology at the center of a new industrial revolution. AI is reshaping
fundamental processes and functions across entire industries, from drug development and airline
scheduling to supply chain optimization and medical imaging. AI is no longer a concept of the future—
it’s a game-changer today. And companies that move ahead decisively and strategically with AI will
gain significant lasting advantages within their industries.
© 2023 Boston Consulting Group 1
Nowhere is this more evident than in the financial sector. AI runs on data, and banks and other
multiline financial institutions (FIs) command vast, high-quality, customer-centric gold mines of data.
In particular, the granular transactional data of a bank’s customer base can provide precise, wide-
ranging insights into behaviors, preferences, needs, and risks in ways that few other industries’ data
sets can. (See Exhibit 1.) For instance, in retail banking, leveraging AI to forecast and tailor future
product offerings on the basis of customer needs and behaviors is rapidly becoming table stakes in
many banking markets.
The rise of generative AI (GenAI) has enriched the broader AI toolkit, accelerating opportunities for
financial institutions to create new value with AI. The ability of GenAI models to digest (understand)
and generate (converse) in plain language makes AI capabilities more universally accessible, extending
the reach of AI assets to nontechnical users throughout organizations.
FI executives should take the arrival of this new phase of technology as an opportunity to commit to AI
and GenAI as key drivers of the industry’s future direction.
In this article, we chart a roadmap for this journey, from integrating GenAI into existing frameworks to
reimagining traditional operations through a complete AI transformation. In the rapidly changing AI
landscape, establishing a firm people strategy is as critical as adeptly navigating the challenges of
governance and regulation. And because the technology progresses daily, a forward-looking AI vision is
imperative for financial leaders shaping the future.
© 2023 Boston Consulting Group 2
Incorporate GenAI in Your Roadmap
Media headlines tend to cast an exaggerated and oen imprecise spotlight on GenAI. The resulting
hype and confusion have caused many executives to question whether GenAI will render their existing
AI strategies and initiatives obsolete. The clear answer is no. Indeed, to the contrary, GenAI
complements AI that is already embedded in existing FI strategies.
Many people in the financial sector informally use the term AI to refer to a subset of AI techniques that
focus on predictive decision-making models. (See Exhibit 2.) Over the past decade, this form of AI has
risen to prominence in FIs primarily because it addresses various prediction and classification
challenges that are pivotal to banking and insurance, such as risk monitoring, optimal pricing, and
product propensity modeling. We refer to this form of AI as predictive AI.
GenAI and predictive AI are powerful tools, but they serve fundamentally different purposes.
Consequently, using them is not an either-or question. A bank’s AI strategy will need to include both of
them going forward, harnessing their respective strengths in different ways.
One way to think of how predictive and generative AI complement each other is on the model of the
two halves of the human brain. (See Exhibit 3.) Predictive AI is comparable to the le side of the brain,
© 2023 Boston Consulting Group 3
wired specifically for logic, measurement, and calculation. This le brain comprises algorithms that
assign probabilities, categorize outcomes, and support decisions. For its part, GenAI acts as the right
brain, wired to excel at creativity, expression, and a holistic perspective—the sorts of skills required to
generate plausibly human-sounding responses in an automated chat.
Rather than negating the fundamentals of existing AI strategies, GenAI adds a new skill set to the mix.
Accordingly, leaders should lean in and consider how GenAI can enhance and extend their current AI
approaches by opening up new opportunities for AI-driven impact.
Many people may initially associate GenAI in banking and insurance with customer service chatbots,
but the technology’s versatility extends far beyond these applications to encompass tasks such as
automated financial analysis and AI-assisted code development. Numerous global banks are exploring
such uses for GenAI models (either built in-house or sourced as a service), and industry giants such as
Goldman Sachs, Deutsche Bank, American Express, and Wells Fargo are already starting to go live
with their solutions.
© 2023 Boston Consulting Group 4
When considering the new opportunities of GenAI alongside existing predictive AI–driven solutions,
leaders should bear in mind that well-proven and potential AI applications now span almost every
aspect of FI workflows, from client-facing roles to back-end operations. (See Exhibit 4.)
To take full advantage of these new GenAI opportunities, financial institutions must sharpen their
methods for identifying, prioritizing, and incubating initiatives that are likely to have the greatest
positive impact on value generation, customers and employees, and quality. Two guiding principles
© 2023 Boston Consulting Group 5
emerge for leaders: be clear about AI’s strengths and weaknesses, and take a disciplined approach to
AI experimentation.
The Boundaries of AI’s Capabilities
As with any tool, it’s important to use AI in suitable applications. In the case of predictive AI, for
example, a credit risk scoring system based on machine learning will make better lending decisions
than most humans when presented with simple credit card applications. But if the task is to assess
loans involving complex structured finance transactions in which every application is unique, it’s better
to let a human decide.
The same holds true for GenAI. A recent study by the BCG Henderson Institute, in collaboration with
leading academics, found that GenAI excels at tasks such as creative product innovation and that
human efforts to improve or enhance model outputs in these areas oen backfired and led to worse
results. On the other hand, for tasks falling outside the technology’s current capabilities, such as
solving business problems, GenAI underperformed against humans, more oen than not hindering the
performance of study participants who leveraged the technology.
In other words, GenAI performs best when humans act as complementors of GenAI output, taking over
tasks that fall outside AI’s domain of expertise (as in the predictive AI example of credit scoring). But
when humans act as enhancers—taking the output and trying to make it better—they can significantly
diminish the value of using AI. (See Exhibit 5.)
Experimental Discipline
© 2023 Boston Consulting Group 6
Evaluating and launching smaller-scale use cases within innovation-driven areas of the business can be
highly beneficial. Creating these types of AI laboratories can help nurture a broader appetite and
greater acceptance for AI solutions within the organization. They also offer a platform to refine new
techniques and build technical capabilities. And they provide a practical way to grapple with key
decisions, such as whether to develop technological foundations in-house, in-source them, form
partnerships, or explore other integration options.
However, the past decade of AI growth and AI experimentation has shown clearly that experimentation
can easily get out of hand. A broad “survival of the fittest” approach—that is, launching a large array
of small use cases to see which few succeed and flourish—oen yields disappointing results. The most
effective AI strategies involve conducting selective experiments in controlled laboratory-style testing
environments. This approach enables leaders to use insights gained from the experiments to pinpoint
a small number of high-impact AI opportunities and rally the organization around them.
As GenAI solutions evolve rapidly, the need for continuous experimentation will remain critical to
harnessing their full potential. At the same time, though, a disciplined approach to experimentation is
essential.
Reimagine AI-Enabled End-to-End Solutions That
Reshape Entire Journeys
The successes and failures of recent AI implementations indicate that companies see greater impact
and capture more value when they holistically reimagine entire processes end-to-end and with AI.
Isolated use cases that focus on a single part of a larger process can shine brightly for a short time, but
they oen burn out young, with the scale and impact of change falling short of expectations. And
incorporating AI into legacy processes built around the needs and capabilities of human workers can
lead to disjointed rollouts and potential friction for employees.
Beyond Tweaking—Transformation
The big wins from AI consistently come from broad transformations that involve rethinking the way an
entire process works as part of an AI landscape. An end-to-end approach isn’t a matter of inserting AI
at every step, but rather of redesigning processes from the ground up with both AI and human roles in
mind for optimized value.
The vast operations of FIs contain a powerful synergy waiting to be unlocked. By leveraging predictive
AI and GenAI in concert with human expertise, FIs can achieve enhanced process efficiency and
effectiveness—an impact greater than the sum of the parts.
Let’s unpack the roles of the two AI domains within FI workflows:
© 2023 Boston Consulting Group 7
• Analytical and Predictive Tasks. These le brain tasks, such as determining the best offer with
which to reach out to a customer, are appropriate for predictive AI.
• Creative and Expressive Tasks. These right brain tasks, such as creating the content and
designing visuals for the customer offer, are better suited for GenAI.
These two simple examples can form the core of a modern hyperpersonalized product marketing
campaign. Predictive AI and GenAI work hand in hand to automate most campaign tasks end-to-end,
from selecting the target customer to deciding on the many parameters and variables of an offering to
writing a tailored message and inserting custom-generated images. But even as AI streamlines many
aspects of the workflow, humans remain integral as complementors, supervising the process and
dealing with exceptions that require human expertise beyond AI’s capability.
Golden Patterns
Although numerous constellations of AI use are possible, many big opportunities that lie within end-to-
end workflows—in particular, opportunities that marry predictive AI and GenAI in complementary
ways—follow basic patterns.
© 2023 Boston Consulting Group 8
One such pattern consists of three steps: (1) process information; (2) evaluate/decide; (3) take creative
action. In practice, this might be the workflow for replying to a customer inquiry, processing a
supplier’s invoice, making a decision on a credit card application, monitoring an account for signs of
money laundering, or writing a section of an investment prospectus. (See Exhibit 6.)
In legacy processes based on human expertise, a human sis through the information, evaluates it,
comes to a decision, and then takes action. But each of these stages in the pattern is an opportunity
for predictive AI and GenAI to team up with the human.
Depending on the specific context, the first step (process information) might offer an opportunity to
use GenAI to synthesize and condense large amounts of information into easily digestible summaries,
or to engage the power of predictive AI to narrow the field of choices by extracting targeted insights
from large data sets.
In the second step (evaluate/decide), a predictive AI model can reliably make automated decisions on
cases that lie within its domain of expertise (typically the lion’s share of cases to be decided) and route
© 2023 Boston Consulting Group 9
the exceptional cases to a human in the loop. Here, the predictive model acts as the central steering
mechanism for the process, independently determining the need for human involvement.
The third step (take creative action), whether it involves composing a loan rejection letter, a suspicious
activity report, or a response to a customer’s question, can oen be turned over to a GenAI model—
for full automation of simple and/or non-mission-critical cases, or at least for preprocessing of
repetitive elements when the occasional imprecision of GenAI is a risk to full automation.
Repetitive, high-volume workflows that follow a golden pattern of this sort in one or more places are
game-changing opportunities to transform the process end-to-end.
Focus the Journey on People and Process, Not Just on
Tech
Rapid advances in AI make it all too easy to become fixated on the technology, the IT implementation,
and the data underlying it. And indeed, leaders face many important challenges here. AI is data-hungry
and can lead to uncontrolled data proliferation, so a clear data strategy is essential. And although a
GenAI model such as ChatGPT is very user-friendly, it is not at all IT-friendly to implement at scale.
But time and again we see instances where soer success factors—the target operating model and its
organizational structures, the approach to AI talent and skills management, and the change
management that must accompany any transformation—are underrepresented and underfunded
within bank’s AI strategies and prove to be the most critical success factors.
Operating Model and Organizational Structure
AI enables significant productivity growth. Work is automated or augmented, and roles must be
redesigned. We see four major types of impact on work that will alter roles across the organization
(and drive the many examples listed in Exhibit 4):
• Repetitive tasks such as low-code/no-code automation
• Knowledge synthesis such as review of all commercial loan agreements
• Data-driven decisions such as automation of vendor negotiations
• Creative tasks such as augmentation of code generation
To adjust to this change, FIs must be bold in rethinking people-driven processes and reimagining whole
functions. This effort will require the creation of more interdisciplinary teams with embedded data,
business analysis, and legal capabilities; the implementation of a flatter and more agile structure for
© 2023 Boston Consulting Group 10
quicker iterations and decisions; and a reduction in spans of control in order to handle the increasingly
complex nature of human work.
Finally, a platform operating model is critical to supporting successful AI adoption. An elevated market
orientation with greater ability to rapidly deploy people, processes, and data will support faster and
more assertive business model innovation and disruption. Cross-functional teams with end-to-end
ownership of products, journeys, and services will support reimagining whole processes, and the
platform operating model’s ability to drive scalability with standardization and without compromising
on customization will be a key enabler.
Talent and Skills
Going forward, nearly every human role will have a relationship with AI:
• Roles that build AI such as technology specialists who create and monitor AI models and
support tech platforms, leveraging deep technical capabilities
• Roles that shape AI such as functional experts who direct AI operations to deliver business
outcomes and integrate models into business processes
• Roles that use AI such as practitioners who work with outputs from AI models, interpreting
resulting content and data to deliver value to customers and employees
• Roles that govern AI such as specialists who monitor AI output to ensure that the soware
drives returns and to verify that the system uses tech safely and ethically
GenAI will have a high degree of impact on certain functions, including marketing, customer service,
legal, and soware development. These functions are likely to see extensive automation, resulting in
significant opportunities for cost reduction, demand generation via higher-quality service, and the
ability to focus resources on higher-value tasks.
Financial institutions must be pragmatic about implementing changes. This entails identifying which
roles have the highest value to their particular GenAI strategy and then developing an appropriate
value-added talent plan. (See Exhibit 7.) To manage the transition to GenAI well across all functions,
executives must integrate GenAI directly into their workforce planning process, defining skills required
in the future state, assessing current workforce potential, devising strategies for filling supply-demand
gaps, and supporting comprehensive culture and change management to inform the organization’s
“build, buy, or borrow” talent strategies.
© 2023 Boston Consulting Group 11
Prioritize Governance, Defining Your Own Rules of
the Road
Achieving transformative impact from AI and gaining acceptance of and trust for AI solutions within
the organization become possible only when the safeguards of a strong AI governance framework are
in place. Without solid governance, both predictive AI and GenAI can easily fall afoul of legal,
regulatory, and reputational hazards. The risk of bias against certain customers, for example, may
increase with large language models (LLMs) that train on biased public data sets obtained from the
internet. Company leaders are struggling with this difficulty, as a recent BCG survey of 2,000 global
executives found. Fully 70% of respondents said that concerns about the limited traceability of sources
of LLMs discouraged them from using GenAI, and 68% said that fear of the black box nature of the
technology and the increased risk of data breaches held back their implementation of GenAI.
© 2023 Boston Consulting Group 12
Regulators around the globe have been busy finalizing specific AI laws, amending them with GenAI
provisions, and updating data privacy, liability, and copyright laws for the new technology. (See Exhibit
8.) But the technology and its effects are evolving faster than ever, so regulatory uncertainty around
GenAI is likely to prevail for some time. Nevertheless, three frameworks are particularly noteworthy for
financial institutions. FIs should expect to receive special scrutiny in all three of these regimes, as their
products are considered essential to citizens and particularly sensitive.
The first and second regimes are the upcoming ASEAN Guide in AI Governance and Ethics (a guiding
framework) and, much more importantly, the EU’s AI Act (a risk-based consumer protection law that is
the first horizontal law on AI in the world). The EU AI Act classifies predictive AI and GenAI
applications into four risk categories. Applications that fall into the “unacceptable risk” category will be
banned from the European market, while applications that fall into the “high-risk” category will be
subject to pre- and post-deployment barriers and obligations. Common predictive AI creditworthiness
assessments will likely be high-risk applications, as will GenAI-powered customer support chatbots. Still
in its final negotiation stages, the EU AI Act is supposed to reach final form by the end of 2023 or early
2024 and, aer a grace period, will apply to all products in the European market. Failure to conform to
its requirements may result in fines of up to 7% of global annual turnover.
The third regime is the US regulators’ approach, which currently aims to adapt existing regulations
rather than to create new laws, and which takes a more national-security-driven perspective on GenAI
© 2023 Boston Consulting Group 13
risks. President Joe Biden’s executive order on AI issued on October 30, 2023, sets in motion a sector-
specific set of checks and balances, along with measures to foster the safe and responsible use of the
technology by companies and by the government itself. It is the first step toward legislation, but when
and how the US will regulate AI remains a subject of debate in Congress and within the administration.
With appropriate guardrails in place to guide AI developers and users, companies should be able to
deploy and quickly scale even rapidly changing technologies, with clear controls on the risks and with
high regulatory compliance. These guardrails should center on a framework that ensures alignment of
AI development and operation with the bank’s purpose and values while still delivering transformative
business impact. We call this approach responsible AI. (See Exhibit 9.)
A holistic and agile responsible AI framework must include five key components:
• Strategy—a comprehensive AI strategy linked to the firm's values as well as to its risk strategy
and ethical principles
• Governance—oversight by a defined responsible AI leadership team, with established escalation
paths to identify and mitigate risks
• Processes—rigorous processes put in place to monitor and review products to ensure that they
meet responsible AI criteria
© 2023 Boston Consulting Group 14
• Technology—data and technology infrastructure established to mitigate AI risks, including
toolkits to support responsible AI by design and appropriate life-cycle monitoring and
management
• Culture—strong understanding among all staff, including AI developers and users, of their roles
and duties in upholding responsible AI, and strict adherence to them
A recent BCG study in collaboration with MIT Sloan Management Review found that organizations
that successfully integrate responsible AI practices into the full AI product life cycle realize more
meaningful benefits. In fact, the likelihood of making full use of the benefits of predictive AI nearly
triples, jumping from 14% to 41%, when companies become leaders in responsible AI.
The rise of AI in the workplace will undoubtedly surface complex and pressing questions related to
human-AI collaboration and will probably elicit strong positions from workers’ unions on process
changes and technology implementation. Questions will arise that the new AI regulations do not
answer. But executives who prepare for this eventuality now by developing a holistic RAI framework
will have a critical advantage and will set up their AI transformations for success.
Aim for the Horizon
Like any foundational new technology, GenAI raises numerous important issues—around how to
realize opportunities for greater efficiency and effectiveness, but also around how to deploy the
technology, how to address the complexities of a new people strategy, and how to keep the technology
within the bounds of safe regulation and good governance.
The temptation to wait and see may be strong, but too much is at stake to play the short game.
Executives must make investigating and adopting AI, including GenAI, a transformational priority for
their organizations, taking a medium- to long-term perspective in their AI strategies, their HR planning,
and their approach to building a robust governance framework around the technology. Players that
actively plan today for the impending AI revolution in their ways of working will be at a decisive
advantage going forward.
© 2023 Boston Consulting Group 15
Authors
Stiene Riemer
MANAGING DIRECTOR & PARTNER
Munich
Michael Strauß
MANAGING DIRECTOR & SENIOR PARTNER
Cologne
Ella Rabener
MANAGING DIRECTOR & PARTNER, BCG X
Berlin
Jeanne Kwong Bickford
MANAGING DIRECTOR & SENIOR PARTNER
New York
Pim Hilbers
MANAGING DIRECTOR & PARTNER
Amsterdam
Nipun Kalra
MANAGING DIRECTOR & PARTNER
Mumbai - Nariman Point
Aparna Kapoor
PARTNER
Singapore
Julian King
MANAGING DIRECTOR & PARTNER, BCG GAMMA
Sydney
© 2023 Boston Consulting Group 16
Silvio Palumbo
MANAGING DIRECTOR & PARTNER
New York
Neil Pardasani
MANAGING DIRECTOR & SENIOR PARTNER
Los Angeles
Marc Pauly
PARTNER & DIRECTOR
Frankfurt
Kirsten Rulf
PARTNER & ASSOCIATE DIRECTOR
Berlin
Michael Widowitz
MANAGING DIRECTOR & PARTNER
Vienna
ABOUT BOSTON CONSULTING GROUP
Boston Consulting Group partners with leaders in business and society to tackle their most important
challenges and capture their greatest opportunities. BCG was the pioneer in business strategy when it was
founded in 1963. Today, we work closely with clients to embrace a transformational approach aimed at
benefiting all stakeholders—empowering organizations to grow, build sustainable competitive advantage,
and drive positive societal impact.
Our diverse, global teams bring deep industry and functional expertise and a range of perspectives that
question the status quo and spark change. BCG delivers solutions through leading-edge management
consulting, technology and design, and corporate and digital ventures. We work in a uniquely collaborative
model across the firm and throughout all levels of the client organization, fueled by the goal of helping our
clients thrive and enabling them to make the world a better place.
© Boston Consulting Group 2023. All rights reserved.
© 2023 Boston Consulting Group 17
For information or permission to reprint, please contact BCG at [email protected]. To find the latest
BCG content and register to receive e-alerts on this topic or others, please visit bcg.com. Follow Boston
Consulting Group on Facebook and X (formerly Twitter).
© 2023 Boston Consulting Group 18 |
115 | bcg | A-Gen-AI-Pathfinder-in-Health-Insurance-2024.pdf | A (Gen)AI
Pathfinder
Unlock the potential of (Gen)AI
in health insurance
August 2024
Dr. Heike Dorninger, Dr. Andreas Klar, Dr. Konstantin
Storms, Dr. Karin Tremp, David Wilhelm, Jakob Gliwa,
Dr. Andreas Benn, Clara Schlegel
Boston Consulting Group partners with leaders
in business and society to tackle their most
important challenges and capture their greatest
opportunities. BCG was the pioneer in business
strategy when it was founded in 1963. Today,
we work closely with clients to embrace a
transformational approach aimed at benefiting all
stakeholders—empowering organizations to grow,
build sustainable competitive advantage, and
drive positive societal impact.
Our diverse, global teams bring deep industry and
functional expertise and a range of perspectives
that question the status quo and spark change.
BCG delivers solutions through leading-edge
management consulting, technology and design,
and corporate and digital ventures. We work in a
uniquely collaborative model across the firm and
throughout all levels of the client organization,
fueled by the goal of helping our clients thrive and
enabling them to make the world a better place.
Executive summary
Three key actions
(Gen)AI—which is the combination of generative
artificial intelligence (GenAI) and “traditional” AI—
for health insurers
will enhance operational efficiency in health
insurance by up to 30% within the next five To fully capitalize on the transformative power of (Gen)AI,
years. (Gen)AI is revolutionizing industries, health insurers need a structured approach. The following
especially health insurance. This cutting-edge are three critical actions to take, ensuring you achieve
technology offers unparalleled opportunities to immediate benefits while laying the groundwork for
boost operational efficiency and significantly sustainable, long-term success:
improve health and financial outcomes.
1. Assess your starting point
Achieve impactful results within the first three Understanding your initial situation is critical. Assess
months of implementation—for example, up to a your technical readiness, existing data sources, and
50% reduction of customer inquiry search times. privacy/security requirements. Make informed build-
Health insurers are already diving into (Gen)AI, discovering or-buy decisions to tailor a unique (Gen)AI approach.
its transformative potential for their operations and ser- Establish a dedicated task force to guide these early
vices. These pioneering efforts showcase immediate bene- steps and support scaling efforts.
fits and set the stage for widespread adoption.
2. Focus on quick wins
Transformative (Gen)AI steps can begin within six Aim to enhance your competitive edge with (Gen)AI.
months—successfully implementing applications Prioritize high-potential applications through small-
initially regardless of technology and data setup. scale proof of concepts, allowing secure, manageable
While experimentation is vital, focusing on quick wins is experimentation. These projects provide invaluable
crucial for momentum. To truly harness (Gen)AI’s power, learning experiences and pave the way for
health insurers must embrace a strategic framework. This broader implementation.
means selecting applications based on business needs and
technical readiness, ensuring the identification of the best 3. Embrace a strategic transformative approach
near-term opportunities and building a foundation for Adopt a strategic transformation mindset from
future advancements. This approach also ensures the start, even with small proofs of concept. Focus
adherence to privacy and security standards. on the four pillars: strategy, organizational and
cultural adaptation, technology, and policy. This
(Gen)AI can be a key lever to address the reduction comprehensive approach ensures ethical and
of the 25% of preventable health care costs, effective (Gen)AI-driven transformation.
effectively closing the gap between identified
potential and actual gain. Scaling (Gen)AI applications The time to act is now, leveraging (Gen)AI to
requires a robust operating model anchored by four pillars: revolutionize the industry and secure a
strategic alignment, organizational and cultural adaptation, competitive advantage for the future.
technology enhancement, and responsible AI policies. By
acting decisively and strategically, health insurers can rapidly By implementing these three key actions, health insurers
unlock (Gen)AI’s potential, reaping immediate, tangible can not only achieve immediate and tangible benefits, but
benefits tailored to their unique circumstances. also position themselves at the forefront of innovation.
3 AGEN(AI)PATHFINDER
(Gen)AI is more than a trend
for health insurers
Health insurers are under increasing pressure to applications, including care management, claims
streamline their processes to effectively manage cost management, and customer service. For example, a
efficiency and the continually rising health care claims. prominent multinational health insurer is utilizing AI to
(Gen)AI is a key lever to drive operational efficiency and offer personalized health care, enhanced patient pathway
enhance health and financial outcomes to address the management, and clinical decision support. Additionally, a
challenges of the evolving healthcare market. tech-savvy US health insurance provider is committed
to simplifying the health care experience with innovative
(Gen)AI will enhance operational efficiency in digital tools and highly personalized customer support.
health insurance by up to 30% within the next
five years. These companies are revolutionizing the industry by
integrating advanced technology to enhance patient care
Nearly all daily tasks of health insurers can be optimized and streamline services. As the industry undergoes
with (Gen)AI (see Deep Dive Box), enhancing efficiency internal and external transformations, insurers must
and competitiveness. Some insurers already adopting strategically invest in (Gen)AI to maintain their competitive
(Gen)AI are achieving significant performance and edge and set future standards.
service advantages.
Examples include local, regional, and global health
insurers insurers utilizing (Gen)AI for a wide range of
Key introducing statements for health insurers:
• Health insurers need both traditional AI and GenAI • Investing in (Gen)AI optimizes tasks and sets
to stay competitive and streamline processes. industry standards, with some insurers already
benefiting significantly.
• (Gen)AI delivers the decisive “Wow factor” enhancing
client engagement and retention by utilizing various
data inputs to provide a personalized experience.
BOSTONCONSULTINGGROUP 4
Deep Dive Box
What is (Gen)AI?
AI, originating in the 1950s, has evolved significantly applications across industries, providing efficiency gains
with increased computing power and the ability to quickly, even with unstructured data.
make predictions from vast data. About 25 years ago, it
advanced to mimic the human brain’s ability to Traditional AI, reliant on structured data and strict
connect experiences, culminating in GenAI, which uses rules, remains crucial in health insurance for
deep learning and Generative Adversarial Networks underwriting, propensity modeling, budget allocation,
(GANs) to generate predictions and create outputs fraud detection, forecasting, and predictive accuracy.
from extensive data. GenAI can complement traditional AI or function
independently to unlock significant value. This
GenAI excels at generating new content, combination of traditional and generative AI is referred
understanding unstructured texts, engaging in to as (Gen)AI.
customer dialogues, and personalizing experiences at
scale, enhancing productivity and creativity with
minimal specialist knowledge required. It offers broad
Where can health insurers start?
(Gen)AI is a powerful technology with a wide range of (Gen)AI can be a key lever to address the reduction
capabilities pertinent to developing innovative solutions in of the 25%2 of preventable health care costs, effec-
the health insurance sector. The following are the five core tively closing the gap between identified potential
functions of (Gen)AI.1 and actual gain.
1. Information retrieval and summarization The numerous (Gen)AI applications along the value chain
present enormous potential to revolutionize this sector;
2. Content creation they must be carefully prioritized to ensure effective and
efficient integration. From working closely with dozens of
3. Conversational interfaces and language services (health) insurers around the globe, we have seen that
companies unlocking significant impact within a short
4. Data and information analysis and interpretation amount of time follow a methodological assessment of
their situation and prioritize (Gen)AI applications
5. Transcription, along with speech-to-text and text-to-voice accordingly. To fully capitalize on these opportunities (see
processing Deep Dive Box for examples), health insurers should start
with applications that are not only impactful for their
In addition, (Gen)AI supports further functions such as specific needs but also straightforward and quick to
multilingual translation/localization and personalized implement, allowing for valuable early experimentation
insights. Together, these core and additional functions form with (Gen)AI.
the building blocks for a wide range of different
applications along the health insurance value chain.
Exhibit 1 shows exemplary applications—from product
development and underwriting to customer service.
5 AGEN(AI)PATHFINDER
Exhibit 1 - Health insurers’ value chain and (Gen)AI application examples
Key functions
Product development
Analyze public data on trends, Optimize product launch based Process market research surveys
health care utilization and on market conditions and customer feedback to ...
competitor offerings to enhance generate new product ideas
own products
Marketing and sales
Recommend competitive rates Create pictures and videos Personalize text for an individual
by comparing competitors‘ for marketing campaigns approach to attract new customers
...
reimbursement rates at scale
Underwriting
Cluster along key criteria to Create text modules and scripts Adapt policy language
identify patterns for pricing for negotiations purposes to client clusters
...
and policy issuance
Care management
Analyze customer-specific data Develop individual care plan Offer AI-supported digital
(risk assessment) to offer based on demographic profiles therapies and coaching programs
preventive service and for selected patient groups ...
intervention measures
Claims management
Draft communications for Flag potential fraud cases with Ensure accurate claims outcomes
standard mailings to claimants intelligent validation rules, with price transparency
...
spotting unusual patterns and patient-friendly explanation
of benefits
Customer service
Cluster incoming requests Suggest customer- agent concise Summarize conversations and
along patterns, incl. sequence and p ersonalized answers store automatically in CRM
...
of request priority
Support functions
IT/Infrastructure: Controlling and Finance: Create HR: Support in the hiring
Generate software code dashboards/forecasts/ process or creation of trainings
...
simulations/business reports and job simulations
BOSTONCONSULTINGGROUP 6
Deep Dive Box
Three exemplary (Gen)AI journeys for health insurers
Deep dives into claims management, care providers and patients in submitting claims to insurers
management, and customer service show how by reducing errors and communication loops due to
(Gen)AI enhances operational efficiency, boosts false or incomplete data. By integrating (Gen)AI into
customer satisfaction, and reduces claims costs for payment processing, complex billing processes and
health insurers. These areas should be prioritized for payment delays can be reduced.
(Gen)AI deployment due to their significant business
potential. At the same time, health insurers should For example, BCG worked with an Asian health insurer
start with applications that meet four key criteria— to build and implement a (Gen)AI audit co-pilot
value, feasibility, risk, and strategic alignment—for application to reduce manual documentation and
valuable early experimentation with (Gen)AI. unlock efficiencies. Within just 10 weeks, the co-pilot
tool was tested and fully implemented, ultimately
Claims management journey. Based on our project reducing the manual writing and documenting
experience, (Gen)AI is already solving everyday processes for auditors by up to 50%.
problems and reducing costs in medical invoicing
management. (Gen)AI enhances claims verification Ultimately, end-to-end leveraging of (Gen)AI
through smart algorithms and shortens lengthy claims applications in claims processes will thoroughly change
settlement negotiations, for example, by selecting the current operations and unlock significant efficiency
most effective negotiation strategy based on historical gains (see Exhibit 2).
settlements. In addition, applications support health
Exhibit 2 - Claims management journey
Document tagging Claims processing Facilitated Subrogation
and summarization incl. outlier detection payment processing drafting
Documentation processing Augmentation of fraud Facilitated payment Efficient drafting of subrogation
and summarization, detection measures with processing, integrated with documents, incorporate case
automatically annotate smart analysis, enriching payment gateways and plans, details for claims involving
documents and images input with further data and secured transactions third parties
Time-intensive document Challenge of outlier� Complex billing action Complexity of subrogation
handling and data entry claims/fraud detection and payment delays processes/doc. drafting
Claims intake Claims verification Claims settlement and Post-processing
and data capture and adjudication payment processing and reporting
Inefficiencies/errors Difficulty of client inform. Complex/lengthy Inefficient communications
during claim submission verification on accuracy claims negotiations of claim status
Assistance for Client information Claims negotiation Automated status update
claim submission verification strategy advancement and communications
drafting
Assistance to submitting Patient information and Using historical claims
health claims, considering document verification of (patterns) for modelling the Automated communications
prior authorization policyholder status, drafting most effective claims and drafting of personalized
and inaccuracies communication requests negotiation strategies communications to
interested parties
7 AGEN(AI)PATHFINDER
Care management journey. Care management factor. By pro-actively addressing at-risk members to
involves numerous data sources that are often see a general practitioner or cardiologist, up to 20%
integrated into disparate, unconnected systems. Here, reduced hospitalizations in heart failure
(Gen)AI can showcase its far-reaching potential by were realized.
structuring, integrating, and analyzing various data
sources to drive personalization of both the customer Additionally, (Gen)AI enables patient-specific selection
and patient experience across all treatment areas (see of appropriate and available providers—including
Exhibit 3). Notably, AI-driven predictive analytics can appointment management through predictive
more accurately and efficiently identify at-risk patients, algorithms that factor in real-time data on cancellations
thereby enabling highly personalized care management and emergencies. These applications highlight
with early preventive health care that ultimately the need for a clear vision and strategy of (Gen)AI
increases overall patient well-being and health. For implementation to realize incremental potential from
example, a German health insurer implemented an AI combining traditional AI and GenAI.
model co-developed with BCG to predict the risk of
hospitalization for heart failure in the next 12 months
out of all hypertension patients with a known risk
Exhibit 3 - Care management journey
Inefficient risk assessment/ Provider finder and Virtual treatment and Patient data
patient monitoring appointment management therapy assistant summarization
Predictive analytics for Efficient management of Patient inquiries response for Patient data summarization
holistic risk assessment, early provider finder (timely and 1st level support with real-time from different sources to
intervention strategies and most suitable) and support, tracking of compliance be brought into a
tailored care plans appointment times and treatment mgmt. comprehensive overview
Provider finder and Virtual treatment and Patient data
Holistic risk assessment appointment management therapy assistant summarization
Initial assessment/ Care planning and Transition and Evaluation and potential
risk stratification implementation continuity management plan adjustment
Complex patient records
and data synchronization; Inconsistent patient Delayed responses
generic care plans compliance with care plans to patient needs
Patient data Personalized Ongoing communication
management health reminder facilitation
Data analysis from multiple Personalized patient care Alert automation and
sources to enable timely support through automated messaging systems for
interventions, chronicles and reminders for medication, communications between
personalized treatment plan treatments, and others care teams
BOSTONCONSULTINGGROUP 8
Customer service journey. This area is often adjustments and new product offers. By supporting
considered the prime example of (Gen)AI applications, agents with reply suggestions and providing 24/7
with several near-off-the-shelf solutions readily available assistance, (Gen)AI is set to revolutionize the customer
and a huge potential for efficiency gains (see Exhibit 4). service experience, addressing the high demand for
Our experience with health insurance clients shows personalized interactions.
that (Gen)AI can automate up to 70% of standard
customer inquiries, thereby freeing up agents to focus For example, together with multiple health insurers
on complex tasks. around the globe, we build and implemented (Gen)AI
customer service applications, enhancing customer
With limited resources, customer inquiries often follow service and customer experience. Once fully
a first-in, first-out principle, causing delays in serious or implemented, these applications have realized up to
time-critical cases. (Gen)AI can prioritize urgent 30% customer service cost reduction through higher
requests and manage complex customer relationships, efficiency and up to 6% topline uplift through better
making real-time recommendations for contract customer retention and acquisition.
Exhibit 4 - Customer service journey
Proactive customer
Inquiry categorization Efficient onboarding Policy modification, relationship management,
and prioritization and training coverage optimization churn prevention
Clustering of incoming Interactive training environment, Analysis of customer data to Analysis of customer interaction
requests along patterns, simulating real-world interactions; recommend adjustments of history, drafting proactive
while identifying sequence of instant feedback and gamification policies to adapt to evolving personalized communications,
request priorities and flagging elements; accessible in idle times needs of patient, boosting preventing termination by
of most critical requests cross- and up-selling effective retention
Long waiting times for Labor-intensive and Complexity of contract, Lack of target and timely
answers due to lack of slow onboarding/training inefficient cross- and outreach for effective
prioritization of new staff up-selling relationship building
Initial contact Resolving Promoting Post-service
and triage customer inquiry additional services processing
Challenging to initiate Difficulty of finding the Generic offerings not aligned Time-consuming
contact, esp. during right information to answer to customer needs, also due post-interaction, such as
non-business hours accurate and detailed to lack of agent expertise details for CRM
24/7 customer Enhanced Additional services: Efficient CRM
assistance response quality customized product/ documentation and
service offerings follow-ups
24/7 customer assistance in Accurate and fast information
clarification and retrieval and response Offering customers products Entry of customer interaction
understanding of coverage, suggestions; routing of inquiries and services based on analyzed details in CRM system, with
benefits, and other topics, to the right support level interactions, thematic interests, suggestions for follow-ups based
providing consistent service local offers; incl. customized on analyzed content/outcomes
payment plans of interactions
9 AGEN(AI)PATHFINDER
Three key factors to consider Understanding which (Gen)AI applications yield the best
short-term results helps health insurers adopt (Gen)AI in
when determining the alignment with their needs and capabilities. This requires
assessing their individual situations, including available
starting point for leveraging
data sources, privacy and security requirements, technical
readiness, and build-or-buy decisions (see Exhibit 5).
(Gen)AI applications
Exhibit 5 - Three key factors to consider when determining the starting
point for leveraging (Gen)AI applications
Data privacy and Technical Build-or-buy
security prerequisites readiness decision
Understanding data Assessing companies’ Narrowing down the
utilization possibilities tech readiness build-or-buy decision
Data privacy and be achieved using methods such as anonymization,
encryption, or access control. Moreover, data selection
security prerequisites for applications extends beyond data privacy and security
prerequisites to include considerations of data availability,
governance, and quality (see Deep Dive Box).
While data protection and security rules vary across the As a straightforward initial step, health insurers can use
globe, the health sector is nearly always among those publicly available data sets and non-sensitive internal data
industries with the most rigorous provisions, as health to minimize compliance risks, unlocking significant
data is by its very nature the most sensitive. Therefore, potential in the process. As they improve their capabilities
when leveraging (Gen)AI, health insurers must consider for data handling with regard to (Gen)AI, insurers can
and ensure IT security as well as adherence to applicable then progress to use more sensitive data. This methodical
data protection and security requirements. In this context, approach not only mitigates the initial risks, but also
insurers must keep in mind that there might be company- enables strategic data usage to be expanded—and
specific data governance guidelines with stipulations facilitates the gathering of the crucial technical data
beyond those legally required. and human resources capabilities needed to reach the
The (Gen)AI applications utilized must meet all regulatory next level.
security and data privacy requirements, which can
BOSTONCONSULTINGGROUP 10
Deep Dive Box
Data sources for (Gen)AI applications
A solid understanding of the necessary security • Internal company data, such as internal
standards and applicable data protection requirements knowledge databases or provider information.
for a given data source is an absolute prerequisite for These can be leveraged for (Gen)AI applications to
the source’s use within (Gen)AI applications. Basically, provide more insightful and relevant results, such
the following are three types of data sources in this as those required for smart chatbots. However, it is
context—each with a broad spectrum of possible crucial to establish authorization concepts to prevent
(Gen)AI applications: confidential information from being shared with
customers.
• Public data—either as open source or possibly
via licenses or payment models. For example, • Personally identifiable information and strictly
public data sources can be leveraged for (Gen)AI, confidential internal company data. It is crucial
creating images or video material for marketing to ensure compliance with all internal and regulatory
campaigns, requiring a thorough vetting (incl. requirements. Therefore, integrating these data
copyright laws) before implementation to ensure sources into (Gen)AI applications requires flawless
high-quality and guideline-compliant results. data governance.
Technical readiness
Achieve impactful results within the first three
months of implementation—for example, up to a
While (Gen)AI applications can be implemented in any 50% reduction of customer inquiry search times.
technical environment, defining an initial scope that takes
into consideration integration complexity ensures quick To determine where to start with (Gen)AI technology,
implementation. Most health insurers, regardless health insurers of all technical maturity levels should
of their technical readiness, can easily start with (Gen)AI consider launching low-threshold solutions that promise
applications that require only minimal tech integration. short-term efficiency gains. Building on this foundation,
Short-term adoption is also feasible for insurers with and based on their technology readiness, health insurers
legacy systems characterized by manual processes and can then gradually develop their own systems into more
isolated data storage, and thereby little scalability (see powerful, more integrated applications. Doing so ensures
Deep Dive Box). that each step is sustainable and in line with their
evolving technological landscape.
Deep Dive Box
Potential methods for (Gen)AI implementation
Standalone solutions. Suitable for insurers with warehouses or data lakes. Example: (Gen)AI chatbot
lower tech maturity. Alongside standalone solutions, integrated into customer relationship management
prerequisites on data applications and infrastructure (CRM) systems but monitored by a human in
levels can be established simultaneously to enable the loop to provide intelligent and effective
deeper integrations. Example: A (Gen)AI knowledge personalized communication.
assistant using company-specific data to draft customer
Future scenario. Health insurers with advanced
responses, reducing processing times.
modular IT systems can leverage (Gen)AI extensively
Modern system clusters. Focus on web/app across business units and launch low-threshold
development and cross-functional areas such as input solutions for short-term gains, then gradually develop
management and customer service software. more integrated applications based on technology
readiness. This ensures sustainable progress aligned
Service-based architecture. For insurers with
with the evolving technological landscape.
modern, flexible tech stacks (APIs, cloud infrastructure),
integrating (Gen)AI more deeply into IT systems
using centralized data hubs or stores such as data
11 AGEN(AI)PATHFINDER
Build-or-buy decisions
• Before deciding to buy, it is essential to evaluate
expected value, total costs, and contract terms, while
With the increasing accessibility of (Gen)AI and its reduced ensuring IT, compliance, and business teams
requirements on data structure compared with traditional collaborate to avoid costly provider lock-in and long
AI, it is expected that the health market will see a lead times.
significant rise in both standalone solutions and
enhancements for widespread proprietary software, While data sources and technical readiness are crucial,
emphasizing the need to carefully consider overall (Gen)AI objectives, costs, speed of implementation,
implementation strategies. and application performance should also be considered in
build-or-buy decisions. Proper assessment can reduce lead
Determining which data should be used for initial (Gen)AI times from months or years to weeks, speeding up the
applications and assessing technical readiness are the realization of potential.
building blocks for key decisions when getting started with
(Gen)AI—what to build and what to buy: In sum, addressing data privacy, security prerequisites,
technical readiness, and build-or-buy decisions enables
• Building (Gen)AI applications in-house is advisable for health insurers to start implementing (Gen)AI and embark
applications that can be modified and reused across on long-term transformation. Beginning with a low-
various scenarios, leveraging widely adopted tools to threshold proof of concept provides a realistic overview of
accelerate the building process. Key advantages (Gen)AI’s impact, allowing insurers to build internal
include have greater control of tools, better integration capabilities and scale applications across an organization.
of proprietary data, technical flexibility, and This step-by-step approach helps insurers realize efficiency
development of internal (Gen)AI capabilities. potential, gain experience, and align their organization,
ultimately unlocking the full potential of (Gen)AI and
• Buying a (Gen)AI tool is advantageous when existing building transformation chains across all business areas
providers offer easily deployable add-ons, minimal globally (see Exhibit 6).
customization and integration are needed, the
application relies on public data, and resources for
building a custom solution are unavailable.
Exhibit6-(Gen)AIpathwayevolution
Boost efficiency
Enable first insights Focus on scaling
and effectiveness
(Gen)AI
maturity
(Gen)AI transformation
Understand Initiate early Explore benefits Leverage early Continuously
(Gen)AI‘s potential (Gen)AI engagement of (Gen)AI learnings and scale optimize and
build chains
Gain a first Start experimenting as Discover how (Gen)AI Refine with first
understanding of how soon as possible Improves efficiencies and learnings and set up Continuously optimize
(Gen)AI can be Develop initial effectiveness operating model (Gen)AI for more impact
applied within your applications for testing Provides a competitive Scale indiv. applications in Build transformative
specific context and learning, enable PoCs advantage selected business areas chains across the entire
organization
BOSTONCONSULTINGGROUP 12
Key (Gen)AI starting point statements for health insurers
• Ensure compliance with data protection standards • Choose between building custom (Gen)AI
for public, internal, and personal data, adhering to applications or buying existing solutions,
legal and internal guidelines. considering cost, value, and technical capabilities.
• Start with simple (Gen)AI solutions for legacy
systems; use more integrated applications for
modern tech stacks.
How can health insurers
leverageandscalethebenefits?
After establishing a starting point and gaining insights culture, technology, and policy, is essential for this transfor-
from small-scale (Gen)AI proofs of concept, a transforma- mation (see Exhibit 7). When supporting our (health)
tion is needed to fully realize (Gen)AI’s potential across an insurance clients on these end-to-end (Gen)AI transforma-
organization. This requires ensuring effective operation tions, addressing the four key transformation pillars has
and widespread utilization of (Gen)AI solutions to derive proven effective to unlock up to 10% margin uplift over
greater value and functionality. Implementing a multi-level 12–18 months.
operating model, based on strategy, organization and
Exhibit 7 - (Gen)AI path – key transformation pillars
Data privacy and Technical Build-or-buy
security prerequisites readiness decision
Strategy Organization and culture Technology Policy
Leverage strategy Future-proof Design tech Shape (Gen)AI to
to define the initial organization through architecture to strengthen corporate
(Gen)AI playing field right-sizing, up-skilling, build and scale values and avoid
and set the right org. and tactical hiring (Gen)AI across regulatory pitfalls
ambition entire organization
13 AGEN(AI)PATHFINDER
Strategy:defininggoalsfor (Gen)AI will transform their company, industry, and
business model. This includes setting ambitious goals
value creation through (Gen)AI for value delivery of (Gen)AI and the required scale of
investment. These new aspirations should be
integrated into the ongoing strategic planning cycle.
A well-defined strategy that incorporates (Gen)AI is
To embed (Gen)AI ambitions into the overall digital
essential for setting the goals to create value through
strategy, it is necessary to break down general goals
(Gen)AI. This strategic foundation is crucial for effectively
into specific objectives for core business functions.
deploying and successfully scaling (Gen)AI within an
organization. By incorporating a robust |
116 | bcg | Banking-on-Generative-AI.pdf | WHITE PAPER
Banking on Generative
AI: Maximizing the
Financial Services
Opportunity
September 2023
By Jeanne Bickford, Rafal Cegiela, Julian King, Kevin Lucas, Neil Pardasani, Ella Rabener,
Benjamin Rehberg, Stiene Riemer, Michael Strauss, Jon Sugihara, and Michael Widowitz
Banking on Generative AI: Maximizing
the Financial Services Opportunity
T
echnological change, like many evolutions, often happens slowly, and then all at once.
Neural networks were conceived in the 1940s and natural language processing algo-
rithms came 20 years later, when the first chatbot, Eliza, was created. There followed
nearly 60 years of gradual development, until massive computing power helped create the
models that have attracted global headlines over recent months.
In the financial industry, experimentation with natural language processing has also pro-
gressed in steps, with banks incrementally adding functionality such as chatbots and auto-
mated document processing. Fast forward to today. Massive computing power has helped
accelerate experimentation and create generative AI (GenAI) models that are set to be
game-changing for financial services.
GenAI is a catch-all term for a range of models, predominant among which are foundation
models (FMs) and their large language models (LLMs) subset. (See “What Is Generative AI?”)
WHAT IS GENERATIVE AI?
Generative AI (GenAI) is a set of algorithms, capable of generating seemingly new, realistic content from unstructured
inputs such as text, images, or audio.
The term GenAI encompasses both foundation models and large language models:
Foundation models are pretrained with large datasets and massive compute power, so they are ready to be used without
additional training. They can be applied to many tasks (unlike traditional AI), including generating text or graphics, predict-
ing, or classifying.
Large language models are a subset of FMs and can ingest and produce text. The terms LLM and FM are not interchange-
able.
1 BANKING ON GENERATIVE AI: MAXIMIZING THE FINANCIAL SERVICES OPPORTUNITY
The promise of FMs is that they present shortcuts to resolving complex challenges. Their
potential puts them on a different plane to other technologies that have been labelled
game-changing over the past decade. However, they also bring risks, including leakage of
confidential data, transmission of bias, and potential for low-quality outputs.
In the banking industry, potential FM use cases range from replacing mundane manual
processes to revolutionizing digital customer interactions, creating highly personalized mar-
keting content, and supporting investment decisions. If applied effectively, we expect these
kinds of applications will boost cost effectiveness on a double-digit scale while significantly
improving customer outcomes. Based on our experience working with clients, there is also
potential to reduce back- and front-office process costs by 20-30% in the next two to three
years and 50% or more in the following three years—freeing up staff to engage in more
valuable and less repetitive activities. (See Exhibit 1.)
Exhibit 1 - Gen AI adoption is picking-up rapidly
Now (2023) Very soon (2025) Soon (2030)
• Bionic customer support • AI-powered conversational • AI-powered conversational
Facing • Mass production of content for channels for mass customer sales channels for premium and
customers hyper-personalization & service commercial lines
• Fact search, question answering
• AI assistant for RM productivity • Natural language coding • Self-coding systems
• Code generation/Testing • Intelligent back-office process • Human augmentation and semi-
Employee • Office productivity automation across repetitive autonomous solutions across all
efficiency
• Knowledge management processes functions
• Generating innovative products • AI-driven expert capacity
augmentation
Source: BCG.
BOSTON CONSULTING GROUP 2
Banks Are Investing in New Use Cases
Amid a growing competition, the largest banks are investing heavily in FMs, leveraging exter-
nal models and open-source offerings, such as Langchain—an emerging standard compo-
nent also used by Google. Use cases range from more generic processes such as code writing
and call center support to those specific to the financial services industry, such as financial
analyses or delivering financial advice. (See Exhibit 2.)
Leading IT solution vendors, including Microsoft, Salesforce, Pega, and ServiceNow, have
created FM roadmaps and are offering new functionalities and products leveraging their own
or third-party FMs. In parallel cloud vendors are unveiling environments for model fine-tun-
ing and developing functionalities that will facilitate use cases. Cloud vendors are offering
both proprietary FMs and FMs obtained through partnerships. Microsoft Azure is teaming up
with OpenAI, Google Cloud Platform has unveiled its proprietary PaLM2, and Amazon Web
Services offers both startup FMs (AI21 Labs, Anthropic, Stability AI) and its own Titan FM.
The power of FMs is centered on four key areas, each of which has applications in more than
one area of banking:
• Summarization (including answering questions, fact finding) refers to the ability to
ingest unstructured text and summarize it to the required level of detail. This may be
used to boost customer service efficiency, identify customer needs and preferences from
documents such as emails, create risk profiles from uploaded files, or document software
source code.
• Content generation is the ability to elaborate on a given topic or create media content
such as images or videos given succinct inputs. This can be used in customer communica-
tions or to generate software source code, among many other examples.
Exhibit 2 - Leading FI players are announcing GenAI use cases every day
Goldman Sachs J.P.Morgan ANZ Commonwealth Bank
Code Writing and Testing Financial Advisory Code Testing Call Center Support
Experimenting with Gen AI Working on a ChatGPT like tool Created a team to explore Gen At the bank’s call centers, a
technologies to assist its which aims to revolutionize AI to augment its code testing Gen AI model is already being
developers in autonomously investment decisions by capabilities as well as to explore used to help staff answer
creating and testing code. In providing advanced AI-powered deeper potential use cases. The complex customer questions by
some cases, developers have assistance in analyzing and bank's 4000 software engineers interrogating 4500 documents
been able to write as much as recommending financial see opportunities to use Gen AI on the bank’s policies in real
40% of their code automatically securities such as stocks, to improve efficiency, reliability time. It’s is making its apps
using Gen AI1 bonds, commodities, and and performance of its code3 smarter, including tailoring new
alternatives2 offers to 7.7 million users4
American Express Deutsche Bank Bloomberg Wells Fargo
Predicting customer behavior Operational Efficiency Financial Analyses Synthetic Data Generation
Aims to predict how customers Testing Google's Gen AI and Blooomberg created a Partnered with synthetic
are going to perform over time, LLMs to provide new insights to dedicated BloombergGPT data-generating platform Hazy,
enabling better financial financial analysts, driving model that is trained for to create a self-service model
planning and decision-making. operational efficiencies and analyzing financial information for generating and using
AmEx exploring ways LLMs execution velocity. This will and data to assist with risk datasets for Wells Fargo data
could be used to analyze the empower employees by assessments, judgning financial scientists The intelligent use
feedback and inquiries increasing productivity while sentyment, and can also be cases it has in mind for
customers provide through helping safeguard customer applied for automating synthetic data include fraud
customer service portals, as data privacy, data integrity, and accounting and auditing tasks7 detection using machine
well as on social media5 system security6 learning model8
Source: BCG.
3 BANKING ON GENERATIVE AI: MAXIMIZING THE FINANCIAL SERVICES OPPORTUNITY
• Conversation is the ability to provide human-like answers to customer prompts while
considering the context and flow.
• Data generation (also provided by “small” and non-text generative AI models) is the
ability to generate complex sets of data with specific characteristics. This is intended
predominantly for test cases to compensate for small numbers of empirical observations,
ranging from fabricated fingerprints to biometric identification, or realistic dialogues with
customers to test chatbots.
These capabilities can be applied along the banking value chain, accelerating development
of (and enriching) sales and marketing communications (such as in call centers), increasing
sales and service efficiency and availability, and lifting software development productivity.
(See Exhibit 3.) Still, for many financial institutions, the first step in using FMs is to support
employees, for example in finding the right answer to an incoming query. These use cases
are low-hanging fruit—which we estimate, based on client engagements, can deliver 10–20%
effectiveness gains.
Exhibit 3 - Generative AI capabilities can be applied across banking value chain
Non-exhaustive
VII. Supporting
I. Marketing II. Distribution III. Product IV. Financial VI. Risk &
V. Servicing corporate
& Sales & Onboarding development Advice compliance
functions
Generating
Copywriting RM productivity: Generate content Composing Compliance IT Code
creative/innovative
& creating visual preparing for for financial personalized monitoring and documenting/
products &
collaterals meetings education emails from RMs report generation generation/ review
features
Mass production
Credit approval: Composing Intelligent Data privacy &
of content for Contract & term IT: Test case
support/ personalized document compliance
hyper- generation generation
automation emails processing checks
personalization
Loan & other Detecting trends HR: Copywriting
Customer: Configuring/ Credit review Fraud detection
products and scenario/ recruiting/
Product search, coding products in support/ with synthetic
application portfolio employer
fact search systems automation transactional data
assistance optimization branding content
Chatbots/ RM productivity:
Individualized Virtual assistant/
voicebots for lead rading gist of
contract & term service chat/ Memo writing
warming & memos, fin.
generation voicebots
conversion reports,
interacting with
Sales trainings analytics
Strategy:
with simulated
Competitor
client
analytics
conversations
Data
HR: Screening
augmentation for
CVs
model training
IT Support:
Knowledge base
search
IT: Test data
generation
Content generation Synthetizing, question answering Conversational interfaces Data generation
Source: BCG.
BOSTON CONSULTING GROUP 4
The next obvious step will be to directly read or listen to customer communications and
recommend answers, while still relying on employees to lead the conversation and make
final checks. These are more challenging tasks, requiring solutions that can get close to
creating customer-ready outputs, and potentially leading to 20–50% productivity improve-
ments.
The longer-term vision would be for FM-based agents to take responsibility for tasks end-to-
end. This would include back and forth dialogues with customers—within clear guardrails
and topic areas—and executing operations directly in internal IT systems. (See Exhibit 4.) In
this context, there would be significant upsides in redeploying full-time employees to cli-
ent-facing and higher-value work.
Exhibit 4 - Use-cases with increasing responsibility entrusted to FM-based
solutions
Channel type Use case Scope work delegated to GenAI Implementation
complexity
Read & Retrieve Make Formulate Respond &
understand knowledge decision response execute service
Reactive employee
assistant
One-way Pro-active employee
channels assistant
• E-mail
• Mail Client Service Bot
Listen & Clarifying Retrieve Make Formulate Finalizing Confirm
understand conversation knowledge decision response conversation & execute
Reactive employee
assistant
Interactive Pro-active employee
channels assistant
• Call-center
• Chat Client Service Bot
Source: BCG.
5 BANKING ON GENERATIVE AI: MAXIMIZING THE FINANCIAL SERVICES OPPORTUNITY
Risks Must Be Managed
The flipside of FMs’ unique capabilities is the potential for new or exacerbated risks. There-
fore, banks must take care in their use, including protecting themselves against malicious
actors. Failures to do so will expose them to both potential regulatory sanction and reputa-
tional damage.
Risk that may play out in the short term include the emergence of shadow AI, or applications
that are not overseen or managed by the organization, leading to potential compliance
issues. Another short-term risk is leakage of confidential data, either through prompt engi-
neering or through vendors using prompts for model training. To protect stakeholders, sensi-
tive data can be encrypted, anonymized, or protected through permissions processes, but
leakage can still occur through ungoverned usage. (See Exhibit 5.)
Looking ahead, a risk especially relevant to the financial industry relates to the potential use
of FMs for fraud. The models may, for example, help criminal actors assume identities, forge
checks, impersonate customers through voice imitation, or even mimic managers to per-
suade employees to bypass security measures (social engineering). There are also potential
legal pitfalls, including infringements of intellectual property rights, for example when gener-
ating graphics that resemble copyrighted work.
Exhibit 5 - Gen AI amplifies existing AI risks and introduces new ones –
which can be resolved over time
Source: BCG.
BOSTON CONSULTING GROUP 6
noitpircseD
Short term Medium-term Long-term
laitnetoP
noituloseR
1 2 3 4 5 6 7 8
Shadow Sensitive Enhanced Uncertain Copyright Biased Environmental Capability
AI data leaks Fraud quality challenges outputs Harm Overhang
More widespread Sensitive data can Constantly Dynamically Unclear data Generative AI SOTA achievable Latent capabilities
and potentially be evolving changing provenance models trained through larger and how they
unconventional/ extracted from capabilities to information from publicly on real world models with might be
non-traditional models via deceive and ecosystem makes scraped datasets data can carry more complex leveraged
uses of AI prompt detect with it difficult to and quality over bias to architectures with depends on the
invisible to engineering or higher stakes over mitigate this issues outputs in hard higher interacting
top-down more sophisticated time from more issue without to detect ways computational ecosystem and
governance abuse/attacks pervasiveness extensive human needs development of
approaches intervention new systems
Clear guidelines Sensitive data Better detection Fact-checking Legal Reinforcement Research work in Better
captured in should be methods with services and precedents will Learning with TinyML with an governance
enforceable anonymized verifiable claims problematic be set as courts Human emphasis on mechanisms and
policies and/or encrypted on performance information figure out IP and Feedback (RLHF) edge-device understanding of
accompanied by and reduced false fingerprinting copyright issues can catch some deployment, how capabilities
cultural change Chained models negatives and can provide which will shape biases but not faster inference, emerge
and top-down can be used to false positives stop-gap response and all. Tech. and cost +
communication block leakage of measures to stem development advancement environmental
sensitive data issues approaches likely needed for considerations
comprehensive
solve
Spontaneous or underinformed use of FM-based functionalities may lead to unexpected
results, including generation of so-called hallucinations. These are responses not justified by
training data, which are often the result of the system misunderstanding the question. Simi-
lar risks could be manifested through use case design flaws, leading to uncertain or volatile
outputs. Finally, the model may produce biased opinions, reflecting similar biases in training
data. Potential antidotes include deeper fact-checking and information finger printing.
On a strategic horizon, potential risks relate to the evolution of business ecosystems, where
business models and functionalities based on browsing the internet may cease, with pages
replaced by conversational interfaces or on-the-fly generated content. Stakeholders will need
to keep a close eye on how capabilities develop and put in place governance frameworks that
are sufficiently flexible to keep pace.
Key Design Variables
Not all FMs are created equal, and financial institutions will need to make a range of selec-
tions depending on the intended use case, hosting preference, memory configuration, and
other factors.
Initial applications often retrofit FMs into existing roles, adding them to tasks like co-author-
ing marketing messages, coding chatbot flows, or converting text to graphics. This approach
allows users to do the same things but faster. Looking ahead, there is a strong argument for
setting the bar higher. This would mean adopting a clean-sheet approach that puts the tech-
nology at the heart of the design process Potential examples include:
• Mass produce marketing content based on product characteristics and personas in the
customer base, instead of manually describing each picture’s details.
• A chatbot delivering human-like interaction, able to handle multithreaded, nonlinear
conversations and accommodate new services without the need to code and maintain
detailed conversation flows.
• Recommendations for lending and investment decisions based on a blend of structured
and unstructured data, including detailed justifications for internal and external purposes.
Decisions regarding hosting of models should be aligned with the type of model employed.
(See Exhibit 6.) General-purpose and specialized models (ChatGPT or OpenAI’s Code DaVin-
ci for coding) are usually adopted as a cloud-based service and accessed via secure API,
under the guarantee that submitted “prompts” will not be used to train the model, cutting
the risk of data leakage. For dedicated adaptors, which facilitate customization, or fine-tuned
models, the range of options includes training and running by a vendor or internal data
science and IT team (often within a cloud tenancy).
7 BANKING ON GENERATIVE AI: MAXIMIZING THE FINANCIAL SERVICES OPPORTUNITY
Exhibit 6 - Multiple design choices need to be taken for each use-case
GenAI solution layers Design options Tech components Non-exhaustive examples
Source: BCG.
Other design choices that will shape the models’ impact include:
• Prompting logic: How will questions to the FM be prepared? Options include directly
passing user requests, building an autonomous agent, or using tools such as internet
search or mathematical calculators.
• Short-term memory: How will the “conversation notepad” be organized? Will it be a
“raw” history of conversation or structured to perform a specific task, for example a list of
outstanding customer requests.
• Long-term memory: How will the most relevant knowledge, such as frequently-asked
question documents or email exchange history, be stored and retrieved to provide data for
the FM?
BOSTON CONSULTING GROUP 8
kcatS
hceT
&
ngised
lacigoL
ataD
&
ledoM
• Accessing CRM Customer 360, executing
operations in Core Banking System or Card
Chat & voice
Environment Executing actions
interfaces/ Data access System
integration in banking systems (RPA, APIs)
channels
• Technologies: Specific to bank's IT
architecture
• Multi-step: Product-emotions-pictures
Prompting Multi-step Agent- • Agent based, eg. Intent-resolution-execution
Direct Templated Algorithmic
logic chains based • Technologies: LangChain, LMQL, Microsoft
Guidance
• Buffers & structures with e.g.: task list of chat-
Short-term Conversation Purpose-structured None bot, working summaries of document parts
memory history memory
• Technologies: LangChain, Meta FAISS
• Vector stores containing e.g.: product
Long-term Similarity-based Search-engine
None information, processes, regulations, FAQs
memory retrieval based
• Technologies: Chroma, Llamaindex, Pinecone
• General: ChatGPT, Google PaLM, A21, Aleph Alpha
• Specialized: Google MedPaLM, Code DaVinci
LLM General- With Fine- Custom- Custom-
Specialized • Custom: BloombergGPT
tailoring purpose adaptor tuned trained built
• Model tuning: Google Model Garden, AWS
Bedrock
• Proprietary models: ChatGPT3, Google PaLM2
Proprietary Opensource • Opensource models: Vicuna, ChatGPT2
LLM sourcing
• Model hosting: Google Model Garden, AWS
& hosting
Embedded Public Managed Self managed Bedrock
• Model repositories: HuggingFace
No PII or banking information Sensitive data & systems Chatbot logic and short-term memory
• Environmental integration: How will the solution be integrated with communication
channels, data, and operational or banking systems? This “classic” IT architecture problem
can easily become a bigger bottleneck on the way to adoption than FM-related challenges.
Four Pillars for Success
While FMs can create game-changing benefits, decision makers at financial institutions
must take care in rolling out use cases and assessing impacts. There are both risks and
opportunities. However, through detailed planning, many of the potential risks associated
with the technology can be mitigated. In Exhibit 7, we identify four pillars of successful ap-
proaches:
• Potential: Where is there most potential to use FMs in the context of the bank’s AI ma-
turity? How do you build off existing data foundations? Which use cases will lead to most
differentiation? And how do you create strategic advantage? Banks that have a clear vision
to start with will be best placed to execute effectively.
• Risk management: What are the risks and how should they be managed? How can you
protect and grow the business by making ethical choices that are aligned with your pur-
pose and values? Decision makers must strike a balance between managing potential
exposures and achieving benefits.
• Foundations: Establishing foundations means acquiring understanding. The aim should
be to go beyond simple retrofitting and consider the skills, ways of working, and tools that
will enable a transition to a new operating model. Among other things, this will likely
mean creation and test cycles for marketing shortened from weeks to hours, and opera-
tions employees predominantly managing “AI assistants” instead of directly performing
operational tasks. As in other technology decisions, the build/buy/partner dilemma will
apply, and it will be incumbent on individual businesses to make choices that reflect their
capabilities and direction of travel, achieving upside while avoiding vendor locks.
Exhibit 7 - Leaders must make choices across four key pillars
Potential Risk Foundation People
Which use cases will differentiate How can the company capture the What are the foundational How to adapt org structures and
your organization? benefits of AI while managing capabilities on data infrastructure prepare employees for deployment?
downside risks? and governance for building AI
solutions/industrialize?
Discover your strategic advantage Protect and grow your business by Develop foundational capabilities Prepare your workforce with
through experimentation deploying AI that is ethical and to enable AI solutions strategic workforce planning and
aligned with your purpose and transforming op models
values
Deep dive in next section Deep dive in section 03 How to organize and implement
Source: BCG.
9 BANKING ON GENERATIVE AI: MAXIMIZING THE FINANCIAL SERVICES OPPORTUNITY
• People: Alongside technology choices, management teams must make capability-focused
decisions. Budgets will need to be assigned to training, technology, and risk management.
There may be a requirement to overcome resistance in some quarters, amid concern over
the impact on roles and job security, while operating models and governance frameworks
may need to be adjusted.
FMs offer banks a significant opportunity to boost workforce effectiveness and automate
numerous processes, while creating a more streamlined and personalized customer expe-
rience. FMs can boost the quality and availability of customer-facing services, refine commu-
nications, and improve risk management and compliance.
Large banks are already rolling out use cases in areas including customer services and soft-
ware engineering. However, these uses case are just the beginning. As the technology contin-
ues to evolve, banks must take a more holistic approach, focusing on the technical and
environmental factors that will dictate model choice and model management. Not least,
there is a strategic task to drill down into the potential of FMs, consider risks, and the lay the
foundations that will promote security and accelerate the journey to scale.
1. Ryan Browne, “Goldman Sachs is using ChatGPT-style A.I. in house to assist developers with writing code,” CNBC, March 22, 2023.
2. Hugh Son, “JPMorgan is developing a ChatGPT-like A.I. service that gives investment advice,” CNBC, May 25, 2023.
3. Tim Hogarth, chief technology officer, ANZ, “How generative artificial intelligence can make engineers more efficient.” bluenotes, May 22, 2023.
4. James Eyers, “CBA goes all in on generative AI,” Australian Financial Review, May 24, 2023.
5. “AI in FinTech: 7 use cases market leaders pursue,’ 8allocate, August 2, 2023.
6. Shritama Saha, “How Deutsche Bank is riding the generative AI wave,” AIM, August 10, 2023
7. “Introducing BloombergGPT, Bloomberg’s 50-billion parameter large language model, purpose-built from scratch for finance,” Bloomberg, March
30, 2023.
8. “Synthetic data and the Wells Fargo-Hazy relationship,” VentureBeat, March 28, 2022.
BOSTON CONSULTING GROUP 10
About the Authors
Jeanne Bickford is a Managing Director and Senior Partner in BCG’s New York office. You
may contact her by email at [email protected].
Rafal Cegiela is a Principal, Data Science in BCG’s Warsaw office. You may contact him by
email at [email protected].
Julian King is a Managing Director and Partner in BCG’s Sydney office. You may contact him
by email at [email protected].
Kevin Lucas is a Managing Director and Partner in the BCG X Sydney office. You may
contact him by email at [email protected].
Neil Pardasani is a Managing Director and Senior Partner in BCG’s Los Angeles office. You
may contact him by email at [email protected].
Ella Rabener is a Managing Director and Partner in BCG’s Berlin office. You may contact
her by email at [email protected].
Benjamin Rehberg is a Managing Director and Senior Partner in BCG’s New York office. You
may contact him by email at [email protected].
Stiene Riemer is a Managing Director and Partner in BCG’s Munich office. You may contact
her by email at [email protected].
Michael Strauss is a Managing Director and Senior Partner in BCG’s Cologne office. You
may contact him by email at [email protected].
Jon Sugihara is a Managing Director and Partner in the BCG X Singapore office. You may
contact him by email at [email protected].
Michael Widowitz is a Managing Director and Partner in BCG’s Vienna office. You may
contact him by email at [email protected].
For Further Contact
If you would like to discuss this report, please contact the authors.
11 BANKING ON GENERATIVE AI: MAXIMIZING THE FINANCIAL SERVICES OPPORTUNITY
Boston Consulting Group partners with leaders in business For information or permission to reprint, please contact
and society to tackle their most important challenges and BCG at [email protected]. To find the latest BCG con-
capture their greatest opportunities. BCG was the pioneer tent and register to receive e-alerts on this topic or others,
in business strategy when it was founded in 1963. Today, please visit bcg.com. Follow Boston Consulting Group on
we work closely with clients to embrace a transformational Facebook and Twitter.
approach aimed at benefiting all stakeholders—empowering
organizations to grow, build sustainable competitive © Boston Consulting Group 2023. All rights reserved. 8/23
advantage, and drive positive societal impact.
Our diverse, global teams bring deep industry and functional
expertise and a range of perspectives that question the
status quo and spark change. BCG delivers solutions
through leading-edge management consulting, technology
and design, and corporate and digital ventures. We work
in a uniquely collaborative model across the firm and
throughout all levels of the client organization, fueled by
the goal of helping our clients thrive and enabling them
to make the world a better place.
bcg.com |
117 | bcg | BCG-Executive-Perspectives-AI-and-GenAI-in-Customer-Engagement-EP8-4Dec2024.pdf | Executive
Perspectives
Transformation through
AI and GenAI
Customer Engagement
December 2024
Introduction In this BCG
Executive Perspective,
As part of our ongoing series of C-suite conversations on AI, we are sharing our most recent
we show you how to
learning in a series designed to help navigate the rapidly changing world of AI. After working
with over 1,000 clients in the past year, we've found that AI is at an inflection point: in 2024,
leverage AI to transform
the focus is on turning AI's potential into real profit.
and create value in
In this edition, we discuss the future of customer engagement and the role AI will play in
customer engagement
turbocharging the way companies interact with customers and generate ideas.We address
key questions including:
• How can AI help ideate new products/services?
• How can AI use data to improve product fit and gauge customer reception?
• How can AI turbocharge personalized customer communication?
• How can AI transform the way we communicate with customers?
• How can AI augment the capabilities of existing teams?
.d
e
v
We identify company archetypes poised for maximum growth through AI-powered
re
se
customer engagement…
r
sth
g
ir
llA
Consumer-focused Large enterprises Organizations Customer service- .p
u
o
companies with high with complex seeking innovation oriented companies rG
g
n
customer interaction sales processes acceleration itlu
s
n
o
C
…yielding the quickest path to value and first-mover advantage n
o
ts
o
B
y
b
This document is a guide for CEOs, customer engagement leaders, and product 4 2
0
2
development teams to cut through the hype around AI in customer engagement ©
th
g
and understand what creates value now and in the future. iry
p
1 o
C
Executive summary | Leveraging GenAI to enhance customer engagement
The rapid advancement of generative AI (GenAI) is revolutionizing how companies engage with customers,
reshaping traditional insights, marketing, sales, and service functions into highly personalized, efficient, and
innovative processes. The transformative potential of GenAI centers on three critical levers:
AI reinvents
• Market mirror – virtual simulation of customer feedback (i.e., what you offer): GenAI empowers
customer
businesses to enhance value propositions through customer intelligence and synthetic persona-driven insights
engagement
• Creativity at scale (i.e., what you communicate): By enabling hyper-personalization at scale, GenAI allows
in three ways
companies to craft tailored content across all channels, significantly increasing engagement and relevance
• Conversational capability (i.e., how you engage): GenAI-driven conversational agents and virtual sellers are
transforming customer interactions, providing seamless, personalized, and efficient experiences
.d
e
The human To fully seize these opportunities, organizations must transform people, processes, and technology. This means v re
se
element is key creating AI-centric roles, upskilling teams, and adopting agile structures. Empowering people is crucial to
r
sth
g
to succeed unlocking AI's full potential
ir
llA
.p
u
o
rG
g
n
itlu
s
n
Transform o
This deck provides a strategic roadmap for implementing GenAI, guiding organizations from initial adoption to fully C
n
o
to lead in integrated (3-18+ months), AI-powered customer engagement. By embracing these innovations, businesses can ts o
B
y
the AI era achieve unprecedented levels of efficiency, innovation, and customer satisfaction, to be leaders in the digital age b
4
2
0
2
©
th
g
iry
p
2 o
C
AI is revolutionizing business, fundamentally reinventing
how companies engage, innovate, and deliver experiences
GenAI is already driving …with a large potential to transform the way companies engage with
near-term enhancement… customers and organize around customer intelligence
Faster and better
Always-on
Select examples innovation grounded
customer intelligence
on richer customer
synthesis engine
insights
•Mass qualitative interviews at scale
•Real-time segmentation and trends
Robust data management .d
Enhanced-customer More productive e v
•A daI- ta as s si es tt sed synthesis and queries of expe vr ii ee wn oc ne cw uit sh
to
3 m60 e°
r
an ad
n
a dn ina tl ey rt aic cs
t
wto
it
o hr nga en wi ze m gea nr eke rati tn eg
d
w cit oh
n
a teu nto t-
re
se r
sth
g
model of information flow ir
llA
.p
u
o
rG
•Interrogation of synthetic panels g
n
itlu
s
n
o
C
n
•Automation of survey drafts Hyper-personalization Iterative feedback loop o
ts
o
of offers/services to on concepts with B
y
b
each customer synthetic customers 4 2
0
2
©
th
g
iry
p
Source: BCG experience 3 o
C
Customer intelligence drives competitive advantage,
yet many fail to turn it into better results
Many types of customer ...informing many ...with
intelligence... customer-centric decisions... proven value
Primary research (quantitative, Revenue
~10-20%
qualitative) growth
First-party data Differentiated Optimized go-to- .d
e
v
value proposition market activation re
Social listening
~15-25%
C
f or
poo
em
rs
a
o
tt
ip
o
ts
nim
sa aiv
z
nei ddn ag cts
ivation
se
r
sth
g ir
llA
.p
u
o
Market data
rG
g
n
itlu
s
n
o
Etc. ~20-40%
Brand C
n
o
ts
Improved customer White space advocacy o B
y
b
experience identification 4 2
0
2
©
th
g
iry
p
Source: BCG experience 4 o
C
What if AI-driven customer intelligence could help
you answer key business questions…
Brand Product/offer …yielding faster, better,
messaging development cheaper results?
How can we position and message What products, services, and
our brand to resonate effectively experiences should we develop
Higher
with our priority demand spaces? to meet our customers' needs?
Innovation
Pricing Go-to-market
quality
Quality
and value execution
How can we align different pricing, What strategies should we use
promotion levels, and service to promote our offerings via 3-5X
.d
e
offerings with our consumer targets? strategic distribution, advertising, v re
and sponsorships? Faster insights-to-
se
r sth
Speed
impact process g ir
llA
.p
Consumer Investment u o
rG
g
engagement decisions
30-40%
n itlu
s
n
o
How can we optimize CRM1, Which partnerships should C
n
o
influencer and digital marketing, we develop to deliver both Cost saved on
ts
o B
y
b
and customer experience to functional and emotional value Cost qualitative research 4 2
0
2
maximize consumer engagement? to our customers? ©
th
g
iry
p
1. Customer Relationship Management. Source: BCG experience 5 o
C
GenAI's transformative power is redefining
insights-driven customer engagement in three ways
What you offer What you communicate How you engage
Market mirror Creativity at scale Conversational capability
Informed by customer insights, BCG's Strategic planning ensures targeted, Engagement strategies optimized with
GenAI platform "market mirror" creative, and scalable hyper- GenAI-driven direct and human-led
drives enhanced value proposition personalization interactions
and innovation
..dd
ee
vv
rree
ssee
rr
sstthh
More compelling value proposition, precise 4x faster content generation and 5x faster 2x customer acquisition at sustained CPA1, gg iirr
llllAA
targeting of customers; increased speed and copywriting, enhancing communication with improved customer satisfaction and ..pp
uu
efficiency of innovation and activation efficiency; decreased cost of acquisition LTV2 (brand loyalty); decreased cost to serve oo rrGG
gg
nn
iittlluu
Example: Strategic insights enable a consumer Example: Strategic GenAI integration allows a Example: A software company implements a ss nn
oo
CC
goods company to turbocharge innovation for biopharma company to enable "always-on” "GenAI sales assistant" to support B2B sales nn
oo
ttss
impact, reaching 10x concepts in 10x shorter time content generation throughout marketing value teams, enhancing lead quality and allowing reps oo BB
yy
bb
chain; delivers more content in less time to focus on relationship building, which improves 44
22
00
customer satisfaction and loyalty 22
©©
tthh
gg
iirryy
pp
1. Cost per Acquisition; 2. Lifetime Value. 6 oo
CC
Market mirror | A comprehensive solution is essential
to harness GenAI for revolutionizing customer engagement…
Core advantages
Ongoing data capture, deep synthesis across diverse data sources, and interactive
Deep “always-on" insights
insights repository (e.g., proprietary business data, market, primary customer insights)
Segmented customer demand space "synthetic twins" trained on key drivers of choice,
Synthetic twins
and a panel of business expert stakeholders to generate and react to ideas
.d
e
Forward-looking market simulation to test, refine, and prioritize ideas of value v re
Continuous market testing se
proposition (e.g., GTM1, innovation, pricing) r sth
g
ir
llA
.p
u
o
Dynamic dashboard for Dynamic interface generating tailored outputs based on user prompt rG
g
n
strategic scenario planning (e.g., messaging, reports, marketing copy and visuals, scenario planning and war gaming)
itlu
s
n
o
C
n
o
ts
o
B
Democratized access to insights, customer-centric value proposition, y
b
360° customer centricity 4
2
innovation, and activation 0 2
©
th
g
iry
p
1. Go to Market. 7 o
C
Market mirror | … leveraging advanced, actionable insights
and dynamic innovation to differentiate and guide strategy
Illustrative
Key
Insight retrieval and synthesis Concept creation and value prop Testing and validation
enablement
Augmented Answer Agentic Product Image Concept and Customer
search refinement concept gen. user query feedback
AI
RAG1
.d
GenAI e v
AI AI re
se
platform
Pre-processing of data through Codifying data inputs into Using data inputs to model
r sth
g
vision pipeline into vector DB guardrails for concept generation synthetic consumers
ir
llA
.p
u
o
rG
Demand Centric Growth output2® g n
Market and customer intelligence (3P/1P) BASES tests itlu
(i.e., emotional and functional needs of demand spaces) s n
o
C
Social listening/posts and reviews A/B tests n
Data (e.g.) Brand-specific checklist o
ts
o
Entire compendium of DCG B
Primary customer insights (2P) y
Legal and R&D guidelines b
output 4 2
0
2
©
th
g
1. Retrieval augmented generation & re-ranking & multi-query; 2. BCG's proprietary Demand Centric Growth offer leverages deep customer insights to unlock iry
p
untapped market potential, driving strategic growth that maximizes revenue and market share, positioning companies for significant competitive advantage 8 o
C
Market mirror in consumer goods E2E demand generation case study |
GenAI turbocharges innovation, achieving 10x concepts in 10x shorter time
Illustrative example
End-to-end Rich insights Rapid ideation Immediate
and innovation
demand generation generation synthetic testing
• Understand customer • Integrate feedback, • Generate product • Test with real-time
needs and the market trends, and proprietary concepts
feedback from synthetic
data
• Strategize for growth • Illustrate with tailored twins to fine-tune
opportunities • Validate insights content strategies
through conversational
• Target demand spaces • Ensure brand and legal
refinement
and analyze consumer compliance
influence pathways
..dd
ee
vv
rree
ssee
rr
sstthh
gg
iirr
llllAA
..pp
uu
oo
rrGG
gg
nn
iittlluu
ss
nn
oo
CC
nn
oo
ttss
oo
BB
IMPACT 5x faster from 10x concepts in 10x shorter 3x more breakthrough innovations, yy bb
44
22
00
22
data to insights time, with 100% brand fit lower lead time and cost ©©
tthh
gg
iirryy
pp
9 oo
CC
Creativity at scale | GenAI delivers personalized,
convenient, values-driven experiences that meet evolving customer needs
Hyper- Real-time Cutting-edge Community
personalization ultraconvenience experience and connection
Explosion of customer solicitation Shift to online and new Beyond physical stores, augmented Connection with communities
emphasizing brand communication technologies raising standards experiences building on around passions/similar interests
and offering relevance especially for checkout, delivery, and customer emotions and brands with similar values
aftersales
.d
e
v
re
se
r
sth
g
Of Millennials are willing Of Millennials had an Of customers say the Of Chinese customers'
ir
llA
.p
63% to share personal data 77% active Amazon 79% experience provided 45% purchase decisions are u o
rG
to get personalized Prime membership is as important as the influenced by key g
n
offers and discounts in the US in 2024 product sold opinion leaders/
itlu
s
n
o
influencers C
n
o
ts
o
B
y
These expectations span customer demographics—with a stronger emphasis by Generation Z and Millennials b
4
2
0
2
©
th
g
iry
p
Sources: Ocean Insight customer trend survey; BCG Social Retail Playbook; BCG customer trend survey; Statista; Salesforce; Shoptalk; desktop research; BCG analysis 10 o
C
Creativity at scale in biopharma case study | Always-on GenAI
content generation delivers more assets with significant time saved
Illustrative example
Marketing content development Material approval process
Launch and performance
Marketing value chain Campaign asset detailing Artifact/asset creation MLR1 and revision
monitoring
• Marketing team creates • Write ad copy • Revise marketing content • Launch/execute campaign
outline on campaign
• Source/generate images • Update claims matrices • Monitor performance
Today from • Brief agency on asset
• Review content pre-MLR and • Generate proof for final • Periodically review by
8-10 weeks development requirements
localize as needed post MLR approval expiration
Current avg
2-3 weeks 2-3 weeks 2-3 weeks ~1 week
time spent
• Tailor brief to the customer • GenAl creates images from • Automatically review • Automatically document and
from business plan campaign briefs, claims promotional material perform checks for final
..dd
matrices approval ee vv
• Rapidly develop concept for rree
ssee
creative builds • Automatically perform QC and • Synthesize data rr sstthh
localize content
• Notify need for periodic review
gg iirr
llllAA
Reduce time to final campaign Reduce agency support and time ..pp
Future with GenAl Reduce agency support and high Reduce agency support and rapid uu oo
design and high first-time right to review content due to high first- rrGG
To 3-5 weeks first-time right content generation documentation and localization gg
content generation time right content nn iittlluu
ss
nn
oo
CC
nn
Expected at-scale oo
ttss
value Estimated 20-40% 30-50% 10-25% 0-25% oo BB
yy
bb
Timeline 1-2 months -2-4-day acceleration -3-6-day acceleration -1-3-day acceleration -0-1-day acceleration 44 22
00
22
For new campaign launch ©©
tthh
gg
iirryy
pp
1. Medical, Legal and Regulatory. Source: BCG experience. 11 oo
CC
Conversational capability | GenAI conversational agents simplify
interactions, cut customer effort, and provide quicker, accurate solutions
Human support
Self-help through
conversational agent
Entry Assignment Resolution
Spending hours browsing Being forced to provide Getting transferred to Waiting on sparse
)
m
public sources to get manual input across multiple agents and communication
t
n o
r
e f generic answers many support portals needing to reshare updates on the
r (
r e
u to create a ticket info each time case's status
t
C a
t
s
.d
e
Resolving common issues in Easily opening a case Troubleshooting Receiving real-time, v re
) se
e
r
o
t
(
minutes via personalized with a single GenAI with one agent who on-demand updates
r
sth
g
u
t e AI responses dialogue interface is fully up-to-speed on recent actions,
ir
llA
u F t a t
s
with the issue turnaround times, etc. .p u o rG
g
n
itlu
s
n
o
C
Overall effort saved: Improved speed and accuracy of diagnosis, eliminating repetitive communication and n
o
ts
o
B
steps y
b
4
2
0
2
©
th
g
iry
p
12 o
C
Conversational capability in sales | GenAI transforms sales with
hyper-personalization and AI-driven roles
B2C sales B2B sales
Conversational commerce Sales planning and operations
Grocery Helper conveys hyper-personalized promos or messages in customers' AI agents execute sophisticated planning to optimize coverage, territory design,
family group chats, understanding purchase drivers of customers involved in and goal setting and provide advanced automation for deal desk, approvals,
conversations, boosting basket size, and enhancing buying experience (can also
performance management functions
be relevant for B2B sales in fragmented trade)
Gen AI sales team support/
work as a team with seller
Seller
Intelligent Sales
Assistant
Customer .d
e
v
re
se
r
sth
Solution Engineer g
ir
llA
.p
u
o
rG
g
n
itlu
Sales Coach s n
o
C
Virtual seller n o
ts
o
Engages directly with customers from B
y
b
customer identification through closure 4
2
0
in an entirely AI-powered channel 2
©
Personalized offers promo Personalized messages and th
g
instant promo in store iry
p
13 o
C
Conversational capability in B2B sales | GenAI reinvents how sales teams
in the field engage with customers across different channels
Future of sales deal life cycle, powered by PredAI + GenAI
Channel Discover Learn Try Buy Use Discover Learn
Sales coach Solution engineer Sales assistant Sales assistant
Slack/IM proactively maps buyer identifies lack of identifies high- creates tailored semi-
power map product adoption and propensity cohort automated campaign
sends summary aligned to new product journey with content,
offering emails, and calls
Phone/
Solution enginee Sales assistant Solution engineer
text summarizes call, identifies budget cycle updates quote via r e
t
compiles quote based and process – adds to phone call from customer a
on needs identified account after in-person meeting
l
s
h
.d
e
t n v re
o se
m
o
r
sth
g
w
T
ir
llA
Email Solution engineer Sales assistant .p
creates tailored answers customer u o
rG
adoption plan, aligned inquiries and schedules
g
n
to buyer values meetings itlu
s
n
o
C
n
o
ts
o
B
y
CRM
S pra el pe ps
i
na gs s fois rt Qan Bt
R 1,
S cra el ae ts
e
sa pss ei rs st oa nn at
li zed
S pua sle hs
e
sc noa otc ih
fication via
S sca hl ee ds
u
a les ss i mst ea en tit
n g
S dro al fu tsti co un
s
te on mg i Sn Oee Wr
2
S pra ol ve is
d
ea ss s hi os lt isa tn ict S ofa fele rss ta os cs ri es ata ten t
stage 0
b
4 2 0
2
generates research and relationship app with commercial and creates a draft of and proposal content performance review, opportunities ©
insights for company development plans construct to accelerate content peer and market
th
g
and attendees close date comparison
iry
p
14 o
C
1. Quarterly Business Review; 2. Statement of Work
Path to implementation | Shift the insights function from data
requestor to insights curator for customer-centric growth
The insights curator: The future of insights goes beyond responding to data requests. Instead, it will
involve curating insights from a broad range of sources—structured or unstructured, requested or
not—to drive more holistic, customer-centric decision making
Requested data Unrequested data
A new approach to
d customer insights
e
r Qualitative data Social data
u
t
c
u Communities Ratings and review data Customer insights are not just
r
t .d
s e
n about gathering data on demand v re
U
se
AI and
r
sth
g
human
Today’s focus is on combining
ir
llA
expertise
.p
u
o
human expertise with AI to
rG
d e
r
Survey data Behavioral data
curate and synthesize data from
g n itlu
s
u n
Search data o
t C
c u Behavioral data with opt-in multiple sources, enabling more n o ts
r CRM data o B
t S informed, customer-centric y b
4
Biometric data Open data 2 0 2
decisions ©
th
g
iry
p
15 o
C
Path to implementation | Navigate the GenAI vendor ecosystem
with rigorous assessment to unlock value
GenAI vendor landscape Build vs. partner assessment
Insights to impact
Example use cases Ex. vendors Build in-house if: Partner assessment criteria
value chain
• Automate AI-driven survey writing and quant analysis
Technical capabilities and
• Boost niche segments with synthetic respondents
Primary performance
research: • Synthesize open-ended questions automatically Custom,
e c n quant • Analyze video surveys with AI insights in-house
a
d iut
u
• Interact with data conversationally GenAI Customization and
g co h Gather and platform flexibility
igg
u synthesize
• Moderate AI conversations for qual insights at scale
e t a ro r h customer rP esri em ara cr hy : • Generate summaries, themes, verbatim analysis
t
s
gt
t n intelligence qual • Conduct large-scale online focus groups
n id
iv o r p
;r
oe m
e g a n
a m Secondary
•
•
L Sa uy mer
m
G ae rin zA
e
I
r
o en
se
s ao rc ci ha l
w
l ii ts hte tn hi en mg ato tio cl s
analysis and
• G bae sn isA I
o
u
f
se case is a
Data security and privacy
.d
e v re se
r sth
t a r g e g n research quote ID d yoif ufe r r ce on mtia pt aio nn y for User experience g ir llA
e t n i sa h c E • Synthesize 1P/3P data and digital engagement • T dah te ar e s ea cr ue rs itig y/n Ii Pfi cant .p u o rG
a s
e v
r2 E
d n
S ty en st th inet gic • T exe ps et rc to an dc ve ip ct es with AI-generated customers and
•
l Te ha ek ra eg ae
r
r ei nsk 'ts
Support and maintenance
g n itlu
s n
ea o
s G C Innovate Innovation • Generate product ideas with GenAI e thff ae tc dti ev le iv v ee rn td ho er s (SLAs) C n o ts
B and activate performance o B y
b
expected 4 2
Marketing • Create marketing content and briefs quickly • Cost of working with Cost and pricing model 0 2 ©
a vendor is too high th
g
Note: This is a small sample of the iry
p
Source: Company websites growing GenAI vendor ecosystem, 16 o
C
with new solutions emerging regularly
Path to implementation | Rethink customer-facing team
structures, streamlining into three unified, AI-driven processes
Not exhaustive
Today, organizational functions These functions will be unified with the emergence of
engaging with customers are
AI and organized across three main processes
distinct (e.g., R&D often
disconnected from the
end users) 1 Insights-driven product development • Customer insights
• Customer research
• AI in data analysis for customer insights
• R&D
• AI-enhanced, demand-driven product
• Inside sales • Direct sales • Product development
Sales
• E-commerce • Channel sales development/R&D process • Growth and pricing
• Customer support
Content generation & personalization at scale
Service • Technical support 2 .d
• Field service • AI-driven content creation • Marketing e v re
se
• Personalized marketing strategies • Personalization strategy r
sth
• Content strategy g
Customer
• Account • Customer • Automated and AI-enhanced customer ir
llA
success
management training interactions .p
u
• Onboarding • Renewals o rG
g
n
itlu
s
Customer • Market research 3 Customer interaction management n o C
research • Customer surveys • Sales n o
• Automated and AI-enhanced customer • Customer support ts o
B
interactions • Customer success y b
4
• Product innovation • Concept • CRM 2 0
R&D/ • Chatbots, virtual assistants, and CRM systems 2
product
• Prototype validation ©
th
development • User testing g iry
p
17 o
C
Path to implementation | Design scalable architecture to
support the expanding GenAI ecosystem
Smart business layer (systems of engagement)
GenAI is embedded across all layers,
…
from customer interaction (smart AI copilots Conversational apps AI assistants
business layer) to data analytics (data
layer) and innovation testing (AI layer) AI layer 4
Guardrails
Systems of engagement include AI-
driven tools like intelligent sales 1 Orchestration Ops and
assistants, virtual sellers, and E2E app Model garden Foundation/other small models monitoring
vendors
autonomous agents, as well as n
2 Model platform o
conversational and cognitive apps for y i t
t a
a seamless and highly personalized i r u r g .d
customer journey Data layer c e S e t n
I
e v re se
r
sth
The core transaction and data layers
3 Data products
Operational
g ir
llA
.p
integrate real-time data activation, Repository and storage data u o rG
g
insights, and advanced analytics,
Distribution and integration
services n itlu
s
supported by the GenAI layer for n o
C
n
predictive and AI-driven innovation o
ts
Core transaction layer o
B
y
b
4
2
0
Infra and cloud layer Public cloud Private cloud Specialized hardware (GPU1 & TPU2) 2 ©
th
g
iry
p
1. Graphics Processing Units; 2. Tensor Processing Units. 18 o
C
Path to implementation | Develop practices to manage risks
and ensure responsible AI use
Provide disclosure Provide transparency Protect sensitive data
Disclose use of AI/GenAI to customers, Provide transparency into data usage at time of Be cautious of inadvertently revealing sensitive
including in cases where they are interacting collection, allowing for explicit opt-out information arising from AI-derived insights
with a GenAI agent or receiving AI-generated (e.g., emergent health issue identifiable from
content recent medication purchases)
.d
e
v
re
Limit types of engagement Prevent bias Ensure quality
se
r
sth
g
ir
llA
.p
Explicitly consider the degree of Identify and mitigate demographic bias (e.g., Ensure GenAI systems, especially those that u
o
rG
personalization allowed based on class of gender, age) in personalized messages/services are directly customer-facing, are fully tested g
n
product (e.g., no segment-of-one or products offered for quality and risk (e.g., offensive language,
itlu
s
n
o
personalization for potentially addictive recommending competitor products, offering C
n
o
products or services, no engagement around products at steep discounts, inaccurately ts o
B
y
products related to death of loved one) answering customer questions) b
4
2
0
2
©
th
g
iry
p
19 o
C
Call to action Identify GenAI opportunities to drive customer centricity
• Evaluate where GenAI can enhance value propositions and boost
productivity, including customer insights, sales, and customer service
Begin your GenAI transformation
today—strategize, upskill, and
Develop a strategic GenAI roadmap
innovate for successful customer
• Elevate customer engagement by progressively embedding GenAI
engagement
across all touch points
• Engage senior leadership to set short-term goals and long-term plans
for GenAI integration
Launch cross-functional centers of excellence to drive
productivity gains
• Create cross-functional centers of excellence that bring together
expertise from various departments (e.g., marketing, sales, customer .d
e
v
re
support, tech, R&D) to drive GenAI implementation
se
r
sth
• Institute human-led best practices and support implementation g ir
llA
across functions to manage risk and ensure responsible AI use .p
u
o
rG
g
n
itlu
Invest in skills, technology, and human-led processes s n
o
C
n
• Upskill your workforce to leverage AI effectively o
ts
o
B
• Establish human-led processes, enhanced with AI y b
4
2
0
• Build necessary tech infrastructure to support GenAI applications 2
©
th
g
iry
p
2200 o
C
BCG experts |
Key contacts
for GenAI in
customer
engagement
Karen Lellouche Lara Koslow Ben Eppler
Tordjman Managing Director Managing Director
Managing Director & Senior Partner & Partner
& Senior Partner
.d
e
v
re
se
r
sth
g
ir
llA
.p
u
o
rG
g
n
itlu
s
n
Lauren Taylor Stephen Edison Greg McRoskey Melike Inonu o C
n
o
Managing Director Managing Director Partner & Sr Manager - Customer ts
o
B
& Partner & Partner Associate Director Demand & Innovation y
b
4
2
0
2
©
th
g
iry
p
2211 o
C |
118 | bcg | ai-maturity-matrix-nov-2024.pdf | The AI Maturity Matrix
Which Economies Are Ready for AI?
November 2024
By Christian Schwaerzler, Miguel Carrasco, Christopher Daniel,
Brooke Bollyky, Yoshihisa Niwa, Aparna Bharadwaj, Akram Awad,
Richard Sargeant, Sanjay Nawandhar, and Svetlana Kostikova
Contents
03 Introduction
04 Key Findings
05 The Relationship Between
Exposure and Readiness
10 T he Archetypes of AI Adoption
15 Strategic Next Steps
17 M ethodology
21 About the Authors
Introduction
V
iews vary on how much AI is changing the world economies are gradually adopting AI, but there is a small,
today, but one thing is clear: the technology is on influential group of AI pioneers that take their place as
course to shape the future of economic development. leaders. Their prize is economic advantage, but they are
Business leaders expect large impacts on operations and also poised to shape how humanity will interact with this
value creation in the 3-to-10-year timeframe, and world- powerfully disruptive technology.
wide spending on artificial intelligence will more than
double to $632 billion by 2028.1 The long-term, expansive By focusing on two pivotal aspects, this report offers a
scale of this growth makes AI an economic priority in every unique approach to viewing the global dynamics of AI
region across the globe. adoption. First, we examine each economy’s exposure to
AI-driven disruptions. We define exposure as the potential
This growth also adds urgency to the questions that for AI to impact a sector in an economy negatively or
policymakers face about AI. Is a society able to build an positively. We then assess each economy’s readiness to
AI-skilled workforce in key sectors? How will a government harness AI’s potential for growth and to mitigate potential
set up resilient, modern infrastructure? How does a nation risks. The resulting matrix brings together these factors to
spur enough investment and R&D to stay competitive? present six archetypes of AI economic development and
potential. We offer recommendations tailored to the
BCG’s new AI Maturity Matrix assesses 73 global different groups to guide policymakers—and provide an
economies to answer some of these key questions.2 This interactive dashboard for a more detailed exploration of
matrix provides a broad view of global adoption: most our analysis.
1. “AI Is Showing ‘Very Positive’ Signs of Eventually Boosting GDP and Productivity,” Goldman Sachs website, May 13, 2024; “Worldwide Spending on
Artificial Intelligence Forecast to Reach $632 Billion in 2028, According to a New IDC Spending Guide,” IDC website, August 19, 2024.
2. Details of the selection process are available in the methodology section.
BOSTON CONSULTING GROUP THE AI MATURITY MATRIX: WHICH ECONOMIES ARE READY FOR AI? 3
Key Findings
Out of 73 economies assessed, only Several economies with high Most economies in the study are
five—Canada, Mainland China, AI readiness are just behind the not ready for AI disruption. More
Singapore, the UK, and the US— pace of AI pioneers. While this than 70% score below the halfway
are categorized as AI pioneers. group of AI contenders includes mark in categories like ecosystem
They have reached a high level of established economies, it also participation, skills, and R&D.
readiness by blending elements like features emerging ones like India, Policymakers must act now to adjust
investment and infrastructure, turning Saudi Arabia, and the UAE that are to a world of AI and boost resiliency,
disruption into a competitive edge. using policy and targeted investments productivity, jobs, modernization, and
They are in a unique position to guide to adopt AI on an advanced level. As competitiveness.
the world forward in innovation, these economies strengthen their
talent development, and AI regulation innovation capabilities, they will grow
and ethics. more competitive and influential in
the AI space.
Distribution of Economies Across the Archetypes of AI Adoption
Sources: BCG Center for Public Economics; BCG analysis.
Note: Within each archetype, economies appear in alphabetical order.
BOSTON CONSULTING GROUP THE AI MATURITY MATRIX: WHICH ECONOMIES ARE READY FOR AI? 4
hgiH
ERUSOPXE
woL
AI contenders
Steady contenders
· Australia · Japan
AI practitioners · Austria · Luxembourg
· Belgium · Malaysia
· Denmark · Netherlands
· Estonia · Norway
· Finland · Portugal
Exposed practitioners · France · South Korea AI pioneers
· Bahrain · Greece · Germany · Spain · Canada
· Bulgaria · Hungary · Hong Kong · Sweden · Mainland China
AI emergents · · C Cy zp ecru hs ia · · K Mu aw lta ait · · I Ir se rala en ld · · S Taw ii wtz ae nrland · · S Ui Kngapore
· Italy
· Algeria · Iraq · US
· Angola · Nigeria
· Ecuador · Venezuela Gradual practitioners Rising contenders
· Ethiopia
· Argentina · Morocco · Brazil · Saudi Arabia
· Chile · Oman · India · Türkiye
· Colombia · Pakistan · Indonesia · UAE
· Dominican · Peru · New Zealand · Vietnam
Republic · Philippines · Poland
· Egypt · Romania
· Iran · Qatar
· Kenya · Slovakia
· Latvia · South Africa
· Lithuania · Thailand
· Mexico · Ukraine
Bottom 10% READINESS Top 10%
The Relationship Between Exposure
and Readiness
T
he future of AI is framed by high expectations. Yet A key place for public sector leaders to start is to under-
adoption is already paying off today with efficiency stand their economy’s level of exposure to AI by sector.
gains and return on investment. Businesses that are Exposure can lead to positive or negative impacts; for
scaling AI have boosted their revenues by 2.5 times com- example, in terms of jobs, exposure could lead to displace-
pared to competitors. When scaled across an entire econo- ment or create new employment opportunities throughout
my, such potential gains elevate AI to a pressing area for a sector. However, job displacement is not the only area of
policymaking—both today and in the years ahead. exposure. (See sidebar, “The Dimensions of AI Exposure.”)
BOSTON CONSULTING GROUP THE AI MATURITY MATRIX: WHICH ECONOMIES ARE READY FOR AI? 5
The Dimensions of AI Exposure
We find that exposure appears on many levels.
Productivity. Recent BCG research shows how AI’s ability
to automate tasks and optimize processes helps both
workers and entire businesses. On one level, AI expands
employee capabilities. BCG research found that in our
organization, GenAI-supported consultants performed 20%
better on data science tasks that fell outside their usual
areas of expertise or training. One biopharma firm used
GenAI to shorten its drug discovery process by 25%.3 On a
broader scale, several economies in this study are exposed
to these potential shifts.
However, AI could also disrupt traditional workflows in
sectors reliant on manual processes, such as manufactur-
ing. Small businesses using classic methods will often
compete with larger companies that deploy AI-driven
automation. Such small businesses might struggle to
match the productivity of optimized firms, impacting the
sector’s overall performance.
Uneven Sectoral Impact. Some sectors may lag in AI
adoption, widening the gap between innovative industries
and slower-moving ones. For instance, even as a tech-ori-
ented sector like finance readily adopts AI, agriculture may
be slower to fit the technology into workflows, failing to
result in the overall productivity gains that could help boost
a nation’s economic performance.
Job Evolution. Most observers expect that AI will make
some job categories obsolete. But new jobs that call for
advanced technical skills, including AI specialists and AI
ethics officers, will offset some of the displacement or
create new employment opportunities in sectors that have
long lagged in hiring.
In our methodology, which represents a snapshot of the
current landscape, we gauge exposure scores through four
major sources:
• A BCG survey of business leaders across sectors on
their perceptions of exposure to AI
• The frequency and intensity of AI discussions during
quarterly earnings calls of publicly listed companies
• The number of AI-related job vacancies on LinkedIn
• GenAI-sourced insights on disruption across various
industries
3. BCG client experience.
BOSTON CONSULTING GROUP THE AI MATURITY MATRIX: WHICH ECONOMIES ARE READY FOR AI? 6
Our study includes several findings about sector exposure sectors also produce AI-related goods and services that
to AI: other industries use or sell. In other words, economies with
strong ICT sectors that produce AI technologies can see
Six sectors are most exposed to AI-driven changes. their GDPs grow.
These include information and communication; high-tech
goods; retail; financial services; public services; and motor For example, semiconductors created by an economy’s
vehicles manufacturing, as shown in Exhibit 1. high-tech goods sector—resulting in more powerful, effi-
cient chips—are used in onboard auto electronics for auton-
ICT sectors (such as information and communication and omous driving, enhanced safety features, and improved fuel
high-tech goods) show high exposure because AI can great- efficiency. Homegrown AI disrupts the economy’s automo-
ly transform how work gets done in these areas. Yet these tive sector, making it more innovative and competitive—
sectors are more than just hotbeds for automation. Such and growth soars for both automakers and chip makers.
Exhibit 1 - Exposure to AI: Heatmap of Sectors
SOURCES LEVERAGED TO GAUGE EXPOSURE
Exposure Survey of Publicly listed Job vacancies GenAI-sourced
Sector to AI business leaders companies on LinkedIn insights
Information and communication
High-tech goods
Retail and wholesale
Financial services
Public services
High
exposure
Motor vehicles and parts
Business services
Accommodation and catering
Machinery and equipment
Transport and storage services
Oil and gas, coke, and refined petroleum
Utilities
Pharmaceuticals
Arts, recreation, union, and personal services
Textiles, leather, and clothing
Mining
Metals
Food, beverages, and tobacco Limited
exposure
Other transport equipment
Nonmetallic minerals
Chemical, rubber, plastics
Construction
Other miscellaneous
Agriculture, forestry, and fishery
Furniture manufacturing
Paper and wood products (without furniture)
Sources: BCG Center for Public Economics; BCG analysis.
Note: For more details on sources, see the report’s methodology section.
BOSTON CONSULTING GROUP THE AI MATURITY MATRIX: WHICH ECONOMIES ARE READY FOR AI? 7
Economies with a high share of sectors that are most Readiness by Degrees
exposed to AI are among the world’s most exposed
to disruption. The three most exposed economies in our Assessing readiness helps an economy understand its
study are Luxembourg (with financial services making up strengths and weaknesses as it manages technology risk
almost 30% of GDP); Hong Kong (22% financial services and makes the most of AI.
and 22% business services); and Singapore (18% business
services; 16% retail; 14% financial services).4 Readiness for AI refers to an economy’s ability to effective-
ly implement and integrate AI. This ability can be mea-
Economies with industry sectors that are less suscep- sured across six dimensions that make up BCG’s ASPIRE
tible to AI disruption are less exposed. Such sectors index: Ambition; Skills; Policy and regulation; Investment;
include construction, agriculture, and furniture manufactur- Research and innovation; and Ecosystem. (See Exhibit 2.)
ing; countries include Indonesia (13% agriculture and 11%
construction of GDP); India (17% agriculture and 8% con- This framework offers a comprehensive view on adoption
struction); and Ethiopia (36% agriculture). But these sectors levers for AI. Ambition assesses whether a country has a
can be fertile ground for economic transformation. specific AI strategy and a government entity to oversee it,
while Skills looks at the availability of AI specialists. (For
Along with boosting efficiency, AI can create positive more details, refer to the methodology at the end of the
spillover effects throughout an economy—especially report.) ASPIRE is useful for assessing the full range of
a less exposed economy. AI can spur growth in adjacent advanced, emerging, and developing economies, some of
sectors, helping a country shift the mix of sectors in its GDP. which are quite prepared for AI. It also showcases the
For example, AI-driven agricultural technology could help imbalances that often form when an economy is highly
optimize supply chains with data on crop yields, weather, advanced in some of these six areas and lacking develop-
and market trends. The country’s transportation sector ment in others. (See Exhibit 3.)
would become more efficient and modernized.
Most economies must do more to prepare them-
Ultimately, exposure to the changes brought by AI is selves adequately for AI disruption. The numbers are
inherent in today’s world. It’s inevitable that AI will show stark: Out of 73 economies assessed, only five—categorized
up somewhere in an economy, even to a limited degree, so as AI pioneers—have achieved a high level of readiness.
every country’s economy has at least some exposure to the More than 70% score below the halfway mark in categories
technology. Yet an economy with high exposure isn’t nec- like ecosystem participation, skills, and R&D.
essarily in a bad spot—on the contrary, some of the most
exposed economies are also the most prepared. Pioneers are out in front in skills, R&D, ecosystems,
and investments. In skills, the US and Singapore stand
out with robust AI talent pools, which are crucial for driving
innovation. The US leads in investing, driven by its sophisti-
cated capital markets and the abundance of AI unicorns. In
the R&D race, Mainland China is leading in patents and AI
academic papers.
Everywhere else, innovation and investment must
catch up. The bulk of economies score below the average
in R&D and investment, hindering their ability to foster
startups or deploy homegrown solutions. Some countries
that perform well in ecosystems, including Japan, Germa-
ny, and the UAE, have good telecommunications and AI
infrastructure; they’ve benefited by accessing new technol-
ogy from ecosystem partners. However, other economies
score lower in innovation and ecosystem participation,
leaving them with fewer options to access new solutions.
The ambition to engage AI is high throughout the
world—but countries need more than ambition. Most
economies, including upper-middle countries like the Domi-
nican Republic and lower-middle-income countries like Kenya,
have stated their national strategies or created national AI
ministries and steering committees. Yet societies will only find
positive outcomes if they move beyond planning and take pro-
active, concrete actions, such as forming test beds for R&D.
And for many actors, it will take time before tangible results
from AI emerge. Ambition must be paired with patience.
4. Sector share percentages are calculated from the total sum of all sectors, or GVA (gross value added), which we use as a proxy for GDP in the
report. GDP equals GVA plus taxes minus any subsidies.
BOSTON CONSULTING GROUP THE AI MATURITY MATRIX: WHICH ECONOMIES ARE READY FOR AI? 8
Exhibit 2 - Readiness for AI: ASPIRE Index
A S P I R E
Ambition Skills Policy and Investment Research and Ecosystem
regulation innovation
· Existence of AI · Concentration of · Regulatory quality · Value of AI unicorns · Research papers · Fixed broadband
strategy AI-related specialists · Governance · Mcap of IT-related published on AI internet traffic per
· Existence of · Pool of AI-related effectiveness and tech-related · AI-related patents capita
specialized AI specialists · Governance of data companies/GDP · Top-ranked · Electricity prices
government · Total public · Economic freedom · Value of trade in ICT universities in data · Telecommunication
agency/ministry contributions in index services (per capita) science and AI fields infrastructure index
GitHub by top 1,000 · AI and democratic · Value of trade in ICT · Number of AI · Average download
users values index goods (per capita) startups speed
· Kaggle Grandmasters · VC availability · Online service index
· Number of Python · Funding of AI · Performance of
package downloads companies economy-wide
per 1,000 people · Computer software statistical systems
spending
Sources: BCG Center for Public Economics; BCG analysis.
Exhibit 3 - Readiness for AI: Measuring Economies
Policy and Research and
Economies Total ASPIRE Ambition Skills regulation Investment innovation Ecosystem
Canada
AI Mainland China 68 10 17 8 8 8 19
Singapore
pioneers United Kingdom 0 100 0 10 0 25 0 10 0 15 0 15 0 25
United States
Australia Japan
Finland Netherlands
France South Korea 58 10 14 8 6 4 16
Top 25% Germany Spain
India Sweden 0 100 0 10 0 25 0 10 0 15 0 15 0 25
Ireland Taiwan
Israel UAE
Austria Malaysia
Belgium New Zealand
Brazil Norway
Denmark Poland 47 9 11 7 4 2 14
Top 50% Estonia Portugal
Hong Kong Saudi Arabia 0 100 0 10 0 25 0 10 0 15 0 15 0 25
Indonesia Switzerland
Italy Türkiye
Luxembourg Vietnam
Argentina Malta
Chile Mexico
Colombia Pakistan
Cyprus Peru 38 10 8 6 2 1 11
Czechia Qatar
Top 75%
Egypt Romania
0 100 0 10 0 25 0 10 0 15 0 15 0 25
Greece South Africa
Hungary Thailand
Latvia Ukraine
Lithuania
Bahrain Kuwait
Bulgaria Morocco 31 7 7 5 2 1 9
Dominican Oman
Top 90%
Republic Philippines
0 100 0 10 0 25 0 10 0 15 0 15 0 25
Iran Slovakia
Kenya
Algeria Iraq 20 4 5 3 1 1 6
AI Angola Nigeria
emergents Ecuador Venezuela 0 100 0 10 0 25 0 10 0 15 0 15 0 25
Ethiopia
Minimum for dimension Maximum for dimension
Average
Sources: BCG Center for Public Economics; BCG analysis.
Note: Economies positioned at the borderline between the top AI pioneers and the top 25% range are considered as part of the top 25% group. Due
to rounding, the dimension scores may not sum up to the total score.
BOSTON CONSULTING GROUP THE AI MATURITY MATRIX: WHICH ECONOMIES ARE READY FOR AI? 9
The Archetypes of AI Adoption
T
he combined analysis of AI exposure and Pioneers will want to amplify their strategies to keep up
readiness reveals six distinct adoption groups. their competitive edge. But as competitive as technology
(See Exhibit 4.) evolution can be, countries everywhere should come to-
gether to address the emerging ethical questions around
AI Pioneers. These are the vanguards of AI adoption, AI. Pioneers can participate in these important ethical
building on strong infrastructures and engaging the tech- efforts in several ways. For one, they are authoring the
nology in diverse sectors. All pioneers invest greatly in world’s first AI-specific regulatory codes, which will likely be
R&D, as shown by the many tech companies, startups, models for regulation in other countries. These leaders
and unicorns in each of the five countries. Job sectors and should also convene nations throughout the world in dis-
education systems are full of highly skilled talent. cussions around AI ethics. (See sidebar, “How Singapore
Became an AI Pioneer.”)
AI will make up progressively larger shares of the pioneers’
GDPs over the next several years, as these actors supply
more and more AI technologies, services, skills, and invest-
ment to the world. For example, the US exports software,
platforms, and essential hardware for AI computing, as
well as cloud-based AI services. Mainland China exports
AI-powered consumer electronics, including autonomous
driving platforms. This presence in the global tech supply
chain allows pioneers to set standards and influence the
entire AI landscape.
BOSTON CONSULTING GROUP THE AI MATURITY MATRIX: WHICH ECONOMIES ARE READY FOR AI? 10
BOSTON CONSULTING GROUP THE AI MATURITY MATRIX: WHICH ECONOMIES ARE READY FOR AI? 11
hgiH
ERUSOPXE
woL
Exhibit 4 - Definitions of the Archetypes
AI practitioners AI contenders
Exposed practitioners Steady contenders
Economies with relatively high Economies with relatively
exposure to AI and insufficient high exposure to Al and AI pioneers
levels of readiness sufficient levels of readiness
Economies able to
for its adoption
meet high levels of
exposure with
AI emergents extremely high
Economies with readiness
extremely low
readiness and
different levels of Gradual practitioners Rising contenders
exposure to AI Economies with relatively low Economies with relatively low
exposure to Al and low exposure to AI despite high
readiness for its adoption readiness for its adoption
Bottom 10% READINESS Top 10%
Sources: BCG Center for Public Economics; BCG analysis.
AI contenders have a relatively high level of AI readiness. Rising Contenders. These are mainly economies with lower
These actors are enjoying efficient operations, lower costs, AI exposure due to a relatively higher share of industrial and/
and other benefits of adoption. Going further and acceler- or resource-based mix of sectors. This lower level of exposure
ating AI across sectors will strengthen their positions; if is the main difference between rising contenders and steady
these economies expand their stakes in niche or special- contenders, but governments in this subgroup push for AI
ized markets, they could compete with AI pioneers in such adoption with the same commitment as steady contenders.
areas. We split AI contenders into two archetypes:
An interesting case here is India, which is grouped with
Steady Contenders. These economies have higher shares several high-income economies because of its high level
of highly exposed service sectors, such as financial ser- of readiness.
vices. However, their exposure is balanced by high readi-
ness. This group is mainly dominated by high-income Eu- • The Indian government has launched several AI-focused
ropean economies like Germany, which has high exposure initiatives, such as the National AI Strategy and the
due to its large ICT and advanced manufacturing sectors. creation of centers of excellence in AI, which aim to inte-
Germany’s technological innovation and strong industrial grate AI into key sectors like agriculture and education.
base attract foreign trade and investment. Combined with
its robust AI strategy, the country has established itself as • India is investing heavily in AI education and training
a strong player in global tech markets. programs to build a large, tech-skilled workforce.
A notable country here from outside Europe is Malaysia. The • India has a rapidly growing startup ecosystem, partic-
strong focus of the Malaysian government on AI is evident in ularly in AI-driven fintech, health care, e-commerce,
its National AI Roadmap, tech hubs, and universities offer- edtech, and agritech.
ing AI training. This shows how public sector leadership can
help an emerging economy reach technology maturity and Two other notable examples in this group are Saudi Arabia
competitiveness on par with high-income economies. and Indonesia. Having focused on building AI foundations
since launching the National AI Strategy in 2020, Saudi
Arabia is now emerging as a global center of excellence in
fields of national priority such as Arabic language AI, indus-
trial and energy-related AI, as well as health care and
education. Indonesia, through its National AI Strategy, is
emphasizing education to meet the needs of its growing
population and foster long-term economic growth.
How Singapore Became an AI Pioneer
Despite its small population, Singapore is a notable exam-
ple of AI adoption due to a successful government strategy
on AI—including talent, regulation, innovation, and invest-
ment. The country launched its National AI Strategy in
2019, with an updated version in 2023 (NAIS 2.0), focusing
on integrating AI across multiple sectors. In February 2024,
Singapore announced a five-year plan to invest more than
US$743 million in AI to strengthen its position as a global
business and innovation hub.
Skilling is a key piece of government efforts. The country’s
TechSkills Accelerator program has upskilled more than
230,000 people since 2016. The country’s AI Apprentice-
ship Program (AIAP) trains Singaporean tech workers on
real-world AI projects. Singapore has also moved to attract
talent; the ONE Pass and Tech@SG programs make it
easier for tech companies to hire international experts by
simplifying the visa process.
Singapore has launched specific AI policies and frame-
works. The Model AI Governance Framework guides com-
panies in the ethical use of AI, ensuring transparency and
accountability. The AI Verify Foundation is a global open-
source community to support companies in deploying AI
responsibly and maintaining stakeholder trust.
The country’s five-year national R&D strategy—the
Research, Innovation, and Enterprise (RIE) plan—funds
innovation with US$19 billion, launched in 2020 across
various sectors, including the digital economy. The AI
Singapore program brings together the country’s research
institutions in an ecosystem of innovation.
Singapore also established the Center for Frontier AI
Research (CFAR), which supports AI R&D related to nation-
al priorities.
BBOOSSTTOONN CCOONNSSUULLTTIINNGG GGRROOUUPP TTHHEE AAII MMAATTUURRIITTYY MMAATTRRIIXX:: WWHHIICCHH EECCOONNOOMMIIEESS AARREE RREEAADDYY FFOORR AAII?? 1122
AI pioneers are the vanguards of
AI adoption, building on strong
infrastructures and engaging the
technology in diverse sectors.
AI practitioners make up a diverse group of countries at AI Emergents. These economies are at the early stages
different levels of economic progress. We split AI practi- of AI adoption. They need to build foundational strategies
tioners into two archetypes: and infrastructure to reach the basic levels of AI integration
and competitiveness.
Gradual Practitioners. These are typically upper-middle
and lower-middle-income countries that are adopting AI at These countries lack a national AI strategy or similar holis-
a moderate pace. Their economies include low-tech sectors tic approaches to AI. Skilled workers and investment are
such as tourism, textiles, wood manufacturing, and agri- often scarce, as is activity related to research papers,
culture, where adopting AI is not yet imperative for com- patents, and startups. Nations in this archetype should
panies. However, countries here can explore how AI brings look outward for international investment and sources of
efficiency or new revenue lines to these sectors. This will talent. They should also establish the basics of a govern-
maintain competitiveness and foster growth as the tech- ment-driven technology strategy.
nology becomes more relevant over time.
However, building competitiveness is not out of reach for
Long reliant on its energy resources, Qatar is using AI countries in this group. Nigeria has leveraged foreign direct
applications in the oil industry—its dominant sector—to investment to lead Africa’s fintech revolution. If the country
optimize production and boost sustainability. This puts focuses on developing AI talent within its growing popula-
Qatar at the leading edge of innovation in the industry. tion and adopts a more holistic approach—such as imple-
menting a national AI strategy—Nigeria could build on its
Exposed Practitioners. This group includes developing fintech momentum and become a key player in the conti-
and developed economies vulnerable to AI disruption due nent’s AI landscape.
to more high-exposure sectors and low readiness. Actors
here will need to accelerate AI adoption and mitigate
potential risks.
While these countries may currently have a gap between
their AI exposure and readiness, they are well positioned to
gain ground quickly with investments in infrastructure and
education. It is a sound strategy to focus on niche and
specialized markets.
• Malta is becoming a leader in AI regulation and block-
chain, building a safe and attractive environment for
tech companies.
• Cyprus is using a skilled workforce to develop AI applica-
tions in tourism and financial services.
Others in the group can build on the lessons learned:
Bahrain and Kuwait can leverage AI in the energy sector,
especially to optimize oil production and manage supply
chains. Greece and Bulgaria have strong academic tradi-
tions in engineering and mathematics, which can serve as
a foundation for building AI expertise. By investing in AI-fo-
cused education and retraining programs, they can en-
hance their readiness.
BOSTON CONSULTING GROUP THE AI MATURITY MATRIX: WHICH ECONOMIES ARE READY FOR AI? 14
Strategic Next Steps
What can governments do to position themselves for ad- • Accelerating AI: customizing the ASPIRE levers for
vantage in the AI-dominated future? We propose a set of AI contenders and AI practitioners
initiatives for each archetype across three themes, as
shown in Exhibit 5: • Amplifying AI: driving the global AI agenda for
AI pioneers
• Enabling AI: establishing the foundational elements
for AI emergents
BOSTON CONSULTING GROUP THE AI MATURITY MATRIX: WHICH ECONOMIES ARE READY FOR AI? 15
Exhibit 5 - Recommendations on AI Adoption for Each Archetype
Enabling AI Accelerating AI Amplifying AI
AI emergents AI contenders AI pioneers
AI practitioners
Enable AI adoption through a Actively oversee AI adoption, Support leading AI
A national AI strategy and a with a focus on addressing industry(ies) across the
Ambition
dedicated entity to oversee lagging topics. tech value chain.
implementation.
Provide basic AI training and Attract and retain AI talent pool Enhance cross-cutting AI
S digital programs to modernize (software developers, engineers) expertise and sector
Skills
the workforce. and focus on big data and specialization among AI
advanced trainings in AI. specialists.
P Policy and Enhance government Focus on AI ethics Ensure centralized
effectiveness to build a and flexible rules for oversight and more flexible
regulation
foundation for AI. experimentation. rules on open data.
Boost investments in R&D, Boost investment in high- Provide tailored support
I university programs, performance computing and for national AI champions,
Investment
workshops, and engage the data centers, and attract FDI unicorns, and startups.
private sector. in AI.
Establish research Create test beds for Focus on applied
R Research and
centers in AI and work developers and startups. research and ensure
innovation to ensure industry cross-industry sharing.
collaboration.
Ensure basic digital Promote AI solutions Enhance cross-cutting
E
infrastructure (e.g., and new technologies for AI application and
Ecosystem
high-speed internet) to strategic sectors. support advanced tech
enable AI adoption. infrastructure.
Sources: BCG Center for Public Economics; BCG analysis.
Note: FDI = foreign direct investment.
These recommendations offer a national-level approach to • Research and innovation. Encourage cross-disci-
AI readiness. Akin to the economy-wide level, economic plinary research in AI and applications in agriculture, lo-
managers can apply this to drive sectoral transformation. gistics and robotics, with an aim to share best practices.
For example, the framework can be used to drive change
across value chains in agriculture, logistics, and robotics: • Ecosystem. Create platforms that facilitate data sharing
between agritech companies and logistics firms; foster
• Ambition. Set national ambition to boost agriculture an ecosystem that connects robotics engineers, agritech
productivity through AI-powered agritech solutions, experts, and industrial sectors to help transition agricul-
robotics, and logistics. tural robotics to adjacent fields.
• Skills. Reskill workers in both agriculture and logistics With BCG’s AI Maturity Matrix, we hope to offer policymak-
sectors to adopt AI-based technologies in the agriculture ers a practical framework to navigate the evolving AI land-
value chain. scape and harness AI’s potential to strengthen economies
and enhance societal well-being.
• Policy and regulation. Develop policies that support
open data access and interoperability between agritech
data and supply chain systems.
• Investment. Invest in AI infrastructure such as Internet
of Things-enabled supply chains and predictive analyt-
ics platforms to optimize logistics using agritech data;
invest in R&D in scalable agricultural robotics.
BOSTON CONSULTING GROUP THE AI MATURITY MATRIX: WHICH ECONOMIES ARE READY FOR AI? 16
Methodology
We performed a comprehensive regional analysis by divid- Our analysis of economy-level AI exposure is based solely
ing the world into five geographical areas: the Americas, on a sectoral evaluation and the composition of the econo-
Asia, the Middle East and Africa, Europe, and Oceania. my across those sectors. It does not consider additional
Each area was further subdivided into relevant subregions. factors such as overseas workers or business process out-
We then selected the top economies by real GDP 2023 to sourcing. As a result, certain |
119 | bcg | 2024-gam-report-may-2024-r.pdf | FINANCIAL INSTITUTIONS
GLOBAL ASSET MANAGEMENT REPORT 2024
22ND EDITION
AI and the Next Wave
of Transformation
May 2024
Introduction
The global asset management industry’s assets
rose to nearly $120 trillion in 2023, reverting from
a decline the year before. However, asset managers
are facing a variety of challenges to their growth.
Investors are gravitating to passively managed funds and that can enhance a three Ps strategy. AI can boost produc-
other products that have lower fees even as asset managers’ tivity by enabling improved decision making and operation-
costs increase. Their efforts to create new products that al efficiencies. It can be leveraged to create and manage
would differentiate them from competitors have largely fallen personalized portfolios at scale and to tailor the customer
short, with investors sticking mostly to established products experience. And AI can enhance the efficiency of deal
with reliable track records. Historically, the industry has been teams in private markets and boost their ability to drive
able to weather these pressures thanks to revenue growth value creation. In adopting AI to facilitate these key moves,
that has been largely driven by market appreciation. In the asset managers should view the technological possibilities
years ahead, however, market appreciation is expected to as transformational tools for their organization.
slow, creating further challenges to the industry.
As part of this year’s report, we surveyed asset managers
In the face of these pressures, asset managers will need to with collectively more than $15 trillion in assets under
rethink the way they operate in order to maintain the management to gather their views on the role of AI in their
growth and profitability of past years. The most viable way business. The vast majority of survey respondents expect
forward is by using an approach that we call the three Ps: to see significant or transformative changes in the short
productivity, personalization, and private markets. Asset term, and two-thirds either have plans to roll out at least
managers should increase productivity, personalize cus- one generative AI (GenAI) use case this year or are already
tomer engagement, and expand into private markets. scaling one or more use cases.
As the artificial intelligence (AI) technological revolution Waiting is not an option when it comes to investing in AI.
gathers momentum, asset managers have an opportunity The technology is rapidly developing, and asset managers
to invest in AI and integrate it into their operations in ways that do not start their journey now risk being left behind.
BOSTON CONSULTING GROUP AI AND THE NEXT WAVE OF TRANSFORMATION 2
Five Fundamental Pressures Persist
At a first glance, the global asset management indus-
try experienced an impressive rebound in 2023. The
industry’s total assets under management (AuM) rose
to nearly $120 trillion, an increase of 12% over 2022, a
year that saw AuM plummet by 9%. (See Exhibit 1.)
All parts of the world participated in the 2023 recovery:
AuM growth ranged from 16% in North America to 5% in
Asia-Pacific markets, excluding Japan and Australia.
(See Appendix 1.)
However, while dramatic, the growth only serves to mask
the asset management industry’s underlying vulnerability.
Industry revenues increased by just 0.2% in 2023, while
costs rose by 4.3% for the year. With these two opposing
forces at play, profits declined by 8.1%. (See Exhibit 2.)
BOSTON CONSULTING GROUP AI AND THE NEXT WAVE OF TRANSFORMATION 3
Exhibit 1 The Global Asset Management Industry’s AuM Grew by 12% in 2023
GLOBAL AUM ($TRILLIONS) NET FLOWS AS A SHARE OF BEGINNING-OF-YEAR AUM (%)
+12%
116.6 118.7 4.4
+6% 103.7 106.3
CAGR
93.5 3.4
83.5 3.1 3.1
80.7 2.9
73.7
69.1
2.1
47.6
1.6
1.5 1.5
37.3
1.2
0.9
2005 2010 2015 2016 2017 2018 2019 2020 2021 2022 2023 2005 2010 2015 2016 2017 2018 2019 2020 2021 2022 2023
to to
2009 2014
Sources: BCG’s Global Asset Management Market Sizing Database, 2024; BCG’s Global Asset Management Benchmarking Database, 2024.
Note: Market sizing corresponds to assets sourced from each region and professionally managed in exchange for management fees; it includes captive AuM
of insurance groups or pension funds that delegate AuM to asset management entities with fees paid. Globally, 44 markets are covered, including offshore
AuM, which is not included in any one of the six regions. (See Appendix 1.) For all countries where the currency is not the US dollar, the end-of-year 2023
exchange rate is applied to all years to synchronize the current and historic data. Values differ from those in prior studies due to exchange rate fluctuations,
revised methodology, and changes in source data.
Exhibit 2 Rising Costs and Stagnant Revenues Drove Profits to Decline
Average AuM Net Revenues Costs Profit pool
INDEX INDEX INDEX INDEX
+1.4% +0.2% +4.3% –8.1%
176 176 172 180 184 169
213 216 134 129 144
100 100 100
142 2010 2015 2022 2023 2010 2015 2022 2023 2010 2015 2022 2023
100
AUM (BASIS POINTS) AUM (BASIS POINTS) NET REVENUES MARGIN (%)
26.1 24.5 22.0 21.7 17.4 15.7 14.9 15.3 34 36 32 30
2010 2015 2022 2023 2010 2015 2022 2023 2010 2015 2022 2023 2010 2015 2022 2023
Source: BCG’s Global Asset Management Benchmarking Database, 2024.
Note: The analysis is based on a global benchmarking study of 80 leading asset managers, representing $69 trillion in AuM, or about 60% of global AuM.
Index totals have been rounded.
BOSTON CONSULTING GROUP AI AND THE NEXT WAVE OF TRANSFORMATION 4
Tight monetary poli cies and In addition, the asset management industry continued to
face structural challenges from the five fundamental pres-
general market uncertainty sures that we identified in last year’s report. These pres-
sures did not subside in 2023. (See Exhibit 3.)
prompted investors to move
Revenue pressure continues. Asset managers cannot rely
into products with lower fees.
on market performance to drive revenue growth in the future
to the same extent that they have in the past. Since 2006,
Money market products had
almost 90% of the industry’s revenue growth has come from
net inflows of $1.3 trillion. market appreciation. This growth coincided with a period of
generally low interest rates. However, as most global central
banks continue their fight against inflation, interest rates are
expected to remain higher, a condition that will likely constrain
asset managers’ revenue growth from market appreciation.
Passive funds are increasingly popular. Passive prod-
ucts continue to capture the lion’s share of net flows. In
2023, passive products attracted 70% of total global mutual
funds and exchange-traded funds (ETFs) net flows (about
$920 billion). That was a sharp rise compared with the
period from 2019 through 2022, when 57% of net flows
went into passive products.
Fee compression is accelerating. Similarly, the pressure
on fees showed no signs of reversing in 2023. The average
fee in 2023 was 22 basis points (bps), down from 25 bps in
2015 and 26 bps in 2010. Continued tight monetary policies,
combined with general market uncertainty, resulted in inves-
tors moving into products with lower fees. Money market
and bond products generated net inflows of $1.3 trillion and
$700 billion, respectively, while public equity had net out-
flows of $200 billion.
Costs are rising. Costs continued on an upward trajecto-
ry, increasing by about 80% since 2010 at a compound
annual growth rate of 5%.
Fewer new products are surviving despite attempts
at innovation. Despite asset managers’ continuing efforts
to develop new offerings, many have not been successful.
In fact, only 37% of all mutual funds launched in 2013 still
existed by 2023. This is a significant decrease, compared
with 2010 when 60% of funds that had been launched a
decade earlier remained active.
BOSTON CONSULTING GROUP AI AND THE NEXT WAVE OF TRANSFORMATION 5
Exhibit 3 Five Fundamental Pressures Weigh on Growth
Revenue pressure continues
Market performance has been the main driver of growth Investors are shifting to products with lower fees
$BILLIONS NEW FLOWS, 2023 ($TRILLIONS)
Revenue from
net flows 89 194
11
89
94 54
1.3
43 89% of total
revenue growth 0.7
Revenue
pressure –0.2
2005 2006–2023 2006–2023 2023 Money Bonds Equity
Revenues Revenues from Revenues Revenues market
net flows, offset from market
by revenue performance
pressure
Passive funds are increasingly popular
Net flows to passively managed funds increased Top ten fund managers captured an increasing share of positive net flows
NET FLOWS TO FUNDS, US ONLY1 POSITIVE NET FLOWS INTO PASSIVE POSITIVE NET FLOWS INTO ACTIVE
MUTUAL FUNDS, US ONLY ($BILLIONS)2 MUTUAL FUNDS, US ONLY ($BILLIONS)2
84%
181 504 382 162
9% 5%
26% 1.1 45% 33%
8%
91% 95%
3.2 6.0 67%
55%
2.4
1.1
0.2
1990–1999 2000–2009 2010–2023 2010 2023 2010 2023
Active Passive Share of passive Top ten3 Rest of industry
Fee compression is Costs are rising Fewer new products are surviving
accelerating despite attempts at innovation
70%
26 bps 64% 180
25 bps
66% 60%
22 bps
129
42%
100 37%
2010 2015 2023 2010 2015 2023 2010 2015 2023
Average fee (net distribution costs) Costs Costs as a share New funds that reach
(index, 2010) of revenue the ten-year mark
Sources: BCG’s Global Asset Management Market Sizing Database, 2024; BCG’s Global Asset Management Benchmarking Database, 2024; ISS
Market Intelligence Simfund; BCG analysis.
Note: bps = basis points. All figures are global unless otherwise noted. Revenue pressure includes the impact of both the shift in product mix and
change in pricing pressure. The scope of the analysis is active core, active specialties, solutions, and passives; it excludes alternatives. Values differ
from those in prior studies due to exchange rate fluctuations, revised methodology, and changes in source data.
1Corresponds to mutual funds, including exchange-traded funds but excluding variable annuities.
2Corresponds to mutual funds, including exchange-traded funds but excluding money market and variable annuities.
3The largest ten firms by the amount of positive net flows received.
BOSTON CONSULTING GROUP AI AND THE NEXT WAVE OF TRANSFORMATION 6
AI Can Accelerate the Three Ps
To remain competitive and boost profitability in AI is being built into a variety of tools that asset managers
the face of the five fundamental pressures, asset can use to improve their operations. The power of such
managers should use an approach that we call tools comes from AI’s ability to rapidly collect, synthesize,
the three Ps—productivity, personalization, and and analyze vast amounts of data from internal and exter-
private markets. nal sources and then generate information on the basis of
patterns found in the data. The subset of AI known as gen-
We introduced this approach in last year’s report and erative artificial intelligence (GenAI) has the ability to inter-
continue to find it the best strategy for spurring growth. pret and analyze unstructured data from a wide range of
Increased productivity can make a big difference in just sources and create original content. Tools that combine the
about every organizational function. Improved personaliza- capabilities of AI and GenAI can communicate with users in
tion can facilitate the development of products tailored to natural language, a feature that simplifies their use and can
the unique needs of customers, enhance the customer accelerate their adoption.
experience, and enable asset managers to distinguish
themselves effectively from competitors. The expansion Both AI and GenAI are becoming critical to asset managers.
into private markets can help asset managers focus on Those that service insurance portfolios are finding these
higher-margin products to diversify revenue. Key to accel- technologies instrumental as they adapt to new pressures
erating each of these elements is AI. on their allocation and risk management strategies. (See
the sidebar “The Future of Risk-Adjusted Performance.”)
BOSTON CONSULTING GROUP AI AND THE NEXT WAVE OF TRANSFORMATION 7
The Future of Risk-Adjusted Performance
In the insurance industry, investment income can represent model, however, can compare portfolio holdings and their
as much as 30% to 50% of a company’s earnings. Each day, risk levels, optimize runoff profiles, and establish new risk
billions of dollars from insurance portfolios move through thresholds all in one step. With the present-day systems,
the financial markets. As a result, any small change in portfolio managers typically lack the resources to perform
performance outcome, even if it’s only a matter of decimal the cumbersome task of analyzing underlying assets from
points, can add or destroy significant amounts of value. third-party sources, such as managers of funds of funds or
ETFs, more often than two or three times a year. AI, howev-
To drive performance and lessen risk, insurance asset er, can quickly detect and analyze that data.
managers have long relied on two main analytical process-
es. Asset and liability management (ALM) is used to inform The rapid turnaround makes it possible to provide insur-
the investment team about the commitments made ance clients with a transparent, multidimensional assess-
through policies, while strategic asset allocation (SAA) ment of which assets are being stacked against which
helps determine how to maximize investment upside while liabilities—and the analysis can be performed monthly or
minimizing risks for the insurer. even weekly. On the liability side, AI has proven itself able
to forecast insurance policy lapse rates. This is a capability
Now, however, asset managers are under pressure to perform that was previously available only to managers of larger
these processes in much greater detail and far more often. portfolios with the means to build analytical models to
scale. Now, however, GenAI models, which can system-
Regulations are one source of pressure. The International atize vast amounts of scattered data from both structured
Financial Reporting Standards (IFRS) amendments 17 and and unstructured sources, can make this capability avail-
9 that took effect in Europe and Asia in 2023 have brought able to asset managers of any size.
a new set of accounting practices. The IFRS requirements
for standardized performance metrics have compelled
many asset managers serving insurance clients in those
regions to rethink their previous asset allocation strategies
so that they can achieve their objectives.
But even greater pressures have arisen from the geopoliti- In one step, AI models can
cal turmoil and resulting market uncertainties that affect
compare portfolio holdings and
every part of the world. In this chaotic climate, insurance
portfolio managers need to be prepared with in-depth
their risk levels, optimize runoff
market intelligence so that they are ready to make adjust-
ments far more frequently. Whereas it used to be typical to
profiles, and establish new
conduct ALM and SAA reviews once or twice a year, quar-
terly reviews are now considered the minimum require- risk thresholds.
ment, and some firms are starting to conduct the process
every month, leveraging a greater amount of internal and
external data.
The most advanced players are starting to use AI and its GenAI models provide the ability to develop and automate
generative AI (GenAI) subset to perform their ALM and many reporting exercises. For example, reconciling the
SAA processes with the depth and frequency that’s now market performance of portfolio holdings with the account-
required, and it is becoming clear that this is where the ing figures is still largely a manual exercise, but it will not
future lies for insurance investment management. AI mod- be for much longer. Using GenAI, an asset manager can
els are especially effective at combining large amounts of connect automatically with all requisite data platforms and
data, including unstructured data from multiple sources, to quickly download a full report. In this case, too, the system
inform the decision-making task. These models make it provides a fully transparent disclosure of where the num-
possible to extract more value from the analytical processes bers come from.
and reduce the associated costs by 5% to 15%. Moreover,
with the exponentially greater efficiency gained from AI, By using AI- and GenAI-powered risk and allocation analyt-
some players have been able to achieve risk-adjusted ics, portfolio managers also gain access to a wider scope
returns that have been 10 to 20 basis points higher than of investment parameters and data, with the ability to
previous performance. react quickly when market changes call for adjustments.
These advanced systems are going to become a source of
AI models are able to boost the efficiency of both the ALM competitive advantage in the next few years—and it will
and SAA processes in a number of ways. Currently, most be mandatory for insurance asset managers to embrace
reviews are still performed using mathematical models that AI to continue producing winning investment results and
optimize the results using only one variable at a time. An AI cost efficiencies.
BOSTON CONSULTING GROUP AI AND THE NEXT WAVE OF TRANSFORMATION 8
AI Can Increase Productivity
Asset managers can achieve significant efficiency gains • Operations and Risk and Compliance. AI can make
across their value chain by using AI. (See Exhibit 4.) a significant impact on reporting and data management
workflows, primarily by accelerating document
• Sales and Marketing. A combination of AI and GenAI preparation. This capability can be used for portfolio
tools can help develop marketing content—drafting performance analysis, management research, and the
white papers on the basis of internal research, for readying of client statements, proxy forms, dividend
example, or creating social media posts that summarize notices, and more. AI can make risk management
the white papers. AI tools can analyze public data about more efficient by using its ability to analyze system
prospective clients and then direct the sales team to the logs and real-time data, identify irregular activities,
most promising leads, increasing conversion efficiency. and proactively flag anomalies to the risk team. AI-
Additionally, AI tools can support sales teams with based alerts can go well beyond simple rules-based
customer interactions. For example, a software-based AI notifications; for example, AI can detect signs of market
agent can provide real-time insights to a human sales instability from news reports and respond before a
agent who is speaking with a client; an AI agent can portfolio value crosses the threshold that would have
even communicate directly with clients. triggered an action.
• Investment Management and Trade Execution. AI • IT. AI can enhance the efficiency and effectiveness of
can support investment teams with thesis development IT infrastructure management by detecting anomalies,
by quickly gathering, synthesizing, and analyzing predicting failures, and automatically troubleshooting
data. AI tools can do this whether the information is internal networks. AI copilots can streamline the coding
proprietary—from internal research, for example—or process, as well as accelerate the development, testing,
is compiled from the web or alternative data sources, and deployment of trading algorithms. AI chatbots can
such as public filings, macroeconomic statistics, and support the internal IT desk, enabling faster problem
geospatial reports. Additionally, the tools can facilitate solving when users experience technical issues.
effective knowledge management and data sharing by
organizing reports, data sets, and research developed • Business Management and Support. AI can improve
by various investment teams. As a result, AI can break decision-making and strategic-planning efficiency
down silos and minimize redundant analyses, which by analyzing performance updates across different
occur frequently when investment teams managing investment teams and generating synthesized insights
different funds or products are exploring similar themes. for executives. Similar insights can be used to generate
fundraising documents and investor presentations.
AI tools can automate the creation and review of
legal documents and contracts, quickly spotting and
addressing potential issues.
Exhibit 4 AI-Enabled Gains Can Improve Productivity Across the Value Chain
100 1.50–2.50
0.50–0.75
2.50–4.50
15–25
30–40
1.00–1.50
0.75–1.50
10–20 0.25–.075
20–25 0.10–0.25 0.10–0.25 0.10–0.25 0.25–0.50 85–95
15–30
15–25 15–25 15–20 10–20
5–10
Estimated efficiency gains (%)
Asset Sales Marketing Investment Operations IT Risk and HR Legal and Finance Management Asset
managers’ management compliance audit and strategy managers’
cost base and trade AI-powered
(indexed) execution cost base
Business management and support
Sources: BCG’s Global Asset Management Benchmarking Database, 2023; expert interviews; BCG analysis.
Note: Individual value chain ranges do not add up to the total range because of rounding.
BOSTON CONSULTING GROUP AI AND THE NEXT WAVE OF TRANSFORMATION 9
AI can enhance productivity These efficiency opportunities represent just a fraction of
AI’s potential; the list of applications continues to expand
by conducting preliminary
alongside the advancement of the technology. Asset man-
agers are already witnessing impressive results from AI
investment research, data
implementations. For example, an AI tool can accelerate
collection, and analysis. investment research. (See Exhibit 5.) This tool could
enhance investment analysts’ productivity by conducting
preliminary data collection and analysis, enabling the
human analysts to concentrate on generating insights. An
analyst may, for example, use the AI tool to summarize a
company’s market position on the basis of its financial
filings and news coverage. After reviewing this initial
research, the analyst can employ the tool for a more
in-depth analysis of selected topics needing further inves-
tigation. Eventually, the tool can be used to draft a report
focused on the key issues.
Exhibit 5 Asset Managers Can Use AI and GenAI to Accelerate
Investment Research in Natural Language
1. Conduct a search 2. Receive selected topics 3. Create a tailored report
AI responds to a research and insights AI drafts a written investment
analyst’s request for information AI delivers a detailed analysis of key report in the required format
about a company and its market elements requested by the analyst
Tell me about [company] Analyze the latest financial Draft a report for
and its market position in information on [company]. [company] in this structure:
under 1,000 words. Is there anything related · SEC filings
to supply chain disruption?
· Conference calls
RESEARCH ANALYST
· Press releases
RESEARCH ANALYST
· Equity research
· EPS consensus
[Company] is a United
States-based Here is information based RESEARCH ANALYST
semiconductor on [company’s] latest
manufacturer. Its quarterly reports and an
current stock price is analyst’s event Here is the report.
$105.92. . . . presentation:
AI AND GENAI TOOLS
· Outpacing industry
AI AND GENAI TOOLS growth at 25% CAGR
· Sales of global equipment
increased 20% annually
since the pandemic. . . .
AI AND GENAI TOOLS
Sources: Expert interviews; BCG analysis.
BOSTON CONSULTING GROUP AI AND THE NEXT WAVE OF TRANSFORMATION 10
AI Can Enable Greater Personalization
AI can enhance personalization by expanding the ability The second development is the efficient scaling of custom-
to create and manage customized portfolios at scale. In ized portfolio management, resolving a major challenge that
addition, it makes it possible to provide a more highly has made bespoke investing a high-cost service that is
tailored customer experience, from acquisition through available only to institutional investors and high-net-worth
retention, at scale. individuals. When investing is nonpersonalized, capital from
a large group of investors goes into a vehicle such as a mutu-
Personalized Portfolios at Scale al fund, an ETF, or a pooled (or commingled) fund, and one
Advancements in AI are leading to two major develop- asset manager efficiently determines the optimal rebalanc-
ments for personalized portfolios. First, AI represents a ing and trading strategy for the whole fund. However, in
step-change in the construction of customized portfolios. personalized portfolio management, optimal rebalancing
Currently, adding tactical tilts or thematic exposures to and trading strategies need to be determined for each inves-
personalized portfolios is primarily based on structured or tor’s portfolio—for example, to respect specific objectives
precurated data sets. For example, a financial advisor can and constraints or to maximize tax loss harvesting opportu-
add an ESG tilt to a client portfolio if the advisor has nities. Consequently, the more clients with personalized
access to a data set that measures companies’ ESG charac- portfolios that an asset management firm has, the more
teristics. AI, and more specifically GenAI, can significantly portfolio managers it will need. It is nearly impossible to
expand the range of tactical tilts and allow a theme-based automate personalized portfolio management using statisti-
selection of securities. This is possible due to GenAI’s cal or rules-based processes across thousands of fully cus-
ability to understand requirements expressed in natural tomized accounts because the number of parameters grows
language and then process and translate that information so quickly.
into investment recommendations.
However, with the latest AI developments, AI-powered agents
For example, an advisor could tell a GenAI tool that a client can be trained to understand the intent and context of portfo-
wants to decrease allocations to companies that are heavi- lio management. Learning from patterns, such agents can
ly exposed to the oil and gas value chain. The tool will tailor their approach to determine the best rebalancing and
quantify the exposure of publicly traded companies by trading strategy for each custom portfolio. The human portfo-
reading through financial filings, transcripts of earnings lio manager can oversee a group of these AI agents, effective-
calls, analyst reports, and news reports. The tool will then ly managing a much larger number of portfolios than they
rank these companies on the basis of their exposure to the otherwise would be able to. With this capability, asset man-
oil and gas value chain and make adjustments in the port- agement firms can potentially offer personalized portfolios to
folio accordingly. a much broader group of investors. (See Exhibit 6.)
Exhibit 6 AI Makes It Possible to Scale Personalized Portfolio Management
Nonpersonalized investing Customized investing Customized investing with GenAI
Clients (retail or institutional) Clients (retail or institutional) Clients (retail or institutional)
Advisor1 Advisor1 Advisor1
Commingled or pooled fund
The first AI agent narrows down the list of
portfolios to trade given the results of its task
A portfolio manager at an asset Several portfolio managers at various
management firm manages the pooled asset management firms—each one
funds of all investors operating at capacity—manage individual
client portfolios
The second AI agent further refines the list of
portfolios to trade given the results of its task
A portfolio manager makes the final decision
KEY CHARACTERISTICS
Scalable but limited; no customization Allows customization, but requires Allows customization and at scale
many portfolio managers
Sources: Expert interviews; BCG analysis. Different types of portfolios
Note: GenAI = generative artificial intelligence.
1Financial advisor, relationship manager, or sales professional.
BOSTON CONSULTING GROUP AI AND THE NEXT WAVE OF TRANSFORMATION 11
Traditional sales approach AI-enabled sales approach
Leads prospect meetings
BOSTON CONSULTING GROUP AI AND THE NEXT WAVE OF TRANSFORMATION 12
noitareneg
daeL
tcatnoc
laitinI
tnemegagne
tneilC
Personalized Customer Experience • Customer Sales. A human sales agent can use AI tools
AI can also enable the hyperpersonalization of customer that analyze information about a potential client before a
engagements at scale. This can be achieved by using AI meeting, identifying the client’s needs and preferences,
tools that analyze data about prospective or existing clients and developing talking points accordingly. For example,
and then develop investment materials and engagements AI tools can help determine the potential client’s risk-
tailored to their needs and circumstances. GenAI can reward profile on the basis of demographics, then present
expand the capabilities of AI by enabling the analysis of the sales agent with discussion points that focus on the
unstructured text-based data—customer social media posts, appropriate products.
for example—and facilitating the creation of investment
information that is tailored to each individual customer. • Customer Engagement and Retention. AI can
enable a shift from periodic scheduled contacts, such
More specifically, AI can enable greater personalization in as biannual touch points, to proactive engagements
three areas of the customer experience. informed by relevant market events. For instance, AI tools
may evaluate the effects of an upcoming fee change for
• Marketing. AI can help the marketing team segment a mutual fund and identify clients who would be likely
potential customers into highly granular groups with to withdraw their funds. The relationship management
clearly defined needs. Then, using natural language, team can then contact these clients and recommend
the team can leverage AI tools to develop marketing investments in other funds with lower fees. This proactive
materials that are tailored to specific customer interests. approach helps ensure better client retention.
For example, if a marketing team writes a white paper,
AI tools can develop personalized emails for a group of Overall, AI can enable client management teams to support
prospective clients, extracting key takeaways tailored to a significantly higher number of clients with more tailored
each recipient. content at relevant moments. The implementation of AI
tools can lead to a decrease in indirect sales activity of as
much as 25% while simultaneously increasing customer
satisfaction. (See Exhibit 7.)
Exhibit 7 How AI Assistants Can Improve Sales Effectiveness
Identifies and prioritizes Forms initial leads through Identifies high-
leads manually industry contacts potential leads
Sends standardized email Reviews and sends Creates personalized
to all leads to leads emails
Monitors outreach and manually Tracks prospects’ responses Creates customized
updates CRM and updates CRM meeting materials
Prepares for meetings with time-
intensive research; leads meetings
Researches internal Synthesizes CRM and Leads client meetings
databases to manually create product information for
meeting materials dossier before meetings
Manually reviews notes and Generates meeting notes and Reviews next steps and
develops next steps proposes next steps follows up with clients
Human agent AI tools 25% Decrease in time spent on 8–12 Number of additional
client touch points
indirect sales activity
per week
Sources: Expert interviews; BCG analysis.
Note: CRM = customer relationship management.
Exhibit 8 AI Tools Increase the Efficiency of a Private-Market Deal Team
Investment committee memo preparation
TIME SPENT USING THE TIME SAVED WITH AI
ACTIVITY CONVENTIONAL PROCESS (ESTIMATE) (ESTIMATE)
Data analysis 20%–25% 30%–40%
Written output 15%–25% 35%–45%
Meetings and calls 15%–25% 20%–30%
Research 10%–15% 35%–45%
Email 5%–15% 20%–30%
Note synthesis 5% 45%–55%
Content review 5%–10% 20%–30%
Scheduling and other calls 5% NA
Sources: Expert interviews; BCG analysis.
Note: NA = not applicable.
AI Can Unlock Private-Market Potential
AI can enhance the efficiency of private-market deal teams Private-market players can also drive value creation by helping
by automating repetitive tasks and synthesizing data for their portfolio companies use AI. This will be especially
enhanced decision making. (See Exhibit 8.) It can improve important for asset managers that invest in companies
deal teams’ productivity across many parts of the value whose industries are expected to be highly disrupted by AI. In
chain. We estimate that in the due diligence process, for biotech, to name one example, AI is expected to increase the
example, AI can shorten the time required for preparing pace of product innovation and create new, efficient ways to
investment committee memos by roughly 30%. AI tools are discover new molecules and compounds. Private-mar |
120 | bcg | generative-ai-in-health-and-opportunities-without-spine98.pdf | Generative AI in health
and opportunities for
public sector organizations
October 2023
By Priya Chandran, Lauren Neal, Julia McBrien, and Shabana Quinton
Generative AI in health and
opportunities for public sector
organizations
Key takeaways: GenAI opportunity landscape across the
health ecosystem
• Generative AI has the potential to transform industries.
GenAI is projected to grow faster in healthcare than any
• The healthcare industry is expected to experience some other industry. With an estimated compound annual
of the most significant benefits from and growth in growth rate of 85%, by 2027 the market value is expected
GenAI investments in the coming years. to reach $22B1. This appetite for investment is driven by
the potential for GenAI to significantly enhance efficiency,
• As practitioners, public sector organizations can leverage reduce costs, and improve health outcomes.
GenAI, and AI more broadly, to improve their operations,
and accelerate delivery against their missions. The power of AI promises value for all stakeholders across
the health ecosystem. For example, for providers, GenAI is
• As enablers, public sector organizations can play a key expected to create $80B+ in savings via automated opera-
role in accelerating the adoption of AI within the health tions, reduced burden, and improved revenue cycle man-
ecosystem, including by establishing policy and regula- agement. AI, and GenAI in particular, has the potential to
tions that promote the responsible use of AI and reduc- provide patients with improved quality, access, affordabili-
ing barriers related to data and technology, workforce, ty, personalization, and equity of healthcare services to-
and infrastructure. ward better patient outcomes. Through GenAI, providers/
hospital systems are offered the potential for quicker,
• To get started, public sector organizations should estab- more effective methods to prevent, diagnose, and treat
lish rules of responsible engagement, define and com- patients. Payers may benefit from improved data sharing
municate their AI strategies, serve as a central coordina- and analysis, communication, and payments, including
tor to improve transparency and facilitate partnerships, preventative healthcare through predictive models and
improve their internal readiness to adopt and scale AI, automation of claims processing. Research & develop-
advance workforce capabilities, build a culture of AI use, ment institutions including pharma, biopharma, and
and provide thought leadership to establish trust in AI biotech can accelerate the product pipeline by applying
among users and the public. GenAI to drug discovery and design, clinical trial planning
and execution, precision medicine therapies, diagnostic
image enhancement & analysis, supply chain risk identifi-
What is Generative AI? cation and process augmentation, and more.
Generative artificial intelligence (GenAI) has emerged as Public sector organizations have an important role to
the biggest buzz word across nearly all industries given the play in both using and advancing the technology:
potential for its broad and deep applications and the speed
at which the technology is maturing. Evolving from prior • As practitioners, they can use GenAI to improve internal
advances in deep learning and machine learning, the most operations and better deliver against their own mission, by
powerful GenAI algorithms are trained on vast quantities deploying the latest technology to modernize processes,
of unlabeled data in a self-supervised way. They learn products, and services to more cost-effectively and effi-
underlying patterns from training data, which enables ciently serve individuals, agencies, and businesses.
them to complete a wide range of tasks, including creating
fully original text, images, audio and more in a matter of • As enablers, they can advance the healthcare AI ecosystem
seconds. With thousands of new GenAI tools being devel- by fostering research, innovation, investment, work-
oped each week, there is tremendous anticipation and force development, and collaboration. Enablement also
excitement about its potential. includes limiting unintended consequences by codifying
best practice standards, policies, and regulations in eth-
ics and safety, data privacy, and security.
1. AI TAM research, Expert interviews, BCG analysis.
1 GENERATIVE AI IN HEALTH AND OPPORTUNITIES FOR PUBLIC SECTOR ORGANIZATIONS
Practitioners: Improving internal operations • The U.S. Centers for Disease Control and Prevention
and mission delivery (CDC) could reduce vaccine misinformation and ad-
vance health equity through AI-driven multi-language
Public sector organizations have started to experiment chatbots to answer questions related to disease control
with AI to improve their internal operations, deploying the and prevention, quickly generate communication mate-
technology to improve ways of working and to optimize rials during an emergency response, and monitor public
business processes. The addition of GenAI will further health data.
enhance these capabilities. For example, the Department
of Veterans Affairs (VA) has used AI to sort incoming • The Administration for Strategic Preparedness and
claims from multiple sources (e.g., mail, fax, and electron- Response (ASPR) may use AI to predict outbreaks,
ic), reducing processing times from 10+ days to less than 1 optimize mobilization of resources during an emergency
day. The Centers for Medicare and Medicaid Services response, and assess the strength of health systems to
(CMS) built an AI pipeline to drastically reduce time spent predict areas most likely to be impacted.
on the Authority to Operate (ATO) security planning pro-
cess, a process which currently requires 540+ hours logged • Lastly, Centers for Medicare and Medicaid Services
for every document submitted. CMS and other federal (CMS) could use AI to forecast needs based on health
health organizations, including the Agency for Healthcare data, improve management of allocations, and detect
Research and Quality (AHRQ) and Health Resources and instances of fraud, waste, and abuse.
Services Administration (HRSA), are using chatbots to
improve customer service by providing knowledge- and While these use cases may be built within the limitations
action-based responses with 24/7 availability. There are of currently available data and technology, there are oppor-
opportunities to expand existing AI use cases to be genera- tunities to evolve them to be generative in nature as the
tive in nature as the depth and breadth of data sets grow, landscape matures and evolves.
more sophisticated models are trained, policies and regula-
tions are established, and new applications are tested.
Enablers: Advancing the health ecosystem
To realize the potential for GenAI internally, selecting through AI
priority use cases that drive step-change improvements in
productivity and effectiveness is the most critical first step. Public sector organizations are also uniquely positioned to
Beyond those described above, Exhibit 1 outlines several support the broader health ecosystem - including patients,
additional use cases that can address common challenges providers, payers, biopharma, MedTech, other federal
related to internal operations, ranging from use cases that agencies - in harnessing the full power of GenAI through
are conceptual to those that have been implemented, the products and services that they deliver. When enabled
tested, and validated. by the public sector, the opportunities for GenAI applica-
tions across the health ecosystem are extensive.
Importantly, public sector organizations should also define
and prioritize AI use cases that directly enable mission The health ecosystem generates massive data stores that
delivery. For example, these may include: can be used to train generative AI models from sources
including electronic health records, imaging, testing &
• Using AI to improve the speed and consistency of med- diagnostics, -omics, biosensors and more. Additionally,
ical product safety reviews for the U.S. Food and Drug GenAI foundational models trained on these data sets will
Administration (FDA) by summarizing drug adverse have a myriad healthcare applications, including virtual
event reports, prioritizing reports for further evaluation, health coaches, precision health monitoring, ‘hospital at
and referencing historical data to track emerging issues. home’ services, disease surveillance, digital twins, digital
clinical trials, and more. These applications have the po-
• AI could be used by the National Institutes of Health tential to impact quality improvement, patient safety,
(NIH) to boost productivity of clinical trial development clinician/patient experience, and access to care, issues that
teams by gathering intelligence from numerous sourc- are often core to agency missions (Exhibit 2).
es (e.g., ClinicalTrials.gov, FDA guidance documents,
PubMed, and other publicly available sources of scien-
tific publications) to provide protocol writing support,
elevate areas of over- or under-investment in research
by reviewing grant submissions against previous awards,
or improving health equity through pre-assessment of
clinical study plans.
BOSTON CONSULTING GROUP 2
Exhibit 1 - Applications of generative AI to improve public sector internal
operations
Non-exhaustive
Functions Potential GenAI applications
Siloed service delivery and Natural and multi-language processing of agency documentation
data; reliance on higher cost and public guidance – ability to personalize interaction in a more
Service human responses natural way without or assisting the human in the loop, and with a
Delivery robust understanding of content, context, and ability to quickly
search for the right data
Front
Office
Time-consuming and menial Automated email generation; personalization of mailings and
approach to composition of correspondence to residents and organizational stakeholders
Public Relations & correspondence Streamlined public consultations from constituents to collect
Communications input and prioritize relevant issues
Missing/incorrect data, Data validation: automating compliance-related coding tasks to
manual processes, high develop software which ensures compliance with regulations
Risk prevalence of fraud, waste, Compliance monitoring and reporting: Identify potential
Management & and abuse in/external compliance breaches through identification of
Middle Compliance non-compliant events; automate drafting of compliance reports
Office Claims risk detection, e.g., rationalizing a claim denial based
on risk of FWA or non-compliance, generating communication
material (claim outcome, appeal responses, etc.), etc.
Poorly managed data (e.g., Assisted data management across data quality improvement,
unstructured, unclean, data lifecycle management, data governance, data integration,
Data fragmented) prevents data processing, data architecture enhancement, etc.
Management timely data to action Automated analytics of managed data, including generation of
& Analytics summaries of findings
Policy research and Policy research: analyzing in/external regulations to benchmark
validation policies
Policy assessment: generating risk assessment scenarios to
Legal
identify potential compliance risks and policy impact
Difficulty across the talent Point solutions: Ingest, categorize, and/or process applications
acquisition and retention with minimal human involvement to significantly reduce
continuum: identifying, processing time; Automated, proactive follow-up to quickly
Back-end
assessing, hiring, training, address application issues
Processing
and retaining talent End-to-end process automation: Augment resource deployment
and increasingly automate processes over time (e.g., automated
self-services Chatbots and digital-first processes with human in
Back
the loop for verification/ transaction finishing)
Office
Manual, time-consuming, Identification: Enhance outreach efforts, documentation, and
and repetitive processes strategy with generative content
HR/People leading to high backlogs Assessment: Predicting hiring success based on prior experience
and user deviation and and skillset; identification of non-traditional candidates and fits
error
Upskill and retain employees with personalized support from
hire to retire in a way that is more approachable and personalized
than existing solutions (e.g., performance management, training)
Improve employee experience by removing mundane/repetitive
aspects of the job
Slow and costly legacy Accelerate productivity and speed throughout each step of the
systems modernization tech modernization journey: 1. Discovery 2. Identification of
and digital transformations dependencies 3. Documentation 4. Coding (generation,
IT specification, conversion) 5. Testing 6. Deployment
Tech. Sub-optimal data Natural language processing of data to better and quickly
Innovation governance and retention understand what to retain and how for compliance and storage
& optimization
Modernization
Slow IT ticket resolution; Automated routing of the IT ticket to best respondent based on
limited self-service issue and respondent expertise (enhanced v. general queue)
Comparison of IT issue against prior to provide validated
resolutions based on prior user feedback (dynamic v. general FAQ)
= Validated = Early stage = conceptual
3 GENERATIVE AI IN HEALTH AND OPPORTUNITIES FOR PUBLIC SECTOR ORGANIZATIONS
Exhibit 2 - Generative AI can create meaningful clinical applications for
stakeholders
Pre-trained model
Model topology Opportunities
Virtual health coach
Precision health
Training data, e.g.,
EHR Hospital at home
Training Fine-tuning
Training of large Leverages specific
Imaging & tests data sets requires domain data, Disease surveillance
significant time & requiring less
resources time & resources
-omics Digital twins
Biosensors
Digital clinical trials
Fine-tuning data
Many others…
Sources: “Multimodal Biomedical AI”, Nature, 2022; “On the Opportunities and Risks of Foundation Models”, Center for Research on Foundation
Models, arXiv, 2021; BCG analysis.
Early use case experimentation across the ecosystem has Successful implementation of these real-world use cases
delivered results. For labs and clinics, large language mod- can address some of the biggest healthcare challenges,
els are being trained on a body of sequences and amino including reducing the burden on providers by providing
acids to generate new protein structures and predict mo- supplemental support to diagnostics; automating filing and
lecular interactions2. The University of Kansas is utilizing fraud detection for payers; aiding in personalized medicine,
technology by Abridge to identify key points from pa- drug discovery, and commercial operations within biophar-
tient-provider conversation and creating EMR-integrated ma; allowing for AI-assisted robots and sensors within
summaries, reducing the documentation burden on physi- MedTech, and more (additional potential use cases are
cians. These are merely examples of many new generative defined in Exhibit 3).
AI applications in healthcare.
2. 2011.13230.pdf (arxiv.org).
BOSTON CONSULTING GROUP 4
Exhibit 3 - Overview of real-world use cases for generative AI across the
health ecosystem
Use case Description and examples
Digital clinical Leverage AI to analyze voice patterns and codify voice biomarkers
1
voice analysis to noninvasively detect abnormalities for clinical diagnosis
Ambient Documentation systems that leverage speech recognition and AI
Providers 2
digital scribe to automate documentation and summarize verbal encounters
Diagnostic image AI imaging interpretation uses deep learning and categorization on
3
interpretation medical images for faster and more accurate image interpretation
Intelligent prior A predictive process that payers utilize to approve care by automating
4
authorization workflows after a provider submits treatment notifications
Payers
Claim fraud ML model to detect fraud patterns by finding connections based
5
detection on different factors from previously processed claims
Precision AI-powered precision medicine provides clinicians with an
6
medicine opportunity to specifically tailor early interventions to each individual
Drug discovery AI algorithms to analyze millions of molecules and potential
7
Biopharma & repurposing interactions with target proteins to develop new drugs
AI in commercial Analytics to increase business impact and efficiency with commercial
8
operations operations functions (sales, customer engagement, marketing, etc.)
Robotic AI-assisted robots to perform sophisticated surgeries with precision and
9
surgeries speed and derive new methods by learning from previous surgeries
Medtech
AI enabled Neuroprosthetic system - AI decoder that learns the user’s intention based
10
prosthetic arm on the nerve signals it senses in the arm to translate movement intent
Source: BCG analysis.
Note: Emerging AI use cases gathered from trend reports and latest Garnter Hype Cycle’s.
5 GENERATIVE AI IN HEALTH AND OPPORTUNITIES FOR PUBLIC SECTOR ORGANIZATIONS
GenAI is projected to grow faster in
healthcare than any other industry.
With an estimated compound annual
growth rate of 85%, by 2027 the market
value is expected to reach $22B.
Exhibit 4 - Requirements to enable to the ecosystem across six dimensions,
ASPIRE
Aspirations need to balance needs and perspectives of residents, public, and private sector organizations
Dimension Requirement
A Articulate AI vision (e.g., global leadership, pioneering ecosystem) and benefits for residents, public,
Ambition
and private sector
S
Skills Attract, develop and retain talent for the workforce to thrive in the new age of AI
P Nurture developments and provide flexibility and certainty over AI activities to provide guidance to
Policy & Regulations
public and private sector. Drive innovation in state and federal regulation of benefits processing.
I
Investment Deploy funding mechanisms to stimulate and attract AI-related private businesses and use cases
R
Research & Innovation Build and enable core research and innovation institutions, public and private, in the domain of AI
E
Ecosystem Stimulate AI adoption through commercialization and industry application
Source: BCG analysis.
Note: Detailed countries benchmarks available in appendix.
Public sector organizations can help ecosystem stakehold- National AI Initiative Act passed in 2020 allocated critical
ers improve their probability of successful implementation funding for AI R&D. Lastly, public sector organizations are
and maximize the impact of their investments in GenAI in enabling the AI ecosystem by making solutions accessi-
several ways, as outlined in BCG’s ASPIRE Framework ble. In 2018, Britain enacted the AI Sector Deal to set out
described in Exhibit 4. actions that would promote the adoption and use of AI in
the UK.
Public sector organizations are already taking strides to
enable AI across the ASPIRE dimensions. For example,
various countries are articulating their AI ambitions and How should public sector organizations get
establishing national AI strategies, including around started with GenAI?
healthcare (e.g., the US Department of Health and Human
Services developed an AI Strategy in 2021). Other countries Generative AI is a new territory, with many organizations
are enabling the ecosystem through investments in skills. still in the experimentation phase, testing and building
For example, Qatar’s AI strategy is highly focused on in- their AI capabilities through a series of pilots. More ad-
vesting in K-12 education, apprenticeship programs, re- vanced organizations are cautiously implementing on a
search funding, and attracting talent. Similarly, public larger scale, while navigating concerns related to accuracy,
sector organizations are enabling AI within the health safety, ethics, and privacy. Some organizations have
ecosystem through the development of policy and regu- well-defined strategies for integrating the use of AI within
lations. The EU is finalizing the terms of the “AI Act” their organizations, while others are still determining how
which will be the world’s first comprehensive regulation of best to jumpstart their AI efforts.
AI, and potentially a global standard. The US government
appointed a committee to improve coordination of federal Depending on the starting point, there are several steps
AI efforts and advise the White House on interagency that public sector organizations can take to accelerate
research and development (R&D) priorities. Countries are responsible implementation internally, and to support
similarly investing in AI, including specific investments wide-scale adoption across the health ecosystem.
into research and innovation. For example, the US
7 GENERATIVE AI IN HEALTH AND OPPORTUNITIES FOR PUBLIC SECTOR ORGANIZATIONS
First, they can establish rules of responsible engage- Critically, public sector organizations must build their
ment. At BCG, we define Responsible AI (RAI) as develop- talent and establish a culture of using AI, ensuring
ing and operating AI systems that align with organizational that employees are familiar with how to operate GenAI-en-
values and widely accepted standards of right and wrong abled applications by developing new educational courses,
while achieving transformative business impact. Public re- and up-skilling current employees, hiring new types of
sector organizations can define AI governance processes, talent (e.g., data scientists), and supporting change man-
policies, and decision rights to guide implementation agement efforts.
internally. Further, they can help the healthcare ecosystem
to more effectively adopt and deploy RAI by providing a Finally, public sector organizations can provide thought
framework that supports organizational decision making leadership on ethics, trust, and regulation to acceler-
and offers guidance for developing and using generative AI. ate AI progress across the health ecosystem. Generative AI
Additionally, by investing in the development of tools to is still little understood by the public, leading to distrust in
monitor and manage generative AI risks, public sector outputs which may be biased, false, or opaque unless
organizations can create feedback mechanisms for users models are reviewed and corrected by human experts and
to report inaccurate or unhelpful results and proactively made more transparent. Furthermore, AI may be misused/
flag issues, such as biased outputs, intellectual property over relied on unless hospitals, clinicians, payers, and other
and copyright infringement, and cybersecurity risks. health ecosystem players clarify how specific solutions
should be used, with clear messaging that AI-generated
Next, they can define and communicate their AI ambi- insights are recommendations rather than mandates.
tion and strategy, identifying top use cases considering Guidance on how to mitigate emerging risks and capture
their potential to impact operational efficiency and mis- early trust and value will depend on setting the right
sion delivery. Prioritization may also be informed by con- guardrails early to experiment in the right way.
ducting a “discovery process”, engaging with ecosystem
stakeholders to understand the big pain points and highest
value GenAI opportunities to enable an efficient and
high-quality health ecosystem. For prioritized use cases,
they can design and launch pilots focused on generating
incremental impact, evaluate their effectiveness, refine as
needed prior to full scale-up, all the while strengthening
their organizational muscle for disruptive change. Further,
public sector organizations may play a role in coordinating
AI investment across the ecosystem, creating awareness of
ongoing investments and facilitating collaboration and
partnership to accelerate outcomes.
To improve the likelihood of successful implementation,
they can improve their “AI readiness” by investing in
infrastructure by integrating data in the cloud, adding
computing power to reduce time to train and run algo-
rithms, provide shared environments where public and
private sector organizations can collaborate, and make
health data resources available to entities across the eco-
system to accelerate equitable healthcare. They can break
down siloed data and technology systems which limit ac-
cess to data needed to train and feed models, impacting
the accuracy of the AI algorithms and quality of the gener-
ated output. They can also integrate new AI models into
legacy technology and operational, policy, and other mis-
sion-specific workflows to drive adoption and improve
usability. Externally, public sector organizations can provide
guidance for industry on ways to improve system interoper-
ability so that systems and AI technologies can work togeth-
er, and varied data sources can be exchanged seamlessly.
BOSTON CONSULTING GROUP 8
About the Authors
Priya Chandran is a Managing Director and Senior Part- Lauren Neal is a Principal based in the Washington, D.C.
ner based in the New Jersey office. You may contact her at office. You may contact her at [email protected].
[email protected].
Julia McBrien is a Project Leader based in the Shabana Quinton is a Partner based in the Washington,
Raleigh-Durham office. You may contact her at D.C. office. You may contact her at Quinton.Shabana@bcg.
[email protected]. com.
For Further Contact Acknowledgements
If you would like to discuss this report, please contact the BCG brings industry-leading talent, deep experience in AI
authors. innovation, speed to value, and vetted partnerships bring-
ing together the best of science, academia, and industry to
enable governments and public sector organizations on
their generative AI journey. Special thank you to Krishna
Srikumar, Steven Mills, Satty Chandrashekhar, and Jona-
than Brice for their contributions to this paper.
9 GENERATIVE AI IN HEALTH AND OPPORTUNITIES FOR PUBLIC SECTOR ORGANIZATIONS
Boston Consulting Group partners with leaders in business For information or permission to reprint, please contact
and society to tackle their most important challenges and BCG at [email protected]. To find the latest BCG con-
capture their greatest opportunities. BCG was the pioneer tent and register to receive e-alerts on this topic or others,
in business strategy when it was founded in 1963. Today, please visit bcg.com. Follow Boston Consulting Group on
we work closely with clients to embrace a transformational Facebook and Twitter.
approach aimed at benefiting all stakeholders—empowering
organizations to grow, build sustainable competitive © Boston Consulting Group 2023. All rights reserved. 10/23
advantage, and drive positive societal impact.
Our diverse, global teams bring deep industry and functional
expertise and a range of perspectives that question the
status quo and spark change. BCG delivers solutions
through leading-edge management consulting, technology
and design, and corporate and digital ventures. We work
in a uniquely collaborative model across the firm and
throughout all levels of the client organization, fueled by
the goal of helping our clients thrive and enabling them
to make the world a better place. |
121 | bcg | generative-ai-for-the-public-sector-from-opportunities-to-value.pdf | Generative AI for
the Public Sector:
From Opportunities
to Value
December 2023
By Miguel Carrasco, Cyma Habib, Frank Felden, Richard Sargeant, Steven Mills,
Simon Shenton, Jamie Ingram, and Gareth Dando
Generative AI for the Public Sector:
From Opportunities to Value
The new tools of generative artificial intelligence (GenAI) are funda-
mentally changing the nature of knowledge work, creating promising
opportunities to significantly increase productivity and spur innova-
tion across entire economies.
Public administration is one sector where GenAI could The market for generative AI products and services is
have the greatest potential. To reap the benefits of the growing exponentially. Since the consortium OpenAI first
technology, public sector leaders need to start by under- announced its language model GPT3 in 2020, GenAI has
standing how GenAI can create value for them. Then they attracted more than $20 billion in venture capital funding.
should set priorities and mobilize to capture its transfor- A recent report by Bloomberg Intelligence predicts the
mative impact. Generative AI refers to a category of artifi- market will grow by over 40% per year for the next ten
cial intelligence capable of creating credible new content, years.
including text, images, audio, code, data, and other media,
based on foundational or generative models. The most The use of GenAI offers significant potential productivity
powerful GenAI tools are trained on large language models gains for the public sector. It can improve the quality and
(LLMs) that process a vast quantity of data to emulate the speed of government decision-making at scale and raise
way people communicate. This capability makes GenAI the efficiency and effectiveness of public policies, pro-
a general-purpose disruptive technology. It expands the grams, and services. The new GenAI tools complement
boundaries of what organizations can do in everyday existing AI capabilities already used in the public sector.
operations, especially in the realm of knowledge work. (See Exhibit 1.)
ChatGPT, one of the first GenAI models for processing
language, has more than 180 million users.
Exhibit 1 - GenAI Expands and Complements Other AI Capabilities
AI Capabilities in Use New GenAI Use Cases
Sense patterns and trends Summarize documents and meetings
Target government investments Review and draft contracts
Segment and design services Engage with customers and citizens
Identify and address risks Generate content and messaging
Optimize public assets and operations Generate code
Detect fraud and anomalies Monitor and use social media
Source: BCG analysis.
1 GENERATIVE AI FOR THE PUBLIC SECTOR: FROM OPPORTUNITIES TO VALUE
Currently, most governments are just beginning to experi- BCG estimates the productivity gains of GenAI for the
ment with the technology. The conditions are not yet in public sector will be valued at $1.75 trillion per year by
place to leverage GenAI at scale and unlock its potential. 2033. (See Exhibit 2.) Our estimate on the impact of GenAI
There are also a number of risks related to issues such as is based on using inputs from Pearson-Faethm modeling. It
accuracy, security, privacy, bias, and intellectual property reflects productivity gains across all national, state or
ownership that need to be managed before fully deploying provincial, and local governments and across all domains
the technology. This article is the first in a three-part BCG such as legislative, administrative, courts, health care,
series exploring how governments can responsibly leverage education, transportation, and security.
GenAI to drive maximum public impact, how they can scale
the technology, and how to assess and mitigate the risks. The impact of GenAI on public sector jobs is more nu-
anced. While some efficiencies may lead to reduced need
for labor, in most cases governments will seek to reinvest
Productivity Value of GenAI for Governments the productivity benefit to address unmet needs of citizens
Estimated at $1.75T or in higher value-added activities that will generate better
outcomes. Some employees will be readily able to adapt
GenAI provides an unprecedented opportunity for govern- and incorporate higher-value work into their role; however,
ments around the world to deliver greater value and public for many workers, reskilling and upskilling will be essential.
impact for citizens, businesses, and government. At a
minimum, these tools could free up many valuable hours
of a public servant’s time on simple and repetitive cogni-
tive tasks and enable that individual to focus on other,
higher-value activities.
Exhibit 2 - Estimated Annual Productivity Benefit from GenAI ($USD Billions)
Sweden (13)
Norway (9) Finland (7)
Canada (22) Netherlands (12) Denmark (11)
UK (89) Germany (62)
France (57) Austria (7)
Japan (20)
USA (358) Spain (20) Italy (29)
Bangladesh (2) South Korea (14)
Switzerland (20)
Mexico (25)
Thailand (20)
UAE (8)
India (15) Malaysia (7)
Saudi Arabia (51) Singapore (2)
Indonesia (7)
Brazil (40)
Australia (20)
South Africa (6)
All other countries (514)
(#) = Total estimated productivity benefit ($B)
Sources: Faethm and Pearson; BCG analysis.
Note: Estimated productivity benefits represent the sum of benefits across national, state, and local governments, with GenAI implemented at scale.
Productivity benefits are calculated for public sector professions using best-case-scenario benchmarks across similar professions.
BOSTON CONSULTING GROUP 2
Five Opportunities for Government Policy and Programs
Expanded capabilities for policy development
Given that the nature of opportunities varies across govern-
ment, we have found it helpful to consider the use cases Stewardship of how a country serves its citizens requires
for GenAI from the perspective of senior executives in five great skill and involves the continuous optimization of policy
different types of government functions. The range of and programs. GenAI provides new tools for improving the
opportunities and the types of changes that may occur at capabilities needed in policy functions, including problem
each level are outlined below, followed by more detailed identification and analysis, policy research and synthesis,
discussion and use case examples. policy and program design, consultation and stakeholder
engagement, and implementation and evaluation.
• Policy and Programs. To better understand current
public policy issues and challenges, as well as the cur- • Enhanced policy and program design. GenAI tools
rent state and root causes, and to design more effective make it possible for policy analysts to rapidly synthe-
policy options, interventions, and programs; optimize size and analyze immense volumes of structured and
policy settings; and strengthen deliberative processes. unstructured data from diverse sources and formats
and across jurisdictions. These may include past policy
• Service Delivery and Operations. To improve the documents, speeches, reports and reviews, white papers,
quality and accessibility of public services to citizens academic studies, journals, articles, budget papers, da-
and businesses, improve efficiency of operations, reduce tabases and datasets, and other research inputs. GenAI
risks, and continuously optimize allocation of resources tools can help to summarize issues, identify options,
to meet policy goals and objectives. present pros and cons, distill and categorize key points,
and draft policy briefs and summaries. This allows policy
• Support Functions. To improve the efficiency of sup- professionals to provide more timely and responsive
port functions, shared services, and corporate services; advice, cover more ground in their research, strengthen
reduce overheads; and improve staff experience. the breadth and depth of evidence that underpins policy
advice, and devote more time to critical thinking about
• Regulators. To improve integrity and compliance with more complex policy challenges—adapting and tailoring
regulations, reduce the cost of monitoring and oversight, to the local context, communication, and messaging,
reduce risks, streamline administration, and make it implementation considerations, and more intensive
easier for citizens, businesses, and other stakeholders to stakeholder management.
comply and meet their obligations.
• Richer public consultation and participatory gov-
• Central Agencies. To develop, implement, and optimize ernance. GenAI can ingest hundreds or even thousands
whole-of-government strategies, priorities, policies, and of submissions received as part of public consultation
standards, and to optimize funding and resource alloca- processes, summarize and categorize recommendations
tion to achieve government objectives. and suggestions, create heatmaps to identify areas of
alignment and divergence, identify consensus views
and unique perspectives, and co-pilot the drafting of
summaries and recommendations. This enables govern-
ments to gather and process a broader range of inputs,
to capture a more extensive and comprehensive range
of views from citizens and stakeholders, and to increase
transparency and engagement in policy formulation
and co-creation. It makes it possible for a wider range of
people to engage in and provide perspectives beyond the
traditional lobby groups and associations with resources
and funding to do so. It enables constituent and stake-
holder engagement to occur more frequently, iteratively,
and interactively, supporting richer and more substan-
tive conversations to occur.
3 GENERATIVE AI FOR THE PUBLIC SECTOR: FROM OPPORTUNITIES TO VALUE
Highlighting the insights from public consultation
BCG developed a simulation called publicconsultation.ai
which uses ChatGPT to analyze and synthesize responses
from written submissions to a public consultation process.
The tool was able to ingest the content of the submissions
and generate a bullet-point summary of recommendations,
along with a matrix showing which stakeholders supported
which ideas. It also generated a list of “outlier” suggestions
not reflected in the summary. The entire process took only
hours versus what would have taken days or weeks, even
for a more complex issue involving hundreds of submis-
sions.
GenAI can enable ordinary citizens who are not typically
engaged in these consultative processes to contribute
more easily by directly involving them in simulated and
natural conversations that draw out their perspectives.
These tools also enable policymakers to synthesize the
views of a much larger and more diverse group of people
than they otherwise could. This promotes greater inclusive-
ness and transparency and improves the quality and rich-
ness of participation in policymaking processes.
BOSTON CONSULTING GROUP 4
• More responsive implementation. Policy profession- Support Functions
als can use GenAI tools to more rapidly translate new Improved internal workings of government
and updated policy into operational changes. For exam-
ple, when a policy is updated, GenAI tools could be used GenAI presents a significant opportunity to enhance the
to generate and update conforming policy and program internal administration of public sector organizations. For
guidelines; generate computer code and implement the Chief Finance Officers, Chief Information Officers, Chief
necessary changes in IT systems; rework operational Legal and Risk Advisors, Chief HR Officers, and heads of
manuals, procedures, and protocols for customer ser- other corporate and support functions, these tools can
vice; and revise government websites—all at once. automate and augment many existing tasks and activities.
Generative AI can streamline procurement, enhance em-
ployee engagement, facilitate better learning and develop-
Service Delivery and Operations ment outcomes, and optimize budgeting and forecasting.
Enhanced service delivery outcomes
• Enhanced learning and development. GenAI can
Many service delivery agencies have long used innovative create customized curricula in line with long-term
technologies, such as virtual assistants and robotic process departmental goals and the personal learning objec-
automation, to provide public services in the most efficient tives of public servants quickly and at scale. It can help
and effective way to meet the needs of citizens and busi- identify thematic and cross-cutting development needs
nesses. GenAI now offers agencies the chance to go further for the workforce based on performance reviews, provide
by implementing new tools for optimizing operations and personalized learning recommendations, tutor people
for designing digital services with more accessible interfac- based on their individual learning styles, and serve as a
es, cross-agency interoperability, and personalized features. thought partner to break down complex problems. L&D
functions can also provide staff with access to training
• Improved customer experience. GenAI can be used and tools translated into multiple languages.
to analyze voice recordings or speech-to-text transcripts
from contact centers to better understand the call • Rapid code development. In the IT function, Ge-
demand, and then make strategic interventions such nAI-enabled co-pilot coding tools can write software
as improved communication and information to reduce code in multiple programming languages. Several
demand. One example is to identify the most frequently published findings from controlled experiments already
asked questions, typical issues, or complaints based on show significant increases in code quality and productiv-
all the calls and inquiries received, and then determine ity of more than 30%, along with other benefits such as
how services can be better designed to avoid this service employee satisfaction and retention. For governments,
failure demand. For example, the Taiwanese government GenAI could be a breakthrough in tackling the mounting
has explored options to better understand citizen pain level of technical debt and legacy system replacements
points and prioritize improvements which will have the needed. It may also be able to assist with the mod-
greatest impact on customer satisfaction. ernization and migration of many undocumented and
out-of-date IT systems—and help bridge the talent gap
• 24/7 accessible services. GenAI-enabled assistants can for high-demand technologist skillsets by improving the
communicate by voice or text across multiple languages productivity of existing staff and serving as a learning
and can be used to provide 24/7 support to citizens from accelerator for new staff, especially in older, niche, and
anywhere, including regional and remote communities. legacy programming languages.
The Indian government, for example, is exploring op-
tions to leverage a GenAI-enabled assistant to help citi- • Improved recruitment processes. With appropriate
zens access policy information. The solution will support safeguards in place, public sector agencies could use
voice memo input and allow citizens to access services GenAI capabilities to adopt more proactive talent acqui-
in multiple languages from anywhere at any time. sition strategies; match potential candidates to the most
relevant openings; conduct and summarize interviews
• More personalized services. The capability for GenAI with potential candidates through interactive conversa-
to pull together data from multiple sources can help tion and testing; and streamline recruitment documen-
government agencies understand the unique context of tation and preparation of employment agreements and
a person, their family, and their “story” across multiple contracts.
contexts. The enhanced understanding of a customer’s
context can be used to generate personalized commu-
nications, such as follow-up emails, without any human
interaction or drafting required. As the technology ma-
tures, governments could leverage GenAI assistants to
provide tailored advice in a multitude of contexts, includ-
ing answering questions on existing complex policies
such as tax, pensions, benefits, visas, and immigration.
5 GENERATIVE AI FOR THE PUBLIC SECTOR: FROM OPPORTUNITIES TO VALUE
Cutting months off the wait time for services
Many governments have backlogs of individuals seeking
disability services. In some countries, citizens expect to
wait more than a year to access disability support ser-
vices—creating high levels of stress for participants and
caregivers and negatively impacting participants’ long-term
functional outcomes.
According to one estimate, GenAI could reduce application
processing times by up to 90% and free-up staff capacity to
engage with residents. GenAI applications in this case
could include ingesting and processing candidate qualifica-
tion criteria; referring customers to other relevant govern-
ment services; providing assistance to walk-through an
application process and pre-fill applications; facilitating
approval and analysis; and generating customized service
plans for case managers to coordinate with applicants.
GenAI could be used to streamline similar government
application processes, such as licenses and permits, build-
ing applications, environmental permits, passports and
visas, grant and rebate applications, and many more.
BOSTON CONSULTING GROUP 6
Regulators Central Agencies
Streamlined regulation development, compliance, Accelerating whole-of-government priorities
and reporting
For the heads of central agencies, such as finance depart-
GenAI presents an exciting opportunity for heads of regula- ments, treasury departments, and cabinet or executive
tory bodies to streamline regulations and compliance mon- offices, GenAI offers a unique opportunity to accelerate the
itoring processes. Regulators can use the tools to analyze a delivery of whole-of-government strategic priorities. Out-
broad range of data to identify trends, patterns, and anom- comes could include:
alies that might be difficult to identify otherwise, and use
the analysis to target compliance and enforcement activi- • Whole-of-government strategies and policies.
ties where the greatest risk exposures are. Central agencies could use GenAI to assist in the aggre-
gation and synthesis of diverse policies and strategies
• Compliance monitoring and detection. Using across government and ensure there is a consistent
rules-based logic and GenAI’s capability to assess large narrative and strong alignment with overall government
amounts of data, governments can expand and optimize objectives and priorities. Officials might also be able to
their oversight and monitoring of compliance. For exam- use GenAI to draft, review, and summarize complex top-
ple, an environmental conservation agency might utilize ics for government consideration and synthesize com-
GenAI to monitor industrial emissions data in real time mentary and input from across government agencies.
and juxtapose it with air quality regulations. The plat-
form could autonomously identify offenders and initiate • Improved communication. Using GenAI, governments
enforcement measures, contributing to the preservation will be able to communicate policies and budgets more
of clean air standards and safeguarding public health. Fi- effectively to citizens. With GenAI, the process could be-
nancial regulators can use similar approaches to analyze come much easier. The tools can synthesize information
transaction, market, trade, and other data and identify from a variety of sources and develop draft descriptions
potential instances of insider trading. and summaries. For example, they can use GenAI to
prepare simpler and more accessible communications
• Streamlining regulations. One use case could be to in text, audio, video, infographics, and interactive media
create simulations of how draft regulations might affect formats. They can support richer, two-way communi-
different constituents and industries. Another oppor- cations with citizens to answer questions and queries;
tunity would be to identify unknown inconsistencies, tailor information for specific stakeholder groups, such
contradictions, gaps or duplication in existing laws and as industries, regional areas, and families; and instantly
legislation, or proposed new legislation. translate material into multiple languages. People can
access information in the language and format of their
choice.
7 GENERATIVE AI FOR THE PUBLIC SECTOR: FROM OPPORTUNITIES TO VALUE
Streamlining procurement at the Department
of Defense
The US Department of Defense is prototyping and testing
a GenAI-powered contract-writing tool called Acqbot. The
tool is designed to assist procurement officers with writing
contracts and the end-to-end lifecycle management of
contracts.
The tool helps them define problem statements, draft
requirements, and prepare end-to-end solicitation docu-
ments. It supports them in building statements of work and
acts as co-pilot to draft and iterate contract agreements.
Finally, it supports quality control, such as by checking for
inclusion of regulation citations.
Procurement officers provide input and remain in the loop
at every step. Future developments still under consider-
ation may include additional functionality to support evalu-
ation of responses from suppliers.
BOSTON CONSULTING GROUP 8
Getting Started with GenAI in the Public Sector Establish guardrails.
Deployed responsibly, GenAI has the potential to deliver
As this article highlights, the rapid advances of GenAI significant value, but it also comes with significant risk.
technology present exciting opportunities for the public This being said, the biggest risk may be if governments fail
sector. We have identified five key success factors that will to adopt GenAI quickly enough or at all. To balance risks
enable government leaders to move beyond the initial and opportunities, government leaders should be seeking
small experiments, identify where to begin their GenAI to establish Responsible AI frameworks which build the
journey, and build their capability to unlock the opportuni- necessary guardrails and create the confidence needed to
ties of GenAI at scale. drive innovation. Recent BCG research shows that when
leaders are actively engaged in Responsible AI, companies
Prioritize the high-value use cases. achieve 58% more business benefits, are 17% more pre-
Explore the landscape of opportunities, but quickly focus pared for investing in Responsible AI, and are 22% more
on a few “golden” use cases—the opportunities with the prepared for emerging AI regulations.
greatest potential value or benefits for citizens and govern-
ment. Develop pilot projects for these use cases, monitor- Encourage innovation.
ing the outcomes carefully. We call this phase “experiment The benefits of GenAI will emerge as knowledge workers
and learn.” It enables governments to build-up valuable explore the technology first-hand. Leadership encourage-
first-hand experience and skills. ment will make a difference. Government leaders must
create a permission space for public sector employees to
Capture and propagate early learning. experiment within reasonable boundaries. One way to do
Governments are still early in experimenting with GenAI this is to demonstrate their own hands-on engagement,
and learning how it can improve public services. One of the working closely with one or two pilots themselves.
most effective things senior leaders can do is establish
mechanisms that encourage the cross-pollination of ideas Public sector adoption of GenAI is still in the early stages,
and learning. To do this, leaders should establish a central but it needs to accelerate. The efficiency and citizen bene-
team whose responsibility is to track and share success fits of an AI-powered government are no longer hypotheti-
stories and synthesize common lessons and hurdles that cal. Private sector implementations of GenAI-augmented
government agencies encounter. As GenAI maturity in- products and services, AI bots and assistants, and even
creases, this role will shift towards removing the hurdles company-specific proprietary trained models show that the
and delivering central enablers. value is real and achievable. Some public sector leaders
around the world are starting to experiment with use
Invest in enablers. cases, but there is a disproportionate focus on the down-
At some point, every government will roll out these technol- side risks. More senior leadership focus and investment
ogies more broadly: refining them, scaling them, and opti- are needed to scale this and maximize the upside poten-
mizing beyond use cases and pilots. Begin preparing for tial. The time to act and capture the immense government
this at the start. Invest in workforce skills, design gover- and citizen benefits of this revolutionary technology is now.
nance mechanisms, and put in place key processes and
technology choices. Build the technology and data capabili-
ties required to enable more sophisticated GenAI use cases.
9 GENERATIVE AI FOR THE PUBLIC SECTOR: FROM OPPORTUNITIES TO VALUE
About the Authors
Miguel Carrasco is a Managing Director and Senior Cyma Habib is a Principal based in the London office.
Partner based in the Sydney office. You may contact him You may contact her at [email protected].
at [email protected].
Frank Felden is a Managing Director and Senior Richard Sargeant is a Managing Director and Partner
Partner based in the Cologne office. You may contact based in the London office. You may contact him at
him at [email protected]. [email protected].
Steven Mills is a Managing Director and Partner Simon Shenton is a Managing Director and Partner
based in the Washington, DC office. You may contact based in the London office. You may contact him at
him at [email protected]. [email protected].
Jamie Ingram is an Engineering Director based Gareth Dando is a Vice President, Data Science
in the London office. You may contact him at based in the Sydney office. You may contact him at
[email protected]. [email protected].
For Further Contact Acknowledgements
If you would like to discuss this report, please contact The authors would like to thank Francisca Browne and
the authors. Brad Goff for their contributions to this report.
BOSTON CONSULTING GROUP 10
Boston Consulting Group partners with leaders in business BCG at [email protected]. To find the latest BCG con-
and society to tackle their most important challenges and tent and register to receive e-alerts on this topic or others,
capture their greatest opportunities. BCG was the pioneer please visit bcg.com. Follow Boston Consulting Group on
in business strategy when it was founded in 1963. Today, Facebook and X (Formerly Twitter).
we work closely with clients to embrace a transformational
approach aimed at benefiting all stakeholders—empowering © Boston Consulting Group 2023. All rights reserved. 12/23
organizations to grow, build sustainable competitive
advantage, and drive positive societal impact.
Our diverse, global teams bring deep industry and functional
expertise and a range of perspectives that question the
status quo and spark change. BCG delivers solutions
through leading-edge management consulting, technology
and design, and corporate and digital ventures. We work
in a uniquely collaborative model across the firm and
throughout all levels of the client organization, fueled by
the goal of helping our clients thrive and enabling them
to make the world a better place.
For information or permission to reprint, please contact
11 GENERATIVE AI FOR THE PUBLIC SECTOR: FROM OPPORTUNITIES TO VALUE
bcg.com
12 GENERATIVE AI FOR THE PUBLIC SECTOR: FROM OPPORTUNITIES TO VALUE |
122 | bcg | future-of-consumer-intelligence-series-2-layout.pdf | future of consumer intelligence #2
Put On Your
Running Shoes
June 2024
By Lara Koslow, Jean Lee, Verena Damovsky, and Kenechukwu Obinwa
Put On Your Running Shoes
Focus groups, surveys, segmentations, customer journey maps, cus-
tomer data platforms, data appends, and the practice over the past
five to eight years of naming C-level customer officers—companies
spend a lot of time and energy on the customer.
Going for Gold In our survey and throughout this article, we define AI as
including both predictive AI (tools that automatically per-
In our first article in this series, “Filling the Empty Chair,” form tasks, learn, and adjust to new information, such as
we detailed how executives and their organizations go to Google’s search or basic customer service bots) and gener-
great lengths to delve into the mind and behaviors of their ative AI (tools that create new, logical content, such as text
consumers because consumer-centric companies deliver or images, on the basis of input information, as seen in
far better results: uplifts of approximately 10% to 20% in tools like ChatGPT and DALL-E).
revenue growth, 15% to 25% in cost savings, and 20% to
40% in brand advocacy, based on BCG experience.
An Endurance Run
The amount of available consumer data has exploded in
recent years, but companies have struggled to harness this Despite all the time and effort organizations invest to
abundance of data to create value. With the rise of AI and deeply know the customer, only 38% of industry leaders
GenAI, however, we believe that organizations have a report that they have achieved a holistic view of their
unique opportunity over the next three to five years to consumers, and many say that they face challenges related
better address these challenges and transform their to obtaining, consolidating, and integrating data. (See
consumer intelligence capability into an enterprise-wide Exhibit 1.) Others point to the lack of suitable talent, silos
ecosystem that creates a more holistic view of the in the organization that hinder coordination, the absence
consumer than ever before. In our first article, we outlined of a clear and unified strategy, and insufficient funding for
how companies can use AI and GenAI to help build this data, research, talent, and technology.
ecosystem—and what it will take to get started.
As a follow-up to that work, from March 7 to March 24, New Shoes
2024, we surveyed C-suite-level and senior executives who
work at companies that have 1,000 or more employees Faced with these challenges, up to 53% of industry leaders
and who are responsible for consumer insights, pricing, across all countries feel that AI will be a game changer for
sales and marketing strategy, and/or product develop- consumer intelligence—elevating their companies’ view of
ment. Our objective was to better understand these lead- their customers to new heights. (See Exhibit 2.)
ers’ views on consumer intelligence and the impact of AI
and GenAI on consumer intelligence, and to identify what
they are doing with AI and GenAI today. The study focused
on five countries—China, France, Germany, the UK, and
the US—but it also included supplemental data (from
Australia, Canada, India, and Singapore) to provide a more
global view. We supplemented this research with expert
interviews with current and former chief marketing offi-
cers at large, US-based consumer goods companies.
1 PUT ON YOUR RUNNING SHOES
Exhibit 1 - Issues with Obtaining and Integrating Data Pose Major
Challenge to Companies
Top challenges faced by respondents in gathering meaningful consumer insights (%)
47%UK
38
33
32
29
Only
26
38% 23
22
Average
of leaders feel
37% they have a 16
holistic view of
US China
consumers1
9
32%
France
Hard to Hard to Fragmented Lack of Lack of No clear Insufficient No clear Not a
30%Germany
obtain integrate data talent coordinationstrategy funding owner priority
Source: AI in Consumer Intelligence, BCG survey (n = 541), March 2024.
1Percentage represents respondents who “strongly agree” with the statement “My company has a holistic, 360-degree view of our consumers” across
all surveyed countries.
Exhibit 2 - Across Assessed Geographies, Executives Firmly Believe that AI
Will Be Pivotal in Transforming Consumer Intelligence
Perceived effect of AI on consumer intelligence, by country (%)
AI will be a game changer1
53 46 39 29 23
Up to
53%
France UK US China Germany
of leaders feel that AI will
be a game changer
for consumer AI will be a big step up or a game changer2
intelligence
78 78 78 82 63
France UK US China Germany
Source: AI in Consumer Intelligence, BCG survey (n = 541), March 2024.
1Percentages in this row represent respondents who selected “Game changer—will elevate consumer intelligence to a whole new level” in response
to the question “At its full potential, how much do you feel AI can enhance consumer intelligence overall?”
2Percentages in this row represent respondents who selected either “Game changer—will elevate consumer intelligence to a whole new level” or “A
lot—a big step change from other options” in response to the question given in footnote 1.
BOSTON CONSULTING GROUP 2
Fueling this sense of optimism about AI and its likely The Medalists
impact on consumer intelligence are expectations of signif-
icant productivity gains (held by varying percentages of Who are the AI leaders in consumer intelligence globally?
leaders in different regions, from 36% in Germany to 62% Invited to self-report on this question, 58% of respondents
in China) and improved output (from 33% in Australia to in France said that they are leaders in AI adoption, versus
55% in China). These positive expectations prevail despite 44% in the US, and 21% in China. (See Exhibit 5.)
some misgivings about data privacy issues (cited by a low
of 25% of leaders in China, and a high of 44% in Australia) Not surprisingly, across countries, large corporations are
and high initial investment costs (cited by just 23% of adopting more AI tools than medium or small companies.
respondents in India but by 36% of respondents in France). Overall, 57% of large companies possess custom GenAI
(See Exhibit 3.) tools and 56% have deployed GenAI add-ins. In contrast,
smaller companies tend to rely disproportionately on
Looking forward, many executives believe that AI will ad- public GenAI tools such as ChatGPT (44%). (See Exhibit 6.)
dress a number of their critical consumer intelligence
issues. (See Exhibit 4.) They hope that AI will facilitate the Looking ahead, we expect this gap in AI adoption between
integration of cross-channel insights, increase the focus on larger and smaller companies to widen. Although more
implications and strategy, provide a better view of likely than 50%+ of companies report that they plan to spend
consumer behavior, expedite product prototyping to enable from 6% to 8% of their revenue on AI this year—and al-
faster market launches, and provide much more rigorous though 57% of smaller companies plan to do so—the
regular predictive modeling and scenario testing to inform absolute investment volumes of smaller companies lag the
better decisions. $250 million-plus that larger companies plan to spend
every year by a factor of 100. (See Exhibit 7.) As such,
smaller firms will likely need to leverage vendors to access
economies of scale in AI development – and may have less
bespoke solutions.
Exhibit 3 - Leaders Expect the Biggest Gains to Involve Productivity and
Output Quality, While Data Privacy and Potential for Misinformation Are
Key Concerns
Key positive drivers (%) Non-aided associations with AI1 Key negative drivers (%)
China Agility Australia
36–62 Productivity gains Trusted Reliable 25–44 D coa nt ca e p rnri svacy
Germany China
Quality
Efficiency
China
Unique
33–55 Improved output Excitement Potential
Australia Smart Convenient Sensitive
France
High initial
Accuracy Productive Risk 23–36 investment costs
Helpful Advanced India
France
Learning/upskilling
26–48 Job loss
opportunity Empowerment Precision
Canada Easy
Innovation
US
France 21–50 Cost-saving potential CSe rcu ere ative Progress 22–40 T mh ir se ina ft o o rmf ation
Canada
Australia Fast
Useful
Size of word represents number of responses
xxx
(non-aided), minimum of 10 responses
Sources: AI in Consumer Intelligence, BCG survey (n = 541), March 2024; BCG analysis.
Note: The paired country names associated each key positive or negative driver are the countries with the highest and lowest percentages on that
driver. Thus, for example, for “Productivity gains” in the “Key positive drivers (%)” column, the percentage for China is 62% and the percentage for
Germany is 36%.
1Multiselect responses to “What are the main drivers of positive/negative sentiment toward AI at your company?” The word cloud presents non-aided
associations with AI.
3 PUT ON YOUR RUNNING SHOES
Exhibit 4 - Executives Have Big Hopes for AI and How It Can Address
Current Challenges
Current state without AI Expected future state with AI
1 Isolated data silos Integrated cross-channel insights 1
2 Considerable time spent on data integration and analysis Increased focus on implications and strategic planning 2
3 Limited understanding of data Comprehensive consumer behavior predictions 3
Extensive time and resources spent on product testing
4 AI-enabled prototyping, leading to faster market readiness 4
and development
5 Reactive postmortems Predictive modeling and scenario testing 5
Sources: Future of Insights study 2023; client interviews.
Exhibit 5 - Leaders in France Are More Likely to Self-Report as ‘Ahead’
Than Leaders in Other Countries on Various AI Engagement Metrics
Respondents within country samples that say they are leaders across AI engagement metrics (%)
AI sentiment AI ambition AI adoption AI investments
60 58%
50
44%
40%
40
32%
30
21%
20
10
0
France UK Germany China US
Sources: AI in Consumer Intelligence, BCG survey (n = 541), March 2024; BCG analysis.
Note: Results track the percentage of respondents, by country, who chose the most positive response on a five-point scale to the question “How
would you rate your company’s overall AI sentiment, AI ambition, AI adoption, AI investment?”
BOSTON CONSULTING GROUP 4
Exhibit 6 - Most Large Companies Have Custom GenAI Tools and GenAI
Add-ins Whereas Smaller Companies Disproportionately Rely on Public
GenAI Tools
Public GenAI tools GenAI add-ins Custom GenAI tools
GPT-3.5 Zoom AI Companion ChatGPT Enterprise
DALL-E Slack AI Microsoft co-pilot
58%
Large average adoption
corporations rate across assessed
tools and practices
58% 56% 57%
39%
average adoption
Small rate across assessed
corporations tools and practices
44% 21% 37%
Sources: AI in Consumer Intelligence, BCG survey (n = 541), March 2024; BCG analysis.
Note: “Large companies” are defined as corporations whose annual revenue exceeds $1 billion; “small companies” are defined as corporations
whose have annual revenue is less than $50 million.
Exhibit 7 - Most Respondents Plan to Invest from 6% to 10% of Revenue in AI
Although small companies lead on planned investment share, they lag in absolute volumes by a factor of 100
Planned AI investment as share of revenue, by company size (%)1
57
$250M–
Yearly spending on AI
50 49 $450M for large companies
40
35
$10M– Yearly spending on AI
23 $25M for medium companies
13
10
7 7 $200k– Yearly spending on AI
for small companies
2 3 $400k
Note: Low N (34) for small
companies by definition
<1% of revenue 1–5% of revenue 6–10% of revenue >10% of revenue
Large company Medium company Small company
Sources: AI in Consumer Intelligence, BCG survey (n = 541), March 2024; BCG analysis.
Note: “Large companies” are defined as corporations whose annual revenue exceed $ 1billion; “medium companies” are defined as corporations
whose annual revenue is from $50 million to $1 billion; “small companies” are defined as corporations whose annual revenue is less than $50
million. The conversion rate of US dollars is taken as €1 = $1.08. The sum of the revenue categories for each company size does not equal100%
because responses of “Do not know” are excluded.
1Estimates are calculated by multiplying proposed investment range as a percentage of revenue by absolute revenue and obtaining median values
from output ranges.
5 PUT ON YOUR RUNNING SHOES
The Starting Blocks A/B testing (37%) and web analytics (35%) appear to be
the types of use cases most frequently rolled out across
In looking specifically at AI in consumer intelligence, we industries. (See Exhibit 9.)
found that industry leaders from across the countries we
surveyed overwhelmingly agree that companies can heavily Even so, priority use cases vary across industries. For exam-
leverage AI across different use cases. (See Exhibit 8.) The ple, A/B testing is a priority in the retail and consumer goods
percentages of leaders agreeing with this proposition range sector, but loyalty programs are most popular in the manu-
from 96% in the UK to 100% in France. Although 37% of facturing and industrial sector (38%). (See Exhibit 10.)
respondents in the US and 70% in France report that their
company has already rolled out one or more such use
cases, about half of those that adopted at least one use
case (from a low of 17% in Australia to a high of 34% in
France) have already rolled out three or more AI use cases,
an indication the early stage of business development that
the technology is in.
Exhibit 8 - Business Leaders Universally Feel That Companies Can
Meaningfully Leverage AI for Consumer Intelligence
AI in consumer intelligence use cases across all surveyed countries
France Australia
100% 17%
UK
UK
96%
96–100% 37–70% 37% 17–34%
France
34%
France
70%
Almost all industry leaders ...and 37% to 70% report having …but only half of them
agree that companies can rolled out at least one have already rolled out three or
meaningfully leverage AI AI-enabled use case for more such AI use cases in
for consumer intelligence use cases1 consumer intelligence2 consumer intelligence2
Sources: AI in consumer intelligence, BCG survey (n = 541), March 2024; BCG analysis.
1Represents respondents who selected at least one consumer intelligence use case in response to the question “Where do you feel AI can be
meaningfully leveraged?”
2Represents the number of consumer intelligence use cases that respondents classified as ”rolled out” in response to the question “Where does your
company use AI?”
BOSTON CONSULTING GROUP 6
Exhibit 9 - A/B Testing and Web Analytics Are the Most Widespread Use
Cases to Date, But Many Other Types of Pilots Are Underway
AI implementation status across select use cases in consumer intelligence (%)
A/B testing at scale 37 34 29
Web analytics 35 38 27
Customer experience 34 34 32
CRM database 32 33 35
Customer-centric design 31 33 36
Loyalty program 31 29 39
Competitor research 30 38 32
Market sizing 28 39 32
Demand/trend forecasting 27 40 34
Personalized marketing 27 40 33
Customer journey 26 43 31
Dynamic/promotional pricing 26 41 32
Customer segmentation 25 35 40
Consumer sentiment 25 31 45
Rolled out Pilot Not yet
Sources: AI in Consumer Intelligence, BCG survey (n = 541), March 2024; BCG analysis.
Note: Classification of consumer intelligence use cases as “rolled out,” “pilot/testing,” or “planning to”/“wish we did”/“no AI” is based on responses
to the question “Where does your company use AI?” CRM = customer relationship management.
Exhibit 10 - Priority Use Cases Differ by Industry
Loyalty programs are a priority in manufacturing and industrial goods, while A/B testing is a
priority in retail and consumer goods
Respondents who have rolled out select use cases (%)
26%
21%
38
35 34
32 31
27 29 28 28 26 26 26 26 24 24 25 25 23
19 20 21 19
16 17 15
9 9 9
Loyalty Demand/ CRM Customer- Competitor Customer Customer
program trend database centric research experience segmentation
forecasting design
Dynamic/ Web Market Personalized A/B testing Customer Consumer
promotional analytics sizing marketing at scale journey sentiment
pricing
Manufacturing and industrial goods Retail and consumer goods Priority use case
Sources: AI in Consumer Intelligence, BCG survey (n = 541), March 2024; BCG analysis.
Note: Classification of consumer intelligence use cases as “rolled out” is based on responses to the question “Where does your company use AI?”
split by industries of interest. CRM = customer relationship management.
7 PUT ON YOUR RUNNING SHOES
Turning Pro Given the speed of AI’s development, the strategic ap-
proach for all companies, regardless of their maturity,
The maturity of AI varies by market, as do the needs of should be to start small, demonstrate impact, and adapt
each market to further evolve. For example, in relatively their organization and technical capabilities iteratively. Of
mature markets such as France and the US, leaders see course, the key challenge will be to keep pace. As one
strengthening AI-capable talent and better understanding former fashion and retail CEO told us, “Today, the speed of
available tools as most critical tasks to harness the power the technology is much faster than the speed of the organi-
of AI. (See Exhibit 11.) But in less mature markets such as zations using it.” But for those who put on their running
China, respondents tend to identify instead the need for shoes as AI continues to fuel the evolution of the consum-
strong leadership guidance. er intelligence capability, it may usher in an era of compre-
hensive consumer understanding that will benefit us all, as
consumers ourselves.
Exhibit 11 - The Most Critical Issues in More AI-Advanced Markets and in
Less AI-Advanced Markets Differ
High
France 49% Better (that is, AI-capable) talent
US 45% Understanding of emerging AI tools and landscape
AI adoption
UK 49% Understanding of emerging AI tools and landscape
Dedicated AI leader/team
China 58%
Germany 50% More leadership attention/prioritization
Low
Sources: AI in consumer intelligence, BCG survey (n=541), March 2024; BCG analysis.
Note: Top-ranked response, by country, to the question “What do you think is most critical for your company to harness the full power of AI with
regard to consumer intelligence?”
BOSTON CONSULTING GROUP 8
About the Authors
Lara Koslow is a managing director and senior partner in Jean Lee is a is a partner and director in the firm’s Seattle
the firm’s Miami office, with a focus on growth strategy, office, focused on customer growth and strategy across a
marketing, branding, consumer insight, and commercial/ range of consumer sectors. She has deep expertise in the
go-to-market topics across industries—in particular, travel travel and tourism sector and served as North America
and tourism, consumer, retail, and automotive. She is the leader for BCG’s Center for Customer Insight for many
global leader of BCG’s customer demand and innovation years. You may contact her by email at [email protected].
business, which includes BCG’s Center for Customer In-
sight (CCI) and AI Lighthouse platform. She is the former
global leader of BCG’s Center for Customer Insight, which
she ran for over 8 years, and the former global leader of
the marketing business. You may contact her by email at
[email protected].
Verena Damovsky is a project leader in the firm’s Lagos Kenechukwu Obinwa is a consultant in BCG’s Lagos
office, with experience working with clients across Europe, office, with experience working with clients in various
Africa and Asia across different industries and functions. industries and functions across Africa and Asia. He has
She has expertise in customer segmentation with a focus experience developing customer engagement strategies,
on the social impact space. You may contact her by email particularly in the social impact space. You may contact
at [email protected]. him by email at [email protected].
For Further Contact Acknowledgments
If you would like to discuss this report, please contact the Thank you to the following for their support crafting this
authors. publication:
• Silvia Mazzuchelli, BCG Senior Advisor
• Amanda Helming, Former Chief Marketing Officer, UNFI
• Marissa Jarratt, Executive Vice President, Chief Market-
ing and Sustainability Officer, 7-Eleven
9 PUT ON YOUR RUNNING SHOES
Boston Consulting Group BCG’s Center for Customer Insight (CCI)
Boston Consulting Group partners with leaders in business The Boston Consulting Group’s Center for Customer
and society to tackle their most important challenges and Insight (CCI) applies a unique, integrated approach that
capture their greatest opportunities. BCG was the pioneer combines quantitative and qualitative consumer research
in business strategy when it was founded in 1963. Today, with a deep understanding of business strategy and
we work closely with clients to embrace a transformational competitive dynamics. The center works closely with BCG’s
approach aimed at benefiting all stakeholders— various practices to translate its insights into actionable
empowering organizations to grow, build sustainable strategies that lead to tangible economic impact for our
competitive advantage, and drive positive societal impact. clients. In the course of its work, the center has amassed
a rich set of proprietary data on consumers from around
Our diverse, global teams bring deep industry and functional the world, in both emerging and developed markets. The
expertise and a range of perspectives that question the CCI is sponsored by BCG’s Marketing, Sales & Pricing and
status quo and spark change. BCG delivers solutions Global Advantage practices. For more information, please
through leading-edge management consulting, technology visit Center for Customer Insight (https://www.bcg.com/
and design, and corporate and digital ventures. We work capabilities/marketing-sales/center-customer-insight).
in a uniquely collaborative model across the firm and
throughout all levels of the client organization, fueled by the
goal of helping our clients thrive and enabling them to make About the Research
the world a better place. The Future of Consumer Intelligence series features
data from an online survey of current C-level and senior
executives at global/national companies with 1,000+
employees, conducted in March 2024. Survey respondents
self-reported a responsibility for consumer (insights), pricing,
sales and marketing strategy, and/or product development.
The survey was produced by the authors and BCG’s Center
for Customer Insight (CCI), in partnership with coding and
sampling provider Dynata, the world’s largest first-party data
and insights platform. We supplemented this research with
expert interviews with current and former chief marketing
officers at large, US-based consumer goods companies. The
goal of the research was to understand industry leaders’
perspective on consumer intelligence in light of a rapidly
evolving AI and GenAI landscape. A team composed of BCG
consultants and experts from CCI completes the survey
analytics.
© Boston Consulting Group 2024. All rights reserved. 6/24
For information or permission to reprint, please contact BCG at [email protected]. To find the latest BCG content
and register to receive e-alerts on this topic or others, please visit bcg.com. Follow Boston Consulting Group on Facebook
and X (formerly known as Twitter).
bcg.com
11 PUT ON YOUR RUNNING SHOES |
123 | bcg | Executive-Guide-to-Investing-in-GenAI.pdf | Executive Guide
to Investing in
Generative AI
A framework to embrace
this disruptive and rapidly
evolving technology
By: Pranay Ahlawat, Drake Watten, Matt Kropp, and Vlad Lukic.
2 Generative AI Economics
Content
Introduction 03
1. Why do I want to use Generative AI and what business value will it create? 04
2. What type of Generative AI should I be using and is that a future-proof decision? 05
3. Do I really need to build this capability, or should I wait and buy? 08
4. What is the Total Cost of Ownership (TCO) equation and how will it evolve over time? 09
Putting the four questions together—how to win in Generative AI 10
Conclusion 11
About the authors 12
3 Generative AI Economics
Introduction
Generative AI’s potential to disrupt industries, revolutionize customer relationships, and
change the way knowledge work gets done has made it a strategic imperative across
companies. The question is “how” to embrace this technology in a way that is right for your
organization and can profitably scale. Based on our research and interviews with executives,
there are four key obstacles that executives must navigate today.
Fragmented approach to Inadequate understanding of
generative AI without a clear generative AI technology trends
business case and implications
Limited understanding of the business Not knowing what technology
value and current feasibility of use cases, as approach to take and misunderstanding
well as unclear prioritization of use cases its risks
Not being clear on when to Not fully understanding
“build” versus “buy” the at-scale economics of
generative AI
Prioritizing the wrong use cases to build,
when buying an out-of-the-box solution Not factoring the second-order costs
might be a better fit of adopting generative AI at scale and
building business cases
This article delves into four pivotal questions that executives should ask to better manage
these challenges and succeed in their generative AI journeys.
4 Generative AI Economics
1
Why do I want to use Generative
AI and what business value will it create?
This is a simple question, but one that gets overlooked in the rush of companies to invest
in generative AI. In our research, many companies have started multiple initiatives without
fully understanding the entire spectrum of generative AI use cases, sequencing or prioritizing
them, and estimating their business impact. We observe many large companies taking a
fragmented and uncoordinated approach to generative AI, where different Business Units
(BUs) or Lines of Business (LOBs) are driving use cases in the absence of an enterprise-
wide strategy. Furthermore, many companies are still discovering the capabilities of current
platforms and tools and are failing to understand their limitations or challenges, partly due to
the noisy hype surrounding generative AI today.
To cut through the noise, companies must keep three things in mind.
• Not all use cases are created equal. Organizations need to start with a clear strategy
based on the business value of the use case, as opposed to taking a scattered approach
across multiple pilots. Some companies are at risk of being disrupted and need to double
down on generative AI to build new offerings and value propositions. In other cases, more
horizontal use cases may be less strategic, and companies should thoughtfully prioritize
these considering taking into consideration levels of investment, ROI, and risk, amongst
other things.
• It’s early days and Generative AI’s capabilities are still evolving. Despite its enormous
potential, generative AI has technical limitations and is still maturing for all use cases,
particularly for modalities outside text (such as video), and more industry-focused ones
(for example, healthcare, manufacturing). In our research, many customer pilots are
running into day-two operational challenges (for example, cybersecurity, machine learning
operations, governance), and some are unable to get past pilots and user acceptance
testing phases because of results that do not meet the bar. To fully understand the benefits
and challenges, organizations should experiment and run pilots within their context, and
with their data securely to ensure the outcomes are salient for the objectives they have set.
• There are risks and associated challenges with generative AI which may drive second-order
costs. These risks are well understood and include data hallucinations, bias, cybersecurity,
and copyright challenges, amongst others. Many of the companies we interviewed are only
starting to grasp these challenges as they scale.
To make thoughtful Generative AI investments that create the most impact, it is important to
think through your strategy and technological approach, understand the alternatives, and be
equipped to pivot quickly.
5 Generative AI Economics
2
What type of Generative AI should I be using
and is that a future-proof decision?
Generative AI is not a one-size-fits-all technology, as it can be deployed in four different ways.
The options, which come with their unique tradeoffs and costs, range from using public APIs
that are turnkey and available for immediate usage, to building and training a custom model
from scratch. What companies choose ought to be driven by six strategic considerations.
Speed to market Data privacy and regulation needs
Companies that are seeking feature parity Customers in industries such as healthcare or
with competitors or who operate in industries financial services will need to evaluate the data
that are getting disrupted (for example, residency or privacy controls requirements
traditional chatbot platforms or enterprise and decide if those could be met by compliant
search vendors) have a much greater need to cloud solutions, or whether they warrant
rapidly respond. building custom models of a generative AI
architecture built for on-premises on-perm.
Customization requirements Latency and performance
requirements
While out-of-box functionalities from Speed of response is a key consideration
commercial model vendors such as OpenAI, for certain real-time use cases, such as live
Anthropic, Amazon Titan, etc., might work for summarization, automated trading, etc. Out-
multiple use cases, more complex or domain- of-box capabilities from commercial model
specialized use cases might benefit from fine- vendors might not be sufficient. The ability
tuning commercial or open models. to optimize the size, and architecture of the
foundation model for latency and speed of
response may be critical considerations for
such use cases.
Volume and scale of intended Operating model implications
use cases
The overall volume and nature of workloads— Availability of talent is a key factor in
whether it is spiky and seasonal or consistently determining how a company implements
high volume—need to be considered, as they generative AI. Building a custom model is not
impact the mode of deployment. a viable option for many enterprises due to
the steep requirements for data science and
machine learning talent. Additionally, the
maturity of a company’s broader operating
model, such as its data pipeline management
as well as machine learning operational
processes is another key factor.
Based on the above considerations, customers can choose one of four generative AI
deployment options.
• Usep ublic APIs (for example, Google Translate, AI21 Summarize, Amazon, etc., or AI21
Paraphrase) for simple, and standard tasks that do not require any customization.
• Consume Models-as-a-Service from model vendors such as OpenAI, AWS Titan, and
Anthropic. Most customers leverage these models with no tuning or customization via
techniques like RAG (Retrieval-Augmented Generation).
• Useo pen models such as Falcon or Sable Diffusion which can be used as-is or
customized for domain-specific use cases.
• Create and train a custom model from scratch (such as Bloomberg GPT) for specialized
tasks and maximum performance control.
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
3202
© thgirypoC
6 Generative AI Economics
Generative AI deployment options
Public APIs and Services Models-as-a-Service (MaaS) Open Models Proprietary Models
Leverage public services e.g.,
Build Gen AI applications on Tune or use publicly available Build custom models
Google Translate, Amazon
Description managed models e.g., or open-source models as is from scratch e.g.,
Polly, Azure Cognitive
Cohere, Anthropic, OpenAI e.g., Falcon, Stable Diffusion Bloomberg GPT, Palm
Services
etc.
Ability to Limited – some vendors High degree of customization Maximum ability to optimize
customize No ability to customize provide ability to tune models – can be used as-is, tuned or every aspect of stack
fully re-trained
Deployment Cloud with option to deploy
Cloud multi-tenant Hybrid - cloud or on-premise Hybrid - cloud or on-premise
modality instance in private VPC
<$1M to $7M $3M to $30M $10M to $50M+
Setup Cost1 <$0.1M
Multitenant cloud Private VPC2 Private VPC2 On premise Private VPC On premise
<$1M to $6M $5M to $1M $7M to $1M
Run (1-year)1 <$0.1M
Varies by usage and deployment Private VPC2 On premise Private VPC2 On premise
Use-cases requiring full control
Use-cases requiring contextu- Specialized use-cases requiring
over the model for data privacy, Simple standard tasks, al learning with limited to no domain specific or organization
Use-cases customization, latency and perfor-
like language translation customization, e.g., personal- specific customizations, e.g.,
mance optimizations, e.g., defense ized marketing campaigns drug discovery acceleration industry use cases
1: Assumed GPT-3 equivalent foundation model with 175B parameters and 300B training tokens. Inferencing assumed for 10B tokens and fine tuning for 20B tokens.
Training, tuning and inference costs calculated as API price per token * number of tokens or (FLOPs per token for GPT-3 / FLOPs per second for each machine) * pricing
per second for each machine. Total costs estimated based on BCG's 70:20:10 framework – 70% effort in AI implementations spent on talent & org change, 20% on
technology and 10% on algorithms. 2. Virtual Private Cloud
7 Generative AI Economics
These decisions come with their own tradeoffs. Models-as-a-Service can speed up time to
market, but it can get expensive at scale and lead to vendor lock-in. On the other hand,
leveraging open-source models or building a model from scratch comes with high up-front
costs and demanding talent needs.
What complicates this decision even more is the fact that the technology and economics
of generative AI are rapidly evolving. To make a future-proof decision, it is imperative for
customers to understand where the proverbial “puck” is going and the technology trends
influencing that direction. Our research found four key trends:
Bigger is not necessarily better: it Model customization and
is important to find the right model tuning continue to get easier
at the right cost/performance
The capabilities of foundation models are
Several optimization techniques, such as
quickly converging. Smaller, well-trained
Low-Rank Adaptation (LoRA), pruning
models are now delivering comparable
and quantization, are lowering the costs of
performance to larger models at a fraction of
training, tuning, and inferencing models. This
the cost, challenging the notion that bigger
is lowering the technology barriers and time
is always better. For example, Chinchilla, a
to market to build custom models, making
70-billion parameter model, delivers the
these architectures more viable.
same performance as Gopher, a 280-billion
parameter model, but at a 75% lower cost.
Custom silicon and accelerators Overall generative AI stack
are set to create a step change in is maturing and driving
performance/costs democratization
Innovations in accelerators and silicon are Tools and open-source libraries such as
improving price/performance ratios. Custom LangChain and HuggingFace Transformers
silicon from Amazon (AWS Trainium and are reducing the barriers to entry to building
AWS Inferentia) are delivering with up to 30- complex generative AI applications.
50% in price performance, and driving more
integrated stacks for machine learning.
As companies evaluate different deployment options, it is crucial to assess the technical,
financial, and organizational impacts of each option. They must think of vendor lock-in risks,
as well as technology risks of making wrong platform bets. In our research, we found that
organizations are experimenting with multiple options at the same time, depending on use
cases. They are using Models-as-a-Service (MaaS) to start, focusing on lower-risk use cases
(for example, summarizations, knowledge management), while also dabbling in open-source
or custom models for specialized use cases like anti-money laundering and fraud detection.
The value and risk of generative AI must be understood within an organization’s context.
8 Generative AI Economics
3
Do I really need to build this capability,
or should I wait and buy?
Beyond choosing the right technology, companies need to ask if building a solution is
warranted at all. Given the rapidly changing technology landscape and where we are
in the generative AI adoption cycle, it is important for organizations to be realistic about
what can be achieved, and be clear about what use cases to build and which ones to buy.
In our experience, there are multiple types of scenarios where investing aggressively and
building solutions early is advisable. The first scenario is one in which generative AI
incrementally enhances the core offering or business model (for example, chat-based
e-commerce, built-in generative AI tools in creative software such as Adobe Firefly, and AI-
enabled workflow tools for enterprise software such as Salesforce Einstein-GPT). A second
scenario is when generative AI creates a novel offering or opens new markets, such as drug
discovery in biopharma. Yet another scenario might be if your industry is facing disruption
(for example, information services, legacy NLP (Natural Language Processing), and chatbot
platforms), moving rapidly and adopting generative AI might be the only option.
On the other end of the spectrum, buying out-of-the-box solutions might be a better option
for more standardized use cases where solutions already exist (for example, coding or writing
assistants), or horizontal use cases where platform solutions will likely emerge (for example,
customer support). Waiting, continuing to evaluate, and keeping one’s options open can be a
good strategic decision as it increases flexibility, keeps a company focused, and reduces the
technological risk of making wrong investments.
There are other vertical and efficiency-driven use cases (for example, network automation in
telcos), where companies need to consider their business value, assess their current ability
to execute (such as lack of specialized machine learning talent), and estimate costs to build.
While prioritizing these use cases can yield significant value, it remains crucial for companies
to continuously assess the rapidly evolving tech landscape and explore commercial solutions
that not only expedite the advancement of their use case but also effectively manage risks
and minimize investments.
Regardless of approach, as a general principle, companies should have a very high bar for
building proprietary foundation models, as this will only make economic sense at a very large
scale. This option is more suited for hyperscalers, given extremely specialized talent and large
upfront capital are required to develop and train state-of-the-art foundation models.
9 Generative AI Economics
4
What is the Total Cost of Ownership (TCO)
equation and how will it evolve over time?
There are two types of costs of generative AI: primary and secondary. Our research suggests
that most companies consistently underestimate the latter.
Primary costs, which are better understood, include fixed set-up costs (for example, hardware,
data curation, costs of engineering, training, and tuning) and variable costs (such as
consumption costs for APIs, and inference).
The secondary costs of generative AI include a lot of hidden, and hard-to-estimate costs,
ranging from maintenance costs (for example, repaying technical debt, re-training, incremental
testing), risk management, organizational change management, as well as legal costs.
The total costs can vary significantly between deployment options (as illustrated in the exhibit
above). Public API options have the lowest setup and run costs (<$0.1M setup costs for API
integration). Maas (Models as a Service) options are generally used without customization
and typically have lower setup costs (<$1M) including setting up data pipelines and vector
databases, but the run costs at scale can get expensive, reaching up to $6M in some cases.
Building proprietary models can be prohibitively expensive—sometimes costing upwards of
$50M depending on the type of model, data curation, and architectural investments—but
they provide the most architectural control over run costs.
Like any large-scale digital transformation, the biggest overlooked secondary cost is
organizational change management. BCG’s 10:20:70 framework explains the relative
investments needed to implement AI at scale: 10% on algorithms, 20% on technology, and
70% on organizational change. The last cost of organizational change is not only the biggest
investment, but also the hardest to estimate and fraught with the risks of internal change.
Compliance and legal costs are also increasing due to the new risks introduced by generative
AI. In our research, some organizations are experiencing a 25% increase in legal, testing, and
compliance costs for their generative AI use cases. We do expect these costs to improve over
time, as companies go up the experience and adoption curve.
Understanding the TCO for generative AI requires modeling different scenarios that weigh
expected benefits and both primary and secondary costs—as well as how each cost may
evolve. Moreover, the evolving technological innovations underscore the need to be flexible in
decision-making. Executives may be better off making decisions with a medium-term horizon
and being prepared to pivot quickly.
10 Generative AI Economics
Putting the four questions together–how to
win with generative AI
There are four strategic control points you need to consider to be on the winning
side of generative AI:
Focus on economic viability Look ahead to solve the data and
at scale when making core day-two operational equation
technology choices
No technology can scale unless you solve
day-two operational challenges. In our
Building and running generative AI
research, customers are starting to run
applications can be very expensive.
into operational challenges across the
Customers must evaluate the cost-efficiency
stack: data quality and data pipeline
and performance expectations of different
management, machine learning operations,
model and deployment options vis-a-vis the
model observability and governance,
use case they are solving. Factors such as
etc. Companies need to plan for and get
data privacy, customization requirements,
ahead of these challenges to deploy AI
latency, and volume must be considered to
at scale.
make the right technology bet at the optimal
price/performance ratio. Moreover, once Assess feasibility and impact before
companies understand their priority use
investing for scale, and be clear
cases, they have to invest in scale to avoid
the cycles of repeated demos. Generative AI about when to buy versus build
can only be strategic if it is done at the right
Like any transformative technology early
cost at scale.
in the adoption cycle, generative AI is rife
with hype, partly driven by news media, and
Invest early in closing marketing messages from some software
the talent gaps and tooling vendors. Companies must
take these claims with a grain of salt, and
instead choose to experiment aggressively to
Talent remains a top-of-mind concern and a understand generative AI’s capabilities and
major stumbling block for most companies impact in their contexts. They must continue
that are implementing generative AI. to make investments in talent and dive
Companies must invest in up-skilling their deep into the right use cases, while carefully
current workforce and hiring to close the assessing “build versus buy” decisions in
talent gap over time. Consider other options these early days. Sometimes, saying “no”
like acquisitions or partnerships to bridge and waiting to get the timing right is more
the gap in the near term. strategic.
11 Generative AI Economics
Conclusion
Generative AI is here to stay and needs to be a strategic priority for businesses
today. There is no doubt that it will disrupt industries and will create a step change in
productivity. However, it is also complex, rapidly evolving, and can be very expensive at
scale. While it is imperative to move quickly, it is equally important to carefully prioritize
where to place your bets.
To avoid missteps and future-proof their generative AI investments, companies
ought to start with a clear strategy and understanding of use cases. They need to
assess if generative AI is a sufficiently mature technology for their use cases and
test within their organizational contexts. They also need to fully understand the
technology options, trends, and tradeoffs to make the right technology bet, from
Models-as-a-Service to building custom models. They need to know when to say
“no” and wait, and fully think through when to build versus buy. Most importantly,
companies must understand the economics and full scope of costs—primary and
secondary—to realistically estimate ROI. And if they do decide to take action, they
must be prepared to solve basic but fundamental problems: data processes, talent,
and day-two operations. Making strategic and focused generative AI bets—without
rushing hastily—can save your company from costly missteps at the very least, or, at
best, dramatically accelerate your company’s position in the market.
12 Generative AI Economics
About the authors
Pranay Ahlawat is a partner and Matt Kropp is managing director
associate director in the firm’s and senior partner in
Washington D.C. office. firm’s San Francisco office.
You may contact him at You may contact him at
[email protected]. [email protected]
Drake Watten is managing director Vlad Lukic is managing director
and partner in firm’s and senior partner in firm’s
San Francisco office Boston office.
You may contact him at You may contact him at
[email protected] [email protected]
For further contact Acknowledgments
If you would like to discuss this report, The authors thank the following for
please contact the authors. their contributions to the development
of this report: Aakash Joshi and Sai
Masipeddi from BCG and Phil Le-
Brun, Archana Vemulapalli, Tom
Adams, Ahmad Tawil, Susane Seitinger,
Adil Soofi, Ritesh Vajariya, Kamran
Khan, Joe Senerchia, Nitin Nagarkatte,
Salman Taherian, Jake Burns, Ross
Richards and Priya Arora from AWS.
13 Generative AI Economics
Boston Consulting Group
Boston Consulting Group partners with
leaders in business and society to tackle
their most important challenges and
capture their greatest opportunities.
BCG was the pioneer in business
strategy when it was founded in 1963.
Today, we work closely with clients to
embrace a transformational approach
aimed at benefiting all stakeholders–
empowering organizations to grow, build
sustainable competitive advantage,
and drive positive societal impact.
Our diverse, global teams bring deep
industry and functional expertise and
a range of perspectives that question
the status quo and spark change. BCG
delivers solutions through leading-edge
management consulting, technology and
design, and corporate and digital ventures.
We work in a uniquely collaborative model
across the firm and throughout all levels of
the client organization, fueled by the goal
of helping our clients thrive and enabling
them to make the world a better place.
© Boston Consulting Group 2023. All rights reserved. 7/23
For information or permission to reprint, please contact BCG at [email protected]. To find the latest BCG content
and register to receive e-alerts on this topic or others, please visit bcg.com. Follow Boston Consulting Group on Facebook
and Twitter. |
124 | bcg | BCG-Wheres-the-Value-in-AI.pdf | Where’s the Value
in AI?
October 2024
By Nicolas de Bellefonds, Tauseef Charanya, Marc Roman Franke, Jessica Apotheker,
Patrick Forth, Michael Grebe, Amanda Luther, Romain de Laubier, Vladimir Lukic,
Mary Martin, Clemens Nopp, and Joe Sassine
Contents
01 Who’s Getting Results from AI 13 The Playbook for Winning
and Why? with AI
• A Steep Curve • Overcoming Tough Challenges
• What Leaders Do Differently • The Capabilities Required for Success
• Incumbents Reap Value • Jump-Starting Your Journey to AI Value
• How Two Companies Applied the
Playbook for Success
05 The Surprising Sources of Value
from AI
17 Appendix
• In the Core
• Sector Matters
20 Acknowledgments
Who’s Getting Results from AI
and Why?
A
fter all the hype over artificial intelligence (AI), the Consider these examples of the value being created by AI,
value is hard to find. CEOs have authorized invest- including generative AI (GenAI), from companies in three
ments, hired talent, and launched pilots—but only different sectors. A financial institution is committed to
26% of companies have advanced beyond the proof-of- achieving $1 billion in productivity improvements, in addi-
concept stage to generate value. This report yields import- tion to enhanced risk outcomes and better client and
ant insights into what AI leaders are doing to drive real employee experiences, by 2030. A biopharma company is
value from the technology, where others fall short, where chasing $1 billion in value potential (revenues and costs)
the value is coming from, how individual sectors are per- by 2027. A major automaker expects to cut its cost of
forming, and how companies can change their own goods sold by up to 2% and accelerate new product devel-
AI trajectories. opment time by 30%.
BOSTON CONSULTING GROUP 1
These results are typical of the value that leaders across A Steep Curve
industries are achieving by building digital capabilities to a
level at which they can implement AI programs at scale. Building AI capabilities is a complex challenge. Our latest
BCG’s latest research into AI adoption, a continuation of research, involving more than 1,000 companies worldwide,
our studies into digital transformation and AI maturity, shows that only 4% have developed cutting-edge AI capa-
found that of the 98% of companies that are at least exper- bilities across functions and are using them to consistently
imenting with AI, only 26% have developed the necessary generate substantial value. (See Exhibit 1.) Another 22%
capabilities to move beyond proofs of concept and begin have an AI strategy and advanced capabilities and are
extracting value. (For more on how we define AI and our starting to generate value. We call these companies lead-
research methodology, see the Appendix.) And only 4% are ers. The remaining 74% have yet to show tangible value
at the forefront of AI innovation, systematically building from their use of AI.
cutting-edge AI capabilities and scaling them across the
organization. These categorical distinctions are important because lead-
ers far outperform the others. Over the past three years,
Here’s our latest look at who the top 26% of companies are leaders’ revenue growth has been 50% greater than the
and how they are generating superior value from AI. The overall average. Their total shareholder returns are 60%
two chapters of the report that follow look at where com- higher, and they gain 40% higher returns on invested capital.
panies are extracting value and what you need to do to These companies also excel on nonfinancial factors, such as
move your company up the AI maturity curve. patents filed and employee satisfaction, and they are in pole
position to benefit as AI platforms and tools mature.
Exhibit 1 - Leaders Have Built the Capabilities Needed to Implement AI at
Scale, Reaping Diverse Benefits over Less Mature Companies
Maturity stage (% of companies)
AI stagnating AI emerging AI scaling AI future-built Value achieved
25 49 22 4
50%
Revenue
higher revenue growth
(3-year average)
Total 60%
shareholder
return higher 3-year TSR
40%
Returns
higher RoIC
(3-year average)
0 25 50 75 100
Are taking minimal Have developed Have developed an AI Are at the forefront 1.9x
or no AI action, foundational strategy and advanced of AI innovation, Innovation
lack foundational capabilities and capabilities, and are systematically more patents
capabilities, and are started initial scaling them effectively building cutting-edge
not generating value experimentation while starting to AI capabilities across
but are struggling generate value functions and 1.4x
Employee
to scale and consistently
generate value generating satisfaction better overall
substantial value Glassdoor indicator
Source: BCG Build for the Future 2024 Global Study (merged with DAI).
Note: “Leaders” include AI future-built and AI scaling companies; “less mature” or “other” companies” include AI stagnating and AI emerging
companies. RoIC = return on invested capital; TSR = total shareholder return.
2 WHERE’S THE VALUE IN AI?
What Leaders Do Differently They invest strategically in a few high-priority oppor-
tunities to scale and maximize AI’s value. Data on AI
Leaders have six differentiating characteristics. adoption shows that leaders pursue, on average, only
about half as many opportunities as their less advanced
They focus on the core business processes as well as peers. Leaders focus on the most promising initiatives,
support functions. A common misconception is that AI’s and they expect more than twice the RoI in 2024 that other
value lies mainly in streamlining operations and reducing companies do. In addition, leaders successfully scale
costs in support functions. In fact, its greatest value lies in more than twice as many AI products and services across
core business processes, where leaders are generating 62% of their organizations.
the value. Leveraging AI in both core business and support
functions gives these companies competitive advantage. They integrate AI in efforts both to lower costs and to
generate revenue. Almost 45% of leaders integrate AI in
They are more ambitious. Leaders’ expectations for their cost transformation efforts across functions (com-
revenue growth from AI by 2027 are 60% higher than those pared with only 10% of nonleaders). And more than a third
of other companies, and they expect to reduce costs by of leaders focus on revenue generation from AI, compared
almost 50% more. Three-quarters of the most forward- with only a quarter of other companies. (See Exhibit 3.)
looking companies focus on company-level innovation core “We have a program under which every business unit is
to the business. In contrast, only 10% of other companies required to submit three to five projects each year—and
do so—and if they leverage AI at all, it is mainly for produc- since 2020, they have all had to focus on AI,” said the
tivity. Leaders look beyond pure productivity plays and back enterprise product director of an alternative energy compa-
their ambitions with investment in AI and workforce en- ny. “These projects need to demonstrate how they would
ablement, doubling down on several aspects of AI, relative improve the company either through cost savings, in-
to their peers. (See Exhibit 2.) They make twice the invest- creased operational efficiency, or revenue generation.”
ment in digital, twice the people allocation, and twice the
number of AI solutions scaled.
Exhibit 2 - Compared with Their Peers, Leaders Are Allocating More of Their
Budget and Resources to Digital and AI Capabilities in 2024
Budget People Innovation
2.0x 1.6x 2.0x 1.9x 1.6x 1.1x 2.2x
19.2%
18.2%
12.3% 13.8%
10.1% 9.1% 9.6% 8.9% 6.3 5.5 8.2%
5.0% 4.6% 5.1% months months
Revenue share 2024 increase Share of FTEs Share of digital Share of FTEs Time to market Share of AI/GenAI
invested in in AI/GenAI dedicated to FTEs dedicated to be upskilled for new digital products scaled
digital and AI investments digital and to AI/GenAI in AI/GenAI and AI products across the
vs 2023 AI work roles today organization
AI stagnating or AI emerging AI scaling or AI future-built
Source: BCG Build for the Future 2024 Global Study (merged with DAI).
Note: FTEs = full-time equivalent employees.
BOSTON CONSULTING GROUP 3
Exhibit 3 - Leaders Integrate AI with Broader Cost Transformation Efforts
and Have a Greater Focus on Revenue
Integration of AI with broader cost transformation efforts (%) AI investment split between cost reduction and revenue growth (%)
4
15 Greater
26 27
More AI 36 revenue
43 43 integration focus
27 21
55
20
40
53 52
47 44
30
17
AI stagnating AI emerging AI leaders AI stagnating AI emerging AI leaders
Without GenAI Exploratory Multiple functions Revenue Equal Cost
Source: BCG Build for the Future 2024 Global Study (merged with DAI).
They direct their efforts more toward people and But AI’s impact extends to all industries. For example, a
processes than toward technology and algorithms. leading automaker used GenAI to accelerate tender docu-
Leaders follow the rule of putting 10% of their resources ment drafting and adjustments by 50% while improving
into algorithms, 20% into technology and data, and 70% document quality and consistency. GenAI also increased
into people and processes, which our data shows are the the automaker’s speed in analyzing competing offers (by
key capabilities underpinning success. 50%) and reduced the time necessary to search knowledge
assets (by 50% to 75%).
They have moved quickly to focus on GenAI. Leaders
use both predictive AI and GenAI, and they are faster in Leaders are blazing the AI trail, but other companies can
adopting GenAI, which opens opportunities in content catch up if they take a page from the leaders’ playbook and
creation, qualitative reasoning, and connecting other tools focus on the areas that offer them the best opportunities
and platforms—in part because their more advanced and on the capabilities they need to build in order to capi-
capabilities facilitate putting the prerequisites (such as talize. We explore these factors in the next two chapters.
large language models) in place.
Incumbents Reap Value
Not all AI leaders are hyperscalers and digital natives, com-
panies that include AI as part of their product or services
offering. More than half of the top-performing 26%, includ-
ing the ones described at the beginning of this chapter, are
traditional incumbents that have strengthened their capa-
bilities and are using them to build differentiated competi-
tive advantage. The sectors with the biggest percentages of
AI leaders tend to be those that were among the first to
experience digital disruption a decade and half ago and got
the earliest start on building digital capabilities. They
include fintech (49% are leaders), software (46%), and
banking (35%).
4 WHERE’S THE VALUE IN AI?
The Surprising Sources of Value
from AI
L
eading companies are dreaming big. By 2027, the top The common narrative for AI involves support functions—
26% of companies in our survey of AI maturity expect HR, IT, legal, and the like—where automating relatively
to achieve 45% more value via cost reduction and 60% low-level and repetitive functions creates significant value.
more value via revenue growth than other firms. Even in But the companies that are generating the most value are
2024, leaders expect to realize more than twice the RoI not only deploying productivity plays in support functions
from AI initiatives than other companies do, resulting in a but also focusing on reshaping their core business process-
5% reduction in addressable operational expenses and a es and inventing new revenue streams. They are achieving
5% increase in addressable revenues. results from AI across a wide range of functions, from R&D
to operations and from sales and marketing to customer
service. Because they have built the necessary capabilities,
they can more readily identify, pilot, and scale up value-
creating use cases. For example, one chemicals company
expects to create more than $500 million in value from an
end-to-end transformation that will implement AI across
operations, site services, and procurement.
BOSTON CONSULTING GROUP 5
In the Core Sales and marketing, for example, is fast emerging as a
major source of AI value in such sectors as software (31%
Overall, the companies in our survey derive 62% of the of AI value generated), travel and tourism (31%), media
value they obtain from AI and generative AI in core busi- (26%), and telecommunications (25%). Specific roles and
ness functions, including operations (23%), sales and the scale of impact differ by industry, but AI offers compa-
marketing (20%), and R&D (13%). Support functions gener- nies a near-term opportunity to reshape the sales function
ate 38% of the value, with customer service (12%), IT (7%,) with next-best action recommendations, talk tracks, and
and procurement (7%) leading the way. basic workflow automation. In the medium term, AI and
GenAI will enable real-time assisted selling and autono-
In some sectors the spread between core and support is even mous selling via digital sales avatars, with limited human
wider. (See Exhibit 4.) Software, media, fintech, insurance, involvement. Such automation will permit human staff to
telecommunications, and biopharma generate 70% to 90% of focus on strategic and relationship selling, while virtual
their AI-related value in core business processes. Although we assistants cover more transactional tasks. As predictive
found wide variation among sectors, the overall results are smart selling becomes the norm, traditional silos dividing
consistent—even most of the sectors in the bottom quartile marketing, sales, and pricing will dissolve. Our experience
generate 40% to 60% of AI value in core processes. indicates that resulting increases in customer lifetime
value and go-to-market efficiencies could almost double
profit margins.
Sector Matters
Companies in different sectors also benefit from identify-
ing the domains in which AI can produce the most value.
Our research shows that they vary widely by industry. (See
“AI in Insurance and Biopharma.”)
Exhibit 4 - To Realize Value from AI, Companies Focus on Core Business
Processes, with Sector-Specific Variability
Where companies are achieving or see business value
Global average: 62%
Sectors
Core business processes (%) Support functions (%)
Software 94 6
Media 87 13
Fintech 85 15
Insurance 77 23
Telecommunications 71 29
Biopharma 70 30
Banking 68 32
Airlines 65 35
Retail 63 37
Automotive 62 38
Transport and logistics 61 39
Medtech 59 41
Consumer products 57 43
Oil and gas 49 51
Chemicals 48 52
Machines and automation 40 60
Power, utilities, and renewables 21 79
Core business function Support function
Source: BCG Build for the Future 2024 Global Study (merged with DAI).
6 WHERE’S THE VALUE IN AI?
Leaders are not only deploying
productivity plays but reshaping
core business processes and
inventing new revenue streams.
The impact on marketing will be equally profound and will In one current instance, a global pharmaceuticals compa-
encompass four key processes: ny is using AI to accelerate its drug discovery capabilities.
The initial vision was to build, test, and validate an AI
• Insight to Innovation. Automated data collection prototype with chemists to quantify the value impact in the
and analysis will speed identification of market oppor- discovery workflow. The company assessed the potential of
tunities and increase marketers’ ability to develop new state-of-the-art models to find new preclinical candidates
product design. faster, and then it built its own machine learning algorithm
to rapidly screen over 1 billion drug compounds and a
• Concept to Creation. Workflows will accelerate asset genetic algorithm to power a lead optimization pipeline for
creation and feedback loops, seamlessly adapting, local- molecular chemists. The project generated value of $100
izing, and disseminating content. million a year through faster launches, including a 25%
reduction in cycle time. The company expanded its library
• Campaign Setup and Execution. Hyper-segmentation of molecules by 100 times, increasing the visibility of novel
and real-time execution that responds to trends and compounds to its researchers.
feedback will speed campaign creation and automatical-
ly track progress against key objectives. Customer service is already a significant source of AI-
generated value in insurance (24% of the value created)
• Marketer Productivity. Marketers will spend less time and banking (18%). Companies are using AI to boost pro-
on time-consuming, repetitive, administrative tasks and ductivity, reducing the need for multiskilled frontline teams
more time on strategic decision making. and redesigned agent journeys. We are seeing near-term
increases of 30% to 40% in productivity and a profit-and-
For example, a leading North American telco is already loss impact of 10% to 20% for the function.
using AI to analyze call recordings to identify opportunities
for cost savings and higher customer satisfaction. The Ambitions run much bigger. Leading companies expect to
company has reduced call center interaction time by 20% realize long-term increases in productivity of up to 60%.
and cut call transfers to live agents by 25%. AI-powered The impact of integrating AI into customer service process-
chatbots now handle 30% of calls, and the telco expects to es will reverberate throughout the value chain. Customer
reduce total costs in the relevant business unit by 25%. service functions will be able to preempt issues and
self-heal by fixing problems before customers detect them,
Predictably, AI is having a big impact in R&D in research- and they will enable customers to resolve their own issues
intensive sectors such as biopharma (27% of value creat- through self-help. If the customer still needs human assis-
ed), medtech (19%), and automotive (29%, in an industry tance, AI will support the agent’s response with augmented
undergoing a major transition to software-driven vehicles). capabilities such as optimizing the conversation in real-
A medtech company vice president told us, “Generative AI time by considering the customer’s needs in context and
has allowed us to generate images for training purposes making offers where relevant.
that mimic real diseases that humans can have. We start-
ed deep diving into generating thousands of images that A leading international bank needed to modernize its
aren’t coming from patients but are being generated by the customer management system to improve service quality,
generative model mimicking real-life cases. Our predictive reduce operational costs, and enhance revenue generation.
AI model improved accuracy by 4% to 5% because of this It turned to GenAI to reshape both customer interactions
generative AI approach.” and backend processes, including deploying GenAI for chat
support, enhancing agent efficiency, improving service
In the R&D function of the future, we expect individual-, quality, and increasing conversion rates. It also integrated
team-, and company-level changes to improve concept GenAI into its APIs and apps for smooth and scalable opera-
R&D, product development and industrialization, and prod- tions. Results included a reduction of almost 20% in interac-
uct evolution. AI will accelerate and automate each step by tion time between customers and agents; a drop of 4 min-
shortening iteration loops, democratizing access to exper- utes in average service time while retaining similar levels of
tise across teams and organizations, fast-tracking explora- customer satisfaction, an increase of 28 points in conversion
tion of new concepts, simulating product designs, and rates, and a doubling in breadth of products sold.
forecasting procurement orders, among other changes.
8 WHERE’S THE VALUE IN AI?
Consumer products and retail companies are making big The critical challenge for companies is to identify the key
gains with AI-driven personalization (19% of the value use cases within each function. For example, 43% of insur-
created for the former and 22% for the latter). About 30% ance companies leverage AI in scoring, fraud assessment,
of consumer companies in our survey have adopted AI for and triage while 42% of biopharma companies use AI in
personalized marketing (among other functions) and are systematic protein and drug molecule generation (at least
seeing productivity gains of about 30% from such activities for pilots and proofs of concept). The highest value use
as marketing content generation, marketing mix and RoI cases typically involve a mix of predictive AI and GenAI.
optimization, and data-driven digital marketing. As a re-
sult, leaders are doubling down in other areas at two to Although companies in each sector may be generating
four times the rate of slower movers, applying AI to genera- the greatest value from use cases in one or two do-
tive product design, and manufacturing optimization. mains, most are still experimenting—and obtaining mea-
surable results in up to half a dozen domains in the core
Within each process or function, it’s critical to define spe- business, including customer relations and experience,
cific use cases and associated business value. In most content production and management, and product man-
sectors, more than half of GenAI’s value potential lies in agement. In more than a few sectors—including oil and
two or three functional domains. In insurance, 55% of the gas, utilities, and machinery and automation—support
value lies in in policy administration, underwriting, and functions are a significant source sources of value, too.
claims management. In biopharma, 57% of the value is
found in R&D and in sales and marketing. There are many routes to value. Chapter 3 explores how
your company can efficiently find its most productive paths.
BOSTON CONSULTING GROUP 9
AI in Insurance and Biopharma
At the business process, function, and use-case level, value
creation from AI is already taking different directions in
different sectors, highlighting the importance to each com-
pany of independently identifying where its best opportuni-
ties lie. Consider the evidence that our survey gathered in
two very different sectors: insurance and biopharma.
The average AI maturity of both sectors falls in the middle
of the maturity curve, not far off the all-sector average.
Companies in both sectors generate an average of 70% or
more of AI value from core business processes and 30% or
less from support functions. But the similarities end there.
Insurance
Insurers are focusing on operations (policy administration,
underwriting, and claims management), customer service,
and marketing and sales. (See the AI factsheet for insur-
ance.) So far, the widest adoption of predictive AI at the
individual-opportunity level has occurred in the areas of
scoring, fraud assessment, and triage and policy automa-
tion. Adoption of GenAI is strongest in the use of chatbots
to resolve questions and summarize customer interactions.
In line with their overall scores, insurers’ biggest challenges
involve people and processes: improving staff AI literacy,
prioritizing opportunities over other concerns, and estab-
lishing RoI for identified opportunities. They also wrestle
with the tasks of integrating AI with existing IT systems
and of increasing the accuracy and reliability of AI models.
An Asian life and health insurance company with a strong
track record in digital transformation sought to demon-
strate the benefits that GenAI could have on its operations
by identifying and executing a couple of high-impact, high-
use cases. The insurer prioritized the possibilities on the
basis of a high-level analysis of potential impact. It select-
ed two opportunities, one in customer-service call center
operations and the other in sales and marketing. The
former achieved a 30% reduction in call center search
times and the latter a 30% to 40% reduction in marketing
and sales material creation time.
10 PUBLICATION TITLE
AI Factsheet for Insurance
Where does insurance stand on the AI maturity curve? Main challenges
Top challenges across people and processes, technology, and algorithms
Maturity stage (% of companies)
AI stagnating AI emerging AI scaling AI future-built Focus areas Key challenges
9 64 25 2 BCG’s 10-20-70 Respondents citing the challenge (%)
model
Algorithms Lack of accurate/reliable models
10%
Lack of access to high-quality data
Insurance average Difficulty integrating with existing IT systems
Global average
Technology
Difficulty ensuring security and compliance
20%
IT budgets limiting investments in AI
Insufficient platform capabilities for
at-scale testing
Insufficient AI literacy
0 25 50 75 100
Difficulty prioritizing opportunities vs other concerns
Insurance companies have emerging AI capabilities slightly ahead
of the global average
Difficulty establishing RoI on identified opportunities
Where are the value pools in my sector? People and Difficulty reimagining workflows and processes
processes
Distribution of AI value potential along functional domains (%)
70% Lack of specialized AI engineers
Core business Support Lack of available talent and skills
Claims Product HR
functions management management functions
77 15 9 23 5 Difficulty measuring predetermined KPIs
Difficulty sequencing opportunities into
Legal 4 a roadmap
Customer service Underwriting Marketing,
and policy sales, IT 6 Procurement 4 Weak governance structures to steer
administration24 16 distribution13 responsible AI
Finance 4
Difficulty identifying short- and
long-term next steps
15 25 35 45 55 65 75
Source: BCG Build for the Future 2024 Global Study (merged with DAI).
Biopharma Once again, the biggest challenges in applying the technol-
ogy relate to people and processes: prioritizing opportuni-
Biopharma tells a different story. More than half of the ties over other concerns, advancing staff AI literacy, acquir-
value in this sector comes from commercial/sales and ing available talent and skills, and establishing RoI on
marketing (30%), and R&D (27%). Biopharma companies identified opportunities. The top algorithm and technology
are using GenAI for systematic protein, drug, and biological issues involve integrating AI with existing IT systems, and
processes generation, real-time hyperpersonalized engage- maximizing the accuracy and reliability of models.
ment with health care practitioners, and personalized
outreach to patients and providers. They are using AI and
GenAI together for analyzing and documenting customer
interactions and for targeting patient identification via
biological data. (See the AI factsheet for biopharma.)
BOSTON CONSULTING GROUP 11
AI Factsheet for Biopharma
Where does biopharma stand on the AI maturity curve? Main challenges
Top challenges across people and processes, technology, and algorithms
Maturity stage (% of companies)
AI stagnating AI emerging AI scaling AI future-built Focus areas Key challenges
27 46 19 8 BCG’s 10-20-70 Respondents citing the challenge (%)
model
Algorithms Lack of accurate/reliable models
10%
Lack of access to high-quality data
Health care average Biopharma average Difficulty integrating with existing IT systems
Global average Technology
20% Difficulty ensuring security and
compliance
Insufficient platform capabilities for
at-scale testing
Difficulty prioritizing opportunities vs other concerns
0 25 50 75 100
Insufficient AI literacy
Biopharma companies have emerging AI capabilities on a
par with the global average
Lack of available talent and skills
Where are the value pools in my sector? People and Difficulty establishing RoI on identified opportunities
processes
Distribution of AI value potential along functional domains (%) Lack of leadership alignment,
70% communications, and behavior modeling
Lack of specialized AI engineers
Core business Support
Research and Finance
functions development functions Difficulty making a business case for
70 27 30 6 scaling initiatives
Lack of a clear AI case for change
IT 4
Commercial/sales Manufacturing Customer Procurement
and marketing service HR 3 Difficulty identifying short- and
long-term next steps
30 13 7 7 Legal 3
Difficulty reimagining workflows
and implementing processes
15 25 35 45 55 65 75
Source: BCG Build for the Future 2024 Global Study (merged with DAI).
12 WHERE’S THE VALUE IN AI?
The Playbook for Winning with AI
L
eading companies are well on their way to creating Meanwhile, the 70% of companies that are struggling, wait-
significant value and advantage from AI. For example, ing, planning, and experimenting have an urgent need to
a consumer products company applied GenAI to re- accelerate their efforts to overcome barriers and catch up as
duce costs by $300 million through productivity gains and their competitors improve their productivity, revenues, and
agency cost savings. A global consumer goods company customer experience. As leaders and aspiring leaders ex-
expects to generate $100 million in additional sales from a pand their AI capabilities and as GenAI models and tools
GenAI-powered virtual conversational assistant, the first in mature, less capable companies will fall farther behind.
its sector. A North American telco achieved a 10% reduc-
tion in call handling time and cut the cost of customer Here’s an AI playbook that all companies can follow.
retention by more than 30%, leading to $200 million in
annualized savings.
BOSTON CONSULTING GROUP 13
Overcoming Tough Challenges Our experience, corroborated by our new research, indi-
cates that about 70% of the challenges relate to people
Our survey highlights the most difficult challenges that and process, about 20% are technology issues, and only
companies face in implementing AI initiatives. They fall 10% involve AI algorithms (which often occupy a lot more
into four groups: organizational time and resources). (See Exhibit 5.) The
survey confirms our long-held view that when companies
• Difficulties in defining clear priority use cases with com- undertake digital or AI transformations, they need to focus
pelling returns for the anticipated investments 70% of their effort and resources on people-related capabil-
ities, 20% on technology, and 10% on algorithms. Too often,
• A host of issues related to moving from plans to action companies make the mistake of prioritizing the technical
and delivering value, such as prioritizing investments, issues over the human ones—which helps explain why many
scaling solutions across functions and businesses, over- of them do not achieve the results they are looking for.
coming resistance to adoption, and realizing the benefits
Challenges evolve over time, of course, as companies build
• People and skills issues, including building specific AI their capabilities. But while less AI-capable companies
skills and broader AI literacy focus on getting the basics right, leaders are more con-
cerned with ensuring security and compliance, implement-
• Integrating AI solutions with existing IT systems, and ing responsible AI, and resolving technical issues such as
enabling access to high-quality data guardrails for large language models, high model latency,
and run costs.
Exhibit 5 - The Biggest Challenges Relate to People and Processes, Such
as Prioritizing Opportunities and Establishing RoI
Focus areas Key challenges
BCG’s 10-20-70 model Respondents citing the challenge (%)
Algorithms Lack of accurate/reliable models 48%
10%
Lack of access to high-quality data 43%
Difficulty integrating with existing IT systems 56%
Technology IT budgets limiting investments in AI 48%
20%
Difficulty ensuring security and compliance 46%
Expensive scaling due to high model run costs 37%
Difficulty establishing RoI on identified opportunities 66%
Difficulty prioritizing opportunities vs other concerns 59%
Difficulty making a business case for scaling initiatives 56%
Difficulty realizing cost takeout/savings 54%
People and Resistance and fear that AI will impact jobs 48%
processes
Lack of a clear AI case for change 42%
70%
Difficulty measuring predetermined KPIs 38%
Lack of leadership alignment and communications 37%
Difficulty reimagining workflows and processes 37%
Insufficient AI literacy 37%
Lack of specialized AI engineers 37%
20 25 30 35 40 45 50 55 60 65 70
Source: BCG Build for the Future 2024 Global Study (merged with DAI); n = 1,000.
14 WHERE’S THE VALUE IN AI?
The Capabilities Required for Success Jump-Starting Your Journey to AI Value
We analyzed the self-reported capabilities of AI leaders After assessing the capabilities and approaches of the lead-
compared with those of other companies. This assessment ing companies, we have compiled a playbook for how any
revealed empirical evidence about the most important c |
126 | bcg | Stariway-to-GenAI-Impact-Automotive-Industry.pdf | Artificial Intelligence | Article
The Stairway to (Gen)AI Impact
in the Automotive Industry
November 2024
By Andrej Levin, Felix Stellmaszek, Alex Xie, Jonathan Nipper, Manuel Kallies, Nina Kataeva, and Tobias Schmidt
Boston Consulting Group partners with leaders
in business and society to tackle their most
important challenges and capture their greatest
opportunities. BCG was the pioneer in business
strategy when it was founded in 1963. Today,
we work closely with clients to embrace a
transformational approach aimed at benefiting all
stakeholders—empowering organizations to grow,
build sustainable competitive advantage, and
drive positive societal impact.
Our diverse, global teams bring deep industry and
functional expertise and a range of perspectives
that question the status quo and spark change.
BCG delivers solutions through leading-edge
management consulting, technology and design,
and corporate and digital ventures. We work in a
uniquely collaborative model across the firm and
throughout all levels of the client organization,
fueled by the goal of helping our clients thrive and
enabling them to make the world a better place.
© Boston Consulting Group 2024. All rights reserved.
BOSTON CONSULTING GROUP 2
Content
04 Key Takeaways
05 Setting Your Company Up to Succeed with (Gen)AI
08 Putting It All Together
11 The Authors
BOSTON CONSULTING GROUP 3
Key Takeaways
Automotive companies have moved quickly to implement (Gen)AI,
but many may be celebrating too early before achieving
measurable benefits. It’s time for a focused approach.
About two-thirds of executives in industrial goods industries (including automotive) say they are not satisfied
with their company’s progress in (Gen)AI, and about half are still expecting cost savings within the year,
according to a BCG survey.
Automotive companies can make an impact on the bottom line with three (Gen)AI value creation levers,
potentially achieving a return on investment (ROI) of 10 to 15 times in less than three years. These are new
revenue streams from an augmented direct sales approach, cost reduction through automation of more
complex tasks or services, and productivity gains through allowing teams to focus on the most critical tasks
of supply chain management or EV/software product innovation.
But to achieve the full potential of (Gen)AI, leaders must identify realistic profitability improvement goals
at the outset and use a rigorous process to prioritize value-creating use cases and establish accountability.
In business technology, generative artificial intelligence (GenAI) is often seen as a new miracle engine that
promises to put early adopters on the track to greater efficiencies, lower costs, and quick wins. But most companies
in the automotive sector find that after a year of investing in endeavors that have underdelivered compared with
expected results, the effect seems more like a collective spinning of wheels. In some cases, automotive companies
have encountered significant challenges with (Gen)AI implementations, such as a major North American
manufacturer whose chatbot mistakenly offered a vehicle for just $1 USD.
In addition, many organizations have celebrated their success too early, before realizing benefits that show up on the
bottom line. As a result, the perceived effects evaporate before they can be substantiated. According to a recent BCG
survey, 67% of executives in the industrial goods sector (including automotive) said they are not satisfied with their
company’s progress in (Gen)AI. Automotive companies should not give up too soon, however—49% of industrial
goods’ respondents (including automotive) still expect the technology to deliver cost savings in 2024. The risk is that
disappointed expectations can lead to disillusionment and lost momentum, when there are real benefits to be had.
Nevertheless, the automotive industry is particularly well-positioned to capitalize on (Gen)AI advancements.
The sector has been a leader in (Gen)AI, as seen with the progress in autonomous driving technologies over the
past decade. However, the current landscape is marked by the shift to electric vehicles (EVs), heightened global
competition, particularly from Asia, and pricing pressures from rising post-pandemic inventories. Such an
increasingly complex environment demands a deeper exploration of (Gen)AI applications. From product
development to supply chain optimization, automakers must leverage (Gen)AI not only to streamline operations
but also to secure a competitive edge.
Moreover, automotive companies have a unique opportunity to unlock further value from (Gen)AI, thanks
to the expansion of connected vehicles, which now generate vast amounts of data. This data, coupled with decades
of investment in building the necessary digital infrastructure, provides automakers with a foundation that few
industries can match. The ability to harness this information and apply (Gen)AI-driven insights across operations
positions the industry to achieve significant gains in both efficiency and revenue generation, setting a clear path
to future growth.
BOSTON CONSULTING GROUP 4
Setting Your Company Up to Succeed
with (Gen)AI
BCG’s experience with automotive clients suggests that many
businesses are failing to realize value from (Gen)AI because they
lack a structured approach and clear value focus from the outset.
Specifically, companies have treated (Gen)AI like a typical technology upgrade or a collection of pilots, with tech
teams leading the way. While this is fine for the technology side of the equation, it fails to achieve real bottom-line
impact. In fact, (Gen)AI requires an even higher level of craftsmanship than other types of transformation. It needs
a greater focus on critical enablers like process and operating-model redesign, training, and employee adoption—
factors that are often overlooked—as well as measurable P&L or balance-sheet improvement goals.
Visible support from leadership is an important success factor. (Gen)AI transformations need to be led from the
C-level, and business unit and functional leaders must be accountable for defining value- creation levers and
achieving results. This can help propel the holistic, people-driven transformation that is required, with specific EBIT
(earnings before interest and taxes) targets that are actionable and traceable.
The automotive industry desperately needs this bottom-line focus to help achieve the agility and efficiency required
to sustain a competitive advantage within a transforming sector. Many players are feeling the pressure of higher
product cost, investments for EV scaling, and at the same time, lower barriers of entry to the playing field in
general. Therefore, new sources of value creation are required—by augmenting new direct sales approaches but
also by accelerating highly complex internal processes like supply chain management or end-to-end value and
capacity planning.
BOSTON CONSULTING GROUP 5
Furthermore, there are several automotive-specific challenges that reinforce the need for leadership, people, and
process engagement in (Gen)AI initiatives. First, many OEMs are working with legacy systems, making human-in-
the-loop engagement critical to ensure that use cases and algorithms are applied correctly and effectively. Second,
legal and privacy concerns surrounding the use of vehicle data are significant, with multiple OEMs in the U.S. facing
lawsuits related to vehicle-data privacy breaches, which must be carefully managed and mitigated. Finally, the
complexity of automotive ecosystems, involving numerous stakeholders such as suppliers, dealers, consumers, and
finance organizations, adds another layer of difficulty in transitioning from pilots to scaled value. These factors
highlight the need for a coordinated, strategic approach to fully leverage (Gen)AI's potential while navigating
industry-specific challenges.
From our work on more than 350 (Gen)AI projects for clients, we have learned that the key is to target EBIT gains
from the outset, not just focus on an implementation program that companies hope will mature into a value
creator. BCG has identified a “(Gen)AI stairway” with four stages that organizations must move through smoothly
to get from (Gen)AI illusion to economic impact. (See Exhibit 1.)
Exhibit 1 | The GenAI Staircase to Success
Companies achieve
real value
MONEY MAKER
Beyond
Execution to P&L
Most companies celebrate tech
success too early Operating model
SHOWMAN change
Process redesign Process redesign
People adoption People adoption
THEORIST
Tech
ILLUSIONIST Scale enablers Scale enablers Scale enablers
GenAI GenAI GenAI GenAI
No value No value No value Value
Source: BCG analysis.
Stage 1: ILLUSIONIST
In this stage, companies across all industries embark on the (Gen)AI journey, attracted by promised efficiency gains
from tools like ChatGPT, Microsoft Copilot, or customized chatbots. We have seen examples of more than 100 use
cases where organizations were running pilots in many departments but without a clear definition of user needs
and without considering technological synergies across functions.
Unfortunately, such isolated deployments fail to generate tangible value because they lack a strategic plan for
employee training and enterprise-wide scaling, leaving the impression that (Gen)AI benefits are just an illusion.
BOSTON CONSULTING GROUP 6
Stage 2: THEORIST
Scaling beyond a few limited tools is necessary, given that meaningful value can come only from meaningful scale.
But we find that most companies that try to scale (Gen)AI throughout the enterprise struggle with low user
adoption. Often, employees feel excluded and do not receive sufficient training; as a result, they fail to incorporate
the tool effectively into their workflow and the (Gen)AI value remains out of reach.
Stage 3: SHOWMAN
As companies struggle with the complexities of (Gen)AI deployment, some realize that success requires more than
just implementing tools. Leading organizations adopt the 10/20/70 principle, referring to how companies should
apportion time and resources, recognizing that, while algorithms (10% of the effort) and technology (20%) are
essential to a (Gen)AI implementation, people and process changes (70%) require the most attention. Companies
that invest in training and change management can typically achieve an adoption rate of about 60% across the
enterprise in our estimation, compared with 30% for those that do not invest in these areas.
Organizations that have successfully undertaken holistic process and workflow changes report efficiency gains
of up to 50%. But celebrating at this stage is premature. Our follow-up with these clients reveals that, within a few
months, the initial gains tend to dissipate as employees’ new free time becomes occupied with backlog tasks
or other emerging priorities. So, while this stage is a significant milestone, it is not enough to drive lasting value.
Stage 4: MONEY MAKER
The final, essential step to achieving measurable value from (Gen)AI is to establish clear links between efficiency
gains and the P&L statement and diligently execute an EBIT-focused transformation. Companies have a variety of
challenges to address with (Gen)AI, and there are numerous potential strategies to convert efficiency gains to
actual profit.
BMW stands out as an exemplary case for achieving EBIT impact, highlighting the importance of establishing and
applying levers beyond technology. (Gen)AI use cases developed in collaboration with BCG streamlined BMW's
procurement processes by automating workflows and optimizing supplier interactions, reducing costs, and
enhancing decision-making. Examples of these applications include the automatic draft generation of RFPs
and tenders, as well as automatic offer evaluation support via (Gen)AI, which improved both speed and accuracy
in procurement operations. This comprehensive approach, spanning marketing and operations, not only improved
efficiency but also created new revenue streams, driving substantial EBIT improvements. Additionally, a first Proof
of Concept at BMW was created to develop hyper-personalized marketing using (Gen)AI, further increasing
customer engagement. BMW rigorously tracks cost savings and revenue gains from the automation processes,
ensuring that the value generated is tied back to measurable financial performance.
For any (Gen)AI strategy, organizations need to establish at an early stage how efficiency gains will affect the P&L
and then follow up with a structured end-to-end transformation. Companies should create a (Gen)AI transformation
office to safeguard value through target setting, tracking, enablement, and accountability enforcement. This allows
the business to truly complete its (Gen)AI journey and achieve the status of (Gen)AI money maker without getting
stuck along the way.
BOSTON CONSULTING GROUP 7
Putting It All Together
The stairway concept of (Gen)AI adoption emphasizes that the initial
step is the only one strictly focusing on technology.
Many companies have begun to move beyond the pure tech phase but find themselves stuck between the second
and third steps. Their experience has been one of experimentation, and while partial steps might help bring
intangible improvements, such as increasing working satisfaction for employees, these actions ultimately lack
significant EBIT impact. We call this “creative (Gen)AI chaos.”
The key to escaping the chaos and bring structure to the implementation is to define EBIT impact targets from the
beginning. The magic of targets comes from imposing a structure and discipline that help maintain focus and guide
the project to a successful result.
In one example of this structured approach, a North American automotive client focused on improving marketing
efficiency by utilizing AI to better target high-propensity and in-market vehicle buyers. From the outset, the use case
was not only technically sound but also backed by a strong business case. The company identified both cost savings
through insourcing the capability—previously reliant on external tools—and performance improvements through
more precise targeting. Additionally, it implemented new processes to ensure consistent use of the AI tool across
the organization and established a tracking mechanism to measure cost savings and performance gains. This clear
focus on both technology and business outcomes illustrates how defining specific targets from the beginning can
lead to measurable, sustainable impact.
In general, a full implementation of (Gen)AI in an automotive firm can result in significant EBIT impact arising
from both overall efficiency improvements as well es top-line growth. Typically, companies can expect up to 50%
efficiency gains in various processes, with faster automation and streamlined workflows. Additionally, top-line
growth can typically increase by 1% to 2% through (Gen)AI applications such as hyper-personalized marketing
or optimized pricing strategies, further contributing to the firm's overall financial performance.
BOSTON CONSULTING GROUP 8
Three key value creation levers can increase the bottom line (see Exhibit 2):
1 Revenue: Automotive companies can boost revenue in various ways, including by augmenting their
salesforce, call centers, and dealers to identify and approach qualified leads with a highly personalized offer
at certain points of their ownership or leasing life cycle. Additionally, smart-feature bundling and the
introduction of smart-pricing mechanisms will play a significant role.
An example of GenAI’s revenue-driving potential is seen in East Asia, where automotive companies are
focusing on enhancing in-vehicle experiences. This includes advanced voice and gesture recognition systems
that tailor cabin features—such as climate control, seat adjustments, and navigation—to specific passengers
based on their location within the car and profile, whether adult or child. By optimizing both comfort and
safety, these enhancements improve customer satisfaction and engagement, establishing a differentiated
brand presence. These GenAI-driven innovations not only support revenue growth but also bolster EBIT
by fostering greater brand loyalty and reducing customer acquisition costs.
2 Expenditures: Spending can be reduced, for instance, by using (Gen)AI to minimize external service costs,
particularly in software development or marketing creation. For example, a leading automotive company we
worked with utilized (Gen)AI to automate key aspects of software development, reducing reliance on third-
party service providers. By streamlining internal development processes and cutting supplier costs for
software development tasks, the company significantly reduced outsourcing expenses and achieved
substantial cost savings. Beyond external cost, (Gen)AI can also help streamline internal administration or
knowledge management to accelerate decision making. Speed and cost are often two sides of the same coin.
3
Productivity: Automating software testing and coding with (Gen)AI has helped a leading automotive
company reduce manual effort and accelerate development. This enabled faster project completion and
allowed engineers to focus on more strategic tasks, significantly boosting overall productivity. Another
example is the automation of the tendering processes, enabling fast drafting and reviewing of key
documents. This in turn helps to reinvest time within the procurement organization to focus on longer-tail
suppliers or deeper assessment of critical supply-chain events to identify additional savings.
Exhibit 2 | How GenAI Contributes to the Bottom Line
Illustrative: Initial estimates from ongoing research; highly dependent on industry and company context
3pp
Technological maturity
People adoption
Process redesign
40%–60% Increase revenue
Operating model reorganization
Others
1pp 10%–30% Cut costs
20%–40% Boost productivity
5- to 10-year Implementation 2-year Impact on EBIT
EBIT potential efforts EBIT potential potential by lever
Source: BCG analysis.
BOSTON CONSULTING GROUP 9
Organizations can achieve the full potential of (Gen)AI only by establishing a proper value realization mechanism.
They should start by setting top-down targets based on an assessment of (Gen)AI’s potential to improve the three
value areas. In addition, implementing a rigorous process to prioritize high-potential (Gen)AI use cases and
establish accountability is essential. Companies can validate progress and achievements from the bottom up, for
example, by conducting workshops to identify value packages. Furthermore, a proper change-management program
will be required because the transformation will significantly affect organizational culture. The final step is to
establish a transformation office to help execute the implementation roadmap, proactively manage roadblocks,
establish governance structures, and track the progress and impact of specific value packages and the overall
program.
We have seen (Gen)AI transformations yield a 1% to 2% increase in revenue and an 8% to 12% cost reduction
compared with the baseline. Companies can potentially achieve an ROI of 10 to 15 times in less than three years.
For example, BCG worked with a client to achieve an EBIT impact of over 10% by evaluating more than 50 (Gen)AI
value packages. In another case, we helped identify (Gen)AI use cases that a client is putting into practice to
capture up to €1 billion in potential EBIT impact by 2028.
When it comes to implementing (Gen)AI, possibly the most significant technology advancement of our generation,
leaders need to be realistic in the face of excessive hype and expectations. (Gen)AI is destined to affect all
companies in all industries, but only those that can implement it with a focus on process, people, and
organizational culture—not just on the technology—will emerge with real success in terms of greater profitability.
A successful transformation begins with identifying achievable EBIT goals and establishing measures such as a
(Gen)AI transformation office to hold key people accountable. To avoid premature celebrations, leaders must stay
focused on bottom-line impact—starting with taking the right steps at the outset of the (Gen)AI journey.
BOSTON CONSULTING GROUP 10
The Authors
Andrej Levin Felix Stellmaszek
Managing Director & Partner Managing Director & Senior Partner
Hamburg Atlanta
[email protected] [email protected]
Alex Xie Jonathan Nipper
Managing Director & Senior Partner Managing Director & Partner
Shanghai Detroit
[email protected] [email protected]
Manuel Kallies Nina Kataeva
Managing Director & Partner Managing Director & Partner
Berlin Vienna
[email protected] [email protected]
Tobias Schmidt
Managing Director & Partner
Hamburg
[email protected]
The authors thank Nicolas de Bellefonds, Lucas Christenson, Elias Kurta, Ivan Tretiakov, Alexander Hoellinger,
Felix Prosenz, Maximilian Tischer, and Katharina Wortberg for their invaluable contributions to this article.
FURTHER CONTACT
If you would like to discuss this report, please contact one of the authors.
BOSTON CONSULTING GROUP 11 |
127 | bcg | how-ai-agents-are-opening-the-golden-era-of-customer-experience.pdf | How AI Agents Are
Opening the Golden
Era of Customer
Experience
By Karen Lellouche Tordjman, Dutch MacDonald, Phil Gerrard,
Rob Derow, Bridget Scott, Kartik Poria, Rob Bell, and Mark Irwin
January 2025
Businesses have long recognized the link The innovations that make reaching these
between improved operational efficiency and CX goals feasible are already taking place
a better customer experience (CX). But the at different levels. Internally, AI is enabling
rapid convergence of various technological companies to streamline processes,
forces—including next-generation hardware transform the employee experience, and
powered by virtual agents—is turning that increase productivity with unprecedented
link into a measurable source of value speed. In customer service operations, for
creation. Increasingly well-documented example, GenAI has helped companies
use cases for generative AI (GenAI) achieve productivity improvements of
are demonstrating that companies can between 15% and 30%, with some aspiring
simultaneously offer a different and vastly to as much as 80% higher productivity.
superior customer experience at a radically
lower cost-to-serve that yields significant Externally, AI is driving shifts in how
improvements in financial performance. customers engage with brands, making
their interactions more human, more
To achieve this “holy grail,” companies personalized, and less tedious and
have begun to look beyond the traditional confusing. The virtual co-pilots of retailers,
customer journey—actively managed by for example, are interacting with customers
shoppers on their apps—to explore the to answer questions, facilitate returns, and
idea of more comprehensive customer create personalized offers. These co-pilots
“missions” managed by a network of free up employees to handle customer
trusted autonomous agents trained to issues that still require complex or nuanced
accomplish specific tasks with minimal human intervention.
human involvement. The first attempts in
the market today will pale in comparison
to what users will see as the convergence
between hardware and agents intensifies.
2
Contents
Linking Process Improvement to Customer Experience 4
From Time-Consuming Apps to Trusted Autonomous Agents 6
Convergence of Technologies Boosts Benefits 9
Today’s No-Regrets Moves for Executives 11
3
Section 1
Linking Process
Improvement to
Customer Experience
4
Linking Process Improvement to Customer Experience
Current investments by Amazon and the While efforts like these will yield significant
Swedish fintech company Klarna show how productivity improvements, they will not
a company’s AI deployments can deliver fully achieve the desired customer, cost,
benefits internally and externally. and financial performance benefits unless
companies also set the bar much higher for
Amazon’s vast and growing population of enhancement of the CX. That’s where the
warehouse robots can pick, pack, and move convergence of next-generation hardware and
merchandise more efficiently than humans autonomous agents creates game-changing
can. AI applications optimize storage opportunities.
positions and routes within the warehouses
and help the robots detect defective or
damaged merchandise more reliably than
humans, who now take on supervisory and
maintenance roles. This efficiency results
in faster and more flexible delivery for
customers. Amazon claimed that in March
2024 it delivered 60% of orders to Amazon
Prime members on the same or next day
in the top 60 US metropolitan areas. At the
same time, it is aiming to improve cost-to-
serve by 25% during peak seasons at its next-
generation warehouses.
Klarna, a global provider of “buy now, pay
later” payment solutions, introduced an
AI customer service assistant powered by
OpenAI in early 2024. With 85 million active
users, the company reported that within
its first month, the AI assistant managed a
workload equivalent to that of 700 full-time
agents. Customer satisfaction levels were on
par with previous satisfaction with human
agents, but customers benefitted in several
other measurable ways. Repeat inquiries
fell by 25% due to greater accuracy in task
resolution. Speed of service also improved,
with customers resolving problems in less
than 2 minutes versus 11 minutes with human
agents. In September 2024, the company
announced new features for the AI assistant,
including open-ended research, searches
for specific products or brands, product
comparisons, product recommendations, and
price research. Klarna has estimated that the
implementation of these assistants will yield
$40 million in additional profits in 2024.
5
Section 2
From Time-Consuming
Apps to Trusted
Autonomous Agents
6
From Time-Consuming Apps to Trusted Autonomous Agents
In his keynote address at Dreamforce 2024, draw complex inferences on its own from
Salesforce CEO Marc Benioff said that the personal information the consumer
service employees waste over 40% of their has allowed it to access. As Exhibit 1
time on low-value and repetitive tasks. We shows, a vertical network of proprietary
have found that customers who want to and third-party agents would work in the
buy a car, remodel their home, or just find background to complete the discrete tasks
something fun to do on a weekend face a of a customer mission—such as “get me
similar set of challenges. Much of their my new car”—rather than simply serve
research conducted via apps, call centers, as the customer’s Q&A machine on a
or web pages can devolve into a confusing, traditional customer journey. The network
frustrating, and error-prone experience. One of agents thus do much of the burdensome
study by Google revealed that a customer work that a consumer would currently
can have more than 700 digital touchpoints need to do manually through an app or a
over a few months as they try to plan a trip. website. As this “agentic AI” learns a user’s
unique preferences, it will deliver products,
Imagine instead that a brand’s chatbot is services, and experiences that are more
an agent that is designed and trained to comprehensive and more personalized.
Exhibit 1 - How AI Agents Work
Source: BCG analysis.
7
When planning a vacation, for example, a
brand’s autonomous agent will use each
family member’s unique interests and
preferences to suggest destinations; find
itineraries; map out routes; recommend
flights, hotels, and restaurants; and even
make reservations for activities. During
the trip, the task-specific networked agents
can also track the family’s progress and—
through a single user interface—make short-
term recommendations for changes based
on traffic, weather, activity opportunities,
meeting friends, or other conditions. This
has the potential to achieve a superior
standard of convenience and service and
alleviate what can sometimes be a dizzying
overabundance of choices while traveling.
If the customer wants to remodel their
home, the agents could find the most
effective methods and materials, make
recommendations, order products, and
even generate self-help videos for tasks the
customer wants to undertake. Autonomous
agents can manage not only major
occasional events (travel, car purchase,
home remodeling) but also day-to-day tasks
such as shopping, diet and workout planning,
pet care, and car and home maintenance.
8
Section 3
Convergence of
Technologies Boosts
Benefits
9
Convergence of Technologies Boosts Benefits
Now imagine an AI agent that is multimodal. and videos, play music, make video calls,
This is where the convergence between livestream what the wearer sees, and
agents and next-generation devices even translate foreign languages. The
comes into play. Besides revolutionizing device anticipated from the collaboration
specialized tasks and solving customer between former Apple designer Jony Ive
problems faster, better, and more efficiently, and OpenAI CEO Sam Altman is expected
agents also untether customers from their to have similar benefits in that the product
screen-based or handheld devices. The will immerse itself into the wearer’s day-to-
shift from app-based operating systems to day life without the interruptions that occur
ambient, natural interfaces—such as voice, using a handheld screen.
augmented reality, and eye movement—
will accelerate as AI-powered devices As agents and hardware converge, we
become more intelligent and predictive. anticipate mutually reinforcing innovations
This transformation will drive hardware that boost both user experience and
innovations that blend into a customer’s customer experience. On the agent side,
day-to-day life rather than being a distinct we expect technical advancements that
task or experience. In the workplace, will bring down the cost of training and
employees can use these devices to interact operating the underlying models, reduce
with the company’s agents that help them the risk of hallucinations, and help build
do their jobs better, providing benefits to stronger and more widespread customer
customers. trust to turn over their personal data and
much of their life management. In addition
The smart glasses from Meta and Ray-Ban to upgraded technology, devices will also
exemplify this. They function as normal offer greater benefits in terms of comfort
eyeglasses, even with prescription lenses, and style, tighter integration into the flow of
but their built-in GenAI can analyze and work or life, and more responsiveness to a
interpret what they see along with the range of sensory cues.
wearer. The glasses can take pictures
10
Section 4
Today’s No-Regrets
Moves for Executives
11
Today’s No-Regrets Moves for Executives
The integration of agents into an organization one—we recommend a more pragmatic
goes beyond a straightforward deployment approach for the short term. Companies
of GenAI to test the value of use cases. In line should start with a small number of deep
with BCG’s progression of Deploy-Reshape- initiatives that lay the foundation for an
Invent, the ability to establish a solid eventual transformation. In a deep initiative,
connection between radical improvements the company tests use cases for how AI and
in operational efficiency and a vastly different GenAI can significantly enhance operational
and superior customer experience creates efficiency and customer experience, then
an imperative for companies to invent new comprehensively assesses the effects on
business models or risk being left behind. the tech stack, the future roles of affected
These models would serve complex data- team members, and the required metrics for
driven customer missions (“buy the perfect defining success from a customer, operational,
car for me”) instead of separate customer and financial standpoint. At the beginning,
journeys (“show me mid-size hybrid SUVs”) the assessments should give greater weight to
as the twin forces of autonomous agents and the operational side, which is a prerequisite
multimodal hardware continue to evolve and for scaling the agents with customers.
converge. (See Exhibit 2.)
In detail, we see several no-regrets moves
But before embarking on a massive customer right now to prepare an organization for a
experience transformation—or even planning larger transformation.
Exhibit 2 - The Convergence of Use Cases and Technologies Reduces Operational Costs
and Transforms Customer Experience
Source: BCG analysis.
12
The vertical integration in Exhibit 1 will take place first, as agents start
to take over parts of the customer journey. At the same time, companies
Build for need to start thinking about their role in broader customer missions. The
journeys and design challenge is to understand these journeys and missions and serve them in
for missions ways designed from the outset to achieve their CX and financial goals.
Executives can begin to rethink traditional CX metrics by incorporating new
KPIs, such as cost-to-serve, alongside their preferred customer satisfaction
Adopt metrics. These metrics should allow an executive team to “connect the
new metrics dots” by making transparent the relationship between lower costs, superior
customer experience, and better financial performance.
Companies have tended to underinvest in efforts that translate
operational improvements into better customer experience, because they
Reevaluate considered short-term gains to be soft and uncertain rather than hard and
investment verifiable. The ability to draw direct links between operational efficiency
priorities and customer experience, and then calculate short- and long-term ROIs,
can attract the investment needed to keep the momentum going.
This approach is necessary to keep pace with the accelerating rate of
innovation as well as convergence. Convergence will take place not only
Learn to technologically—such as between hardware and agents—but also within
converge the organization as it finds ways to integrate teams, functions, and skills
in ways that drive greater efficiency and a better CX. It means launching
deep proof-of-concept tests of GenAI’s potential in different functions,
starting with low-hanging fruit, to identify potential pitfalls and address
them early in the implementation process.
These actions will enable the company’s leadership team to start preparing a roadmap for
integrating GenAI into the customer-experience strategy—from short-term wins to long-
term, full front-to-back transformation.
13
The convergence between ever-improving hardware
and larger networks of trusted autonomous agents will
allow companies to achieve the holy grail of improved
productivity, higher customer satisfaction through a
superior and differentiated customer experience, and
better financial performance. The familiar link between
improved efficiency and better customer experience
will become a nexus of innovation and value creation in
ways we can only begin to imagine, as the technologies
integrate more seamlessly into day-to-day life.
The authors are grateful to their BCG colleagues Oli Shaw, Melanie Stetter Hernandez, and Aylin Ozcan
for their insights.
14
bcg.com
15 |
128 | bcg | WEF_Frontier_Technologies_in_Industrial_Operations_2025.pdf | In collaboration with Boston Consulting Group
Transformation of Industries in the Age of AI
Frontier Technologies
in Industrial Operations:
The Rise of Artificial
Intelligence Agents
W H I T E P A P E R
J A N U A R Y 2 0 2 5
Images: Getty Images
Contents
Reading guide 3
Foreword 4
Executive summary 5
Introduction 6
1 The next leap: reinventing industrial operations through 7
frontier technologies
1.1 Entering the next frontier: the path towards self-control 8
1.2 Redefining the role of humans: from operators 8
to AI-enabled orchestrators
2 AI agents fuelling the transformation of operations 10
2.1 Virtual AI – paving the way for autonomous systems 12
2.2 Embodied AI – igniting a new era in robotics 15
3 Strategic imperatives for industrial operations transformation 18
3.1 Paving the way for successful use of AI agents 18
in industrial operations
3.2 Staying at the forefront of AI agent innovations 19
3.3 Building the foundations: organizational and technological 19
Conclusion 21
Contributors 22
Endnotes 25
Disclaimer
This document is published by the
World Economic Forum as a contribution
to a project, insight area or interaction.
The findings, interpretations and
conclusions expressed herein are a result
of a collaborative process facilitated and
endorsed by the World Economic Forum
but whose results do not necessarily
represent the views of the World Economic
Forum, nor the entirety of its Members,
Partners or other stakeholders.
© 2025 World Economic Forum. All rights
reserved. No part of this publication may
be reproduced or transmitted in any form
or by any means, including photocopying
and recording, or by any information
storage and retrieval system.
Frontier Technologies in Industrial Operations 2
Reading guide
The World Economic Forum’s AI Transformation of This white paper series explores the transformative
Industries initiative seeks to catalyse responsible role of AI across industries. It provides insights
industry transformation by exploring the strategic through both broad analyses and in-depth
implications, opportunities and challenges of explorations of industry-specific and regional deep
promoting artificial intelligence (AI)-driven innovation dives. The series includes:
across business and operating models.
Cross industry Regional specific
Impact on industrial ecosystems Impact on regions
AAIll iGanocveernance IPnw cCollaboration with AAIll iGanocveernance AAIll iGanocveernance AAIll iGanocveernance
A E T FJIn T ALrc xr Io aa NA l nl pa i ns UGb nf eo o s ASr r a m r fAt RHi i oo a m Yn t Ici row P mn eti 2t h o i 0W nofA I 2c In t Hc n nd 5ae u In : dtst T u t irr B o u Ee ies n se Pin t y A rtth o o y Pe n EAg d Re o Sf A EI RIES L f W S a INo Nne Oc Sr o d Vv e IrJ Ee Gnk a Mor Haf a Bb F Tro i Eg o r r RRaAi c sn E mu ,e 2Pg 0g C e OP 2G m w Ra 4r Tso oe e e rdn n k e Sut a fr c t oa ut t rit dio vi Av in eit ce s ya t i:A on I nd A E C WJIn T Ac r n Hr ho a Nl n tl Ia e as Uib Tf fio o lr AElr ra egcmt R io a n Pyi Yn t ai Ao gw n Pli 2Pt e h o 0 I EfA a s nI 2c n Rc rd 5te aaun est nu t dlrre i l de i os g i xn Oe t :h n pe B A c pg ae e o o l’ af rs A tnI u c nin itig e s A a B a WJICna ATc nnp Hr aro aa Nl dtc nl d Ilai s a Uitb Ty f fi o o An EC Rr r Ca cme Rt cni eo a Pt i y Yrn t iae i w Anow, b nU li 2Pgt nh ao e 0i I Efvt h n re I 2r Rne Rr ds d s 5tGit u i seyl so es ob t lr kfa c i lO el i sC sx g u fy io nbr re ed tr ih n tS ee yAc cu g :r e e it y o f AI B C I WJIn n T Ac r Hhlo da Nul nl Ia i us Unb Tefo o s AEr ar pa m tt R ’io a rs Pr Yn t yi i A ow n P ni 2Pt T h o t 0a EfA r I 2c t atn Rc o hd 5e nun st su tt Arr oe i fe ocs Ai rn t m it Ih o -e P a nAg to:e i o wof n A eI red
AI in Action: Beyond Leveraging Artificial Intelligence’s Artificial Intelligence Blueprint to Action:
Experimentation to Generative AI for Job Energy Paradox: and Cybersecurity: China’s Path to
Transform Industry Augmentation and Balancing Challenges Balancing Risks AI-Powered Industry
Workforce Productivity and Opportunities and Rewards Transformation
Industry or function specific
Impact on industries, sectors and functions
Advanced Media,
manufacturing Financial entertainment
and supply chains services and sport Healthcare Transport Telecommunications Consumer goods
F i T I WJIn nn T Ac hrr Ho ta Nol enl I e Ia s n Ub Tn lf o lo Ai ERdr r gta mt
R
ii u io a ee Ps Yn ti As no rw e ni 2Ptt c h To r 0o EfB e ieI 2o n Ra f s d 5ct Ao Au ln s h gt OC rrio ne t en s ips fi o niu n cl et l ti tn h o sig r ae a G gA lr go tieu iep oof s nA I s: A i WJIn nAA T AcIl r Hrl oi a G Na l n tl Fn Io a s Uc iv b Tfee fi io or AEnrn ra cma t
R
ain o a Pic Yn t ae i n Aow n li 2Pct h o
0
I EfA i nI a 2c n Rc d 5te lun est Su t lrre i le eis g rin e v th ine c A c ege e so f AI A E WJInAA T AcIl r n Hrl oi aG Na l n tln Io a ts Uc iv b Tefee fio or AErn r ra cma t
R
tin o a Pic a Yn t ae i Aow in li 2n Pt h o
0
I EfA m nI 2c n Rc d 5te eun estu nt lrre i le tis g i an e t nh ne d A cg e e S o f p iA nI o M rtedia, T L WJIn A T ec h Hro a Nl aenl Ia Us Tb df o o AEFr r ia m n R t ui Po a Ygn t ti A ow u n 2Pi tt rh o h 0E efB 2I eo Rn s 5d ot o u Wn s f tC ri Ao e an ss yIiu n -lt i t En hg e n G Ar g ao eup bof lA eI d Health: D GAI G WJIMn n
A
Tcc I HerK lro ta Noi Iln en al e cas Us Tbbe f sel ao oy AElr ar &a i nm r
R
t g a Ci lb Po a e Yon t e Li Am o o Cw rp n
o
2n Pi nt a h o a 0n F
Eg
tf iy z 2tI Run iaT 5sed tu lr
t
ys u ia t cr si re n sts e sin t: o pth e o Ag re t o ,f AI
Frontier Technologies Artificial Intelligence Artificial Intelligence in The Future of Intelligent Transport, Upcoming Upcoming
in Industrial in Financial Services Media, Entertainment AI-Enabled Health: Greener Future: industry report: industry report:
Operations: The and Sport Leading the Way AI as a Catalyst Telecommunications Consumer goods
Rise of Artificial to Decarbonize
Intelligence Agents Global Logistics
Additional reports to be announced.
As AI continues to evolve at an unprecedented and stakeholders engaged in AI strategy and
pace, each paper in this series captures a unique implementation across organizations.
perspective on AI – including a detailed snapshot
of the landscape at the time of writing. Recognizing Together, these papers offer a comprehensive
that ongoing shifts and advancements are already view of AI’s current development and adoption,
in motion, the aim is to continuously deepen and as well as a view of its future potential impact.
update the understanding of AI’s implications Each paper can be read stand-alone or alongside
and applications through collaboration with the the others, with common themes emerging
community of World Economic Forum partners across industries.
Frontier Technologies in Industrial Operations 3
January 2025 Frontier Technologies in Industrial Operations:
The Rise of Artificial Intelligence Agents
Foreword
Kiva Allgood
Head, Centre for Advanced Daniel Küpper
Manufacturing and Supply Managing Director and
Chains; Member, Senior Partner, Boston
Executive Committee, Consulting Group (BCG)
World Economic Forum
Amid a landscape of exponential technological advantages and catalyse sustainable growth.
change, society is entering the Intelligent Age1 – Manufacturers that fail to fully harness the
an era defined by far more than technology alone. transformative power of frontier technologies in
The Intelligent Age is characterized by a mass operations and supply chains will surely fall behind.
revolution transforming all aspects of society, and
we’re already beginning to witness profound shifts. Although the pursuit of frontier technologies is not
Alongside technological intelligence, environmental, novel, the stakes are now higher than ever. The
social and geopolitical intelligence will be challenges associated with identifying and evaluating
fundamental to success in this age. these technologies and integrating them into long-
term strategies have grown more complex as the
In this new era, industrial operations are being pace of innovation accelerates. Forward-thinking
redefined. To better understand emerging industries, technology leaders and academic
opportunities and explore potential responses, institutions are pioneering such advancements.
the World Economic Forum – in collaboration with Yet, even with the growing accessibility of tools
Boston Consulting Group (BCG) – launched the like generative AI, manufacturers still face a crucial
global initiative Frontier Technologies for Operations: question – how will frontier technologies drive real,
AI and Beyond. Building on the success of our measurable impact in day-to-day operations?
previous AI-Powered Industrial Operations initiative
from 2022, this new scheme aims to equip This white paper presents a bold yet actionable
manufacturers with the insights and tools necessary vision of one such frontier technology – AI agents.
to shape the future of industrial operations in the It additionally outlines methods by which this
Intelligent Age. technology could be applied to create tangible value
in industrial operations. The paper focuses on the
It’s important to address two pressing questions transformative potential of two types of AI agents
– why focus on frontier technologies, and why – virtual AI agents and embodied AI agents – and
now? The answer is simple yet profound – these provides insights and case studies from leading
innovations, like others in the Intelligent Age, drive industries while challenging conventional thinking
boundary-breaking advancements that push and inspiring new strategies. Its aim is to highlight
the limits of what’s currently possible, facilitating innovative perspectives to help manufacturers
collaborative intelligence and amplifying human unlock the full potential of AI agents and spearhead
ingenuity. In doing so, they provide competitive operational transformation.
Frontier Technologies in Industrial Operations 4
Executive summary
It’s essential that manufacturers embrace
frontier technologies to secure a thriving,
sustainable future in manufacturing.
The manufacturing landscape is becoming – Virtual AI agents – advancing autonomous
increasingly complex, and this trend is projected to software systems: Virtual AI agents enable
accelerate in the coming years. Labour shortages, software applications to autonomously achieve
rising cost pressures and shifting customer demands, defined goals in the digital environment, acting
geopolitical dynamics and decarbonization goals as assistants, advisers or automation agents.
necessitate significant operational transformation. These agents support workers and can also
independently control and steer processes
Current technologies will be insufficient to drive and machinery.
the required levels of flexibility, sustainability and
excellence needed to facilitate this change. To – Embodied AI agents – ushering in a new
succeed, manufacturers can embrace frontier era of robotics: Embodied AI agents equip
technologies that push the limits of innovation. physical systems, such as robots, with the
However, navigating this rapidly evolving ability to perceive and act within the physical
technological landscape is challenging, as many environment, allowing for dynamic and complex
manufacturers need to address immediate operational movements. These advancements will be
needs and plan for the future of their operations. crucial for overcoming the current limitations
of robotic automation.
Industrial operations are likely to evolve towards
an artificial intelligence (AI)-centric model, where AI Successfully navigating the transition to near-
drives self-controlling, near-autonomous systems autonomous, AI-agent-driven operations requires
while empowering humans. While near-autonomous a comprehensive, value-driven approach to
operations may become common in some industries, technology adoption. Solutions should be scalable
human involvement will remain crucial. The role of and aligned with long-term business objectives.
humans will be redefined, with workers transitioning Establishing strong organizational and technological
from hands-on operators to orchestrators, stepping foundations that support this vision will be
in when judgment or creativity is required. This shift crucial for manufacturers looking to capture the
will boost operational efficiency, allowing humans technology’s full potential.
to focus on strategic tasks and ethical decision-
making to drive innovation and growth. The insights presented in this paper are focused
on manufacturing and founded on the collective
In the broad landscape of frontier technologies, expertise of the initiative community, drawing
AI – and more specifically, rapidly evolving AI from consultations with senior executives and
agents – have the potential to propel manufacturers academic experts. Moving forward, the community
towards this future, and unlock novel opportunities will continue to work closely with manufacturing
in operations across many industries. This report stakeholders across industries to deliver a global,
focuses on two types of AI agents: virtual AI comprehensive outlook on the future of industrial
agents and embodied AI agents. These agents are operations. This effort will concentrate on recent
expected to enhance both digital applications and and future frontier technologies, with an emphasis
physical systems, and perform complex tasks with on responsible transformation approaches.
minimal human intervention.
Frontier Technologies in Industrial Operations 5
Introduction
AI agents are transforming industrial
operations, driving efficiency and unlocking
competitive advantages.
As new frontier Frontier technologies have pushed the limits – Where is the real value in this transformation?
technologies of what is possible in industrial operations
emerge, over the past decades, significantly boosting – Which frontier technologies will address
productivity, reducing costs and improving the key challenges?
manufacturers face
work environment. Innovations like robotics and
the key challenge
the industrial internet of things (IIoT) have been – What steps need to be taken to realize value
of discerning
instrumental in modernizing operations and laying at scale?
which innovations
the foundation for the next wave of breakthroughs.
will bring lasting
Drawing on insights from experts and executives
value at scale, and
Today, the technological landscape is evolving at across operations and technology, this white paper
which are merely an unprecedented pace. This progress is primarily provides a strategic perspective on these questions,
transient trends. driven by the exponential increases in computing with a focus on AI-agent-enabled transformation.
power and breakthroughs in artificial intelligence It presents a forward-looking vision of AI-driven,
(AI) society is currently witnessing. As new frontier near-autonomous industrial operations. It explores
technologies emerge, manufacturers face the key the role of AI agents in enabling this vision,
challenge of discerning which innovations will bring specifically virtual AI and embodied AI agents,
lasting value at scale, and which are merely transient offering concrete examples and case studies to
trends. This creates uncertainty around where to demonstrate their value. Additionally, it outlines the
focus development efforts and investments. strategic imperatives necessary for successfully
scaling these technologies. While AI agents hold
Overcoming these challenges is essential. transformative potential, it is crucial to recognize that
Harnessing the value of frontier technologies is now they are not yet fully developed. Leading companies
vital for manufacturers as they seek to maintain are running pilots to test their capabilities, with their
a competitive edge and tackle industry-specific at-scale impact to be realized in the coming years.
obstacles. To retain a leading position in the evolving
landscape, companies must not only adopt these Although not covered in this white paper, other
innovations but also understand the transformative frontier technologies – such as biotechnology and
impact on the future of operations. Success in this quantum technology – are generating significant
journey hinges on answering a few key questions: interest. These technologies hold the potential
to revolutionize manufacturing operations, either
– What will the future of industrial operations directly or indirectly, but remain in earlier stages
look like? of development.
BOX 1 The two types of AI agents
Virtual AI agents Embodied AI agents
Software-based AI agents that operate AI agents integrated into physical systems
entirely in the digital environment and enable – such as robots – that interact with the
digital applications to autonomously achieve physical environment
defined goals
Frontier Technologies in Industrial Operations 6
The next leap:
1
reinventing industrial
operations through
frontier technologies
Preparing for the challenges ahead
requires operational transformation
driven by frontier technologies.
Manufacturers face a complex operating – Geopolitical dynamics: Tariffs and fragmented
environment with growing challenges: production across multiple geographies
hinder economies of scale, leading to greater
– Cost competitiveness: Rising labour costs, complexity in supply chains, dispersed know-
supply chain disruptions and international how and increased risks.
competition necessitate improved efficiency
and lowered structural costs. – Sustainability: To meet decarbonization goals,
it’s crucial to optimize energy and resource use
– Labour shortages: More than 2.1 million while reducing emissions through robust supply
manufacturing jobs are projected to remain chain management.
unfilled in the US alone until 2030,2 driving
workforce risks and productivity challenges. Addressing these challenges requires a shift in
operational excellence, breakthrough innovation,
– Customer demands: Consumers’ expectations structural optimization, supply chain diversification
for greater customization and faster delivery and investment in regional manufacturing clusters.
drive the need for more flexible production
systems and better demand forecasting.
Frontier Technologies in Industrial Operations 7
1.1 E ntering the next frontier:
the path towards self-control
Although The industrial sector stands at a pivotal juncture. performance monitoring can be centralized in virtual
the extent of Frontier technologies, such as AI agents, are control centres rather than dispersed throughout
automation will capable of performing complex activities. This the shop floor.
ultimately depend paves the way for increasingly AI-driven, near-
autonomous operations, within which many Self-controlling factories and supply chains will
on the return of
machines and AI-enabled systems will function with deliver significant improvements such as:
investment, many
minimal human intervention. Success depends on
factories may
cultivating a trusted human-machine interaction, – Efficiency: Predictive analytics will shift operations
converge towards
where both collaborate seamlessly. from reactive to proactive management,
autonomy, driven
anticipating issues and implementing necessary
by the need to Currently, automation is often reserved for simple, adjustments immediately. Real-time adjustments
remain competitive. repetitive tasks that still require manual oversight will enhance machine uptime, quality control
to ensure continuous operation. In the past, and cost efficiency.
the expansion of automation was hindered by
technological hurdles (such as an inability to handle – Flexibility: Advanced robotics and AI will
unsorted flexible parts like cables automatically) enable highly personalized manufacturing
and financial constraints. However, more advanced and swift reconfigurations, making production
technologies and decreasing costs are poised lines adaptable to varying product demands.
to enable wider deployment across factories, Autonomous systems will self-organize for
with autonomous systems taking control of optimal factory layout and performance, further
routine operations. These autonomous systems enhancing flexibility. They will also increase
– encompassing machines, robots and virtual supply chain agility and responsiveness.
systems – may manage routine tasks ranging
from material handling to quality control and – Sustainability: Autonomous systems
production planning. Such systems may optimize will optimize energy consumption and
and adjust production parameters on machines in minimize waste. Real-time analytics will
real time to align with business needs, enhancing monitor environmental impacts, ensuring
flexibility. Although the extent of automation will that sustainability goals are met without
ultimately depend on the return of investment sacrificing efficiency.
across industries and regions, many factories may
converge towards autonomy, driven by the need to – Worker empowerment: AI-driven tools and
remain competitive. automation will enhance workforce capabilities
and facilitate human-machine interactions,
The shift towards autonomy may also revolutionize enabling workers to quickly understand
factory design. Future AI-centric factories might production issues and make more well-
prioritize machine-optimized layouts that enhance informed decisions.
production efficiency and flexibility. For instance,
valuable ground-floor space can be freed up by The transformation to near-autonomous industrial
storing unfinished parts in automated multi-storage operations requires coordinated changes across
shelves, manual processes can be accelerated and both human and technological dimensions.
1.2 R edefining the role of humans: from
operators to AI-enabled orchestrators
Human involvement will remain essential in productivity breakthroughs. For example, one
industrial operations of the future, as workers individual supported by assistant systems can
may transition from hands-on operators to AI- supervise multiple functions such as quality,
enabled orchestrators who oversee autonomous inspection and production simultaneously.
systems and provide judgment or ingenuity as Maintenance activities that require physical dexterity
required. As machines advance in natural language – such as checking for leaks or replacing parts
comprehension, human-machine interactions inside a machine – may partially remain human-led
will become more fluid and intuitive, enabling but can be significantly augmented by virtual agents.
Frontier Technologies in Industrial Operations 8
In a future In a future with largely self-controlling systems, yield deviations that systems cannot resolve,
with largely humans may partner with machines, harnessing humans can step in to address the issue.
self-controlling collaborative intelligence to focus on higher-value
systems, humans tasks, such as: – Continuous improvement involves solving
complex problems and optimizing processes.
may partner
– Strategic decision-making involves using For instance, in a chemical processing plant,
with machines,
AI-driven recommendations to make business- engineers may use AI to identify inefficiencies
harnessing
critical decisions. For instance, in an automotive in mixing or reaction processes. They can then
collaborative
plant, AI may recommend adjustments to redesign workflows or machine configurations
intelligence to
production schedules or shift planning. A human to optimize output and reduce waste.
focus on higher- planner may weigh these recommendations
value tasks. against factors such as projected customer – Creativity and innovation involve developing
demand or current labour availability. new production processes and rethinking
factory layouts. For instance, in a consumer
– Performance supervision involves monitoring electronics plant, a maintenance worker might
and adjusting autonomous systems as needed. introduce creative ideas to streamline tool
For instance, in a semiconductor plant, operators changes by mounting additional supports that
may monitor autonomous systems handling have been employed in other industries.
wafer fabrication. If performance metrics show
BOX 2 Industry example: Shifting role of technicians and supervisors
A global wheel manufacturer has experienced a improvement by optimizing the plan-do-check-act
shift in the role of their technicians and supervisors (PDCA) cycle. Supervisors, in turn, are evolving
with the introduction of a prescriptive AI solution into AI users, interpreting AI-driven insights and
for process parameter adjustment developed by a guiding operators towards more efficient problem-
Cape Town-based AI solution provider. Instead of solving. This transition enables both operators and
managing process details, technicians now focus supervisors to concentrate on long-term, systemic
on identifying root causes and driving continuous improvements rather than routine, reactive tasks.
BOX 3 Industry example: Elevating planner roles with AI-supported decision making
A Fortune 500 technology manufacturer elevated routine decisions in inventory management
the role of its planners from executors to architects while routing exceptions to human experts with
of its supply chain decision-making process. contextual data, analysis and recommendations.
Previously relying solely on humans, the company The platform optimized stock levels and ensured
struggled with delayed decision-making, resulting supply was matched to regional demand. As
in large inventories and long lead times. By a result, 77% of agent recommendations were
harnessing an AI agent solution from a US-based automatically executed and 90% were accepted
decision intelligence company, they automated without change.
This evolution will require manufacturers to anticipate a transition in workforce skills
and cultural identity, making early engagement of operators in the transformation
journey critical for success.
Frontier Technologies in Industrial Operations 9
AI agents fuelling
2
the transformation
of operations
Virtual and embodied AI agents could drive
the transition towards near-autonomous
operations in both software and robotics.
Realizing the transformative vision of AI-centric and embodied AI agents have the potential to deliver
operations requires a thorough assessment and significant value, unlock new opportunities and drive
evaluation of the potential of AI agents. Both virtual the transition towards near-autonomous operations.
AI will transform from a data-centric front end to an agent-centric
user end, relying on domain-specific data sources to optimize
industrial operations. These domain-based agents will drive new
growth of AI across different industries. The interactive agents
will further transform the new large knowledge model, fostering
the development of AI ecosystems with advanced technologies,
tools and talents.
Jay Lee, Clark Distinguished Professor;
Director, Industrial AI Center, University of Maryland
Frontier Technologies in Industrial Operations 10
BOX 4 The basics of AI agents
AI agents amplify the impact of large language models The roles can be predefined, or agents can be flexible
(LLMs) by giving them access to tools and enhancing their and dynamically adapt to new roles.
ability to observe, plan and execute actions.3 Traditional AI
algorithms, such as machine learning, are task-specific and – Reasoning module: Agents have limited reasoning
require human input for defining tasks, providing data and capabilities. The underlying LLM is capable of
interpreting results. In contrast, AI agents, once trained, decomposing the agent’s prompts and returning an
can operate and achieve specific objectives autonomously, actionable plan. It extracts key insights and makes
continuously observing their environment, planning actions logical connections by replicating reasoning steps
and harnessing tools to execute complex tasks. AI agents observed in training data. This enables agents to
function in a continuous observe, plan and act cycle, which decide on the required next steps by breaking down
makes them particularly valuable for operations. Each step is complex tasks into small actions to achieve their
enabled by interfaces or modules:4 objectives. Recent studies have shown that current
LLMs are not yet capable of formal reasoning. Real-
– Observe: Agents collect and process data from the world solutions thus require other types of AI and
environment, including multimodal data, user input or solvers and cannot solely rely on existing LLMs.5
data from other agents. For example, an agent can
perceive deviations in production quality and underlying – Act: Agents execute actions by harnessing internal
parameters in real time. or external tools and systems. For example, an agent
accesses the machine controller and changes the defined
– Agent-centric interfaces: Agents require protocols, machine parameters.
application programming interfaces (APIs) and
specifically designed interfaces to input multimodal – Action module: Agents decide which tools to use,
data or perceive real-time data from multiple sources. using access mechanisms such as APIs, system
integrations or other agents as needed.
– Memory module: Agents have short- and long-term
memory, which allows them to remember general Functioning in this cycle, agents continuously learn from
knowledge, past actions and decision-making. self-reflection or external feedback. Through goal-oriented
learning approaches, such as reinforcement learning, agents
– Plan: Agents and their underlying LLMs evaluate possible continuously adapt and refine their strategies over time.
actions to prioritize them through logical reasoning, in This makes them particularly valuable in complex, dynamic
accordance with their objectives. In the example above, environments where conditions and objectives are constantly
the agent reviews possible actions to improve quality and shifting. Such environments can be found widely across
decides to change production parameters. industrial operations. As part of multi-agent systems, in
which specialized agents work together by dividing complex
– Profile module: Agents have defined attributes, problems among themselves, they can automate entire
identities, roles or behavioural patterns. processes end-to-end.
AI agents function in a continuous observe, plan and act cycle
Observe
Collect and process
data from environment
Agent
Act Plan
Execute by leveraging Evaluate possible
internal or external actions to prioritize
tools/systems them through reasoning
Source: Boston Consulting Group (BCG).
Frontier Technologies in Industrial Operations 11
2.1 V irtual AI – paving the way for autonomous systems
Virtual AI agents can manage a wide range of agents have applications across all operation
software-based tasks, from routine operations functions, including production, maintenance,
and research to advanced analytics and task quality, engineering, logistics and planning.
automation. In industrial operations, they can
enhance responsiveness, improve execution quality, The maturity of virtual AI agents can be categorized
boost productivity and reduce operational mistakes. into three levels: assistant, recommendation and
Unlike traditional machine learning programmes, automation. The distinct objectives at each maturity
they can make context-sensitive decisions in real level are pursued by specialist agents:
time and adapt through feedback loops. These
FIGURE 1 The four types of virtual AI agents
Recommendation Automation
Assistant
Maturity level (proposing scenarios (autonomously performing
(executing manual tasks)
and actionable insights) activities)
Specialist
Knowledge agent Adviser agent Automation agent
agents
Meta agents Meta agent
Source: Boston Consulting Group (BCG), World Economic Forum.
Knowledge agents support workers as intelligent optimize machine performance, adjust production
assistants. They analyse and synthesize vast parameters, recode instructions or modify
amounts of data to provide real-time operational production plans. They surpass existing RPA
insights, flag anomalies and create content such as (robotic process automation) by automating not
reports and code. By accessing multiple tools and only individual tasks but also entire human activities
real-time data sources, such as machine logs and that require understanding, planning and execution.
sensor data, they add value to functions that require
quick insights – for example, in maintenance, quality Meta agents orchestrate specialist agents in the
and logistics. They can also support engineering context of multi-agent systems to achieve broader
with machine code generation. objectives, enabling area- or even factory-wide
steering. The long-term vision for meta agents is to
Adviser ag |
129 | bcg | the-blueprint-for-ai-powered-marketing.pdf | The Blueprint for
AI-Powered Marketing
December 2024
By Derek Rodenhausen, Ray Yu, Trevor Sponseller, Paola Scarpa, Henry Leon,
Javier Perez Moiño, Val Elbert, and Jordan Baker
The Blueprint for
AI-PoweredMarketing
AI is reshaping the marketing landscape, offering breakthrough
capabilities like unlimited permutations of personalized marketing
for modern consumer journeys, predictive insights, and real-time
decision making. But for most marketers, the whole idea of AI
feels overwhelming.
Marketers are constantly bombarded with pitches from WhatAIExcellenceLooksLike
tech providers and platforms, each claiming that their AI
solution will revolutionize the way their brand engages with Most brands are still in the early innings with AI: more
customers. They wonder: Where do I start? Which areas than 80% of our survey respondents are exploring off-the-
should I prioritize? shelf solutions, adopting use cases, or experimenting.
Two-thirds claim they are stymied by either a lack of knowl-
To address these questions, we conducted one of the larg- edge or the sheer number of options. But a select few have
est studies of its kind, surveying more than 2,000 market- started to figure it out: about 20% of respondents have
ers globally and speaking with over 50 marketing leaders. integrated AI tools deeply into their marketing workflows.
(See “About the Study.”) Through our research, we devel- They are testing AI-assisted decisions and personalization
oped a clear framework to help marketers cut through the approaches in such areas as content creation, predictive
noise, pinpoint where AI can drive the most impact, and analytics, and synthetic research methodologies. In the
build a roadmap for growth. In this article, we share in- past 12 months, these leaders have achieved powerful
sights from the study. We then outline the essential steps results: they report 60% greater revenue growth than their
for making AI an integral element of the marketing func- peers and are adapting to consumer trends twice as fast as
tion to create an AI flywheel that seamlessly integrates their peers.
capabilities across media planning, creative, activation, and
measurement to boost efficiency, deliver results, and keep So what are the 20% doing right? We’ve identified six key
brands ahead of the curve. actions. (See Exhibit 1.)
2 THEBLUEPRINTFORAI-POWEREDMARKETING
About the Study
In August 2024, BCG partnered with Google to research
how leading marketers are using AI, leveraging Google’s
global scale and expertise in AI-enabled advertising solu-
tions. Our goal: to build a globally consistent framework for
marketers to use on their AI adoption journey. We surveyed
2,000 marketing executives across 10 key global markets
and more than a dozen industries, from retail and consum-
er packaged goods to banking and pharmaceuticals. Com-
pany size ranged from small and medium-sized businesses
to multinationals.
The survey consisted of more than 50 questions about
companies’ maturity level, AI capabilities and practices,
and the extent to which their AI-based marketing was
integrated across the enterprise. We asked about such
issues as:
• Their use of AI for developing consumer insights (for
example, automating traditional audience research and
creating synthetic insight tools such as conversational
personas that represent different target audience seg-
ments and their preferences)
• Their audience segmentation and media-bidding strat-
egies (e.g., setting inputs for AI models to conduct auto-
mated bidding and using predictive AI to build audiences)
• Their use of GenAI across creative and content tasks
(such as developing creative concepts, auto-populating
briefs, and designing assets)
• Their people and process strategy (including talent
development, governance, decision making, and their
enterprise strategy for AI)
To gain deeper qualitative insights, we also conducted
more than 50 interviews with marketing decision makers,
inquiring about the impact of their AI program on ROI and
overall performance over the previous 12 months. Correlat-
ing practices with performance, we then distilled the capa-
bilities list down to the essential six and identified the
most important actions marketers can take at each stage
to forge a path to AI excellence.
BOSTONCONSULTINGGROUP 3
Exhibit1-WithTheseSixActions,LeadersAchieve60%Higher
RevenueGrowth
(Percentage of leading AI marketers who take each action)
Measurement & Insights Media & Personalization Creative & Content People & Process
24% 24% 35% 20% 9% 10%
Data Robust Media Audience AI-Powered X-Functional
Foundations Testing Budgeting Strategy Workflows Advocates
Derive insights Accelerate Run Generate Use GenAI across Build
using integrated testing with outcomes-based real-time audiences creative tasks cross-functional AI
customer view GenAI media allocation advocates
The payoff
60%
higher revenue growth than peers
Source: Google/BCG,“PathtoAIExcellence,”September2024,Global,n=2,135.
Note: RespondentsincludemarketingAIdecisionmakers/influencersatsmalltolargecompanies.
They have an integrated view of the customer. The An insurance company’s experience illustrates AI’s dra-
modern consumer journey is less linear and more varied matic impact on speed. The company used GenAI to ana-
than ever. Leaders understand that AI can help them ad- lyze various data sources, propose testable hypotheses,
dress that complexity, starting with a comprehensive view and design structured experiments to reduce bias. Tests
of the customer—a view built on first-party data that feeds are now out into the market in half the time, and the
AI models and connects to media activation. company reduced its analysis time from more than eight
hours to 30 minutes.
Take, for example, the bank that wanted to present person-
alized offers but lacked a unified customer view. It integrat- They shift budget allocations dynamically to max-
ed first-party data from branch interactions, call logs, and imize outcomes. Capitalizing on opportunity requires
emails and then connected systems with segmentation, fast action and spending efficiency—the ability to redirect
rules, and content assignments. Predictive AI models limited resources to where they will have the greatest
powered cross-selling campaigns based on the individual effect. AI’s ability to automate budget shifts unlocks new
customer’s history. The bank was able to cut offer launch potential for both efficiency and effectiveness. Some 35%
times by more than 60% and test as many as seven vari- of companies we surveyed are able to shift budgets across
ants for different use cases. platforms and channels, dynamically adjusting their mar-
keting allocations to take advantage of new opportunities.
They accelerate test design, execution, and analysis.
Fewer than 25% of respondents have tapped into AI’s One e-commerce company used a predictive AI model to
potential to support always-on testing by accelerating forecast outcomes based on channel allocations (e.g., paid
insights. Leaders, however, are using AI to create new search, social media) and outreach timing. The company
campaign ideas with greater speed and volume than applied these forecasts to inform its decisions about cou-
traditional approaches. Consumer and B2B marketers pons, promotions, and loyalty points and tracked results
alike are creating breakthrough models and accelerating quarterly. It cut media budget-planning time by 66% and
their testing. Because it can quickly produce content increased brand awareness by 11%.
variants, AI helps companies scale high-impact
strategies—in other words, invest in what works.
4 THEBLUEPRINTFORAI-POWEREDMARKETING
They’re able to reach real-time audience segments To foster cross-functional partnerships, one global consumer
with personalized messages. AI helps marketers define packaged goods company adopted a “squad” model that
the most valuable audiences based on real-time signals spanned marketing, finance, demand planning, and trade,
and target them with the right messages. While more with support from legal, R&D, and sales. This approach
than half of the companies surveyed define their strategic reshaped demand and improved brand positioning, helping
audiences based on precise signals (such as proximity to to unlock personalization across the consumer journey.
purchase), only 20% have integrated real-time, AI-powered
segmentation into their activation strategies.
ChartingthePathfromVisiontoValue
Consider the B2B software company that wanted to devel-
op new audiences without raising costs. It used its custom- From our survey analysis and qualitative findings, we devel-
er data platform to group high-value customers and pros- oped a data-derived view of the path to AI marketing suc-
pects, creating a seed audience. This seed audience was cess. For each of four maturity stages—essentials, scaling,
fed into a predictive AI model, which generated new look- leading, and transforming—we identify the most high-im-
alike audiences—expanding the company’s outreach and pact actions to take. (See Exhibit 2.)
cutting lead costs by 25%.
Stage 1: Essentials—Building the AI Foundation. At
They use AI to develop creative throughout the whole this stage, companies address gaps in foundational capa-
creative lifecycle. Although 48% of companies surveyed bilities, implementing AI-ready data management tools
frequently use GenAI for tasks like copywriting or develop- and testing AI-powered campaigns with first-party data.
ing taglines, only 9% have adopted AI for their end-to-end They redesign creative processes to translate hero assets
creative workflow. More than half lack content manage- into modular, channel-agnostic content to be read by AI.
ment systems, so they are unable to organize creative in a They use built-in AI tools within ad platforms to make fast
way that AI can access. Leaders, however, take advantage of progress. Further, they establish a view from the top about
AI’s ability to dramatically accelerate the creative process, where in the organization AI can add the most value and
improving both its speed and quality. engage stakeholders across functions for early support.
Leaders take advantage of AI’s ability to dramati- Stage 2: Scaling—Implementing Top-Priority Use
cally accelerate the creative process, improving Cases. In the scaling stage, organizations focus on ex-
both its speed and quality. panding AI-powered use cases across media and creative,
ensuring content compliance and performance tracking.
An online retailer, for example, used AI to analyze perfor- They expand the scope of the data fed to AI, improving the
mance data on hundreds of creative assets, probing the granularity and speed of insights for decision making. In
effectiveness of such elements as color and calls to action. addition, they introduce balanced marketer-AI workflows
The insights they gained improved add-to-cart rates and backed by responsible AI governance.
conversions, and then included in new briefs so the cre-
ative team could use them to develop and test campaigns Stage 3: Leading—Developing Leading Capabilities
and predict performance pre-launch. Development time to Integrate AI Workflows. This stage involves building
fell by 75%. real-time audiences and targeting them by shifting funds
fluidly across channels as needed. Leaders increase the
They’re building an AI culture by enlisting advocates volume and relevance of creative by using AI throughout
across core functional areas. Currently, only 26% of com- the creative lifecycle, from ideation to measurement. They
panies enlist four or more key functional areas in their AI use predictive AI to forecast outcomes and inform media
initiatives. But AI marketing leaders understand that scaling activation efforts. Finally, marketing teams collaborate with
AI successfully calls for building strong partnerships with stakeholders to implement a next-generation talent strate-
virtually every core function. These partnerships are essen- gy that supports the new AI-driven processes.
tial for promoting new workflows, securing funding, and
implementing talent strategies. IT, for example, overhauls
legacy systems, streamlines processes, and redesigns tech
infrastructure to support future needs. Finance validates AI
business cases, ensuring sustained resource allocation. HR
secures the talent with the necessary AI marketing skills.
Data engineering advises on buy-or-build tech decisions,
while legal assesses data privacy, IP, and compliance risks—
crucial tasks in a rapidly evolving environment.
BOSTONCONSULTINGGROUP 5
Only 9% of marketers have
adopted AI for their end-to-end
creative workflow.
Exhibit2-TheMostImportantActionsatEachStageofthePathto
AIExcellence
Transforming
Leading
Critical near-term goal
Create an integrated AI
Essentials Scaling Use predictive AI to flywheel that . . .
forecast outcomes;
experiment, feeding
Measurement Establish foundational data A dan ta al ;y iz me pa r ob vro ea id ne sir g s hc to pe of insights to media activation . t r. e. nle dv se ara ng de s g aA tI h t eo r sp ir ne sd ii gc ht ts
& Insights practices & KPIs granularity & speed Build & target real-time in real time . . .
audiences; make fluid
Scale AI-powered allocations based on AI . . . Auto-populates
Media & Start testing campaigns & personalize insights campaign plans and
Personalization AI-powered campaigns with first-party data high-value audiences . . .
Produce more (and more
relevant) creative; use AI . . . Generates 1:1 content
Creative Build modular, Pilot GenAI for creative end-to-end automatically with AI based
& Content channel-agnostic content across text, image, on social learning . . .
and video
Adjust organizational
Fully embeds AI across
structure and
re-imagined marketer + AI
People Define the most value-adding Introduce balanced upskill people workflows
& Process areas; get C-suite approval; marketer-AI workflows;
engage functional areas identify talent needs
Percentage of
38% 43% 19% <1%
respondents
in stage
Source: Google/BCG,“PathtoAIExcellence,”September2024,Global,n=2,135.
Note: RespondentsincludemarketingAIdecisionmakers/influencersatsmalltolargecompanies.
Stage 4: Transforming—Creating a Transformative To get started, CMOs can follow a few simple steps:
Marketer + AI Flywheel. No company has truly reached
this stage, as technical limitations still exist. But in the 1. Conduct an AI excellence assessment to understand
next five years, marketers and machines will increasingly your starting point relative to peers.
play complementary roles to create a flywheel whereby
AI and marketers together power campaign planning and 2. Identify key steps in your end-to-end workflow where
execution. (See Exhibit 3.) Marketers will drive transforma- AI could play a role (for example, in measurement and
tion by embedding AI into reimagined workflows, defining insights, media, or creative).
strategy, priority audiences, and business goals. AI will play
a much greater role in analysis and execution and in sup- 3. Set two to three AI goals for the quarter (such as estab-
porting strategy development. It will predict trends, gather lishing your data foundations or piloting three use cases
insights in real time, auto-populate campaign plans, select with key partners using off-the-shelf tools).
high-value audiences for 1:1 messaging, generate personal-
ized content, and suggest real-time adjustments. Leading 4. Launch a cross-functional AI task force that includes
companies believe they are twice as close to achieving this marketing, engineering, IT, finance, HR, legal, and
future vision than their peers. agency partners.
In the next five years, marketers and machines Reaching higher levels of AI integration requires more than
will increasingly play complementary roles to just deploying the latest technologies and tools. Marketers
create a flywheel that powers campaign planning must rethink their role, shifting their emphasis from the
and execution. tactical and executional to the strategic. They must become
orchestrators, guiding AI to deliver better outcomes while
trusting the machine to handle the complexity of execution.
BOSTONCONSULTINGGROUP 7
AI leaders see 60% higher
revenue growth.
Exhibit3-TheMarketer-AIFlywheelFreesMarketerstoFocusMore
onStrategy
The Marketer + AI Vision
Marketer
Building to
Derive insights Generate the future
using integrated real-time audiences
NOW
customer view
Leading
Provide AI marketers are
strategic
Measure & 2x
input & machine monitor
plan to
outcomes
maximize
Run
outcomes Accelerate
outcomes-based
testing
media allocation closer to
with AI
achieving
this vision
than peers
Use Gen AI across
creative tasks
Build
&
c r reo is ms- afu gn inc eti o wn oa rl
k
A flI
o
a wd
svo c
ate s
Source: Google/BCG,“PathtoAIExcellence,”September2024,Global,n=2,135.
What’sontheHorizon? By harnessing the power of AI effectively, CMOs can not
only improve the efficiency of their operations; they can
Although the AI revolution in marketing is still in its early also unlock new avenues for growth. The path to AI
stages, the sophistication and quality of AI output is im- excellence—and to realizing the power of the marketer-AI
proving literally by the week. As technology advances, the flywheel—is clear. Marketers who take bold steps today
gap between leaders and laggards will only widen; and at will be shaping the future of their organizations.
the speed of change we’re witnessing, that gap may—
sooner rather than later—become unbridgeable. Reach out to the team for information on how to
take an AI self-assessment.
Over the next 12 months, leaders are planning to expand
AI use cases: 46% will use GenAI for video creative, 36% This content is the result of a joint research effort between Google
will launch AI chatbots (internal, customer-facing, or both), and BCG.
and 35% will customize creative by platform. As the num-
ber of advanced use cases grows, the system becomes
more sophisticated and faster.
BOSTONCONSULTINGGROUP 9
AbouttheAuthors
Derek Rodenhausen is a managing director and partner Ray Yu is a managing director and partner in BCG’s
in the New York City office of BCG. You may contact him at Atlanta office. You may contact him at [email protected].
[email protected].
Trevor Sponseller is a principal in BCG’s New York City Paola Francesca Scarpa is a managing director and
office. You may contact him at [email protected]. partner in the Milan office of Boston Consulting Group.
You may contact her by email at [email protected].
Henry Leon is a managing director and partner in the Javier Perez Moiño is a managing director and partner in
firm’s London office. You may contact him by email at BCG’s Madrid office. You may contact him by email at
[email protected]. [email protected].
Val Elbert is a managing director and partner in the firm’s Jordan Baker is a project leader in BCG’s New York City
New Jersey office. You may contact him by email at office. You may contact her by email at
[email protected]. [email protected].
ForFurtherContact
If you would like to discuss this report, please contact the
authors.
10 THEBLUEPRINTFORAI-POWEREDMARKETING
Boston Consulting Group partners with leaders in business For information or permission to reprint, please contact
and society to tackle their most important challenges and BCG at [email protected]. To find the latest BCG con-
capture their greatest opportunities. BCG was the pioneer tent and register to receive e-alerts on this topic or others,
in business strategy when it was founded in 1963. Today, please visit bcg.com. Follow Boston Consulting Group on
we work closely with clients to embrace a transformational Facebook and X(formerlyknownasTwitter).
approach aimed at benefiting all stakeholders—empowering
organizations to grow, build sustainable competitive © Boston Consulting Group 2024. All rights reserved. 12/24
advantage, and drive positive societal impact.
Our diverse, global teams bring deep industry and functional
expertise and a range of perspectives that question the
status quo and spark change. BCG delivers solutions
through leading-edge management consulting, technology
and design, and corporate and digital ventures. We work
in a uniquely collaborative model across the firm and
throughout all levels of the client organization, fueled by
the goal of helping our clients thrive and enabling them
to make the world a better place.
bcg.com
12 THEBLUEPRINTFORAI-POWEREDMARKETING |
130 | gartner | gen-ai-planning-workbook.pdf | GenAI Planning
Workbook
4 steps to implementing
generative AI in your enterprise
© 2023 Gartner, Inc. and/or its affiliates. All rights reserved.Gartner is a registered trademark of Gartner, Inc. or its affiliates.This presentation, including all supporting materials,
is proprietary to Gartner, Inc. and/or its affiliates and is for the sole internal use of the intended recipients. Because this presentation may contain information that is confidential,
proprietary or otherwise legally protected, it may not be further copied, distributed or publicly displayed without the express written permission of Gartner, Inc. or its affiliates.
Focus GenAI conversations on real business
problems and achievable use cases
Generative AI (GenAI) is suddenly on everyone’s
radar, but some organizations already have
extensive experience and success in deploying
AI techniques across multiple business units
and processes.
Gartner research shows these mature AI
organizations represent just 10% of those
currently experimenting with AI, but would-be
GenAI adopters can learn a lot from them.
Use this planning workbook to focus
conversations among business and IT leaders
around best practices that help you focus on
GenAI initiatives that are both valuable and
feasible. To get there, take a strategic approach.
RESTRICTED DISTRIBUTION
2 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved.
Actions related to the 4 pillars of GenAI strategy
Establish your vision for GenAI Remove barriers to
capturing value
How GenAI will drive your
enterprise goals, what benefits What organizational barriers
you expect and how you will could hinder your success and
measure success. what actions are needed to
remove those hurdles.
Prioritize adoption Identify the risks
Which are the best GenAI What regulatory, reputational,
initiatives to pursue, based on competency, technology and
their value and their feasibility — other risks you may need to
as agreed to by both IT and assess and mitigate.
business leaders.
RESTRICTED DISTRIBUTION
3 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved.
Key components of your AI strategy framework
Vision Value Risks Adoption
• Goals • Business impact • Regulatory • Use cases and
value maps
• Benefits • Change • Reputational
management • AI decision
• Success metrics • Competency
framework
• People and skills
• Decision
governance
RESTRICTED DISTRIBUTION
4 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved.
AI VISION
First, state clearly how GenAI objectives
link to enterprise goals
Don’t underestimate the need to level-set with
stakeholders from the outset:
Stating AI goals clearly is key
1. Restate the corporate vision of your
to encouraging and enabling
enterprise:
organizationwide fluency and
“……………………………………………”
adoption of AI. It will also help
2. State how AI will support that vision:
you to fund the right use cases —
ones that will deliver clear return
– e.g., AI will enable better business value
on investment and lead to further
in these areas in these ways
innovation.
– e.g., We will use AI to achieve fairer
outcomes
RESTRICTED DISTRIBUTION
5 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved.
AI VISION
Then, specify how GenAI will drive business goals
Ask why you are pursuing GenAI and what value you expect it to bring based on your major business goals, how you will
measure success and what use cases could maximize that value (you will verify the value/feasibility of those use cases in the
“adoption” phase).
Illustrative
Use Cases to Pursue
Goal How AI/GenAI Enables That Goal
(Illustrative Examples)
Topline revenue growth Business model change inspired or supported by AI creates Behavioral analytics, contract life cycle
net-new business initiatives. management
Improved customer Greater ability to conduct customer behavior analytics Virtual customer assistants
satisfaction increases proximity to the customer.
Reduced costs Task and process automation reduce operational costs. Risk/fraud mitigation, asset performance
management
Staff augmentation and Augmented AI and automation increase productivity by shifting Knowledge management and training, content
increased productivity people away from managing mundane tasks. generation, code generation
Improved service Data-driven predictive analytics tools advance digital services. Predictive maintenance, proactive threat
availability management
RESTRICTED DISTRIBUTION
6 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved.
AI VISION
Also, set AI success metrics
To measure the value of individual use cases, you’ll need success metrics that tie into your overarching business
goal. Select metrics like those listed here that relate to specific key success factors and provide a timeframe in
which you expect to demonstrate value.
Business Goal Appropriate Success Metric Completion
Improved customer satisfaction Customer satisfaction index/Net Promoter Score Date
Topline revenue growth Revenue growth for product lines Date
New business initiatives Number of new business initiatives Date
Task or process automation Reduction in processing time Date
Reduce costs Reduction in CapEx and OpEx Date
Staff augmentation and increased Workforce productivity metrics, such as time spent on value-added tasks Date
productivity
Improved service availability % of annual availability Date
RESTRICTED DISTRIBUTION
7 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved.
AI VALUE
Remove organizational barriers to capturing value
Having identified potential benefits to the business (in the vision stage), surface any strategic concerns that could hinder your
ability to capture value in the way you have identified it. Also, identify solutions, responsibilities and actions as illustrated here.
Executive(s) What the Organization
Strategic Concern Solution
Responsible Will Do
Projects aligned to corporate Document goals and require a CIO • Indicate which corporate goals should be
goals are more likely to succeed portfolio approach to AI opportunity. addressed.
and mature. • Size portfolio (five or fewer pilots and
minimum viable products).
Metrics deliver credibility for Select metrics as proxies for financial CFO • Collaborate with your chief data and analytics
project maturity. and risk results or direct such officer to discuss what will be most
measurements. measurable and educational for future
projects.
Formal structures of accountability Help complete a RACI (responsible, Chief data (and • Draft a RACI matrix for all aspects of AI
bolster AI results. accountable, consulted and analytics) officers, project and product development.
informed) matrix for AI strategy CIO
development and execution.
RESTRICTED DISTRIBUTION
8 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved.
AI RISKS
Assess and mitigate risks
Any type of AI comes with a range of risks, including those illustrated here. GenAI carries specific new types of risks, such
as hallucinations and biased and inaccurate results. Log all such major risks so you can properly assess and mitigate each.
Key Types Risk Executive(s)
Action Plan
of Risks Category Responsible
Regulatory Adhere to CIO/CTO and Understand the continuously Enable collaboration between Create an AI governance
regulations CRO evolving regulatory landscape. AI practitioners and legal, risk office, which serves an
and security members to independent audit committee
evaluate use case feasibility to review results.
and acceptable risks.
Reputational Secure and CIO/CTO Acknowledge the threats Bolster security across Leverage external resources
safe against AI posed by both enterprise security controls, to help secure your AI
malicious and benign actors data integrity and AI model systems.
in your organization. monitoring.
Competencies Technical debt CIO/CTO Align AI strategy with cloud Create a technology roadmap Create a startup accelerator
strategy and explore cloud to modernize data and program to reduce technical
as foundation for AI. analytics infrastructures to debt and innovate
align with AI goals and incrementally.
timeline.
RESTRICTED DISTRIBUTION
9 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved.
AI ADOPTION
Prioritize projects that are valuable and feasible
Rate the feasibility and value of each project using simple criteria like those shown here, and actually score each so you can rank projects
against one another. Typically, executives are keen to pursue initiatives where value is high (and risk also tends to be high, i.e., feasibility is
low) but avoid projects where feasibility is so low that it makes the project impossible. A use case with a seemingly outstanding contribution
to business value and strong feasibility is either a breakthrough, or the market is missing a great opportunity.
TECHNICAL FEASIBILITY FACTORS BUSINESS VALUE FACTORS
Overall
Business Overall Technical
Architecture Have Skills/ Aligns With
Access to Sponsor KPIs Value Feasibility
Project andTechnology People to Our Mission Ranking
Labeled Data Support Measurable (Scale of 1 to (Scale of 1 to 10;
Feasibility Execute andValues
10; 10 Being 10 Being High)
High)
Name Yes/Maybe/No Yes/Maybe/No Yes/Maybe/No Yes/Maybe/No Yes/Maybe/No Yes/Maybe/No
Name Yes/Maybe/No Yes/Maybe/No Yes/Maybe/No Yes/Maybe/No Yes/Maybe/No Yes/Maybe/No
Name Yes/Maybe/No Yes/Maybe/No Yes/Maybe/No Yes/Maybe/No Yes/Maybe/No Yes/Maybe/No
Name Yes/Maybe/No Yes/Maybe/No Yes/Maybe/No Yes/Maybe/No Yes/Maybe/No Yes/Maybe/No
Name Yes/Maybe/No Yes/Maybe/No Yes/Maybe/No Yes/Maybe/No Yes/Maybe/No Yes/Maybe/No
RESTRICTED DISTRIBUTION
10 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved.
Actionable, objective insight
Position your IT organization for success. Explore these additional
complimentary resources and tools for IT leaders:
Resource Center Webinar eBook eBook
The Top Generative AI Beyond the Hype: The Augmenting Decisions Essential Guide to
Questions Answered Practical Applications With Artificial Data Fabric
by Gartner Experts & Use Cases of Intelligence
Find out why data fabric belongs
Generative AI in your data management
Access benefits, applications Decision automation can drive
thinking.
and risks of generative AI. competitive advantage. Know
Explore the future of
when and how to use it.
generative AI and understand
the many use cases.
Learn More Watch now Download Now Download Now
Already a client?
Get access to even more resources in your client portal. Log In
RESTRICTED DISTRIBUTION
11 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved.
Connect With Us
Get actionable, objective insight to deliver on your most critical priorities. Our expert guidance and
tools enable faster, smarter decisions and stronger performance. Contact us to become a client:
U.S.: 855 811 7593
International: +44 (0) 3330 607 044
Become a Client
Learn more about Gartner for IT Leaders
gartner.com/en/information-technology
Stay connected to the latest insights
RESTRICTED DISTRIBUTION
12 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved. |
134 | ibm | ibm-5-trends-for-2025-report.pdf | IBM Institute for Business Value | Research Brief
5 Trends
for 2025
Ignite innovation with
people-powered AI
Introduction
AI democratizes data—and 2024 was a year of letting go. As a combination of conflict and transformation threw old
assumptions into doubt, leaders had to reassess their appetite for risk. They had to weigh
the need for speed against the safety of proven processes—then change the habits that were
redefines decision-making.
holding them back.
Generative AI was at the center of this shift, introducing a world of new opportunities, as well
How can leaders empower people to innovate as uncharted risks. Agentic AI, which refers to systems and programs that perform a variety
of functions autonomously, can act on behalf of employees while they do other work. By
without putting the business at risk? giving AI agents specific permissions and rights, they can automate decision-making,
problem-solving, and other tasks that go beyond the data the system’s machine learning
models were trained on in a way that most AI assistants don’t.
And as digital labor evolves, it puts the power of transformation firmly in employee’s hands.1
It makes it possible for individuals to increase productivity and redefine workflows—and
challenges preconceived notions about what it means to lead.
5 Trends for 2025: Ignite innovation with people-powered AI 2
Figure 1
The fact is, leaders don’t have time to vet every innovation. As agentic AI augments roles 2024 2025
across the organization, they need to delegate more decisions to truly pick up the pace.
Leaders still need to define the destination—and the rules of the road—but they must
empower teams to rethink workflows and deploy AI agents in new ways to improve
performance at scale.
6%
In this environment, leaders are walking a tightrope between agility and security, striking
Experimenting 30%
a balance between resilience and risk. It’s no easy feat. To learn how they’re gaining ground,
the IBM Institute for Business Value (IBM IBV), in partnership with Oxford Economics,
surveyed 400 global leaders across 17 industries and six geographies in October and
November 2024. We asked them about the challenges they must overcome to succeed
in an AI-fueled competitive landscape, how they’re preparing their people to drive change,
and what opportunities they expect to accelerate progress most. Scaling & Optimizing 44% 46%
We paired these results with the insights we’ve gained from dozens of surveys, in-depth
interviews, and client engagements conducted in 2024 to map out the trends that will
reshape the AI roadmap in 2025 (see “Research methodology,” page 5).
We found that leaders are still struggling to transform the business with their AI
investments—but they believe they’re on the cusp of a major breakthrough. In fact, 63% Innovating 24% 44%
of executives say their AI portfolio will have a material financial impact on their organization
in the next one to two years.
Source: 5 Trends for 2025 global pulse survey. Q: Which of the following best
describes your organization’s approach to adopting AI, today and next year?
Note: Sums don’t equal 100% because “reevaluating” and “none of the
above” were also potential responses.
5 Trends for 2025: Ignite innovation with people-powered AI 3
To deliver on these expectations, organizations plan to push teams forward at a rapid clip.
Today, 30% of executives say their organizations are primarily experimenting with AI, testing
In the coming year, it’s likely that some organizations will begin to set themselves
its use in low-risk, non-core functions to gain experience, build confidence, and identify
apart. Will yours be one of them? Explore these five trends for 2025 to learn what
potential pain points. Only 24% say they’re innovating with AI to advance new opportunities
leaders need to know to overcome the obstacles that lie ahead—and what they can
and create new business models.
do to gain a competitive edge.
In 2025, leaders expect to see a major shift. 46% of executives say their organizations will be
scaling AI, using it to optimize existing processes and systems, while 44% expect to use AI
to innovate. Only 6% say their organizations will still be experimenting.
1 Agentic AI will transform your 4 The rapid pivot to AI has
To turn that momentum into real business value, leaders will need to empower people to business—but first you must upended IT budgets, but
make the most of the technology at their fingertips. That means democratizing decision- reskill your people. self-funding is imminent.
making and giving people the tools and training they need to succeed. People are the secret
ingredient to winning with AI—but they can’t succeed without strategic reskilling, security
guardrails, and decision support. 2 Despite efforts to slow its 5 AI product and service
growth, technical debt innovation is the #1 CEO
continues to increase. goal, yet business models
aren’t keeping up.
3 In the age of AI, location
is everything.
5 Trends for 2025: Ignite innovation with people-powered AI 4
Research
methodology
This study is part of the IBM IBV’s “Five We also surveyed 400 global leaders Each edition of “Five Trends” highlights the
Trends” series, now in its sixth year, which across 17 industries (including banking, key challenges and opportunities expected
offers strategic insights to help organizational government, and telecommunications) to drive significant business impact in the
leaders plan for what’s next. and six geographies (the US, the UK, coming year. This year’s report identifies
Germany, India, Australia, and Singapore) the trends that will shape industries and
Based on comprehensive research
in October and November 2024. organizations in 2025, providing actionable,
conducted over the past 12 months,
Participants were asked a range of research-backed insights—based on
the report draws on data from 55 surveys
questions about forward-looking business in-depth research and comprehensive
covering hybrid cloud and AI, general
and technology strategies in various client engagements—to help leaders
business, finance and technology, and
formats (multiple choice, numerical, and navigate and thrive in an increasingly
specific industries, including insights from
Likert scale). Any datapoints not otherwise complex and dynamic environment.
more than 43,000 executives and 4,000
cited were sourced from this five trends for
consumers worldwide. This portfolio of
2025 global pulse survey. Sample size was
research was used to inform the trends
too small to support group comparisons
we explored in a pulse survey conducted
within this set of participants.
by the IBM IBV, in partnership with
Oxford Economics.
5 Trends for 2025: Ignite innovation with people-powered AI 5
Agentic AI will transform
your business—but
first you must reskill
your people.
The future of work is being rewritten with AI. But many
employees are unprepared for what comes next—and
progress will stall if too many are left behind.
5 Trends for 2025: Ignite innovation with people-powered AI 6
While roughly 5% of the global workforce
consistently needs to be reskilled each
While roughly 5% of the global workforce
year, the rapid evolution of AI has sent
consistently needs to be reskilled each
this figure skyrocketing. In 2024, global
year, the rapid advancement of AI has
sent this figure skyrocketing. CEOs estimated that, on average, 35%
of their workforce needed to be reskilled.
That translates to more than a billion
workers worldwide.2
What exactly is creating this chasm? The
escalating need for true transformation.
Instead of automating specific roles
wholesale, organizations are pairing people
with domain-specific AI agents to improve
their performance. In fact, 87% of executives
expect jobs to be augmented rather than
replaced by generative AI.3 This means,
rather than learning a new skill or tool,
workers must completely rethink how they
do their jobs to make the most of gen AI.
In 2024, global CEOs estimated that, on
average, 35% of their workforce needed
to be reskilled. That translates to more
than a billion workers worldwide.
Source: The 2024 CEO Study: 6 hard truths CEOs must face. IBM Institute for Business Value.
5 Trends for 2025: Ignite innovation with people-powered AI 7
In this environment, 64% of CEOs say that succeeding with AI will depend more on people’s This “third wave” of AI promises to transform workflows wholesale.7 In fact, nine in 10
adoption than the technology itself.4 However, 64% say their organization must take executives now say their organization’s workflows will be digitized with intelligent
advantage of technologies that change faster than people can adapt5 —and 47% of executives automation and AI assistants by 2026—and 77% of executives believe gen AI will enable
say their people lack the knowledge and skills to effectively implement and scale AI across connected assets to make autonomous decisions by 2026.8 Executives also report that the
the enterprise. volume of decision-making by digital assistants will increase by 21% in the next two years
due to generative AI.9 This will have huge implications for operating models, as organizations
Part of the problem is insufficient training. While executives say AI literacy is the most critical
must create new structures that give employees oversight over autonomous decision-
capability their workforce will need in 2026,6 only 22% strongly agree that their organization
making—and manage the new risks it creates.
has integrated AI knowledge, skills, and abilities into employee professional development
plans. And less than half say their organization has implemented a formal change management It’s a lot to work through, but 67% of CEOs say the potential productivity gains from
program to enable the integration of AI assistants and agents into daily workflows. automation are so great that they must accept significant risk to stay competitive.10 What’s
more, 82% of executives agree that the benefits they expect from gen AI will exceed the
That’s a big issue, since agentic AI is quickly transforming the role of individual contributors.
potential risks.11 But employees will need targeted training and skills development to deliver
As simple AI assistants are supplemented by AI agents with more advanced capabilities,
on this promise—and deliver the competitive advantage that executives expect.
employees will need to manage entire teams of agents that are completing tasks
autonomously—and learn how to work with chat-based supervisory AI agents that can help
streamline this process.
Organizations must create new structures
that give employees oversight over
autonomous decision-making—and manage
the new risks it creates.
5 Trends for 2025: Ignite innovation with people-powered AI 8
What to do
1 2 3
Make AI literacy a must-have—and Unlock your team’s collective genius. Future-proof your workforce.
double down on agentic AI skills. Demolish siloed thinking and establish Establish new roles, such as process
Launch comprehensive educational collaboration sandboxes where AI-enabled orchestrators and digital librarians, to
initiatives blended with hands-on projects workflows can be rigorously tested and manage how AI assistants, models, and
aimed at rapidly advancing AI literacy and refined—encouraging people to get their governance guidelines are used and shared
getting teams comfortable with agentic hands dirty without fear of failure. Hold across the organization. Introduce checks
AI. Mandate AI skills training across all leaders from business units, IT, and HR and balances that provide oversight for
roles and create a culture where AI jointly responsible for AI outcomes to autonomous decisions made by agentic
proficiency is non-negotiable to enable underscore the strategic importance AI. Regularly host hackathons that bring
smarter collaboration and responsible of enterprise-wide adoption. Make together diverse perspectives to
integration of AI agents and assistants governance integral to collaborative conceptualize creative uses of AI assistants
into everyday workflows. innovation efforts and reimagine the and agents. Establish performance- and
operating model to integrate agentic readiness-based compensation and
AI effectively and responsibly. incentives that align with business goals
and gen AI adoption priorities.
5 Trends for 2025: Ignite innovation with people-powered AI 9
Despite efforts to slow its
growth, technical debt
continues to increase.
Time is money. And leaders are always looking for
ways to save both. But the workarounds that accelerate
transformation in the short term often create technical debt
that limits long-term innovation and growth.
Technical debt refers to the long-term costs and inefficiencies caused by quick, suboptimal
technical decisions made to expedite development or delivery. And the growing demand
for digital products, services, and experiences is compounding this debt much faster than
organizations can address it. As a result, 55% of executives now say technical debt is either
a major obstacle or a real roadblock to achieving business goals.12
5 Trends for 2025: Ignite innovation with people-powered AI 10
Think about the automotive industry. While the lifetime of a car could be 15 years
or more, the digital experience in the driver’s seat is often outdated within 18 months.
If manufacturers don’t design and install software in a way that can easily be updated
as technology evolves, customer satisfaction will suffer.13
The same is true for enterprise IT. To deliver
the innovations that customers, employees,
and partners expect, organizations must
77%
build solutions within a modern architecture.
That’s because traditional systems don’t
tend to play well with next-gen apps,
software, and infrastructure.
of executives say they need
to adopt gen AI quickly to
This is particularly relevant for generative AI
keep up with competitors.
and agentic AI. Organizations need robust
infrastructure that can handle the data and But only
computational requirements of AI to go from 25%
pilots to enterprise-wide solutions. Yet,
while 77% of executives say they need to
adopt gen AI quickly to keep up with
competitors14 —only 25% strongly agree that of executives strongly agree
that their organization’s IT
their organization’s IT infrastructure can
infrastructure can support
support scaling AI across the enterprise.
scaling AI across the enterprise.
Sources: The 2024 CEO Study: 6 hard truths CEOs must face. IBM
Institute for Business Value; 5 Trends for 2025 global pulse survey.
5 Trends for 2025: Ignite innovation with people-powered AI 11
One of the biggest barriers is the quality, accessibility, and security of enterprise data. Flexibility must be part of the equation, as well. To get the greatest benefit from
Training gen AI models with internal and proprietary data is critical to help organizations gain cloud-based technologies, organizations need to be able to run each system and
an edge with them. Yet, only 16% of tech leaders are confident that their current cloud and application in the right public or private cloud environment—an approach we call hybrid
data capabilities can support gen AI,15 and just 21% of executives strongly agree that their by design. On average, IT executives from companies adopting hybrid by design for their
organization has the data it needs to scale AI across the enterprise. digital transformation efforts reported 3X higher ROI than those that don’t.16
To scale AI systems and incorporate agentic AI without compounding technical debt, That’s a pretty strong argument for planning ahead. But today, two-thirds of CEOs say they’re
organizations must incentivize teams to modernize traditional systems and change the way meeting short-term targets by reallocating resources from longer-term efforts.17 If leaders
they develop new solutions. By linking long-term productivity gains and performance metrics don’t shift this mindset, technical debt could preclude progress, even if quick wins drive
to every new solution, CIOs, CFOs, and other key leaders can measure the potential benefits of growth or profitability today.
modernization and put a price tag on taking shortcuts. When these leaders join forces, they can
help teams decide when accumulating technical debt for the sake of speed makes sense—and
when it’s better to build the right architecture from the start, especially as upskilling and
reskilling efforts increase productivity over time.
By linking long-term productivity gains and
performance metrics to new solutions, leaders
can measure the benefits of modernization—
and put a price tag on shortcuts.
5 Trends for 2025: Ignite innovation with people-powered AI 12
What to do
1 2 3
Bridge the gap between Incentivize scalability. Architect for agility.
vision and reality. Empower IT leaders to educate the Establish a nerve center that is focused
Identify the missing architectural pieces business on the full cost associated with on designing solutions for modularity and
needed to succeed with AI at scale. the tech architecture required to scale AI. scalability and is charged with deploying
Connect AI business cases to associated Quantify the cost of taking shortcuts—and each AI use case in the most appropriate
modernization costs to avoid unexpected the business value that comes with environment. Build a composable
expenses. Intentionally invest in the AI developing pilots that can quickly scale. platform that decouples models, tools,
initiatives that will deliver the most Celebrate teams that think holistically infrastructure, and apps, creating
business value in the long term, establish about AI innovation—and propose projects flexibility and cost-effectiveness in your AI
a cross-functional AI board responsible for that limit the creation of future technical ecosystem. Prioritize making high-quality
determining ROI from a line-of-business debt—to change the organization’s data accessible across platforms.
perspective, and develop a workforce behavioral economics.
strategy that helps your people innovate
without increasing tech debt.
5 Trends for 2025: Ignite innovation with people-powered AI 13
In the age of AI, location
is everything.
Perpetual disruption is here to stay. But that doesn’t mean
it’s predictable. To navigate complexity wherever it rears its
head, leaders must be able to see the big picture—and the
market-level minutiae—in one sweeping view. They must
strategically adjust operations based on market-level shifts,
without overreacting to local disruptions as they occur.
5 Trends for 2025: Ignite innovation with people-powered AI 14
Trend 3
And striking the right balance is getting
harder every day. Looking to the future, 60%
of government leaders believe that shocks
are likely to increase in frequency and 70%
believe they’re likely to increase in intensity
and impact.18 This is forcing business leaders
to assess where their data is housed and
rethink how—and where—their organizations 96% 93%
should operate. In 2024, 86% of executives
said their location strategy was impacted by
of executives say data privacy, security, expect AI to impact
geopolitical disruption—and that figure is and regulations will determine where they their location
expected to rise to 93% in 2026. locate operations in 2026. strategy in 2026.
Location strategies, which define where a
company’s key resources and capabilities
reside, are also being influenced by the AI
revolution. As organizations seek out the
talent, data ecosystems, and infrastructure
needed to scale AI effectively, they’re moving
operations to places they believe will provide
the greatest strategic advantage. In 2024,
67% of executives say their organization’s
use of AI changed where it operated—and
a whopping 93% expect AI to impact their
location strategy in 2026. Similarly, 96% of
executives say data privacy, security, and
Source: 5 Trends for 2025 global pulse survey.
5 Trends for 2025: Ignite innovation with people-powered AI 15
regulations will determine where they locate operations in the next two years. However, it’s At the same time, splintering AI regulation has business leaders looking at certain aspects
important to note that many privacy regulations are not so restrictive as to necessitate data of the business through a different lens. For instance, 37% of executives say they will manage
localization. Leveraging hybrid cloud environments enables organizations to ensure compliance their data strategy and governance more regionally in 2025, with 26% saying they will take
with data privacy requirements while maintaining operational flexibility. a more global approach. Still, 69% of executives expect their organization to receive a
regulatory fine due to generative AI adoption.19
Think of it this way: As organizations expand into new markets to drive growth, they need to use
customer data to drive product development and deliver personalized experiences with AI. But As regulations become more widely defined and adopted, executives expect this risk to
they also need to comply with local regulations—and cultural expectations—regarding how AI decrease. For example, 57% of CEOs say the guidelines provided by the EU AI Act increases
is used and how private data is secured. So, they’re prioritizing markets that will offer the ideal their willingness to invest in AI.20 The predictive capabilities of gen AI can help organizations
mix of skilled talent, computing capabilities, supportive regulations, and customer demand to manage disruption, as well. In fact, 77% of executives say gen AI models can successfully
foster growth. identify geopolitical and climate risks, enabling proactive mitigation.21
As a result, 89% of executives agree with two statements about their 2025 location
strategy that may seem at odds. They say they’re widening their reach and extending
operations globally and they’re primarily focused on a few core markets. This shows that
organizations are being selective as they plan their growth strategy—but international
markets are still a priority.
As organizations seek out the talent, data
ecosystems, and infrastructure needed to
scale AI, they’re moving operations to places
that provide the greatest strategic advantage.
5 Trends for 2025: Ignite innovation with people-powered AI 16
What to do
1 2 3
Stress-test your strategy. Innovate through volatility. Bake in regulatory preparedness.
Develop AI models specifically for navigating Leverage hybrid cloud and open AI Get obsessed with documentation. Assess
global instability, using predictive analytics approaches to enable global AI strategies. where data is housed and how this could
to stay ahead of regulatory changes, supply Combine secure, scalable hybrid cloud affect operations. Ensure AI-generated
chain challenges, and shifting labor markets platforms with open-source AI frameworks assets can be traced back to the foundation
in volatile regions. Align the most crucial and to drive innovation, facilitate interoperability model, dataset, or other inputs by creating
differentiating supply chain workflows with across markets, and address compliance an inventory of every instance where AI is
your early predictive generative AI use with diverse regional regulations while being used. Seed this source information
cases. Use digital twins and simulations to fostering collaboration and inclusivity in into digital asset management and other
identify latent weaknesses and bottlenecks. AI development worldwide. systems to help teams comply with
extensive existing and emerging legislation
in areas such as data privacy, security, and
consumer protection.
5 Trends for 2025: Ignite innovation with people-powered AI 17
The rapid pivot to AI has
upended IT budgets, but
self-funding is imminent.
Generative AI has made the traditional IT budgeting process
untenable. It’s sending shockwaves through technology
and finance teams as they rush to reevaluate their spending
priorities—and move money where it’s needed most.
5 Trends for 2025: Ignite innovation with people-powered AI 18
Trend 4
75%
of business leaders are thinking of gen AI
more like an innovation investment than
traditional IT today.
Leaders know they need to invest in gen AI to keep up with the competition, but these solutions
have yet to deliver production-level ROI. This has led to widespread cannibalism of broader IT
budgets. In 2024, one in three organizations pulled funding for gen AI from other IT initiatives,
with only 18% of tech execs funding these projects with net-new spend.22 Of course, there’s
some overlap between the investments gen AI requires and other IT priorities, with executives
But
reporting that infrastructure, cloud, and data account for more than 40% of gen AI costs.23
71%
But that still leaves a large funding gap—one that executives are rushing to fill. Nearly all
executives (95%) say gen AI will be at least partially self-funded by 2026, with a focus on
of executives say gen AI
driving future profitability. While three in four business leaders are thinking of gen AI more
should be self-funded to
like an innovation investment than traditional IT today, 71% of executives say gen AI should
justify its investment.
be self-funded to justify its investment.24
Source: IBM Institute for Business Value generative
AI and innovation spend pulse (AI Academy) survey.
5 Trends for 2025: Ignite innovation with people-powered AI 19
So, what will it take to move gen AI out of the innovation sandbox and into the revenue stream? That’s one reason the average gen AI investment takes almost 14 months to deliver positive
It starts with focusing gen AI investments in the areas with the greatest potential and the ROI, compared to just 10 months for other technology investments.26 But that won’t be the
lowest-risk applications, rather than spreading funds evenly across the portfolio. In 2024, 71% case for much longer. As more organizations embrace fit-for-purpose AI models—combined
of gen AI spending went to HR, finance, customer service, sales and marketing, and IT, where with open source and agile—the cost side of the equation will start to shrink. Over the next three
investments were expected to cut costs. Only 29% went to product-related business functions, years, executives expect their AI model portfolios to include 63% more open models than they
where growth-driving innovations incubate. This makes it difficult to define business cases that use today, which will play a large role in driving down development costs.27
break the mold.25
While revenue growth has been the least effective metric for gauging gen AI success
to date, it will be the primary way businesses measure differentiation in the long term.
But to get revenue and ROI metrics where they need to be, leaders must make data-driven
decisions about which gen AI plays they expect to do the most to advance strategic
objectives—and fund them accordingly.
As organizations embrace fit-for-purpose
AI models—combined with open source and
agile—the cost side of the equation will start
to shrink.
5 Trends for 2025: Ignite innovation with people-powered AI 20
What to do
1 2 3
Unify infrastructure, amplify impact. Invest like a shark. Own the open-source advantage.
Clearly define what each AI investment is Dive into the data to identify the projects Create an open-source program office that
worth to the business—and what they will that seem most likely to deliver real manages your organization’s consumption
cost to implement. Identify the business value—then cut the dead weight of—and contributions to—open-source
infrastructure upgrades needed to roll out that’s holding you back. Commit sufficient code. Build a warehouse of open-source
solutions at scale and bundle projects that resources to successfully scale AI and code that has been carefully reviewed to
can share these costs to boost ROI. Create allocate your AI budget based on one thing: streamline access to preferred offerings.
centralized control centers that unify growth potential. Think ecosystems, not Incentivize developers to actively
isolated AI efforts across enterprise silos, and engage your most valuable contribute to open-source projects critical
functions to measure business outcomes customers and strategic IT partners in bold to your business, especially IT
more accurately. conversations about where gen AI can add infrastructure modernization, to gain
the most value. influence over the projects your
organization relies on.
5 Trends for 2025: Ignite innovation with people-powered AI 21
AI product and service
innovation is the #1 CEO
goal, yet business models
aren’t keeping up.
As generative AI supercharges innovation, the pipeline of new
products and services is bursting at the seams. But many
organizations are too wedded to old business models to tap
into new opportunities to drive growth.
5 Trends for 2025: Ignite innovation with people-powered AI 22
2023 2024
And CEOs are feeling the crunch. In 2024, they cited business model 1
Top CEO priorities Productivity or profitability Product and service innovation
innovation as the top challenge they expect to face over the next three
years—up from 10th place in 2023—while also naming product and Tech modernization Tech modernization
service innovation as their top priority for the same timeframe.28
Customer experience Cybersecurity and data privacy
Business leaders understand that, to make the most of innovative
offerings, they’ll also need to rethink how they turn a profit. Cybersecurity and data privacy Forecast accuracy
In fact, 62% of CEOs say they must rewrite their organizational Environmental sustainability Productivity or profitability
playbook to win in the future.29 AI will play a major role in this shift.
6
Product and service innovation Customer experience
Over the next three years, 85% of executives say AI will enable
business model innovation and 89% say it will drive product and
service innovation.30 What does this look like? It starts with analyzing
1
customer and market data faster and more comprehensively than ever Top CEO challenges Environmental sustainability Business model innovation
before—then changing strategies to keep up with shifting demands.
Cybersecurity and data privacy Productivity or profitability
This will require centering business models on the careful design of
human-machine interaction—and building strong supporting Tech modernization Scalability of service delivery
governance structures—as well as rethinking organizational structures
Talent recruiting and retention Marketing and sales effectiveness
and workflows.
Diversity and inclusion Forecast accuracy
Getting it right can help companies stay ahead of the competition—and
strengthen customer relationships. For example, nine in 10 executives Forecast accuracy Environmental sustainability
already using gen AI for product idea generation say it differentiates
Ecosystem and partnerships Diversity and inclusion
their company by helping it respond to market shifts faster. Going
forward, they also believe generative AI will positively impact product
Market share growth Cybersecurity and data privacy
differentiation (88%), product trust (83%), and product quality (80%).31
Marketing and sales effectiveness Supply chain
10
Business model innovation Talent recruiting and retention
5 Trends for 2025: Ignite innovation with people-powered AI Source: The 2024 CEO Study: 6 hard truths CEOs must face. IBM Institute for Business Value. 23
Standing out will be crucial as a flood of AI-inspired products and services compete for By partnering with organizations that offer complementary capabilities, companies can tap
eyes in the marketplace. As customers are barraged with new options from every direction, into a vast network of expertise and resources that enhance innovation. It’s no longer about
executives across 13 industries agree that a single differentiating factor does the most to being the best at everything. It’s about being the best at what you do best—and tapping
move the needle on ROI: customer loyalty.32 partners for everything else.
And nothing keeps customers coming back like bespoke experiences. In fact, executives
expect personalization and customization to be the top customer demands that will disrupt
how their organization delivers products and services.33 But accommodating rapidly evolving
consumer preferences requires more than just clever algorithms and data analysis. It takes
open business models built on ecosystem partnerships.
Nothing keeps customers coming back like
bespoke experiences.
5 Trends for 2025: Ignite innovation with people-powered AI 24
What to do
1 2 3
Shatter departmental divides. Lock onto the moving target Stop trying to go it alone.
Build multidisciplinary teams to blend |
135 | ibm | ibm-why-invest-in-ai-ethics-and-governance.pdf | IBM Institute for Business Value | Research Insights
Why invest in AI ethics
and governance?
Five real-world origin stories
In collaboration with the Notre Dame—IBM Tech Ethics Lab
How IBM
can help
Clients can realize the potential of AI, analytics, and
data using IBM’s deep industry, functional, and
technical expertise; enterprise-grade technology
solutions; and science-based research innovations.
For more information:
AI services from IBM Consulting
ibm.com/services/artificial-intelligence
AI solutions from IBM Software
ibm.com/Watson
AI innovations from IBM Research®
research.ibm.com/artificial-intelligence
The Notre Dame-IBM Tech Ethics Lab
techethicslab.nd.edu/
2
Key
takeaways
Organizations that measure Embracing AI ethics is essential.
the value of AI ethics
It’s not just about loss aversion. 75% of executives
could be a step ahead. view AI ethics as an important source of competitive
differentiation.1 More than 85% of surveyed consumers,
Our holistic AI ethics
citizens, and employees value AI ethics.2
framework considers
three types of ROI. Longer-term, proactive AI ethics
strategies can generate value across
the organization.
A majority of companies (54%) expect AI ethics to be
very important strategically,3 with executives citing
involvement of 20 different business functions.4
Investing in AI ethics has the
potential to create quantifiable value.
Organizations that measure the value of AI ethics could
be a step ahead. Our holistic AI ethics framework
considers three types of ROI: economic impact
(tangible), reputational impact (intangible), and
capabilities (real options ROI).
1
Introduction
Generative AI is revolutionizing industries, but its dizzying ascendance
has also raised significant ethical concerns. Balancing the potential
benefits with ethical and regulatory implications is crucial.
But it’s not easy. In IBM Institute for Business Value (IBM IBV) research, 80%
of business leaders see AI explainability, ethics, bias, or trust as major roadblocks
to generative AI adoption.5 And half say their organization lacks the governance
and structures needed to manage generative AI’s ethical challenges.6
In the face of this uncertainty and risk, many CEOs are hitting pause. More than half
(56%) are delaying major investments in generative AI until they have clarity on AI
standards and regulations,7 and 72% of executives say their organizations will actually
forgo generative AI benefits due to ethical concerns.8
Yet there is a path forward—if executives broaden their outlooks and view AI ethics
as an opportunity. Even better: ongoing research suggests that investing in AI ethics
has the potential to create quantifiable benefits.
In order to unlock this potential, organizations need to embrace a new perspective
as they evaluate the ROI of AI ethics investments. In part one of this report, we
identify three key types of ROI that apply to AI ethics—in other words, a holistic AI
ethics framework. In part two and part three, we explore two distinct but valuable
ways to justify AI ethics investments right now. (We plan to build on this work by
conducting additional research in 2025 that explores quantification in greater depth.)
Finally, we offer an action guide for bringing the holistic AI ethics framework to life
inside the organization.
We also include stories from five executives on the front lines of AI ethics,
as part of an ongoing collaborative project among the IBM IBV, the Notre Dame—IBM
Tech Ethics Lab, the IBM AI Ethics Board, and the IBM Office of Privacy and
Responsible Technology. Some interviews were conducted in collaboration with
Oxford Economics.
2
Part one
Exploring a holistic
AI ethics framework9
AI ethics and governance investments can span broadly
across the enterprise, from an AI ethics board to an
ethics-by-design methodology, from an integrated
governance program to training programs covering AI
ethics and governance, among many other endeavors.10
(See “AI ethics: Stories from the front lines” on page 13.
Also refer to our IBM IBV study The enterprise guide
to AI governance at ibm.co/ai-governance.)
So how do organizations begin measuring the impact of such
initiatives? We developed a holistic AI ethics framework to meet
this need, validating it through an extensive series of conversations
with over 30 organizations. This approach can help organizations
understand the value of their AI ethics and governance investments.
Traditionally, investments are justified by calculating ROI in financial
terms alone. AI ethics investments are more challenging to evaluate,
providing both tangible and intangible benefits as well as helping
build longer-term capabilities.
“Our work has to not just contribute to the mission of the organization—
it also has to contribute to the profit margin of the organization,”
notes Reggie Townsend, VP of the Data Ethics Practice at SAS.
“Otherwise, it comes across as a charity, and charity doesn’t get
funded for very long.”
We developed a holistic AI ethics
framework, validating it through
an extensive series of conversations
with over 30 organizations.
3
A holistic AI ethics framework identifies three
types of ROI that organizations should consider
with AI ethics investments.
Economic impact (tangible ROI)
refers to the direct financial benefits
of AI ethics investments, such as cost
savings, increased revenue, or reduced
cost of capital. For example, an
organization might avoid regulatory
fines by investing in AI risk management.
Reputational impact (intangible ROI)
can involve important yet
difficult-to-quantify elements, such
as an organization’s brand and culture
that support positive returns or impact
on an organization’s reputations with
shareholders, governments, employees,
and customers. Examples include
improved environmental, social,
and governance (ESG) scores;
increased employee retention;
and positive media coverage.
Capabilities (real options ROI)
alludes to the long-term benefits
of building capabilities that, established
first for AI ethics, can disseminate
broader value throughout an organization.
For example, technical infrastructure
or specific platforms for ethics may allow
organizations to modernize in ways
that lead to further cost savings
and innovation.
Source: “The Return on Investment in AI Ethics:
A Holistic Framework.” Proceedings of the 57th
Annual HICSS Conference on Systems Sciences.
January 2024.
4
The holistic AI ethics framework depicted above
describes three paths to understanding the impact
of investments in AI ethics with regards to
stakeholders: the direct path through economic
return, and indirect paths through capabilities
and reputation. This framework encompasses
and describes the relationships, stakeholders,
and potential returns that exist when organizations
make investments in AI ethics.11
At a high level, how might this approach work in
practice? Consider the investment in an AI Ethics
Board infrastructure and staff. This investment helps
prevent regulatory fines (tangible impact); increases
client trust, partner endorsements, and business
opportunities (intangible impact); and helps enable
the development of management system tooling that
improves automated documentation and data
management (capabilities). The holistic AI ethics
framework illustrates how AI ethics is interwoven
throughout an organization, both in terms of practices
and outcomes.
The holistic AI ethics
framework illustrates how
AI ethics is interwoven
throughout an organization,
both in terms of practices
and outcomes.
5
Part two
The value of “loss aversion”
What is AI ethics? A senior vice president with responsibility
for data policy at Fidelity Investments puts it succinctly:
“It’s using AI technology in a responsible form to be able
to distinguish between right and wrong as we communicate
with our customers, prospects, and other clients.”
In recent IBM IBV research, 72% of executives said they’ll step back from
generative AI initiatives if they think the benefits might come at an ethical
cost. These same organizations are 27% more likely to outperform on
revenue growth—a correlation that is hard to ignore.12
Yet noble AI intentions are often talked about more than they are acted
on. While over half of organizations in our research have publicly endorsed
principles of AI ethics, less than a quarter have operationalized them.13
Fewer than 20% strongly agree that their organizations’ actions
and practices on AI ethics match (or exceed) their stated principles
and values.14
“It’s all good to want to do it, but you need to actually do it,” says a senior
leader responsible for AI governance at a global financial services firm.
“But to do it, you need resources, which requires funding. More important
than that, you need the will of senior executives.”
So, what is the business justification for investing in AI ethics? It often
starts with a loss aversion approach: avoiding costs associated with
regulatory compliance or retaining revenue that might be lost if
customers move their business to enterprises that prioritize AI ethics.
Noble AI intentions are often talked
about more than they are acted on.
6
The fact that these motivations reflect a short-term A prod from AI regulators
strategy does not detract from their significance.15
Loss aversion generates near-immediate results.
AI regulations are a catalyst for action. The EU
As the senior leader responsible for AI governance at
AI Act is the first comprehensive AI regulation by
a global financial services firm notes, “The business
a major entity. One strategy manager at Deutsche
case is all about decreasing reputational risk.”
Telekom says, “The EU AI Act could change the face
of AI ethics globally. If, for instance, an American
company is working with us, they also have to comply
with the EU AI Act.”
The EU’s effort is only the beginning. Organizations
Examples of loss aversion include:16
such as the Partnership on AI, the Global Partnership
on AI, the World Economic Forum, the United Nations,
and the Organisation for Economic Co-operation and
Regulatory justifications
Development (OECD) have all published principles
Avoid a regulatory fine. and guidelines on a responsible approach to AI.17
In a survey by the Centre for the Governance of AI
Avoid legal costs. of over 13,000 people across 11 countries, 91%
agreed that AI needs to be carefully managed.18
Implement required technical
compliance mechanism.
Given this emphasis on regulations, oversight,
and responsible approaches to AI, a focus on loss
Enable business
aversion isn’t just sensible but necessary.
for required compliance.
Customer/partner/
competitor justifications
Allay stakeholder concerns.
Avoid threat to business model.
Meet specific customer request or need.
Protect brand reputation.
Keep pace with competitors.
7
Part three
Leveraging AI ethics
to generate value
The benefits of investments in AI ethics aren’t exclusive
to cost avoidance or damage control. They also help to build
useful capabilities and tangible innovations that can enable
an organization’s long-term strategies.19 Such value
generation can be more indirect than loss aversion and
requires an expanded view of ROI. It also won’t happen
overnight and can take time to see measurable outcomes.
But organizations that are sophisticated about their understanding of AI
ethics can use the investments to:20
– Enable long-term plans to scale AI responsibly.
– Build unique and valuable organizational capabilities that can lead
to differentiation.
– Improve employee efficiency or productivity.
– Align with values to advance as an industry leader.
– Seize a market opportunity.
– Protect vulnerable individuals and communities.
– Increase customer satisfaction.
– Demonstrate trustworthiness and maturity.
– Support Environmental, Social, and Governance (ESG) efforts.
– Increase ability to manage risk over the long term.
– Innovate for a competitive advantage.
As AI technology matures, organizations can not only integrate AI into their
operations, they can repurpose that technology toward new innovations. A
senior director from a leading health and consumer goods retailer explains,
“Based on the measures we took from the AI standpoint to create and
enrich the customer experience, we have seen returns in terms of adoption
of those brands, sales growth, customer retention, and customer growth.”
8
Combining the best
of both worlds
Organizations that embrace a holistic approach
that encompasses both loss aversion and value
generation will be more efficient, effective, and
successful—as well as more ethical.
Reactive Proactive
Loss aversion Value generation
Regulatory compliance justifications Create technologies, infrastructures,
and platforms that can support AI ethics
efforts and be repurposed
Avoid a regulatory fine.
Avoid legal costs. Enable long-term plans to scale
AI responsibly.
Implement required technical
compliance mechanism. Build unique and valuable
organizational capabilities that
Enable business for required compliance.
can lead to differentiation.
Improve employee efficiency
or productivity.
Justifications relating to clients, Align with values to advance
partners, and competitors as an industry leader.
Seize a market opportunity.
Allay stakeholder concerns.
Protect vulnerable individuals
Avoid threat to business model.
and communities.
Meet specific customer request or need.
Increase customer satisfaction.
Protect brand reputation.
Demonstrate trustworthiness
and maturity.
Keep pace with competitors.
Support Environmental, Social,
and Governance (ESG) efforts.
Increase ability to manage risk over
the long term.
Innovate for a competitive advantage.
Source: “On the ROI of AI Ethics and Governance Investments: From Loss Aversion
to Value Generation.” California Management Review. July 29, 2024.
9
The senior vice president of Fidelity Investments observes:
“What companies don’t realize is that up-front investment actually
pays significant ROI, not just in terms of ethics, but from a total cost of
implementation on any of your use cases. Because if you don’t lay that
foundation, you spend a lot more money with everybody implementing
one pillar at a time and not benefiting from any reuse.”
A preliminary step to this evolution, of course, is to actually develop
AI use cases that align with and support organizational strategy.
Notes the strategy manager at Deutsche Telekom, “Either you could
create AI solutions for the customer, or you could create AI solutions
for your internal infrastructure.”
Out of the starting block, it’s instinctive and reasonable to adopt
a “defensive” loss-aversion posture to avoid the pitfalls we’ve
described, such as regulatory fines, legal costs, and reputational risks.
But fertile ground can be found in the pivot to value generation.
Organizations need to create technologies, infrastructures,
and platforms with the versatility to support AI ethics initiatives
and to fuel broader corporate innovation.
Procuring support and budget for these strategies can be tricky.
To persuade skeptics and surmount obstacles, organizations should
clearly pinpoint potential value generated, including metrics of economic
returns. This can be done through a process of identifying relevant loss
aversion and value generation justifications as the organization plans
and then evaluates potential investments21—essentially, using the
holistic AI ethics framework.
Fertile ground can be found in the pivot
to value generation. Organizations need
to create technologies, infrastructures,
and platforms with the versatility to
support AI ethics initiatives and to fuel
broader corporate innovation.
10
Action guide
How to reap the rewards
of AI ethics investments
Investing in AI ethics is not just the right thing to do, it can also be a sound
business decision. By using the holistic AI ethics framework, organizations can
make informed choices about allocating resources to AI ethics, helping boost
the trustworthiness and potential of AI programs overall.
According to IBM IBV research, 75% of executives view ethics as an important
source of competitive differentiation.22 A study from the Economist
Intelligence Unit echoes those results, pointing to a competitive edge through
product quality, talent acquisition and retention, and new revenue sources.23
These studies underscore the criticality of a proactive approach to AI ethics.
Organizations must consider how governance of AI differs from that of
previous technologies, permeating every corner of their culture, ecosystem,
and customer engagement.
“You educate the AI engine based on what humans are thinking,” says the
senior director at a leading health and consumer goods retailer, “because
they are the better judge from an ethics standpoint.”
Along those lines, Reggie Townsend of SAS observes: “We have a diverse set
of folks who have come from a variety of different backgrounds and life
experiences. We do hard work, but we do heart work. I don’t hire anyone who
doesn’t have a heart for what we’re doing. We have passionate people on our
team, and we bring that passion to the work. That’s fundamentally important.”
11
Here’s our five-step guide for optimizing
your AI ethics investments
1
Engage your savviest AI ethics experts to educate the C-suite on
differences between loss aversion and value generation approaches
to AI ethics. Help executives envision the potential of leveraging AI
ethics technology, platforms, and infrastructure for broader use.
2
Identify specific value generation justifications for AI ethics and
governance that may apply to the AI use cases at hand. Examples
include the ability to responsibly improve the answers to customers
and increased employee productivity and job satisfaction.
3
Think through the anticipated stakeholder impacts of the AI
use case and identifying potential indicators. These include:
– Direct economic returns (for example, the value
of an expanded customer base)
– Intangible reputational returns (for example, earned media
value of customer reviews)
– Capabilities and knowledge returns from real options
(for example, improved customer response quality that leads
to more first-contact resolutions).
4
Create an AI ethics implementation strategy that can deliver on value
generation justifications. Using the analysis in action 3, identify the
potential returns holistically. Doing so can help optimize the potential
returns on your investments in AI ethics and governance while
simultaneously benefitting stakeholders, ecosystems, and society.
5
Turn value generation into a competitive advantage. Focusing on value
generation can provide a competitive advantage in an environment
where regulatory compliance is business as usual.
For additional information and actions on the holistic AI ethics framework, refer to “On the ROI
of AI Ethics and Governance Investments: From Loss Aversion to Value Generation,” California
Management Review, at https://cmr.berkeley.edu/2024/07/on-the-roi-of-ai-ethics-and-
governance-investments-from-loss-aversion-to-value-generation/ and “The Return on
Investment in AI Ethics: A Holistic Framework” at https://arxiv.org/abs/2309.13057.
12
AI ethics
Stories from the front lines
Deutsche
Deutsche Telekom’s data initiatives are closely tied with monetizing data through AI
Telekom applications and monitoring the EU AI Act. One strategy manager at the company leads
a team that is involved in virtually every AI conversation in the organization and is
therefore able to provide a holistic overview of the company’s approach to AI ethics.
Preparing for the EU
AI Act with internal Deutsche Telekom has created a team of high-level executives responsible for
governance and evaluating current and future AI initiatives—in effect, an organized governance group.
education The group’s most important purpose is to help ensure that the company complies
with data privacy and security procedures both internally and externally—including
the EU AI Act.
“AI is all about data. It’s a fundamental element of any AI product,” says the manager,
adding that he regards data as a crucial component of the AI ethics approach as well.
Before incorporating any data into its products, the organization considers who is
exposed to the data and how customer data is protected. Beyond their customers,
Deutsche Telekom also must protect certain data segments in terms of sustainability
and energy practices.
Critically, Deutsche Telekom heavily invests in educating employees about AI and
its ethical use, often in the form of internal workshops, including training related
to the EU AI Act.
“Training colleagues is definitely a return on investment because it reduces the time
to market and we come up with more innovative products,” he says. And with its
continual efforts to improve, Deutsche Telekom experiences greater innovation
and enhanced customer trust.
13
AI ethics: Stories from the front lines
Fidelity
Responsible AI initiatives are embedded into each phase of AI use cases at
Fidelity Investments, beginning with robust data management practices
Investments
and feeding into a dynamic review process driven by the company’s AI
Center of Excellence. The financial services firm invested heavily in these
Reaping ROI through
initiatives to make AI ethics one of its foundational pillars—rather than a
repurposed use case
compliance box-checking exercise.
implementation
Each business line at Fidelity has a dedicated team for AI use case
development and vendor management. This work is guided by the
expertise of external consultants and actively monitored by the firm’s
compliance and risk officers, who receive specialized AI training.
The AI Center of Excellence is involved in each step of this process, from
vendor selection to model evaluation. It resides in Fidelity’s data function
and includes representation from each business unit at the firm, with roles
ranging from risk compliance and audit to legal and even information
security. This process also allows Fidelity to confidently answer clients’
increasing demands for information on its AI use and governance.
Resistance to responsible AI initiatives is inevitable, as they can delay
projects or limit use cases. “You have to explain that the reason controls
are so important is not just some random compliance policy, but that
there are implications to the firm if we get this wrong,” says a senior vice
president with responsibility for data policy at the firm. Fidelity has
been able to minimize pushback by framing these initiatives as
integral to the success of AI projects and by streamlining the overall
governance process.
“You have to explain that the reason controls
are so important is not just some random
compliance policy, but that there are
implications to the firm if we get this wrong.”
A senior vice president with responsibility for data policy at Fidelity Investments.
14
AI ethics: Stories from the front lines
SAS Reggie Townsend, VP of the Data Ethics Practice at SAS, leads a team tasked
with coordinating responsible innovation principles, operational workflows and
Ensuring an AI-driven governance structures across a global organization. It all began with questions
future that is built and investigating.
for all of us
Prompted by risks to vulnerable populations and the increasing sophistication of
AI, Townsend and close colleagues began digging deeper into responsible AI
and data ethics at SAS. They were empowered by SAS leadership to formalize
the company’s longtime commitment to responsible innovation. Consequently,
SAS created the Data Ethics Practice (DEP).
With a philosophy of “ethical by design,” the DEP guides the company’s efforts
to help employees and customers deploy data-driven systems that promote
human well-being, agency, and fairness. This approach compels individuals
to answer three basic questions:
– For what purpose?
– To what end?
– For whom might it fail?
The team helps build Trustworthy AI capabilities and workflows to help
customers and developers pursue their responsible AI goals. AI governance
advisory services from the DEP are helping customers put AI into action
responsibly. The DEP also provides critical counsel to employees on product
development, marketing, and more.
When Townsend’s role and team were created, the hope was their work would
bolster trustworthiness of products, processes, and people. This, in turn,
would enhance the brand’s reputation as a trusted AI leader.
Profits are important, of course. But according to Townsend, his team’s
guiding principle is that wherever SAS software shows up, it does no harm.
“Sometimes,” Townsend observes, “you just have to take action because
it’s the right action to take.”
“Sometimes, you just have to take
action because it’s the right action
to take.”
Reggie Townsend
Director and VP, Data Ethics branch, SAS
15
AI ethics: Stories from the front lines
Global financial For one senior leader responsible for AI governance at a global financial services firm,
AI development and ethics starts with education. He advocates hosting workshops
services firm
that discuss ethical principles and values—empowering leadership to understand
trade-offs. “We need to talk about AI in a way that interests leadership, not just in
Justifying positive
processes and procedures,” he observes.
returns with a lowered
reputational risk In discussing how to measure the return on investments in AI ethics, the senior leader
offers the “creepy line” metaphor. Often, organizations find themselves in situations in
which they are doing something perfectly legal that is highly profitable, yet still feel
uncertain about the ethicality of their actions—a sense of crossing the “creepy line.”
In such situations, he says that organizations must examine the activity through the
lens of both current and future generations, in conjunction with all comprehensive
ethical considerations. As long as these considerations are covered satisfactorily,
the organization should feel reassured that the “creepy line” is not breached.
He also notes, “Reputational risk is a key factor in justifying positive returns.
We aim to decrease reputational risk while applying data and AI ethics principles.”
For example, his team conducted an ethical fairness review of loan pricing involving a
credit scoring algorithm. In conducting this review, the team analyzed all 165 features
of the model, asking if there were any potential causal mechanisms for why that
particular data feature may correlate with an individual’s ability to pay back a loan.
Ultimately, three data features were removed because a causal link did not exist,
thus avoiding the lack of fairness in using this AI technology.
“We need to talk about AI in a way
that interests leadership, not just in
processes and procedures.”
Senior leader responsible for AI governance at a global financial
services firm
16
AI ethics: Stories from the front lines
A leading health
A senior director at this organization instituted an AI initiative to provide solutions
via vendors and internal products. A recent conversation with him covered three main
and consumer
operational areas.
goods retailer
A rigorous governance process. The retailer’s AI governance group is a centralized
body that helps ensure all AI initiatives fulfill their required steps for approval. In that
Driving success with a
vein, it conducts sessions in which project teams present how they’ve aligned their
thorough AI ethics and
compliance measures with the group’s control plan. If approved, the projects move
governance strategy
forward. The director notes that, as a sizeable enterprise engaging with large numbers
of partners, suppliers, customers, and other ecosystems, it must be extremely careful
in building their AI capabilities.
The AI ethics engine. Whether the retailer invests in SaaS-, vendor-, or open-source-
based products, they ensure all ethical parameters are met prior to deployment.
Its internal audit process is referred to as “the AI ethics engine.” In engaging a vendor,
the organization first conducts a background check, looking at the health of its
industry, clients, reputation, and capabilities. This process can span two to four
months. Once the retailer picks its vendor, it engages in a pilot. If success and ethics
measures are met, the partnership proceeds.
Stakeholder success. The organization has heavily invested in AI capabilities
to enhance the customer engagement experience and drive market strategies
and customer growth. The director notes, “AI by itself or a human by itself cannot
be successful, but if you combine those two together, the outcome is successful
and accurate.”
At this particular retailer, AI capabilities implemented in customer service, for example,
will not replace customer service employees. Rather, the organization invests in
providing these employees with additional skills, resulting in employee retention.
This approach can create benefits for the customers, employees, and company’s
economic returns.
“AI by itself or a human by itself cannot
be successful, but if you combine those
two together, the outcome is successful
and accurate.”
Senior director at a leading health and consumer goods retailer
17
Authors
Nicholas Berente Marianna Ganapini
Senior Associate Dean for Academic Programs Associate Professor, Philosophy
Professor of IT, Analytics and Operations Union College
University of Notre Dame, Mendoza College of Business linkedin.com/in/marianna-b-ganapini-769624116/
linkedin.com/in/berente/ [email protected]
[email protected]
Brian Goehring
Marialena Bevilacqua Associate Partner, AI Research Lead
PhD Student in Analytics IBM Institute for Business Value
University of Notre Dame, Mendoza College of Business linkedin.com/in/brian-c-goehring-9b5a453/
linkedin.com/in/marialena-bevilacqua-6848b9132/ [email protected]
[email protected]
Francesca Rossi
Heather Domin IBM Fellow and AI Ethics Global Leader
Global Leader, Responsible AI Initiatives, IBM IBM Research
Associate Director, Notre Dame—IBM Tech Ethics Lab linkedin.com/in/francesca-rossi-34b8b95/
linkedin.com/in/heatherdomin/ [email protected]
[email protected]
Contributors
Sara Aboulhosn, Angela Finley, Rachna Handa,
Jungmin Lee, Stephanie Meier, and Lucy Sieger
18
About Research Insights The right partner for
a changing world
Research Insights are fact-based strategic insights
for business executives on critical public- and At IBM, we collaborate with our clients, bringing
private-sector issues. They are based on findings together business insight, advanced research, and
from analysis of our own primary research studies. technology to give them a distinct advantage in
For more information, contact the IBM Institute for today’s rapidly changing environment.
Business Value at [email protected].
IBM Institute for Related reports
Business Value
The enterprise guide to AI governance
IBM Institute for Business Value. October 2024.
For two decades, the IBM Institute for Business Value
ibm.co/ai-governance
has served as the thought leadership think tank for
IBM. What inspires us is producing research-backed,
The CEO’s guide to generative AI:
technology-informed strategic insights that help
Responsible AI & ethics
leaders make smarter business decisions.
IBM Institute for Business Value. October 2023.
From our unique position at the intersection ibm.co/ceo-generative-ai-responsible-ai-ethics
of business, technology, and society, we survey,
interview, and engage with thousands of executives, AI ethics in action
consumers, and experts each year, synthesizing IBM Institute for Business Value. April 2022.
their perspectives into credible, inspiring, and ibm.co/ai-ethics-action
actionable insights.
To stay connected and informed, sign up to receive
IBV’s email newsletter at ibm.com/ibv. You can also
find us on LinkedIn at https://ibm.co/ibv-linkedin.
19
Notes and
sources
1 Goehring, Brian, Francesca Rossi, and Beth Rudden. 11 Bevilacqua, Marialena, Nicholas Berente, Heather
AI ethics in action: An enterprise guide to progressing Domin, Brian Goehring, and Francesca Rossi.
trustworthy AI. IBM Institute for Business Value. April “The Return on Investment in AI Ethics: A Holistic
2022. https://ibm.co/ai-ethics-action Framework.” Proceedings of the 57th Annual HICSS
Conference on Systems Sciences. January
2 Ibid. 2024. https://arxiv.org/abs/2309.13057. “OECD AI
Principles overview.” OECD. Accessed November 15,
3 Ibid. 2024. https://oecd.ai/en/ai-principles
4 Ibid. 12 The CEO’s guide to generative AI: Customer and
employee experience. IBM Institute for Business Value.
5 2023 Institute for Business Value generative AI state of
August 2023. https://www.ibm.com/thought-
the market survey. 369 global CxOs. April/May 2023.
leadership/institute-business-value/en-us/report/
Unpublished information.
ceo-generative- |
136 | ibm | ibm-embedding-ai-in-your-brands-dna.pdf | IBM Institute for Business Value | Research Insights
Embedding AI
in your brand’s DNA
Innovate from products to ecosystem—
and everything in between
How IBM
can help
IBM has been providing expertise to help retail and consumer products
companies win in the marketplace for more than a century. Our researchers
and consultants create innovative solutions that help clients become more
consumer-centric by delivering compelling brand and store experiences,
collaborating more effectively with channel partners, and aligning demand
and supply. With a comprehensive portfolio of solutions for merchandising,
supply chain management, omnichannel retailing, and advanced analytics,
IBM helps deliver rapid time to value. With global capabilities that span 170
countries, we help brands and retailers anticipate change and profit from
new opportunities. For more information on our retail and consumer
products solutions, please visit: ibm.com/industries/retail, ibm.com/
consulting/retail, and ibm.com/industries/consumer-goods.
2
Key
takeaways
Brands are evolving Over the next year, retail and consumer
beyond mere AI adoption, products executives expect to expand
embedding it in their DNA AI significantly throughout all areas
to harness their distinct of the business, from brand-defining
AI-driven advantage. activities to core operations.
But to be AI-centric, organizations need an open
mindset for how AI can deliver transformation beyond
productivity gains.
Across 13 areas of the business,
executives plan to augment most
activities with AI over the next
12 months.
But they only project 31% of their workforce will need
to reskill or develop new skills in that same time frame,
underestimating what’s needed to support employees
in the AI transformation.
Almost 9 in 10 executives claim to have
clear organizational structures, policies,
and processes for AI governance.
But fewer than one-quarter of organizations have fully
implemented and continuously review tools on AI
governance, putting brand trust at risk.
1
Industry executives project that
AI’s contribution to revenue
growth will increase 133%
from 2023 to 2027.
Consumers are ready for AI.
Are you?
Consumers are tech-savvy trendsetters and brands need to keep up to
stay relevant. Today, customers and shoppers are actively engaged with
AI in their daily lives, from using AI-powered search engines to creating
content with generative AI tools.
In the 2024 IBM Institute for Business Value (IBM IBV) consumer research study,
nearly two-thirds of consumers said they have used or want to try AI applications.1
This interest sets the stage for retail and consumer products companies to hasten
integration of AI across their business while keeping an eye toward becoming AI-led
brands—leveraging the technology to reimagine operations, inspire loyalty, and
expand the size of customers’ wallets for long-term competitive advantage.
Our latest survey of 1,500 global retail and consumer products executives finds
organizations are accelerating their adoption. AI—both traditional and generative—has
permeated all functions in the enterprise to some degree. From marketing and customer
service, to supply chain and procurement, to finance and IT operations, AI use cases
span brand-defining, business-enabling, and corporate operations. Looking ahead
through 2025, most executives are thinking big, expecting AI to be used extensively
across the business (see Figure 1). Industry leaders also report AI spending is on the
rise (see Perspective, “AI spending moves outside of IT”), and they project that AI’s
contribution to revenue growth will increase 133% from 2023 to 2027.
Retail and consumer products organizations are at a pivotal point in their AI journey.
The question is: are they taking enough of the right steps to become AI-led brands,
or are they just tacking on ad hoc AI solutions that deliver short-term gains? It’s time to
move beyond just productivity and efficiency and extend AI’s power enterprise-wide
to boost process effectiveness, spark new business models and ecosystems, and
ignite engagement with innovative employee and customer experiences.
2
FIGURE 1
Retail and consumer
products organizations
plan to use AI extensively
in 2025.
Figure 1
Retail and consumer products organizations use AI extensively in 2025.
Percent of organizations planning to use AI to a moderate or significant extent over the next 12 months
Marketing and customer experience 89%
Digital commerce 86%
Merchandising 86%
Customer service 85%
Brand-defining
areas
Stores
79%
Product design and development 76%
Supply chain operations 90%
Sustainability 87%
Procurement 86%
Business-enabling
areas
Production and manufacturing 83%
IT and security 90%
Finance 90%
Corporate
operations
HR 88%
Percentages represent an average of responses for a set of tasks in each functional area, based on the question: “To what extent
do you use AI or gen AI in this activity?” Respondents replied “to a moderate extent” or “to a significant extent.”
3
Perspective
In this report, we discuss three factors that will help
AI spending
organizations make a fundamental change in their
DNA, where AI emerges as the driving force behind shifts beyond
every decision, innovation, and strategy. In part one,
IT budgets
we discuss balancing the marathon with the sprint to
shift from plus-AI to AI-first. In part two, we examine
the need to prepare the workforce for the planned
rapid and aggressive AI adoption, and in part three,
we address the imperative to safeguard consumer AI budget allocation is undergoing a significant shift.
trust. Each section includes an illustrative case study While IT budgets will still play a role, retail and
and concludes with an action guide of steps brands consumer products executives report a growing
can take to accelerate progress. portion of AI spending is moving outside of traditional
IT budgets. As AI becomes more than just a tech tool,
functional areas are identifying their needs for AI as
part of larger business solutions, from creative
Definitions
marketing tools to empowering store associates to
new warehouse management systems.
Traditional artificial intelligence
Executives project their IT budget dedicated to AI
spend will increase by 19% over the next year, but
Systems that understand, reason, learn, and
spending on AI outside of the IT budget is expected
interact. AI technology includes machine
to surge 52%. As a percent of revenue, IT spending
learning (ML) approaches, but also other
on AI will be 1.04% and AI spending outside of IT
techniques such as reasoning, planning,
will be 2.28% by 2025. Taken together, 3.32% of
scheduling, and optimization.
revenue could be dedicated to AI spending next year.
For a $1 billion company, that equates to $33.2
Generative AI
million for total AI spend.
A class of machine learning that generates
With at least 13 functional areas that span retail and
content or data, including audio, code,
consumer products organizations, executives across
images, text, simulations, 3D objects, and
the C-suite must keep tabs on the investments being
videos—usually based on unsupervised or
made in each area, coordinating platforms and tools
self-supervised learning. Recent examples of
to provide transparency across the enterprise.
generative AI include GPT-4 (language), DALL-E
IT and the business lines must work together to avoid
(images), GitHub Copilot (code), and AlphaFold
duplication of effort and to help ensure consistent
(scientific protein folding).
alignment with the overall business strategy.
4
Part one
Building an intelligent brand
that endures
Consumer organizations need to take a long-term view
of their AI journey while moving with urgency and intent.
Nearly all industry executives are banking on AI for innovation in products and
services (89%) as well as business models (85%). But a mere 54% expect AI to
influence operational innovation. Transforming operations with AI across supply
chains, manufacturing, distribution, finance, and compliance is the very essence of
being an AI-centric brand. This remodel is both a marathon and a sprint—moving from
simple AI use cases to orchestrating AI across functions to deliver sustainable value.
Many organizations are in the early stage of adoption, integrating AI within a single
function. For example, 88% use AI to a moderate or significant extent in demand
forecasting, 87% for HR help desks, 84% for IT support and issue remediation, 84%
in creating and managing trade promotions, 81% in inventory and order management,
and 80% in managing production activities. These are quick wins that can deliver
a more immediate impact on daily operations.
But companies are keen on expanding to more sophisticated uses of AI over the next
12 months. They will be transitioning from internal departmental use cases with
limited system integration to multifaceted ones that require external collaboration,
more complex system integrations, and more human intervention and oversight.
Take virtual assistants as an example (see Figure 2). Initially, they responded to
simple, predefined queries such as order and shipment status. As they have become
more integrated with data in ordering systems, they can identify delays or missing
orders as well as back-order options and in-store availability. Adding customer
shopping history and generative AI capabilities to their arsenal, they can dynamically
recommend offerings and personalized content for individual customers. Camping
Only 54% of executives
World’s virtual assistant, Arvee, illustrates the value of integrating platforms such
expect AI to help their
companies innovate in as Oracle and Salesforce so that the assistant can access customer information
operations. efficiently to address queries faster.2
5
Executives expect to expand rapidly to more sophisticated AI use cases
across the enterprise. For example, those leveraging AI to a significant
extent for personalized responses and follow-up actions in customer
service plan to increase their usage by 236% over the next 12 months.
Similarly, they want to grow significant AI usage in integrated business
planning by 82% and in talent acquisition by 300%.
Figure 2
BFIrGaUnRdEs 2 and retailers plan to expand use of AI/gen AI
into more sophisticated use cases over the next year.
Brands are fueling virtual assistants with more
comprehensive, relevant enterprise data to enable
increasingly personalized responses to customers.
I can provide When I am connected to the When I have access to
shipment status order management warehouse customer profiles and shopping
and tracking and store inventory system, history, I can dynamically
information. I can provide options for recommend offerings and
back orders and in-store personalized content for
pickup options. individual engagement.
6
Case study
As organizations progress with their initiatives, they Kroger uses AI
are investing in platforms to integrate AI tools and
to elevate customer
models. Today, as they establish their AI foundation,
they are primarily focused on data and analytics pickup experiences3
platforms (65%), innovation platforms (64%), and
skills/learning platforms (62%). Building on these
existing platforms and expanding to others will
enable federation and orchestration of AI across
Kroger has long depended on data and advanced
functions, facilitating cross-functional learning to
analytics to fuel business innovation. Since its
support scaling AI across the enterprise.
inception decades ago, its loyalty program has
Executives plan to integrate AI capabilities with delivered a trusted value exchange enabled by
business partners over the next three years, and they permission-based information. Today, using machine
predict the use of ecosystem platforms will surge learning algorithms, Kroger delivers valuable
from 52% today to 89%. Take the product compliance personalized offers and communications across
ecosystem as an example. By integrating end-to-end 150 million customer touchpoints and through
AI-driven compliance, brands can ensure all facets 1.9 billion unique coupons customized for millions
of the product lifecycle align with evolving regulatory of loyal customers.
requirements, consumer safety, and sustainability
Most recently, Kroger has been exploring ways
expectations. This ecosystem prioritizes accelerated
to use AI to help improve the customer experience,
product lifecycle management with an advanced
specifically order pickups. Using AI-enabled dynamic
business rules engine and touchless bill-of-materials
batching, an AI solution sorts through 200,000 totes
generation, helping ensure products are market-
per second to build the most efficient pickup trolley.
ready with minimal manual intervention.
It drives a 10% reduction in steps by identifying the
most efficient pick route through the store. With
dynamic batching of orders, these tools are providing
associates the most efficient pick routes, so Kroger
can dramatically reduce pickup lead time in its
highest volume stores.
Executives expect their use
of ecosystem platforms for
AI tool and model integration
to surge from 52% today to
89% in the next three years.
7
Action guide
Intentionally embed AI in operations
to deliver a sustainable brand advantage.
In the 2024 IBM IBV CEO study, 70% of retail and consumer products
CEOs said that to win the future, they must rewrite their organizational
playbook.4 As you redefine your core operational strategies and processes
to capitalize on AI, concentrate on how to achieve lasting value.
Tailor AI to your As you move beyond AI-driven productivity gains, you need a clear vision and strategy
brand’s priorities. for where AI and gen AI can help you distinguish yourself from competitors or shore
up weaknesses. But keep in mind that consumers expect you to stay true to your core
values as you innovate. If a strong customer experience is your focus, use AI to
personalize customer service and optimize in-store experiences. If product innovation
is a differentiator, tap into AI for product design, customer preferences, and vendor
capabilities to facilitate faster ideation and development cycles. The key is to
concentrate on what’s most important—not everything that’s possible.
Invite finance, Becoming an AI-centric brand requires purposefully aligning IT with long-term
technology, and business goals, not just the hottest tech. For example, organizations that consider
business leaders applications and infrastructure holistically in support of business needs (known as
to the same table. “hybrid-by-design” principles) can generate more than three times higher ROI over
five years.5 Tear down the silos between finance, technology, and business leaders so
that together, they can build solid business cases for where AI can deliver a long-term
competitive edge.6
Venture beyond Traditional strategic partnerships focused on physical distribution of supplies and
tried-and-true products are no longer enough in the age of AI. Tech companies, startups, and other
partnerships. nontraditional partners are needed for model development, platforms, and tools.
For example, other IBM IBV research found that 65% of organizations are already
working with or planning to work with a strategic partner to build a large language
model for generative AI initiatives.7 Prioritize partners who understand your goals and
share your vision. Identify those with a proven record for integration and loop them
into your processes early. Think outside the box, imagining new partners that create
new opportunities for growth.
8
Part two
Priming the augmented workforce
AI is transforming the nature of work from the store to the factory floor,
but industry executives undervalue workforce reskilling.
AI is diffused throughout the retail and consumer products workplace. Nearly all
(96%) executives say their teams are using AI and gen AI to a moderate or significant
extent at work. When virtually everyone is using a new and powerful technology such
as AI, then virtually everyone needs training to optimize the value and understand the
risks that could damage the brands. Yet, leaders project only 31% of their workforce
will need to reskill or develop new skills over the next 12 months, with this number
climbing to just 45% in the next three years—a significant miscalculation.
Both hard and soft skills—from prompt engineering and data analytics to critical
thinking and problem solving—are essential to ushering in the age of the augmented
workforce where AI won’t replace people, but people who use AI will replace people
who don’t.8 The talent transformation is an ongoing training and education process
that must be defined and started sooner rather than later. If not, 67% of employees
have said they will leave for another employer that provides better training on new
technologies, according to an IBM IBV survey of more than 21,000 workers.9
Executives recognize the workforce will be increasingly augmented, while automation
remains crucial for rules-based tasks and repetitive work. Across 13 functional areas
from marketing and commerce to supply chain, HR, and IT, they plan to more
frequently augment than automate activities over the next 12 months (see Figure 3).
Industry leaders know that many brand-defining areas demand human intuition,
creativity, emotional intelligence, and expertise that can be complemented by AI.
For example, in product design and development, AI can accelerate idea generation
and ideation, even providing visualizations. Likewise, operational areas have vast
Leaders project only
amounts of data where decisions require human oversight, such as supply planning,
31% of their workforce
where 54% plan to augment their employees. In this activity, AI can quickly access
will need to reskill
or develop new skills and analyze a broader range of data to help the supply planner confidently resolve
over the next year. shortages in minutes, knowing important information is not missed.
9
FIGURE 3
Retail and consumer products
executives know that
automation has its place but
Figure 3
see a future of augmentation.
Retail and consumer goods executives know that
automation has its place but see a future of augmentation.
Percent of activities that will be automated, augmented, or
have no impact from AI in each area over the next 12 months
No impact Automated Augmented
Digital commerce
12% 31% 58% & B2B sales
Product design, development, and
14% 28% 57% product lifecycle management
Merchandising /
14% 32% 54% category management
Brand-defining Marketing
13% 35% 52%
areas
Customer service
15% 35% 50%
Stores
21% 32% 47%
Sustainability
11% 35% 54%
Procurement
14% 33% 54%
Supply chain operations
Business-enabling 12% 37% 51%
areas
Production
8% 43% 49% and manufacturing
HR
10% 35% 54%
Finance
12% 36% 52%
Corporate
operations IT
9% 40% 51%
Percentages represent an average of responses for a set of tasks in each
functional area, based on the question: “To what extent do you use AI
or gen AI in this activity?” Respondents replied “to a moderate extent”
or “to a significant extent.”
10
Case study
Ultimately, brands will be finding the sweet spot for Japanese retailer empowers
automation and augmentation. Take managing the
people with AI to boost profits
seasonal workforce as one example. AI-powered
automation can streamline hiring, onboarding, while reducing waste10
and scheduling processes, reducing administrative
burdens and helping control costs. Managers can use
AI-powered tools that provide real-time insights into
staffing needs, predict demand fluctuations, and
optimize schedules. Similarly, in inventory A leading retail company in Japan was grappling with
management, AI-powered sensors and cameras costly problem: food and consumer-goods waste was
automatically monitor inventory levels in real time, eating away at their profits. The client’s field staff
while providing employees with the insights needed needed data-driven insights to make more informed
to reduce the risk of stockouts or overstocking. pricing decisions.
Even areas that have a high degree of automation, For a wide variety of products and the company’s
such as customer self-service, can benefit from operations, price optimization relied more heavily
augmented employees. As executives expand use on human judgment than data, leading to variations in
of AI for personalized responses and follow-up customer forecasts, stock levels, and discount rates.
actions over the next 12 months, they say 55% These variations resulted in excessive and inadequate
of the activities will be augmented versus 30% stocking, irregular discount amounts and timings,
being automated. and large profit losses due to food waste and missed
sales opportunities.
The company worked with IBM to develop a specialized
price optimization AI system to analyze vast amounts
of data, predict customer numbers and purchase
patterns, and suggest optimal discount amounts and
timings. Now the client’s field staff can combine their
own expertise with data to improve pricing decisions.
The pricing optimization system was designed to adapt
to different product categories and sell-by durations,
making it a versatile, scalable solution that can support
Brands are finding the
a diverse product range.
sweet spot for automation
and augmentation.
11
Action guide
Prepare your workforce
to power your AI-centric brand.
AI is clearly impacting virtually the entire retail and consumer products
workforce—from the person stocking the shelves to those who sit with
you in the C-suite. It’s being built into many of the tools employees use
every day, such as AI-powered sales forecasting tools or AI-driven
design tools. Leaders need to ensure all employees are prepared
to optimize the value AI can deliver.
Connect HR, IT, Executives report leadership for reskilling efforts is divided among an AI center of
and business lines competence (31%), HR (22%), AI committees (18%), and IT (17%). This disjointed
to define reskilling approach is risky and can create confusion and frustration among employees.
strategies. Leadership from HR, IT, and the business must join forces to shape an effective
reskilling strategy. HR brings both an understanding of how to manage change and
culture along with tactical implementation expertise. IT brings the technology
knowledge, and business leaders can work directly with employees to define how AI
can augment the workforce within each business domain. Have the joint team report
directly into the C-suite and define measures to hold them accountable.
Predict every If you only expect a third of your workforce will need reskilling or upskilling over the
employee’s next few years, you aren’t thinking big enough. Just as you forecast product demand,
potential. predict what employees will need to succeed in a rapidly evolving workplace.
Look beyond just current skills to employee potential. Use AI-powered HR tools to
anticipate how an individual might develop, perform, or contribute based on skills,
talents, personality traits, experiences, and educational background.11
Share a blueprint You may not know exactly what lies ahead, but you can communicate your vision
for the workplace for the future of work. From routine business operations to brand-defining areas,
of tomorrow. AI creates anxiety as employees worry about being replaced or not having the skills
they need. Share your plans for automation versus augmentation with your workforce
and help them see how AI will create new opportunities and enable them to do their
jobs faster and better—from designing products to creating promotions to managing
inventory. Consider how employees will use—and benefit from—technology as
carefully as you consider the tech investment itself.
12
Part three
Safeguarding brand trust
With so many products vying for consumers’ attention,
AI can either bolster or undermine a brand’s trust.
Trust is paramount for both consumers and industry CEOs. Our 2024 consumer
research report showed that 9 out of 10 consumers value trust when choosing a
brand.12 Similarly, 73% of retail and consumer products CEOs in our 2024 CEO study
said trust will have a greater impact on their organization’s success than any specific
product or service.13
But AI adds new dimensions to the issue of trust, with risks impacting both
business partner and customer relationships. Consumers are already wary of AI in
general—only 53% trust the technology, falling from 61% over the past five years.14
And within the partner ecosystem, companies need to know that each member is
practicing trustworthy AI.
Retail and consumer products executives recognize that AI creates risks that can
erode trust. Nine in 10 say misuse, such as creating misleading information, is their top
worry associated with AI models, followed by privacy (85%), fairness and bias (80%),
explainability (76%), and transparency (73%). For example, biased models can
alienate customers. One consumer survey revealed that almost two-thirds of
consumers avoid AI-fueled recommendations because they are biased or
stereotypical.15
At the same time, these risks are slowing progress with generative AI opportunities.
57% of executives say data accuracy and bias is a barrier to gen AI adoption. 55% also
cite privacy and confidentiality of data and 54% are concerned about cybersecurity.
Despite these concerns, organizations are struggling to enable the tools that can help
them manage the risks. Companies have created a foundation: 87% of executives say
they have clear AI governance structures. But less than a quarter of companies have
advanced implementation of tools to assess, monitor, and manage AI governance
90% of executives
(see Figure 4). “Showing your work”—designing solutions with explainability and
cite misuse as
transparency built in—will be critical to instilling confidence in consumers regarding
their top concern
your use of AI.
with AI models.
13
FIGURE 4
Few brands have advanced implementation of
tools to help them manage their AI governance
policies and activities.
Figure 4
Few brands have robust implementation of tools to help
them manage their AI governance policies and activities.
Approach to AI governance Advanced implementation of tools
84%
have defined roles and
responsibilities for all stakeholders
involved in AI
11%
have advanced implementation of AI
accountability tools
91%
conduct ethical impact assessments
to evaluate the impact of AI
initiatives on different stakeholders
16%
have advanced implementation of AI
bias and fairness tools
87%
have established clear organizational
structures, policies, and processes
for AI governance
23%
have advanced implementation of AI
governance frameworks or policy tools
90%
build explainable models that can be
easily understood and audited
24%
have advanced implementation of AI
transparency and explainability tools
77%
conduct regular risk assessments to
identify potential security threats
26%
have advanced implementation of AI
risk and safety tools
Q. To what extent do you agree with the statements about your organization’s approach to AI governance? Percentages represent
those who agree and strongly agree. Q. To what extent has your organization implemented tools to assess, monitor, and manage
the following? Percentages represent those who responded “fully implemented, reviewed, and updated regularly.”
14
Case study
PepsiCo models a structured approach that enables Using gen AI to streamline
it to scale AI responsibly. The company began
regulatory management
by establishing a formal responsible AI framework
and assembled a dedicated team to support it. The across regions18
team then developed comprehensive policies and
standard operating procedures to operationalize
their AI principles. The governance board assesses,
validates, and approves gen AI use cases against
its responsible AI principles, sharing best practices A multibillion-dollar global consumer products
and accelerators, and helping mitigate risks. The company operates in the highly regulated agricultural
company is also building a platform that provides products industry. It devotes significant resources
comprehensive governance of models, inputs, to managing compliance with local regulations,
and outputs.16 staying current with continuously changing
guidelines, and integrating compliance into the
Regulations are also intended to support trustworthy product development process.
AI, but a lack of consistent guidelines across
jurisdictions complicates implementation and stalls To help its product compliance and development
plans. In fact, nearly half (46%) of industry CEOs said teams reduce heavy manual workloads and free
their concern about regulations as a barrier to gen AI up more time to work strategically, the company
has increased in the last six months.17 worked with IBM to develop a generative AI-powered
regulations assistant. This solution features a
However, AI can help companies manage the conversational user interface and provides a single
complexity. By automating the monitoring and source of truth for over 1,000 regulations impacting
analysis of detailed regulatory requirements, AI worldwide operations.
enables organizations to quickly identify potential
issues and take corrective action. Executives plan The regulations assistant enables product
to significantly increase their use of AI/gen AI in compliance employees to predict the impact
regulatory compliance over the next year. In product of regulatory intent, summarize regulatory
design and development, the percent increases from requirements, and compare regulations globally,
53% to 79%, for sustainability, 74% to 88%, and for faster than with manual processes. The AI tool also
financial and regulatory monitoring and reporting, enables product developers to analyze the impact
66% to 94%. of regulations on product portfolios, review
solution options, and query product specifications
in a conversational journey.
To date, the regulations assistant has demonstrated
that generative AI can orchestrate regulations data
quickly and drive closer collaboration across borders
to leverage regulatory success across the business.
In product design and The tool also has the potential to increase efficiency
by 8% to 13%, increase productivity by 10% to 15%,
development, executives
and increase profits by over $165 million during the
plan to increase use of AI next five years.
and gen AI to manage
regulatory compliance from
53% today to 79% in the
next year.
15
Action guide
Make trusted AI a brand differentiator.
Customer-obsessed businesses need to deliver on what their written
policies dictate for responsible AI practices. Build confidence in
responsible internal uses of AI before expanding to customer-facing
use cases where broken trust can damage your brand.
Purge bias from To provide transparency and explainability, define clear guidelines to monitor for
your algorithms. discriminatory patterns. For example, conduct regular audits on historical purchasing
and customer data that may reflect stereotyping and societal biases. Facilitate
human-AI collaboration and oversight with training that helps employees understand
and recognize fairness and bias. Prioritize diversity on your AI development teams.
Establish a data governance framework to support data provenance, helping ensure
your data is authentic and trustworthy. Maintain detailed records of bias mitigation
efforts, create dedicated channels for bias-related feedback, and regularly
incorporate insights into system improvements.
Leverage AI to To stay ahead of an AI regulatory environment that is evolving at varying paces
proactively navigate globally, use AI solutions to capture regulatory intent across multiple channels
regulations. and forecast its impact. Choose AI development tools that build in governance and
regulatory compliance management end to end. Proactively compare old and new
regulations to quickly identify key focus areas within impact assessments. Automate
tools to stay up-to-date and streamline audit processes.
Be open about Build trust with customers by being up-front about data collection as well
your use of AI as how and where you are using AI. Offer opt-out options and avoid tech-speak
with customers in your explanations. Exchange AI roadmaps and strategies with business partners.
and partners. Demonstrate your commitment to responsible AI practices and request the same
of your partners.
16
Authors
Dee Waddell Contributors
Global Industry Leader
Consumer, Travel & Transportation Industries The authors would like to thank the following
IBM Consulting for their contributions to this report:
[email protected]
From IBM Consulting:
linkedin.com/in/waddell/
Arnab Bag, Distribution Market HCT Service
Joe Dittmar Line Leader
Senior Partner Rich Berkman, Vice Pr |
137 | ibm | ibm-the-intuitive-supply-chain-report.pdf | IBM Institute for Business Value | Research Brief
The intuitive
supply chain
Predict disruption, deliver growth
Key takeaways
Generative AI can preempt supply Generative AI has put supply chains in flux. 64% of Chief Supply
Chain Officers say gen AI is completely transforming workflows.
chain disruption and unleash
Supply chain teams must work differently. 60% of operations and
growth opportunities. automation executives say AI assistants will handle most traditional
and transactional processes by 2025.
More decisions will be automated. Operations and automation
executives say generative AI will increase the volume of decision-
making by digital assistants by 21% in the next two years.
Predictions will improve, igniting sustainable innovation. 76% of
supply chain and operations leaders say gen AI will help innovate their
product design and make product lifecycles more sustainable.
The intuitive supply chain: Predict disruption, deliver growth 2
Introduction
Make agility your supply
chain superpower
Would a peek at next week’s headlines change
your supply chain strategy today?
The intuitive supply chain: Predict disruption, deliver growth 3
Supply chain certainty is an elusive target. The combined power of generative AI and
With so many fault lines stretching across cloud computing could make that possible.
the business landscape, it seems By harnessing the potential of machine
impossible to accurately predict what will learning, automation, and advanced
happen tomorrow. Supply chain leaders analytics in a hybrid cloud environment,
must often adopt a siege mentality, looking organizations can gain a sixth sense,
for ways to limit their losses as plan B anticipating everything from demand
quickly gives way to plans C, D, and E. fluctuations to sourcing delays. With this
foresight, they can reinvent their supply
But what if you could spend this time
chain strategies, shifting from a reactive
spurring growth? What if you could predict
to a proactive stance.
the future accurately enough to give your
business a competitive edge?
17% 72%
Leaders in gen AI adoption and data-led
innovation—those who view gen AI capabilities
as the primary driver of their automation
report higher annual revenue report greater annual
investments—are reaping outsized rewards. growth than the competition net profits
The intuitive supply chain: Predict disruption, deliver growth 4
Already, leaders in gen AI adoption and That’s a big problem for supply chain executives, and automation executives employees. Part two explains how
data-led innovation—those who view gen AI leaders, who know they need to invest from organizations that are currently accelerating supply chain intelligence can
capabilities as the primary driver of their in next-gen tech today to make their implementing AI-enabled automation. help companies leverage real-time data
automation investments—are reaping operations more agile and resilient for We discovered that these leaders are faster and more effectively than ever
outsized rewards. They report 72% greater an uncertain future—from dynamically focused on creating what we call “the before. And in part three, we’ll explore how
annual net profits and 17% higher annual rerouting shipments and adjusting intuitive supply chain”—agile, adaptive, and gen AI-enabled digital twins, or virtual
revenue growth than the competition. And all production schedules in real time to perpetually prepared, safeguarding brand models, can help organizations improve
the supply chain leaders we surveyed expect identifying bottlenecks and risks before reputation, customer satisfaction, and the their position in the competitive landscape,
their revenue growth from AI-enabled they materialize. bottom line. as well as in the eyes of customers. We
operations to more than double over the next conclude with an action guide that outlines
How can gen AI solve these persistent In this paper, we’ll lay out the steps
three years.1 how to plan, prioritize, and perform to
supply chain problems? To find out, the IBM organizations are taking to get there. In part
make every move count.
Looking at these numbers, it’s no surprise that IBV, in partnership with Oxford Economics, one, we’ll explore the role of AI assistants,
72% of the top-performing CEOs we surveyed surveyed more than 2,000 global Chief which are quickly becoming less like
for the IBM Institute for Business Value (IBM Supply Chain Officers (CSCOs), operations chatbots and more like full-time
IBV) 2024 CEO Study say competitive
advantage now depends on who has the most
advanced gen AI. But the high-speed race to
meet short-term goals is hindering their
progress. Overall, global CEOs agree that a
focus on short-term performance is their top
barrier to innovation—and 66% say their
Supply chain leaders need to invest in
organization is currently meeting short-term
targets by reallocating resources from next-gen tech today to make their operations
longer-term efforts.2
agile and resilient for an uncertain future.
The intuitive supply chain: Predict disruption, deliver growth 5
Part one
Lean into the power
of decision support
Employees paired with AI assistants will deliver
more business value than either could alone.
The intuitive supply chain: Predict disruption, deliver growth 6
Today’s supply chain teams are drowning in a sea of disconnected data.
They increasingly have access to the long-awaited real-time information
they need to make smarter, faster, decisions—but there’s so much to sift
through that many opportunities go unnoticed until it’s too late. to ask for the information they need—and find out where it came
from—with a few simple prompts.
Gen AI-powered digital assistants are changing all that. With their
ability to analyze vast stores of data almost instantaneously, they can For example, AI assistants can analyze which supplier is contributing
bubble up critical insights for supply chain teams to skim from the the most to delays and identify issues causing disruption, such as
surface. Plus, their natural language skills make it easy for employees weather, financial obstacles, or transportation bottlenecks. Then,
AI-fueled predictive models can outline how the situation is most likely
to evolve, allowing AI assistants to offer targeted recommendations
that help supply chain teams prepare for what’s next.
Already, 60% of executives say AI assistants will handle most
traditional and transactional processes by 2025.3 And 90% say their
60% organization’s supply chain workflows will incorporate intelligent
automation and AI assistants by 2026.4
of executives say AI assistants
will handle most traditional and
transactional processes by 2025.
And
90%
say their organization’s supply chain
workflows will incorporate intelligent
automation and AI assistants by 2026.
The intuitive supply chain: Predict disruption, deliver growth 7
When employees use gen AI assistants to It’s not just about explaining how materials By leveraging AI assistants, CSCOs can
quickly query their supply chain platform will get from point A to point B. It’s also aggregate and distill intel, bringing insight
for credible data, rather than manually measuring the supply chain cost of every to the boardroom quickly and confidently
searching multiple systems, they can business decision—and making sure those and making sure supply chain implications
manage change faster—and pivot more costs are considered from the start. Beyond continue to inform strategies as they
precisely. Instead of using the dedicated the sales a new SKU will drive, product evolve. As decisions are made and then
procurement solution to change purchase development strategies should account for tested in the market, AI assistants can
order delivery dates, for instance, the total cost of ownership, forecasting the accelerate the feedback loop, giving
employees can simply ask their assistant cost of delivering a new item in conjunction executives the real-word, real-time data
to make the change for them. with the losses that come from holding on they need to see if their strategies are
to products that don’t sell. delivering the desired results—and change
But that’s only the beginning. Supply chain
tactics quickly if they aren’t.
teams aided by AI assistants are cultivating Then there’s the sustainability dimension.
a new human-technology dynamic that will As both consumers and regulators demand
touch virtually every point of the supply more comprehensive reporting on
chain, from planning to sourcing to environmental impact, supply chain leaders
manufacturing to distribution. In fact, must be able to track sustainability metrics
64% of CSCOs say gen AI is completely all the way to the last mile—and do the hard
transforming their supply chain workflows. work of designing more eco-friendly product
And CSCOs and automation executives say lifecycles. This is another place where gen
gen AI will increase the volume of decision- AI can help, with 76% of supply chain and
making by digital assistants by 21% in the operations leaders agreeing that it will help
With gen AI assistants, employees can manage
next two years. innovate their product design and make
product lifecycles more sustainable.
change faster—and pivot more precisely.
The intuitive supply chain: Predict disruption, deliver growth 8
Case study
Building an intelligent supply chain
using a supply chain AI assistant
IBM employs supply chain staff in 40 which impeded collaboration and At a high level, the IBM supply chain The system uses IBM’s AI technology to
countries and makes hundreds of thousands real-time data transparency. digital transformation revolves around enable natural language queries and
of customer deliveries and service calls in building sense-and-respond capabilities. responses, which accelerates the speed of
IBM supply chain management set out
over 170 nations. IBM also collaborates with This was accomplished by democratizing decision-making and offers more options
a bold transformation vision to build a
hundreds of suppliers across its multitiered data—automating and augmenting to correct issues. Users can ask, in natural
cognitive, intelligent supply chain more
global network to build highly configurable decisions by combining a cognitive control language, about part shortages, order
than a decade ago. The aim was to have
and customized products to customer tower, a cognitive advisor, demand-supply impacts, and potential trade-offs. To date,
an agile supply chain that extensively
specifications. Historically, the IBM supply planning, and risk-resilience solutions. IBM has saved $388 million related to
uses data and AI to lower costs, exceed
chain ran on legacy systems spread Now, the cognitive control tower has reduced inventory costs, optimized
customer expectations, ruthlessly
across different organizational silos, evolved into an enhanced generative AI shipping costs, faster decision-making,
eliminate or automate non-value-add
making information-sharing slow and intelligent layer using a supply chain and time savings (days to hours to minutes
work, and exponentially improve the
incomplete. Employees also performed digital assistant. to seconds).
experience of supply chain colleagues.5
much of their work on spreadsheets,
The intuitive supply chain: Predict disruption, deliver growth 9
Part two
Accelerating supply
chain intelligence
If your data could talk, what would it say?
Supply chain teams are about to find out.
The intuitive supply chain: Predict disruption, deliver growth 10
Whether disruption is caused by In fact, the executives we surveyed
geopolitical conflict, climate catastrophes, anticipate operational performance, 73%
or increasing complexity, supply chain enterprise agility, and strategic advantage
leaders will be judged by their ability to find to be the top three benefits of using gen AI
effective workarounds. And they’re looking investments in their supply chain. And 73%
to gen AI to make their supply chain more say gen AI is already accelerating their
agile, adaptive, and future-proofed. high-impact automation initiatives. of executives say gen AI
is already accelerating
their high-impact
automation initiatives.
The intuitive supply chain: Predict disruption, deliver growth 11
Perspective
The key is to make the entire ecosystem The convergence of gen AI and cloud-based Future-proof your supply chain with
more responsive. By allowing gen AI solutions has also enabled autonomous
cloud-enabled innovation
assistants to interact directly with the automation (see “Future-proof your supply
intelligent layer of the supply chain system— chain with cloud-enabled innovation”).
With the combined power of cloud computing and generative AI, companies can
the cognitive core that pulls insights from In addition to automating workflows,
accelerate supply chain innovation and improve business outcomes to a degree that
vast stores of data—internal and external gen AI assistants can automate the process
wasn’t previously possible.
teams can collaborate more seamlessly. of workflow reinvention. They can learn from
supply chain metrics and transaction history, Deploying gen AI on the cloud lets companies train and deploy models faster and
The goal is for AI assistants to continually
make proactive recommendations, and even at scale, without the need for expensive hardware or infrastructure. It lets multiple
communicate the intelligent layer’s findings
repurpose or redefine new workflows based teams collaborate on the development of gen AI models, moving them between
to the appropriate part of the supply chain
on what they’ve learned. different cloud environments and integrating them with other cloud-based services
team, along with recommended actions.
and applications seamlessly.
While the enterprise resource planning (ERP) This helps streamline workflows to make
system remains the system of record and them more efficient, cost-effective, and Then, of course, there’s cost to consider. With pay-as-you-go pricing, cloud
core transaction engine, supply chain teams environmentally responsible. In fact, 63% infrastructure can ease capital expenditure constraints, allowing companies to
no longer need to interact with it directly. of supply chain and operations leaders say focus on innovation, rather than the financial implications of investing in new tech.
And that goes for other specialized supply integrating sustainability and circularity into When applied strategically, this tech combo can improve efficiency, reduce costs,
chain apps, from procurement to warehouse workflows is a key reason their organization and increase agility. Here are a few ways your supply chain can benefit from cloud-
management to transportation logistics, is investing in automation. enabled innovation powered by gen AI:
as well. This approach lets employees drill
deeper, allowing for real-time analysis and – Forecast future demand. Optimize inventory levels, reduce stockouts or over-
optimization each step of the way. stocking, and improve cash flow.
– Optimize delivery routes. Reduce fuel consumption, lower emissions, provide
dynamic distribution, and improve delivery times.
– Manage supply chain risk. Predict the likelihood of disruption and recommend
proactive mitigation measures.
– Increase supply chain visibility. Identify bottlenecks and recommend corrective
actions teams can take to keep operations from being disrupted.
The intuitive supply chain: Predict disruption, deliver growth 12
Case study
Achieve end-to-end visibility
with AWS Supply Chain
Supply chains are vast, interconnected networks. The multitude of The cloud-based AWS Supply Chain Address data fragmentation
business application directly addresses
participants, disparate systems, and lack of seamless data sharing A supply chain data lake harmonizes
these challenges. By harmonizing disparate
disparate data into a flexible, scalable
make it difficult to accurately forecast future demand, track inventory
data sources into a unified supply chain
canonical data model that aggregates and
levels, and align supply. The fragmentation of data hinders supply data lake, it lays the foundation for
associates supply chain information into
improved end-to-end visibility, forecasting
chain planners’ ability to understand fluctuations, predict future needs a unified data asset. By incorporating a
accuracy, inventory optimization, and
generative AI-powered data onboarding
precisely, and position optimal inventory where it’s needed most. overall supply chain resilience.6 Here are a
agent, companies can also automate data
few of the key business benefits of moving
transformation from any native format into
to this type of cloud-based solution:
the data lake’s canonical model. Customers
can seamlessly extract and upload raw
data, with the agent leveraging large
language models for automated data
mapping through a guided, module-driven
user interface experience.
The intuitive supply chain: Predict disruption, deliver growth 13
Case study (continued)
Increase forecast accuracy Improve supply chain visibility Improving supplier visibility Simplify sustainability
and collaboration compliance processes
Machine learning-powered forecasting The AWS business application can examine
capabilities can help organizations improve warehouses, distribution centers, and stores The AWS application analyzes supplier lead Cloud-based sustainability features create
forecast accuracy and reduce excess in detail, showing on-hand, in-transit, and times, makes future projections compared a more secure and efficient way to obtain
inventory levels. Machine learning at-risk inventory levels. It then uses machine to orders and forecasts, then identifies mandatory documents and datasets from
algorithms can incorporate variables such learning algorithms to automatically issues. It displays all connected trading your supplier network. You can request,
as seasonality, product characteristics, generate, score, and rank multiple inventory partners, enabling supply chain leaders to collect, and export artifacts, such as
vendor characteristics, and destination- rebalancing recommendations to mitigate view and collaborate across multiple tiers. product lifecycle assessments, certificates
origin sites, along with historical order risks. Gaining visibility into network-wide Built-in chat and messaging capabilities on product safety, or reports on hazardous
history, to train the model. inventory levels, movement patterns, and also facilitate seamless communication substances used at any point in the supply
potential risks empowers organizations to and data sharing. chain. Amazon’s Global Trade and Product
optimize inventory positioning and mitigate Compliance (GTPC) team used the AWS
imbalances, overstocks, and stockouts. application’s sustainability features to
transform their compliance data
management process and now expect to
save approximately 3,000 operational
hours per year.
The intuitive supply chain: Predict disruption, deliver growth 14
Part three
Visualize
the future
By using supply chain data to
fuel gen AI-powered virtual models,
companies can unlock a new level of
operational efficiency and resilience.
The intuitive supply chain: Predict disruption, deliver growth 15
Supply chain leaders have long imagined
a future where real-time data flows
seamlessly between IT and operational
By 2026,
technology (OT) systems, enabling a more
77%
agile approach that reacts to constant
change. And their dream is finally
becoming reality.
Think of a manufacturing facility, where
of executives believe gen AI will
operations teams already use AI sensors
enable connected assets to
to detect changes in vibration patterns, make autonomous decisions.
temperatures, power consumption, and
even sound patterns. While traditional AI
can alert teams to signals as they appear—
and even predict when breakdowns are
about to occur—employees must manage 76%
necessary adjustments or repairs based
on this information.
With generative AI, that’s no longer the
case. When paired with vision sensors, of executives say they expect to
gen AI lets connected machines use gen AI to derive differentiated
outcomes from connected assets
self-predict and self-adjust in a harmonious
in the next two years.
fashion, unlocking unprecedented levels of
productivity and efficiency.
The intuitive supply chain: Predict disruption, deliver growth 16
In fact, by 2026, 77% of executives expect It works like this: First, data from drones, support “what-if” risk analysis by predicting With the right perspective, supply chain
gen AI will enable connected assets to robots, cameras, and other connected potential problems—from raw material leaders can look beyond productivity plays
make autonomous decisions. And when assets flow into a unified platform with shortages to multiple supplier plant closings to pull the levers that drive growth. By using
complex asset ecosystems work in a geospatial layer, an information layer, simultaneously—and recommending gen AI to orchestrate multiple data sources,
harmony, they can help businesses achieve and an orchestration layer. Time-lapsed respective contingency plans. systems, and tools, they can inspire
results that weren’t previously possible. visualizations then let supply chain teams innovation across the ecosystem—and
These simulations can also inform product
Executives recognize this potential, with see how specific changes have impacted inform the strategic decisions that set their
development by helping teams identify
76% saying they expect to use gen AI to the ecosystem in the past—and make organization apart.
where waste and inefficiencies can be
derive differentiated outcomes from real-time decisions as situations unfold
removed from the process. This is a key
connected assets in the next two years.7 in the present.
concern for executives, who say visibility
But boosting efficiency is just the first step. Gen AI-enabled virtual models can then of full product lifecycle management and
Businesses can derive much deeper value help teams simulate how future events environmentally sustainable products and
from interconnected data when they use it could affect supply chain operations. services are two of their top automation
to visualize the end-to-end supply chain— They use real-world data and algorithmic priorities for their operations functions over
and simulate how disruption could impact techniques to visualize how the dominos the next three years.
operations each step of the way. will fall in response to different disruptions
to help teams plan accordingly. They
Look beyond productivity plays to pull the
levers that drive growth.
The intuitive supply chain: Predict disruption, deliver growth 17
Case study
Improving pharma supply chain
visibility for patient safety8
Amid the increasing proliferation of counterfeit, falsified, or Seeking safety through transparency players in the prescription drug supply chain.
By connecting through these APIs, Pulse
substandard prescription medications, the US government passed Working with IBM Consulting and AWS,
users can search for trading partners, verify
NABP built a new digital platform called
the Drug Supply Chain Security Act (DSCSA) with the aim of protecting
trading partner status, exchange digital
Pulse that lets its member users track and
patients. It’s rooted in the idea that transparency—the ability to credentials, and perform electronic tracing.
share each prescription drug’s ownership
accurately trace prescription meds throughout the pharmaceutical transaction records, providing increased
The platform enables visibility and
supply chain visibility.
supply chain—is essential to preserving its integrity. collaboration, eliminates tedious
administrative work, and, most importantly,
One key design aspect of the platform—
Just as important is the idea that all the major players in the pharmaceutical ecosystem— creates a more secure supply chain to
which runs on the AWS cloud—is the
manufacturers, wholesalers, dispensaries, and regulators—need a way to share information protect patients.
integration of APIs from providers of the
collaboratively to make it happen. Prompted by the challenge of multiple industry segments
“point” tracking solutions used by most
needing to cooperate to address DSCSA, the National Association of Boards of Pharmacy
(NABP) sought to create a digital platform that would bridge the interoperability gaps
between systems, making compliance with DSCSA faster and easier.
The intuitive supply chain: Predict disruption, deliver growth 18
Action Guide
Make every move count
In the complex game of supply chain chess, executives must always
think several steps ahead. Modernizing supply chains isn’t just about
adopting new technologies or processes—it’s about embracing a new
way of thinking, one that’s rooted in scientific inquiry, experimentation,
and a relentless pursuit of progress.
By applying the scientific method at scale, enterprises can tap into the vast potential of data
and gen AI to drive critical improvements in business strategy, product development, and
global supply chain operations. In fact, 62% of CSCOs say gen AI will accelerate the pace
of discovery, leading to new sources of product and service innovation.9
With the promise of discovery as their guiding light, companies can unlock the full potential
of their supply chains, power ecosystem partnerships, and drive sustainable profitability and
growth. Here’s what leaders across the supply chain ecosystem should do to predict and plan
for endless disruption—and profit from the opportunities volatility can create.
The intuitive supply chain: Predict disruption, deliver growth 19
Action Guide
1. Plan Identify benefits you want to deliver. management to improve decision-making Understand skills requirements
efficiency and speed-to-action. Invest to and gaps.
Investigate the key drop-out points
bring the vision to life and facilitate a
between analysis and action, identifying Create user personas across the range of
seamless and fulfilling experience across
how improvements could flow through into supply chain workflows. Outline how digital
the entire supply chain.
financial and operational performance. assistants will help create new workflows
Outline the productivity KPIs that will be Know the specific functionality and and enhance existing ones. Identify the
targeted for improvement and define systems architecture you need. gaps in skills between these personas and
success criteria. the current state, then define training and
Identify the solutions that will provide
upskilling plans.
Define your employee every feature. Then use an orchestration
experience vision. engine as a process conductor, issuing Keep your eyes on the prize.
precise commands to multiple agents
Provide easy access to relevant AI Align supply chain innovation to your
based on user prompts. Leverage
analytics, recommendations based on role, market offering and the capabilities needed
synthesized data from the integration layer
and intelligent transactional workflows in to deliver it. Prioritize these areas and be
to create dynamic, intelligent workflows
the employee portal. Find ways to integrate confident in delivering them.
that deliver the desired outcomes.
supply chain processes into the employee
experience framework, such as
streamlining logistics and inventory
The intuitive supply chain: Predict disruption, deliver growth 20
Action Guide
2. Prioritize Define supply chain workflows Don’t try to cut your way to growth. Define rules of engagement.
that have the greatest potential
Make the investments needed to Be clear about who is accountable and
for automation.
fundamentally transform ways of working. responsible for specific workflows—and
Map the key points across the workflow Focus spending on the areas that can make who gets a say. Set ground rules for using
that cause rework and manual analysis. your supply chain more agile and resilient. digital assistants and make sure everyone
Be honest about the true nature of your knows how they’re expected to evolve.
Prioritize getting to scale.
processes, not the idealized version that
may be documented somewhere. Invest in initiatives that can quickly
transition from pilot to deployment at scale.
Stop looking for a silver bullet.
Use success in specific areas to build
Be honest about where investment is momentum for the wider transformation.
needed within your current technology
landscape. Set specific timelines for
upgrades or the deployment of new
solutions. Don’t let time and effort that
have been invested in previous solutions
become an anchor that prevents you
from achieving future success.
The intuitive supply chain: Predict disruption, deliver growth 21
Action Guide
3. Perform Feed generative AI data that Review and align to Keep score.
supports supply chain productivity. changing conditions.
Track benefits as they’re delivered to build
Map the full range of data initiatives needed Cultivate a supply chain that can sway with momentum and confidence in new
to connect people and technology. Upskill the winds of change to deliver a competitive technologies. Demonstrate ROI to secure
employees and train tools to speed advantage. Adopt a technology architecture continued investment. Make data-driven
decisions. Identify the key touchpoints to that allows new capabilities to be plugged decisions that can fuel growth and
use gen AI to boost productivity. in without disrupting the user experience. performance improvements.
Put trust in data.
Don’t let people tinker with the workflow
outputs from the system. Where processes
are automated and tested, let the system
run and do its job. Don’t allow competing
forms of analysis designed to suit individual
agendas interfere. Instead, encourage
employees to engage in advanced analysis,
using their assistants to innovate and
address the complexities of interconnected
operations and systems.
The intuitive supply chain: Predict disruption, deliver growth 22
Authors Research methodology products, electronics, telecommunications, IBM Institute for
government, healthcare/life sciences,
Business Value
Amar Sanghera The IBM Institute for Business Value (IBM consumer products, retail, and
AWS Supply Chain Solutions Global Leader, IBV), in conjunction with Oxford Economics, transportation/logistics, each comprising
For two decades, the IBM Institute for
Digital Supply Chains Go-to-Market Strategy interviewed and surveyed more than 2,000 5% to 15% of our total respondent sample.
Business Value has served as the thought
executives with equivalent roles and titles, The size of organizations surveyed, in terms
Michael Mowat including Chief Supply Chain Officer (CSCO), of revenue, ranged from $500 million to leadership think tank for IBM. What inspires
Supply Chain Strategy and Operations Chief Operations Officer (COO), Chief $500 billion, with a mean of $26 billion. us is producing research-backed,
Leader, Finance and Supply Chain Automation Officer (CAO), Chief technology-informed strategic insights that
Transformation, IBM Consulting Information Officer (CIO), and Chief The IBM IBV ran a series of contrast help leaders make smarter business
Financial Officer (CFO). analyses, including pairwise comparisons, decisions. From our unique position at the
Karen Butner highlighting results and differences as
intersection of business, technology, and
Global Research Leader, AI and Automation; In 2024, CSCOs, COOs, and automation shown in this report. Statistical significance society, we survey, interview, and engage
Supply Chain Operations, IBM Institute for executives were also polled about their for all pairwise comparison contrasts was with thousands of executives, consumers,
Business Value, IBM Consulting investments, priorities, and use cases to set at the (p = .05) level, meaning there is
and experts each year, synthesizing their
assess the current impact of generative AI only a 5% chance that the observed
perspectives into credible, inspiring, and
Contributors initiatives, as well as the results they expect differences or relationships between the
actionable insights. To stay connected and
to see in the next two to three years. The groups are due to random variation.
informed, sign up to receive IBV’s email
goal of these surveys was to understand
IBM Consulting The right partner for newsletter at ibm.com/ibv. You can also
how global executives view the impact of
Chris Moose, Lead Client Partner NABP, find us on LinkedIn at https://ibm.co/
gen AI on their organizations’ performance a changing world
Public Sector ibv-linkedin.
and competitive advantage across the
Jonathan Wright, General Manager,
NCE Europe supply chain. At IBM, we collaborate with our clients,
bringing tog |
138 | ibm | ibm-consulting-reimagined-powered-by-ai-report.pdf | IBM Institute for Business Value | Research Brief
Consulting
reimagined,
powered by AI
Foreword
On the cusp
of convergence
We are embarking on one of the most That is why this particular study from the At IBM we are embracing this change. We
significant transformations of our time as a IBM Institute for Business Value is so are supercharging our 160,000
result of AI, across nearly every industry and unique. We decided to take on the consulting consultants’ expertise with AI support so
profession. Consulting is no different—in and AI topic directly, by diving into insights we can create better outcomes, faster and
fact, it is likely to be one of the most from organizations across the globe that at scale, for our clients. We call our
disrupted, given the labor-based business engage with consultancies, to better approach the science of consulting—and we
at its core. We are on the cusp of a new era understand the expectations buyers have of believe, as the only global consultancy at
of consulting, one where science and their consulting partners in this AI era. This scale inside a technology company, we are
Mohamad Ali technology are being combined with skills study describes how clients expect unique in our ability to embrace this
Senior Vice President and expertise to create extraordinary value consultants to use AI to help them reimagine opportunity so we can be better partners.
IBM Consulting faster. Business models are changing as the what’s possible—and they also want AI to
We welcome your input and your thoughts.
consulting and client relationship evolves to help them get more for their money.
deliver new forms of value creation.
Consulting reimagined, powered by AI 2
Key takeaways
Software and talent are AI optimism Consulting Trust is central
converging to create new
abounds. buyers say to success.
value. Are your consultants
no AI, no deal.
accelerating progress—or
sticking to the status quo?
75% 66% 70%
of consulting buyers expect AI to say they’ll stop working with consulting of buyers say the use of AI in consulting
have a positive impact on their organizations that don’t incorporate AI will make them buy from fewer, more
use of consulting. into their services. trusted organizations.
Consulting reimagined, powered by AI 3
Introduction
Redefining As AI makes it easier for humans to wield Economics, surveyed global executives So, it’s no surprise that 86% of consulting
technology, software is supercharging who buy consulting services (see “Research buyers say they’re actively looking for services
consulting value talent—allowing people to create methodology” on page 6). We found that that incorporate AI and technology assets.
business value in ways that weren’t 75% expect AI to have a positive impact on Two-thirds go even further: They’ll stop
previously possible. their use of consulting. working with consulting organizations that
for an AI future
don’t incorporate AI into services provided.
In this environment, organization leaders
expect to work with consultants who are
Executives plan to increase using AI to tackle complex problems with
unprecedented speed and precision. They
overall consulting spend—but Figure 1
want trustworthy partners who are
they expect consultants to Consulting buyers want AI-powered services
obsessed with results—and can help them
deliver greater value with AI. adapt to rapid change.
To understand how expectations are
evolving, the IBM Institute for Business
Value, in collaboration with Oxford
89% 86% 80%
expect consulting say they’re actively say they want more
services to incorporate looking for services digital delivery models.
AI for improved that incorporate AI
productivity and quality. and technology assets.
Consulting reimagined, powered by AI 4
Figure 2
The evolution of consulting delivery
That’s because consulting buyers know But AI as a tool can only take organizations
they can’t run tomorrow’s business with so far. It needs to be married to the right
today’s skills, processes, or technologies. expertise, skills, and capabilities—and easy
In key functions—led by customer service, for people to use—to deliver real business 2000s 2010s Now
Globalization Cloud Generative AI
IT, research and development, and value. And this dependency promises to
marketing—organizations are already fundamentally change how consulting
tapping into AI to drive productivity and services are delivered and consumed. Business Expansion from Increased demand for New demand for
unlock resources for cycles of innovation need data center to cloud integration, digital labor and assets
business processes application migration, to deliver consulting
and new revenue streams. This explains
and modernization and IT services
why enterprise spending on AI surged 78%
between December 2022 and March 2024.1
Cost of Lower cost of skilled Lower cost of technical Scaled expertise and
Supply labor in offshore development and better quality at reduced
locations (e.g., India) deployment cost of service delivery
AI must be married to the right expertise, skills,
and capabilities—and easy for people to use—to
deliver real business value.
Consulting reimagined, powered by AI 5
Rather than replacing consultants, Research methodology
AI makes them even more essential.
The point of view developed in this paper has been informed
by insights obtained from a survey of 400 C-level executives
across 14 industries and 6 countries from June to August
2024. This was an anonymous, double-blind survey
The cost of reinvention conducted by Oxford Economics. That means respondents
were not aware that IBM originated the survey, nor does
IBM have direct visibility into the specific people or
organizations that responded. All respondents fit within
New technology improves productivity, of course. But truly transformational technologies, specified parameters: they lead organizations with an
including AI, also create opportunities to reimagine what can be done, creating whole new average annual revenue of approximately $33 billion and
businesses, industries, and economies (see “Innovation and expectations,” page 7). are large consumers of consulting services. The survey
asked questions about their current use of consultants and
Rather than replacing consultants, AI makes them even more essential. In fact, 86% of
how they view the introduction of AI in consulting. The data
consulting buyers expect to spend more on consulting in the future, and 94% expect AI to
from the survey has been complemented with insights from
positively impact consulting efficiency.
IBM client engagements, as well as interviews with
technology and consulting executives involved in the
As a consequence, average consulting spend, which is 2.8% of total revenue today, is
development and use of AI assets for consulting.
expected to climb above 4% by 2026—an increase of more than $500 billion annually when
applied across the Fortune Global 500.
But not all consultants are created equal. We’ll discuss how the consulting operating model
is changing, how consulting relationships need to adapt—and what business leaders should
look for as they build trusted, long-term partnerships for an AI-enabled future.
Consulting reimagined, powered by AI 6
Perspective
The truth about innovation
In 1930, John Maynard Keynes famously suggested that, within a
hundred years, people would only work 15 hours per week due to
the remarkable pace of technological progress.2 Things didn’t turn
out that way.
That’s because new innovations and software development, and cybersecurity— consultants to deliver existing services New technologies do more
technologies do more than make existing while giving bankers more time to focus on faster and cheaper. But they’re also creating
than make existing work more
work more efficient. They also increase our less transactional client needs, such as a vast range of new opportunities for
efficient. They also increase our
expectations about what can be done and building wealth. Automation, including consulting companies to deliver more and
what we want. robotic assembly lines, self-service kiosks, different value to clients—personalizing each expectations about what can be
and customer service chatbots, continue engagement beyond what was possible for
done and what we want.
Consider ATMs and online banking. While
to generate new opportunities and ways humans without AI assistance. This is
these innovations reduced the need for
of working. ushering in a new innovation model that is
bank tellers who handled deposits and
iterative, agile, and experience-centric—
withdrawals, they created new jobs in IT, The same is happening with consulting. AI
setting a new bar for executives to measure
and other digital technologies are enabling
consulting partners against for the future.
Consulting reimagined, powered by AI 7
A new consulting dynamic While buyers expect consultants to use AI opportunities to improve business
to help them reimagine what’s possible, they performance and designing solutions to
also want AI to help them get more for their deliver specific outcomes. Instead of
money. 78% of consulting buyers say they contracting for time-boxed engagements
Say goodbye to the one-and-done
expect cost savings and delivery benefits. with specific, limited deliverables,
consulting delivery model.
organization leaders and their consultants
It’s the dream scenario—more value faster,
build long-term partnerships that deliver
and at a lower cost. But can consultants
continuous value.
deliver on these demands? Only if they take
As AI opens a new world of business opportunity, the companies
a technology-driven approach focused on Consulting firms become less focused on
that reinvent themselves the fastest can gain a competitive edge. evidence-based experimentation, completing point-solution projects and more
continuous iteration, and user-centric focused on delivering sustained business
And executives expect their consultants to help them rise to
design. To achieve impact at scale they results across strategy and advisory,
this challenge. 89% of buyers now expect consulting services to
must embed expertise into AI models and systems integration, BPO and business
incorporate AI for improved productivity and quality—and 80% say codify knowledge so that it is consistent process operations, and cloud-managed
and replicable across engagements. services. Supported by AI assistants that
they want more digital delivery models.
In this model, consulting partnerships can manage much of the basic work involved
become much more symbiotic. Consultants with implementation, consultants can focus
aren’t contracted to solve a single problem on the complex, mission-critical tasks that
or streamline an isolated process. They’re require human expertise.
charged with identifying widespread
Consulting reimagined, powered by AI 8
Just as the emergence of software-as-a- language models to creating ready-to-use Figure 3
service made specialized software available AI offerings. The ability to access and use
Redefining the consulting
to a wider audience on a routine basis, AI high-quality data will be what sets these operating model
assistants have democratized many consulting organizations apart.
technical skills. When consultants are
Most importantly, buyers need to partner
supercharged with a network of easy-to-use
with providers that take a multi-modal and
AI tools, they can do more to drive
multi-model approach. They need access
transformation and accelerate innovation.
to platforms that integrate multiple
They can help address barriers that keep
capabilities into different workflows,
organizations from deploying new delivery
deployed in the environment that best
models, including organizational inertia,
meets their needs. Consulting organizations Traditional Consulting
security concerns, and resistance to change.
that have consumable, safe, and secure AI consulting reimagined
In this landscape, consultants with deep models, agents, and assistants working in
Opinion-led Evidence-driven
tech know-how are needed to gain a collaboration with each other and with
competitive edge. For instance, anyone human consultants can tap the full range
Unpredictable Reliable outcomes
could use an AI assistant to code new of AI capabilities—whether developed
applications, but an experienced software in-house or accessed through ecosystem
Siloed knowledge Democratized expertise
developer can use AI to develop higher partners—and deliver much greater value
quality code much faster. to consulting buyers.
Exhaustive planning Quick-to-iterate
To capitalize on new opportunities to drive It’s not just about having the right
differentiated innovation, consulting technology. It’s about harnessing the
Time-based Results-obsessed
buyers must look for providers that have wisdom of the crowd, as well as the
already made meaningful headway with expertise of consultants and partners, to
Encumbered by data Enhanced by AI
AI, from incorporating AI assistants into get the best possible solution to every
daily tasks to fine-tuning their own large problem, every time.
Linear processes Open collaboration
Consulting reimagined, powered by AI 9
Supercharge consulting
relationships with AI
AI assistants, reusable assets, and proven methods
power a continuous improvement flywheel.
So, what are consulting buyers All this can change how consulting
In every area of the business, buyers expect consultants to evolve looking for? relationships work. For example, when an
organization needs to redesign part of its
their people, partnerships, and pricing to reflect the possibilities People: Individual consultants should have
operation, a consultant using AI can
a vast repository of AI-enabled assets at
introduced by AI. For strategic development, for example, 63% of quickly tap into its data to quantify the
their disposal, as well as deep technology
potential impact of different initiatives.
buyers expect AI to be used as a supporting tool to a large extent expertise. They’ll be able to generate
Recommendations that used to take weeks
while only 24% see it as a source of automation. For IT support custom AI tools tailored to a specific client
to create can now be delivered in days.
or situation—and tap a knowledge base
services, 58% anticipate AI will be used for support while 38%
that can augment their experience at a
But consulting buyers should still expect a
expect it to be used for automation. speed, breadth, and depth unimaginable
personal touch. Consultants can use AI as a
up to this point.
sparring partner to encourage innovative
Consulting reimagined, powered by AI 10
Figure 4
Navigating the convergence of software and services
and creative thinking. But that only works Pricing: Today, consulting services are
Services Assets and platforms
when consultants have a strong tech typically charged based on hourly or daily
background to build on. The right people fees associated with the people delivering Productization of Assetization of services
services for value for productivity
still make the difference. the work. As consulting shifts from solely
people-based to a blend of human
Partnerships: Consulting buyers have
expertise and technology assets, pricing Labor-led consulting Product-led technology
become more discerning, looking for
will begin to reflect the value a partnership Skilled talent and expertise Automation of function
strategic partners that provide greater
delivers—not just the time it takes to get a
long-term value. 73% say the use of AI in 95% 5% 90% 10%
specific job done. Outcome-based pricing
consulting will make them more critical of
will become more important, as will value
the consulting services they buy and 70%
realization tracking and the incorporation
say it will make them buy from fewer–and
of asset licenses into consulting fee
more trusted–organizations. They expect
structures. Already, 73% of consulting Asset-led delivery
strategic partners to deliver more, help AI-enabled productivity
buyers say they want new pricing models
them develop the in-house skills needed to
from vendors because of their use of AI.
80-90% 10-20%
use their technology more effectively—and
bring a broader range of capabilities to each
engagement. The most in-demand
consultants will be those who act as
orchestrators and conduits for clients,
bringing the best of what consulting can
Value assets at scale
offer through a network of in-house and
Industry transformation
ecosystem capabilities.
50-70% 30-50%
Consulting reimagined, powered by AI 11
Building on a
foundation of trust
As with anything AI, governance will be key to AI-enabled consulting. More and bigger models and assets are concerned about the unethical use of AI
not always better. It’s about using the in consulting services.
Left unchecked, vast security and ethical risks can proliferate—with
right tools at the right time for what
costs spiraling out of control. Consulting buyers will need to be you’re trying to do. Good governance, As in the rest of the world, when it comes
strategic and responsible in selecting partners and using AI models trust, and transparency should be at the to consulting services, trust comes from
core of how assets are used. experience. That’s why business leaders
and platforms that fit their needs.
should look for consultants who spend
Business leaders across the board are every day on the front lines of applying
already demanding this discipline. 90% cutting edge technology to business.
say clear governance around AI in They’re most likely to know where AI can
consulting services is important. 93% be most effective—and its limits. Tapping
say they’ll only use consulting services experts who have been part of the
from organizations that are transparent development of technology assets is key
in their use of AI. And 82% say they’re (see “Engaging everyone for greater impact:
The watsonxTM challenge” on page 13).
Consulting reimagined, powered by AI 12
Case study
Engaging everyone for greater
impact: The watsonxTM challenge
IBM conducted a “watsonx challenge,” a hands-on experience After collaborating with team members to The assistants helped the team complete a
achieve one of these goals, many IBMers real-world project, generating user stories,
designed to bring the AI-building capabilities of watsonx to all IBM
shared that they gained a better tasks, source code, test scripts, and more.
employees.3 With more than 141,000 participants, employees
understanding of how AI can be applied to Using Consulting Advantage resulted in a
formed teams that focused on quickly crafting a solution to one of business needs. The interactive, hands-on 95% reduction in the delivery timeline.
challenge let participants move beyond Plus, by automating each step of the
several challenges, including:
abstract generative AI concepts to being process, the team estimates it reduced
able to design gen AI-infused solutions bugs by 97%.
– Create a custom productivity workflow – Create a RAG-based generative AI for realistic ways of working and
to enhance the already-automated workflow. (RAG is a method that customer challenges.
AskIBM experience. improves the quality of LLM-generated
For example, the team that won the
– Create new or combine existing AI responses by grounding the model on
Chairman’s Award, the highest prize the
assistants to address a client or role- external sources of knowledge.)
challenge had to offer, collaborated to show
based business need in a new or – Add new knowledge and skills to improve
how IBM Consulting Advantage, IBM’s AI
innovative way. the accuracy of answers provided by
assistant and agent platform, could be used
– Build a generative AI application for a use IBM’s AI tools.
to complete urgent projects.
case supporting internal team productivity.
Consulting reimagined, powered by AI 13
Action guide
1. Raise your expectations 3. Open your technology aperture
– Expect more—and expect it for less. Move – Embrace open technology, interoperability,
from a narrow focus on service-level and open standards to allow you to tap the
A new paradigm for the
agreements (SLAs) to SLAs plus objectives wide range of AI capabilities and assets. Use
and key results (OKRs). Encourage consulting open-source resources to reuse and adapt
services economy partners to look for continuous opportunities existing assets instead of creating new ones
to advance shared objectives and design for every need.
solutions focused on outcomes rather than
– Define how you want to share data and make
chase short-term fees and invoicing targets.
your systems accessible to consultants and
– Shift your consulting spend to enduring assets. Determine what data can be shared,
AI-enabled transformation is destined to have major ramifications
value. Make consumable technology assets a with whom, and for what purposes.
for operating models across industries and around the world. key part of your services purchase model to
deliver continuous value for your organization.
Organizations will start adapting their own internal services models Adapt procurement processes to allow for 4. Strengthen internal capabilities
and consume internal services in the same way they are changing iterative and faster consult-to-operate cycles. – Enhance your organization’s absorption
capacity. Prepare your people for new service
their use of consulting. delivery at higher speed through change
2. Focus on trusted relationships
management, training, and engagement.
We’re already witnessing significant impacts in functions including HR, finance, and – Select strategic partners that act as gateways
– Embed AI-enabled consulting in your internal
customer service, where AI has become the critical transformative ingredient. Blending to deeper and wider capabilities and and external services model and adapt the
assets. Ensure they blend their in-house
people and technology. Working collaboratively across AI and software—and organization. way you deploy services within the enterprise.
and ecosystem skills and technology for
Augment internal services functions with user-
These principles are becoming central to how services are designed, delivered, and
your advantage. Reduce coordination and
centric AI assets and new ways of working.
consumed. And there are important moves you need to make now. fragmentation challenges by placing the
responsibility for orchestration on trusted
Here are four key steps to help you start working with partners and advisors who take preferred partners.
a technology-fueled approach to consulting. – Establish governance, ethics, and security
guardrails for AI and make them the core
criteria for how you engage with consulting
organizations. Demand transparency in how
AI is used by consultants working for your
organization.
Consulting reimagined, powered by AI 14
Authors Contributors End Notes
Matthew Candy 1. As share of IT spend. Goehring, Brian, Manish Goyal, Ritika Gunnar,
Mihai Criveti
Global Managing Partner, Generative AI Anthony Marshall, and Aya Soffer. The ingenuity of generative AI:
Nduwuisi Emuchay
IBM Consulting Unlock productivity and innovation at scale. IBM Institute for
Karen Feldman
https://www.linkedin.com/in/mattcandy/ Business Value. June 2024. https://ibm.co/scale-generative-ai
Teresa Hamid
Blaine Dolph Chris Hay 2. Keynes, John Maynard. “Economic possibilities for our
grandchildren.” Essays in Persuasion. Macmillan & Co, London, 1933.
IBM Fellow and CTO, IBM Consulting Assets Amy Hutchins
and Industries
AB Vijay Kumar 3. Internal IBM data.
https://www.linkedin.com/in/blaine-dolph-
Eileen Lowry
5078b96/
Michelle Mattelson
Salima Lin Luq Niazi
Managing Partner, Strategy, Anthony Marshall
M&A, Transformation, Cindy Anderson
and Thought Leadership Tegan Jones
IBM Consulting
https://www.linkedin.com/in/salima-lin-
b17bb71/
Jacob Dencik
Research Director
IBM Institute for Business Value
https://www.linkedin.com/in/jacob-
dencik-126861/
Consulting reimagined, powered by AI 15
© Copyright IBM Corporation 2024
IBM Corporation
New Orchard Road
Armonk, NY 10504
Produced in the United States of America | October 2024
IBM, the IBM logo, ibm.com and Watson are trademarks of International Business Machines Corp.,
registered in many jurisdictions worldwide. Other product and service names might be trademarks
of IBM or other companies. A current list of IBM trademarks is available on the web at “Copyright
and trademark information” at: ibm.com/legal/copytrade.shtml.
This document is current as of the initial date of publication and may be changed by IBM at any time.
Not all offerings are available in every country in which IBM operates.
THE INFORMATION IN THIS DOCUMENT IS PROVIDED “AS IS” WITHOUT ANY WARRANTY, EXPRESS
OR IMPLIED, INCLUDING WITHOUT ANY WARRANTIES OF MERCHANTABILITY, FITNESS FOR A
PARTICULAR PURPOSE AND ANY WARRANTY OR CONDITION OF NON-INFRINGEMENT. IBM products
are warranted according to the terms and conditions of the agreements under which they are provided.
This report is intended for general guidance only. It is not intended to be a substitute for detailed research
or the exercise of professional judgment. IBM shall not be responsible for any loss whatsoever sustained
by any organization or person who relies on this publication.
The data used in this report may be derived from third-party sources and IBM does not independently
verify, validate or audit such data. The results from the use of such data are provided on an “as is” basis
and IBM makes no representations or warranties, express or implied.
Consulting reimagined, powered by AI 10a99804c32fd389-USEN-00 16 |
139 | ibm | ibm-generative-ai-governance.pdf | IBM Institute for Business Value | Research Brief
The enterprise
guide to AI
governance
Three trust factors that can’t be ignored
Foreword
Putting generative AI
governance in context
This Research Brief is part of an ongoing series of reports published by
the IBM Institute for Business Value (IBM IBV) about generative AI and
the opportunities and challenges it presents to organizations worldwide.
As business leaders adopt generative AI to boost competitiveness and increase productivity,
leaders need information on the ever-shifting landscape. Other reports include: 5 trends for
2024: Deep tech requires deep trust, The CEO’s guide to generative AI: Risk management, and
The CEO’s guide to generative AI: Responsible AI & Ethics.
The enterprise guide to AI governance: Three trust factors that can’t be ignored 2
Introduction
Why AI governance
matters more than ever
In less than two years, generative AI, the latest evolution of artificial Gen AI is cutting coding time from days to minutes, boosting content creation, personalizing
customer and employee interactions, automating cybersecurity operations, and optimizing
intelligence, has moved from novelty to business necessity. With gen
processes. But at the same time, AI-related risks are also on the rise: compliance and
AI strategies shifting rapidly from exploring to focusing to expanding,
regulation, data bias and reliability, and a loss of trust when users don’t understand AI model
77% of business leaders in a recent IBM IBV study say they are operation and governance.
convinced that gen AI is not only market ready, but that quick adoption
Governance refers to the principles, policies, and responsible development practices that
is necessary to maintain competitiveness.1 align AI tools and systems with ethical and human values. It establishes the frameworks,
rules, and standards that direct AI research, development, and application according to the
principles that organizations deem worthy. Governance mitigates the potential risks
associated with AI—bias, discrimination, and harm to individuals—through sound AI policy,
data governance, and well-trained and maintained datasets.2
The enterprise guide to AI governance: Three trust factors that can’t be ignored 3
Effective governance is key to building a foundation of trust. Monitoring how AI models are Figure 1
trained and managed helps organizations not only build better models but reassures GFioguvree r2nance can also be a catalyst for growth—for instance,
employees, customers, partners, and other stakeholders that the information and services bGyo fvaecrnilaitnactein cga nm aolsreo mbee aa ncaintaglfyuslt cfoorn gnreocwttiho—n sf owr iitnhs tcaunscteo, mers.
by facilitating more meaningful connections with customers.
they use are reliable.
Governance can also be a catalyst for growth— for instance, by facilitating more meaningful
connections with customers. It’s part of a management mindset that goes beyond risk
management and compliance to unlock opportunity. What’s troubling is that only 21% of
executives in our research say their organization’s maturity around governance is systemic
or innovative (see Figure 1).
From training and tuning to inference and outputs, risks can crop up at every phase of AI
development. In fact, MIT researchers recently compiled a list of over 750 AI risks to help However, only
identify gaps and uncertainties in how organizations perceive the AI risk landscape.3 74% 21%
of respondents believe believe that their
governance will have high organization’s maturity
impact in next 3 years as gen AI around governance is leading.
adoption barriers are removed.
The enterprise guide to AI governance: Three trust factors that can’t be ignored 4
Over 65% of data leaders at a recent Gartner conference highlighted data governance as Figure 2
Figure 1
their top focus in 2024.4 Executives across the C-suite admit that their organizations need to Why AI governance matters more than ever.
Why AI governance matters more than ever.
do better: 60% of CEOs say they’re looking into mandating additional AI policies to mitigate
risk. While 63% of CROs and CFOs say they are focused on regulatory and compliance risks,
only 29% say these risks have been sufficiently addressed5 (see Figure 2). Roughly 27% of
public companies cited AI regulation as a risk in recent filings with the Security and
Exchange Commission.6
of CROs and CFOs say they
So how do organizations move gen AI initiatives forward as fast as possible to capture 63%
are focused on regulatory
business value—while also constructing governance guardrails to keep gen AI on track? and compliance risks.
In this report, we will explore this central leadership challenge, and provide specific
recommendations for action, through the lens of three major trust factors.
of CEOs say they are looking
60%
into mandating additional AI
policies to mitigate risk.
However, only
29%
of CROs and CFOs say
these risks have been
The enterprise guide to AI governance: Three trust factors that can’t be ignored sufficiently addressed. 5
Trust factor 1 Trust factor 2 Trust factor 3
Trustworthy AI depends Accountability: Who is in Transparency: How do you Explainability: How do you
on effective governance. charge of AI governance? assess sources of data and explain the output
Incorporating these three trust what is shared about them? of AI systems and models?
Effective AI governance must be a funded
factors is the starting point for
mandate from senior leadership.
Diverse and multidisciplinary teams Deep collaboration between people and
building governance frameworks. Organization-wide adoption requires
should be deployed to assess data used to AI systems is built on transparency and
flexible governance frameworks to mitigate
build models, matching the broad range of explainability, adding the human touch to
risks and achieve business goals.
needs and expectations of AI users. This AI-informed decision-making. Sharing
will answer questions about how models the provenance of models is fundamental
are audited and how they perform to trust.
compared to humans.
Clear AI governance practices and policies
are at the core of addressing these trust
factors. Without governance, the adoption
of trustworthy and ethical AI systems can
be inhibited. At the same time, gen AI itself
can help improve governance—all across
the enterprise.
The enterprise guide to AI governance: Three trust factors that can’t be ignored 6
Three key AI governance-
related terms
Transparency, explainability, and provenance
Effective governance—delivered through corporate instructions, staff, Transparency is the ability to perceive how people can use and trust. Within its own
an AI system is designed and developed, governance framework, each organization
processes, and systems—helps assure that AI systems operate as
typically supported by the sharing of adopts an explainability approach to meet
an organization intends, while meeting stakeholder expectations
appropriate details about the AI system.7 its objectives.
and regulatory requirements. To enable AI users to direct, evaluate, To build a trustworthy AI model, algorithms
Provenance refers to the ability to explain
cannot be perceived as black boxes. AI
monitor, and take corrective action at all stages of the AI lifecycle,
and verify the origins of the data that trains
developers, users, and stakeholders must
governance relies on transparency, explainability, and provenance. AI models throughout their lifecycles.9 It is
understand the inner workings of AI to trust
vital for ensuring authentic data inputs and
its results.
for enhancing trust in AI-generated insights
Explainability, in the context of AI, is a and decisions. By recording metadata from
set of practices, tools and design the data’s source, provenance provides
principles that that makes AI decisions historical context and supports data
more comprehensible to humans.8 validation and auditing, leading to more
The more explainable an AI system is, the accurate and trustworthy AI outputs.
greater its ability to provide insights that
The enterprise guide to AI governance: Three trust factors that can’t be ignored 7
Trust factor 1 | Accountability
Who is in charge of
AI governance?
The first step is clear accountability. In our research, 60% of C-suite
executives say they have placed clearly defined gen AI champions
throughout their organization. And almost as many—59%—say they
have a direct report responsible for organization-wide AI integration.
What’s more, 80% of C-suite executives say they have a separate
risk function dedicated to using AI or gen AI. They want to be sure
that in developing and deploying AI, they are mitigating the risks of
unintended harm and unwanted biases.
The enterprise guide to AI governance: Three trust factors that can’t be ignored 8
Trust factor 1 | Accountability
Another IBM IBV survey of C-suite leaders, Supporting governance requires a retrofitted after deployment.13 Yet less
Spending on AI ethics has
creative executives, creative managers, and commitment of resources. Spending on AI sophisticated players and newer entrants
designers revealed that 47% of ethics has steadily increased from 2.9% of AI to AI struggle with the complex choices that steadily increased from
respondents have established a generative spending in 2022 to 4.6% in 2024. This share governance can raise. The solution is often
2.9%
AI ethics council to create and manage is expected to increase to 5.4% in 2025. flexible AI governance frameworks, which
in 2022
ethics policies and mitigate generative AI can help adapt to changing markets,
Our research indicates that more
risks.10 The goal of these councils is to mitigate risks, and encourage greater
technologically mature organizations tend
address the risk of “lawful but awful” AI.11 adoption to realize potential.14
to prioritize AI governance. For instance, to
The establishment of enterprise-wide
68% of CEOs in an IBM IBV survey say
governance frameworks helps streamline
governance for gen AI must be integrated
the process of detecting and managing
upfront in the design phase, rather than
technology ethics concerns in AI projects.12 4.6%
in 2024
This share is expected
to increase to
5.4%
in 2025
The enterprise guide to AI governance: Three trust factors that can’t be ignored 9
Action guide | Accountability
Build robust 1. Empower a senior-level executive to 3. Ensure that senior leadership aligns 5. Foster collaboration with stakeholders
lead AI and data governance initiatives. principles with practices. and ecosystem partners.
AI governance
Send the message that governance Align values related to the development Include stakeholders across the entire
is a senior management priority. and procurement of AI. Organizations organization—all working toward the
frameworks Championing AI and data governance achieve the outcomes they measure, same goals. Collaboration with
from an enterprise’s highest levels and aligning principles with practices governments, trade and industry
under an minimizes the risk of failure due supports measurement of progress associations, and other groups helps
to fragmented ownership and towards responsible AI adoption. establish AI governance guidelines,
fuzzy accountability. best practices, and regulations for
executive
4. Develop a cultural foundation for
responsible AI use. Ensure third-party
governance structures.
2. Prioritize and build on responsible AI software vendors and partners with
mandate.
development and deployment. embedded AI are subject to audits and
Without a strong cultural foundation, AI
other governance processes.
governance structures cannot gain
Give leaders who will be accountable for
traction. Healthy cultures have success
AI governance the authority to do the
measurements, incentives, messaging
work and provide their teams with the
and communications, diversity and
necessary resources to support this
inclusiveness, psychological safety,
mandate. Teams can also build on and
proactive employee training, and a
evolve governance frameworks already
holistic approach to AI literacy.
in place.
The enterprise guide to AI governance: Three trust factors that can’t be ignored 10
Case study | Data provenance standards
The Data & The Data & Trust Alliance (D&TA) was “The value of AI depends on the quality of data. To
established in 2020 by CEOs from leading
realize and trust that value, we need to understand
companies, based on a shared conviction
Trust Alliance
that the future of business will be powered
where our data comes from and if it can be used,
by the responsible use of data and AI. The
27 members of the Alliance–including legally. That’s why the members of the Data & Trust
Enhancing AI business
Deloitte, GM, IBM, Johnson & Johnson,
value and trust with data Mastercard, Meta, Nike and UPS– Alliance created a new business practice through
provenance standards15 represent 18 industries, employ over
cross-industry data provenance standards.”
four million people and earn $2 trillion in
annual revenue.
Saira Jesani
Executive Director, Data & Trust Alliance
The enterprise guide to AI governance: Three trust factors that can’t be ignored 11
Case study | Data provenance standards
The Alliance creates tools and practices to most essential metadata—required to IBM saw increases in both efficiency—time
enhance trust in data, models, and the understand more about a dataset’s origin, for clearance—and overall data quality,
processes around them. In 2023, the its method of creation and whether it can with a 58% reduction in data clearance
Alliance developed the first set of cross- be legally used—were selected. processing time for third-party data and
industry data provenance standards, a 62% reduction in data clearance
In early 2024, IBM tested the D&TA data
including 22 metadata fields that provide processing time for IBM owned or
provenance standards as part of a clearance
essential information about the origin of generated data. The D&TA standards were
process for datasets used to train
data and associated rights. a meaningful contributing factor to these
foundational models. IBM’s data governance
improvements. IBM is now adopting the
These standards were created with two program already included a data clearance
D&TA data provenance standards into
objectives in mind: business value and process that applied relevant controls,
its business data standards, where
implementation feasibility. By adopting documented lineage, and defined guidelines
appropriate, to further optimize enterprise
D&TA data provenance standards, for use and re-use. The challenge was a need
data governance.
businesses can better understand datasets to respond to an increasing volume of data
before purchase or use—and have a basis to clearance requests. The organization tested
decline data or request changes from third the standards to optimize the process for
parties. To encourage adoption, only the greater efficiency and accuracy.
12
Trust factor 2 | Transparency
How do you assess AI data
sources and what is shared
about them?
90% of the data available in the world was generated in the last two Before people can use AI, trust must be For transparency to be effective,
earned, and the most effective way to earn organizations must provide explainability—
years—just as gen AI went from curiosity to ubiquity. Approximately 400
user trust is through transparency. With the ability of an AI system to provide
million terabytes of data are created every day, with 150 zettabytes
respect to personal data, transparency insights that people can use to understand
estimated to be generated in 2024.16 But managing such huge amounts of is a key privacy principle. It requires the causes of the system’s predictions.
organizations to be open and forthcoming Clear explanations must be provided about
data presents huge challenges. Almost half of surveyed CEOs say they are
about their data processing practices. This accountability, data, models, algorithms,
concerned about accuracy and bias—an issue that could create as many
enables people to determine how they performance, audits, and related factors
problems as generative AI promises to solve. want their data used and shared. (see Figure 3). Otherwise, organizations
take on a tremendous exposure to risk.
The enterprise guide to AI governance: Three trust factors that can’t be ignored 13
Trust factor 2 | Transparency
Figure 3
Here are important questions for which
organizations must provide clear explanations.
Figure 3
Here are important questions for which organizations must provide clear explanations.
Transparency eliminates black box opacity To help ensure assumptions are not
and supports accurate and fair decision- overlooked, AI governance should
making. As artifacts of the human experience, include diverse, multidisciplinary teams Who is accountable Was the data gathered
virtually all data is biased. AI mirrors our to both build and govern these models. for models? with consent?
biases. The question is: which biases do not Outside experts in psychology,
reflect our values? If bias aligns with an anthropology, law, philosophy,
organization’s values, there must be linguistics, and other disciplines can also Do domain and
Does it represent all the experiential experts
transparency about why that dataset and help ensure that AI is used to augment communities served? agree that correct
approach were chosen over others. If they human intelligence in ways that align data was used?
don’t align, a different approach is needed. with human values. Governance teams
also need a psychological safety net How much better does What goes
when having challenging conversations the model perform into algorithm
compared to a human? recommendations?
about potential disparate impacts of an
AI model.
How is the model
How often is the
31 audited and what
model audited?
is it audited for?
The enterprise guide to AI governance: Three trust factors that can’t be ignored 14
Action Guide | Transparency
Assemble a 1. Establish a multidisciplinary AI 3. Ask questions and think beyond 4. Embrace ideas from outside
governance team. regulatory compliance. the organization.
Dream Team
To avoid blind spots, include a broad Best practices for governance go
Learn, follow, and look for opportunities
range of expertise from technical, ethical, deeper than rules and regulations
to participate in the development of
to build and social domains. This diversity can and can open doors to innovation.
intergovernmental and international
help identify gaps faster, help leverage Compliance starts with policies,
standards. AI principles promulgated by
effective AI existing governance mechanisms and procedures, and industry standards for
the OECD provide a useful starting-off
enable an organization to proactively AI. Building a framework for compliance
point.17 Additional national and global
head off unintended impacts. enables efficient incorporation of new
governance. standards are expected as gen AI
rules and regulations.
adoption grows.
2. Train everyone in transparency.
Give employees at all levels opportunities
to receive the training they need to build
or procure AI models responsibly in their
own domain, as well as awareness of
when to seek help when working outside
their domain, such as audits. Engage the
workforce by creating a culture that
celebrates openness and inclusion.
The enterprise guide to AI governance: Three trust factors that can’t be ignored 15
Case study | Transparency
Australia Post With annual revenues over $5.8 billion, Generative AI is a key part of the Post’s But mindful that while the public has
Australia Post provides postal services mission to boost customer service and concerns about AI, the Post is embracing
from 4,310 locations. As a government- efficiency. After testing and reviewing transparency, because gen AI is well
owned corporation, the Post must use thousands of customer calls and employee underway. The Post has already conducted
Delivering a more
transparency and rules to maintain keystrokes, generative AI is now routing a review of all its data and is now creating
efficient future18
customer trust. Much of the data handled customer queries and answering business- strict procedures around data governance.
is personal and sensitive. as-usual questions. The Post is working It is committed to not just aligning with
toward a goal where gen AI could handle regulatory frameworks but hardening its
between 40% and 60% of calls, delivering privacy and security protocols.
a better customer experience while
significantly reducing costs.
Australia Post is working toward a goal where gen AI
could handle between 40% and 60% of calls to improve
customer experience and reduce costs.
The enterprise guide to AI governance: Three trust factors that can’t be ignored 16
Trust factor 3 | Explainability
How do you explain the output
of AI systems and models?
As more organizations adopt AI, AI acceptance is at a crossroads.
While 35% of respondents to the 2024 Edelman Trust Barometer survey
say they accept this innovation, almost as many—30%—reject it.19
Demonstrating the trustworthiness of AI will be key to optimizing AI’s
impact. Trustworthy AI can also contribute to new ideas that separate
innovators from those doing the bare minimum.
The enterprise guide to AI governance: Three trust factors that can’t be ignored 17
Trust factor 3 | Explainability
A key element of trustworthy AI is not just limited to explaining how a gen AI Figure 4
provenance—the ability to explain and model renders outputs. In higher risk use MFigousret 4executives in our research say they recognize the
Most executives in our research say they recognize the importance of explainability.
verify the origins and history of data cases, it is appropriate to have every importance of explainability.
throughout its lifecycle. When training AI output provide an explanation of its data
models, provenance is essential for lineage informed by provenance, along
78%
ensuring that the data is authentic and with evidence.
trustworthy. Authentic data inputs into AI
Maintaining explainability. We
Most executives in our research say they
models enhances the trustworthiness of maintain robust documentation
AI-generated insights and decisions. recognize the importance of explainability; with explainability.
78% maintain robust documentation; 74%
For people to trust what goes into and what conduct ethical impact assessments;
comes out of AI models, explainability—the and 70% conduct user testing for risk 74%
ability to understand and trust AI outputs— assessment and mitigation (see Figure 4).
is informed by provenance. Explainability is Assessing ethical impact. We conduct ethical
impact assessments to evaluate the potential of
these initiatives on different stakeholders.
70%
Assessing and mitigating risks. We
conduct user testing for risk assessment
and mitigation.
The enterprise guide to AI governance: Three trust factors that can’t be ignored 18
Action Guide | Explainability
Keep humans 1. Design AI systems that facilitate 2. Prioritize AI output that is explainable 3. Incentivize employees to speak up and
human-AI collaboration and oversight. and auditable. speak out if AI output is confusing.
in the loop.
Create and scale repeatable patterns to Invest in applied training that improves Make sure those who build, design, and
ensure AI systems and their transparent AI literacy and provides clear guidance procure AI adopt a human-centric
metadata are accessible to the people on designing and developing human- instead of a data-centric approach and
using them, no matter what their level of centric systems. consider how outputs can be evaluated
technical understanding. after the fact. Provide appropriate
communication so employees feel
empowered and competent to ask about
potential disparate impacts related to
the AI models they work with.
The enterprise guide to AI governance: Three trust factors that can’t be ignored 19
Governance and
building trust
Can generative AI be trusted? Understanding the data used to train, 79% of global respondents say it is
Can trust guardrails balance tune, and make inferences from AI important for their CEOs to speak out
models is essential. What a company about the ethical use of technology.20
the power of gen AI?
does with AI is defined, in large part,
Ultimately, AI governance is about much
The answer is yes—but only by how it selects, governs, analyzes,
more than rules, restrictions, regulations,
and applies data across the enterprise.
if organizations approach AI
and requirements. It’s about a shared
Communicating that process
governance with commitment understanding of practices for effective
transparently is how trust is built and
and enthusiasm. collaboration that can reduce uncertainty
maintained over time.
and increase predictability—practices which
Governance needs to be embedded may actually accelerate development. When
at every phase of the generative AI sponsored and promoted at the leadership
lifecycle— not in functional silos but level, AI governance will no longer be seen
across the enterprise. It must be as just another IT issue, but as a core
championed by top leadership that strategy for value creation, growth,
provides strategic guidance, innovation, and developing the potential
recognition, and feedback. According of human-AI collaboration.
to the 2024 Edelman Trust Barometer,
The enterprise guide to AI governance: Three trust factors that can’t be ignored 20
Authors Contributors IBM Institute for Related reports
Business Value
Phaedra Boinodiris Sara Aboulhosn, Lee Cox, Rachna Handa, The CEO’s Guide to Generative AI: Manage
Global Leader for Trustworthy AI Christina Montgomery, Shyam Nagarajan, unpredictable risks
For two decades, the IBM Institute for
IBM Consulting Dasha Simons, Michael Tucker, and
Business Value has served as the thought
[email protected] Kush Varshney. https://www.ibm.com/thought-leadership/
leadership think tank for IBM. What inspires
https://www.linkedin.com/in/phaedra/ institute-business-value/en-us/report/
us is producing research-backed,
ceo-generative-ai/ceo-ai-risk-management
The right partner for
Brian Goehring technology-informed strategic insights that
Associate Partner and AI Research Lead a changing world help leaders make smarter business The CEO’s Guide to Generative AI:
Institute for Business Value decisions. From our unique position at the Responsible AI & ethics
IBM Consulting At IBM, we collaborate with our clients, intersection of business, technology, and
[email protected] https://www.ibm.com/thought-leadership/
bringing together business insight, advanced society, we survey, interview, and engage
https://www.linkedin.com/in/brian-c-
institute-business-value/en-us/report/
research, and technology to give them a with thousands of executives, consumers,
goehring-9b5a453/
ceo-generative-ai/responsible-ai-ethics
distinct advantage in today’s rapidly and experts each year, synthesizing their
Milena Pribic changing environment. perspectives into credible, inspiring, and
The ingenuity of generative AI
Design Principal, Ethical AI Practices actionable insights. To stay connected and
IBM Software informed, sign up to receive IBV’s email https://www.ibm.com/thought-leadership/
[email protected]
newsletter at ibm.com/ibv. You can also institute-business-value/en-us/report/
https://www.linkedin.com/in/milenapribic/
find us on LinkedIn at https://ibm.co/ scale-generative-ai
ibv-linkedin.
Catherine Quinlan
Vice President, AI Ethics
IBM Chief Privacy Office
[email protected]
https://www.linkedin.com/in/
catherinemquinlan/
The enterprise guide to AI governance: Three trust factors that can’t be ignored 21
Notes and sources 10. Disruption by design: Evolving experiences in the age of generative AI. IBM Institute for Business
Value. June 2024. https://www.ibm.com/thought-leadership/institute-business-value/en-us/report/
generative-ai-experience-design
11. Foody, Kathleen. “Explainer: Questioning blurs meaning of ‘lawful but awful’”. AP. April 7, 2021.
https://apnews.com/article/death-of-george-floyd-george-floyd-cba9d3991675231122e2b68fb
d5b4b00
12. Montgomery, Christina and Francesca Rossi. “A look into IBM’s AI ethics governance framework.”
1. Goehring, Brian, Manish Goyal, Ritika Gunnar, Anthony Marshall, and Aya Soffer. The ingenuity of
IBM Blog. December 4, 2023. https://www.ibm.com/
generative AI. IBM Institute of Business Value. June 2024. https://www.ibm.com/thought-
blog/a-look-into-ibms-ai-ethics-governance-framework/
leadership/institute-business-value/en-us/report/scale-generative-ai
13. 6 hard truths CEOs must face. IBM Institute for Business Value. May 2024. https://www.ibm.com/
2. Mucci, Tim and Stryker, Cole. “What is AI governance?” IBM Blog. November 28, 2023. https://
thought-leadership/institute-business-value/en-us/c-suite-study/ceo
www.ibm.com/topics/ai-governance
14. A Flexible Maturity Model for AI Governance Based on the NIST Risk Management Framework. IEEE
3. Constantino, Tor. “AI’s Risky Business, MIT Researchers Catalogue Over 750 AI Risks.” Forbes.
USA. July 2024. https://ieeeusa.org/product/a-flexible-maturity-model-for-ai-governance/
September 11, 2024. https://www.forbes.com/sites/torconstantino/2024/09/11/
ais-risky-business-mit-researchers-catalogue-over-750-ai-risks/ 15. The Data & Trust Alliance. Accessed September 28, 2024. https://dataandtrustalliance.org/about
4. “Data Governance is a Top Priority for 65% of Data Leaders-Insights From 600+ Data Leaders For 16. Duarte, Fabio. “Amount of data created daily (2024).” June 13, 2024. Exploding Topics. https://
2024.” Humans of data. March 28, 2024. https://humansofdata.atlan.com/2024/03/ explodingtopics.com/blog/data-generated-per-day
future-of-data-analytics-2024/ 17. OECD AI Principles overview. OECD.AI Policy Observatory. May 2019. https://oecd.ai/en/ai-principles
5. The CEO’s guide to generative AI: Risk management. IBM Institute for Business Value. August 2024. 18. Internal IBM case study.
https://www.ibm.com/thought-leadership/institute-business-value/en-us/report/
19. Edelman, Margot. “Why the human touch is needed to harness AI tools for communications.” World
ceo-generative-ai/ceo-ai-risk-management
Economic Forum. June 18, 2024. https://www.weforum.org/agenda/2024/06/
6. Lin, Belle. “AI Regulation Is Coming. Fortune 500 Companies Are Bracing for Impact.” The Wall human-touch-harness-ai-tools-communications/
Street Journal. August 27, 2024. https://www.wsj.com/articles/
20. Ibid.
ai-regulation-is-coming-fortune-500-companies-are-bracing-for-impact-94bba201
7. IBM Design for AI guidelines and definitions.
8. What is explainable AI? IBM. https://www.ibm.com/topics/explainable-ai
9. What is data provenance? IBM. https://www.ibm.com/think/topics/data-provenance
The enterprise guide to AI governance: Three trust factors that can’t be ignored 22
© Copyright IBM Corporation 2024
IBM Corporation
New Orchard Road
Armonk, NY 10504
Produced in the United States of America | October 2024
IBM, the IBM logo, ibm.com and Watson are trademarks of International Business Machines Corp.,
registered in many jurisdictions worldwide. Other product and service names might be trademarks
of IBM or other companies. A current list of IBM trademarks is available on the web at “Copyright
and trademark information” at: ibm.com/legal/copytrade.shtml.
This document is current as of the initial date of publication and may be changed by IBM at any time.
Not all offerings are available in every country in which IBM operates.
THE INFORMATION IN THIS DOCUMENT IS PROVIDED “AS IS” WITHOUT ANY WARRANTY, EXPRESS OR
IMPLIED, INCLUDING WITHOUT ANY WARRANTIES OF MERCHANTABILITY, FITNESS FOR A
PARTICULAR PURPOSE AND ANY WARRANTY OR CONDITION OF NON-INFRINGEMENT. IBM products
are warranted according to the terms and conditions of the agreements under which they are provided.
This report is intended for general guidance only. It is not intended to be a substitute for detailed research
or the exercise of professional judgment. IBM shall not be responsible for any loss whatsoever sustained
by any organization or person who relies on this publication.
The data used in this report may be derived from third-party sources and IBM does not independently
verify, validate or audit such data. The results from the use of such data are provided on an “as is” basis
and IBM makes no representations or warranties, express or implied.
This document is printed on chlorine-free 100% post-consumer paper meeting Forest Stewardship
Council (FSC) responsible forestry certification. The energy used to manufacture this paper was
generated through renewable green energy. Please recycle.
The enterprise guide to AI gover |
140 | ibm | the-intuitive-supply-chain-report.pdf | IBM Institute for Business Value | Research Brief
The intuitive
supply chain
Predict disruption, deliver growth
Key takeaways
Generative AI can preempt supply Generative AI has put supply chains in flux. 64% of Chief Supply
Chain Officers say gen AI is completely transforming workflows.
chain disruption and unleash
Supply chain teams must work differently. 60% of operations and
growth opportunities. automation executives say AI assistants will handle most traditional
and transactional processes by 2025.
More decisions will be automated. Operations and automation
executives say generative AI will increase the volume of decision-
making by digital assistants by 21% in the next two years.
Predictions will improve, igniting sustainable innovation. 76% of
supply chain and operations leaders say gen AI will help innovate their
product design and make product lifecycles more sustainable.
The intuitive supply chain: Predict disruption, deliver growth 2
Introduction
Make agility your supply
chain superpower
Would a peek at next week’s headlines change
your supply chain strategy today?
The intuitive supply chain: Predict disruption, deliver growth 3
Supply chain certainty is an elusive target. The combined power of generative AI and
With so many fault lines stretching across cloud computing could make that possible.
the business landscape, it seems By harnessing the potential of machine
impossible to accurately predict what will learning, automation, and advanced
happen tomorrow. Supply chain leaders analytics in a hybrid cloud environment,
must often adopt a siege mentality, looking organizations can gain a sixth sense,
for ways to limit their losses as plan B anticipating everything from demand
quickly gives way to plans C, D, and E. fluctuations to sourcing delays. With this
foresight, they can reinvent their supply
But what if you could spend this time
chain strategies, shifting from a reactive
spurring growth? What if you could predict
to a proactive stance.
the future accurately enough to give your
business a competitive edge?
17% 72%
Leaders in gen AI adoption and data-led
innovation—those who view gen AI capabilities
as the primary driver of their automation
report higher annual revenue report greater annual
investments—are reaping outsized rewards. growth than the competition net profits
The intuitive supply chain: Predict disruption, deliver growth 4
Already, leaders in gen AI adoption and That’s a big problem for supply chain executives, and automation executives employees. Part two explains how
data-led innovation—those who view gen AI leaders, who know they need to invest from organizations that are currently accelerating supply chain intelligence can
capabilities as the primary driver of their in next-gen tech today to make their implementing AI-enabled automation. help companies leverage real-time data
automation investments—are reaping operations more agile and resilient for We discovered that these leaders are faster and more effectively than ever
outsized rewards. They report 72% greater an uncertain future—from dynamically focused on creating what we call “the before. And in part three, we’ll explore how
annual net profits and 17% higher annual rerouting shipments and adjusting intuitive supply chain”—agile, adaptive, and gen AI-enabled digital twins, or virtual
revenue growth than the competition. And all production schedules in real time to perpetually prepared, safeguarding brand models, can help organizations improve
the supply chain leaders we surveyed expect identifying bottlenecks and risks before reputation, customer satisfaction, and the their position in the competitive landscape,
their revenue growth from AI-enabled they materialize. bottom line. as well as in the eyes of customers. We
operations to more than double over the next conclude with an action guide that outlines
How can gen AI solve these persistent In this paper, we’ll lay out the steps
three years.1 how to plan, prioritize, and perform to
supply chain problems? To find out, the IBM organizations are taking to get there. In part
make every move count.
Looking at these numbers, it’s no surprise that IBV, in partnership with Oxford Economics, one, we’ll explore the role of AI assistants,
72% of the top-performing CEOs we surveyed surveyed more than 2,000 global Chief which are quickly becoming less like
for the IBM Institute for Business Value (IBM Supply Chain Officers (CSCOs), operations chatbots and more like full-time
IBV) 2024 CEO Study say competitive
advantage now depends on who has the most
advanced gen AI. But the high-speed race to
meet short-term goals is hindering their
progress. Overall, global CEOs agree that a
focus on short-term performance is their top
barrier to innovation—and 66% say their
Supply chain leaders need to invest in
organization is currently meeting short-term
targets by reallocating resources from next-gen tech today to make their operations
longer-term efforts.2
agile and resilient for an uncertain future.
The intuitive supply chain: Predict disruption, deliver growth 5
Part one
Lean into the power
of decision support
Employees paired with AI assistants will deliver
more business value than either could alone.
The intuitive supply chain: Predict disruption, deliver growth 6
Today’s supply chain teams are drowning in a sea of disconnected data.
They increasingly have access to the long-awaited real-time information
they need to make smarter, faster, decisions—but there’s so much to sift
through that many opportunities go unnoticed until it’s too late. to ask for the information they need—and find out where it came
from—with a few simple prompts.
Gen AI-powered digital assistants are changing all that. With their
ability to analyze vast stores of data almost instantaneously, they can For example, AI assistants can analyze which supplier is contributing
bubble up critical insights for supply chain teams to skim from the the most to delays and identify issues causing disruption, such as
surface. Plus, their natural language skills make it easy for employees weather, financial obstacles, or transportation bottlenecks. Then,
AI-fueled predictive models can outline how the situation is most likely
to evolve, allowing AI assistants to offer targeted recommendations
that help supply chain teams prepare for what’s next.
Already, 60% of executives say AI assistants will handle most
traditional and transactional processes by 2025.3 And 90% say their
60% organization’s supply chain workflows will incorporate intelligent
automation and AI assistants by 2026.4
of executives say AI assistants
will handle most traditional and
transactional processes by 2025.
And
90%
say their organization’s supply chain
workflows will incorporate intelligent
automation and AI assistants by 2026.
The intuitive supply chain: Predict disruption, deliver growth 7
When employees use gen AI assistants to It’s not just about explaining how materials By leveraging AI assistants, CSCOs can
quickly query their supply chain platform will get from point A to point B. It’s also aggregate and distill intel, bringing insight
for credible data, rather than manually measuring the supply chain cost of every to the boardroom quickly and confidently
searching multiple systems, they can business decision—and making sure those and making sure supply chain implications
manage change faster—and pivot more costs are considered from the start. Beyond continue to inform strategies as they
precisely. Instead of using the dedicated the sales a new SKU will drive, product evolve. As decisions are made and then
procurement solution to change purchase development strategies should account for tested in the market, AI assistants can
order delivery dates, for instance, the total cost of ownership, forecasting the accelerate the feedback loop, giving
employees can simply ask their assistant cost of delivering a new item in conjunction executives the real-word, real-time data
to make the change for them. with the losses that come from holding on they need to see if their strategies are
to products that don’t sell. delivering the desired results—and change
But that’s only the beginning. Supply chain
tactics quickly if they aren’t.
teams aided by AI assistants are cultivating Then there’s the sustainability dimension.
a new human-technology dynamic that will As both consumers and regulators demand
touch virtually every point of the supply more comprehensive reporting on
chain, from planning to sourcing to environmental impact, supply chain leaders
manufacturing to distribution. In fact, must be able to track sustainability metrics
64% of CSCOs say gen AI is completely all the way to the last mile—and do the hard
transforming their supply chain workflows. work of designing more eco-friendly product
And CSCOs and automation executives say lifecycles. This is another place where gen
gen AI will increase the volume of decision- AI can help, with 76% of supply chain and
making by digital assistants by 21% in the operations leaders agreeing that it will help
With gen AI assistants, employees can manage
next two years. innovate their product design and make
product lifecycles more sustainable.
change faster—and pivot more precisely.
The intuitive supply chain: Predict disruption, deliver growth 8
Case study
Building an intelligent supply chain
using a supply chain AI assistant
IBM employs supply chain staff in 40 which impeded collaboration and At a high level, the IBM supply chain The system uses IBM’s AI technology to
countries and makes hundreds of thousands real-time data transparency. digital transformation revolves around enable natural language queries and
of customer deliveries and service calls in building sense-and-respond capabilities. responses, which accelerates the speed of
IBM supply chain management set out
over 170 nations. IBM also collaborates with This was accomplished by democratizing decision-making and offers more options
a bold transformation vision to build a
hundreds of suppliers across its multitiered data—automating and augmenting to correct issues. Users can ask, in natural
cognitive, intelligent supply chain more
global network to build highly configurable decisions by combining a cognitive control language, about part shortages, order
than a decade ago. The aim was to have
and customized products to customer tower, a cognitive advisor, demand-supply impacts, and potential trade-offs. To date,
an agile supply chain that extensively
specifications. Historically, the IBM supply planning, and risk-resilience solutions. IBM has saved $388 million related to
uses data and AI to lower costs, exceed
chain ran on legacy systems spread Now, the cognitive control tower has reduced inventory costs, optimized
customer expectations, ruthlessly
across different organizational silos, evolved into an enhanced generative AI shipping costs, faster decision-making,
eliminate or automate non-value-add
making information-sharing slow and intelligent layer using a supply chain and time savings (days to hours to minutes
work, and exponentially improve the
incomplete. Employees also performed digital assistant. to seconds).
experience of supply chain colleagues.5
much of their work on spreadsheets,
The intuitive supply chain: Predict disruption, deliver growth 9
Part two
Accelerating supply
chain intelligence
If your data could talk, what would it say?
Supply chain teams are about to find out.
The intuitive supply chain: Predict disruption, deliver growth 10
Whether disruption is caused by In fact, the executives we surveyed
geopolitical conflict, climate catastrophes, anticipate operational performance, 73%
or increasing complexity, supply chain enterprise agility, and strategic advantage
leaders will be judged by their ability to find to be the top three benefits of using gen AI
effective workarounds. And they’re looking investments in their supply chain. And 73%
to gen AI to make their supply chain more say gen AI is already accelerating their
agile, adaptive, and future-proofed. high-impact automation initiatives. of executives say gen AI
is already accelerating
their high-impact
automation initiatives.
The intuitive supply chain: Predict disruption, deliver growth 11
Perspective
The key is to make the entire ecosystem The convergence of gen AI and cloud-based Future-proof your supply chain with
more responsive. By allowing gen AI solutions has also enabled autonomous
cloud-enabled innovation
assistants to interact directly with the automation (see “Future-proof your supply
intelligent layer of the supply chain system— chain with cloud-enabled innovation”).
With the combined power of cloud computing and generative AI, companies can
the cognitive core that pulls insights from In addition to automating workflows,
accelerate supply chain innovation and improve business outcomes to a degree that
vast stores of data—internal and external gen AI assistants can automate the process
wasn’t previously possible.
teams can collaborate more seamlessly. of workflow reinvention. They can learn from
supply chain metrics and transaction history, Deploying gen AI on the cloud lets companies train and deploy models faster and
The goal is for AI assistants to continually
make proactive recommendations, and even at scale, without the need for expensive hardware or infrastructure. It lets multiple
communicate the intelligent layer’s findings
repurpose or redefine new workflows based teams collaborate on the development of gen AI models, moving them between
to the appropriate part of the supply chain
on what they’ve learned. different cloud environments and integrating them with other cloud-based services
team, along with recommended actions.
and applications seamlessly.
While the enterprise resource planning (ERP) This helps streamline workflows to make
system remains the system of record and them more efficient, cost-effective, and Then, of course, there’s cost to consider. With pay-as-you-go pricing, cloud
core transaction engine, supply chain teams environmentally responsible. In fact, 63% infrastructure can ease capital expenditure constraints, allowing companies to
no longer need to interact with it directly. of supply chain and operations leaders say focus on innovation, rather than the financial implications of investing in new tech.
And that goes for other specialized supply integrating sustainability and circularity into When applied strategically, this tech combo can improve efficiency, reduce costs,
chain apps, from procurement to warehouse workflows is a key reason their organization and increase agility. Here are a few ways your supply chain can benefit from cloud-
management to transportation logistics, is investing in automation. enabled innovation powered by gen AI:
as well. This approach lets employees drill
deeper, allowing for real-time analysis and – Forecast future demand. Optimize inventory levels, reduce stockouts or over-
optimization each step of the way. stocking, and improve cash flow.
– Optimize delivery routes. Reduce fuel consumption, lower emissions, provide
dynamic distribution, and improve delivery times.
– Manage supply chain risk. Predict the likelihood of disruption and recommend
proactive mitigation measures.
– Increase supply chain visibility. Identify bottlenecks and recommend corrective
actions teams can take to keep operations from being disrupted.
The intuitive supply chain: Predict disruption, deliver growth 12
Case study
Achieve end-to-end visibility
with AWS Supply Chain
Supply chains are vast, interconnected networks. The multitude of The cloud-based AWS Supply Chain Address data fragmentation
business application directly addresses
participants, disparate systems, and lack of seamless data sharing A supply chain data lake harmonizes
these challenges. By harmonizing disparate
disparate data into a flexible, scalable
make it difficult to accurately forecast future demand, track inventory
data sources into a unified supply chain
canonical data model that aggregates and
levels, and align supply. The fragmentation of data hinders supply data lake, it lays the foundation for
associates supply chain information into
improved end-to-end visibility, forecasting
chain planners’ ability to understand fluctuations, predict future needs a unified data asset. By incorporating a
accuracy, inventory optimization, and
generative AI-powered data onboarding
precisely, and position optimal inventory where it’s needed most. overall supply chain resilience.6 Here are a
agent, companies can also automate data
few of the key business benefits of moving
transformation from any native format into
to this type of cloud-based solution:
the data lake’s canonical model. Customers
can seamlessly extract and upload raw
data, with the agent leveraging large
language models for automated data
mapping through a guided, module-driven
user interface experience.
The intuitive supply chain: Predict disruption, deliver growth 13
Case study (continued)
Increase forecast accuracy Improve supply chain visibility Improving supplier visibility Simplify sustainability
and collaboration compliance processes
Machine learning-powered forecasting The AWS business application can examine
capabilities can help organizations improve warehouses, distribution centers, and stores The AWS application analyzes supplier lead Cloud-based sustainability features create
forecast accuracy and reduce excess in detail, showing on-hand, in-transit, and times, makes future projections compared a more secure and efficient way to obtain
inventory levels. Machine learning at-risk inventory levels. It then uses machine to orders and forecasts, then identifies mandatory documents and datasets from
algorithms can incorporate variables such learning algorithms to automatically issues. It displays all connected trading your supplier network. You can request,
as seasonality, product characteristics, generate, score, and rank multiple inventory partners, enabling supply chain leaders to collect, and export artifacts, such as
vendor characteristics, and destination- rebalancing recommendations to mitigate view and collaborate across multiple tiers. product lifecycle assessments, certificates
origin sites, along with historical order risks. Gaining visibility into network-wide Built-in chat and messaging capabilities on product safety, or reports on hazardous
history, to train the model. inventory levels, movement patterns, and also facilitate seamless communication substances used at any point in the supply
potential risks empowers organizations to and data sharing. chain. Amazon’s Global Trade and Product
optimize inventory positioning and mitigate Compliance (GTPC) team used the AWS
imbalances, overstocks, and stockouts. application’s sustainability features to
transform their compliance data
management process and now expect to
save approximately 3,000 operational
hours per year.
The intuitive supply chain: Predict disruption, deliver growth 14
Part three
Visualize
the future
By using supply chain data to
fuel gen AI-powered virtual models,
companies can unlock a new level of
operational efficiency and resilience.
The intuitive supply chain: Predict disruption, deliver growth 15
Supply chain leaders have long imagined
a future where real-time data flows
seamlessly between IT and operational
By 2026,
technology (OT) systems, enabling a more
77%
agile approach that reacts to constant
change. And their dream is finally
becoming reality.
Think of a manufacturing facility, where
of executives believe gen AI will
operations teams already use AI sensors
enable connected assets to
to detect changes in vibration patterns, make autonomous decisions.
temperatures, power consumption, and
even sound patterns. While traditional AI
can alert teams to signals as they appear—
and even predict when breakdowns are
about to occur—employees must manage 76%
necessary adjustments or repairs based
on this information.
With generative AI, that’s no longer the
case. When paired with vision sensors, of executives say they expect to
gen AI lets connected machines use gen AI to derive differentiated
outcomes from connected assets
self-predict and self-adjust in a harmonious
in the next two years.
fashion, unlocking unprecedented levels of
productivity and efficiency.
The intuitive supply chain: Predict disruption, deliver growth 16
In fact, by 2026, 77% of executives expect It works like this: First, data from drones, support “what-if” risk analysis by predicting With the right perspective, supply chain
gen AI will enable connected assets to robots, cameras, and other connected potential problems—from raw material leaders can look beyond productivity plays
make autonomous decisions. And when assets flow into a unified platform with shortages to multiple supplier plant closings to pull the levers that drive growth. By using
complex asset ecosystems work in a geospatial layer, an information layer, simultaneously—and recommending gen AI to orchestrate multiple data sources,
harmony, they can help businesses achieve and an orchestration layer. Time-lapsed respective contingency plans. systems, and tools, they can inspire
results that weren’t previously possible. visualizations then let supply chain teams innovation across the ecosystem—and
These simulations can also inform product
Executives recognize this potential, with see how specific changes have impacted inform the strategic decisions that set their
development by helping teams identify
76% saying they expect to use gen AI to the ecosystem in the past—and make organization apart.
where waste and inefficiencies can be
derive differentiated outcomes from real-time decisions as situations unfold
removed from the process. This is a key
connected assets in the next two years.7 in the present.
concern for executives, who say visibility
But boosting efficiency is just the first step. Gen AI-enabled virtual models can then of full product lifecycle management and
Businesses can derive much deeper value help teams simulate how future events environmentally sustainable products and
from interconnected data when they use it could affect supply chain operations. services are two of their top automation
to visualize the end-to-end supply chain— They use real-world data and algorithmic priorities for their operations functions over
and simulate how disruption could impact techniques to visualize how the dominos the next three years.
operations each step of the way. will fall in response to different disruptions
to help teams plan accordingly. They
Look beyond productivity plays to pull the
levers that drive growth.
The intuitive supply chain: Predict disruption, deliver growth 17
Case study
Improving pharma supply chain
visibility for patient safety8
Amid the increasing proliferation of counterfeit, falsified, or Seeking safety through transparency players in the prescription drug supply chain.
By connecting through these APIs, Pulse
substandard prescription medications, the US government passed Working with IBM Consulting and AWS,
users can search for trading partners, verify
NABP built a new digital platform called
the Drug Supply Chain Security Act (DSCSA) with the aim of protecting
trading partner status, exchange digital
Pulse that lets its member users track and
patients. It’s rooted in the idea that transparency—the ability to credentials, and perform electronic tracing.
share each prescription drug’s ownership
accurately trace prescription meds throughout the pharmaceutical transaction records, providing increased
The platform enables visibility and
supply chain visibility.
supply chain—is essential to preserving its integrity. collaboration, eliminates tedious
administrative work, and, most importantly,
One key design aspect of the platform—
Just as important is the idea that all the major players in the pharmaceutical ecosystem— creates a more secure supply chain to
which runs on the AWS cloud—is the
manufacturers, wholesalers, dispensaries, and regulators—need a way to share information protect patients.
integration of APIs from providers of the
collaboratively to make it happen. Prompted by the challenge of multiple industry segments
“point” tracking solutions used by most
needing to cooperate to address DSCSA, the National Association of Boards of Pharmacy
(NABP) sought to create a digital platform that would bridge the interoperability gaps
between systems, making compliance with DSCSA faster and easier.
The intuitive supply chain: Predict disruption, deliver growth 18
Action Guide
Make every move count
In the complex game of supply chain chess, executives must always
think several steps ahead. Modernizing supply chains isn’t just about
adopting new technologies or processes—it’s about embracing a new
way of thinking, one that’s rooted in scientific inquiry, experimentation,
and a relentless pursuit of progress.
By applying the scientific method at scale, enterprises can tap into the vast potential of data
and gen AI to drive critical improvements in business strategy, product development, and
global supply chain operations. In fact, 62% of CSCOs say gen AI will accelerate the pace
of discovery, leading to new sources of product and service innovation.9
With the promise of discovery as their guiding light, companies can unlock the full potential
of their supply chains, power ecosystem partnerships, and drive sustainable profitability and
growth. Here’s what leaders across the supply chain ecosystem should do to predict and plan
for endless disruption—and profit from the opportunities volatility can create.
The intuitive supply chain: Predict disruption, deliver growth 19
Action Guide
1. Plan Identify benefits you want to deliver. management to improve decision-making Understand skills requirements
efficiency and speed-to-action. Invest to and gaps.
Investigate the key drop-out points
bring the vision to life and facilitate a
between analysis and action, identifying Create user personas across the range of
seamless and fulfilling experience across
how improvements could flow through into supply chain workflows. Outline how digital
the entire supply chain.
financial and operational performance. assistants will help create new workflows
Outline the productivity KPIs that will be Know the specific functionality and and enhance existing ones. Identify the
targeted for improvement and define systems architecture you need. gaps in skills between these personas and
success criteria. the current state, then define training and
Identify the solutions that will provide
upskilling plans.
Define your employee every feature. Then use an orchestration
experience vision. engine as a process conductor, issuing Keep your eyes on the prize.
precise commands to multiple agents
Provide easy access to relevant AI Align supply chain innovation to your
based on user prompts. Leverage
analytics, recommendations based on role, market offering and the capabilities needed
synthesized data from the integration layer
and intelligent transactional workflows in to deliver it. Prioritize these areas and be
to create dynamic, intelligent workflows
the employee portal. Find ways to integrate confident in delivering them.
that deliver the desired outcomes.
supply chain processes into the employee
experience framework, such as
streamlining logistics and inventory
The intuitive supply chain: Predict disruption, deliver growth 20
Action Guide
2. Prioritize Define supply chain workflows Don’t try to cut your way to growth. Define rules of engagement.
that have the greatest potential
Make the investments needed to Be clear about who is accountable and
for automation.
fundamentally transform ways of working. responsible for specific workflows—and
Map the key points across the workflow Focus spending on the areas that can make who gets a say. Set ground rules for using
that cause rework and manual analysis. your supply chain more agile and resilient. digital assistants and make sure everyone
Be honest about the true nature of your knows how they’re expected to evolve.
Prioritize getting to scale.
processes, not the idealized version that
may be documented somewhere. Invest in initiatives that can quickly
transition from pilot to deployment at scale.
Stop looking for a silver bullet.
Use success in specific areas to build
Be honest about where investment is momentum for the wider transformation.
needed within your current technology
landscape. Set specific timelines for
upgrades or the deployment of new
solutions. Don’t let time and effort that
have been invested in previous solutions
become an anchor that prevents you
from achieving future success.
The intuitive supply chain: Predict disruption, deliver growth 21
Action Guide
3. Perform Feed generative AI data that Review and align to Keep score.
supports supply chain productivity. changing conditions.
Track benefits as they’re delivered to build
Map the full range of data initiatives needed Cultivate a supply chain that can sway with momentum and confidence in new
to connect people and technology. Upskill the winds of change to deliver a competitive technologies. Demonstrate ROI to secure
employees and train tools to speed advantage. Adopt a technology architecture continued investment. Make data-driven
decisions. Identify the key touchpoints to that allows new capabilities to be plugged decisions that can fuel growth and
use gen AI to boost productivity. in without disrupting the user experience. performance improvements.
Put trust in data.
Don’t let people tinker with the workflow
outputs from the system. Where processes
are automated and tested, let the system
run and do its job. Don’t allow competing
forms of analysis designed to suit individual
agendas interfere. Instead, encourage
employees to engage in advanced analysis,
using their assistants to innovate and
address the complexities of interconnected
operations and systems.
The intuitive supply chain: Predict disruption, deliver growth 22
Authors Research methodology products, electronics, telecommunications, IBM Institute for
government, healthcare/life sciences,
Business Value
Amar Sanghera The IBM Institute for Business Value (IBM consumer products, retail, and
AWS Supply Chain Solutions Global Leader, IBV), in conjunction with Oxford Economics, transportation/logistics, each comprising
For two decades, the IBM Institute for
Digital Supply Chains Go-to-Market Strategy interviewed and surveyed more than 2,000 5% to 15% of our total respondent sample.
Business Value has served as the thought
executives with equivalent roles and titles, The size of organizations surveyed, in terms
Michael Mowat including Chief Supply Chain Officer (CSCO), of revenue, ranged from $500 million to leadership think tank for IBM. What inspires
Supply Chain Strategy and Operations Chief Operations Officer (COO), Chief $500 billion, with a mean of $26 billion. us is producing research-backed,
Leader, Finance and Supply Chain Automation Officer (CAO), Chief technology-informed strategic insights that
Transformation, IBM Consulting Information Officer (CIO), and Chief The IBM IBV ran a series of contrast help leaders make smarter business
Financial Officer (CFO). analyses, including pairwise comparisons, decisions. From our unique position at the
Karen Butner highlighting results and differences as
intersection of business, technology, and
Global Research Leader, AI and Automation; In 2024, CSCOs, COOs, and automation shown in this report. Statistical significance society, we survey, interview, and engage
Supply Chain Operations, IBM Institute for executives were also polled about their for all pairwise comparison contrasts was with thousands of executives, consumers,
Business Value, IBM Consulting investments, priorities, and use cases to set at the (p = .05) level, meaning there is
and experts each year, synthesizing their
assess the current impact of generative AI only a 5% chance that the observed
perspectives into credible, inspiring, and
Contributors initiatives, as well as the results they expect differences or relationships between the
actionable insights. To stay connected and
to see in the next two to three years. The groups are due to random variation.
informed, sign up to receive IBV’s email
goal of these surveys was to understand
IBM Consulting The right partner for newsletter at ibm.com/ibv. You can also
how global executives view the impact of
Chris Moose, Lead Client Partner NABP, find us on LinkedIn at https://ibm.co/
gen AI on their organizations’ performance a changing world
Public Sector ibv-linkedin.
and competitive advantage across the
Jonathan Wright, General Manager,
NCE Europe supply chain. At IBM, we collaborate with our clients,
bringing tog |
141 | ibm | the-ingenuity-of-generative-ai.pdf | IBM Institute for Business Value | Research Insights
The ingenuity of
generative AI
Unlock productivity and
innovation at scale
How IBM can help
Clients can realize the potential of AI, analytics, and data using
IBM’s deep industry, functional, and technical expertise;
enterprise-grade technology solutions; and science-based
research innovations.
For more information about AI services from IBM Consulting,
visit ibm.com/services/artificial-intelligence
For more information about AI solutions from IBM Software,
visit ibm.com/watson
For more information about AI innovations from IBM Research,
visit research.ibm.com/artificial-intelligence
2
Key
takeaways
Business leaders must Generative AI investment is surging.
translate experimentation
Spend increased more than 10 times in 12 months,
into enterprise-grade while IT spend grew at only half the rate of inflation.1
investments that deliver
Financial returns from AI have
value at scale.
solidly surpassed the cost of capital.
Average AI ROI hit 13% in 2022—and early generative
AI wins (led by successful pilots) boosted it to 31%
in 2023.
Early generative AI experiments are
gravitating toward low-risk, noncore
use cases.
But organizations can deliver more value by focusing on
business areas that are more closely related to their
competitive advantage.
The biggest gains may come from
stepping into the unknown.
Over the next three years, more than half of executives
expect generative AI to enable types of work that
weren’t previously possible.
1
“There’s no safe space in the corporate world where
you can just hang out and enjoy your winnings
from the past. You’ve got to always be driving
forward to the next horizon.”
Bill Anderson, CEO, Bayer AG
From media sensation to
market-ready solution
Generative AI has seemed almost too good to be true. It cuts coding
time from days to minutes, personalizes products down to the tiniest
detail, and spots security vulnerabilities almost as soon as they appear.
And it’s helped skyrocket AI ROI from 13% to 31% since 2022.
While this largely reflects the success of pilots, sandbox experimentation, and other
small-scale investments, these early results have business leaders rethinking what’s
possible. Our latest proprietary survey of 5,000 executives across 24 countries and
25 industries reveals that most executives are more optimistic about the generative
AI opportunity than they were last year. More than three in four (77%) say generative
AI is market ready, up from just 36% in 2023, and nearly two-thirds (62%) now say
generative AI is more reality than hype (see Figure 1).
More than three-quarters of executives say they need to adopt generative AI quickly
to keep up with competitors. And 72% of the highest performing CEOs say
competitive advantage depends on who has the most advanced generative AI,
according to the IBM Institute for Business Value (IBM IBV) 2024 CEO study.2
Already, business leaders have begun to discover how generative AI boosts the
bottom line. Operating profit gains directly attributable to AI doubled to nearly 5%
from 2022 to 2023—and executives expect that figure to hit 10% by 2025. And
embedded generative AI in existing enterprise software workflows also promises
to deliver more sustainable ROI, according to forthcoming IBM IBV research.3
Still, despite these early signals, some analysts are skeptical. They anticipate that
this hype-driven adoption spike will be followed by a “trough of disillusionment,”
where organizations back away from the complexity involved with deploying
generative AI in core business functions.4 And in some instances, it’s true. One in
three companies pause an AI use case after the pilot phase—but two in three don’t.
2
One in three companies pause an
AI use case after the pilot phase—
but two in three don’t.
In this setting, how can business leaders best translate successful experimentation
into enterprise-grade investments that deliver value at scale? This paper offers
a roadmap to help companies answer this question, accompanied by case studies
illustrating effectiveness in action. First, we outline where generative AI is currently
delivering the highest ROI. Then we explore how executives can capitalize on its
long-term potential and overcome key challenges, from organization structure
to security. Finally, we offer an action guide on how to transform business with
generative AI—regardless of where you are on your AI journey.
FIGURE 1
From skepticism
to confidence
Executives see the true
2023 2024
potential of generative AI
taking shape
We need to adopt generative 38% 77%
AI quickly to keep up with
competitors
Generative AI is market ready 77%
36%
Generative AI is more reality 33% 62%
than hype
3
Case study
Bayer AG thinks
big along the AI
continuum5
Bayer AG chief executive Bill Anderson has an expansive vision for the future of
generative AI: “I think some of the biggest applications we’ll use it for are related
to how are we going to feed two billion more people in the world in the next
20 years, with less land available, less water, and a need to use less chemicals.”
Anderson’s resume—he has an advanced degree in chemical engineering from
MIT and joined Bayer after a stint as CEO at Roche Pharmaceuticals—suggests
a disciplined, evidence-based approach to big predictions. His confidence
in generative AI’s eventual impact is grounded in an understanding of its place
on the continuum of technologies, such as artificial intelligence and machine
learning, that have been remaking his company and industry for some time.
“It’s just starting, but it’s not up for debate,” he says of the fast-blooming new
generation of applications. “We’re definitely moving out of the realm of theory
into application.”
Generative AI at work
The first big win for generative AI at Bayer is coming in enhanced productivity,
a process that is already underway. “It’s replacing a lot of manual labor already,
and we’re just getting started,” says Anderson. Collecting, checking, and crunching
data to better understand patient populations, for example, can yield meaningful
if incremental benefits in terms of testing site and participant selection.
None of this is easy. Counterfeit and simulated products, for example, are a major
risk, with generative AI giving criminals the ability to work fast while evading
security measures. Deepfakes and false reporting are threats, as well.
But Anderson remains convinced of the potential of generative AI to accelerate
drug discovery. In two or three years, he says, a new cancer drug will be in stage
three clinical trials because of work being done now with generative AI. “That’s
really fast,” he says.
44
Bayer AG thinks big along the
AI continuum (continued)
Seeding the future
Over time, Anderson sees generative AI helping Bayer’s €25 billion crop science
division address the tough challenges of crop protection in a time of climate change.
Developing a new insecticide can be even harder than developing a new cancer
drug, because a cancer drug affects only the human body while an insecticide
can have impacts across entire ecosystems.
“We have to simulate the performance of a new crop protection chemical in
100 different environments—being able to use generative AI to make predictions
about which ones are likely to perform best can save us huge amounts of trials.”
Before generative AI can fully address such audacious goals, it must be integrated
across the Leverkusen, Germany-based company’s pharmaceutical, consumer
products, and crop science units. Anderson, who took the helm at the global life
sciences giant in 2023, is expected to be a change agent. He believes this enterprise
transformation is possible—and necessary.
“You don’t last 160 years by resting on your successes in the past,” he says of the
storied business, which operates in 83 countries and brings in €50 billion in annual
revenue. “There’s no safe space in the corporate world where you can just hang out
and enjoy your winnings from the past, right? You’ve got to always be driving forward
to the next horizon.”
55
Focusing generative AI adoption in
essential business functions helps
organizations create transformative,
top-line growth.
Where is generative AI delivering
the most value today?
Generative AI promises to be a powerful catalyst for business
transformation—but it’s not a panacea.
It must be implemented with careful consideration of cost, data governance, and
ethical implications, as well as an eye toward talent and skills. Because generative AI’s
biggest strength is to augment human work rather than automate it, culture change
is essential to deliver sustained value. In fact, 64% of CEOs say succeeding with
generative AI will depend more on people’s adoption than the technology itself.6
Instead of applying generative AI as a solution for every problem, leaders need to
understand how different tools work together, with traditional AI techniques,
generative AI models, and automation each playing their own part. They must
break out of the use case mindset and focus on using generative AI to transform how
employees work every day. Getting there is a journey—and how much experience
an organization has with AI influences where it should start.
Organizations are taking two main approaches to drive the systemic change needed
to deliver sustained AI ROI.
1. Experimentation: Finding efficiencies in low-risk, non-core functions.
Prioritizing generative AI adoption in low-risk areas where traditional AI is already
delivering clear business value helps accelerate transformation and can drive
incremental profitability. Roughly two-thirds of executives say their organizations
are adopting generative AI in customer service (70%), IT (65%), and product
development (65%) functions, which is consistent with what we saw in
mid-2023.7
2. Focus: Augmenting essential business functions to spark broader transformation.
The risk of using generative AI in business operations closer to the core may
be higher—but this is where the promise of business transformation begins to
take shape. Those willing to focus on the previously underexplored areas of
sales; information security; and supply chain, logistics, and fulfillment are
seeing higher ROI.
6
Of course, for many organizations, it makes sense to start a generative AI journey
by experimenting in lower-risk areas. They benefit from marginal gains while teams
learn how to make the most of the technology. But staying in the shallows also keeps
organizations from realizing the more transformative, top-line growth generative AI
can create. Only by setting their sights on enterprise-wide innovation—and focusing
their efforts in areas with the greatest potential—can organizations achieve
long-term, scalable success.
FIGURE 2
Mapping the generative AI journey
Focusing closer to the core does more
to drive enterprise transformation
Strategic Company Percentage of
Focus importance of AI types respondents
Foundational to Essential AI start-ups
business model
Business Central to business Various hyperscalers,
model-centric strategy software/ hardware companies
New business Key to transformation Companies using AI to help
model enabler and platform strategy achieve platform economics
17%
Product- Core to organic growth Companies embedding
embedded and innovation AI into core products
and R&D to differentiate
33%
Vertically Important to business Companies integrating AI within
integrated unit effectiveness function unit(s) to improve customer
experience/operational effectiveness
Horizontally Important to functional Companies deploying AI in functions
deployed effectiveness to improve customer experience/
operational effectiveness
31%
Opportunistic Ad hoc Companies experimenting
in low-risk, noncore
business areas
19%
Experimentation
7
noitamrofsnart
fo
eergeD
Perspective
Breadth versus depth
Organizations are implementing generative AI differently based on their starting point.
AI luminaries are leveraging their experience to drive Generative AI opportunists have low-to-medium
wider transformation with generative AI. They’re levels of adoption for AI overall—though their
already operationalizing and optimizing traditional AI adoption spikes in areas where they have
and are primarily using generative AI to improve on experience with traditional AI. They’re
existing AI capabilities. They have the highest experimenting with generative AI in three key
adoption maturity across functions for both functions: IT, customer service, and information
traditional and generative AI and are delivering higher security. By exploring areas where they see the
ROI with traditional AI than their peers. In most greatest potential, they’re delivering higher ROI
functions, at least 60% have implemented generative from generative AI than their peers.
AI, which means they have an opportunity to focus on
the areas that are already delivering the most value.
AI luminaries Generative AI opportunists
83% Customer service 55%
76% Information technology 67%
72% Information security 59%
67% Research and innovation 46%
64% Manufacturing 1%
64% Marketing 29%
61% Finance 33%
61% Sales 29%
Human resources
60% 19%
Supply chain, logistics,
60% 3%
and fulfillment
59% Procurement 5%
57% Product development 17%
55% Risk and compliance 26%
Q: Where is your organization in its adoption of generative AI for the following functional areas? Percentages include
88 respondents who selected implementing, operating, and optimizing.
Productivity gains that provide
an advantage today will be
table stakes tomorrow.
How to deliver long-term value
The path from experimentation to enterprise-scale innovation
isn’t a straight line. How adoption evolves depends on where an
organization is starting from, which capabilities it has developed,
and how prepared its workforce is to adapt.
At the same time, as generative AI matures, it’s likely that competitive capabilities
will begin to converge—making it more difficult to gain a competitive edge. That’s
why organizations must do the hard work of addressing the obstacles and challenges
that come with generative AI. And they need to do it quickly. What provides an
advantage today will be table stakes tomorrow.
For those early in the journey, deploying generative AI in low-risk functions can
help jumpstart progress toward business transformation. Experimentation and
small-scale wins can streamline workflows and increase efficiency while teams gain
their footing. Our research highlights two key areas as smart places to start:
Customer service
Our analysis suggests that in both generative AI adoption and ROI, customer service
leads the way. Many companies already have a solid foundation of traditional AI to
build upon, such as conversational AI that answers customer queries in natural
language. Recent IBM IBV research found that, on average, organizations using
generative AI in customer service see higher AI ROI than those that don’t.8 But it’s
not unambiguous. One trap to be aware of: Most customer-service use cases only
focus on making existing workflows more efficient. That will change quickly. By the
end of 2024, executives point to three rising opportunities: generating test cases for
training conversational AI (78%), generating dialogue for conversational AI (74%),
and generating dialogue for human agents (69%) (see “AI shifts customer service
into overdrive,” page 11).
IT
Developers are leaning on generative AI to help streamline routine tasks. For instance,
77% of companies that have adopted generative AI in IT are using it to generate code.
They’re also using it to automate code testing by identifying and fixing bugs and
helping ensure the code works as intended. Generative AI also speeds the process
of creating required documentation, including user manuals and other technical
materials that accompany software development and cybersecurity reviews.9
9
These areas are starting points for delivering long-term ROI, offering productivity
gains with meaningful impact. But over time, the biggest gains will come from
focusing generative AI deployment in business functions closer to the core.
Our research indicates that leading organizations are beginning to use generative AI
in previously unexplored areas, such as sales and supply chain, to rethink how work
gets done:
Sales and marketing
Generative AI can boost sales team performance by tapping customer data to
provide insights into their behavior. It identifies quality leads within high-value
market segments, making marketing strategies and outreach efforts more effective.
In fact, 85% of companies that have adopted generative AI in marketing are using
it to summarize market intelligence. Sales and marketing teams are also saving time
by using generative AI to write and edit creative content for emails, blogs, social
media posts, and websites in minutes—not hours—and then invest the time they’ve
saved into finding new ways to build customer relationships.10
Supply chain
As supply chain disruption intensifies, generative AI helps spot potential snags
and find workarounds before issues impact delivery. It enables intuitive
conversations between supply chain decision makers and AI assistants—making
their impact more tangible and relevant by providing the information they need in
real time. By automating mundane tasks and augmenting workflows, generative
AI also lets supply chain professionals focus on complex problem resolution and
process improvement.11 For example, 80% of companies adopting generative AI
in supply chain use it to generate operations documents.
But for some organizations, transformative opportunities like these seem out
of reach. That’s why some business leaders are considering a platform approach
to generative AI that pools resources and gains across departments or partner
organizations as a lower-cost, simpler-to-implement option. This way, leaders can
avoid starting from scratch in each area and embed generative AI quickly and more
strategically across functions that have the greatest potential, including finance,
supply chain and manufacturing, human resources, and sales and marketing.
However, leaders taking this approach also need to consider the unique needs of
each function and find ways to fine-tune generative AI applications accordingly.
75% of organizations are at least
piloting generative AI in five or
more functions.
10
Perspective
AI shifts customer
service into overdrive
From chatting with customers to creating targeted content to optimizing call center
performance, generative AI is taking the transformation of customer service to the
next level. Using natural language generation, it answers customer questions with
more fluent, contextually relevant responses. It can also tap into a customer’s
interaction history to tailor responses and deliver a more personalized experience.
These capabilities let customers chat with generative AI assistants in the same way
they would engage a human agent.
What’s more, the applications of generative AI go far beyond direct interactions
with customers. This technology can enhance the customer service function
more generally by supporting human agent training, increasing personalization,
translating content, and predicting future customer behavior. It can also support
customer-facing conversational AI by generating test cases and dialogue, as well
as reviewing interactions to identify opportunities for improvement.12
These use cases help generative AI supercharge conversational AI with less human
intervention. Using generative AI to create test cases—steps used to verify that an AI
model is working as intended—and responses to a variety of customer queries helps
teams training and fine-tuning conversational AI handle a wide range of scenarios,
user inputs, and edge cases.
11
Case study
Zebra Technologies
empowers the
augmented workforce13
Bill Burns, CEO of Zebra Technologies, expects generative AI to have a
positive impact on the way people work, including the company’s employees
and the users of the rugged mobile devices they produce. “Our business is
focused on the frontline worker,” he says of the $4.5 billion manufacturer-
turned-digital-solutions-provider that enables businesses to intelligently
connect data, assets, and people in industries. The company makes smart
tracking, marking, and printing devices for logistics and other functions in
industries including retail, manufacturing, transportation, healthcare and
public service.
Early notions about generative AI making people obsolete are themselves
outdated, he believes. “It’s not replacing the worker, it’s automating select
tasks within the workflow to augment and return time to the worker,
ultimately empowering the worker and allowing them to focus on higher
value activities.”
Zebra Technologies is methodical in its generative AI investments and has a
high standard for acting on use cases across the enterprise. Mr. Burns cites
the organization’s approach of “Sense, Analyze, Act” as a guide to process
and a way to avoid succumbing to the hype. The goal is to understand actions
and changes in workflow that drive improved outcomes, such as speed of
operations, accuracy, consistency, and overall productivity, then quantify
the impact and articulate an ROI. “Prove it to me and demonstrate that
there’s a business case,” he says.
This measured strategy must be weighed against the rapid maturation of
generative AI and the demands of the marketplace. “You have to have an
urgency around everything you do and operate with two speeds. Speed one
is deliberate, focused on execution and getting solutions into the hands of
our customers for those use cases we are confident will generate value.
Speed two is less structured and experimental, co-innovating with
customers and discovering new areas that can benefit from AI innovation,”
he says. “If we don’t do it, somebody else will.”
“Generative AI will make
employees’ jobs easier and improve
the customer experience.”
Bill Burns, CEO, Zebra Technologies
12
Zebra Technologies empowers the “You have to have an urgency
augmented workforce (continued)
around everything you do...If we
don’t do it, somebody else will.”
Bill Burns, CEO, Zebra Technologies
Mr. Burns is keeping stakeholders across the Creating the future of work
company close as Zebra begins its generative AI
Mr. Burns expects the net impact of generative AI
evaluations. The plan: “Educate ourselves on
to include many good job opportunities, with new
generative AI while connecting with large strategic
positions created that are less stressful, and let
tech partners, form a cross-functional team with the
employees focus on more meaningful work, develop
CTO of the organization looking externally and the
enhanced abilities, and learn and grow on the job
CIO looking internally, and work together to define
quickly. He senses a shift in the way executives are
responsible and ethical AI principles across the
talking about what comes next. “It has evolved from
organization as the space evolves,” he says. “One of
all this hype of ‘workers are going to be replaced’ to
the keys involves communication and change
tasks being automated,” he says. “Generative AI will
management to ensure everyone knows of these
make employees’ jobs easier and improve the
teams and embraces new ways of operating—and
customer experience.”
that starts at the top.”
He points to the example of software developers
Critical decisions for implementation are made after
who can now use the technology to write code.
a thorough review of expected benefits, and of
“Developers will not lose their jobs but instead can
costs. “People think it’s all free because today they
spend more time on value-added tasks or simply
go on to ChatGPT and it’s free,” he says. “If you want
getting more done, especially given underlying labor
to use it at scale inside an enterprise, it’s no longer
availability and cost challenges,” he says.
free, as these solutions consume real resources in
the cloud. But Zebra is exploring and has developed The technology should empower people to move
a frontline worker application that runs the gen AI quickly, he says, by giving them easy access to
model on a mobile device. This reduces costs, actionable insights derived from both structured and
improves security, and protects data.” unstructured data. Faster training and reduced time
to proficiency obviously pay off for employers, too.
Zebra Technologies has already identified many
For instance, think of businesses in the retail sector,
internal use cases for generative AI that have the
which can see very high job turnover, or fields where
potential to change employee workflows. These
new employees have traditionally needed extended
include everything from building marketing
training periods. “With a mobile device as a window
campaigns quickly with multiple languages to
into a generative AI assistant, the newest employee
enabling its customer service teams to provide a
can quickly become as proficient as a much more
more personalized customer experience with
experienced employee while benefitting from
quicker issue resolution times.
greatly improved job satisfaction.”
In terms of product development focused on
customers, the strategy is to use open-source large
language models residing on the company’s
next-generation mobile devices. Zebra then
fine-tunes these models by use case using its own
data while leveraging a platform for customers to
populate the models with their own data and tie into
their systems via Zebra mobile devices.
13
14
Organizations that build a solid
foundation for generative AI today will
be able to pivot and build momentum
as new opportunities arise.
Building a springboard for growth
Despite generative AI’s progress, challenges remain. Almost half of
executives say they’re concerned about accuracy and bias—an issue
that could create as many new problems as generative AI promises
to solve.
Many leaders are also concerned that inadequate expertise, unclear business
cases, and insufficient proprietary data could preclude progress with
generative AI (see Figure 4).
FIGURE 4
A confluence of challenges
Organizations must overcome many obstacles to make
headway with generative AI—and executives’ top concerns
are shifting as it matures
45% Concerns about data accuracy or bias
42% Insufficient proprietary data available to customize models
42% Inadequate generative AI expertise
42% Inadequate financial justification/business case
40% Concerns about privacy/confidentiality of data and information
40% Limited access to technology
36% Requires too much investment
32% Concerns about intellectual property
31% Irrelevant or unclear use cases
31% Security of data/concerns about cybersecurity
19% Constrained by regulation/compliance
12% Not aligned to business strategy
1% Inadequate infrastructure/no barriers
April 2023 August 2023 March 2024
15
Overcoming challenges requires an interconnected effort that brings
together leaders from technology, finance, security, legal, and AI ethics.14
It’s complex work, but avoiding it comes with serious consequences. From
increasing liability to introducing new security vulnerabilities to damaging
brand reputation, leaders must understand and mitigate a litany of new risks
as they integrate generative AI into business operations.
Some organizations are already making efforts to manage these threats:
80%
have a separate part of their risk function dedicated
to risks associated with the use of AI or generative AI.
81%
conduct regular risk assessments to identify
potential security threats introduced by generative AI.
78%
maintain robust documentation to enhance explainability
of how generative AI models work and were trained.
76%
establish clear organizational structures, policies,
and processes for generative AI governance.
72%
develop policies and procedures for managing data
and addressing potential risks.
These activities should be part of any robust generative AI risk management
strategy. But identifying where their organization needs to focus its attention
should be a top priority for leaders as they begin to use generative AI in areas
that are core to their competitive advantage.
Reimagine what’s possible
AI has the potential to transform the business world, economies, and
societies in ways that are hard to imagine. By building the right capabilities
today, organizations can bring these new opportunities into focus.
It’s not just about automating things people are doing today. It’s about doing
things that were never possible before.15 From helping develop cures for
diseases to combatting climate change, generative AI could solve problems
that have confounded people for centuries. More than half of the executives
in our survey say that, in the next three years, generative AI will make
entirely new types of work possible (see Figure 5).
It’s difficult to imagine what these new use cases might look like‚ but that’s
kind of the point. The generative AI applications that could deliver the
greatest value tomorrow have yet to be discovered. The organizations that
build a solid, capability-rooted foundation for generative AI today will
be able to pivot and build momentum as new opportunities arise.
16
“Process automation is not about replacing
an individual. It’s about enhancing the
value of individuals—making human
work more human.”
Javier Tamargo, CEO, 407 ETR
By providing a platform that lets employees experiment safely,
organizations can unlock the collective genius of their workforce.
Leaders will need to foster a growth and innovation mindset—and
encourage employees to look beyond what’s worked in the past—to
pioneer groundbreaking innovation, outpace the competition, and
drive transformative growth at scale with generative AI.
FIGURE 5
Forging a new frontier
Generative AI opens a new world of opportunity
New
opportunities
Augmented
human/machine
tasks
Human tasks
Note: Figure is conceptual in nature. Proportions are not derived from data.
17
Perspective
IBM and NASA are
helping humanity adapt
to a changing climate16
Nearly a quarter of the world’s population now lives in a flood zone, and that number
is expected to climb as rising seas and heavier storms triggered by a changing
climate put more people at risk. The ability to accurately map flooding events can
be key to not only protecting people and property now but steering development
to less-risky areas in the future.
IBM and NASA’s geospatial foundation models are designed to enable important
steps toward this goal by converting NASA’s satellite observations and data into
customized maps of natural disasters and other environmental changes. Potential
applications include helping to estimate climate-related risks to crops, buildings,
and other infrastructure; monitoring and valuing forests for carbon-offset programs;
generating renewable energy forecasts; and developing predictive models
to help enterprises create strategies to mitigate and adapt to climate change.
As part of a Space Act Agreement, IBM and NASA set out to build the first-ever
foundation model for analyzing geospatial data in early 2023. Previously, users
had to train a new model for each task, which required extensive data curation
and compute. Rather than train a foundation model on words, IBM Research taught
a model to understand satellite images. The team then fed the model hand-labeled
examples to teach it to recognize the extent of historic floods and fire burn scars,
changes in land-use and forest biomass, and more.
IBM and NASA expanded the family of models in 2024, developing a foundation
model for weather and climate data. They customized this model for more specific
tasks, such as creating highly localized wind forecasts for renewable energy
planning and increasing the resolution of climate simulations to better understand
and plan for the local effects of climate change.
Using the foundation model is designed to be as simple as selecting a region,
a mapping task, and a set of dates. For example, if a user types “Port-de-Lanne,
France” into the search bar and selects a date range of December 13 to 15, 2019,
the model highlights in pink how far the flood waters extended. Users can overlay
other datasets to see where crops or buildings were inundated. The models and
accompanying visualizations can help with future planning during similar disaster
scenarios: they provide information that could help mitigate flood impacts, inform
insurance and risk management decisions, define infrastructure plans, improve
disaster response, and protect the environment.
18
IBM and NASA are helping humanity
adapt to a changing climate (continued)
IBM and NASA built both models using a masked autoencoder for |
142 | ibm | sg248573.pdf | Front cover
Simplify Your AI Journey:
Ensuring Trustworthy AI with
IBM watsonx.governance
Deepak Rangarao Mohit Sharma
Upasana Bhattacharya Mark Simmonds
Savitha Chinnappareddy PhD Jasmeet Singh
Larry Coyne Martijn Wiertz
David Cruz
Shuvanker Ghosh
Prem Piyush Goyal
Vasfi Gucer
Amna Jamal PhD
Warren Lucas
Karen Medhat
Bob Reno
Artificial Intelligence
Data and AI
Redbooks
IBM Redbooks
Ensuring Trustworthy AI with IBM watsonx.governance
January 2025
SG24-8573-00
ii Ensuring Trustworthy AI with IBM watsonx.governance
Contents
Notices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .vii
Trademarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viii
Foreword. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix
Preface. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .x
Authors. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .x
Now you can become a published author, too! . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii
Comments welcome. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii
Stay connected to IBM Redbooks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiv
Chapter 1. Challenges and opportunities in AI governance for responsible AI . . . . . . 1
1.1 What is AI governance? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Governance as a key enabler for realizing AI value . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2.1 Concern 1: Governance is a brake on AI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2.2 Concern 2: Governance does not scale . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2.3 Concern 3: Governance does not contribute to value generation. . . . . . . . . . . . . . 6
1.3 Challenges with governance of enterprise AI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3.1 Generative AI has changed the governance game. . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3.2 Bring together diverse stakeholder perspectives . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.3.3 Technical complexity is increasing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.3.4 Regulatory and risk complexity is increasing . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.4 An example of legislation and standards related to AI . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.4.1 AI-specific legislation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.4.2 General regulations that apply to AI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.4.3 Technical standards for AI governance. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Chapter 2. Introduction to IBMwatsonx.governance . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.1 Introduction to the IBM watsonx platform and its core components . . . . . . . . . . . . . . . 16
2.2 Introduction to IBM watsonx.ai . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.3 Introduction to IBM watsonx.data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.4 Introduction to IBM watsonx.governance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.4.1 Key capabilities. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.4.2 Use cases. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.4.3 Benefits of watsonx.governance. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.5 Reference architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.5.1 Data Onboarding. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.5.2 Data Preparation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.5.3 AI Building and Deployment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.5.4 AI Lifecycle Management and Governance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
Chapter 3. Implementing AI governance strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.1 Understanding the end-to-end AI lifecycle governance process. . . . . . . . . . . . . . . . . . 28
3.2 Elements of model risk governance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.2.1 Personas. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.2.2 Objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.2.3 Workflows. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.3 Considerations to implement AI governance strategy. . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.3.1 Understanding organizational characteristics. . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.3.2 Configuring AI governance. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
© Copyright IBM Corp. 2025. iii
3.3.3 Leveraging out-of-the-box product content. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.3.4 Example use case. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
Chapter 4. Onboarding a new foundation model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.1 Key considerations to onboard a foundation model . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.1.1 Data transparency. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.1.2 Model evaluation and validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.1.3 Model security and robustness. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.1.4 Ensuring model health and performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.2 Considerations for legal team for approving a new foundation model . . . . . . . . . . . . . 44
4.2.1 Model licensing. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
4.2.2 Legal obligations on the part of the vendor. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.2.3 A final note on legal considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.3 Ethical considerations for approving a new foundation model . . . . . . . . . . . . . . . . . . . 47
4.3.1 Fairness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.3.2 Transparency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.3.3 Privacy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.3.4 Explainability. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.3.5 Robustness. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.3.6 Third-party help. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.4 Considerations for financial stakeholders for approving a new foundation model . . . . 49
4.4.1 Total cost of ownership. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
4.4.2 Return on investment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
4.4.3 Build or buy. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.4.4 Exit strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.4.5 Other factors. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
Chapter 5. Assessing a new use case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
5.1 Business process workflow. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
5.2 Approval workflow. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
5.3 Risk identification assessment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
5.4 Applicability assessment. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
Chapter 6. Governing the end-to-end lifecycle of an AI asset . . . . . . . . . . . . . . . . . . . 61
6.1 What is the AI lifecycle? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
6.2 Metrics in watsonx.governance. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
6.2.1 Drift detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
6.2.2 Explainability. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
6.2.3 Model health. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
6.2.4 Generative AI quality. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
6.2.5 RAG quality metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
6.3 How to implement Lifecycle Governance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
6.3.1 Getting started: Setting up your AI use cases. . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
6.4 Lifecycle implementation and considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
6.4.1 UI-driven implementation of lifecycle governance. . . . . . . . . . . . . . . . . . . . . . . . . 68
6.4.2 Considerations for lifecycle governance for traditional ML hosted on watsonx.ai. 70
6.4.3 Considerations for prompt templates from another platform. . . . . . . . . . . . . . . . . 72
6.4.4 Considerations for traditional ML from another platform. . . . . . . . . . . . . . . . . . . . 74
6.4.5 Governing AI embedded in a business application. . . . . . . . . . . . . . . . . . . . . . . . 74
Chapter 7. Use cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
7.1 Overview of use case 1- Banking credit risk management. . . . . . . . . . . . . . . . . . . . . . 78
7.1.1 Banking credit risk management use case. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
7.1.2 Business context. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
iv Ensuring Trustworthy AI with IBM watsonx.governance
7.1.3 Client need . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
7.1.4 Client challenges. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
7.1.5 Business benefits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
7.1.6 Pilot solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
7.2 Overview of use case 2 - Automated governance for universal bank's AI chatbot. . . . 80
7.2.1 Business context. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
7.2.2 Client need . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
7.2.3 Client challenges. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
7.2.4 Business benefits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
7.2.5 Pilot solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
7.3 Overview of use case 3 - Belgian biopharmaceutical company . . . . . . . . . . . . . . . . . . 81
7.3.1 Business context. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
7.3.2 Client need . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
7.3.3 Client challenges. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
7.3.4 Business benefits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
7.3.5 Pilot solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
Related publications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
IBM Redbooks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
Online resources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
Help from IBM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
Contents v
vi Ensuring Trustworthy AI with IBM watsonx.governance
Notices
This information was developed for products and services offered in the US. This material might be available
from IBM in other languages. However, you may be required to own a copy of the product or product version in
that language in order to access it.
IBM may not offer the products, services, or features discussed in this document in other countries. Consult
your local IBM representative for information on the products and services currently available in your area. Any
reference to an IBM product, program, or service is not intended to state or imply that only that IBM product,
program, or service may be used. Any functionally equivalent product, program, or service that does not
infringe any IBM intellectual property right may be used instead. However, it is the user’s responsibility to
evaluate and verify the operation of any non-IBM product, program, or service.
IBM may have patents or pending patent applications covering subject matter described in this document. The
furnishing of this document does not grant you any license to these patents. You can send license inquiries, in
writing, to:
IBM Director of Licensing, IBM Corporation, North Castle Drive, MD-NC119, Armonk, NY 10504-1785, US
INTERNATIONAL BUSINESS MACHINES CORPORATION PROVIDES THIS PUBLICATION “AS IS”
WITHOUT WARRANTY OF ANY KIND, EITHER EXPRESS OR IMPLIED, INCLUDING, BUT NOT LIMITED
TO, THE IMPLIED WARRANTIES OF NON-INFRINGEMENT, MERCHANTABILITY OR FITNESS FOR A
PARTICULAR PURPOSE. Some jurisdictions do not allow disclaimer of express or implied warranties in
certain transactions, therefore, this statement may not apply to you.
This information could include technical inaccuracies or typographical errors. Changes are periodically made
to the information herein; these changes will be incorporated in new editions of the publication. IBM may make
improvements and/or changes in the product(s) and/or the program(s) described in this publication at any time
without notice.
Any references in this information to non-IBM websites are provided for convenience only and do not in any
manner serve as an endorsement of those websites. The materials at those websites are not part of the
materials for this IBM product and use of those websites is at your own risk.
IBM may use or distribute any of the information you provide in any way it believes appropriate without
incurring any obligation to you.
The performance data and client examples cited are presented for illustrative purposes only. Actual
performance results may vary depending on specific configurations and operating conditions.
Information concerning non-IBM products was obtained from the suppliers of those products, their published
announcements or other publicly available sources. IBM has not tested those products and cannot confirm the
accuracy of performance, compatibility or any other claims related to non-IBM products. Questions on the
capabilities of non-IBM products should be addressed to the suppliers of those products.
Statements regarding IBM’s future direction or intent are subject to change or withdrawal without notice, and
represent goals and objectives only.
This information contains examples of data and reports used in daily business operations. To illustrate them
as completely as possible, the examples include the names of individuals, companies, brands, and products.
All of these names are fictitious and any similarity to actual people or business enterprises is entirely
coincidental.
COPYRIGHT LICENSE:
This information contains sample application programs in source language, which illustrate programming
techniques on various operating platforms. You may copy, modify, and distribute these sample programs in
any form without payment to IBM, for the purposes of developing, using, marketing or distributing application
programs conforming to the application programming interface for the operating platform for which the sample
programs are written. These examples have not been thoroughly tested under all conditions. IBM, therefore,
cannot guarantee or imply reliability, serviceability, or function of these programs. The sample programs are
provided “AS IS”, without warranty of any kind. IBM shall not be liable for any damages arising out of your use
of the sample programs.
© Copyright IBM Corp. 2025. vii
Trademarks
IBM, the IBM logo, and ibm.com are trademarks or registered trademarks of International Business Machines
Corporation, registered in many jurisdictions worldwide. Other product and service names might be
trademarks of IBM or other companies. A current list of IBM trademarks is available on the web at “Copyright
and trademark information” at https://www.ibm.com/legal/copytrade.shtml
The following terms are trademarks or registered trademarks of International Business Machines Corporation,
and might also be trademarks or registered trademarks in other countries.
IBM® IBM Watson® Redbooks (logo) ®
IBM Cloud® OpenPages®
IBM Research® Redbooks®
The following terms are trademarks of other companies:
Microsoft, and the Windows logo are trademarks of Microsoft Corporation in the United States, other
countries, or both.
OpenShift, Red Hat, are trademarks or registered trademarks of Red Hat, Inc. or its subsidiaries in the United
States and other countries.
RStudio, and the RStudio logo are registered trademarks of RStudio, Inc.
Other company, product, or service names may be trademarks or service marks of others.
viii Ensuring Trustworthy AI with IBM watsonx.governance
Foreword
This trilogy of IBM® Redbooks® publications positions and explains IBM watsonx, the
IBM strategic AI and Data platform. Each book focuses on one of the three main components
of the watsonx platform:
(cid:2) IBM watsonx.ai: A next-generation enterprise studio for AI developers to train, validate,
tune, and deploy both traditional ML and new generative AI capabilities powered by
foundation models.
(cid:2) IBM watsonx.data: A fit-for-purpose data store built on an open-lakehouse architecture,
optimized for different and governed data and AI workloads.
(cid:2) IBM watsonx.governance: A set of AI governance capabilities enabling trusted AI
workflows, helping organizations implement and comply with ever-changing industry and
government regulations.
Organizations have long recognized the value that IBM Redbooks provide in guiding them
with best practices, frameworks, clear explanations, and use cases as part of their solution
evaluations and implementations.
This trilogy of books was only possible due to the close collaboration involving many skilled
and talented authors that were selected from our IBM global technical sales, development,
Expert Labs, Client Success Management, and consulting services organizations, using their
diverse skills, experiences, and technical knowledge across the watsonx platform.
I would like to thank the authors, contributors, reviewers, and the IBM Redbooks team for their
dedication, time, and effort in making these publications a valuable asset that organizations
can use as part of their journey to AI.
I also want to thank Mark Simmonds and Deepak Rangarao for taking the lead in shaping this
request into yet another successful IBM Redbooks project.
It is my sincere hope that you enjoy this watsonx trilogy as much as the team who wrote and
contributed to them.
Steve Astorino, IBM General Manager - Development, Data, AI and Sustainability.
© Copyright IBM Corp. 2025. ix
Preface
IBM® watsonx™ is the IBM strategic AI and Data platform. This book focuses on
watsonx.governance, a key component of the platform.
IBM watsonx.governance offers a comprehensive solution for governing data and AI
workloads within a secure and scalable environment. Built on an open architecture, it
empowers organizations to manage data access, compliance, and security across hybrid
multi-cloud deployments. IBM watsonx.governance simplifies data governance with built-in
automation tools and integrates seamlessly with existing databases and tools, streamlining
workflows and enhancing user experience
This IBM Redbooks publication provides a broad understanding of watsonx.governance
concepts and architecture, and the services that are available in the product. In addition,
several common use cases and scenarios are included that should help you better
understand the capabilities of this product.
This publication is for watsonx customers who seek best practices and real-world examples of
how to best implement their solutions while optimizing the value of their existing and future
technology, AI, data, and skills investments.
Note: Other books in this series are:
(cid:2) Simplify Your AI Journey: Unleashing the Power of AI with IBM watsonx.ai, SG24-8574
(cid:2) Simplify Your AI Journey: Hybrid, Open Data Lakehouse with IBM watsonx.data,
SG24-8570
Authors
This book was produced by a team of specialists from around the world working with the
IBM Redbooks, Tucson Center.
Deepak Rangarao is an IBM Distinguished Engineer and CTO responsible for Technical
Sales-Cloud Paks. Currently, he leads the technical sales team to help organizations
modernize their technology landscape with IBM Cloud® Paks. He has broad cross-industry
experience in the data warehousing and analytics space, building analytic applications at
large organizations and technical pre-sales with start-ups and large enterprise software
vendors. Deepak has co-authored several books on topics, such as OLAP analytics, change
data capture, data warehousing, and object storage and is a regular speaker at technical
conferences. He is a certified technical specialist in Red Hat OpenShift, Apache Spark,
Microsoft SQL Server, and web development technologies.
Upasana Bhattacharya is a Senior Product Manager for watsonx.governance, based in
Markham, Canada. In this role she defines the product vision, guides its development,
collaborating with cross-functional teams. In her previous role she was a Product Manager for
Data and AI. Upasana holds a Bachelor of Arts in Economics and Foreign Affairs from the
University of Virginia and an MBA from the McCombs School of Business at the University of
Texas.
x Ensuring Trustworthy AI with IBM watsonx.governance
Savitha Chinnappreddy, PhD is a Senior AI Engineering Manager at IBM with over 17 years
of experience in AI and Data Analytics. She holds a PhD in AI and Data Analytics and is
currently pursuing a post-doctorate focused on Human & AI Collaboration: Governance
strategies for trustworthy AI & Safe AI systems. She has extensive Experience in managing
and scaling large AI and Data Science teams, she has worked closely with architecture and
infrastructure teams to establish compliant pipelines for AI and analytics, delivering impactful
solutions to global customers. With 11 publications in esteemed journals and conferences, as
well as holding a patent, she is also an active guest speaker and participant in faculty
development programs, committed to sharing her knowledge and inspiring the next
generation of AI professionals.
Larry Coyne is a Project Leader at the IBM International Technical Support Organization,
Tucson, Arizona, center. He has over 35 years of IBM experience, with 23 years in
IBM storage software management. He holds degrees in Software Engineering from the
University of Texas at El Paso and Project Management from George Washington University.
His areas of expertise include client relationship management, quality assurance,
development management, and support management for IBM storage management software.
David Cruz is a Data Scientist and AI Engineer working under IBM’s Client Engineering
team. In this role, David has been dedicated to the Federal Market where he works to
implement a wide range of AI solutions for federal clients. In his prior role, he worked under
the Data Science Elite team where he gained skills with IBM platforms for Governance,
namely IBM OpenScale, and this has translated into a growing skill set with watsonx
governance. He is constantly working to implement the cutting edge of AI and AI Governance
technology, and has written various blog posts on topics ranging from Unsupervised Learning
techniques, to RAG how-to guides for beginners.
Shuvanker Ghosh is a certified Executive Architect and Worldwide Platform Leader for Data
and AI in Worldwide Solution Architecture in IBM Technology Expert Labs. With 18 years of
experience at IBM, he serves as a trusted advisor to clients, offering thought leadership on
IBM's Data and AI portfolio. He guides organizations in their responsible AI journey, helping
them adopt best practices. His current focus is on defining solution blueprints and
architectural patterns that assist clients in addressing their business challenges through
responsible and trustworthy AI solutions. He possesses extensive expertise in the IBM Data
and AI portfolio, including the watsonx platform and Cloud Pak for Data. Shuvanker has
successfully led and delivered complex programs that involve multiple teams, providing
technical management, architecture, technology thought leadership, and software
development methodologies and processes. His experience spans various industries,
including retail, finance, insurance, healthcare, telecommunications, and government
Prem Piyush Goyal is a problem solver with extensive experience in developing cutting-edge
technologies at IBM. Specializing in full-stack development, cloud-based microservices, and
AI solutions, he has worked on high-impact projects like IBM Watson® Data Platform and
IBM Watson OpenScale. His expertise spans Python, JavaScript, React, Kubernetes, and
AI-driven solutions like Explainable AI and Concept Drift Detection. Passionate about building
transparent and scalable AI, he continually enhances user experience and optimizes
performance for enterprise applications. His innovative mindset and problem-solving abilities
help drive trust and transparency in AI systems.
Vasfi Gucer leads projects for the IBM Redbooks team, leveraging his 20+ years of
experience in systems management, networking, and software. A prolific writer and global
IBM instructor, his focus has shifted to storage and cloud computing in the past eight years.
Vasfi holds multiple certifications, including IBM Certified Senior IT Specialist, PMP, ITIL V2
Manager, and ITIL V3 Expert.
Foreword xi
Amna Jamal PhD is a seasoned Data and AI Subject Matter Expert (SME) at IBM, boasting
over 8 years of expertise in data management and data science. With a Ph.D. in Engineering
from the National University of Singapore, she brings a wealth of knowledge and experience
to the field, driving innovation and excellence in the intersection of data and artificial
intelligence.
Warren Lucas is a member of IBM Expert Labs. Prior to his time at IBM, Warren has spent
nearly a decade working in Regulatory Compliance, Operational Risk, and Model Risk
Governance supporting a number of Fortune 50 companies in their efforts to redesign and
implement internal governance processes. As a Solution Architect, Warren has specialized in
Governance Console (IBM OpenPages®) for over seven years, where he has personally
performed development, design, advisory, and configuration within the platform. Warren has a
current patent submission for a novel approach in governance and confidence assessments
in large language models (LLMs); he holds a degree in Quantitative Economics.
Karen Medhat is a Customer Success Manager Architect in the UK and the youngest
IBM Certified Thought Leader Level 3 Technical Specialist. She is the Chair of the IBM
Technical Consultancy Group and an IBM Academy of technology member. She holds an
MSc degree with honors in Engineering in AI and Wireless Sensor Networks from the Faculty
of Engineering, Cairo University, and a BSc degree with honors in Engineering from the same
faculty. She co-creates curriculum and exams for different IBM professional certificates. She
also created and co-created courses for IBM Skills Academy in various areas of
IBM technologies. She serves on the review board of international conferences and journals
in AI and wireless communication. She also is an IBM Inventor and experienced in creating
applications architecture and leading teams of different scales to deliver customers' projects
successfully. She frequently mentors IT professionals to help them define their career goals,
learn new technical skills, or acquire professional certifications. She has authored
publications on Cloud, IoT, AI, wireless networks, microservices architecture, and Blockchain.
Bob Reno is a Principal Technical Sales Specialist with over 30 years of experience in Data
Warehousing, Analytics, and AI. As a member of the IBM World Wide Data and AI Technical
Sales team, Bob is a watsonx.governance leader working with customers to enable their
organizations to embrace responsible AI. Bob has contributed to the creation of several
IBM Certification Tests and written several workshops in the watsonx, Cloud Pak for Data and
Data Warehousing space to enable customers and the IBM Technical Community. Prior to
joining IBM, Bob has held roles as a Developer, Technical Architect, and Director of Data
Warehousing and Analytics.
Mohit Sharma is an AI engineering lead on the Client Engineering watsonx team in
Bangalore, India. Prior to this, Mohit was associated with IBM consulting, and worked on
client production projects involving classical ML and deep learning. Mohit has around 14
years of experience in AI, and worked at Hewlett Packard, Wipro (where he conceptualized
the Holmes AI platform) and Accenture before joining IBM in 2018. An AI practitioner having
experience in design and development of AI-based solutions using both open-source and
commercial technologies, Mohit is interested in both data and the science behind it. He has 4
published patents to his credit, and has filed his first patent at IBM.
Mark Simmonds is a Program Director in IBM Data and AI. He writes extensively on AI, data
science, and data fabric, and holds multiple author recognition awards. He previously worked
as an IT architect leading complex infrastructure design and corporate technical architecture
projects. He is a member of the British Computer Society, holds a Bachelor’s Degree in
Computer Science, is a published author, and a prolific public speaker.
xii Ensuring Trustworthy AI wit |
144 | ibm | embedding-ai-in-your-brands-dna.pdf | IBM Institute for Business Value | Research Insights
Embedding AI
in your brand’s DNA
Innovate from products to ecosystem—
and everything in between
How IBM
can help
IBM has been providing expertise to help retail and consumer products
companies win in the marketplace for more than a century. Our researchers
and consultants create innovative solutions that help clients become more
consumer-centric by delivering compelling brand and store experiences,
collaborating more effectively with channel partners, and aligning demand
and supply. With a comprehensive portfolio of solutions for merchandising,
supply chain management, omnichannel retailing, and advanced analytics,
IBM helps deliver rapid time to value. With global capabilities that span 170
countries, we help brands and retailers anticipate change and profit from
new opportunities. For more information on our retail and consumer
products solutions, please visit: ibm.com/industries/retail, ibm.com/
consulting/retail, and ibm.com/industries/consumer-goods.
2
Key
takeaways
Brands are evolving Over the next year, retail and consumer
beyond mere AI adoption, products executives expect to expand
embedding it in their DNA AI significantly throughout all areas
to harness their distinct of the business, from brand-defining
AI-driven advantage. activities to core operations.
But to be AI-centric, organizations need an open
mindset for how AI can deliver transformation beyond
productivity gains.
Across 13 areas of the business,
executives plan to augment most
activities with AI over the next
12 months.
But they only project 31% of their workforce will need
to reskill or develop new skills in that same time frame,
underestimating what’s needed to support employees
in the AI transformation.
Almost 9 in 10 executives claim to have
clear organizational structures, policies,
and processes for AI governance.
But fewer than one-quarter of organizations have fully
implemented and continuously review tools on AI
governance, putting brand trust at risk.
1
Industry executives project that
AI’s contribution to revenue
growth will increase 133%
from 2023 to 2027.
Consumers are ready for AI.
Are you?
Consumers are tech-savvy trendsetters and brands need to keep up to
stay relevant. Today, customers and shoppers are actively engaged with
AI in their daily lives, from using AI-powered search engines to creating
content with generative AI tools.
In the 2024 IBM Institute for Business Value (IBM IBV) consumer research study,
nearly two-thirds of consumers said they have used or want to try AI applications.1
This interest sets the stage for retail and consumer products companies to hasten
integration of AI across their business while keeping an eye toward becoming AI-led
brands—leveraging the technology to reimagine operations, inspire loyalty, and
expand the size of customers’ wallets for long-term competitive advantage.
Our latest survey of 1,500 global retail and consumer products executives finds
organizations are accelerating their adoption. AI—both traditional and generative—has
permeated all functions in the enterprise to some degree. From marketing and customer
service, to supply chain and procurement, to finance and IT operations, AI use cases
span brand-defining, business-enabling, and corporate operations. Looking ahead
through 2025, most executives are thinking big, expecting AI to be used extensively
across the business (see Figure 1). Industry leaders also report AI spending is on the
rise (see Perspective, “AI spending moves outside of IT”), and they project that AI’s
contribution to revenue growth will increase 133% from 2023 to 2027.
Retail and consumer products organizations are at a pivotal point in their AI journey.
The question is: are they taking enough of the right steps to become AI-led brands,
or are they just tacking on ad hoc AI solutions that deliver short-term gains? It’s time to
move beyond just productivity and efficiency and extend AI’s power enterprise-wide
to boost process effectiveness, spark new business models and ecosystems, and
ignite engagement with innovative employee and customer experiences.
2
FIGURE 1
Retail and consumer
products organizations
plan to use AI extensively
in 2025.
Figure 1
Retail and consumer products organizations use AI extensively in 2025.
Percent of organizations planning to use AI to a moderate or significant extent over the next 12 months
Marketing and customer experience 89%
Digital commerce 86%
Merchandising 86%
Customer service 85%
Brand-defining
areas
Stores
79%
Product design and development 76%
Supply chain operations 90%
Sustainability 87%
Procurement 86%
Business-enabling
areas
Production and manufacturing 83%
IT and security 90%
Finance 90%
Corporate
operations
HR 88%
Percentages represent an average of responses for a set of tasks in each functional area, based on the question: “To what extent
do you use AI or gen AI in this activity?” Respondents replied “to a moderate extent” or “to a significant extent.”
3
Perspective
In this report, we discuss three factors that will help
AI spending
organizations make a fundamental change in their
DNA, where AI emerges as the driving force behind shifts beyond
every decision, innovation, and strategy. In part one,
IT budgets
we discuss balancing the marathon with the sprint to
shift from plus-AI to AI-first. In part two, we examine
the need to prepare the workforce for the planned
rapid and aggressive AI adoption, and in part three,
we address the imperative to safeguard consumer AI budget allocation is undergoing a significant shift.
trust. Each section includes an illustrative case study While IT budgets will still play a role, retail and
and concludes with an action guide of steps brands consumer products executives report a growing
can take to accelerate progress. portion of AI spending is moving outside of traditional
IT budgets. As AI becomes more than just a tech tool,
functional areas are identifying their needs for AI as
part of larger business solutions, from creative
Definitions
marketing tools to empowering store associates to
new warehouse management systems.
Traditional artificial intelligence
Executives project their IT budget dedicated to AI
spend will increase by 19% over the next year, but
Systems that understand, reason, learn, and
spending on AI outside of the IT budget is expected
interact. AI technology includes machine
to surge 52%. As a percent of revenue, IT spending
learning (ML) approaches, but also other
on AI will be 1.04% and AI spending outside of IT
techniques such as reasoning, planning,
will be 2.28% by 2025. Taken together, 3.32% of
scheduling, and optimization.
revenue could be dedicated to AI spending next year.
For a $1 billion company, that equates to $33.2
Generative AI
million for total AI spend.
A class of machine learning that generates
With at least 13 functional areas that span retail and
content or data, including audio, code,
consumer products organizations, executives across
images, text, simulations, 3D objects, and
the C-suite must keep tabs on the investments being
videos—usually based on unsupervised or
made in each area, coordinating platforms and tools
self-supervised learning. Recent examples of
to provide transparency across the enterprise.
generative AI include GPT-4 (language), DALL-E
IT and the business lines must work together to avoid
(images), GitHub Copilot (code), and AlphaFold
duplication of effort and to help ensure consistent
(scientific protein folding).
alignment with the overall business strategy.
4
Part one
Building an intelligent brand
that endures
Consumer organizations need to take a long-term view
of their AI journey while moving with urgency and intent.
Nearly all industry executives are banking on AI for innovation in products and
services (89%) as well as business models (85%). But a mere 54% expect AI to
influence operational innovation. Transforming operations with AI across supply
chains, manufacturing, distribution, finance, and compliance is the very essence of
being an AI-centric brand. This remodel is both a marathon and a sprint—moving from
simple AI use cases to orchestrating AI across functions to deliver sustainable value.
Many organizations are in the early stage of adoption, integrating AI within a single
function. For example, 88% use AI to a moderate or significant extent in demand
forecasting, 87% for HR help desks, 84% for IT support and issue remediation, 84%
in creating and managing trade promotions, 81% in inventory and order management,
and 80% in managing production activities. These are quick wins that can deliver
a more immediate impact on daily operations.
But companies are keen on expanding to more sophisticated uses of AI over the next
12 months. They will be transitioning from internal departmental use cases with
limited system integration to multifaceted ones that require external collaboration,
more complex system integrations, and more human intervention and oversight.
Take virtual assistants as an example (see Figure 2). Initially, they responded to
simple, predefined queries such as order and shipment status. As they have become
more integrated with data in ordering systems, they can identify delays or missing
orders as well as back-order options and in-store availability. Adding customer
shopping history and generative AI capabilities to their arsenal, they can dynamically
recommend offerings and personalized content for individual customers. Camping
Only 54% of executives
World’s virtual assistant, Arvee, illustrates the value of integrating platforms such
expect AI to help their
companies innovate in as Oracle and Salesforce so that the assistant can access customer information
operations. efficiently to address queries faster.2
5
Executives expect to expand rapidly to more sophisticated AI use cases
across the enterprise. For example, those leveraging AI to a significant
extent for personalized responses and follow-up actions in customer
service plan to increase their usage by 236% over the next 12 months.
Similarly, they want to grow significant AI usage in integrated business
planning by 82% and in talent acquisition by 300%.
Figure 2
BFIrGaUnRdEs 2 and retailers plan to expand use of AI/gen AI
into more sophisticated use cases over the next year.
Brands are fueling virtual assistants with more
comprehensive, relevant enterprise data to enable
increasingly personalized responses to customers.
I can provide When I am connected to the When I have access to
shipment status order management warehouse customer profiles and shopping
and tracking and store inventory system, history, I can dynamically
information. I can provide options for recommend offerings and
back orders and in-store personalized content for
pickup options. individual engagement.
6
Case study
As organizations progress with their initiatives, they Kroger uses AI
are investing in platforms to integrate AI tools and
to elevate customer
models. Today, as they establish their AI foundation,
they are primarily focused on data and analytics pickup experiences3
platforms (65%), innovation platforms (64%), and
skills/learning platforms (62%). Building on these
existing platforms and expanding to others will
enable federation and orchestration of AI across
Kroger has long depended on data and advanced
functions, facilitating cross-functional learning to
analytics to fuel business innovation. Since its
support scaling AI across the enterprise.
inception decades ago, its loyalty program has
Executives plan to integrate AI capabilities with delivered a trusted value exchange enabled by
business partners over the next three years, and they permission-based information. Today, using machine
predict the use of ecosystem platforms will surge learning algorithms, Kroger delivers valuable
from 52% today to 89%. Take the product compliance personalized offers and communications across
ecosystem as an example. By integrating end-to-end 150 million customer touchpoints and through
AI-driven compliance, brands can ensure all facets 1.9 billion unique coupons customized for millions
of the product lifecycle align with evolving regulatory of loyal customers.
requirements, consumer safety, and sustainability
Most recently, Kroger has been exploring ways
expectations. This ecosystem prioritizes accelerated
to use AI to help improve the customer experience,
product lifecycle management with an advanced
specifically order pickups. Using AI-enabled dynamic
business rules engine and touchless bill-of-materials
batching, an AI solution sorts through 200,000 totes
generation, helping ensure products are market-
per second to build the most efficient pickup trolley.
ready with minimal manual intervention.
It drives a 10% reduction in steps by identifying the
most efficient pick route through the store. With
dynamic batching of orders, these tools are providing
associates the most efficient pick routes, so Kroger
can dramatically reduce pickup lead time in its
highest volume stores.
Executives expect their use
of ecosystem platforms for
AI tool and model integration
to surge from 52% today to
89% in the next three years.
7
Action guide
Intentionally embed AI in operations
to deliver a sustainable brand advantage.
In the 2024 IBM IBV CEO study, 70% of retail and consumer products
CEOs said that to win the future, they must rewrite their organizational
playbook.4 As you redefine your core operational strategies and processes
to capitalize on AI, concentrate on how to achieve lasting value.
Tailor AI to your As you move beyond AI-driven productivity gains, you need a clear vision and strategy
brand’s priorities. for where AI and gen AI can help you distinguish yourself from competitors or shore
up weaknesses. But keep in mind that consumers expect you to stay true to your core
values as you innovate. If a strong customer experience is your focus, use AI to
personalize customer service and optimize in-store experiences. If product innovation
is a differentiator, tap into AI for product design, customer preferences, and vendor
capabilities to facilitate faster ideation and development cycles. The key is to
concentrate on what’s most important—not everything that’s possible.
Invite finance, Becoming an AI-centric brand requires purposefully aligning IT with long-term
technology, and business goals, not just the hottest tech. For example, organizations that consider
business leaders applications and infrastructure holistically in support of business needs (known as
to the same table. “hybrid-by-design” principles) can generate more than three times higher ROI over
five years.5 Tear down the silos between finance, technology, and business leaders so
that together, they can build solid business cases for where AI can deliver a long-term
competitive edge.6
Venture beyond Traditional strategic partnerships focused on physical distribution of supplies and
tried-and-true products are no longer enough in the age of AI. Tech companies, startups, and other
partnerships. nontraditional partners are needed for model development, platforms, and tools.
For example, other IBM IBV research found that 65% of organizations are already
working with or planning to work with a strategic partner to build a large language
model for generative AI initiatives.7 Prioritize partners who understand your goals and
share your vision. Identify those with a proven record for integration and loop them
into your processes early. Think outside the box, imagining new partners that create
new opportunities for growth.
8
Part two
Priming the augmented workforce
AI is transforming the nature of work from the store to the factory floor,
but industry executives undervalue workforce reskilling.
AI is diffused throughout the retail and consumer products workplace. Nearly all
(96%) executives say their teams are using AI and gen AI to a moderate or significant
extent at work. When virtually everyone is using a new and powerful technology such
as AI, then virtually everyone needs training to optimize the value and understand the
risks that could damage the brands. Yet, leaders project only 31% of their workforce
will need to reskill or develop new skills over the next 12 months, with this number
climbing to just 45% in the next three years—a significant miscalculation.
Both hard and soft skills—from prompt engineering and data analytics to critical
thinking and problem solving—are essential to ushering in the age of the augmented
workforce where AI won’t replace people, but people who use AI will replace people
who don’t.8 The talent transformation is an ongoing training and education process
that must be defined and started sooner rather than later. If not, 67% of employees
have said they will leave for another employer that provides better training on new
technologies, according to an IBM IBV survey of more than 21,000 workers.9
Executives recognize the workforce will be increasingly augmented, while automation
remains crucial for rules-based tasks and repetitive work. Across 13 functional areas
from marketing and commerce to supply chain, HR, and IT, they plan to more
frequently augment than automate activities over the next 12 months (see Figure 3).
Industry leaders know that many brand-defining areas demand human intuition,
creativity, emotional intelligence, and expertise that can be complemented by AI.
For example, in product design and development, AI can accelerate idea generation
and ideation, even providing visualizations. Likewise, operational areas have vast
Leaders project only
amounts of data where decisions require human oversight, such as supply planning,
31% of their workforce
where 54% plan to augment their employees. In this activity, AI can quickly access
will need to reskill
or develop new skills and analyze a broader range of data to help the supply planner confidently resolve
over the next year. shortages in minutes, knowing important information is not missed.
9
FIGURE 3
Retail and consumer products
executives know that
automation has its place but
Figure 3
see a future of augmentation.
Retail and consumer goods executives know that
automation has its place but see a future of augmentation.
Percent of activities that will be automated, augmented, or
have no impact from AI in each area over the next 12 months
No impact Automated Augmented
Digital commerce
12% 31% 58% & B2B sales
Product design, development, and
14% 28% 57% product lifecycle management
Merchandising /
14% 32% 54% category management
Brand-defining Marketing
13% 35% 52%
areas
Customer service
15% 35% 50%
Stores
21% 32% 47%
Sustainability
11% 35% 54%
Procurement
14% 33% 54%
Supply chain operations
Business-enabling 12% 37% 51%
areas
Production
8% 43% 49% and manufacturing
HR
10% 35% 54%
Finance
12% 36% 52%
Corporate
operations IT
9% 40% 51%
Percentages represent an average of responses for a set of tasks in each
functional area, based on the question: “To what extent do you use AI
or gen AI in this activity?” Respondents replied “to a moderate extent”
or “to a significant extent.”
10
Case study
Ultimately, brands will be finding the sweet spot for Japanese retailer empowers
automation and augmentation. Take managing the
people with AI to boost profits
seasonal workforce as one example. AI-powered
automation can streamline hiring, onboarding, while reducing waste10
and scheduling processes, reducing administrative
burdens and helping control costs. Managers can use
AI-powered tools that provide real-time insights into
staffing needs, predict demand fluctuations, and
optimize schedules. Similarly, in inventory A leading retail company in Japan was grappling with
management, AI-powered sensors and cameras costly problem: food and consumer-goods waste was
automatically monitor inventory levels in real time, eating away at their profits. The client’s field staff
while providing employees with the insights needed needed data-driven insights to make more informed
to reduce the risk of stockouts or overstocking. pricing decisions.
Even areas that have a high degree of automation, For a wide variety of products and the company’s
such as customer self-service, can benefit from operations, price optimization relied more heavily
augmented employees. As executives expand use on human judgment than data, leading to variations in
of AI for personalized responses and follow-up customer forecasts, stock levels, and discount rates.
actions over the next 12 months, they say 55% These variations resulted in excessive and inadequate
of the activities will be augmented versus 30% stocking, irregular discount amounts and timings,
being automated. and large profit losses due to food waste and missed
sales opportunities.
The company worked with IBM to develop a specialized
price optimization AI system to analyze vast amounts
of data, predict customer numbers and purchase
patterns, and suggest optimal discount amounts and
timings. Now the client’s field staff can combine their
own expertise with data to improve pricing decisions.
The pricing optimization system was designed to adapt
to different product categories and sell-by durations,
making it a versatile, scalable solution that can support
Brands are finding the
a diverse product range.
sweet spot for automation
and augmentation.
11
Action guide
Prepare your workforce
to power your AI-centric brand.
AI is clearly impacting virtually the entire retail and consumer products
workforce—from the person stocking the shelves to those who sit with
you in the C-suite. It’s being built into many of the tools employees use
every day, such as AI-powered sales forecasting tools or AI-driven
design tools. Leaders need to ensure all employees are prepared
to optimize the value AI can deliver.
Connect HR, IT, Executives report leadership for reskilling efforts is divided among an AI center of
and business lines competence (31%), HR (22%), AI committees (18%), and IT (17%). This disjointed
to define reskilling approach is risky and can create confusion and frustration among employees.
strategies. Leadership from HR, IT, and the business must join forces to shape an effective
reskilling strategy. HR brings both an understanding of how to manage change and
culture along with tactical implementation expertise. IT brings the technology
knowledge, and business leaders can work directly with employees to define how AI
can augment the workforce within each business domain. Have the joint team report
directly into the C-suite and define measures to hold them accountable.
Predict every If you only expect a third of your workforce will need reskilling or upskilling over the
employee’s next few years, you aren’t thinking big enough. Just as you forecast product demand,
potential. predict what employees will need to succeed in a rapidly evolving workplace.
Look beyond just current skills to employee potential. Use AI-powered HR tools to
anticipate how an individual might develop, perform, or contribute based on skills,
talents, personality traits, experiences, and educational background.11
Share a blueprint You may not know exactly what lies ahead, but you can communicate your vision
for the workplace for the future of work. From routine business operations to brand-defining areas,
of tomorrow. AI creates anxiety as employees worry about being replaced or not having the skills
they need. Share your plans for automation versus augmentation with your workforce
and help them see how AI will create new opportunities and enable them to do their
jobs faster and better—from designing products to creating promotions to managing
inventory. Consider how employees will use—and benefit from—technology as
carefully as you consider the tech investment itself.
12
Part three
Safeguarding brand trust
With so many products vying for consumers’ attention,
AI can either bolster or undermine a brand’s trust.
Trust is paramount for both consumers and industry CEOs. Our 2024 consumer
research report showed that 9 out of 10 consumers value trust when choosing a
brand.12 Similarly, 73% of retail and consumer products CEOs in our 2024 CEO study
said trust will have a greater impact on their organization’s success than any specific
product or service.13
But AI adds new dimensions to the issue of trust, with risks impacting both
business partner and customer relationships. Consumers are already wary of AI in
general—only 53% trust the technology, falling from 61% over the past five years.14
And within the partner ecosystem, companies need to know that each member is
practicing trustworthy AI.
Retail and consumer products executives recognize that AI creates risks that can
erode trust. Nine in 10 say misuse, such as creating misleading information, is their top
worry associated with AI models, followed by privacy (85%), fairness and bias (80%),
explainability (76%), and transparency (73%). For example, biased models can
alienate customers. One consumer survey revealed that almost two-thirds of
consumers avoid AI-fueled recommendations because they are biased or
stereotypical.15
At the same time, these risks are slowing progress with generative AI opportunities.
57% of executives say data accuracy and bias is a barrier to gen AI adoption. 55% also
cite privacy and confidentiality of data and 54% are concerned about cybersecurity.
Despite these concerns, organizations are struggling to enable the tools that can help
them manage the risks. Companies have created a foundation: 87% of executives say
they have clear AI governance structures. But less than a quarter of companies have
advanced implementation of tools to assess, monitor, and manage AI governance
90% of executives
(see Figure 4). “Showing your work”—designing solutions with explainability and
cite misuse as
transparency built in—will be critical to instilling confidence in consumers regarding
their top concern
your use of AI.
with AI models.
13
FIGURE 4
Few brands have advanced implementation of
tools to help them manage their AI governance
policies and activities.
Figure 4
Few brands have robust implementation of tools to help
them manage their AI governance policies and activities.
Approach to AI governance Advanced implementation of tools
84%
have defined roles and
responsibilities for all stakeholders
involved in AI
11%
have advanced implementation of AI
accountability tools
91%
conduct ethical impact assessments
to evaluate the impact of AI
initiatives on different stakeholders
16%
have advanced implementation of AI
bias and fairness tools
87%
have established clear organizational
structures, policies, and processes
for AI governance
23%
have advanced implementation of AI
governance frameworks or policy tools
90%
build explainable models that can be
easily understood and audited
24%
have advanced implementation of AI
transparency and explainability tools
77%
conduct regular risk assessments to
identify potential security threats
26%
have advanced implementation of AI
risk and safety tools
Q. To what extent do you agree with the statements about your organization’s approach to AI governance? Percentages represent
those who agree and strongly agree. Q. To what extent has your organization implemented tools to assess, monitor, and manage
the following? Percentages represent those who responded “fully implemented, reviewed, and updated regularly.”
14
Case study
PepsiCo models a structured approach that enables Using gen AI to streamline
it to scale AI responsibly. The company began
regulatory management
by establishing a formal responsible AI framework
and assembled a dedicated team to support it. The across regions18
team then developed comprehensive policies and
standard operating procedures to operationalize
their AI principles. The governance board assesses,
validates, and approves gen AI use cases against
its responsible AI principles, sharing best practices A multibillion-dollar global consumer products
and accelerators, and helping mitigate risks. The company operates in the highly regulated agricultural
company is also building a platform that provides products industry. It devotes significant resources
comprehensive governance of models, inputs, to managing compliance with local regulations,
and outputs.16 staying current with continuously changing
guidelines, and integrating compliance into the
Regulations are also intended to support trustworthy product development process.
AI, but a lack of consistent guidelines across
jurisdictions complicates implementation and stalls To help its product compliance and development
plans. In fact, nearly half (46%) of industry CEOs said teams reduce heavy manual workloads and free
their concern about regulations as a barrier to gen AI up more time to work strategically, the company
has increased in the last six months.17 worked with IBM to develop a generative AI-powered
regulations assistant. This solution features a
However, AI can help companies manage the conversational user interface and provides a single
complexity. By automating the monitoring and source of truth for over 1,000 regulations impacting
analysis of detailed regulatory requirements, AI worldwide operations.
enables organizations to quickly identify potential
issues and take corrective action. Executives plan The regulations assistant enables product
to significantly increase their use of AI/gen AI in compliance employees to predict the impact
regulatory compliance over the next year. In product of regulatory intent, summarize regulatory
design and development, the percent increases from requirements, and compare regulations globally,
53% to 79%, for sustainability, 74% to 88%, and for faster than with manual processes. The AI tool also
financial and regulatory monitoring and reporting, enables product developers to analyze the impact
66% to 94%. of regulations on product portfolios, review
solution options, and query product specifications
in a conversational journey.
To date, the regulations assistant has demonstrated
that generative AI can orchestrate regulations data
quickly and drive closer collaboration across borders
to leverage regulatory success across the business.
In product design and The tool also has the potential to increase efficiency
by 8% to 13%, increase productivity by 10% to 15%,
development, executives
and increase profits by over $165 million during the
plan to increase use of AI next five years.
and gen AI to manage
regulatory compliance from
53% today to 79% in the
next year.
15
Action guide
Make trusted AI a brand differentiator.
Customer-obsessed businesses need to deliver on what their written
policies dictate for responsible AI practices. Build confidence in
responsible internal uses of AI before expanding to customer-facing
use cases where broken trust can damage your brand.
Purge bias from To provide transparency and explainability, define clear guidelines to monitor for
your algorithms. discriminatory patterns. For example, conduct regular audits on historical purchasing
and customer data that may reflect stereotyping and societal biases. Facilitate
human-AI collaboration and oversight with training that helps employees understand
and recognize fairness and bias. Prioritize diversity on your AI development teams.
Establish a data governance framework to support data provenance, helping ensure
your data is authentic and trustworthy. Maintain detailed records of bias mitigation
efforts, create dedicated channels for bias-related feedback, and regularly
incorporate insights into system improvements.
Leverage AI to To stay ahead of an AI regulatory environment that is evolving at varying paces
proactively navigate globally, use AI solutions to capture regulatory intent across multiple channels
regulations. and forecast its impact. Choose AI development tools that build in governance and
regulatory compliance management end to end. Proactively compare old and new
regulations to quickly identify key focus areas within impact assessments. Automate
tools to stay up-to-date and streamline audit processes.
Be open about Build trust with customers by being up-front about data collection as well
your use of AI as how and where you are using AI. Offer opt-out options and avoid tech-speak
with customers in your explanations. Exchange AI roadmaps and strategies with business partners.
and partners. Demonstrate your commitment to responsible AI practices and request the same
of your partners.
16
Authors
Dee Waddell Contributors
Global Industry Leader
Consumer, Travel & Transportation Industries The authors would like to thank the following
IBM Consulting for their contributions to this report:
[email protected]
From IBM Consulting:
linkedin.com/in/waddell/
Arnab Bag, Distribution Market HCT Service
Joe Dittmar Line Leader
Senior Partner Rich Berkman, Vice Pr |
145 | ibm | ceos-guide-to-generative-ai-second-edition.pdf | T he CEO’s
Guide to
Generative AI
What you need to know and do to win
with transformative technology
Second edition
The CEO’s
Guide to
Generative AI
What you need to know and do to win
with transformative technology
Second edition
Foreword
Contents Section one
On the edge AI-powered data and technology
1
Chapter 1 Digital product engineering 3
of reinvention
Chapter 2 IT automation 9
Chapter 3 AI model optimization 15
Chapter 4 Cost of compute 21
Just how fast can an organization evolve? Generative AI is pushing CEOs to find out. Jonathan Adashek Mohamad Ali
Chapter 5 Platforms, data, and governance 27
Senior Vice President, Senior Vice President
When the pace of change accelerates to breakneck speeds, businesses begin to strain Marketing and Communications IBM Consulting Chapter 6 Open innovation and ecosystems 33
under the pressure. Bottlenecks cause back-ups. Organizational structures buckle. Chief Communications Officer
IBM Chapter 7 Application modernization 39
Growth engines stall.
Chapter 8 Responsible AI and ethics 45
In this environment, CEOs say business model innovation is the top challenge they must
Chapter 9 Tech spend 51
overcome.1 For many, reinvention is the only option. To ensure their organizations will
achieve operational excellence no matter how hard the winds of change blow, CEOs
must be ready to rip faulty support structures down to the foundation and rebuild. Section two
AI-fueled operations
Gen AI can power this revolution. Over the next three years, executives say traditional 57
and gen AI will support business and operating model innovation by providing access
Chapter 10 Enterprise operating model 59
to additional data (88%), generating new insights from existing data (86%), expanding Kelly Chambliss James J. Kavanaugh
access to new markets (85%), and accelerating product and services development (84%).2 Senior Vice President Senior Vice President and Chapter 11 Business process automation for operations 65
IBM Consulting, Americas Chief Financial Officer
Chapter 12 Finance 71
It will supercharge people and skyrocket productivity, shifting business from a labor-based IBM
model to one that is asset-enabled. It will also open up new markets by enabling workers Chapter 13 Procurement 77
to create high-value solutions that previously weren’t feasible or affordable. Chapter 14 Risk management 83
The key is selecting use cases that drive value—and not spreading the organization too thin. Chapter 15 Physical asset management 89
Rather than looking broadly at applications and opportunities, CEOs should ask how gen AI Chapter 16 Supply chain 95
can help solve the company’s biggest problems. The leaders that win the day will be the
Chapter 17 Marketing 101
ones who stay aligned to their strategic plans and execute the fastest.
Chapter 18 Cybersecurity 107
To see how executives are making the most of this rapidly evolving technology, the IBM
Salima Lin Rob Thomas
Chapter 19 Sustainability 113
Institute for Business Value (IBM IBV) interviewed more than 10,000 CEOs and other
Managing Partner, Strategy, Senior Vice President,
members of the C-suite globally in 2023 and 2024. We asked them where they expect M&A, Transformation, Software
gen AI to make the biggest impact, how they plan to invest, and what obstacles they will and Thought Leadership Chief Commercial Officer
Section three
IBM Consulting IBM
need to overcome along the way.
AI-enabled people
119
Our findings paint the future in an auspicious light. These insights highlight a multitude of new
Chapter 20 Talent and skills 121
challenges, but also showcase strategies that can help CEOs capitalize on the gen AI moment.
Chapter 21 Customer service 127
This book combines IBM’s decades of experience working with clients to apply AI and other
Chapter 22 Customer and employee experience 133
technologies in meaningful ways with the results of our ongoing rapid-response research.
IBM’s long history of using technology to make the world work better puts us in a unique
position to help executives make gen AI work FOR them, rather than becoming something
that happens TO them. Joanne Wright Kareem Yusuf Conclusion 139
Senior Vice President, Senior Vice President,
Explore the following 22 chapters, packed full of potential applications and action items, Transformation Product Management
to learn how gen AI can redefine your customer and employee engagement strategies, and Operations and Growth
IBM Finance and IBM Software
accelerate enterprise transformation with data-driven tech, and build resilient operations
Operations
for a future defined by disruption and change.
Section one
AI-powered
data and
technology
Gone are the days when conversations about data and technology were relegated to the realm of IT.
As generative AI makes it possible for companies to deliver the integrated experiences and hyper-personalized
products and services customers demand, CEOs must understand how their technology is holding them
back—and where their data could offer a competitive edge.
As companies rush to gain a gen AI advantage, CEOs must demystify data and technology to make the most
of their finite tech spend. With the right intel, they can flow funds to the platforms, tools, and applications that
offer the greatest growth potential and retire those delivering diminished returns.
CEOs who have a good understanding of what makes gen AI tick will be best positioned to answer tough
“‘How can we use generative AI?’ is not the right question. questions from customers, regulators, and skeptics as the landscape evolves. If they can explain what data
It’s, ‘What use cases have we got that we need the most was used to train their gen AI models, how those outputs are used, and who is responsible for managing ethical
issues, CEOs will be prepared to address the challenges that are sure to come. Find out how CEOs can develop
help with and what role could different areas of technology
the gen AI expertise they need in the following chapters.
and data analytics play?’”
Bernie Hickman
CEO, Legal & General Retail
1
Chapter 1
“For me, it is all about the flow of data. Digital product engineering + generative AI
People want information. How do we deliver
that information to those customers in a way
that’s meaningful for that individual?”
Eliminate the
Paul Graham
CEO and Managing Director, Australia Post
guesswork in
product development
What do customers really want? To crack that ever-changing cipher, digital product teams must sift through
mountains of data, from market research and user surveys to device metrics, all while navigating complex code
bases and enterprise architectures. It’s a perpetual, painstaking process, and there’s no guarantee they’ll get
it right. Even when market signals and metrics seem to point to a sure-fire win, products can inexplicably flop.
Or a release flying under the radar can lead to an unexpected spike in adoption.
Generative AI helps businesses optimize the product development process—from streamlining ideation
to rapidly testing and validating features—saving money and accelerating speed-to-market. At the same time,
it frees humans to focus on solving complex engineering challenges and differentiating products through design,
UX, and UI—the creative tasks that have the biggest impact on customer loyalty and satisfaction.
Gen AI can help digital product teams hit the mark more consistently by analyzing vast stores of data faster
and more effectively than human teams ever could. Using machine learning algorithms to identify patterns
and trends in customer behavior, gen AI can quickly uncover unmet needs, suggest dozens of features
or new products that could fill a gap—and even validate these options against specific business criteria.
It also makes it possible to develop dynamic products and hyper-personalized experiences that can quickly
adapt to shifting customer demands and rapidly validate changes with customers. Given these game-changing
capabilities, it’s not surprising that 86% of executives say gen AI is now a critical part of digital product design
and development.
Research methodology
The statistics informing the insights in this chapter are sourced from a proprietary survey conducted by the IBM Institute
for Business Value in collaboration with Oxford Economics. The survey queried 450 global digital product leaders in 15
industries on their AI adoption for digital products and its impact on metrics. It was conducted from December 2023 to
2 Section 1 AI-powered data and technology 3
February 2024.
Chapter 1: Digital product engineering
1. Hyper-personalization
The three things to know
2. Ideation
and the three things to do 3. Design
IBM Institute for Business Value research has identified
three things CEOs need to know and do right now.
What you need to know What you need to do
Generative AI helps products hit Redesign product development to
1. Hyper-
personalization the high bar of hyper-personalization derive high-value product insights
at scale. from every customer interaction.
2. Ideation
What to know
Generative AI helps Imagine a world where every product is tailored to a specific Stop letting market trends catch you by surprise. Bypass
What to know 3. Design
products hit the high bar customer—where mobile devices, subscription services, the competition by cultivating proprietary data inputs and
Teams using generative AI
of hyper-personalization What to know and the Internet of Things work together to curate differentiating how you use generative AI. Continuously
can conceptualize and
at scale. experiences for an audience of one. This is the world of learn and generate the experiences, products, and content
evaluate new products Rapid code generation
hyper-personalization, and it’s no longer a distant dream. customers want—at exactly the right time.
in minutes—not days. frees teams to double
down on design. As generative AI comes of age, executives expect it to pave Think beyond cross-sell and upsell. Capitalize on
What to do the way for personalized experiences at a scale we’ve never the UX/UI potential of hyper-personalization by using
seen. By analyzing every click, swipe, and interaction, gen AI gen AI to create dynamic interfaces that adapt based
Redesign product
What to do
can stitch together bespoke product experiences for every on user behavior, preferences, and context. Customize
development to derive What to do
Build augmented teams customer. But only 30% of organizations have been able everything—search results, product designs, and even
high-value product
to prepare for an influx Upskill product to harness this power, tapping gen AI to quickly analyze pricing—to increase customer engagement and drive
insights from every
of generative AI-infused teams on experience and summarize customer feedback. Those leading the way revenue.
customer interaction.
workflows. and innovation. have an early edge: They’re 86% more likely to be creating
Invite customers to incorporate their data into
hyper-personalized experiences than their counterparts.
product experiences on their own terms. Let customers
While only a quarter of organizations are using gen AI opt in to sharing their data and clearly communicate how
to create hyper-personalized digital product experiences it will be used and protected. Use gen AI to uncover hidden
today, that figure is expected to more than double to customer preferences and use predictive analytics to
64% by the end of 2024. Using gen AI in tandem with IoT forecast what customers will want in the future.
could be a powerful way for companies to deliver true
Tap into customer data to create hyper-personalized
hyper-personalization at scale. IoT devices can feed torrents
experiences. Orchestrate disparate data, including from
of data into AI models, which may be why executives say IoT
IoT devices, to enrich the user experience. Use gen AI to
will be a top digital product disruptor, after traditional and
map your product priorities to data-driven customer pain
gen AI, over the next five years.
points. Keep your product roadmap relevant and targeted
“The speed of innovation these days is mind-boggling. Looking ahead, 70% of executives expect gen AI to improve by using gen AI to continually refine a backlog that will
the personalization of their digital product portfolio. How deliver the most business value.
There almost isn’t a week when there aren’t two
far they go—and how fast they get there—will likely decide
or three new developments in the enterprise.”
who gains a competitive edge. In the generative AI future,
products will need to be functional and personal, adapting
Amit Bendov
to meet every customer’s unique preferences, needs, and
CEO and Co-founder, Gong
expectations, no matter how rapidly they change.
4 Section 1 AI-powered data and technology 5
Chapter 1: Digital product engineering Chapter 1: Digital product engineering
1. Hyper-personalization 1. Hyper-personalization
2. Ideation 2. Ideation
3. Design 3. Design
What you need to know What you need to do What you need to know What you need to do
Teams using generative AI Build augmented teams to prepare Rapid code generation frees teams Upskill product teams on experience
can conceptualize and evaluate for an influx of generative AI-infused to double down on design. and innovation.
new products in minutes—not days. workflows.
Consumer expectations are evolving at breakneck speeds— Identify obvious time and money drains in the build and test
and product teams are racing to keep up. Tapping gen AI cycle that can be powered by generative AI. Redistribute
Generative AI has turned the traditional product design process Leverage gen AI to both ideate and rapidly validate a high
for rapid code generation can help them roll out prototypes these resources in a way that supports the development
on its head. Gone are the days of endless brainstorming and volume of ideas with customers. Focus the product team’s
faster without sacrificing the quality and design that of better UX/UI and more innovative products.
exhausting pitch sessions. Today, gen AI can use large data talent on reviewing, enhancing, and building out the ideas
customers demand.
sets to incubate ideas that have high market potential in that seem most likely to succeed in the market.
Liberate developers and designers from traditional
seconds—freeing teams to validate with customers and focus
How does it work? Gen AI speeds up the coding process, skill limitations. Encourage teams to experiment with
Treat generative AI as a team member. Embed gen AI
on the best opportunities.
letting teams test and iterate faster to increase their new training models that will boost their gen AI acumen
to create team workflows that are truly augmented.
speed-to-market—if development teams know how so they can use it creatively. Allocate dedicated research
As this technology matures, two-thirds of executives anticipate Define which inputs and outputs team members and gen AI
to use it responsibly. With the right training, governance, and development days and sponsor hackathons to give
that gen AI will inform—or even create—their product roadmap assistants are responsible for, respectively. Ask gen AI
and adoption incentives, gen AI can help teams move faster teams opportunities to enhance their skills.
by 2026. Already, nearly one-third of organizations are using to carry out discrete activities. Use it to analyze feedback,
while managing risk, freeing up resources to focus on the
gen AI for digital product idea generation. Companies that have generate design options, cut development time, or reduce
Offer more training on creativity and customer context.
creative aspects of UX and UI design.
embraced this early use case delivered a 17% revenue premium wasted effort.
Advocate for all team members to gain domain expertise
for new products and 5% greater revenue from existing product
Today, 87% of executives say their organizations sink at least in experience design. Encourage collaboration within
Reinvent the review process to lower costs and improve
enhancements in 2023.
a fair amount of effort into testing code, while 83% say the cross-functional teams to enable strategic innovation.
efficiency. Implement an idea management system to track
same for developing new features quickly in short release Provide opportunities for experimentation without fear
But the revenue boost is just the beginning. Nine in 10 a high volume of AI-generated ideas, patterns, and trends,
cycles. And they’re eager to relieve themselves of this burden. of failure.
executives already using gen AI for product idea generation including KPIs that help predict success. Streamline
say it differentiates their company by helping it respond to the process of generation, evaluation, and implementation
More than six in 10 leaders plan to use gen AI for code Expand the role of testers into user research.
market shifts faster. Going forward, they also believe gen AI of ideas.
generation in their digital products by 2025, rising to more Reskill quality assurance testers to support higher-value
will positively impact product differentiation (88%), product
than nine in 10 by 2026. But there’s a real benefit in starting activities, such as concept validation and usability testing
Augment repetitive tasks to drive down testing costs
trust (83%), and product quality (80%).
early. Only a quarter of organizations have implemented with customers.
as the pace of innovation increases. Generate and execute
gen AI for digital product code generation so far, but these
Organizations already using gen AI for product idea generation test cases based on code and product requirements to
pioneers are already seeing real results.
are building the foundation needed to augment human work: reduce the likelihood of bugs and defects in rapidly
29% more are focused on building interdisciplinary teams and evolving digital products.
They’re 35% more likely to outperform their peers in revenue
39% more are focused on governance. But executives say
growth and 48% more likely to say their teams dedicate
the skills shortage is the top constraint that could hold digital
significant effort to UX and UI design—focus areas that do
product initiatives back.
more to differentiate them from the competition. What’s more,
only 30% of executives at organizations already using gen AI
While gen AI can create product ideas at lightning speed,
for code generation say UX and UI design is a challenge,
it is humans who must review, validate, refine, and perfect
compared to 45% of those that plan to do so by 2026.
them. This means people will be more important than ever
as human-machine partnerships evolve.
6 Section 1 AI-powered data and technology 7
Chapter 2
IT automation + generative AI
Outdated technology
is dragging you down
Technical debt is back in the spotlight. It erodes profitability, drains resources, inhibits growth, and stifles
creativity. It’s an albatross CEOs carry, impeding their push to accelerate transformation with generative AI.
As a result, many CEOs find themselves mortgaging the future to survive in the present. In fact, the 2024
IBM IBV CEO study found that two-thirds of CEOs say they’re meeting short-term targets by reallocating
“Technology today as a stand-alone resources from longer-term initiatives.3
function does not make sense; technology There is a better way. CEOs can have their cake and eat it too. But how?
is there to reimagine and power the
It starts by changing how we think about IT spending. Rather than viewing IT as a cost center—an expense
business. And this requires a much businesses must eat to keep the lights on—we need to rethink how technology can better boost ROI.
closer integration and collaboration That means automating more than simple tasks that offer quick productivity boosts. Instead, leaders
must assess entire IT workflows, looking for ways to improve processes with a combination of automation
with business leaders.”
and augmentation.
Mohammed Rafee Tarafdar
It’s a big mindset shift. Today, a typical organization spends just 23% of its tech budget to drive revenue,
CTO, Infosys according to recent IBM IBV research.4 But generative AI changes the equation. Three-fourths of IT
executives say the value created from gen AI will be reallocated to new investments that drive business
innovation and growth.
This is why CEOs shouldn’t view tech upgrades as a series of isolated IT costs. They need to connect
IT automation to business strategies that will drive improved performance. And then invest accordingly.
By deliberately upgrading their IT estate with business priorities in mind—applying what we call
hybrid-by-design principles to IT programs—IBM analysis suggests that organizations can increase
ROI three-fold over five years.5
Research methodology
The statistics informing the insights in this chapter are sourced from three proprietary surveys conducted by the IBM Institute
for Business Value in collaboration with Oxford Economics. The first surveyed 207 US-based executives in 25 industries about
generative AI and IT automation in May and June 2024. The second surveyed 2,000 global executives in 10 industries about
AI and automation more broadly from April to July 2023. The third surveyed 216 US-based executives in 17 industries about
8 Section 1 AI-powered data and technology 9
generative AI and application modernization in July 2023.
Chapter 2: IT Automation
1. Innovation
The three things to know
2. Transformation
and the three things to do 3. Prediction
IBM Institute for Business Value research has identified
What you need to know What you need to do
three things CEOs need to know and do right now.
IT automation is the launchpad Break away from
1. Innovation for business innovation. the “break-fix” model.
2. Transformation
What to know Generative AI streamlines the work IT does every day, from CEOs need to focus on modernizing all aspects of the IT
IT automation software deployment to network configuration to capacity estate to enable greater automation. Empower teams
What to know 3. Prediction
is the launchpad management. These tasks are essential to keep operations to move beyond fixing what’s broken to focus on more
for business innovation. Anyone can become running smoothly—but they rarely boost the bottom line. strategic work. Ensure that IT systems are aligned with
What to know
a generative AI genius. strategic business goals and specific operational and
Generative AI automation When IT automation liberates teams from the day-to-day drudgery
financial metrics.
makes IT clairvoyant. of maintenance and support, they’re freed to envision a future built
What to do on new transformative technologies—including, of course, Automate to make hard work easier. Identify the
What to do generative AI. Gen AI also fuels their creative fire, sparking ideas systems, applications, and data flows that must be
Break away from
Make tech for new digital products and revenue streams. And most integrated to streamline and automate work. Give IT teams
the “break-fix” model.
What to do
less techy. companies have hit the ground running. access to a generative AI platform and tools they can use
Conquer complexity to quickly create the code and APIs needed to connect
Today, 62% of IT executives say their organizations are using gen
with intelligent visibility. disparate systems. Encourage teams to identify new ways
AI for code generation—and that figure will jump to 87% by 2026.
to automate and augment routine tasks.
65% of tech leaders expect gen AI solutions to automatically
resolve IT issues with little to no human intervention. And 82% Get more out of every IT automation dollar. Align tech
of IT executives expect generative AI to improve DevSecOps, spend with business objectives—and fast track initiatives
the automated workflows that incorporate security practices that accelerate performance improvement. Go beyond
throughout the development lifecycle, over the next two years. finding efficiencies to invest in tech that will create new
revenue streams and promote rapid growth.
Organizations that see automation as essential to fast-track
gen AI capabilities are already gaining an edge. They outperform Measure what matters. Establish a feedback loop
in workforce agility, profitability and efficiency, innovation, and to continually monitor and improve gen AI model
revenue growth—demonstrating how AI-powered automation performance. Look past traditional IT metrics, such
can transform IT into a business incubator and foster an as uptime and downtime, to gauge success. Instead,
entrepreneurial culture. tie automation efforts to business-centric metrics,
such as user satisfaction, revenue growth, and
By giving everyone access to generative AI tools and expertise,
“Our objective is not to reduce workforce. speed-to-market.
IT democratizes innovation, empowering employees to develop
We just want to let people spend their time
their own ideas to unlock business value—and predict which are
more productively and more creatively, most likely to succeed.
so that they can also be happier.” Gen AI not only drives growth but also attracts and retains top
talent, who are drawn to organizations that prioritize creativity and
Hiroshi Okuyama
autonomy. And if IT leaders funnel this curiosity into a shared,
Director and Member of the Board, Chief Digital Officer, Yanmar Holdings Co., Ltd.
collaborative platform, they can feed a vibrant innovation pipeline
that can help the organization meet ambitious growth targets
quarter after quarter.
10 Section 1 AI-powered data and technology 11
Chapter 2: IT Automation Chapter 2: IT Automation
1. Innovation 1. Innovation
2. Transformation 2. Transformation
3. Prediction 3. Prediction
What you need to know What you need to do What you need to know What you need to do
Anyone can become Make tech less techy. Generative AI automation makes Conquer complexity with
a generative AI genius. IT clairvoyant. intelligent visibility.
Embed IT in the boardroom. Make technology
and automation central to every business strategy—and
Employees don’t need to be IT experts to transform business AI systems already help IT teams accurately predict and Use gen AI-enabled digital twins to model the effects
challenge leaders to connect performance metrics to the
with technology. But they do need IT experts to provide the prevent system failures and bottlenecks. But with gen AI, of specific disruptions across the enterprise and the
systems, platforms, and tools that enable their success.
tools and platforms that put the power of gen AI automation businesses see even farther into the future. ecosystem. Improve ROI with more accurate estimates
at their fingertips. of how much investments in technology and automation
Assemble multi-disciplinary dream teams. Build squads
By deploying gen AI and AIOps in tandem, teams gain
will cost—and how much value they will deliver.
of people with diverse skills and backgrounds, including
If IT provides the right low-code and no-code platforms, intelligence that lets them anticipate and prepare for
data scientists, engineers, domain experts, and business
anyone can create or modernize web and mobile apps—a scenarios that might otherwise catch them by surprise. Hunt for treasure across your IT estate. Provide operations
stakeholders, to collaborate on gen AI projects. Organize
process that, until recently, required a team of developers. For instance, by automatically identifying and mapping visibility into applications and infrastructure by using gen AI
workshops, hackathons, and other competitions that spur
At the same time, gen AI code assistants let developers relationships across the IT estate—a process known as to uncover the relationships that are key to building
innovative thinking and knowledge sharing.
quickly translate code from one language to another, topology discovery—teams can quickly spot dependencies resilience and driving growth. Discover hidden riches
reducing the need for some hard-to-find technical skills. between different systems and components. by modeling different improvements—and investing
Empower DIY developers to get creative with automation.
in the IT automation solutions that promise the best returns.
Evaluate and select a low-code or no-code platform that
While IT must be the catalyst of this transformation, This process reveals how problems in one area can cascade
aligns with the enterprise’s technology stack and gen AI
the benefits will extend across the business: 81% of executives across the business—and lets IT limit the domino effect. Head off hazards at the pass. Get out in front of risks
platform. Establish guidelines for data management,
say gen AI will fundamentally change how people do their jobs. It also makes it easier to optimize network performance, by automating the process of predicting how different
security, and compliance—then push people to explore
And IT executives are up for the challenge, with 70% saying their strengthen security, and keep teams across the organization scenarios could influence complex systems. Use gen AI
what they’re capable of.
organizations will design AI systems to seamlessly collaborate in lockstep. to simulate potential outcomes and validate crisis response
with humans by 2026. plans, then forge confidently into unexplored frontiers.
Challenge cultural norms and let digital natives drive
IT leaders can also use gen AI to supercharge simulations.
change. Flatten hierarchical decision-making to give younger
To produce the best results, they’ll need to bring employees Gen AI-enabled digital twins can model multiple dimensions Right-size technology spend with IT automation—then
team members a stronger voice. Launch reverse mentorship
along for the ride. Technology can be intimidating for simultaneously, letting teams test response strategies more right-size your team. Broaden FinOps capabilities
programs that pair people just entering the workforce with
non-technical teams, but training and reskilling can demystify effectively. Rather than wondering how well their plans will to provide visibility into costs and spending across all AI,
senior leaders. Give them space to ask why. And why not.
gen AI and encourage people to try something new. And work, they can see them in action. hybrid cloud, and application modernization investments.
providing this support is more important than ever. In 2024, Optimize, automate, and augment IT operations to avoid
Gen AI also helps IT more confidently estimate the business
global CEOs said 35% of their workforce would require the financial and environmental costs of overprovisioning.
value of different IT automation investments. Today, 57%
retraining and reskilling over the next three years—up from Realign your tech team to shed expensive talent that
of IT executives are already using generative AI to predict
just 6% in 2021.6 you no longer need.
outcomes, efficiency gains, and ROI in IT and network
For years we’ve been saying that IT needs to work more closely automation initiatives—and this figure will grow to 75%
with the business—and the business needs to work more closely by 2026. This level of visibility can help manage the cost side
with IT. Gen AI could finally make this a reality: 68% of executives of the equation, as well: 76% of IT executives say they will
say it will bridge the gap between IT and the business. By use gen AI to enhance FinOps practices for more precise
providing a shared canvas for collaboration, gen AI helps IT control of cloud costs.
develop a deeper understanding of business problems and
business teams harness the full power of technical solutions.
12 Section 1 AI-powered data and technology 13
“We have more than 40 proprietary Chapter 3
AI models that we train and fine-tune
AI model optimization + generative AI
for revenue teams with our customer
interaction data. The results are more
There’s a gen AI
accurate and meaningful.”
Amit Bendov
CEO and Co-founder, Gong
model for that
ChatGPT made everyone feel like an AI expert. But its simplicity is deceptive. It masks the complexity
of the generative AI landscape that CEOs must consider when building their AI model portfolio.
Gen AI models come in many flavors. What they can do, how well they work—and how much they cost—varies
widely. Who owns the model, how it was developed, and the size of its training dataset are just a few of the
variables that influence when and how different models should be used.
With the massive amount of data and resources it takes to train a single large language model (LLM), the question
of size is monopolizing many conversations about gen AI. As a result, many CEOs wonder whether they should
scale large gen AI models for their business. Or if they should dev |
146 | ibm | why-invest-in-ai-ethics-and-governance.pdf | IBM Institute for Business Value | Research Insights
Why invest in AI ethics
and governance?
Five real-world origin stories
In collaboration with the Notre Dame—IBM Tech Ethics Lab
How IBM
can help
Clients can realize the potential of AI, analytics, and
data using IBM’s deep industry, functional, and
technical expertise; enterprise-grade technology
solutions; and science-based research innovations.
For more information:
AI services from IBM Consulting
ibm.com/services/artificial-intelligence
AI solutions from IBM Software
ibm.com/Watson
AI innovations from IBM Research®
research.ibm.com/artificial-intelligence
The Notre Dame-IBM Tech Ethics Lab
techethicslab.nd.edu/
2
Key
takeaways
Organizations that measure Embracing AI ethics is essential.
the value of AI ethics
It’s not just about loss aversion. 75% of executives
could be a step ahead. view AI ethics as an important source of competitive
differentiation.1 More than 85% of surveyed consumers,
Our holistic AI ethics
citizens, and employees value AI ethics.2
framework considers
three types of ROI. Longer-term, proactive AI ethics
strategies can generate value across
the organization.
A majority of companies (54%) expect AI ethics to be
very important strategically,3 with executives citing
involvement of 20 different business functions.4
Investing in AI ethics has the
potential to create quantifiable value.
Organizations that measure the value of AI ethics could
be a step ahead. Our holistic AI ethics framework
considers three types of ROI: economic impact
(tangible), reputational impact (intangible), and
capabilities (real options ROI).
1
Introduction
Generative AI is revolutionizing industries, but its dizzying ascendance
has also raised significant ethical concerns. Balancing the potential
benefits with ethical and regulatory implications is crucial.
But it’s not easy. In IBM Institute for Business Value (IBM IBV) research, 80%
of business leaders see AI explainability, ethics, bias, or trust as major roadblocks
to generative AI adoption.5 And half say their organization lacks the governance
and structures needed to manage generative AI’s ethical challenges.6
In the face of this uncertainty and risk, many CEOs are hitting pause. More than half
(56%) are delaying major investments in generative AI until they have clarity on AI
standards and regulations,7 and 72% of executives say their organizations will actually
forgo generative AI benefits due to ethical concerns.8
Yet there is a path forward—if executives broaden their outlooks and view AI ethics
as an opportunity. Even better: ongoing research suggests that investing in AI ethics
has the potential to create quantifiable benefits.
In order to unlock this potential, organizations need to embrace a new perspective
as they evaluate the ROI of AI ethics investments. In part one of this report, we
identify three key types of ROI that apply to AI ethics—in other words, a holistic AI
ethics framework. In part two and part three, we explore two distinct but valuable
ways to justify AI ethics investments right now. (We plan to build on this work by
conducting additional research in 2025 that explores quantification in greater depth.)
Finally, we offer an action guide for bringing the holistic AI ethics framework to life
inside the organization.
We also include stories from five executives on the front lines of AI ethics,
as part of an ongoing collaborative project among the IBM IBV, the Notre Dame—IBM
Tech Ethics Lab, the IBM AI Ethics Board, and the IBM Office of Privacy and
Responsible Technology. Some interviews were conducted in collaboration with
Oxford Economics.
2
Part one
Exploring a holistic
AI ethics framework9
AI ethics and governance investments can span broadly
across the enterprise, from an AI ethics board to an
ethics-by-design methodology, from an integrated
governance program to training programs covering AI
ethics and governance, among many other endeavors.10
(See “AI ethics: Stories from the front lines” on page 13.
Also refer to our IBM IBV study The enterprise guide
to AI governance at ibm.co/ai-governance.)
So how do organizations begin measuring the impact of such
initiatives? We developed a holistic AI ethics framework to meet
this need, validating it through an extensive series of conversations
with over 30 organizations. This approach can help organizations
understand the value of their AI ethics and governance investments.
Traditionally, investments are justified by calculating ROI in financial
terms alone. AI ethics investments are more challenging to evaluate,
providing both tangible and intangible benefits as well as helping
build longer-term capabilities.
“Our work has to not just contribute to the mission of the organization—
it also has to contribute to the profit margin of the organization,”
notes Reggie Townsend, VP of the Data Ethics Practice at SAS.
“Otherwise, it comes across as a charity, and charity doesn’t get
funded for very long.”
We developed a holistic AI ethics
framework, validating it through
an extensive series of conversations
with over 30 organizations.
3
A holistic AI ethics framework identifies three
types of ROI that organizations should consider
with AI ethics investments.
Economic impact (tangible ROI)
refers to the direct financial benefits
of AI ethics investments, such as cost
savings, increased revenue, or reduced
cost of capital. For example, an
organization might avoid regulatory
fines by investing in AI risk management.
Reputational impact (intangible ROI)
can involve important yet
difficult-to-quantify elements, such
as an organization’s brand and culture
that support positive returns or impact
on an organization’s reputations with
shareholders, governments, employees,
and customers. Examples include
improved environmental, social,
and governance (ESG) scores;
increased employee retention;
and positive media coverage.
Capabilities (real options ROI)
alludes to the long-term benefits
of building capabilities that, established
first for AI ethics, can disseminate
broader value throughout an organization.
For example, technical infrastructure
or specific platforms for ethics may allow
organizations to modernize in ways
that lead to further cost savings
and innovation.
Source: “The Return on Investment in AI Ethics:
A Holistic Framework.” Proceedings of the 57th
Annual HICSS Conference on Systems Sciences.
January 2024.
4
The holistic AI ethics framework depicted above
describes three paths to understanding the impact
of investments in AI ethics with regards to
stakeholders: the direct path through economic
return, and indirect paths through capabilities
and reputation. This framework encompasses
and describes the relationships, stakeholders,
and potential returns that exist when organizations
make investments in AI ethics.11
At a high level, how might this approach work in
practice? Consider the investment in an AI Ethics
Board infrastructure and staff. This investment helps
prevent regulatory fines (tangible impact); increases
client trust, partner endorsements, and business
opportunities (intangible impact); and helps enable
the development of management system tooling that
improves automated documentation and data
management (capabilities). The holistic AI ethics
framework illustrates how AI ethics is interwoven
throughout an organization, both in terms of practices
and outcomes.
The holistic AI ethics
framework illustrates how
AI ethics is interwoven
throughout an organization,
both in terms of practices
and outcomes.
5
Part two
The value of “loss aversion”
What is AI ethics? A senior vice president with responsibility
for data policy at Fidelity Investments puts it succinctly:
“It’s using AI technology in a responsible form to be able
to distinguish between right and wrong as we communicate
with our customers, prospects, and other clients.”
In recent IBM IBV research, 72% of executives said they’ll step back from
generative AI initiatives if they think the benefits might come at an ethical
cost. These same organizations are 27% more likely to outperform on
revenue growth—a correlation that is hard to ignore.12
Yet noble AI intentions are often talked about more than they are acted
on. While over half of organizations in our research have publicly endorsed
principles of AI ethics, less than a quarter have operationalized them.13
Fewer than 20% strongly agree that their organizations’ actions
and practices on AI ethics match (or exceed) their stated principles
and values.14
“It’s all good to want to do it, but you need to actually do it,” says a senior
leader responsible for AI governance at a global financial services firm.
“But to do it, you need resources, which requires funding. More important
than that, you need the will of senior executives.”
So, what is the business justification for investing in AI ethics? It often
starts with a loss aversion approach: avoiding costs associated with
regulatory compliance or retaining revenue that might be lost if
customers move their business to enterprises that prioritize AI ethics.
Noble AI intentions are often talked
about more than they are acted on.
6
The fact that these motivations reflect a short-term A prod from AI regulators
strategy does not detract from their significance.15
Loss aversion generates near-immediate results.
AI regulations are a catalyst for action. The EU
As the senior leader responsible for AI governance at
AI Act is the first comprehensive AI regulation by
a global financial services firm notes, “The business
a major entity. One strategy manager at Deutsche
case is all about decreasing reputational risk.”
Telekom says, “The EU AI Act could change the face
of AI ethics globally. If, for instance, an American
company is working with us, they also have to comply
with the EU AI Act.”
The EU’s effort is only the beginning. Organizations
Examples of loss aversion include:16
such as the Partnership on AI, the Global Partnership
on AI, the World Economic Forum, the United Nations,
and the Organisation for Economic Co-operation and
Regulatory justifications
Development (OECD) have all published principles
Avoid a regulatory fine. and guidelines on a responsible approach to AI.17
In a survey by the Centre for the Governance of AI
Avoid legal costs. of over 13,000 people across 11 countries, 91%
agreed that AI needs to be carefully managed.18
Implement required technical
compliance mechanism.
Given this emphasis on regulations, oversight,
and responsible approaches to AI, a focus on loss
Enable business
aversion isn’t just sensible but necessary.
for required compliance.
Customer/partner/
competitor justifications
Allay stakeholder concerns.
Avoid threat to business model.
Meet specific customer request or need.
Protect brand reputation.
Keep pace with competitors.
7
Part three
Leveraging AI ethics
to generate value
The benefits of investments in AI ethics aren’t exclusive
to cost avoidance or damage control. They also help to build
useful capabilities and tangible innovations that can enable
an organization’s long-term strategies.19 Such value
generation can be more indirect than loss aversion and
requires an expanded view of ROI. It also won’t happen
overnight and can take time to see measurable outcomes.
But organizations that are sophisticated about their understanding of AI
ethics can use the investments to:20
– Enable long-term plans to scale AI responsibly.
– Build unique and valuable organizational capabilities that can lead
to differentiation.
– Improve employee efficiency or productivity.
– Align with values to advance as an industry leader.
– Seize a market opportunity.
– Protect vulnerable individuals and communities.
– Increase customer satisfaction.
– Demonstrate trustworthiness and maturity.
– Support Environmental, Social, and Governance (ESG) efforts.
– Increase ability to manage risk over the long term.
– Innovate for a competitive advantage.
As AI technology matures, organizations can not only integrate AI into their
operations, they can repurpose that technology toward new innovations. A
senior director from a leading health and consumer goods retailer explains,
“Based on the measures we took from the AI standpoint to create and
enrich the customer experience, we have seen returns in terms of adoption
of those brands, sales growth, customer retention, and customer growth.”
8
Combining the best
of both worlds
Organizations that embrace a holistic approach
that encompasses both loss aversion and value
generation will be more efficient, effective, and
successful—as well as more ethical.
Reactive Proactive
Loss aversion Value generation
Regulatory compliance justifications Create technologies, infrastructures,
and platforms that can support AI ethics
efforts and be repurposed
Avoid a regulatory fine.
Avoid legal costs. Enable long-term plans to scale
AI responsibly.
Implement required technical
compliance mechanism. Build unique and valuable
organizational capabilities that
Enable business for required compliance.
can lead to differentiation.
Improve employee efficiency
or productivity.
Justifications relating to clients, Align with values to advance
partners, and competitors as an industry leader.
Seize a market opportunity.
Allay stakeholder concerns.
Protect vulnerable individuals
Avoid threat to business model.
and communities.
Meet specific customer request or need.
Increase customer satisfaction.
Protect brand reputation.
Demonstrate trustworthiness
and maturity.
Keep pace with competitors.
Support Environmental, Social,
and Governance (ESG) efforts.
Increase ability to manage risk over
the long term.
Innovate for a competitive advantage.
Source: “On the ROI of AI Ethics and Governance Investments: From Loss Aversion
to Value Generation.” California Management Review. July 29, 2024.
9
The senior vice president of Fidelity Investments observes:
“What companies don’t realize is that up-front investment actually
pays significant ROI, not just in terms of ethics, but from a total cost of
implementation on any of your use cases. Because if you don’t lay that
foundation, you spend a lot more money with everybody implementing
one pillar at a time and not benefiting from any reuse.”
A preliminary step to this evolution, of course, is to actually develop
AI use cases that align with and support organizational strategy.
Notes the strategy manager at Deutsche Telekom, “Either you could
create AI solutions for the customer, or you could create AI solutions
for your internal infrastructure.”
Out of the starting block, it’s instinctive and reasonable to adopt
a “defensive” loss-aversion posture to avoid the pitfalls we’ve
described, such as regulatory fines, legal costs, and reputational risks.
But fertile ground can be found in the pivot to value generation.
Organizations need to create technologies, infrastructures,
and platforms with the versatility to support AI ethics initiatives
and to fuel broader corporate innovation.
Procuring support and budget for these strategies can be tricky.
To persuade skeptics and surmount obstacles, organizations should
clearly pinpoint potential value generated, including metrics of economic
returns. This can be done through a process of identifying relevant loss
aversion and value generation justifications as the organization plans
and then evaluates potential investments21—essentially, using the
holistic AI ethics framework.
Fertile ground can be found in the pivot
to value generation. Organizations need
to create technologies, infrastructures,
and platforms with the versatility to
support AI ethics initiatives and to fuel
broader corporate innovation.
10
Action guide
How to reap the rewards
of AI ethics investments
Investing in AI ethics is not just the right thing to do, it can also be a sound
business decision. By using the holistic AI ethics framework, organizations can
make informed choices about allocating resources to AI ethics, helping boost
the trustworthiness and potential of AI programs overall.
According to IBM IBV research, 75% of executives view ethics as an important
source of competitive differentiation.22 A study from the Economist
Intelligence Unit echoes those results, pointing to a competitive edge through
product quality, talent acquisition and retention, and new revenue sources.23
These studies underscore the criticality of a proactive approach to AI ethics.
Organizations must consider how governance of AI differs from that of
previous technologies, permeating every corner of their culture, ecosystem,
and customer engagement.
“You educate the AI engine based on what humans are thinking,” says the
senior director at a leading health and consumer goods retailer, “because
they are the better judge from an ethics standpoint.”
Along those lines, Reggie Townsend of SAS observes: “We have a diverse set
of folks who have come from a variety of different backgrounds and life
experiences. We do hard work, but we do heart work. I don’t hire anyone who
doesn’t have a heart for what we’re doing. We have passionate people on our
team, and we bring that passion to the work. That’s fundamentally important.”
11
Here’s our five-step guide for optimizing
your AI ethics investments
1
Engage your savviest AI ethics experts to educate the C-suite on
differences between loss aversion and value generation approaches
to AI ethics. Help executives envision the potential of leveraging AI
ethics technology, platforms, and infrastructure for broader use.
2
Identify specific value generation justifications for AI ethics and
governance that may apply to the AI use cases at hand. Examples
include the ability to responsibly improve the answers to customers
and increased employee productivity and job satisfaction.
3
Think through the anticipated stakeholder impacts of the AI
use case and identifying potential indicators. These include:
– Direct economic returns (for example, the value
of an expanded customer base)
– Intangible reputational returns (for example, earned media
value of customer reviews)
– Capabilities and knowledge returns from real options
(for example, improved customer response quality that leads
to more first-contact resolutions).
4
Create an AI ethics implementation strategy that can deliver on value
generation justifications. Using the analysis in action 3, identify the
potential returns holistically. Doing so can help optimize the potential
returns on your investments in AI ethics and governance while
simultaneously benefitting stakeholders, ecosystems, and society.
5
Turn value generation into a competitive advantage. Focusing on value
generation can provide a competitive advantage in an environment
where regulatory compliance is business as usual.
For additional information and actions on the holistic AI ethics framework, refer to “On the ROI
of AI Ethics and Governance Investments: From Loss Aversion to Value Generation,” California
Management Review, at https://cmr.berkeley.edu/2024/07/on-the-roi-of-ai-ethics-and-
governance-investments-from-loss-aversion-to-value-generation/ and “The Return on
Investment in AI Ethics: A Holistic Framework” at https://arxiv.org/abs/2309.13057.
12
AI ethics
Stories from the front lines
Deutsche
Deutsche Telekom’s data initiatives are closely tied with monetizing data through AI
Telekom applications and monitoring the EU AI Act. One strategy manager at the company leads
a team that is involved in virtually every AI conversation in the organization and is
therefore able to provide a holistic overview of the company’s approach to AI ethics.
Preparing for the EU
AI Act with internal Deutsche Telekom has created a team of high-level executives responsible for
governance and evaluating current and future AI initiatives—in effect, an organized governance group.
education The group’s most important purpose is to help ensure that the company complies
with data privacy and security procedures both internally and externally—including
the EU AI Act.
“AI is all about data. It’s a fundamental element of any AI product,” says the manager,
adding that he regards data as a crucial component of the AI ethics approach as well.
Before incorporating any data into its products, the organization considers who is
exposed to the data and how customer data is protected. Beyond their customers,
Deutsche Telekom also must protect certain data segments in terms of sustainability
and energy practices.
Critically, Deutsche Telekom heavily invests in educating employees about AI and
its ethical use, often in the form of internal workshops, including training related
to the EU AI Act.
“Training colleagues is definitely a return on investment because it reduces the time
to market and we come up with more innovative products,” he says. And with its
continual efforts to improve, Deutsche Telekom experiences greater innovation
and enhanced customer trust.
13
AI ethics: Stories from the front lines
Fidelity
Responsible AI initiatives are embedded into each phase of AI use cases at
Fidelity Investments, beginning with robust data management practices
Investments
and feeding into a dynamic review process driven by the company’s AI
Center of Excellence. The financial services firm invested heavily in these
Reaping ROI through
initiatives to make AI ethics one of its foundational pillars—rather than a
repurposed use case
compliance box-checking exercise.
implementation
Each business line at Fidelity has a dedicated team for AI use case
development and vendor management. This work is guided by the
expertise of external consultants and actively monitored by the firm’s
compliance and risk officers, who receive specialized AI training.
The AI Center of Excellence is involved in each step of this process, from
vendor selection to model evaluation. It resides in Fidelity’s data function
and includes representation from each business unit at the firm, with roles
ranging from risk compliance and audit to legal and even information
security. This process also allows Fidelity to confidently answer clients’
increasing demands for information on its AI use and governance.
Resistance to responsible AI initiatives is inevitable, as they can delay
projects or limit use cases. “You have to explain that the reason controls
are so important is not just some random compliance policy, but that
there are implications to the firm if we get this wrong,” says a senior vice
president with responsibility for data policy at the firm. Fidelity has
been able to minimize pushback by framing these initiatives as
integral to the success of AI projects and by streamlining the overall
governance process.
“You have to explain that the reason controls
are so important is not just some random
compliance policy, but that there are
implications to the firm if we get this wrong.”
A senior vice president with responsibility for data policy at Fidelity Investments.
14
AI ethics: Stories from the front lines
SAS Reggie Townsend, VP of the Data Ethics Practice at SAS, leads a team tasked
with coordinating responsible innovation principles, operational workflows and
Ensuring an AI-driven governance structures across a global organization. It all began with questions
future that is built and investigating.
for all of us
Prompted by risks to vulnerable populations and the increasing sophistication of
AI, Townsend and close colleagues began digging deeper into responsible AI
and data ethics at SAS. They were empowered by SAS leadership to formalize
the company’s longtime commitment to responsible innovation. Consequently,
SAS created the Data Ethics Practice (DEP).
With a philosophy of “ethical by design,” the DEP guides the company’s efforts
to help employees and customers deploy data-driven systems that promote
human well-being, agency, and fairness. This approach compels individuals
to answer three basic questions:
– For what purpose?
– To what end?
– For whom might it fail?
The team helps build Trustworthy AI capabilities and workflows to help
customers and developers pursue their responsible AI goals. AI governance
advisory services from the DEP are helping customers put AI into action
responsibly. The DEP also provides critical counsel to employees on product
development, marketing, and more.
When Townsend’s role and team were created, the hope was their work would
bolster trustworthiness of products, processes, and people. This, in turn,
would enhance the brand’s reputation as a trusted AI leader.
Profits are important, of course. But according to Townsend, his team’s
guiding principle is that wherever SAS software shows up, it does no harm.
“Sometimes,” Townsend observes, “you just have to take action because
it’s the right action to take.”
“Sometimes, you just have to take
action because it’s the right action
to take.”
Reggie Townsend
Director and VP, Data Ethics branch, SAS
15
AI ethics: Stories from the front lines
Global financial For one senior leader responsible for AI governance at a global financial services firm,
AI development and ethics starts with education. He advocates hosting workshops
services firm
that discuss ethical principles and values—empowering leadership to understand
trade-offs. “We need to talk about AI in a way that interests leadership, not just in
Justifying positive
processes and procedures,” he observes.
returns with a lowered
reputational risk In discussing how to measure the return on investments in AI ethics, the senior leader
offers the “creepy line” metaphor. Often, organizations find themselves in situations in
which they are doing something perfectly legal that is highly profitable, yet still feel
uncertain about the ethicality of their actions—a sense of crossing the “creepy line.”
In such situations, he says that organizations must examine the activity through the
lens of both current and future generations, in conjunction with all comprehensive
ethical considerations. As long as these considerations are covered satisfactorily,
the organization should feel reassured that the “creepy line” is not breached.
He also notes, “Reputational risk is a key factor in justifying positive returns.
We aim to decrease reputational risk while applying data and AI ethics principles.”
For example, his team conducted an ethical fairness review of loan pricing involving a
credit scoring algorithm. In conducting this review, the team analyzed all 165 features
of the model, asking if there were any potential causal mechanisms for why that
particular data feature may correlate with an individual’s ability to pay back a loan.
Ultimately, three data features were removed because a causal link did not exist,
thus avoiding the lack of fairness in using this AI technology.
“We need to talk about AI in a way
that interests leadership, not just in
processes and procedures.”
Senior leader responsible for AI governance at a global financial
services firm
16
AI ethics: Stories from the front lines
A leading health
A senior director at this organization instituted an AI initiative to provide solutions
via vendors and internal products. A recent conversation with him covered three main
and consumer
operational areas.
goods retailer
A rigorous governance process. The retailer’s AI governance group is a centralized
body that helps ensure all AI initiatives fulfill their required steps for approval. In that
Driving success with a
vein, it conducts sessions in which project teams present how they’ve aligned their
thorough AI ethics and
compliance measures with the group’s control plan. If approved, the projects move
governance strategy
forward. The director notes that, as a sizeable enterprise engaging with large numbers
of partners, suppliers, customers, and other ecosystems, it must be extremely careful
in building their AI capabilities.
The AI ethics engine. Whether the retailer invests in SaaS-, vendor-, or open-source-
based products, they ensure all ethical parameters are met prior to deployment.
Its internal audit process is referred to as “the AI ethics engine.” In engaging a vendor,
the organization first conducts a background check, looking at the health of its
industry, clients, reputation, and capabilities. This process can span two to four
months. Once the retailer picks its vendor, it engages in a pilot. If success and ethics
measures are met, the partnership proceeds.
Stakeholder success. The organization has heavily invested in AI capabilities
to enhance the customer engagement experience and drive market strategies
and customer growth. The director notes, “AI by itself or a human by itself cannot
be successful, but if you combine those two together, the outcome is successful
and accurate.”
At this particular retailer, AI capabilities implemented in customer service, for example,
will not replace customer service employees. Rather, the organization invests in
providing these employees with additional skills, resulting in employee retention.
This approach can create benefits for the customers, employees, and company’s
economic returns.
“AI by itself or a human by itself cannot
be successful, but if you combine those
two together, the outcome is successful
and accurate.”
Senior director at a leading health and consumer goods retailer
17
Authors
Nicholas Berente Marianna Ganapini
Senior Associate Dean for Academic Programs Associate Professor, Philosophy
Professor of IT, Analytics and Operations Union College
University of Notre Dame, Mendoza College of Business linkedin.com/in/marianna-b-ganapini-769624116/
linkedin.com/in/berente/ [email protected]
[email protected]
Brian Goehring
Marialena Bevilacqua Associate Partner, AI Research Lead
PhD Student in Analytics IBM Institute for Business Value
University of Notre Dame, Mendoza College of Business linkedin.com/in/brian-c-goehring-9b5a453/
linkedin.com/in/marialena-bevilacqua-6848b9132/ [email protected]
[email protected]
Francesca Rossi
Heather Domin IBM Fellow and AI Ethics Global Leader
Global Leader, Responsible AI Initiatives, IBM IBM Research
Associate Director, Notre Dame—IBM Tech Ethics Lab linkedin.com/in/francesca-rossi-34b8b95/
linkedin.com/in/heatherdomin/ [email protected]
[email protected]
Contributors
Sara Aboulhosn, Angela Finley, Rachna Handa,
Jungmin Lee, Stephanie Meier, and Lucy Sieger
18
About Research Insights The right partner for
a changing world
Research Insights are fact-based strategic insights
for business executives on critical public- and At IBM, we collaborate with our clients, bringing
private-sector issues. They are based on findings together business insight, advanced research, and
from analysis of our own primary research studies. technology to give them a distinct advantage in
For more information, contact the IBM Institute for today’s rapidly changing environment.
Business Value at [email protected].
IBM Institute for Related reports
Business Value
The enterprise guide to AI governance
IBM Institute for Business Value. October 2024.
For two decades, the IBM Institute for Business Value
ibm.co/ai-governance
has served as the thought leadership think tank for
IBM. What inspires us is producing research-backed,
The CEO’s guide to generative AI:
technology-informed strategic insights that help
Responsible AI & ethics
leaders make smarter business decisions.
IBM Institute for Business Value. October 2023.
From our unique position at the intersection ibm.co/ceo-generative-ai-responsible-ai-ethics
of business, technology, and society, we survey,
interview, and engage with thousands of executives, AI ethics in action
consumers, and experts each year, synthesizing IBM Institute for Business Value. April 2022.
their perspectives into credible, inspiring, and ibm.co/ai-ethics-action
actionable insights.
To stay connected and informed, sign up to receive
IBV’s email newsletter at ibm.com/ibv. You can also
find us on LinkedIn at https://ibm.co/ibv-linkedin.
19
Notes and
sources
1 Goehring, Brian, Francesca Rossi, and Beth Rudden. 11 Bevilacqua, Marialena, Nicholas Berente, Heather
AI ethics in action: An enterprise guide to progressing Domin, Brian Goehring, and Francesca Rossi.
trustworthy AI. IBM Institute for Business Value. April “The Return on Investment in AI Ethics: A Holistic
2022. https://ibm.co/ai-ethics-action Framework.” Proceedings of the 57th Annual HICSS
Conference on Systems Sciences. January
2 Ibid. 2024. https://arxiv.org/abs/2309.13057. “OECD AI
Principles overview.” OECD. Accessed November 15,
3 Ibid. 2024. https://oecd.ai/en/ai-principles
4 Ibid. 12 The CEO’s guide to generative AI: Customer and
employee experience. IBM Institute for Business Value.
5 2023 Institute for Business Value generative AI state of
August 2023. https://www.ibm.com/thought-
the market survey. 369 global CxOs. April/May 2023.
leadership/institute-business-value/en-us/report/
Unpublished information.
ceo-generative- |
147 | ibm | IBM-AI-POV_FINAL2.pdf | Precision Regulation for Artificial Intelligence
By Ryan Hagemann, IBM Policy Lab co-Director (Washington, DC) & Jean-Marc Leclerc, IBM Policy Lab co-Director (Brussels)
Among companies building and deploying artificial treated fairly and equitably by AI-based determinations in
intelligence, and the consumers making use of this sensitive use-cases.
technology, trust is of paramount importance. Companies
want the comfort of knowing how their AI systems are That is why today we are calling for precision regulation
making determinations, and that they are in compliance of AI. We support targeted policies that would increase
with any relevant regulations, and consumers want to the responsibilities for companies to develop and operate
know when the technology is being used and how (or trustworthy AI. Given the ubiquity of AI — it touches all of
whether) it will impact their lives. us in our daily lives and work — there will be no one-size-
fits-all rules that can properly accommodate the many
62% of Americans and 70% Europeans unique characteristics of every industry making use of
this technology and its impact on individuals. But we can
prefer a precision regulation approach for
define an appropriate risk-based AI governance policy
technology, with less than 10% in either
framework based on three pillars:
region supporting broad regulation of tech.
85% of Europeans and 81% of Americans
• Accountability proportionate to the risk
support consumer data protection in some
profile of the application and the role of the
form, and 70% of Europeans and 60% of entity providing, developing, or operating an AI
Americans support AI regulation. system to control and mitigate unintended or
harmful outcomes for consumers.
As outlined in our Principles for Trust and Transparency,
• Transparency in where the technology is
IBM has long argued that AI systems need to be
deployed, how it is used, and why it provides
transparent and explainable. That’s one reason why we
certain determinations.
supported the EU and the OECD AI Principles, and in
particular the focus on transparency and trustworthiness
• Fairness and security validated by testing
in both.
for bias before AI is deployed and re-tested
as appropriate throughout its use, especially
Principles are admirable and can help communicate a
in automated determinations and high-risk
company’s commitments to citizens and consumers. But
applications.
it’s past time to move from principles to policy. Requiring
disclosure — as appropriate based on use-case and
end-user — should be the default expectation for many
Wisely, the OECD AI Principles suggest a solid
companies creating, distributing, or commercializing AI
accountability bedrock for this framework, arguing that
systems. In an earlier Policy Lab essay, we articulated a
“[g]overnments should promote a policy environment
disclosure requirement for law enforcement use-cases of
that supports an agile transition from the research and
facial recognition technology. Something similar should
development stage to the deployment and operation
be required of AI more generally in order to provide the
stage for trustworthy AI systems.” This implicit recognition
public with appropriate assurances that they are being
ibm.com/policy 1
of the fundamental difference in accountability between complexity and potential impact of AI systems increases,
stages of AI development can help appropriately assign so too must the accountability embraced by different
responsibility for providing transparency and ensuring organizations providing various functions in the AI lifecycle.
fairness and security, based on who has better control A market environment that prioritizes the adoption of
over the protection of privacy, civil liberties, and harm- lead AI ethics officials, or other designated individuals, to
prevention activities in a given context. oversee and manage this increasing complexity could help
to mitigate risks and improve public acceptance and trust
In the lifecycle of AI capabilities in the marketplace, of these systems, while also driving firms’ commitment
organizations may contribute research, the creation to the responsible development, deployment, and overall
of tooling, and APIs; in later stages of operation, stewardship of this important technology.
organizations will train, manage, and control, operate, or
own the AI models that are put to real-world commercial 2. Different rules for different risks. All entities
use. These different functions may allow for a distinction providing or owning an AI system should conduct an
between “providers” and “owners,” with expectations of initial high-level assessment of the technology’s potential
responsibilities based on how an organization’s role falls for harm. As noted previously, such assessments should
into one or both categories. be based on the intended use-case application(s),
end-user(s), how reliant the end-user would be on the
Differentiating accountability can help to better mitigate technology, and the level of automation. Once initial risk
potential harm by directing resources and oversight to is determined, a more in-depth and detailed assessment
specific applications of AI based on the severity and should be undertaken for higher-risk applications. In
likelihood of potential harms arising from the end-use and certain low-risk situations, a more cursory appraisal
user of such systems. Risk-based regulatory approaches would likely suffice. For those high-risk use-cases, the
like this — which also allow for more manageable and assessment processes should be documented in detail,
incremental changes to existing rules — are ideal means be auditable, and retained for a minimum period of time.
to protect consumers, build public trust in AI, and provide
innovators with needed flexibility and adaptability. 3. Don’t hide your AI. Transparency breeds trust; and the
best way to promote transparency is through disclosure.
Building from these pillars, we propose a precision Unlike other transparency proposals, this approach does
regulation framework that incorporates 5 policy not entail companies revealing source code or other
imperatives for companies, based on whether they are forms of trade secrets or IP. Instead it focuses on making
a provider or owner (or both) of an AI system. These the purpose of an AI system clear to consumers and
policies would vary in robustness according to the level businesses. Such disclosures, like other policy imperatives
of risk presented by a particular AI system, which would here, should be reasonably linked to the potential risk
be determined by conducting an initial risk assessment and harm to individuals. As such, low-risk and benign
based on potential for harm associated with the intended applications of AI may not require the type of disclosure
use, the level of automation (and human involvement), that higher-risk use-cases might require.
and whether an end-user is substantially reliant on the AI
system based on end-user and use-case. 4. Explain your AI. Any AI system on the market that
is making determinations or recommendations with
1. Designate a lead AI ethics official. To ensure potentially significant implications for individuals should
compliance with these expectations, providers and owners be able to explain and contextualize how and why it
should designate a person responsible for trustworthy arrived at a particular conclusion. To achieve that, it
AI, such as a lead AI ethics official. This person would is necessary for organizations to maintain audit trails
be accountable for internal guidance and compliance surrounding their input and training data. Owners and
mechanisms, such as an AI Ethics Board, that oversee operators of these systems should also make available —
risk assessments and harm mitigation strategies. As the as appropriate and in a context that the relevant end-user
ibm.com/policy 2
can understand — documentation that detail essential To achieve this, governments should:
information for consumers to be aware of, such as
confidence measures, levels of procedural regularity, and • Designate, or recognize, existing effective co-
error analysis. regulatory mechanisms (e.g. CENELEC in Europe
or NIST in the U.S.) to convene stakeholders
74% of American and 85% of EU and identify, accelerate, and promote efforts to
create definitions, benchmarks, frameworks and
respondents are in agreement that
standards for AI systems. Ideally, standards that
artificial intelligence systems should be
are globally recognized would help create consistency
transparent and explainable, and strong and certainty for consumers, communicating to end-
users that the AI is trustworthy;
pluralities in both countries believe
that disclosure should be required for
• Support the financing and creation of AI testbeds
companies creating or distributing AI with a diverse array of multi-disciplinary
systems. Nearly 3 in 4 Europeans and two- stakeholders working together in controlled
environments. In particular, minority-serving
thirds of Americans support regulations
organizations and impacted communities should be
such as conducting risk assessments,
supported in their efforts to engage with academia,
doing pre-deployment testing for bias and government, and industry. Working together, these
stakeholders can accelerate the development
fairness, and reporting to consumers and
and evaluation criteria of AI accuracy, fairness,
businesses that an AI system is being used
explainability, robustness, transparency, ethics,
in decision making. privacy, and security; and
5. Test your AI for bias. All organizations in the AI • Incentivize providers and owners to voluntarily
developmental lifecycle have some level of shared embrace globally recognized standards,
responsibility in ensuring the AI systems they design certification, and validation regimes. One such
and deploy are fair and secure. This requires testing potential mechanism is by providing various levels of
for fairness, bias, robustness and security, and taking liability safe harbor protections, based on whether and
remedial actions as needed, both before sale or how an organization adheres and certifies to globally
deployment and after it is operationalized. Owners should recognized best practices and standards.
also be responsible for ensuring use of their AI systems is
aligned with anti-discrimination laws, as well as statutes Finally, any action or practice prohibited by anti-
addressing safety, privacy, financial disclosure, consumer discrimination laws should continue to be prohibited when
protection, employment, and other sensitive contexts. it involves an automated decision-making system. Whether
For many use-cases, owners should continually monitor, a decision is fully rendered by a human or a determination
or retest, the AI models after the product is released is assisted by an automated AI system, impermissibly
to identify and mitigate against any machine-learning biased or discriminatory outcomes should never be
resulting in unintended outcomes. Policies should create considered acceptable. But whereas correcting the bias of
an environment that incentivizes both providers and humans is a daunting and difficult task, in AI systems it may
owners to do such testing well. This can be done without be a matter of addressing historical bias in some training
creating new and potentially cumbersome AI-specific data by testing for, and correcting, statistical failures in the
regulatory requirements, but rather by adhering to a set model. While this will take time, AI offers us the promise
of agreed-upon definitions, best practices, and global of a world where bias and discrimination may one day
standards. fade away. With precision regulations helping to promote
trustworthy AI, that future could be sooner than we think.
ibm.com/policy 3
Since day one, IBM has pushed the boundaries of technology to address the challenges
of tomorrow. We’ve done this while earning our clients’ trust to innovate responsibly and
carefully stewarding their data. We’ll continue to drive forward new technological advances
with the values of accountability, transparency, and trust that our clients and government
partners have relied on since 1911.
The world — and IBM — has changed a lot over the past century. We’ve seen the march of progress move
humanity from an analog era to the digital age and explosive innovation in both bits and atoms contribute to a
wave of disruptive change. At IBM, we’re optimistic about what the future holds, and the crucial role technological
advancement will play in driving economic growth and societal well-being. Already, cloud computing has changed
how work gets done and how connections are made, artificial intelligence has revolutionized our daily routines,
and we can find information on practically anything at the touch of a button. Technology will fundamentally
change society, bring us closer together, improve lives around the world and help us tackle some of our greatest
challenges.
But no journey comes without challenges. We have already seen concerns materialize across
emerging technologies on the implications of opaque AI systems making safety- and life-
critical decisions; the growing pains of new digital platforms leading to the spread of illegal
and harmful content online; and fears that a fully-automated future will displace more jobs
than it creates. All of this comes amid a wave of global challenges to modern society, from
the spread of protectionist impulses to the failure to address climate change.
But at IBM, we’ve seen how technological progress has improved the human condition over the past 100 years.
We were optimistic about the future then, and we remain optimistic about the future to come. While there are
challenges ahead, we believe there are clear and practical ways through them.
As businesses and governments break new ground and deploy technologies that are
positively transforming our world, we want to work collaboratively to make sure public
policy adapts to meet the challenges of tomorrow. That’s why we’ve created the IBM Policy
Lab, a new forum that provides a vision with actionable recommendations to harness
the benefits of innovation while ensuring trust in a world reshaped by data.
Led by co-directors Ryan Hagemann and Jean-Marc Leclerc — two long-standing experts in tech and public policy
— IBM Policy Lab convenes leading thinkers in public policy, academia, and technology to develop the concrete,
common-sense policy ideas leveraging technology to tackle some of the most pressing issues facing our world.
Our approach is grounded in the belief that tech can continue to disrupt and improve civil society while protecting
individual privacy.
ibm.com/policy 4
What We Do How We’re Different
• Develop Industry-Leading Policy Positions that • While some traffic in grandiose policy
don’t just respond to the spot issues of today but recommendations that stand little chance of
look forward to the opportunities of tomorrow becoming reality, IBM has always believed that
and the ways public-private cooperation can big challenges require practical solutions. That’s
pave the way for an even brighter future. With the precisely what IBM Policy Lab has been chartered to
full benefits of artificial intelligence, blockchain, create.
quantum computing and more still untapped,
we’ll put forward bold visions for public policy that • Our policy recommendations will be concrete.
harnesses innovation. Specific. Actionable. We will have big ideas, but they
will be ideas that policymakers can implement on
• Collaborate with Global Thinkers, Stakeholders day one.
and Leaders to collect input and share perspective
from the diverse voices that must inform public • We will also convene government, industry and
policy. civil society experts to think big about upcoming
challenges and make space for collaborative
• Produce Data-Driven Studies and Research to solutions.
guide policymaking with specific, common-sense
recommendations, and help industry leaders make • Serious times call for serious solutions, and that’s
critical decisions on policies impacting our future. precisely what leaders in government, business and
civil society can expect from IBM Policy Lab.
As technological innovation races ahead, our mission
to raise the bar for a trustworthy digital future could
not be more urgent. IBM Policy Lab is committed to
developing and advocating the right policies that meet
the demands of the moment and harness the power of
technology as a force for good in the world.
Jean-Marc Leclerc joined IBM’s Government and Regulatory Affairs team in 2015, where he leads the EU Affairs team. Jean-Marc
is the Chair of the EMEA Policy Committee at the Business Software Association (BSA), and he is a Vice-Chair of the Digital Economy
Committee at the American Chamber of Commerce to the EU. Before joining IBM, he was a Policy Director at Digitaleurope 2013-
15. He has also managed an association representing the music industry in Brussels 2006-13. Jean-Marc is a graduate from the
universities of Paris III, Sciences Po, the Catholic Institute of Paris, and the College of Europe in Bruges.
Ryan Hagemann is the Co-Director of the IBM Policy Lab and a Technology Policy Executive on IBM’s Government and Regulatory
Affairs team. He was previously a senior policy fellow at the International Center for Law & Economics. Before joining ICLE, he was
a senior fellow at the Niskanen Center, where he also served as the senior director for policy and director of technology policy. His
policy expertise focuses on regulatory governance of emerging technologies, as well as a broader research portfolio that includes
genetic modification and regenerative medicine, bioengineering and healthcare IT, artificial intelligence, autonomous vehicles,
commercial drones, the Internet of Things, and other issues at the intersection of technology, regulation, and the digital economy.
His work on “soft law” governance systems, autonomous vehicles, and commercial drones has been featured in numerous academic
journals, and his research and comments have been cited by The New York Times, MIT Technology Review, and The Atlantic, among
other outlets. He has been published in The Wall Street Journal, Wired, National Review, The Washington Examiner, U.S. News &
World Report, The Hill, and elsewhere. Ryan graduated from Boston University with a B.A. in international relations, foreign policy,
and security studies and holds a Master of Public Policy in science and technology policy from George Mason University.
ibm.com/policy 5 |
148 | ibm | Enterprise+AI+Development+Survey.pdf | Enterprise AI Development:
Obstacles & Opportunities
JANUARY 2025
© 2024Morning Consult.All rightsreserved.
Methodology Table of Contents
3
This survey was conducted October 31 to November 1, 2024, Key Findings
among a total sample of 1,063 Enterprise AI Developers in the
US. The interviews were conducted online, and the data is
unweighted. Results from the full survey have a margin of error of
+/- 3%.
4
AI Skills & Developer Challenges
To qualify as an Enterprise AI Developer survey respondents must
meet the following requirements:
9
✓ Be employed (Full-time, Part-time, or Self-employed) Technology Landscape | Developer Tools
✓ Work in one of the following roles: Data Scientist,
Application Developer, System Developer, AI Developer,
ML Engineer, Software Engineer, Software Developer, AI
Engineer, or IT Engineer
16
Agents
✓ Contribute at least occasionally to the development of
enterprise AI applications in their current role
2
KEY FINDINGS
AI skill gaps AI tool gaps Simplicity is the solution
Generative AI skill levels vary The most important tool qualities for Developers crave tools that are easy to
significantly among developers. Most building enterprise AI are also the master. And when it comes to
developers surveyed do not view rarest, hampering the development developer productivity, AI-powered
themselves as experts, despite being process. Meanwhile, developers must coding solutions are incredibly
on the front lines of generative AI juggle a roster of tools. popular.
adoption.
➢ Only one third of respondents are
➢ Less than one quarter (24%) of ➢ Performance (42%), Flexibility (41%),
willing to invest more than two hours
application developers surveyed Ease of Use (40%), and Integration
in learning a new AI development tool.
ranked themselves as "experts" in (36%) are the most essential qualities
generative AI. in enterprise AI development tools,
➢ 99% are using coding assistants in
according to respondents. Yet over a
some capacity for AI development.
➢ One third of respondents list the lack third also said those very same traits
And most commonly, developers said
of a standardized AI development are the rarest.
these tools saved them 1-2 hours per
process as a top challenge.
day (41% of developers).
➢ A majority (72%) of respondents use
➢ 99% of respondents are exploring or between 5 and 15 tools to create an AI
developing AI agents. enterprise application.
3
SECTION 1
AI Skills & Developer
Challenges
4
AI SKILLS & DEVELOPER CHALLENGES
A lack of standard processes and trust are inhibiting the development of generative AI
applications for business
Top Challenges of Developing Gen AI at the Enterprise Level
All Enterprise AI Developers, Showing % selected
Lack of a standardized AI development process 33%
Developing an ethical and trusted lifecycle that ensures transparency and traceability
33%
of data
Customization 32%
Rate of change 31%
Infrastructure/stack complexity 29%
Establishing governance and ensuring compliance 28%
Lack of skills/experience 26%
Lack of clarity on business outcome/objective 26%
Interoperability of tools 23%
LLM quality 19%
Other 0%
IBM3: What are the top obstacles associated with developing gen AI applications for enterprise use? Please select up to 3 options.
Sample size: Enterprise AI Developers = 1,063n 5
AI SKILLS & DEVELOPER CHALLENGES
Expertise in generative AI varies greatly by developer role
Most application developers don’t view themselves as experts in generative AI
Expert Experience Level with Generative AI
Enterprise AI Developers by Role, Showing % selected ‘Expert’
54%
51%
48%
46%
All Enterprise AI 43% 43%
40%
Developers Average: 44% 38%
24%
AI Developer Data Scientist Software System ML Engineer Software IT Engineer AI Engineer Application
Engineer Developer Developer Developer
IBM1: How would you rank your proficiency and professional experience these types of AI?
Sample size: Data Scientist = 39n, Application Developer = 37n, System Developer = 52n, AI Developer = 82n, ML Engineer = 46n, Software Engineer = 277n, Software Developer = 273n, AI Engineer = 95n, IT Engineer = 162n 6
Note: Please note that sample sizes for some roles are small and therefore data should be interpreted directionally
AI SKILLS & DEVELOPER CHALLENGES
Developers feel more comfortable with Generative AI versus Classical AI
Experience Level with Types of AI
All Enterprise AI Developers
Expert Advanced / Experienced Intermediate / Some Experience Novice / Entry Level No Experience
Generative AI 44% 40% 11% 4%
17% difference
Classical AI 27% 40% 17% 9% 8%
IBM1: How would you rank your proficiency and professional experience these types of AI?
Sample size: Enterprise AI Developers = 1,063n 7
AI SKILLS & DEVELOPER CHALLENGES
Time spent on project phases
The number of hours developers spend on tasks becomes less predictable later in the project cycle
Average Number of Hours Spent on Project Cycle Tasks/Stages
All Enterprise AI Developers, Showing % within specific tasks/stage
0-4 hours 5-9 hours 10-19 hours 20-29 hours 30-39 hours 40+ hours Average Hours Median Hours
Infrastructure set-up / T
runtime configuration
18% 20% 21% 19% 13% 8% 18 12 mi
e
v
a
Model selection 20% 25% 22% 16% 7% 10% 17 11 ir
a
n
c
e
Model customization 19% 20% 22% 13% 10% 16% 21 14 ni
c
r
e
a
s
Prompt engineering 21% 20% 25% 16% 10% 8% 17 12 e
s
t
h
r
Orchestration and o
20% 20% 21% 16% 15% 9% 19 12 u
integrations g
h
o
u
Deployment 20% 22% 27% 11% 9% 11% 18 11 t
t
h
e
c
Evaluation and y
18% 20% 26% 14% 8% 15% 21 12 c
Observability l
e
IBM2: During a typical project cycle, how many hours do you spend on the following tasks/stages? Please provide your best estimate. [NUMERIC OPEN END]
Sample size: Enterprise AI Developers = 1,063n 8
SECTION 2
Technology Landscape |
Developer Tools
9
TECHNOLOGY LANDSCAPE | DEVELOPER TOOLS
Enterprise AI Developers face tool sprawl
The majority of those surveyed use between 5 and 15 tools to create an AI enterprise application;
13% of developers are using 15 or more
Average Number of Tools Used to Create an AI Enterprise Application
All Enterprise AI Developers, Showing % Selected
1-5 16%
5-10 35%
10-15 37%
15-20 11%
Over 20 2%
IBM6: On average, how many tools do you use to create an AI enterprise application?
Sample size: Enterprise AI Developers = 1,063n 10
TECHNOLOGY LANDSCAPE | DEVELOPER TOOLS
When it comes to tooling, the most critical tool qualities are also among the rarest
Most Essential Traits in Enterprise AI Development Tools Top Missing Traits in Enterprise AI Development Tools
All Enterprise AI Developers, Showing % selected All Enterprise AI Developers, Showing % selected
Performance 42% Performance 37%
Flexibility 41% Flexibility 37%
Ease of use 40% Integration with existing tool 37%
Integration with existing tool 36% Ease of use 36%
Documentation quality 34% Cost-effectiveness 36%
Cost-effectiveness 33% Documentation quality 35%
Open source 32% Community support and resources 32%
Community support and resources 30% Open source 31%
IBM4 Which of these traits do you consider most necessary to exist in the tools/frameworks that you use to develop enterprise-grade AI systems? Please select up to 3 options. // IBM5 And which of these traits do you find most
commonly lacking in the tools/frameworks that you use to develop enterprise-grade AI systems? Please select up to 3 options. 11
Sample size: Enterprise AI Developers = 1,063n
TECHNOLOGY LANDSCAPE | DEVELOPER TOOLS
Enterprise AI Developers crave tools that are easy to master;
Only one third of those surveyed are willing to invest more than two hours in learning a new AI
development tool
Time Willing to Invest in Learning New AI Tools
All Enterprise AI Developers, Showing % Selected
Less than 10 minutes 0%
10 to 29 minutes 4%
30 to 59 minutes 20%
1 to 2 hours 42%
3 to 5 hours 22%
More than 5 hours 11%
IBM9: How much time in total are you willing to invest learning to use a new AI development tool before moving on?
Sample size: Enterprise AI Developers = 1,063n 12
TECHNOLOGY LANDSCAPE | DEVELOPER TOOLS
Developers are leaning into a variety of tools to streamline AI development
More than half are using no-code or low-code tools; 73% are leaning into pro-code tools
Types of Development Tools | Extent of Usage by Level
All Enterprise AI Developers, Showing T2B (Often + Very Often)
Pro code 73%
y
t
l
u
c Low code 65%
i
f
f
i
D
No code 59%
IBM11: To what extent are you using the following for AI development?
Sample size: Enterprise AI Developers = 1,063n 13
TECHNOLOGY LANDSCAPE | DEVELOPER TOOLS
AI-powered Coding Assistants are saving developers significant amounts of time
41% of developers said these tools saved them 1-2 hours per day
Extent of Coding Assistant Usage for AI Time Saved [Per Day] Using AI-Assisted Coding
Development Tools
All Enterprise AI Developers All Enterprise AI Developers, Showing % Selected
Very often Often Occasionally Rarely Not at all It doesn't save me time. 0%
Less than 15 minutes 2%
15 to 29 minutes 9%
35% 43% 15% 5%
30 to 59 minutes 25%
1 to 2 hours 41%
3 to 4 hours 18%
Nearly all enterprise AI developers are
using coding assistants – and 78% are
More than 4 hours 4%
using them often or very often.
I'm not sure 1%
I don't use AI-assisted coding tools 0%
IBM11: To what extent are you using the following for AI development? Coding assistants // IBM10: On average, how much time do you estimate you save per day by using AI-assisted coding tools?
Sample size: Enterprise AI Developers = 1,063n 14
TECHNOLOGY LANDSCAPE | DEVELOPER TOOLS
Exploration of new AI development tools is limited, with nearly 80% only occasionally
experimenting with new tools
Frequency of Experimenting with New AI Development Tools
All Enterprise AI Developers
Every month Every 1 to 3 months Every 3 to 6 months Rarely or never
21% 46% 31% 2%
Enterprise AI Developers experiment with ~6 new
tools on average during these time frames.
IBM7: How often do you experiment with new AI development tools? // IBM8 How many new AI development tools do you usually experiment with during that timeframe? [NUMERIC OPEN END] Showing Average
Sample size: Enterprise AI Developers = 1,063n, Enterprise AI Developers experimenting with new AI development tools at least every 3 to 6 months = 1,043n 15
SECTION 3
Agents
16
AGENTS
Just about everyone is exploring or developing AI agents
Exploration/Development of Use Cases for AI Agents
All Enterprise AI Developers
1% not currently exploring/developing use
cases for AI agents
99% exploring/developing use cases
for AI agents
IBM12: What use cases is your enterprise exploring or developing for AI agents?
Sample size: Enterprise AI Developers = 1,063n 17
AGENTS
Trustworthiness emerges as the top concern when it comes to scaling agents
Top Concerns for Scaling AI Agents in Enterprise
All Enterprise AI Developers, Showing % selected
Trustworthiness: Ensuring outputs are accurate and void of bias 31%
Introducing new attack vectors: AI Agents being compromised by
23%
malicious actors
Adhering to compliance and regulations 22%
Rogue AI agents aren't keeping
Developers up at night. Only 22% of
Becoming overly autonomous: Humans lose oversight and visibility
22%
into systems those surveyed said agents becoming
overly autonomous was a top concern.
I have no concerns 3%
IBM14: What are you most concerned about when it comes to AI agents scaling in the enterprise?
Sample size: Enterprise AI Developers = 1,063n 18
AGENTS
Customer service, project management, and content creation are top use-cases currently
being explored for agents
AI Agent Use Cases Being Explored
All Enterprise AI Developers, Showing % selected
Customer Service and Support 50%
Project Management / Personal Assistant 47%
Content Creation 46%
HR 43%
Transportation 32%
Healthcare 28%
Not currently exploring/developing 1%
IBM12: What use cases is your enterprise exploring or developing for AI agents?
Sample size: Enterprise AI Developers = 1,063n 19
© 2024 Morning Consult. All rights reserved. |
150 | ibm | Christina-Montgomery-Senate-Judiciary-Testimony-5-16-23.pdf | Testimony of Christina Montgomery, Chief Privacy and Trust Officer, IBM
Before the
U.S. Senate Judiciary Committee
Subcommittee on Privacy, Technology, and the Law
Hearing on “Oversight of AI: Rules for Artificial Intelligence”
Tuesday, May 16, 2023
Chairman Blumenthal, Ranking Member Hawley, members of the Subcommittee:
Thank you for today’s opportunity to present before the subcommittee. My name is
Christina Montgomery, and I am IBM’s Chief Privacy and Trust Officer. I also co-
chair our company’s AI Ethics Board.
Introduction
AI is not new, but it has advanced to the point where it is certainly having a
moment. This new wave of generative AI tools has given people a chance to
experience it first-hand. Citizens are using it for help with emails, their homework,
and so much more.
While IBM is not a consumer-facing company, we are just as active – and have
been for years – in helping business clients use AI to make their supply chains
more efficient, modernize electricity grids, and secure financial networks from
fraud. IBM’s suite of AI tools, called IBM Watson after the AI system that won on
TV’s Jeopardy! more than a decade ago, is widely used by enterprise customers
worldwide. Just recently we announced a new set of enhancements, called
watsonx, to make AI even more relevant today.1 Our company has extensive
experience in the AI field in both an enterprise and cutting-edge research context,
and we could spend an entire afternoon talking about ways the technology is being
used today by business and consumers.
But the technology’s dramatic surge in public attention has, rightfully, raised
serious questions at the heart of today’s hearing. What are AI’s potential impacts
1 See https://newsroom.ibm.com/2023-05-09-IBM-Unveils-the-Watsonx-Platform-to-Power-Next-Generation-
Foundation-Models-for-Business.
2
on society? What do we do about bias? What about misinformation, misuse, or
harmful and abusive content generated by AI systems?
Senators, these are the right questions, and I applaud you for convening today’s
hearing to address them head-on.
IBM has strived for more than a century to bring powerful new technologies like
artificial intelligence into the world responsibly, and with clear purpose. We follow
long-held principles of trust and transparency that make clear the role of AI is to
augment, not replace, human expertise and judgement. We were one of the first in
our industry to establish an AI Ethics Board, which I co-chair, and whose experts
work to ensure that our principles and commitments are upheld in our global
business engagements.2 And we have actively worked with governments
worldwide on how best to tailor their approaches to AI regulation.
It’s often said that innovation moves too fast for government to keep up. But while
AI may be having its moment, the moment for government to play its proper role
has not passed us by. This period of focused public attention on AI is precisely the
time to define and build the right guardrails to protect people and their interests.
It is my privilege to share with you IBM’s recommendations for those guardrails.
Precision Regulation
The hype around AI has created understandable confusion among some in
government on where intervention is needed and how regulatory guardrails
should be shaped. But at its core, AI is just a tool, and tools can serve different
2 See https://www.ibm.com/artificial-intelligence/ethics.
3
purposes. A wrench can be used to assemble a desk or construct an airplane, yet
the rules governing those two end products are not primarily based on the wrench
— they are based on use.
That is why IBM urges Congress to adopt a “precision regulation” approach to
artificial intelligence. This means establishing rules to govern the deployment of AI
in specific use-cases, not regulating the technology itself.
A precision regulation approach that we feel strikes an appropriate balance
between protecting Americans from potential harms and preserving an
environment where innovation can flourish would involve:
• Different Rules for Different Risks – A chatbot that can share restaurant
recommendations or draft an email has different impacts on society than a
system that supports decisions on credit, housing, or employment. In
precision regulation, the more stringent regulation should be applied to the
use-cases with the greatest risk.
• Clearly Defined Risks – There must be clear guidance on AI end uses or
categories of AI-supported activity that are inherently high-risk. This
common definition is key to ensuring that AI developers and deployers have
a clear understanding of what regulatory requirements will apply to a tool
they are building for a specific end use. Risk can be assessed in part by
considering the magnitude of potential harm and the likelihood of
occurrence.
• Be Transparent, Don’t Hide Your AI – Americans deserve to know when
they are interacting with an AI system, so Congress should formalize
disclosure requirements for certain uses of AI. Consumers should know
when they are interacting with an AI system and whether they have recourse
4
to engage with a real person, should they so desire. No person, anywhere,
should be tricked into interacting with an AI system. AI developers should
also be required to disclose technical information about the development
and performance of an AI model, as well as the data used to train it, to give
society better visibility into how these models operate. At IBM, we have
adopted the use of AI Factsheets – think of them as similar to AI nutrition
information labels – to help clients and partners better understand the
operation and performance of the AI models we create.
• Showing the Impact – For higher-risk AI use-cases, companies should be
required to conduct impact assessments showing how their systems
perform against tests for bias and other ways that they could potentially
impact the public, and attest that they have done so. Additionally, bias
testing and mitigation should be performed in a robust and transparent
manner for certain high-risk AI systems, such as law enforcement use-
cases. These high-risk AI systems should also be continually monitored and
re-tested by the entities that have deployed them.3
IBM recognizes that certain AI use-cases raise particularly high levels of concern.
Law enforcement investigations and credit applications are two often-cited
examples. By following the risk-based, use-case specific approach at the core of
precision regulation, Congress can mitigate the potential risks of AI without stifling
its use in a way that dampens innovation or risks cutting Americans off from the
trillions of dollars of economic activity that AI is predicted to unlock.
Generative AI
The explosion of generative AI systems in recent month has caused some to call for
3 See https://www.ibm.com/policy/ai-precision-regulation/.
5
a deviation from a risk-based approach and instead focus on regulating AI in a
vacuum, rather than its application. This would be a serious error, arbitrarily
hindering innovation and limiting the benefits the technology can provide. A risk-
based approach ensures that guardrails for AI apply to any application, even as
this new, potentially unforeseen developments in the technology occur, and that
those responsible for causing harm are held to account.4
When it comes to AI, America need not choose between responsibility,
innovation, and economic competitiveness. We can, and must, do all three now.
Business’ Role
This focus on regulatory guardrails established by Congress does not – not by any
stretch – let business off the hook for its role in enabling the responsible
deployment of AI.
I mentioned that IBM has strong AI governance practices and processes in place
across the full scope of our global enterprise. We have principles grounded in
ethics and people-centric thinking, and we have strong processes in place to bring
them to life. This is also good business: IBM has long recognized ethics and
trustworthiness are key to AI adoption, and that the first step in achieving these is
the adoption of effective risk management practices.
Companies active in developing or using AI must have (or be required to have)
strong internal governance processes, including, among other things:
4 See https://newsroom.ibm.com/Whitepaper-A-Policymakers-Guide-to-Foundation-Models.
6
• Designating a lead AI ethics official responsible for an organization’s
trustworthy AI strategy, and
• Standing up an AI Ethics Board or similar function to serve as a centralized
clearinghouse for resources to help guide implementation of that strategy.
IBM has taken both steps and we continue calling on our industry peers to follow
suit.
Our AI Ethics Board plays a critical role in overseeing our internal AI governance
process, creating reasonable internal guardrails to ensure we introduce technology
into the world in a responsible and safe manner. For example, the board was
central in IBM’s decision to sunset our general purpose facial recognition and
analysis products, considering the risk posed by the technology and the societal
debate around its use. IBM’s AI Ethics Board infuses the company’s principles and
ethical thinking into business and product decision-making. It provides
centralized governance and accountability while still being flexible enough to
support decentralized initiatives across IBM’s global operations.
The board, along with a global community of AI Ethics focal points and advocates,
reviews technology use-cases, promotes best practices, conducts internal
education, and leads our participation with stakeholder groups worldwide. In
short, it is a mechanism by which IBM holds our company and all IBMers
accountable to our values, and our commitments to the ethical development and
deployment of technology.
We do this because we recognize that society grants our license to operate. If
businesses do not behave responsibly in the ways they build and use AI, customers
will vote with their wallets. And with AI, the stakes are simply too high, the
7
technology too powerful, and the potential ramifications too real. AI is not some
fun experiment that should be conducted on society just to see what happens or
how much innovation can be achieved.
If a company is unwilling to state its principles and build the processes and teams
to live up to them, it has no business in the marketplace.
Conclusion
Mr. Chairman, and members of the subcommittee, the era of AI cannot be another
era of move fast and break things. But neither do we need a six-month pause –
these systems are within our control today, as are the solutions. What we need at
this pivotal moment is clear, reasonable policy and sound guardrails. These
guardrails should be matched with meaningful steps by the business community
to do their part. This should be an issue where Congress and the business
community work together to get this right for the American people. It’s what they
expect, and what they deserve.
IBM welcomes the opportunity to work with you, colleagues in Congress, and the
Biden Administration to build these guardrails together.
Thank you for your time, and I look forward to your questions.
8 |
151 | ibm | IBM+Global+AI+Adoption+Index+Report+Dec.+2023.pdf | IBM GLOBAL AI ADOPTION
INDEX – ENTERPRISE
REPORT
NOVEMBER 8 – 23, 2023
© 2023 Morning Consult, All Rights Reserved.
METHODOLOGY & AUDIENCE
REPRESENTATIVE SAMPLE OF IT PROFESSIONALS IN MARKET
• 2,342 IT Professionals at enterprises (organizations with > 1,000 employees)
• This study was conducted in Australia, Canada, China, France, Germany, India, Italy, Japan, Singapore, South Korea, Spain, UAE,
UK, US, and LATAM (Brazil, Mexico, Peru, Argentina, Chile, Colombia)
• Market sample sizes range from 92 to 316
• To qualify for this audience, participants must be employed full-time, work at companies with more than 1,000 employees, work in a manager
or higher level role, and have at least some knowledge about how IT operates and is used by their company.
• Survey conducted online through MC’s proprietary network of online providers.
COMPANY SIZE BREAKDOWN
• 50% of respondents came from firms with 1,001 to 5,000 employees
• 50% of respondents came from firms with more than 5,000 employees
RESPONDENTS REPRESENTED A MIX OF SENIORITY
• All respondents were required to have significant insight or input into their firm’s IT decision-making
• 20% of the sample was at a VP level or above (including CIOs, etc.)
• The remainder of the sample represented a mix of directors and senior manager-level employees
2
IBM GLOBAL AI ADOPTION INDEX
Key Findings
1. AI adoption and exploration, covering both general AI and 3. AI is contributing to multiple facets of organizational operations
generative AI, continues to be a substantial focus for enterprises at enterprises, with IT process automation and security and
globally one year after the release of GPT-3. Many of those large threat detection being the most popular applications. IT
companies already exploring or deploying AI have accelerated their Professionals are at the forefront of AI usage at their enterprises and
roll-out of AI in the past two years, with ‘Research and Development,’ note the importance of being able to build and run AI projects
‘Workforce Upskilling,’ and ‘Building Proprietary AI Solutions’ wherever their data resides. Confidence in these capabilities is high,
emerging as top investment priorities. In the dynamic landscape of as most IT Professionals are confident that their enterprise has the
generative AI, enterprises tend to utilize in-house technology over right tools to find data across the business.
open-source technology.
4. Trustworthy and responsible AI practices are of utmost
importance to both consumers and enterprises at various stages
2. As enterprises enter the AI landscape, many have already
of AI implementation. In fact, most large organizations already
established some form of an AI strategy. This adoption is fueled by
exploring or deploying AI are actively taking steps like safeguarding
factors such as increased accessibility, cost-cutting through
data privacy through the entire lifecycle to ensure that. Insufficient
automation, and growing AI integration in business apps. Globally,
expertise for reliable AI management and development and lack of an
Enterprise IT Professionals highlight accessible tools, %), the
AI strategy are among the biggest barriers enterprises face as they
increased prevalence of AI related skillsets, and AI-tailored solutions
strive to develop trustworthy AI.
as key industry changes. However, challenges like limited knowledge,
too much data complexity, and ethical concerns hinder adoption. In 5. AI has a predominantly positive influence on the workforce.
the context of generative AI, additional obstacles emerge, including Numerous enterprises are investing in AI training, and IT
data privacy and trust/transparency concerns. Professionals note employee enthusiasm for new AI and automation
tools. Additionally, AI plays a crucial role in addressing labor and skills
shortages by equipping large companies with the tools to streamline
tasks and automate self-service interactions.
Methodology: This poll was conducted from Nov. 8 – 23, 2023 among a sample of 2,342 IT Professionals at enterprises (organizations with > 1,000 employees) in Australia, Canada, China, France, Germany, India,
Italy, Japan, Singapore, South Korea, Spain, UAE, UK, US and LATAM (Brazil, Mexico, Peru, Argentina, Chile, Colombia). Global results have a margin of error of +/- 2 percentage points at a 95% confidence level.
3
AGENDA
AI AD OPT ION & IN VEST MENTS
D R IVER S & BAR R IER S OF AI
CURRENT USES OF AI
AI ET H IC S AN D R ESPON SIBIL IT Y
AI’S IMPAC T ON EMPL OYEES
AI ADOPTION & INVESTMENTS
Over the past four years, AI adoption at enterprises has remained steady, with 42% of IT Professionals reporting AI
deploying and an additional 40% reporting active exploration in November 2023.
Has your company adopted or explored using Artificial Intelligence (AI) as part of its business operations and digital transformation?
My company is not currently
My company is exploring, My company has actively
using, or exploring the
but has not deployed, AI in deployed AI as part of its Don't know/Not sure
use of, AI in its business
its business operations business operations
operations
Oct. 2019 16% 34% 45% 5%
Apr. 2021 12% 37% 44% 6%
Apr. 2022 12% 38% 46% 3%
Apr. 2023 16% 34% 45% 5%
Nov. 2023 15% 40% 42% 3%
Base IT Professionals at Enterprises (organizations > 1,000 employees): October 2019 = 1,358n, April 2021 = 1,550n, April 2022 = 2,362n, April 2023 = 2,247n, November 2023 = 2,342n 5
AI ADOPTION & INVESTMENTS
Although there is a similar global AI Adoption trend from April 2023, there are some country specific outliers worth
noting.
Has your company adopted or explored using Artificial Intelligence (AI) as part of its business operations and digital transformation?
My company is not currently My company has actively Increases in AI Adoption
using, or exploring the deployed AI as part of its
use of, AI in its business business operations
operations
My company is exploring,
Don't know/Not sure
but has not deployed, AI in
The UAE, UK, and LATAM all saw an uptick in
its business operations
enterprises deploying AI in November 2023.(UAE: 48%
Global Enterprise 15% 40% 42%
Apr. ‘23, 58% Nov. ‘23) (UK: 29% Apr. ‘23, 37% Nov.
Australia* 17% 50% 29%
‘23) (LATAM: 40% Apr. ‘23, 47% Nov. ‘23).
Canada 12% 48% 37%
China 14% 36% 50%
France 19% 45% 26% 10%
Decreases in AI Adoption
Germany 21% 44% 32%
India 13% 27% 59% China (66% Apr. ‘23 to 50% Nov. ‘23) and Japan (49%
Italy 23% 38% 36% 4% Apr. ‘23 to 34% Nov. ‘23) both experienced drops in AI
deployment, with larger proportions of IT Professionals
Japan 15% 46% 34% 5%
reporting AI exploration (China: 19% Apr. ‘23, 36% Nov.
Singapore 6% 41% 53%
‘23) (Japan: 27% Apr. ‘23, 46% Nov. ‘23).
South Korea* 6% 48% 40% 5%
Spain 18% 51% 28% AI deployment in Italy dropped from 52% in April 2023
UAE 10% 32% 58% to 36% in November 2023. Italian IT Professionals were
UK 17% 41% 37% 6% more likely to report in the second half of the year that
their business is not currently using or exploring AI
US 19% 38% 33% 10%
(13% Apr. ‘23, 23% Nov. ‘23).
LATAM 16% 34% 47%
Base IT Professionals at Enterprises (organizations > 1,000 employees): Global Enterprise = 2,342n, Australia = 92n, Canada = 147n, China = 316n, France = 151n, Germany = 154n, India = 215n, Italy = 112n, Japan = 169n, Singapore = 148n, South Korea = 94n, Spain = 6
101n, UAE = 168n, UK = 145n, US = 126n, LATAM = 204n
*Sample size is between 50 and 99
AI ADOPTION & INVESTMENTS
Enterprises within the financial services are most likely to be using AI, with nearly half of IT Professionals in that
industry reporting their enterprise has actively deployed AI.
Has your company adopted or explored using Artificial Intelligence (AI) as part of its business operations and digital transformation?
My company is not currently
My company is exploring, My company has actively
using, or exploring the
but has not deployed, AI in deployed AI as part of its Don't know/Not sure
use of, AI in its business
its business operations business operations
operations
Global Enterprise 15% 40% 42%
Financial Services Industry 15% 33% 49%
Telecommunications Industry 16% 45% 37%
24% 49% 18% 9%
Government Industry
Energy, Environment, Utilities Industry* 21% 51% 23% 5%
Automotive Industry* 13% 44% 37% 6%
Industrial Industry 11% 46% 42%
Healthcare Industry 20% 47% 25% 8%
Retail Industry 21% 42% 31% 7%
Travel & Transportation Industry* 13% 53% 31%
Base IT Professionals at Enterprises (organizations > 1,000 employees): Global Enterprise = 2,342n, Financial Services = 218n, Telecommunications = 103n, Government = 148n, Energy = 75n, Automotive = 68n, Industrial = 302n, Healthcare = 154n, Retail = 130n, Travel = 68n 7
*Sample size is between 50 and 99; Note: Media & Entertainment, Chemicals/Oil/Gas, and Aerospace & Defense Industry samples sizes are too low to show
AI ADOPTION & INVESTMENTS
About 2-in-5 IT Professionals indicate that their enterprise is implementing generative AI (38%), and another 42%
are currently exploring generative AI (42%).
ChatGPT has quickly raised awareness of generative AI. Is your company using generative AI?
We are actively We are currently exploring We are not exploring nor Don't know/Not sure
implementing generative AI generative AI actively implementing
generative AI
Apr. 2023 34% 40% 14% 12%
Nov. 2023 38% 42% 12% 8%
Base IT Professionals at Enterprises (organizations > 1,000 employees): April 2023 = 2,247n, November 2023 = 2,342n 8
AI ADOPTION & INVESTMENTS
Since April ’23, reported implementation of AI has gone up in Japan (+13%), Singapore (+14%), South Korea (+16%),
and the UK (+21%).
ChatGPT has quickly raised awareness of generative AI. Is your company using generative AI?
We are actively We are currently exploring We are not exploring nor Don't know/Not sure
implementing generative AI generative AI actively implementing
generative AI
Global Enterprise 38% 42% 12% 8%
Australia* 20% 50% 20% 11%
Canada 22% 55% 13% 10%
China 63% 34%
France 19% 44% 23% 15%
Germany 33% 46% 12% 8%
India 61% 34%
Italy 26% 41% 16% 17%
Japan 25% 47% 18% 10%
Singapore 43% 41% 11% 5%
South Korea* 27% 48% 16% 10%
Spain 30% 36% 21% 14%
UAE 52% 39% 7%
UK 32% 46% 14% 9%
US 29% 36% 14% 21%
LATAM 37% 45% 9% 10%
Base IT Professionals at Enterprises (organizations > 1,000 employees): Global Enterprise = 2,342n, Australia = 92n, Canada = 147n, China = 316n, France = 151n, Germany = 154n, India = 215n, Italy = 112n, Japan = 169n, Singapore = 148n, South Korea = 94n, Spain = 9
101n, UAE = 168n, UK = 145n, US = 126n, LATAM = 204n
*Sample size is between 50 and 99
AI ADOPTION & INVESTMENTS
Generative AI adoption is driven by enterprises already deploying AI in their business operations. 63% of IT
Professionals at large companies currently deploying AI also report that their company is implementing generative AI,
compared to only 17% of those at companies only exploring AI.
ChatGPT has quickly raised awareness of generative AI. Is your company using generative AI?
We are actively We are currently exploring We are not exploring nor Don't know/Not sure
implementing generative AI generative AI actively implementing
generative AI
Currently Deploying AI 63% 30% 4%
Exploring AI 17% 62% 12% 9%
Base IT Professionals at Enterprises (organizations > 1,000 employees): Currently Deploying AI = 984n, Exploring AI = 930n 10
AI ADOPTION & INVESTMENTS
4-in-10 or more of IT Professionals within the financial services, telecommunications, and industrial industries
indicate that their enterprise is implementing generative AI.
ChatGPT has quickly raised awareness of generative AI. Is your company using generative AI?
We are actively We are currently exploring We are not exploring nor Don't know/Not sure
implementing generative AI generative AI actively implementing
generative AI
Global Enterprise 38% 42% 12% 8%
Financial Services Industry 40% 42% 7% 11%
Telecommunications Industry 41% 40% 12% 8%
15% 45% 25% 15%
Government Industry
Energy, Environment, Utilities Industry* 27% 47% 17% 9%
Automotive Industry* 37% 46% 9% 9%
Industrial Industry 43% 45% 9% 4%
Healthcare Industry 23% 43% 21% 13%
Retail Industry 22% 41% 24% 14%
Travel & Transportation Industry* 37% 43% 16% 4%
Base IT Professionals at Enterprises (organizations > 1,000 employees): Global Enterprise = 2,342n, Financial Services = 218n, Telecommunications = 103n, Government = 148n, Energy = 75n, Automotive = 68n, Industrial = 302n, Healthcare = 154n, Retail = 130n, Travel = 68n 11
*Sample size is between 50 and 99; Note: Media & Entertainment, Chemicals/Oil/Gas, and Aerospace & Defense Industry samples sizes are too low to show
AI ADOPTION & INVESTMENTS
Companies with 1,000 or fewer employees are less likely than enterprises to be adopting general AI and generative AI.
Has your company adopted or explored using Artificial Intelligence (AI) as part of its business operations and digital transformation?
ChatGPT has quickly raised awareness of generative AI. Is your company using generative AI?
G ener al AI Adopt ion
My company is not currently using, or exploring the use of, AI in its business operations My company is exploring, but has not deployed, AI in its business operations
My company has actively deployed AI as part of its business operations Don't know/Not sure
Company ≤ 1,000 25% 46% 24% 5%
Enterprise (Company > 1,000) 15% 40% 42% 3%
G ener at ive AI Adopt ion
We are not exploring nor actively implementing generative AI We are currently exploring generative AI We are actively implementing generative AI Don't know/Not sure
Company ≤ 1,000 22% 44% 25% 9%
Enterprise (Company > 1,000) 12% 42% 38% 8%
Base IT Professionals: Enterprises (organizations > 1,000 employees) = 2,342n, Company ≤ 1,000 = 6,242n 12
AI ADOPTION & INVESTMENTS
Investment in AI has remained relatively stable since April 2022.
Which of the following best describes your company's AI investment over the last 24 months? [Among IT Professionals at companies currently exploring
or deploying AI]
We have There has been
We have paused
We have accelerated stopped/decreased no change in my
our rollout and/or None of the above
our rollout of AI our rollout and/or company's investment
investment in AI
investment in AI and/or rollout of AI.
Apr. 2022 60% 12% 6% 20%
Apr. 2023 59% 13% 6% 19% 2%
Nov. 2023 59% 12% 6% 21% 2%
Base IT Professionals at Enterprises (organizations > 1,000 employees) exploring/deploying AI: April 2022 = 2,005n, April 2023 = 1,767n, November 2023 = 1,914n 13
AI ADOPTION & INVESTMENTS
59% of IT Professionals at enterprises deploying or exploring AI indicate that their organization has accelerated the
AI rollout in the past 24 months, and only around 1-in-5 (21%) say that their investment has remained unchanged.
Which of the following best describes your company's AI investment over the last 24 months? [Among IT Professionals at companies currently exploring
or deploying AI]
We have There has been
We have paused
We have accelerated stopped/decreased no change in my
our rollout and/or None of the above
our rollout of AI our rollout and/or company's investment
investment in AI
investment in AI and/or rollout of AI.
Global Enterprise 59% 12% 6% 21%
Australia* 38% 10% 7% 41% 4%
Canada 35% 16% 9% 36% 4%
China 85% 5% 4% 6%
France 45% 10% 7% 36%
Germany 52% 16% 4% 24%
India 74% 11% 12%
Italy* 61% 10% 5% 24%
Japan 50% 7% 31% 9%
Singapore 60% 8% 4% 28%
South Korea* 49% 12% 11% 27%
Spain* 48% 28% 6% 14% 5%
UAE 72% 13% 5% 8%
UK 40% 25% 6% 25% 4%
US* 46% 17% 4% 28% 6%
LATAM 67% 10% 8% 15%
Base IT Professionals at Enterprises (organizations > 1,000 employees) exploring/deploying AI: Global Enterprise = 1,914n, Australia = 73n, Canada = 125n, China = 272n, France = 107n, Germany = 117n, India = 184n, Italy = 82n, Japan = 135n, Singapore = 138n, South 14
Korea = 83n, Spain = 80n, UAE = 151n, UK = 112n, US = 90n, LATAM = 165n
*Sample size is between 50 and 99
AI ADOPTION & INVESTMENTS
IT Professionals in the automotive and industrial industries are most likely to report their enterprise has accelerated
AI investments in the past two years.
Which of the following best describes your company's AI investment over the last 24 months? [Among IT Professionals at companies currently exploring
or deploying AI]
We have There has been
We have paused
We have accelerated stopped/decreased no change in my
our rollout and/or None of the above
our rollout of AI our rollout and/or company's investment
investment in AI
investment in AI and/or rollout of AI.
Global Enterprise 59% 12% 6% 21%
Financial Services Industry 54% 13% 4% 26%
Telecommunications Industry* 45% 14% 6% 33%
35% 15% 8% 38%
Government Industry*
Energy, Environment, Utilities Industry* 47% 24% 9% 15% 5%
Automotive Industry* 73% 18% 5%
Industrial Industry 65% 9% 6% 18%
Healthcare Industry 39% 22% 10% 25% 4%
Retail Industry* 49% 13% 6% 31%
Travel & Transportation Industry* 46% 18% 7% 28%
Base IT Professionals at Enterprises (organizations > 1,000 employees) exploring/deploying AI: Global Enterprise = 1,914n, Financial Services = 179n, Telecommunications = 84n, Government = 99n, Energy = 55n, Automotive = 55n, Industrial = 266n, Healthcare = 110n, 15
Retail = 94n, Travel = 57n
*Sample size is between 50 and 99; Note: Media & Entertainment, Chemicals/Oil/Gas, and Aerospace & Defense Industry samples sizes are too low to show
AI ADOPTION & INVESTMENTS
Research and development (44%), reskilling/workforce development (39%), and building proprietary AI solutions
(38%) are the top AI investments at large organizations exploring or deploying AI.
How does your company plan to invest in AI adoption over the next 12-months? Please select all that apply. [Among IT Professionals at companies
currently exploring or deploying AI]
Global South
Australia* Canada China France Germany India Italy* Japan Singapore Spain* UAE UK US* LATAM
Enterprise Korea*
Research & Development 44% 49% 41% 41% 36% 35% 67% 32% 27% 51% 51% 36% 45% 43% 51% 48%
Reskilling and workforce development 39% 36% 42% 42% 33% 32% 55% 24% 30% 43% 37% 22% 44% 36% 38% 38%
Build proprietary AI solutions 38% 30% 23% 53% 28% 39% 53% 40% 34% 37% 23% 30% 44% 33% 29% 35%
Augmenting human tasks with digital labor 33% 34% 26% 40% 16% 40% 40% 26% 24% 33% 33% 31% 39% 39% 33% 30%
Off-the-shelf AI applications 32% 21% 22% 39% 25% 36% 26% 26% 38% 25% 24% 32% 44% 21% 28% 45%
Embed AI into current applications and
29% 33% 28% 26% 26% 30% 42% 18% 24% 40% 27% 22% 24% 21% 18% 41%
processes
Off-the-shelf tools to build our own
29% 14% 19% 43% 16% 32% 32% 17% 31% 32% 25% 16% 30% 21% 20% 38%
applications and models
Don't know/Not sure 4% 8% 7% 0% 5% 3% 1% 5% 8% 3% 1% 5% 0% 7% 12% 2%
Other 0% 1% 1% 0% 0% 1% 0% 0% 1% 1% 0% 1% 0% 0% 0% 0%
Base IT Professionals at Enterprises (organizations > 1,000 employees) exploring/deploying AI: Global Enterprise = 1,914n, Australia = 73n, Canada = 125n, China = 272n, France = 107n, Germany = 117n, India = 184n, Italy = 82n, Japan = 135n, Singapore = 138n, South
Korea = 83n, Spain = 80n, UAE = 151n, UK = 112n, US = 90n, LATAM = 165n 16
*Sample size is between 50 and 99
Note: dark green shading indicates the most-chosen statements while light green shading indicates the least-chosen statements within a specific market
AI ADOPTION & INVESTMENTS
Among enterprises implementing or exploring generative AI, most are using either in-house technology (43%) or
open-source technology (32%), with reported use of each remaining relatively unchanged since April 2023.
Are you using in-house technology, open source technology, or working with technology partner/provider? [Among IT Professionals at companies
currently exploring or implementing generative AI]
We are using in-house technology We are using open source technology We are working with a technology
partner/provider
Apr. 2023 45% 32% 23%
Nov. 2023 43% 32% 25%
Base IT Professionals at Enterprises (organizations > 1,000 employees) exploring/implementing generative AI : April 2023 = 1,660n, November 2023 = 1,873n 17
AI ADOPTION & INVESTMENTS
IT Professionals at companies exploring or deploying generative AI in Australia, Italy, Japan, and the UK are more
likely than the global average to report that their companies are using open-source technology.
Are you using in-house technology, open source technology, or working with technology partner/provider? [Among IT Professionals at companies
currently exploring or implementing generative AI]
We are using in-house technology We are using open source technology We are working with a technology
partner/provider
Global Enterprise 43% 32% 25%
Australia* 22% 45% 33%
Canada 30% 35% 35%
China 50% 23% 27%
France* 43% 35% 22%
Germany 51% 30% 20%
India 46% 32% 22%
Italy* 29% 55% 16%
Japan 34% 47% 19%
Singapore 41% 25% 34%
South Korea* 31% 41% 27%
Spain* 50% 29% 21%
UAE 51% 25% 24%
UK 39% 42% 19%
US* 48% 32% 21%
LATAM 49% 24% 27%
Base IT Professionals at Enterprises (organizations > 1,000 employees) exploring/implementing generative AI : Global Enterprise = 1,873n, Australia = 64n, Canada = 113n, China = 305n, France = 95n, Germany = 122n, India = 204n, Italy = 75n, Japan = 122n, Singapore = 18
124n, South Korea = 70n, Spain = 66n, UAE = 153n, UK = 112n, US = 82n, LATAM = 166n
*Sample size is between 50 and 99
AI ADOPTION & INVESTMENTS
In-house technology is most likely to be utilized in the financial services, telecommunications, energy, and travel
industries.
Are you using in-house technology, open source technology, or working with technology partner/provider? [Among IT Professionals at companies
currently exploring or implementing generative AI]
We are using in-house technology We are using open source technology We are working with a technology
partner/provider
Global Enterprise 43% 32% 25%
Financial Services Industry 48% 28% 23%
Telecommunications Industry* 46% 39% 16%
37% 31% 31%
Government Industry*
Energy, Environment, Utilities Industry* 45% 36% 18%
Automotive Industry* 36% 32% 32%
Industrial Industry 41% 26% 33%
Healthcare Industry 34% 44% 23%
Retail Industry* 37% 40% 23%
Travel & Transportation Industry* 50% 24% 26%
Base IT Professionals at Enterprises (organizations > 1,000 employees) exploring/implementing generative AI : Global Enterprise = 1,873n, Financial Services = 179n, Telecommunications = 83n, Government = 89n, Energy = 55n, Automotive = 55n, Industrial = 265n, 19
Healthcare = 101n, Retail = 81n, Travel = 54n
*Sample size is between 50 and 99; Note: Media & Entertainment, Chemicals/Oil/Gas, and Aerospace & Defense Industry samples sizes are too low to show
AI ADOPTION & INVESTMENTS
Enterprises with more established generative AI practices are more likely to be using in-house technology over open-
source technology. Similarly, in-house technology is most likely to be used by companies with more than 1,000
employees exploring or implementing generative AI.
Are you using in-house technology, open source technology, or working with technology partner/provider? [Among IT Professionals at companies
currently exploring or implementing generative AI]
We are using in-house technology We are using open source technology We are working with a technology partner/provider
Company ≤ 1,000 38% 41% 21%
Enterprise (Company > 1,000) 43% 32% 25%
Enterprise Implementing Generative AI 59% 20% 21%
Enterprise Exploring Generative AI 29% 44% 28%
Base IT Professionals at companies exploring/implementing generative AI: Enterprises (organizations > 1,000 employees) = 1,873n, Company ≤ 1,000 = 4,288n, Enterprise Implementing Generative AI = 894n, Enterprise Exploring Generative AI = 979n 20
AGENDA
AI AD OPT ION & IN VEST MENTS
D R IVER S & BAR R IER S OF AI
CURRENT USES OF AI
AI ET H IC S AN D R ESPON SIBIL IT Y
AI’S IMPAC T ON EMPL OYEES
DRIVERS & BARRIERS OF AI
Most enterprises actively exploring or deploying AI have some form of AI strategy, with 27% reporting that their
company has an AI strategy for limited/specific use cases and about a third (32%) stating that their organization
already has a holistic strategy in place. 32% are in the process of developing an AI strategy.
Which of the following best describes your company's AI strategy? [Among IT Professionals at companies currently exploring or deploying AI]
My company had
My company has My company an AI strategy
an AI strategy, has a holistic but had to
My company does My company is
but it is strategy for discard it None of the
not have an AI developing an AI
focused on how it will use or have not above
strategy strategy
limited/specific AI across the been able to
use cases organization implement it
effectively
Global Enterprise 4% 32% 27% 32% 5%
Australia* 4% 44% 30% 19%
Canada 6% 37% 30% 25%
China 24% 27% 46%
France 7% 38% 33% 18% 4%
Germany 38% 32% 22% 4%
India 26% 19% 42% 11%
Italy* 5% 35% 32% 24% 4%
Japan 7% 30% 27% 30%
Singapore 33% 28% 28% 8%
South Korea* 31% 31% 34%
Spain* 5% 45% 24% 21% 4%
UAE 20% 25% 52%
UK 7% 40% 21% 19% 13%
US* 9% 38% 27% 19% 6%
LATAM 4% 34% 27% 31% 4%
Base IT Professionals at Enterprises (organizations > 1,000 employees) exploring/deploying AI: Global Enterprise = 1,914n, Australia = 73n, Canada = 125n, China = 272n, France = 107n, Germany = 117n, India = 184n, Italy = 82n, Japan = 135n, Singapore = 138n, South 22
Korea = 83n, Spain = 80n, UAE = 151n, UK = 112n, US = 90n, LATAM = 165n
*Sample size is between 50 and 99
DRIVERS & BARRIERS OF AI
Enterprises exploring AI are more likely to be in the beginning stages of AI strategy, while large organizations
deploying AI are more likely to have a holistic strategy in place.
Which of the following best describes your company's AI strategy? [Among IT Professionals at companies currently exploring or deploying AI]
My company had
My company has My company an AI strategy
an AI strategy, has a holistic but had to
My company does My company is
but it is strategy for discard it None of the
not have an AI developing an AI
focused on how it will use or have not above
strategy strategy
limited/specific AI across the been able to
use cases organization implement it
effectively
Currently Deploying AI 18% 23% 50% 7%
Exploring AI 6% 48% 31% 12%
Base IT Professionals at Enterprises (organizations > 1,000 employees) exploring/deploying AI: Currently Deploying AI = 984n, Exploring AI = 930n 23
DRIVERS & BARRIERS OF AI
Larger organizations exploring or deploying AI are more likely than smaller organizations to have a holistic AI
strategy in place (32% vs. 22%).
Which of the following best describes your company's AI strategy? [Among IT Professionals at companies currently exploring or deploying AI]
My company had
My company has My company an AI strategy
an AI strategy, has a holistic but had to
My company does My company is
but it is strategy for discard it None of the
not have an AI developing an AI
focused on how it will use or have not above
strategy strategy
limited/specific AI across the been able to
use cases organization implement it
effectively
Company ≤ 1,000 8% 36% 29% 22% 3%
Enterprise (Company > 1,000) 4% 32% 27% 32% 5%
Base IT Professionals at companies exploring/deploying AI: Enterprises (organizations > 1,000 employees) = 1,914n, Company ≤ 1,000 = 4,402n 24
DRIVERS & BARRIERS OF AI
Advances in AI making it more accessible (45%) is the top external driver of AI adoption at enterprises currently
exploring or deploying AI, followed by the need to reduce costs and automate key processes (42%) and the increasing
amount of AI embedded into standard off the shelf business applications (37%).
What external factors, if any, are helping drive AI adoption in your organization? Please select all that apply. [Among IT Professionals at companies
currently exploring or deploying AI]
Global South
Australia* Canada China France Germany India Italy* Japan Singapore Spain* UAE UK US* LATAM
Enterprise Korea*
Advances in AI that make it more accessible 45% 48% 46% 39% 32% 41% 59% 35% 42% 52% 52% 40% 41% 51% 42% 55%
Need to reduce costs and automate key
42% 48% 46% 35% 31% 40% 48% 35% 54% 49% 52% 31% 40% 37% 39% 41%
processes
The increasing amount of AI embedded into
37% 32% 34% 48% 29% 44% 47% 30% 25% 41% 27% 26% 35% 36% 27% 41%
standard off the shelf business applications
Competitive pressure 31% 41% 30% 24% 28% 33% 39% 28% 23% 41% 20% 16% 41% 30% 36% 27%
Directives from leadership 26% 30% 26% 20% 20% 23% 32% 13% 16% 33% 27% 20% 36% 26% 28% 35%
Labor or skills shortages 25% 32% 30% 22% 19% 32% 28% 9% 47% 24% 22% 19% 28% 29% 36% 9%
Pressure from consumers 25% 29% 20% 29% 18% 22% 34% 17% 9% 30% 20% 15% 33% 23% 24% 28%
Company culture 23% 19% 13% 28% 7% 21% 26% 27% 19% 26% 20% 29% 23% 25% 26% 26%
Environmental pressures 19% 15% 10% 23% 14% 13% 26% 15% 14% 20% 23% 15% 27% 26% 17% 16%
Legal and regulatory compliance pressures 18% 21% 16% 16% 18% 21% 22% 13% 15% 18% 18% 10% 19% 21% 23% 13%
Supply chain issues 18% 18% 19% 22% 12% 17% 28% 7% 13% 20% 14% 9% 25% 22% 22% 9%
Demands due to the Covid-19 pandemic 15% 10% 10% 21% 9% 9% 24% 5% 11% 19% 19% 11% 20% 17% 10% 14%
None of the above 1% 1% 1% 3% 1% 1% 1% 1% 1% 0% 1% 1% 0% 3% 2% 1%
Other 0% 1% 0% 0% 0% 0% 0% 0% 1% 1% 0% 1% 0% 1% 1% 1%
Base IT Professionals at Enterprises (organizations > 1,000 employees) exploring/deploying AI: Global Enterprise = 1,914n, Australia = 73n, Canada = 125n, China = 272n, France = 107n, Germany = 117n, India = 184n, Italy = 82n, Japan = 135n, Singapore = 138n, South
Korea = 83n, Spain = 80n, UAE = 151n, UK = 112n, US = 90n, LATAM = 165n 25
*Sample size is between 50 and 99
Note: dark green shading indicates the most-chosen statements while light green shading indicates the least-chosen statements within a specific market
DRIVERS & BARRIERS OF AI
Compared to AI projects 2 to 3 years ago, Enterprise IT Professionals consider accessible AI solutions (43%), the
increased prevalence of data, AI, and automation skills (42%), and AI tailored solutions (41%) the most important
changes in the industry.
Compared to AI projects 2-3 years ago, what are the most important changes you see in the industry? Please select no more than three.
Global South
Australia* Canada China France Germany India Italy Japan Singapore Spain UAE UK US LATAM
Enterprise Korea*
AI solutions are more accessible and easier to
43% 40% 41% 47% 38% 40% 57% 38% 36% 41% 49% 26% 48% 38% 40% 50%
deploy
Data, AI and automation skills are more prevalent,
teams are positioned to build, deploy, and manage 42% 35% 39% 53% 24% 34% 55% 27% 27% 49% 50% 41% 52% 43% 36% 43%
AI
AI solutions are better designed to fit the needs of
41% 34% 35% 51% 30% 37% 49% 38% 28% 47% 40% 38% 53% 32% 39% 46%
businesses
Businesses have clear data and AI strategies 31% 27% 24% 36% 21% 34% 38% 21% 21% 37% 29% 31% 48% 32% 22% 34%
Businesses have ethical guidelines in place for their
27% 32% 19% 35% 20% 21% 33% 21% 30% 26% 17% 16% 35% 32% 24% 25%
AI adoption
Don't know/Not sure 6% 10% 8% 1% 11% 8% 0% 5% 20% 3% 6% 6% 1% 4% 14% 2%
Other 0% 0% 0% 0% 0% 1% 0% 1% 1% 0% 1% 1% 0% 1% 1% 0%
Base IT Professionals at Enterprises (organizations > 1,000 employees): Global Enterprise = 2,342n, Australia = 92n, Canada = 147n, China = 316n, France = 151n, Germany = 154n, India = 215n, Italy = 112n, Japan = 169n, Singapore = 148n, South Korea = 94n, Spain =
101n, UAE = 168n, UK = 145n, US = 126n, LATAM = 204n 26
*Sample size is between 50 and 99
Note: dark green shading indicates the most-chosen statements while light green shading indicates the least-chosen statements within a specific market
DRIVERS & BARRIERS OF AI
Barriers to successful AI adoption have stayed consistent from April, although high prices are less likely to be a
hinderance in November (April ‘23 26% vs. Nov. ‘23 21%).
What, if anything, is hindering successful AI adoption for your business? Please select all that apply. [Among IT Professionals at companies currently
exploring or deploying AI]
33%
We have limited AI skills, expertise or knowledge
32%
25%
We have too much data complexity
24%
23%
We have ethical concerns
19%
22%
AI projects are too complex or difficult to integrate and scale
25%
21%
We have a lack of tools/platforms for developing AI models
23%
21%
The price is too high
26%
We do not have the use cases defined or the end user research 17% Nov. 2023
needed to get started 16%
Apr. 2023
17%
We do not have a holistic AI strategy in place
18%
17%
We do not have the ability to properly govern our AI models
17%
We are locked-in to one vendor (AI and Cloud tied to one single 13%
vendor) 16%
3%
None of the above
4%
0%
Other
1%
Nothing is technically hindering successful AI adoption for my 11%
business 10%
Base IT Professionals at Enterprises (organizations > 1,000 employees) exploring/deploying AI: April 2023 = 1,767n, November 2023 = 1,914n 27
DRIVERS & BARRIERS OF AI
Despite the increased prevalence in AI related skills, IT Professionals at enterprises exploring or deploying AI are
most likely to express that limited A |
153 | mit_edu | Advanced_20Technology_20Adoption_20-_20Selection_20of_20Causal_20Effects.pdf | AEA Papers and Proceedings 2023, 113: 210–214
https://doi.org/10.1257/pandp.20231037
ROBOT AND AUTOMATION: NEW INSIGHTS FROM MICRO DATA‡
Advanced Technology Adoption: Selection or Causal Effects?†
By Daron Acemoglu, Gary Anderson, David Beede, Catherine Buffington,
Eric Childress, Emin Dinlersoz, Lucia Foster, Nathan Goldschlag,
John Haltiwanger, Zachary Kroff, Pascual Restrepo, and Nikolas Zolas*
Advanced technologies, including robotics, Our work documented these facts:
artificial intelligence AI , and software sys-
( )
tems, are thought to be spreading rapidly in i The share of adopting firms remains
()
industrialized economies. In Acemoglu, Aaron, low for AI and robotics 3.2 percent
(
et al. 2022 , we used the 2019 Annual Business and 2 percent of firms, respectively and
( ) )
Survey ABS to provide a comprehensive over- rises to 19.6 and 40.2 percent for equip-
( )
view of the adoption of AI, robotics, dedicated ment and software, respectively.
equipment, specialized software, and cloud
computing for US firms in all sectors during ii Adoption is concentrated in large firms.
( )
2016–2018.
iii As a result, a high share of workers is
( )
exposed to these technologies, espe-
cially in manufacturing. For exam-
‡Discussants: Betsey Stevenson, University of Michigan; ple, 12–64 percent of US workers
James Bessen, Boston University; Gino Gancia, Queen and 22–72 percent of US manufac-
Mary University of London; Susan Helper, Case Western turing workers are exposed to these
Reserve University. technologies.
* Acemoglu: MIT email: [email protected]; Anderson:
( )
National Center for Science and Engineering Statistics email: iv A significant share of adopters, rang-
[email protected]; Beede: US Census Bureau ( email: ( )
) ( ing from 30 percent for specialized
[email protected]; Buffington: US Census Bureau
) software to 65 percent for robotics by
email: [email protected]; Dinlersoz:
( US Census Bureau email: emin.m.dinl) ersoz@census. employment weight, report using these
(
gov; Foster: US Census Bureau email: lucia.s.foster@cen- advanced technologies for automation.
) (
sus.gov ); Goldschlag: US Census Bureau (email: nathan. In total, 30.4 percent of US workers and
[email protected]; Kroff: US Census Bureau email:
) ( 52 percent of manufacturing workers are
[email protected]; Zolas: US Census Bureau
email: nikolas.j.zolas@c) ensus.gov; Childress: George employed at firms using these technolo-
( )
Mason University email: [email protected]; Haltiwanger: gies for automation.
( )
University of Maryland email: [email protected]; Restrepo:
( )
Boston University email: [email protected]. Any opinions and
( ) v Consistent with the use of these advanced
conclusions expressed herein are those of the authors and do ( )
technologies for automation, adopters
not reflect the views of the US Census Bureau. All results have
been reviewed to ensure that no confidential information is have higher labor productivity and lower
disclosed. The Census Bureau’s Disclosure Review Board and labor shares.
Disclosure Avoidance Officers have reviewed this data product
for unauthorized disclosure of confidential information and
vi Firms report that these technologies
have approved the disclosure avoidance practices applied to ( )
this release. DRB Approval Numbers: CBDRB-FY21-058, increase their demand for skills but do
CBDRB-FY21-316, CBDRB-FY22-057, CBDRB-FY22- not necessarily expand employment.
ESMD006-011, CBDRB-FY22-411, CBDRB-FY23-034,
CBDRB-FY23-112. DMS number 7508509.
This paper revisits the second fact—the rea-
† Go to https://doi.org/10.1257/pandp.20231037 to visit
sons why firms adopting advanced technologies
the article page for additional materials and author disclo-
sure statements. are larger. In principle, this could be for two
( )
210
VOL. 113 ADVANCED TECHNOLOGY ADOPTION: SELECTION OR CAUSAL EFFECTS? 211
different reasons. Either adoption of advanced
technologies causally expands employment, or 7
selection leads larger firms to more adoption.
6
For example, already-large firms may have a
greater likelihood of adopting advanced tech- 5
nologies because of fixed costs, or firms that are
growing fast for other reasons may also be better 4
at adopting and using these technologies.
These two explanations have different impli- 3
cations. The former would suggest that advanced
2
technologies contribute to employment growth,
at least at the firm level the i ndustry-level impli- 1
(
cations could differ from the firm-level ones, as
pointed out in Acemoglu, Lelarge, and Restrepo 0
2020 and Koch, Manuylov, and Smolka 2021 .
)
The latter would weigh in favor of limited
employment gains even in adopting firms and
would caution against firm-level explorations
using ordinary least squares or event study strat-
egies to uncover the effects of advanced technol-
ogy adoption.
Our results favor the selection interpretation.
Using data from the Longitudinal Business
Database LBD , we document that adopters
( ) were already large and growing faster before
AI, robotics, cloud computing, and specialized
software systems became broadly available.1
We also find that employment trends at adopt-
ing firms remained largely unchanged after the
widespread use of these technologies. Persistent
size and growth differences between adopters
and nonadopters imply that fi rm-level estimates
of the effects of advanced technologies must be
interpreted with caution.
I. Adoption and Firm Size
We first provide graphical evidence on the
relationship between firm size and the adoption
of AI and robotics. We focus on these technol-
ogies because they have received considerable
attention in recent empirical work. Figure 1 plots percentiles within detailed s ix-digit industries.2
adoption rates for firms in 36 size and age cate- The figure also reports the average adoption rate
gories, defined in terms of employment and age for firms in each size class.
2 We assign firms to their main six-digit North American
Industry Classification System industry in terms of payroll
1 These statements refer to employment. We document in across all its establishments. Employment percentiles are
Acemoglu, Anderson, et al. 2022 that firms’ adoption of defined based on the employment distribution in each indus-
( )
advanced technologies is associated with an increase in sales try. By construction, Figure 1 isolates differences in adop-
and a reduction in their labor share. The same pattern for tion rates across firms of different size operating in the same
French manufacturing is documented in Acemoglu, Lelarge, narrowly defined industry and controls for size differences
and Restrepo 2020. between manufacturing and nonmanufacturing firms.
( )
tnecrep
smrif
erahS
)
(
Share of firms using AI, 2016–2018
0–50 50–75 75–90 90–95 95–99 99–100
Firm size percentiles
8
7
6
5
4
3
2
1
0
tnecrep
smrif
erahS
)
(
Age percentiles
0–25 90–95
25–50 95–100 50–75 Overall
75–90
Share of firms using robotics, 2016–2018
Age percentiles
0–25 90–95
25–50 95–100
50–75 Overall
75–90
0–50 50–75 75–90 90–95 95–99 99–100
Firm size percentiles
Figure 1. Adoption of AI and Robotics for Firms in
Different Size and Age Categories
Notes: The figure plots adoption rates for AI and robotics by
firm age and size percentiles within detailed six-digit indus-
tries. See Acemoglu, Anderson, et al. 2022 for similar fig-
( )
ures for the remaining technologies.
Source: 2019 ABS
212 AEA PAPERS AND PROCEEDINGS MAY 2023
Adoption rises with size for all technologies
in the ABS: 5.5 percent of firms in the top per-
centile of their industries’ employment distribu-
4.8
tion use AI, 5.1 percent use robots, 31.4 percent
4.6
use dedicated equipment, 67.4 percent use spe-
4.4
cialized software, and 63.5 percent use cloud
4.2
computing. In contrast, the adoption rate among
4
firms in the fiftieth to seventy-fifth percentile
3.8
of industries’ employment distribution is much 3.6
lower: 3.1 percent for AI, 1.7 percent for robots, 3.4
18.6 percent for dedicated equipment, 39.6 per-
3.2 cent for specialized software, and 33.4 percent 3
for cloud. 2.8
II. Firm Employment Histories
The previous section documented sizable dif-
ferences in employment levels between adopt- Figure 2. Employment Trends for Establishments in
Robot-Using Firms and Others for 1978–2018
ing and nonadopting firms for robotics and AI .
( )
We now explore whether employment histories,
Notes: The figure plots the inverse hyperbolic sine of
in terms of both levels and trends, differ between employment in establishments associated with firms using
adopters and nonadapters. robots in the 2019 ABS lines with circles and those asso-
( )
Because LBD does not contain consistent ciated with nonrobot users in the 2019 ABS (dashed lines ).
For each cohort, we report employment numbers for the
information on firm-establishment histories, we
years following its entry into the LBD.
create a p seudo–firm establishment panel that
tracks employment in all establishments asso- Sources: 2019 ABS and 1978–2018 LBD
ciated with each firm in the ABS technology
module in 2018. We then conduct our empiri-
cal analysis at the level of these establishments
between 1978 and 2018.3 late 1990s and early 2000s. Third, employment
Figure 2 focuses on the differential employ- dynamics of adopters’ establishments seem
ment histories of adopters and nonadopters of unaffected by rising adoption of robots in recent
robotics for illustration purposes. It plots the decades.
evolution of average employment by cohort for To explore these patters for all technologies,
establishments in adopting and nonadopting we turn to the following regression model:
firms.4 The figure reveals three key patterns.
First, establishments in adopting firms are ini- 1 y Adopter
( ) j,i,c,t = αc+βi,t+γc× j
tially larger have higher employment than
( )
establishments in nonadopting firms. These size Adopter ,
+ δt× j+ϵj,i,c,t
differences are present at an early age and grow
over time, especially for early cohorts. Second, for an establishment j in industry i , cohort c ,
differences in employment levels and growth in year t . The left-hand-side variable is the
rates precede the period of rapid robot adoption inverse hyperbolic sine IHS of establishment
( )
in the United States, which took place in the employment, which allows us to include zeros in
our analysis. The right-hand-side variables are
cohort dummies ; industry-by-year dummies
αc
, which account for differences in employ-
3 In particular, this p seudopanel follows the same estab- βi,t
lishments over time, even though some of these establish- ment trends by four-digit industries; and cohort
ments may not have belonged to the firm in question in the and growth effects depending on adopter status
past. See Foster et al. 2016 for more details on this strategy as measured by the adopter dummy Adopter .
to track activity of firm( s bac) k in time. ( j)
These terms allow adopters to have different ini-
4 The first year in the LBD is 1976. We do not observe
tial levels differences by cohort and different
the exact age of establishments that existed at this point and ( )
assign them to a “ pre-77” cohort. growth dynamics different time effects .
( )
tnemyolpme
tnemhsilbatse
fo
SHI
Employment trends for establishments of
robot adopters and nonadopters, 1978–2018
Adopters Nonadopters
Pre 77
Pre 77 85–91
77–84
92–98
77–84 85–91 92–98 99–05 99–05
06–12 06–12
1970 1975 1980 1985 1990 1995 2000 2005 2010 2015 2020
VOL. 113 ADVANCED TECHNOLOGY ADOPTION: SELECTION OR CAUSAL EFFECTS? 213
adopting firms is significantly greater than the
size of nonadopters at the same point in time.
35 For example, establishments at r obot-adopting
30
firms from the 1977–1984 cohort were initially
25 24.3 percent larger than establishments of firms
20 not adopting robotics technology. The same dif-
15 ference is 14.7 percent for robot-adopting firms
from the 1999–2005 cohort.
10
Panel B depicts the estimates of , which mea-
5 δt
sures the differential establishment employ-
0 ment growth of adoptin( g firms. It con) firms that
5 establishment employment for adopters grew
−
10 more rapidly than it did for n onadopters. For
−
example, from 1978–1984 to 1992–1998, estab-
lishments of robot-adopting firms expanded
their employment by 11.1 percent more than
nonadopters. Notably, for most technologies,
these differential growth experiences long pre-
dated the periods of high adoption in the United
States as a whole. Indeed, robotics, AI, special-
ized software systems, and cloud computing were
not spreading rapidly before the late 1990s.6 For
example, the adoption of AI concentrates in the
2016–2018 period see Acemoglu, Autor, et al.
( 2022 , while robot adoption gained prominence
)
in the late 1990s and the 2000s see Acemoglu
(
and Restrepo 2020 . Yet, establishments of AI
)
and robot-adopting firms were larger and grew
more rapidly than those of nonadopters decades
before these periods.
Panel B also shows that the differential
employment growth of adopters relative to
nonadopters is unaffected by the increased
adoption of these technologies in recent years.
If anything, establishments in adopting firms
grew at more comparable rates to establish-
ments in nonadopting firms in recent years. For
example, our estimates in Panel B imply that
the yearly growth differential for establishments
Figure 3 depicts estimates from equation 1 in robot-adopting firms relative to nonadopters
( )
separately for the five technologies in the ABS. went from 0.8 percent per year in 1978–1998 to
Panel A presents estimates of , which 0.4 percent in 1999–2018.
γc+δc
compare the initial establishment size of adopt-
ing firms of cohort c to the size of nonadopting III. Discussion
firms at the time of entry.5 The results in this
panel show that, consistent with our discussion Figures 2 and 3 show that establishments in
for robotics adoption in Figure 2, the initial adopting firms were initially larger and grew
size in terms of establishment employment of more rapidly than nonadopters, even before
( )
5 The interaction terms give employment differences at 6 The exception is dedicated equipment, such as
γc
the base period. Adding gives an estimate of employ- computer–numerically controlled machines, whose wide-
γc+δc
ment differences in the first period each cohort enters the spread adoption dates back to the early 1970s and is studied
LBD. in detail in Boustan, Choi, and Clingingsmith 2022.
( )
,secnereffid
tnemyolpme
laitinI
sretpodanon
susrev
sretpoda
)cδ
+
cγ(
Panel A. Differences in initial employment
Pre-7777–84 85–91 92–98 99–05 06–12 13–18
Cohorts
25
20
15
10
5
0
,secnereffid
htworg
tnemyolpmE
sretpodanon
susrev
sretpoda
)tδ(
Technology
AI
Robotics
Equipment
Software
Cloud computing
Panel B. Differences in employment growth
Technology
AI Robotics
Equipment
Software
Cloud computing
85–91 92–98 99–05 06–12 13–18
Period
Figure 3. Differential Employment Dynamics for
Establishments in Adopting Firms Relative to Others
Notes: Panel A plots estimates of from equation 1 ,
γc+δc( ( )) which measures the differential establishment employment
size for adopter firms relative to nonadopters. Panel B plots
, which measures the differential establishment employ- δt
ment growth for adopter firms relative to nonadopters.
Sources: 2019 ABS and 1978–2018LBD
214 AEA PAPERS AND PROCEEDINGS MAY 2023
the adoption of advanced technologies intensi- REFERENCES
fied in recent years. These patterns support the
view that adopters of advanced technologies are Acemoglu, Daron, Gary W. Anderson, David N.
differentially selected and were already large Beede, Cathy Buffington, Eric E. Childress,
and on differential growth trajectories. Emin Dinlersoz, Lucia S. Foster, et al. 2022.
The figures also document that the difference in “Automation and the Workforce: A Firm-Level
employment dynamics between adopting firms’ View from the 2019 Annual Business Survey.”
establishments and others has remained largely NBER Working Paper 30659.
unchanged or become less pronounced in recent Acemoglu, Daron, David Autor, Jonathon Hazell,
years as adoption intensifies. This is the opposite and Pascual Restrepo. 2022. “Artificial Intel-
of what one would expect if advanced technolo- ligence and Jobs: Evidence from Online
gies caused adopting firms to expand their employ- Vacancies.” Journal of Labor Economics 40
ment. Instead, it points to small or negative effects S1 : S293–340.
( )
of automation technologies on firm employment Acemoglu, Daron, Claire Lelarge, and Pascual
trajectories. Restrepo. 2020. “Competing with Robots:
The possibility that technology does not lead Firm-Level Evidence from France.” AEA
to large employment expansions at adopting Papers and Proceedings 110: 383–88.
firms aligns with the fact that a significant share Acemoglu, Daron, and Pascual Restrepo. 2020.
of adopters report using advanced technologies “Robots and Jobs: Evidence from US Labor
for automation. In contrast to other applications Markets.” Journal of Political Economy 128
of advanced technologies, automation reduces 6 : 2188–244.
( )
production cost by displacing workers from their Boustan, Leah Platt, Jiwon Choi, and David Cling-
roles, creating an ambiguous effect on firm-level ingsmith. 2022. “Automation after the Assem-
employment. This possibility also aligns with bly Line: Computerized Machine Tools,
firms’ self-assessments on the effects of these Employment and Productivity in the United
technologies, which point to ambiguous effects States.” NBER Working Paper 30400.
of advanced technologies on employment levels Foster, Lucia, John Haltiwanger, Shawn Klimek,
Acemoglu, Anderson et al. 2022 . C.J. Krizan, and Scott Ohlmacher. 2016. “The
( )
One challenge when interpreting our findings Evolution of National Retail Chains: How We
is that we do not know the exact adoption date Got Here.” In Handbook on the Economics of
of these technologies. Currently, the ABS data Retailing and Distribution, edited by Emek
only tell us whether a firm used a technology Basker, 7–37. Cheltenham, UK: Edward Elgar
in 2016–2018. Future waves of the ABS tech- Publishing.
nology module will measure year of adoption, Koch, Michael, Ilya Manuylov, and Marcel
providing a more accurate picture of how tech- Smolka. 2021. “Robots and Firms.” Economic
nology changes firm employment dynamics. Journal 131 638 : 2553–84.
( ) |
154 | mit_edu | Regulating_20Transformative_20Technologies.pdf | AER: Insights 2024, 6(3): 359–376
https://doi.org/10.1257/aeri.20230353
Regulating Transformative Technologies†
By Daron Acemoglu and Todd Lensman*
Transformative technologies like generative AI promise to acceler-
ate productivity growth across many sectors, but they also present
new risks from potential misuse. We develop a multisector technol-
ogy adoption model to study the optimal regulation of transforma-
tive technologies when society can learn about these risks over time.
Socially optimal adoption is gradual and typically convex. If social
damages are large and proportional to the new technology’s pro-
ductivity, a higher growth rate paradoxically leads to slower opti-
mal adoption. Equilibrium adoption is inefficient when firms do not
internalize all social damages, and sector-independent regulation is
helpful but generally not sufficient to restore optimality. JEL D21,
(
H21, H25, O31, O33
)
Recent breakneck advances in generative artificial intelligence have simultane-
( )
ously raised hopes of productivity gains in many sectors and fears that this technol-
ogy will be used for nefarious purposes, even posing an existential risk comparable
to nuclear war.1 Some experts have called to slow down or pause the development
and adoption of AI technologies,2 partly because a slower rollout might provide
time to identify danger areas and craft appropriate regulations. However, there is
little economic analysis of these issues, and it is unclear whether slowing the devel-
opment and adoption of a promising, transformative technology ever makes sense.
In this paper, we develop a framework to provide a first set of insights on these
questions. We consider a multisector economy that initially uses an old technology
but can switch to a new, transformative technology. This technology is transforma-
tive both because it enables a higher growth rate of output and because it is general
purpose and can be adopted across all sectors of the economy. It also poses new
risks. We model these by assuming that there is a positive probability of a disaster,
meaning that the technology will turn out to have many harmful uses. If a disaster
is realized, some of the sectors that had started using the new technology may not
be able to switch away from it, despite the social damages. Whether there will be a
* Acemoglu: Massachusetts Institute of Technology, Department of Economics email: [email protected];
( )
Lensman: Massachusetts Institute of Technology, Department of Economics email: [email protected]. Peter
( )
Klenow was the coeditor for this article. We thank Glen Weyl for several useful discussions; Joshua Gans,
Chad Jones, and three anonymous referees for comments; and the Hewlett Foundation and the National Science
Foundation for financial support.
† Go to https://doi.org/10.1257/aeri.20230353 to visit the article page for additional materials and author
disclosure statements.
( )
1 https://www.nytimes.com/2023/05/30/technology/ai-threat-warning.html
2 https://futureoflife.org/open-letter/pause-giant-ai-experiments/
359
360 AER: INSIGHTS SEPTEMBER 2024
disaster is initially unknown, and society can learn about it over time. Critically, we
assume that the greater are the new technology’s capabilities, the more damaging it
will be when used for harmful purposes.3
In this environment, we study socially optimal and equilibrium adoption deci-
( )
sions. We first show that it is optimal to adopt the new technology gradually because
this enables greater learning. If all sectors immediately adopted and the disaster
transpired, many of them would not be able to switch back and avoid the social
damages. Gradual adoption allows society to gain from the new technology while
updating its beliefs about whether it will have socially damaging uses. As more time
passes without disaster, the belief that there will be a disaster declines “no news
(
is good news” . As society becomes more optimistic, it is optimal to adopt the new
)
technology across a larger number of sectors. Under weak conditions, this adop-
tion path is slow and convex, accelerating only after society is fairly certain that a
disaster will not occur. A simple quantitative example indicates that, for reasonable
parameters for the new technology’s growth advantage and disaster risk, optimal
adoption can be very slow.
Perhaps surprisingly, we demonstrate that adoption should be slower when the
new technology has a higher growth rate and damages from a disaster are large.
This is for two reasons. First, since damages after a potential disaster increase with
the new technology’s capabilities, a higher growth rate means that damages also
grow more quickly. Second, with a higher growth rate, the effective discount rate for
future output declines, so that short delays in adoption are not very consequential
for discounted utility.
Compared to optimal adoption, equilibrium adoption is inefficiently fast if private
firms internalize only part of the social damages from a disaster. Even the order in
which sectors adopt the new technology can differ between the equilibrium and the
optimum—sectors that have high social damages are not necessarily those that have
high private damages for adopters.4 Finally, we discuss how regulatory schemes can
help to close the gap between optimal and equilibrium adoption. Pigouvian taxes,
use taxes, or adoption taxes that are sector specific can fully implement optimal
adoption. When s ector-specific policies are not feasible, it is generally not possible
to implement optimal technology choices, but regulation can still increase welfare
by prohibiting use of the new technology in the sectors with the largest potential for
harm until the risk of a disaster is sufficiently low.
This paper is a first attempt to study the consequences and regulation of trans-
formative technologies that can be used for good or bad. Our conclusions naturally
depend on our modeling assumptions and should be interpreted with caution.
There are three literatures on which we build. The first is a growing literature
on economic disasters e.g., Rietz 1988; Barro 2006, 2009; Weitzman 2009, 2011;
(
Martin and Pindyck 2015, 2021 , which explores how the risk of rare economic
)
3 These assumptions can be motivated with generative AI applications. For irreversibility, once large language
models like ChatGPT are deployed in secondary education, it may be impossible to roll back their use, even after
it becomes clear that they harm student learning. For the damages rising with productivity, many experts fear that
these technologies either pose existential risks or will be misused, both of which would be more damaging when
they have greater capabilities e.g., Shevlane et al. 2023.
4 For example, if AI is us( ed to create pervasive dis) information on social media, this may be disastrous for
democracy but profitable for social media platforms.
VOL. 6 NO. 3 ACEMOGLU AND LENSMAN: REGULATING TRANSFORMATIVE TECHNOLOGIES 361
disasters affects asset prices and c ost-benefit analysis but does not focus on ques-
tions of technology adoption.
The second is a literature on technology adoption e.g., Katz and Shapiro 1986;
(
Parente and Prescott 1994; Foster and Rosenzweig 1995, 2010; Acemoglu, Aghion,
and Zilibotti 2006; Acemoglu, Antràs, and Helpman 2007; Comin and Mestieri
2014 . Early work touching on AI includes Galasso and Luo 2019 and Agrawal,
) ( )
Gans, and Goldfarb 2019 , but these papers do not focus on issues of learning about
( )
social damages from new technologies.
Third, there is a nascent literature focusing on damages from certain technologies
e.g., Bovenberg and Smulders 1995; Acemoglu et al. 2012 . Most closely related
( )
to our paper are a few works that discuss the dilemma between growth and existen-
tial risk from new technologies, including AI. Jones 2023 develops a one-sector
( )
growth model in which AI can be used to raise the aggregate growth rate but with
small probability causes human extinction. Whether it is optimal to use AI depends
crucially on the coefficient of relative risk aversion and whether consumption utility
is bounded. Aschenbrenner 2020 incorporates existential risk into Jones’s 2016
( ) ( )
model of growth and mortality and argues that existential risk rises with consump-
tion unless new mitigation technologies are developed. His model thus exhibits an
“existential risk Kuznets curve” in which existential risk optimally increases until
sufficient R&D resources are shifted toward mitigation. These two papers share
our focus on the costs and benefits of transformative technologies, but they do not
address the speed of adoption across sectors and do not feature learning about risks
over time.
The rest of the paper is organized as follows. Section I presents our benchmark
model. Sections II and III characterize optimal and equilibrium technology choices.
Section IV discusses the conditions under which optimal technology choices can
be restored through regulatory taxes, and Section V concludes. Omitted proofs and
extensions are in the online Appendix.
I. Setup
We consider a continuous-time economy that linearly produces a final good from
a continuum of sectors i 0,1 :
∈ [ ]
1
Y
= ∫ 0
Y
i
𝑑i
.
A representative household has r isk-neutral preferences defined over this final good
and discounts the future at rate 0 .
ρ >
Each sector can use an old technology O or a new, transformative technology N .
We write Q t 0 for the quality of technology j O,N at time t , x t 1 if
j ( ) > ∈ { } i ( ) =
sector i switches its production process to technology N and x t 0 otherwise.
i ( ) =
Sectoral output is
Y 1 x Q x Q ,
i = ( − i ) O + i αi N
362 AER: INSIGHTS SEPTEMBER 2024
where designates the comparative advantage of the new technology, which may
αi
vary if the new technology is b etter suited for some sectors than others. Given tech-
nology choices x x and qualities Q Q ,Q , final output is
= ( i ) i ∈[ 0,1 ] = ( O N )
1
Y x,Q 1 x Q x Q di
( ) = ∫0 ( − i ) O + i αi N .
The new technology is transformative, both because it is general purpose and can
be applied across all sectors and because it enables not just the production of more
output but a higher growth rate:
g g 0
N > O ≥ .
As a result of its restructuring impact on the economy, it also poses new risks. We
model these by assuming that there may be a disaster whereby the new technol-
ogy generates negative effects. If a disaster happens, then there will be damages of
Q 0 in units of the final good in the sectors that are using the technology.
δ i N > ( )
We assume that use of the new technology may be irreversible, so that with proba-
bility 0,1 , sector i cannot switch to technology O if it is using technology N
ηi ∈ ( )
when the disaster strikes. The realization of this reversibility event is independent
across sectors. We assume that damages are proportional to Q because the negative
N
effects correspond to misusing the better capabilities of the new technology.
In what follows, we reorder sectors so that is increasing and assume that i
δi
denotes the quantiles of the distribution, so that we can take this distribution to be
δ
uniform over some interval
[
_ , _
]
. Overall damages then become
δ δ
1
D (x, Q
) = ( ∫0 δi
x
i
𝑑i
)
Q
N .
The economy will experience a disaster with probability – 0,1 , and if there
μ ∈ ( )
is a disaster, its arrival time T is distributed exponentially with rate . We let t
λ μ( )
denote the planner’s or society’s posterior belief at t that there will be a disas-
( )
ter, assuming one has not yet arrived. We impose rational expectations, so that
0 – and the posterior belief evolves according to Bayes’ rule:
μ ( ) = μ
1 ˙ t t [1 t ]
( ) μ( ) = −λμ( ) −μ( ) .
A few comments are in order. First, we model damages in each sector i by the
reduced-form function Q to capture a broad range of potential harms. In the con-
δ i N
text of AI, these include the spread of disinformation that harms democracy; mass
unemployment; and the disruption of production in many sectors from AI-aided
cyber attacks.5 Second, as suggested above, the assumption that damages are pro-
portional to Q is related to the transformative nature of this new technology. For
N
5 Our functional form assumptions also impose that the rate of substitution between gross consumption and
damages in utility is constant and equal to one. Jones 2023 points out that this may not hold in the case of existen-
( )
tial risk and explores the implications for optimal use of a life-threatening new technology.
VOL. 6 NO. 3 ACEMOGLU AND LENSMAN: REGULATING TRANSFORMATIVE TECHNOLOGIES 363
example, damages from disinformation from AI will be higher when it can gen-
erate better language. Third, we assume that the arrival rate of the disaster—and
hence learning about the negative effects of the new technology—is independent
of how many sectors switch to the new technology. This is for simplicity but is
not unreasonable since many of the potential misuses of a new technology can be
gradually recognized without widespread adoption.6 Fourth, it can be verified that
our results remain identical if, instead of a single e conomy-wide disaster, there are
sector-specific disasters and beliefs about each sector’s disaster follow 1 .
( )
II. Socially Optimal Technology Choice
In this section, we set up, solve, and provide comparative statics for the social
( )
planner’s problem.
A. Social Planner’s Problem
Given risk neutrality, the planner’s objective is
(2
)
V (0
) =
E
μ ( 0 ) [ ∫0
∞ exp
(
−ρt
)
[Y (t
) −
D
(
t
) ]
𝑑t
]
,
where Y t and D t denote output and damages at time t and the expectation E is
( ) ( ) μ( 0 )
with respect to the prior belief 0 over the disaster’s arrival time T . To ensure that
μ ( )
the objective is well defined, we assume that
3 g ,
( ) ρ > N
which rules out the case in which the new technology grows so quickly that dis-
counted utility becomes infinite.
It is more convenient to work with the recursive formulation of 2 , which has
( )
the following state variables: the posterior belief of disaster, ; the time-varying
μ
qualities of the old and new technologies, Q ; and, after the disaster, the set of
sectors that were already using the new technology and for which this use is irre-
versible. We track these sectors using the vector x– x– , where x– 1 if
sector i uses technology N irreversibly and x– 0 o= th e ( r w i ) i i s∈ e[ 0 ., 1 L] et V ,Q i = de note
i = (μ )
predisaster social welfare, and let W x– ,Q denote postdisaster welfare. Then the
( )
Hamilton-Jacobi-Bellman HJB equations for the planner are
( )
(4
)
ρV (μ, Q
) =
max {Y
(
x, Q
) + μλ[
E
[
W
(
x– , Q
) |
x
] −
V
(
μ, Q
) ] } +
V˙
(
μ, Q
)
,
x 0,1
i ∈{ }
(5
)
ρW
(
x– , Q
) =
xma
x–
x
,1
{Y
(
x, Q
) −
D
(
x, Q
) } +
W˙
(
x– , Q
) .
i ∈{ i }
6 Alternative assumptions are discussed in Section V.
364 AER: INSIGHTS SEPTEMBER 2024
Equation 5 imposes that x cannot be less than x– because x– 1 implies that
( ) i i i =
sector i ’s use of the new technology is irreversible. V then depends on the condi-
tional expectation of welfare after a disaster given the current technology choices x ,
denoted by E [W
(
x– , Q
) |
x
]
.7 In (4 ), we also use the fact that the arrival rate of the
disaster, given the posterior , is .
μ μλ
To characterize the planner’s technology choices, suppose first that the disaster
has occurred. The planner’s problem in 5 is linear, so the solution is
( )
x
1 if x–
i =
1 or
( αi − δi )
Q
N >
Q
O
,
i = { 0 else
.
This expression assumes, without loss of generality, that the planner sticks with the
old technology if indifferent. It also imposes the constraint that x 1 when x– 1 .
i = i =
Even when unconstrained, it may be optimal to set x 1 if the output produced by
i =
technology N exceeds its damages plus the output that can be produced by technol-
ogy O . We first assume that damages are sufficiently large that, whenever possible,
the planner chooses technology O after a disaster:
6
( ) αi ≤ δi .
This enables us to focus on the most interesting case, where damages exceed the
benefits of the new technology. We return to the general case in Section IIC.
Integrating the HJB equation 5 and taking expectations with respect to x – , we
( )
have
E [W
(
x– , Q
) |
x
] = ∫0
1
[
(1
−
x
i ηi ) _ ρ
−1
g O
Q
O +
x
i ηi _ ρ
α i
−
−
g
δ
N
i Q
N ]
𝑑i
.
Before the disaster, it is optimal from 4 to use technology N in sector i if and
( )
only if
(7
) αi
Q
N −
Q
O > μληi [ _ ρ
−1
g O
Q
O − _ ρ
α i
−
−
g
δ Ni Q
N ] .
Intuitively, the left-hand side is the flow gain from using technology N in sector
i , while the right-hand side is the expected loss due to the disaster, including both
the discounted value of lost output and the irreversible damages. These losses are
multiplied by the posterior arrival rate of the disaster and the probability of irre-
μλ
versibility . Since is decreasing and Q Q is increasing, for any initial state
ηi μ N / O
0 ,Q 0 , there exists a time t such that technology O is used in sector i
(μ( ) ( ) ) i < ∞
before t and technology N is used thereafter.
i
7 To determine this conditional expectation, we use P rx– 1 x 1 and Prx– 1 x 0 0 .
( i = | i = ) = ηi ( i = | i = ) =
VOL. 6 NO. 3 ACEMOGLU AND LENSMAN: REGULATING TRANSFORMATIVE TECHNOLOGIES 365
B. Socially Optimal Technology Adoption
To determine how socially optimal use of technology N changes over time,
( )
denote the fraction of sectors that use technology N , or total adoption, by
1
X
(
μ, q
) = ∫0
x
i
(μ, q
)
𝑑i
.
Here, q log Q Q is the quality gap between the technologies and
= ( N / O )
x ,q 1 if and only if it is optimal to use technology N in sector i in state
i (μ ) =
,q . For simplicity, we assume that and are constant across sectors and
(μ ) αi ηi
equal to and the general case is studied in online Appendix B . This implies
α η ( )
that there exists a damage threshold L ,q such that it is optimal to adopt the new
(μ )
technology in sector i if and only if L ,q . Letting F denote the cumulative
δi < (μ )
distribution function of the uniform distribution over
[
_ , _
]
, total adoption is then
δ δ
just the fraction of sectors below the damage threshold:
X ,q F L ,q
(μ ) = ( (μ ) ) .
The following proposition is immediate from 7 , and we omit its proof.
( )
PROPOSITION 1: Suppose that 6 holds and and are constant across sectors.
( ) αi ηi
It is socially optimal to use technology N in sector i if and only if L ,q , where
δ i < ( μ )
L ,q exp q exp q
(8
) _
( μ
ρ
−) g−
N
α
= _
α
_
_−
_ μ__ λ _
η(
_
− ___)
− _ ρ
−( −
g O
)
.
L ,Q and thus X ,q is increasing in and q ; decreasing in g , , and ; and
(μ )( (μ )) α O λ μ
decreasing in g , provided that L ,q .
N (μ ) > α
Given 6 , the condition L ,q is satisfied as soon as there is any adoption.
( ) ( μ ) > α
Proposition 1 then implies that when the new technology enables faster growth, its
adoption should be slower. This is because of a precautionary motive—even though
the planner is r isk neutral, she would like to avoid irreversible damages from the
new technology. The faster the new technology grows, the greater are the potential
net output losses, strengthening this precautionary motive.
The comparative statics in Proposition 1 are partial because they hold the state
,q fixed. Full comparative statics must account for how parameter changes
(μ )
affect the evolution of the state t ,q t . The belief t does not depend on the
( μ( ) ( ) ) μ( )
growth rates g and g , but the quality gap q t q 0 g g t does. The
O N ( ) = ( ) + ( N − O )
damage threshold L ,q is increasing in the quality gap, so any change in growth
( μ )
rates affects adoption at each t 0 through both the direct effects described in
>
Proposition 1 and the indirect effects through changes in the quality gap q t . The
( )
next proposition characterizes these total effects.
366 AER: INSIGHTS SEPTEMBER 2024
PROPOSITION 2: Suppose that 6 holds and and are constant across sectors.
( ) αi ηi
i X t ,q t is decreasing in g .
( ) (μ( ) ( ) ) O
–
ii There exists an earliest time t such that X t ,q t is decreasing in
( ) – – < ∞ ( μ( ) ( ) )
g if t t . The time t is decreasing in g .
N > N
iii Adoption falls to zero as g approaches —that is, lim X t ,q t 0 .
( ) N ρ g N ↑ ρ ( μ( ) ( ) ) =
The first part of Proposition 2 establishes that the comparative static for g from
O
Proposition 1 generalizes in the presence of the indirect effects through q t —the
( )
quality gap q t is declining in g , reinforcing the direct effect and decreasing adop-
( ) O
tion. The second part shows that the new technology’s growth rate has more nuanced
implications: adoption is not always decreasing in g , but it is decreasing after some
–
N
critical time t , and this time itself is a decreasing function of g . This holds because
N
the precautionary motive highlighted above must compete with the fact that the
quality gap q t is increasing in g , but this indirect effect can dominate only at short
( ) N
time horizons.
The third part of the proposition establishes that as g increases toward the dis-
N
count rate, adoption almost stops. This might appear paradoxical initially but is also
intuitive. When g is approximately equal to , the benefits from the new technol-
N ρ
ogy are very high, leading to nearly infinite discounted utility provided no disaster
arrives. Delay in adoption thus has little effect on these benefits. However, a disaster
will have huge negative consequences, and avoiding it now takes precedence.
The next proposition further characterizes the shape of the adoption curve. Since
F is uniform, X ˙ ,q f L˙ ,q , where f is the constant density of F . Hence, the
( μ ) = ( μ )
curvature of technology adoption is
X¨ ,q L¨ ,q
( μ ) ( μ )
_ X˙_ _ ,_ q_ = _ L˙_ _ ,_ q_ .
( μ ) (μ )
We therefore have the following proposition.
PROPOSITION 3: Suppose that 6 holds.
( )
i L˙ ,q 0 is decreasing in g , and it is decreasing in g if and only if the
( ) ( μ ) > O N
quality gap is sufficiently large—that is,
g g g
α
exp
(
q
) −
1
> _
( ρ
_
_−
_ _
N
_
)
1_
−
_ − _
(
_ μ _
N
_
−
__
_O
_
)
( _
λ1
+ _ ρ
−μη
g O ) .
ii There exists a positive constant G ,q such that if exp q 1 ,
( ) (μ ) α ( ) >
L¨ ,q is positive if and only if g g G ,q . G ,q is independent
( μ ) N − O < ( μ ) ( μ )
of g and increases to infinity over time.
N
The intuition for the first part is the same as for Proposition 2. The damage
threshold increases as the posterior belief falls and the quality gap q grows. Faster
μ
growth for technology O slows the rate of increase of the quality gap and raises the
VOL. 6 NO. 3 ACEMOGLU AND LENSMAN: REGULATING TRANSFORMATIVE TECHNOLOGIES 367
1
0.75
0.5
0.25
0
0 10 20 30
t
opportunity cost of using technology N after the disaster. Consequently, the damage
threshold grows less quickly in each state. Faster growth for technology N raises
both the rate of increase in the quality gap and the net output losses from technology
N after the disaster. The latter effect dominates when the quality gap is sufficiently
large because additional improvements in technology N relative to O have only a
negligible impact on the planner’s technology choice.8
The second part of the proposition proves that adoption of the new technology
will eventually have a convex segment where adoption accelerates because eventu-
(
ally g g will be below G , q . This result holds even though the learning rate
| μ˙ | fa lN ls − at a O greater than expo(μ nent) i) al rate when μ < _1 2 (in particular, _ dd t | μ˙ | = −λ|
˙ 1 2 . This is because expected damages from technology N in sector i are
μ |( − μ))
proportional to the posterior , and as declines, larger increases in the damage
μ μ
threshold L ,q are needed to balance the expected damages and benefits in the
(μ )
“marginal” sector.9
To illustrate these results, we depict the time path of adoption in a couple of
parameterized cases in Figure 1. We set g 2 percent in line with trend GDP
O =
growth in developed economies and 0.04 to produce a risk-free interest rate
ρ =
of 4 percent. We choose two values for g based on Chui et al. 2023 , who fore-
N ( )
cast an increase in the growth rate of 0.6 –3.6 percent in the United States between
2023 and 2040 from AI and other automation technologies. We take the lower end
of this range, g g 0.6 percent , and a higher but still conservative estimate
N − O =
8 The latter effect also dominates regardless of the quality gap whenever L ,q 0 and g g g .
9 In online Appendix B, we verify this intuition by showing that learning dy( nμ am ) ic s> f avor con c N a − ve aO d o≥ pt ioρ n − w h eN n
sectors are heterogeneous according to instead of .
αi δi
t
X
)
(
% %
6 8
2. 3.
= =
gN gN
40 50 60
Figure 1. Socially Optimal Adoption Curves
Notes: Adoption curves X t X t,qt for different values of g . The remaining parameter values are
( ) ≡ ( μ( ) () ) – N
ρ =
0.04 ,
λ =
0.05 ,
η =
0.5 ,
α =
1 , g
O =
0.02 , _
δ =
1 , and
δ =
5 . The initial state is
μ (
0
) =
0.2 and
q0 0 .
( ) =
368 AER: INSIGHTS SEPTEMBER 2024
from the middle of the range, g g 1.8 percent while still satisfying 3 . We
N − O = ( ( ))
take the two technologies to have the same quality in year t 0 , thus q 0 0 .
= ( ) =
We supp_ose that damages range from one to five times gross sectoral output
_ 1 , 5 , and we set 0.5 so that half of all sectors using the new tech-
(δ = δ = ) η =
nology cannot switch back after a disaster. We set the expected arrival time of a
disaster if one exists to be 20 years, which gives 0.05 . Finally, a recent survey
( ) λ =
of AI experts reports a median estimate of existential risk of about 10 percent,10 and
since we are interested in nonexistential misuses of AI as well, we choose the initial
disaster probability to be twice as large, 0 20 percent. Figure 1 shows that
μ ( ) =
optimal adoption is slow, taking about 40 years until full adoption when g 2.6
N =
percent and almost 60 years when g 3.8 percent .
N =
C. Optimal Adoption with Small Damages
We have so far imposed 6 , ensuring that the postdisaster damages from the new
( )
technology are large and exceed its gross output within each sector. This is a natural
benchmark since our analysis is motivated by significant potential harms from AI.
We now relax this assumption and allow a sector’s damages to be small relative to
its output under the new technology .
( δ i < α)
In online Appendix C, we show that socially optimal adoption is again char-
acterized by a damage threshold L ,q , and we prove the following analogue to
(μ )
Proposition 2 for small damages.
PROPOSITION 4: Suppose that and are constant across sectors. For all t with
α i ηi
L t ,q t :
(μ( ) ( ) ) < α
i X t ,q t is decreasing in g .
( ) (μ( ) ( ) ) O
ii X t ,q t is increasing in g .
( ) (μ( ) ( ) ) N
iii If q 0 is sufficiently low and X t ,q t F , adoption is bounded
( ) ( ) (μ( ) ( ) ) < ( α)
below F as g approaches —that is, lim X t ,q t F .
(α) N ρ g N ↑ ρ ( μ( ) ( ) ) < ( α)
Adoption among sectors with small damages is still decreasing in g , but in con-
O
trast to the case with large damages, it is increasing in g . Gradual adoption remains
N
optimal even when g increases toward the discount rate . With small damages,
N ρ
using technology N is always optimal in the long run. Nevertheless, gradual adop-
tion is optimal to learn about the probability of a disaster before one occurs and
( )
to delay the adoption of technology N in case of a disaster until the quality gap
becomes sufficiently large. This strategy thus avoids temporary costs of irreversibil-
ity. Further analysis of this case is presented in online Appendix C.
Finally, we note that if damages are uncertain, any chance of large damages leads
to longer optimal delay, even if expected damages are small, in order to avoid the
possibility that damages turn out to be large and adoption is irreversible.
10 https://aiimpacts.org/2022-expert-survey-on-progress-in-ai
VOL. 6 NO. 3 ACEMOGLU AND LENSMAN: REGULATING TRANSFORMATIVE TECHNOLOGIES 369
In summary, the optimal adoption of a new, transformative technology should be
gradual, particularly when its superior capabilities also make its potential damages
greater and there is learning about the likelihood of misuse a “disaster” .
( )
III. Equilibrium Technology Choice
We now characterize equilibrium technology adoption when private firms do not
fully internalize social damages.
A. The Firm’s Problem
Suppose now that in each sector, the choice of technology is made by a private
representative firm that seeks to maximize expected discounted profits. To sim-
( )
plify, we assume that the firm in sector i appropriates all output of its intermediate
as profits but only internalizes private damages . This textbook externality
γ i ≤ δi
leads to excessively fast adoption of the new technology before the disaster, and
our main results below describe how the equilibrium and socially optimal adoption
curves differ.
Firm i ’s profit maximization problem can be formulated recursively in the same
way as the planner’s problem in the previous section. The state variables before the
disaster are again and Q , and after the disaster they are x – and Q . Let ,Q
μ i Π i (μ )
denote the firm’s predisaster value, x– ,Q its postdisaster value, and Y x,Q its
Φ i ( i ) i ( i )
gross output. The HJB equations for the firm are
( )
(9
) ρ Π i
(μ, Q
) =
max
{
Y
i (
x
i
, Q
) + μλ[
E
[ Φi (
x–
i
, Q
) |
x
i ] − Πi
(μ, Q
) ] }
x 0,1
i ∈{ }
˙ ,Q ,
+ Π i (μ )
10 x– ,Q max Y x,Q x Q ˙ x– ,Q
( ) ρ Φ i ( i ) = x x– ,1 { i ( i ) − i γi N } + Φ i ( i ).
i ∈{ i }
These value functions differ from the planner’s 4 and 5 because the firm internal-
( ) ( )
izes only a fraction of the flow damages from technology N .
γ i / δi
We now impose a stronger version of 6 : private damages are sufficiently large
( )
that firm i will always choose technology O after the disaster if possible:11
11
( ) αi ≤ γi .
Similar to the planner’s solution, it is privately optimal for firm i to use technology
N if and only if
Q Q 1 Q αi − γi Q
α i N − O > μληi [ _ ρ − g O O − _ ρ − g N N ] .
11 Without this assumption, an additional inefficiency would arise in equilibrium as firms would use the new
technology in some reversible sectors even after a disaster.
( )
370 AER: INSIGHTS SEPTEMBER 2024
The only difference between this condition and the planner’s optimality condition
7 is that private damages appear instead of social damages on the right-hand
( ) γi δ i
side. Firm i internalizes fewer damages from technology N and thus begins using it
earlier.
B. Equilibrium Technology Adoption
We denote total equilibrium adoption by
1
X
̃
(μ, q
) = ∫0
x
̃ i
(μ, q
)
𝑑i,
where x ,q 1 if and only if firm i uses technology N in state ,q . Again
̃ i (μ ) = ( μ )
assuming that and are constant across sectors, it is immediate that firm i will
αi η i
adopt the new technology if and only if private damages are lower than the damage
threshold, L ,q . Equilibrium adoption is then
γ i < (μ )
X ,q F L ,q ,
̃ (μ ) = γ ( (μ ) )
where F is the cumulative distribution function of .
γ
γi
This characterization implies that all comparative statics results from Section IIB
apply to equilibrium adoption. The results in Propositions 1 and 3 concern only the
damage threshold L ,q and hold exactly as stated, while Proposition 2 applies
( μ )
after replacing X ,q with X ,q .
(μ ) ̃ ( μ )
PROPOSITION 5: Suppose that 11 holds and and are constant across sectors.
( ) αi ηi
i X t ,q t is decreasing in g .
( ) ̃ (μ( ) ( ) ) O
ii There exists an earliest time t such that X t ,q t is decreasing in
( ) ̃ < ∞ ̃ (μ( ) ( ) )
g if t t . The time t is decreasing in g .
N > ̃ ̃ N
iii Adoption falls to zero as g increases to : lim X t ,q t 0 .
( ) N ρ g N ↑ ρ ̃ ( μ( ) ( ) ) =
In the remainder of this section, we characterize how the optimal and equilibrium
adoption curves differ. We first observe that similar adoption curves do not imply
that the equilibrium is optimal because the order in which sectors adopt the new
technology matters. For example, private and social damages may be negatively
affiliated, meaning that high social damage sectors have low private damages. In this
case, the order in which the new technology spreads in equilibrium is exactly the
opposite of the optimal order.
Even when the equilibrium and optimal orders of adoption coincide, the equilib-
rium can be inefficient. To see this, suppose that social and private damages are pos-
itively affiliated, so that there exists a n onnegative and strictly increasing function
( )
with . We can then write equilibrium adoption as
κ γ i = κ( δi ) ≤ δi
X ,q F 1 L ,q
̃ ( μ ) = ( κ − ( (μ ) ) ) .
VOL. 6 NO. 3 ACEMOGLU AND LENSMAN: REGULATING TRANSFORMATIVE TECHNOLOGIES 371
1
0.75
0.5
0.25
0
0 10 20 30
t
Figure 2. Comparing Socially Optimal and Equilibrium Adoption Curves
Notes: Socially optimal and equilibrium adoption curves, Xt and |
155 | mit_edu | Dynamo20Case202024.pdf | Case Study
Dynamo AI
Written by Audrey Woods
In a world of booming generative AI applications, business leaders across the economy worry that they need to
incorporate AI or risk falling behind. However, there are many factors to consider before designing, using, and/
or launching AI systems, including security, privacy, hallucinations and legal compliance. Between increasing
regulation and shifting consumer expectations, it’s more important than ever for companies to be aware of and
address the risks at every level of their evolving AI stack.
MIT CSAIL Startup Connect member Dynamo AI aims to help companies navigate this change with a suite of
products designed to offer end-to-end privacy, security, and compliance solutions. Broken into three pillars of
AI-focused support—DynamoEval, DynamoEnhance, and DynamoGuard—Dynamo AI seeks to support the
democratization of AI technologies by making them accessible, reliable, and safe to implement.
GETTING THEIR START
As CEO and Co-Founder Vaikkunth Mugunthan tells it, Dynamo AI began as the last chapter of his PhD thesis
at MIT. When he first started his graduate studies under CSAIL Principal Research Scientist Lalana Kagal,
he was focused on theoretical privacy. But when a CSAIL Alliances poster session landed him a summer
internship with JPMorgan, he was introduced to federated learning, which trains AI by sharing the model itself
instead of sharing data, thereby offering valuable privacy guarantees. There was a pressing industry demand
for this technology, especially in finance and healthcare, and the interest he experienced made him confident
enough to launch the company in 2021.
While Dynamo AI was originally focused only on providing a “plug and play” tool for federated learning—in fact
the company’s original name was DynamoFL—the team soon realized that there were many other aspects of
AI implementation that customers were concerned about. New AI-related laws emerging around the world, the
risks, both known and unknown, of launching AI systems, and the ongoing process of protecting users and
companies from those who might misuse AI programs either intentionally or not were all issues that Mugunthan
realized Dynamo AI could help with. Therefore, in late 2023 and early 2024, the company pivoted from a
focus exclusively on federated learning to a broader, more comprehensive set of features designed to assist
companies in both designing and launching safe and compliant AI systems.
Now Dynamo AI is partnering with Fortune 500 businesses to test their tools in various industry functions. They
went through the YCombinator startup accelerator and got “the first interview on the first day,” Mugunthan
says, and the company has since gone on to raise $19.4 million. Their two different TechCrunch features—one
in 2022 and one in 2023—each brought a fresh wave of publicity that helped grow the company to where it is
today, with 45 employes and customers such as Lenovo, Qualcomm, and Aisin. As a self-described “research-
heavy” company, Dr. Mugunthan explains how their priority is to “hire masters or PhDs from the best schools”
and become the most trusted name in the field of AI support.
DYNAMO AI: SECURING THE AI STACK
When explaining the need that Dynamo AI aims to meet in the industry, Head of Growth
and Strategic Partnerships Kavi Arora says, “if we really want to democratize these
technologies in different GenAI-based consumer-focused, internally focused applications,
and agent focused applications, there are a many risks that exist in privacy, security, and
hallucinations.” Dynamo AI, he explains, is “addressing the need for privacy, security, and
compliance throughout that AI stack [by] testing applications for risks, remediating those
risks, and then real-time guardrailing applications as they go into production.” These
services fall into three main “modules:” DynamoEval, DynamoEnhance, and DynamoGuard.
DynamoEval evaluates LLMs and generative AI programs as they’re being designed to make
sure a given program complies to emerging regulatory standards. Providing automated
stress testing, DynamoEval generates the needed documentation for regulatory audits and
checks a system’s weaknesses in privacy, security, and hallucination. DynamoEnhance
offers support at the next phase of development, fixing and remediating the identified risks.
This module offers several easy-to-use techniques to improve privacy (such as federated
learning), mitigate hallucinations, and bolster program safety. And finally, DynamoGuard
supports AI programs going into production by creating real time guardrails that are
customizable in natural language, for every organization’s bespoke policies. “For example,”
Arora says, ”multiple financial services institutions are deploying internal chatbots leveraged
by different portfolio managers. We’re enabling them to create guardrails where you can
allow certain levels of portfolio managers to drive certain types of insights and others
to not.” That level of granularity helps businesses enforce governance policies, prevent
misuse, and easily audit their LLMs.
Taken altogether, Dynamo AI’s various modules offer a platform for enabling secure, private,
hallucination-free, and regulation-compliant AI models.
THE CSAIL CONNECTION
One thing Dr. Mugunthan makes clear is how much he attributes his success to Lalana
and CSAIL. From the very beginning, it was Dr. Kagal’s encouragement that inspired him
to take a chance and apply to MIT, and he describes his time at CSAIL as “a fantastic
experience.”
“I really enjoyed the collaborative nature of projects,” he says, highlighting his ability to get
a minor from Harvard and his deep roots in the CSAIL community.
For Dr. Mugunthan, his link to CSAIL is more than nostalgic; it’s a pivotal part of his
company’s strategy. He says through MIT, “we have access to the best talents in the world,”
an advantage he’s used to hire some of the PhDs the company now employs. Dynamo AI
is also planning to launch internships and create a Dynamo AI ambassador program with
CSAIL, which would deepen this connection. Dr. Mugunthan adds, “I wouldn’t have been
at this stage [without CSAIL], so I want to give it back as well.”
continued Dynamo
AI
CASE
STUDY
For more information about CSAIL Alliances industry engagements, please visit:
cap.csail.mit.edu
Beyond recruitment, Dynamo AI is utilizing their connection with CSAIL Alliances to
maximize the company’s exposure and vet potential business partners. Dr. Mugunthan has
been invited to present at several Alliances conferences, which has led to “a good number
of client leads” and helped Dr. Mugunthan understand specifically which companies were
interested. “We were able to tailor our product toward what they needed,” he explains,
which helped Dynamo AI create even more market traction. Because of that, he’s eager to
take part in future CSAIL Alliances events.
“CSAIL Alliances has been super helpful,” Dr. Mugunthan says, calling Sr. Client Relations
Coordinator Philip Arsenault “a fantastic friend of mine.”
LOOKING FORWARD
When asked what they’re focused on next, Dr. Mugunthan and Arora explain that Dynamo
AI is now looking to expand the company’s reach into new sectors, exploring how their
suite can be applied to different industries. With the AI market growing in nearly every
economic sector, Dynamo AI hopes to use this momentum to their advantage and support
positive technological change.
Dr. Mugunthan says the end goal for Dynamo AI is “to make sure that when it comes to
privacy preserving machine learning, the first company that comes to anyone’s mind is us.”
With that in mind, Dr. Mugunthan calls his association with MIT an “added advantage,”
showing clients that Dynamo AI has the best people on the job.
continued Dynamo
AI
CASE
STUDY
For more information about CSAIL Alliances industry engagements, please visit:
cap.csail.mit.edu |
156 | mit_edu | Integrating-AI-in-organizations-for-value-creation-through-Human-AI-teaming-A-dynamic-capabilities-approach.pdf | JournalofBusinessResearch182(2024)114783
Contents lists available at ScienceDirect
Journal of Business Research
journal homepage: www.elsevier.com/locate/jbusres
Integrating AI in organizations for value creation through Human-AI
teaming: A dynamic-capabilities approach
Cristina Simo´ na,*, Elena Revillaa, Maria Jesús Sa´ enzb
aIE Business School, IE University, Spain
bMassachusetts Institute of Technology, United States
A R T I C L E I N F O A B S T R A C T
Keywords: Although the potentialities of artificial intelligence (AI) are motivating its fast integration in organizations, our
Human-AI teaming knowledge on how to capture organizational value out of these investments is still scarce. Relying on an
Productive dialogue approach to dynamic capabilities that focuses on the team level, we examine how humans and AI create in-
Dynamic capabilities
teractions that engage both agents in productive dialogue for value co-creation. Our analysis is based on a
Artificial intelligence
longitudinal case of the development of a recruitment algorithm at a national subsidiary of Santander bank. Our
results allow to identify three main sets of human-AI teaming interactions: achieving interoperability, building
trust, and producing mutual knowledge gains. We elaborate a set of propositions on how the value of AI is
increased when such interactions are created through productive dialogue, opening the scope for further research
on the teaming dynamics that turn the collaboration between both agents into a source of value creation for
companie
1. Introduction capabilities (DCs) framework to explain how firms can transform their
resources to persistently maintain value creation (Ambrosini &
The growing availability of artificial intelligence (AI) is stimulating Bowman, 2009; Teece, 2014). DCs have traditionally been studied both
its rapid drive for integration into organizations (Kinkel et al., 2022). AI at the macro-level of organizational routines (Felin & Powell, 2016;
is fundamentally related to autonomous decision-making (Berente et al., Fainshmidt et al, 2016; Bingham et al., 2015) and micro-level of exec-
2021), and it has therefore attracted companies’ interest because of its utive decisions (Day & Schoemaker, 2016; Kor & Mesko, 2013). How-
potential to extend their scope to domains that have been exclusively ever, scholars have proposed an emergent, meso-level of DC
human (Dwivedi et al., 2021), revolutionizing how business creates development that considers team learning as the source of their dyna-
value for organizations (Mikalef & Gupta, 2021). Thus, the imple- mism and acts as a link between the macro and micro ones (Harvey
mentation of these technologies is gaining momentum (Ångstro¨m et al., et al., 2022; Salvato & Vassolo, 2018). DCs are enabled at this meso-level
2023), and senior executives seem to agree on the criticality of AI as a when teams establish high-quality interactions through productive
game changer in the current business scenarios (Ångstro¨m et al., 2023; dialogue (Tsoukas, 2009). Thus, teammates are motivated to change
van de Wetering et al., 2022). However, organizations are still struggling how they work and produce joint learning that is translated into adap-
with the issue of how to capture value from AI (Berg et al., 2023), with tive organizational routines (Enholm et al., 2022; Harvey et al., 2020).
only 20 percent of companies declaring an impactful exploitation of AI In the present paper we contend that this meso-level of analysis be-
applications (Akter et al., 2021) and studies showing that investments in comes particularly consequential when examining how to capture value
this technology may even negatively impact market value (Lui et al., out of the integration of AI in organizations. In this scenario, AI is an
2022). In this context, scholars are shifting the focus from the benefits of agent in full who should collaborate closely with humans to improve
AI from a technological perspective toward a more holistic approach to performance through the augmentation of their capabilities (Raisch &
how to leverage AI and what its real sources of value are (Enholm et al., Krakowski, 2021). Considering AI’s peculiarities regarding volatility,
2022). opaqueness, or elusiveness of human control (Hassija et al., 2023;
From a theoretical perspective, scholars have relied on the dynamic- Mikalef & Gupta, 2021) we face unique inquiries in terms of how to team
* Corresponding author at: Maria de Molina, 13, 28006 Madrid, Spain.
E-mail addresses: [email protected] (C. Simo´n), [email protected] (E. Revilla), [email protected] (M. Jesús Sa´enz).
https://doi.org/10.1016/j.jbusres.2024.114783
Received 25 July 2023; Received in revised form 8 June 2024; Accepted 12 June 2024
Availableonline27June2024
0148-2963/©2024ElsevierInc.Allrightsarereserved,includingthosefortextanddatamining,AItraining,andsimilartechnologies.
C. Sim´on et al. J o u r n a l o f B u s i n e s s R e s e a r c h 182(2024)114783
up with the technology to leverage its value (Ångstro¨m et al., 2023). 2. Theoretical background
While the scholar literature is examining how to develop DCs through
human-AI teaming both from a micro perspective (Weber et al, 2022; 2.1. Dcs
Brau et al, 2023) and also at a macro level (Mikalef & Gupta, 2021), to
the best of our knowledge there are no studies of human-AI teaming Recent studies propose that, when properly integrated into an or-
which place the focus on the meso-level of DCs. Following the call for ganization’s socio-technical system, AI creates value for the firm by
research on the mechanisms that reveal the interplay that underlies enacting DCs (Drydakis, 2022; Mikalef & Gupta, 2021; Schoemaker
human-AI teaming (Ångstro¨m et al., 2023; O’Neill et al., 2022), we et al., 2018). By contrast with ordinary capabilities, DCs are defined as
contend that this meso-level is key to explain how companies can “the firm’s ability to integrate, build, and reconfigure internal and
leverage the value of AI, and rely on the literature on productive dia- external competences to address rapidly changing environments.”
logue as a source of value creation (Keeling et al., 2021; Tjosvold et al., (Teece, 2007, p.316). Importantly, DCs cannot be acquired but can only
2014; Tsoukas, 2009) to discuss how to achieve quality teaming in- be developed internally (Easterby-Smith et al., 2009), and they do not
teractions between humans and AI. Given these considerations, the directly impact organizational performance but through a change of
following research question is posed: How should human-AI teams develop already-existing, ordinary enterprise capabilities (Chatterji & Patro,
through productive dialogue to create value for an organization? Based on 2014). In addition, DCs are path-dependent, going through iterative
the productive dialogue literature, we explore how humans and AI cycles of sensing-seizing-reconfiguration through which opportunities
create interactions that engage both agents at the team-based, meso-level are grasped and developed, thus leading to organizational learning and
of DC development in gaining leverage on the technology. Specifically, resource transformation (Easterby-Smith et al., 2009; Helfat & Peteraf,
we identify three categories of interactions, aiming at (i) negotiating the 2009). Therefore, the passage of time is critical for these capabilities to
terms of interoperability among the agents, (ii) nurturing trust to enrich reach their maturity stage and become embedded in organizational
the outcomes of the teaming relationship and (iii) creating mutual processes, learnings, and routines (Pan et al., 2022).
learning along the way. Along the lines of the previous arguments, among the premises
In attempting to deal with these issues, we rely on a longitudinal case derived from the DC framework is that value creation does not depend
study on how a human-AI team develops over a period of four years. on a company’s investment of resources but on how such resources are
Given the relevance of the evolving nature of these capabilities, we draw combined and deployed to create organizational learning and dynamic
on a process ontology (Tsoukas & Chia, 2002) to develop our case study, adaptation (Lockett, 2005). Until recently, DC scholars have examined
which emphasizes the temporal evolution of phenomena (Sharma & this phenomenon at two distinct levels. On the one hand, there is ample
Bansal, 2020). We focus on a particular type of AI application, a literature concentrated on the macro-level of operational routines and
machine-learning (ML), supervised-classification algorithm for decision-making systems, given their relevance in providing reliability
screening candidates in a large bank’s recruitment processes. The hiring to companies (Teece, 2007, 2014). For example, Helfat and Peteraf
function offers a particularly rich context for delving into human-AI (2003) examined the high-level origins of organizational capabilities
teaming (Chowdhury, Dey, et al., 2023) because experiences thus far and the specific sources of heterogeneity that support competitive
manifest how AI may destroy value through unintended consequences, advantage. Also from a macro perspective, DCs have been approached
thus demanding intensive interactions with humans for monitoring and from the view point of organizational learning (Denford, 2013; Kaur,
correction purposes (Soleimani et al., 2022; Teodorescu et al., 2021). In 2019) or of their role in adapting to disruptions at the company and
addition, only a small proportion of companies report being able to market levels (Karimi & Walter, 2015). Researchers have also adopted a
integrate these applications into their processes (Laurano, 2022), mainly distinct approach to DCs focusing on the micro-level of managerial
due to a lack of knowledge of what it represents for the human-resource (Adner & Helfat, 2003) or top executive decisions (Day & Schoemaker,
(HR) department in terms of assumed risks and work to be done (Hocken 2016). By contrast to the macro-level, this approach opens the scope for
& King, 2023). We could systematically monitor the implementation of change as decisions are flexible and subject to change; however, it limits
AI in a talent-acquisition department from its inception, focusing on the the view of the DC-creation process to individual decisions and therefore
quality of the interaction within the human-AI team that enabled a conceals our understanding of how routines are created (Helfat &
valuable integration of the technology into the organization. Peteraf, 2015; Salvato, 2021).
This study extends our current knowledge on the organizational Although both approaches to DC have led to relevant insights,
integration of AI-based applications in several ways. First, we contribute neither explains how individual decisions may be aggregated in a
to the development of the meso-level approach to DCs integrating AI as a manner that reconfigures resources and transforms routines at the
unique, new team member, therefore adding to the scarce literature in organizational level. To address this issue, further conceptual de-
this field (Harvey et al., 2022; Salvato & Vassolo, 2018). Our findings velopments propose a meso-level in which interpersonal connections
illustrate how, while engaged in cycles of interactions, the team facili- among employees act as connectors between the micro- and macro-
tated the dynamic sensing-seizing-reconfiguring pathway characteristic levels (Salvato & Vassolo, 2018). In this approach, companies leverage
of DCs (Chirumalla, 2021; Krakowski et al., 2023). Second, our research the joint effect of the adaptation to change triggered by top-level de-
points to the relevance of productive dialogue between humans and AI cisions and the stability created by organizational routines when em-
to allow for the creation of organizational value (Keeling et al, 2021; ployees are interconnected through teamwork, engaged to envision
Tjosvold et al, 2014). Thus, our study shows that it is not only the opportunities for improvement and willing to act, thus constituting the
quantity but also the quality of interactions between humans and AI that source of the dynamization of organizational capabilities (Peteraf et al.,
captures value out of the technology. We contribute to the human-AI 2013).
teaming literature by elaborating a set of propositions on how the This meso-level perspective emphasizes productive dialogue as a
value of AI is increased when such interactions are developed through critical aspect of the teaming-development process: “We see productive
productive dialogue. This way, our study suggests novel extensions on dialogue as the means through which individual employees’ proposals
the teaming strategies that engage both agents in value creation. for change become aggregated into a firm-level dynamic capability”
Finally, our study contributes to the practical approach to human-AI (Salvato & Vassolo, 2018, p. 1730). Productive dialogue assumes that
teaming at the meso-level of analysis by providing managerial insights agents recognize the “otherness” of their mates, act candidly, engage in
into how to design team interactions that facilitate the integration of AI, mutual interaction, and stimulate action out of collective learnings
augmenting the capacity of humans and leveraging the value that the (Berkovich, 2014; Tsoukas, 2009). As a result of productive dialogue,
technology can create for the organization. employees genuinely engage in improving current routines by devising
new reconfigurations of resources and prompting action, their joint
2
C. Sim´on et al. J o u r n a l o f B u s i n e s s R e s e a r c h 182(2024)114783
efforts resulting in changes in how a unit operates (Salvato & Vassolo, through flexible adaptation” (Kaplan & Haenlein, 2019a, p. 17). It is this
2018) and creating value (Keeling et al., 2021). Therefore, the sensing- capacity of AI to optimize itself through learning that shows the greatest
seizing of opportunities and final reconfiguration of resources that potential to dynamically transform organizations’ operating architec-
characterize DCs and create business value (Teece, 2014) emerge from ture and redefine how they capture and share value (Ångstro¨m et al.,
these inner teaming flows that evolve in collective learning cycles over 2023).
time (Harvey et al., 2022). Such a myriad of interactions brings about While these benefits are already apparent for companies, it is also
process and resource reconfigurations that are unique and difficult to accepted that the performance of AI applications to realize value is
imitate, thus reinforcing competitive advantage (Molloy & Barney, limited by several factors (Ångstro¨m et al., 2023; Revilla et al., 2023).
2015). One key factor concerns the quantity and quality of the data used for
This complex system of interactions through productive dialogue training it (Kaplan & Haenlein, 2019; Sarker, 2021; Vial et al., 2021).
that characterizes the meso-level reveals as particularly relevant when Although data collection is rapidly growing in organizational contexts,
exploring the integration of AI in organizations. The research literature available databases may not be appropriate for AI’s learning process
has growingly recognized that, to realize its benefits, AI cannot work in (Berente et al., 2021), and there are also data-privacy and related legal
isolation but only woven into human’s daily practices, which demands a issues that may limit its use (Van Den Broek et al., 2022). Furthermore,
close collaboration on the part of both agents (Moser et al, 2022; unlike humans, AI lacks the ability to interpret contextual cues and
Johnson & Vera, 2019). Additionally, the teaming between humans and anticipate the consequences of its decisions (Krakowski et al., 2023;
AI poses specific challenges involving the adaptation to the specificities Lindebaum & Ashraf, 2023), and this introduces relevant margins of
¨
of this new sociotechnological scenario (Musick et al, 2021; Angstrom et error when facing ill-structured problems under conditions of
al, 2023). Scholars have approached these aspects of human-AI teaming complexity, ambiguity, and scarce information (Madni & Madni, 2018).
in different ways as well. At the macro level, Mikalef et al (2021) AI may also produce senseless outcomes in problems involving social
identify the routines that companies should develop to best support B2B issues because it lacks the ability for moral deliberation (Hasija & Esper,
via AI, and others follow similar approaches at a more general level of 2022; Moser et al., 2022). Finally, a severe limitation of AI is its lack of
application of AI (e.g., Kemp, 2023). More recently, Akter et al. (2023) transparency and explainability regarding how data are integrated and
develop a framework for the application of AI to service innovation the knowledge that is gained by processing them (Chowdhury et al.,
focusing on the most relevant organizational capabilities that turn AI 2023). This “black boxing” prevents humans from understanding its
into a competitive advantage in the area. Conversely, working at the intentions, reasoning, and performance (Hasija & Esper, 2022; Vo¨ssing
micro level, Weber et al (2023) interview a group of experts in the field et al., 2022), and it consequently creates mistrust and reluctancy to
of AI with distinct background and degrees of experience regarding their collaborate with the technology (Dorton & Harper, 2022).
perspectives on the design of processes for developing organizational These unique characteristics of AI demand an intense, high-quality
capabilities that allow for an effective implementation of the technol- collaboration with humans to identify and correct flaws to make the
ogy. Similarly, Brau et al. (2023) analyze the effectiveness of different most of its potentialities (Balasubramanian et al., 2022; Weber et al.,
executive profiles to examine how their AI-based decisions determine 2023). However, scholars have claimed that, when this teaming is
the performance of digitized retail supply chains. A recent survey effective, AI can in turn augment human capabilities and improve de-
(Ångstro¨m et al., 2023) also collects the opinions of a wide sample of cision making through a mutual learning process (Weiss & Spiel, 2022).
executives with AI expertise on the challenges they face when inte-
grating this technology and the decisions that delineate a successful, 2.3. Human-AI teaming, sources of dynamism, and value creation
value-creating implementation. Finally, scanty studies deal partially
with human-AI teaming at the meso-level of analysis, e.g., examining the The application of a meso-level approach to the creation of DCs in the
patterns of interdependency that connect the different actors involved in context of AI calls for the above type of human-AI teaming, which in-
AI performance (Jacobides et al., 2021) or exploring specific aspects of volves the interaction of “at least one human and one autonomous agent
human-AI collaboration such as the effects of interactions being volun- where the autonomous agent has a significant role and is treated as a full
tary or hierarchically imposed (Bezrukova et al., 2023). teammate instead of a simple tool” (Schelble et al., 2022). Such a defi-
Our study focuses on how human-AI teaming develops at this meso- nition recognizes that human-AI teaming means not simply adding a
level perspective of DC creation. We argue that quality interactions new resource but also undergoing an internal redesign of the human
based on productive dialogue motivate effective teaming development. team operations in light of the new capacity (Mikalef & Gupta, 2021;
Through productive dialogue, teammates prepare to change how they Saenz et al., 2020), and accumulating evidence shows that the consid-
perform their work and based on such interactions, organizations enact eration of AI as one more group member significantly impacts the team’s
DCs. We contend that researching at this meso-level requires to first performance (Hauptman et al., 2023).
examine the unique features that AI may provide to organizations and Scholars have mostly tried to understand the terms of the interaction
then review the current body of knowledge on how such distinct but between humans and AI relying on the extensively studied field of
complementary agents engage in human-AI teaming for value creation. human teams (Endsley et al., 2022; Johnson & Vera, 2019). These an-
alyses reveal three main factors that have proven fully applicable to the
2.2. AI as a potential source of organizational value type of quality interaction that the meso-level model of DCs discusses
when it proposes productive dialogue as a core mechanism for enacting
Although scholars agree that there is not a single, univocal definition the dynamism of organizational capabilities (Salvato & Vassolo, 2018).
of AI, most concur on referencing it to human intelligence. Therefore, AI The first one is the recognition of interdependency between the team-
has recently been defined as “the ability of a system to identify, inter- mates (Kozlowski, 2015) to determine the workflow structure and terms
pret, make inferences and learn from data to achieve predetermined of the exchanges (Kozlowski & Ilgen, 2006). To engage in productive
organizational and societal goals” (Mikalef & Gupta, 2021, p. 3). dialogue, AI should integrate in a team’s activities while simultaneously
Operating in this way, AI speed and information-processing capacities demonstrating a level of agency in its outcomes that convinces humans
have proven to outperform those of humans in different scenarios, such of the unique value that this technology can provide beyond a tradi-
as in the management of routine and codifiable work (O’Neill et al., tional information and telecommunications (IT) application (Musick
2022), prediction tasks (Choudhury et al, 2020), and situations that et al., 2021; O’Neill et al., 2022). From this starting point, humans and
demand the fitting of models to large sets of alternatives (Weber et al., AI should share a profound understanding of each other’s capacities and
2023). Yet, the most distinct feature of AI is its ability to “learn from complementarities (Hauptman et al., 2023) and for this purpose, the
such data, and to use those learnings to achieve specific goals and tasks issues of explainability and transparency are critical for humans to
3
C. Sim´on et al. J o u r n a l o f B u s i n e s s R e s e a r c h 182(2024)114783
intervene in a timely manner (Endsley, 2023; Endsley et al., 2022). critical (Soleimani et al., 2022; Teodorescu et al., 2021). Furthermore,
Humans should be able to understand why a system makes specific de- the accessibility to multiple data sources allowed a synergistic collection
cisions, which represents a challenge because AI procedures and out- of evidence to warrant the validity of our findings (Eisenhardt, 1989).
comes are typically opaque (Ångstro¨m et al., 2023; Kellogg et al., 2020) From project inception, we could systematically and regularly observe
and AI can change its capacities in unpredictable, non-obvious ways the interactions leading to human-AI teaming. We combined these ob-
(Endsley, 2023). A recent review of empirical research on human-AI servations with interviews and archival data, mainly internal commu-
teaming (O’Neill et al., 2022) revealed that interdependence and nications and project presentations in public fora. This strengthened the
training in each other’s awareness were positive for the team (Johnson grounding of the theory, which is an important point because the
et al., 2021; Li et al., 2022; Xiong et al., 2023), and that low levels of theoretical development in the field is limited (Musick et al., 2021). The
reliability in AI could be balanced by increasing transparency (Chowd- longitudinal data allowed us to follow a process-based approach, which
hury, Joel-Edgar, et al., 2023; Vo¨ssing et al., 2022). is considered important for our objectives given the consideration of DCs
Another exportable aspect of human teams for achieving the pro- as evolving change-management phenomena and the need to focus on
ductive dialogue that creates efficient teaming with AI is trust, regarded the “know-how knowledge” of their development (Langley et al., 2013).
as an important antecedent of mutual understanding and team cohesion
(Feitosa et al., 2020). Trust, defined as “the attitude that an agent will 2.4. Case setting
help achieve an individual’s goals in a situation characterized by un-
certainty and vulnerability” (Dorton & Harper, 2022; Lee & See, 2004), In 2016, motivated by the fast-growing introduction of big data and
is fundamental for humans to become willing to accept AI as an equal analytics in business organizations, the TA manager of Santander Spain
partner (McNeese et al., 2021). Trust has also proven to be key to bank started to explore the application of these technologies in hiring
reinforcing interactions and extracting value from them (Hoff & Bashir, processes within the bank’s national market. As a result of the de-
2015). However, trust is not a binary phenomenon but rather operates as partment’s long experience in the practice, they were clear that the final
a continuum where there is a continuous calibration over interactions “hire” decision should be in the hands of a human. However, a whole set
between the agents (Yang et al., 2023), and how this works for AI is yet of opportunities emerged as some tasks could greatly benefit from the
to be fully explored (Dorton & Harper, 2022a). For example, trust grows use of AI. First of all, the volume of applications for junior positions was
with increased connection among humans; in human-AI teaming, very high due to the prestigious-employer branding of Santander bank
however, it may decline if interactions reveal flaws or malfunctions of and its widespread commercial network. When the project started, the
the technology (Glikson & Woolley, 2020); it may also rise if humans are department received an average of 4,000 applications per sales vacancy;
seen to identify and correct the failure to utilize an opportunity to learn under time pressure to fill the positions, they could review and reach out
more about the boundary conditions of the technology (Dorton & to only approximately 900, thus rendering a significant amount of po-
Harper, 2022). tential talent unexplored. Another relevant opportunity for the TA team
Finally, in the course of this continuous interdependency-based came up from the fact that candidates’ profiles were highly heteroge-
interaction reinforced by trust, learning emerges as a third decisive neous, given the applicants’ lack of experience. This made the CV-
attribute at the meso-level (Harvey et al., 2020). As a result of gaining screening process time-consuming; for every candidate considered
collective experience in decision-making and problem-solving through valid for a telephone interview, they had to go through more than 30
productive dialogue, team members engage in a process of mutual applications. Finally, the screening was conducted on a “first-come-first-
learning, in which two agents adapt their behavior and/or mental states served” basis, which might have left out talented late applicants. Due to
during continuous interaction (Peeters et al., 2020), and they in turn its unrelenting information-processing capability, the algorithm could
reinforce the acknowledgement of their respective agencies in the process the information on the candidates and update the ranking
decision-making processes (Weiss & Spiel, 2022). In the context of regardless of the exact time the application came in; the TA team would
human–human interaction, mutual learning is a natural phenomenon thus obtain the best matches at the top of the list in real-time throughout
because teammates recognize the need to co-adapt and become pre- the entire selection process, which constituted a huge opportunity to
dictable and explainable to facilitate collaboration (Harvey et al., 2022). improve the efficiency of the department’s operations.
In the case of human-AI interactions, it should be noted that learning After exploring the market, a decision was made to start collabora-
extends well beyond the training of the algorithm, and that every tion with IIC, a research-and-development (R&D) institute specializing
interaction is an opportunity to extract knowledge about how to best in big data. The objective was to develop an algorithm to support the
perform the task at hand by learning from each other’s mental models screening of the massive selection processes in the bank, focusing on the
and consequences of their chosen course of action (van Zoelen et al., sales force, who constituted the bulk of the positions in the more than
2021). 12,000 branches in the country. From the different types of AI-based
learning models, a decision was made to choose an ML supervised-
Research design learning application, mostly used by companies because its mode of
Case-selection strategy operation is relatively understandable and the final objective of the
process is more controllable by humans (Balasubramanian et al., 2022;
We adopted an interpretive case-study approach, which focuses on Kaplan & Haenlein, 2019). In the most extended uses of supervised ML,
revealing how a theory applies to a particular context (Eisenhardt & humans feed the algorithm with a set of predetermined categories and a
Graebner, 2007). Our setting for the case study is the Talent Acquisition large set of data, and the machine learns to estimate the correspondence
(TA) department of Santander bank in Spain, which provides recruit- of cases to one of the categories. The development of these algorithms
ment and selection services to source a national region of 20,000 em- comprises a training stage in which the AI builds the learning model, and
ployees. We followed several criteria for selecting this case. First, it is a further testing phase involving the AI making decisions autonomously
“particularly suitable for illuminating and extending relationships and with a human monitoring the quality of the outcomes (Campesato,
logic among constructs” (Eisenhardt & Graebner, 2007, p.27). The case 2020).
is theoretically representative of the organizational context of a large
company that plans for the implementation of an AI-based application; 2.5. Data collection
therefore, it faces the challenges and opportunities of developing
human-AI teaming to create value for the organization. Additionally, the The data collection included the entire process of AI-teaming
AI application is an algorithm that supports hiring processes, for which development over a period of four years. The most differential aspect
the collaboration between humans and developers has been regarded as of the case study was the monitoring of the development and
4
C. Sim´on et al. J o u r n a l o f B u s i n e s s R e s e a r c h 182(2024)114783
implementation of the algorithm from its inception by the first author, users’ feedback. Finally, attendance at conference presentations
who met with the TA manager by-monthly for a systematic follow-up. constituted a relevant source of observations of team members in a
This allowed us to meet one of the prior conditions in case study scenario in which they formally reflected upon the dynamics of the
research, which is “the development of testable, relevant and valid project and summarized what they jointly considered to be the key
theory requires intimate connection with the real world” (Verleye, takeaways from the project in terms of interactions, performance, and
2019). Our monitoring focused on the meso-level of analysis of DC mutual learning.
development, that is, the set of collective interactions based on pro-
ductive dialogue that might eventually translate into transformation of 2.6. Data analysis
the department’s routines.
The project team was composed of five persons: two on the side of the We followed a theory-elaboration model of analysis (Verleye, 2019)
developers (a project manager, psychologist with postgraduate training because we rely on theoretical insights of the meso-level DC framework
in AI-based applications, and a data scientist with experience in machine to identify the teaming mechanisms that create value but challenge them
learning), and two domain experts (selection technicians with a long by including AI as a unique member of the team. The data analysis was
experience in hiring processes) and the TA manager on the side on the conducted in three main stages using themati |
158 | deloitte | data_ai_trends_report.pdf | Data and AI
Trends Report
2024
The impact of generative AI
Data and AI Trends Report 2024
New opportunities,
new technologies,
new skills.
Gen AI is here – and it's a game-changer! This revolutionary Changes are rippling across the entire data stack in response
technology will disrupt industries and transform our lives more to this new era. To learn more about how technologies are
profoundly than ever before. Data is the fuel for AI, and what shifting, Google surveyed hundreds of business and IT leaders
powers its effectiveness. To truly take advantage of gen AI in with questions about their goals and strategies for harnessing
your enterprise, you need the ability to access, manage, and gen AI. This report delves into their perspectives for 2024 and
activate your structured and unstructured data across a variety beyond, offering valuable insights for organizations looking to
of systems. capitalize on gen AI within their enterprise.
Furthermore, your data can also benefit from AI and machine
learning (ML) for deeper understanding, to enhance models, or
improve customer experiences. Success hinges on achieving all
of this while maintaining a high level of data quality and
security, while upholding responsible data use principles.
Page 2
Data and AI Trends Report 2024
1 2
Top 5 trends
Gen AI will speed the
delivery of insights The roles of data
at a glance:
across organizations. and AI will blur.
5 minute read 6 minute read
3 4 5
Operational data
AI innovation will will unlock gen AI 2024 will be the year
hinge on strong data potential for of rapid data platform
governance. enterprise apps. modernization.
3 minute read 4 minute read 5 minute read
Page 3
Data and AI Trends Report 2024
Trend 1 Trend 2 Trend 3 Trend 4 Trend 5
Gen AI will speed the
delivery of insights
across organizations.
Page 4
Data and AI Trends Report 2024
Trend 1 Trend 2 Trend 3 Trend 4 Trend 5
Nearly 2/3 of data decision
makers expect a democratization
of access to insights in 2024.
84% believe gen AI will help their
organization access insights faster.
Google Cloud Customer Intelligence Trends Research Survey, 2024. Page 5
Data and AI Trends Report 2024
Trend 1 Trend 2 Trend 3 Trend 4 Trend 5
Why should you care?
“Moody's deep expertise in
understanding financial data,
It’s almost impossible to overstate how Modern BI tools were already developing ways
disclosures, and reporting
significantly gen AI has changed the to bring data to everyone who needed it;
uniquely position us to anchor
technological landscape. In the case of reports embedded in the most relevant
business intelligence (BI), as tools become context for the data, such as account insights
development of fine-tuned large
more accessible, even non-technical team appearing in a salesperson’s CRM, is an easy
members will be able to benefit from these example. But those insights have always language models. Google
insights; driving productivity and needed to be carefully curated by an analyst.
Cloud’s gen AI will help our
disseminating knowledge faster than ever The end user has always been a step removed
before. That means better data literacy across from the data. Connecting a large language customers and employees
your organization, smarter decisions being model to your business data closes that gap.
produce new insights faster than
made, and ultimately greater success in the Team members can interact with your data
market.
intuitively and conversationally, or create
ever before.”
reports and dashboards by simply ‘talking’ to
52% of non-technical users are already
your data or making a simple search across
using gen AI to draw out insights today.
your business. In fact, many of the
Nick Reed
organizations surveyed for this report are
Chief Product Officer, Moody's Corporation
already putting this into practice.
Page 6
Data and AI Trends Report 2024
Trend 1 Trend 2 Trend 3 Trend 4 Trend 5
AI is already used by
both the most advanced
data scientists and within
lines of business.
Essentially, tools that connect people to key intelligent visualizations, which will be
business data through natural language will be integrated with productivity tools and
a major force in bridging existing gaps in business applications. As many applications
organizational skill sets.
allow users to see how others found
successful answers to questions, people will
Throughout 2024 and beyond, expect to see
also be able to benefit from aggregate
more business users ‘talking’ to their data
knowledge, as well as gaining insight into
using search and leveraging a conversational
which interactions have had the greatest
UI to create reports, dashboards and intuitive
impact over a day, a quarter, or a year.
Page 7
Data and AI Trends Report 2024
Trend 1 Trend 2 Trend 3 Trend 4 Trend 5
AI for all.
62%
Large language models connected to business data will allow even non-technical team members
to ‘talk’ to their data using search. Plus, a conversational UI can be used to create reports,
dashboards and intuitive intelligent visualizations, which will be integrated with productivity tools
and business applications.
47%
This shift is already underway. Here's how data decision makers responded to the question:
"Which type of non-technical users in your organization have been leveraging generative AI
42%
41%
to draw insights from your organization's data?"
37%
36%
33%
32%
23%
19%
16%
Security Logistics Administrative Business Customer & Finance & Human Product
Operations Sales Marketing,
Development Account service Accounting Resources Management Advertising & PR
Page 8
Google Cloud Customer Intelligence Trends Research Survey, 2024.
Data and AI Trends Report 2024
Trend 1 Trend 2 Trend 3 Trend 4 Trend 5
“Wendy’s introduced the first modern pick-up window in the industry
more than 50 years ago, and we’re thrilled to continue our work with
Google Cloud to bring a new wave of innovation to the drive-thru
experience. Google Cloud’s gen AI technology creates a huge
opportunity for us to deliver a truly differentiated, faster, and
frictionless experience for our customers, and allows our employees
to continue focusing on making great food and building relationships
with fans that keep them coming back time and again.”
Todd Penegor
President and CEO, Wendy’s
Page 9
Data and AI Trends Report 2024
Trend 1 Trend 2 Trend 3 Trend 4 Trend 5
“Gen AI isn’t just enhancing BI, it’s reinventing it. In 2024, insights
won’t be uncovered – they’ll be proactively surfaced, with more
nuance and context than ever before.”
Irfan Khan
President & Chief Product Officer, SAP HANA Database & Analytics, SAP
Page 10
Data and AI Trends Report 2024
Trend 1 Trend 2 Trend 3 Trend 4 Trend 5
The roles of data
and AI will blur.
Page 11
Data and AI Trends Report 2024
Trend 1 Trend 2 Trend 3 Trend 4 Trend 5
80% of respondents agree
that the lines between data
roles are starting to blur.
Google Cloud Customer Intelligence Trends Research Survey, 2024. Page 12
Data and AI Trends Report 2024
Trend 1 Trend 2 Trend 3 Trend 4 Trend 5
Why should you care?
“When I was little, my mom
would spend hours with a travel
As the use of AI becomes more widespread, As processes are streamlined, the roles of
agent planning our vacations.
the speed at which companies can go from data and AI will become increasingly blurred;
Working with Google Cloud to
raw data to AI will become increasingly meaning formerly siloed teams will need to
important. work more closely together than ever before.
incorporate generative AI allows
The organizations that master this process will us to create a bespoke travel
be able to make better decisions, launch new
concierge within our chatbot.
products and services faster, and provide
superior customer experiences.
We want our customers to go
beyond planning a trip and help
them curate their unique travel
experience.”
Martin Brodbeck
CTO, Priceline
Page 13
Data and AI Trends Report 2024
Trend 1 Trend 2 Trend 3 Trend 4 Trend 5
According to research, more than half
Many data analysts are now taking on (54%) of digital leaders say skills
shortages prevent them from keeping up
responsibilities that were traditionally
with the pace of change.
reserved for data scientists and vice versa.
Nash Squared Digital Leadership Report, 2023.
Data analytics and engineering, AI, and This interlocking allows users to:
business analytics are the most scarce skills
within organizations. Gen AI presents an
Have a common workspace for data
opportunity to boost productivity of existing
engineers, analysts and scientists that
data teams and workloads, thus assisting with
supports multiple coding languages such as
this widening skills gap. To be able to
SQL, Python, and Spark.
seamlessly use data and AI platforms allows
organizations to improve productivity, and
Extend software development best practices
innovate faster by accelerating their data to AI
such as CI/CD, version history and source
journey.
control to data assets, enabling better
collaboration and hand-offs.
Data and AI tools are also becoming
increasingly interconnected in order to help
users streamline data and AI workflows.
Page 14
Data and AI Trends Report 2024
Trend 1 Trend 2 Trend 3 Trend 4 Trend 5
Gen AI is also providing employees with
ways to accomplish more technical tasks.
For instance, tools can suggest the new lines of
code required to update a financial-reporting
system or outline the A and B versions of a
marketing campaign or otherwise create first
drafts that human employees can take and
implement into live production environments.
The organization of the future: Enabled by gen AI, driven by people, McKinsey & Company, 2023.
Page 15
Data and AI Trends Report 2024
Trend 1 Trend 2 Trend 3 Trend 4 Trend 5
Bringing AI directly to data can activate its
full potential regardless of its format.
A significant challenge hindering organizations The advent of advanced AI and Machine
from fully utilizing the potential of data lies in Learning (ML) technologies has revolutionized
the substantial amount of untapped, the way organizations leverage their data.
unstructured data generated today. This These cutting-edge technologies offer
includes formats such as images, documents, unparalleled opportunities to unlock the full
and videos. It is estimated to cover roughly up potential of all data, regardless of its format;
to 80% of all data, which has so far remained structured, semi-structured, or unstructured.
untapped by organizations.
Similarly, multi-modal AI has opened up a world
of possibilities for organizations, unlocking new
Structured data, characterized by its
levels of efficiency and accuracy when tuning 80% of the global datasphere
organization in fixed fields and columns, such
and grounding models in their enterprise data.
will be unstructured by 2025.
as in spreadsheets or databases, can be easily
Text embeddings enable vector searches
processed and analyzed using traditional
directly on data, without the need for complex
methods. However, unstructured data - think
and time-consuming preprocessing steps. This
VentureBeat, 2022.
social media posts, emails, customer call
simplifies the process of finding relevant
recordings, clinical documentation, and sensor
information, identifying patterns and trends,
readings - is often complex and challenging to
and clustering similar unstructured data in
interpret, making it difficult to extract
sources like documents.
meaningful insights.
Page 16
Data and AI Trends Report 2024
Trend 1 Trend 2 Trend 3 Trend 4 Trend 5
“The most transformative aspect of 2024's data and AI landscape
isn't just about efficiency – it's about democratization. By seamlessly
interconnecting these technologies, we empower not just data
scientists, but business users across the organization to unlock
actionable insights and drive innovation.”
Ali Golshan
CEO, Gretel
Page 17
Data and AI Trends Report 2024
Trend 1 Trend 2 Trend 3 Trend 4 Trend 5
AI innovation will hinge on
strong data governance.
Page 18
Data and AI Trends Report 2024
Trend 1 Trend 2 Trend 3 Trend 4 Trend 5
66% of organizations have at
least half of their data dark,
posing significant risk.
Five Factors For Planning A Data Governance Strategy, Forbes, 2023 & Gartner Glossary, Dark Data, 2024.
Page 19
Data and AI Trends Report 2024
Trend 1 Trend 2 Trend 3 Trend 4 Trend 5
Fewer than half of respondents
Why should you care?
(44%) are fully confident in their
organization’s data quality.
Google Cloud Customer Intelligence Trends Research Survey, 2024.
This explosion of new technology has its Similarly, most respondents (54%) consider
drawbacks, too. Many organizations are their organizations only somewhat mature
discovering new vulnerabilities and when it comes to data governance and only
weaknesses, especially when it comes to the 27% consider their organizations either
quality of their data. It’s not enough to just extremely or very mature.
apply LLMs to data – these models need to be
On the plus side, many organizations are
grounded in good quality enterprise data or
already taking steps to ensure data accuracy,
otherwise risk hallucinations. Organizations
data quality, and trust. The majority of
which take a practical approach to data
organizations surveyed
governance, quality, and trust will be in a
strong position to deliver tangible business
believe they are building a data driven
outcomes with AI.
culture
Most respondents are only somewhat
are centralizing data governance oversight
confident (45%) in their organization's data
quality, and another 11% are even less than are building centralized policy
somewhat confident. management, monitoring, and auditing.
Google Cloud Customer Intelligence Trends Research Survey, 2024. Page 20
Data and AI Trends Report 2024
Trend 1 Trend 2 Trend 3 Trend 4 Trend 5
Snap Inc uses Google’s Data Cloud to deliver a business domain-specific,
self-service data platform across distributed data, with decentralized data
ownership but centralized governance and visibility. With increased data
efficiency they can focus on improving the user experience and boosting
engagement.
Carrefour uses Google's Data Cloud to achieve zero trust network
protections, improving data security and strengthening secure access to
business-critical applications. Their data-centric infrastructure provides
flexibility to make changes very quickly and deliver the highest quality
service to their customers.
Page 21
Data and AI Trends Report 2024
Trend 1 Trend 2 Trend 3 Trend 4 Trend 5
What should
69%
organizations look for?
It’s key that organizations look for secure-by- End-to-end data lineage.
design data platforms that fully integrate data Automatically generated lineage to track data
encryption. The right platform should flows, perform impact analysis, and use
automatically catalog the data you own and lineage as a foundation for governance and 31%
give you the ability to logically unify and compliance across data and AI models.
organize your data leveraging metadata. This
Unified governance for data and AI assets.
enables you to centrally secure and govern
Central policy management, monitoring, and
data, based on your business context, and use
auditing for data authorization, retention, and
built-in automation and intelligence around
classification.
data profiling, quality, lineage, and more to
better manage data at scale. This enables:
Data quality. Auto-generate data quality rules 69% of employees had bypassed
to measure for completeness, accuracy, and
their organization’s cybersecurity
validity of your data.
guidance in the past 12 months.
Gartner Predicts Nearly Half of Cybersecurity Leaders Will Change Jobs by 2025, 2023.
Page 22
Data and AI Trends Report 2024
Trend 1 Trend 2 Trend 3 Trend 4 Trend 5
“2024 is a watershed moment for generative AI. Organizations that
fail to manage the risks throughout the AI development lifecycle will
be left behind. Those who proactively establish strong AI governance
practices are the ones who will unlock the true potential
of this technology.”
Felix Van De Maele
CEO and Co- Founder, Collibra
Page 23
Data and AI Trends Report 2024
Trend 1 Trend 2 Trend 3 Trend 4 Trend 5
Operational data will
unlock gen AI potential
for enterprise apps.
Page 24
Data and AI Trends Report 2024
Trend 1 Trend 2 Trend 3 Trend 4 Trend 5
71% of organizations plan
to use databases integrated
with gen AI capabilities.
Google Cloud Customer Intelligence Trends Research Survey, 2024. Page 25
Data and AI Trends Report 2024
Trend 1 Trend 2 Trend 3 Trend 4 Trend 5
Why should you care?
Businesses are excited about the potential Operational databases and warehouses with
with large language models (LLMs). They have vector support help bridge the gap between
all experienced the power of tools like Gemini LLMs and enterprise gen AI apps. This is why
and other large language models, but they we’re seeing so much interest in vector search
also recognize that the creative nature of and vector databases and why Retrieval-
these tools is not a good fit for most Augmented Generation (RAG) is an important
enterprise use-cases. Enterprise gen AI technique for enhancing and augmenting
applications face a variety of challenges that LLMs and gen AI models. We’re seeing a lot of
LLMs alone do not address; they need to innovation across the industry and much of it
provide accurate and up-to-date information, is driven by the open source community
offer contextual user experiences and do all including PostgreSQL, one of the most popular
this while not breaking the bank. databases for developers.
Page 26
Data and AI Trends Report 2024
Trend 1 Trend 2 Trend 3 Trend 4 Trend 5
What do organizations want
from AI-powered
databases?
Seamless connectivity to AI models, the ability to ground LLMs using techniques like RAG and the
ability to use natural language for database administration are the most important capabilities
when using AI in databases.
45%
40% 40%
38%
35%
33%
29%
Seamless Ability to ground LLMs using Ability to use natural language Ability to use natural Simplifying database Tight integration with AI Built-in, high-performance
connectivity to AI techniques such as retrieval for database administration, language to generate code migration and code tooling and frameworks vector search
models augmented generation (RAG) governance, and compliance conversion
Google Cloud Customer Intelligence Trends Research Survey, 2024. Page 27
Data and AI Trends Report 2024
Trend 1 Trend 2 Trend 3 Trend 4 Trend 5
Get it right and reap
the rewards.
The true power of gen AI is unlocked when Databases that fail to integrate
gen AI capabilities are likely to
operational data is integrated with gen AI to
become obsolete.
deliver real-time, hyper-personalized, and
contextually-relevant experiences across
Having AI closer to the operational data will
enterprise applications. Simply put, gen AI-
also allow developers to iterate quickly and
enabled operational databases holding
enhance the experience with all available data.
relevant business data will be the key to
You can do this where your data already lives
unlocking gen AI in the enterprise.
because databases are already powering all
applications, so organizations don’t have to
Successful databases will evolve to be AI-first,
learn or set up an entirely new system and it is
and deeply integrate technologies such as;
significantly more cost effective. In addition,
vector search, seamless connectivity to AI
with open source technologies like
models, support for natural language to SQL,
PostgreSQL, developers can get started
and tight integrations with AI tooling and open
sourc6e 9fra%me woorfk se. Amll thepselo wiyll bee enastiv ehlya d bquyicpklya wsitsh feamdil iar tools and capabilities.
built into operational databases and will
their organization’s cybersecurity
become table stakes.
guidance in the past 12 months.
IDC FutureScape Worldwide AI and Automation 2022 Predictions.
Page 28
Data and AI Trends Report 2024
Trend 1 Trend 2 Trend 3 Trend 4 Trend 5
“We explored several new entrants in the database market that focus on
storing vectors and ended up trialing several. And given Linear’s existing
data volume and our goals for finding a cost-efficient solution,
we opted for Cloud SQL for PostgreSQL once support for pgvector was
added. We were impressed by its scalability and reliability. This choice
was also compatible with our existing database usage, models,
ORM, etc. This meant the learning curve was non-existent for our team.”
Tom Moor
Head of US Engineering, Linear
Page 29
Data and AI Trends Report 2024
Trend 1 Trend 2 Trend 3 Trend 4 Trend 5
“Customers are looking to leverage the power of LLMs by
augmenting them with their domain knowledge and enterprise data.
To support these new use cases, cloud-based database solutions
that also embrace open-source gen AI orchestration frameworks will
provide application developers with the capabilities to help them
quickly and more efficiently build Retrieval Augmented
Generation (RAG) applications.”
Harrison Chase
CEO, Langchain
Page 30
Data and AI Trends Report 2024
Trend 1 Trend 2 Trend 3 Trend 4 Trend 5
2024 will be the year
of rapid data platform
modernization.
Page 31
Data and AI Trends Report 2024
Trend 1 Trend 2 Trend 3 Trend 4 Trend 5
Only 14% of organizations
are satisfied with their
legacy databases’ support
for AI, indicating there is a lot
of room for improvement.
Google Cloud Customer Intelligence Trends Research Survey, 2024. Page 32
Data and AI Trends Report 2024
Trend 1 Trend 2 Trend 3 Trend 4 Trend 5
Why should you care?
As more and more organizations seek to take The gen AI boom is bringing new urgency to
advantage of the opportunities gen AI brings, database modernization because the most
many are discovering their legacy databases popular AI tools for working with vectors,
are holding them back due to lagging models, and data run in the cloud and are
technology and poor user experience. In based on open source database technologies
addition to outdated technology and a poor such as PostgreSQL. In addition, the most
developer experience, legacy databases have advanced AI models run only on major cloud
also caught the attention of C-level executives platforms.
because of their expensive, unfriendly
licensing and vendor lock-in, which often
result in millions of dollars in unnecessary
annual costs.
Page 33
Data and AI Trends Report 2024
Trend 1 Trend 2 Trend 3 Trend 4 Trend 5
Top challenges with
legacy databases.
23%
19%
17%
13%
12%
10%
4%
Licensing costs Lack of cloud-first Lack of integration with Vendor lock-in Limited data model Limited pool of Lack of community
and practices architecture cloud services options practitioners motivated to support
work with this technology
Google Cloud Customer Intelligence Trends Research Survey, 2024. Page 34
Data and AI Trends Report 2024
Trend 1 Trend 2 Trend 3 Trend 4 Trend 5
Smooth transitions are more
possible than ever.
Thankfully, migrating from legacy databases is IT decision makers are now comfortable
becoming easier with database migration approving large modernization projects as
tools and programs continuing to improve and they look to embrace open technologies,
mature. We’re also seeing AI help to augment including gen AI, as part of their innovation
these tools, to the point where breaking free roadmaps.
from legacy databases is now much easier
with AI-assisted code conversion, code
completion, and improved efficiencies.
Page 35
Data and AI Trends Report 2024
Trend 1 Trend 2 Trend 3 Trend 4 Trend 5
“Data migration tools have been around forever, but more recently,
they've been getting smarter with the ability to do AI assisted code
conversion and code completion. The hardest part of the migration
is transforming the data and training the new applications to fit in
the new database. Both are made easier through gen AI. You can
use a model to look at a source database and find out how to
transform the data into the destination database. You can get
some quick wins, and ultimately get developer productivity. There's
still a ton of legacy stuff, and gen AI is bringing the bar down to
simplifying migrations.”
Andrew Storrs
VP Data Engineering, Aritzia
Page 36
Data and AI Trends Report 2024
Trend 1 Trend 2 Trend 3 Trend 4 Trend 5
“Character.AI is a pioneer in the design and development of open-
ended conversational applications. Our gen AI platform utilizes our
own advanced neural language model to generate human-like text
responses and engage in contextually relevant conversations.
When we found AlloyDB for PostgreSQL, we were stuck between a
rock and a hard place. Usage of our service had scaled
exponentially, putting unique stresses to various parts of our
infrastructure, especially our databases. Google Cloud's AlloyDB
and Spanner provide a solid foundation, delivering reliability,
scalability, and price performance for our workloads, from
engagement and operations, to AI and analytics.”
James Groeneveld
Research Engineer, Character.AI
Page 37
Data and AI Trends Report 2024
Trend 1 Trend 2 Trend 3 Trend 4 Trend 5
“Legacy systems were built for a different era. Gen AI's potential lies
in its ability to find novel connections and insights. In 2024,
organizations that liberate their data from outdated systems will be
best positioned to stay ahead of the curve.”
Gopal Srinivasan
Principal - Generative AI, Deloitte Consulting
Page 38
Data and AI Trends Report 2024
Trend 1 Trend 2 Trend 3 Trend 4 Trend 5
An entire generation of
developers is building AI
applications and leveraging AI
for more efficient coding,
improved database
performance insights, and
enhanced security posture.
Are you one of them?
Page 39
Data and AI Trends Report 2024
How Google
Cloud can help.
Google Cloud helps organizations unify data We also provide cutting-edge AI/ML and
and connect it with groundbreaking AI to generative AI capabilities that are readily
unleash transformative insights and available for your data, enabling all of your
personalized experiences. By harnessing the people to easily and quickly access the data
simplicity, scalability, security, and intelligence they need and unlock its true value. All of this
of Google's unified data and AI approach, is delivered with enterprise-grade efficiency.
businesses can unlock the full potential of It’s this unique combination that makes
their data in a single, streamlined solution.
Google Cloud an unparalleled partner for
turning raw data into organizational value.
Because Google Data Cloud consolidates
workloads and manages the entire data life
cycle, data teams are empowered to develop
modern, data-driven applications using
popular open-source engines and models.
Page 40
Data and AI Trends Report 2024
Inside our one-of-a-kind approach.
Fully connected data and AI. A unified data foundation. The most open data platform for Enterprise-grade efficiency and
security at scale.
modernization.
As new ways of interacting with systems and Google Cloud's unified data foundation is built Google Cloud is committed to being the most Google Data Cloud is an industry leader in
data emerge, it’s clear that organizations need on BigQuery, and brings your data together open cloud provider, letting you build modern, efficiency, security, and scale; catering to
AI models grounded in quality enterprise data into one place, integrating structured and data-driven applications wherever your organizations of all sizes and adhering to the
that allows for analytical insights and unstructured data with AI to deliver insights workloads are. We support open source and most stringent enterprise requirements.
augmented experiences. across your entire data estate. This unified
open standards, and offer managed database
data foundation allows you to manage your We help make it easy for organizations to
services that are fully compatible with popular
With Google’s Data Cloud, data teams can use entire data lifecycle and help make data share data safely and securely across
open-source engines and models.
gen AI tools to activate their enterprise data access, management, governance, and organizational boundaries, run queries across
across BigQuery and AlloyDB, and use built-in analysis easier for different types of users With AlloyDB Omni and BigQuery Omni, you exabytes of data with blazing speed, and
features to easily apply AI/ML directly to their within an organization, effectively removing can utilize data and modernize your process billions of transactions – all with
data. For instance, BigQuery ML allows data data silos. applications across Google Cloud, AWS, Azure, generally lower cost.
teams to construct ML models straight on
and Google Distributed Cloud, without
their BigQuery data simply using SQL, and Our highly scalable architecture unifies
incurring the costs, security risks, and
even call foundation models in Vertex AI. Built- transactional and analytical systems, enabling
governance concerns associated with data
in vector embedding capabilities in AlloyDB tightly integrated data services across
migration. It’s now easier than ever to get
also allow users to store and generate BigQuery, AlloyDB, and Spanner. This allows
started with gen AI on a data platform that
embeddings within their data stores to help easy data analysis from Spanner to BigQuery,
meets you where you are on your
augment their LLMs and support their gen AI with virtually no impact to the underlying
modernization journey.
use cases. transactional workloads.
Page 41
Data and AI Trends Report 2024
So,
Ready to join
the party?
what’s next?
If you’ve got any questions about the content of this report, or
want to know more about how Google Cloud can support your
organization, our experts are always on hand.
Talk to an expert
Clearly, 2024 is shaping up to be an exciting - Naturally, many of these new opportunities
and pivotal - year for many organizations. require new skills, and for existing processes
Those who are able to prepare their people to be refined. Organizations that embrace the
and platforms to fully embrace new need to upskill and fully equip their people will
capabilities made possible with gen AI will not quickly find that this investment pays
only see short-term productivity gains, but dividends in the form of almost limitless
begin to effectively future-proof their potential.
organization against ever-evolving
competition.
You can also take our Data & AI Strategy Assessment to discover
how ready your organization is for AI-powered digital transformation
and receive expert recommendations to get you there faster.
Take the assessment
Page 42
Data and AI Trends Report 2024
Methodology.
The Google Cloud Customer Intelligence Region Company size Role
team conducted a global research study on
NORAM 180 1,000 to 4,999 19% Business Development 2%
Data & AI Trends with 410 Data Decision
Makers from 12/18/2023 - 1/17/2024. Active EMEA 104 5,000 to 9,999 20% IT or IS (Information Technology, 40%
Computer Engineering, Security,
recruitment was paused from 12/23 to 1/1 for etc.)
JAPAC 76 10,000 to 49,999 30%
winter break. Respondents included a mix of
Software Development 1%
LATAM 50 50,000+ 31%
Data, IT, and business leadership roles with
Technology Strategy or Product 11%
seniority ranging from C-level to Manager. All
Development
Industry Role level
respondents were employed at 1,000+
Marketing/Advertising/PR 16%
employee organizations currently using data Financial services 74 C-level 14%
Operations 3%
products & services. Respondents did not
Retail (e.g. Grocers, Stores, 42 VP or equivalent 20%
Boutiques, Franchises,
know Google was the research sponsor and Product Management 3%
Restaurants, etc.)
Director 43%
the identity of participants was not revealed
Research/Analytics/Strategic 6%
Technology 85
Manager 7% Planning
to Google.
Other 103
Lead / Head 4% Sales 1%
Data Science 15%
Interaction with data products and services
Hands on 40%
Strategic/oversight 60%
Page 43
Data and AI Trends Report 2024 |
159 | deloitte | amc-ai-the-future-of-public-engagement.pdf | Artificial Intelligence (AI) and
the Future of Public Engagement
How government and public sector organizations are using AI to
streamline communication and serve increasingly diverse populations.
January 2025
As the use of AI continues to erupt populations they serve. In just over
across industries, constituent a year since the launch of OpenAI’s
engagement and targeted outreach GPT-4, our research1 showed that
are proven front-runners for early nearly 1 in 4 commercial organizations
adoption of the technology. This is had already begun using AI in their
good news across all industries, but marketing and outreach operations.
especially for Government, which relies This statistic, and many like it, reveal a
heavily on constituent engagement clear truth: AI isn’t merely a buzzword
for establishing trust, transparency, or a fleeting tech trend; it represents a
accessibility, and reliability. paradigm shift in how we understand,
process, and disseminate information.
Despite the importance of this field
in fulfilling mission requirements Despite this truth, the wealth of
for Government agencies, public information available on AI can make
sector organizations face unique it difficult to understand the use cases
challenges and expectations as it and applications that are right for you.
relates to how they engage with To cut through the noise, leaders from
the people they serve, including Deloitte’s GPS Advertising, Marketing,
heightened expectations among & Commerce2 (AM&C) practice have
consumers, expanded diversity of identified three key opportunities to
audiences served, demand for timely enhance existing outreach practices
and accurate communications, and with AI. The case studies that follow will
increased complexity of regulatory describe real-world examples of public
environments in which they operate. sector organizations that have taken
advantage of these AI opportunities
Fortunately, the AI revolution
and as a result have built trust with
presents exciting opportunities
the public, created efficiencies for
to improve efficiencies for public
their audiences and employees, and
sector organizations at all levels and
increased impact of their messaging
transform how they engage with the
and outcomes.
2
USING AI TO AI offers new opportunities for public Public sector organizations don’t have
sector communicators at each step of to re-invent the wheel — they can
ENHANCE the marketing and outreach process. infuse AI into their Existing Outreach
AI-powered analytics can uncover and Engagement Frameworks.
PUBLIC behavior patterns and preferences
with precision in real-time, Generative At Deloitte, we see three
OUTREACH
AI (GenAI) tools can produce hundreds clear ways for public sector
of tailored options, and Machine organizations to utilize AI in existing
AND
Learning (ML) technologies can now frameworks, reducing barriers to
make real-time data-driven decisions creating impactful outreach and
ENGAGEMENT
to maximize every dollar. engagement with their constituents.
AI AI AI
OPPORTUNITY OPPORTUNITY OPPORTUNITY
1 2 3
Gaining Deeper Developing Hyper- Optimizing Messaging
Audience Insights Personalized Content Delivery
EXISTING OUTREACH PROCESS: EXISTING OUTREACH PROCESS: EXISTING OUTREACH PROCESS:
UNDERSTAND DEVELOP CONTENT DISTRIBUTE CONTENT
YOUR AUDIENCES FOR YOUR AUDIENCES TO YOUR AUDIENCES
Public sector organizations Citizens demand messaging that It isn’t only about crafting the perfect
are charged with serving resonates with their individual lives and message — public sector organizations
incredibly diverse populations situations. Gone are the days of a “one- must also make sure that message
while simultaneously delivering size fits all” outreach approach. While lands in front of their primary targets at
personalized services. The key this challenge is a universal hurdle the right time. This goes beyond effort
to success is developing a deep for all communicators to overcome, it and money; it’s about guaranteeing
understanding of their audience, takes center stage with public sector that every ounce of investment results
but achieving this is not always as organizations that must, at times, in a maximum impact when reaching
feasible as it sounds. address a population of 334+ million. the public.
AI Injection: AI-powered AI Injection: GenAI tools and AI Injection: Machine-learning
analytics tools can provide AI-powered Chatbots can deliver models can analyze performance
insight into GPS organization’s hyper-personalized content to of content in real-time and
audiences that no human eye or audiences, more efficiently optimize distribution to maximize
focus group can conjure up. than ever. campaign impact.
3
AI
Gaining Deeper
OPPORTUNITY
1
Audience Insights
PUBLIC SECTOR ORGANIZATIONS and how best to reach them: How old are they? Do they
live in urban or rural areas? How do they get their news?
PROVIDE DEEPLY PERSONAL
What do they care about most?
SERVICES TO HIGHLY DIVERSE
POPULATIONS.
AI-POWERED ANALYTICS TOOLS CAN HELP PUBLIC SECTOR
Populations aren’t just growing — they’re aging and ORGANIZATIONS UNDERSTAND THEIR AUDIENCES’ UNIQUE
diversifying. Nearly 1 in 6 Americans are over the age of 65, NEEDS BETTER THAN EVER BEFORE.
a proportion projected to rise to 1 in 5 by 20303. Further, the
2020 Census4 showed that 67.8 million US residents (almost AI-powered analytics tools use advanced algorithms and
20%) speak another language other than English at home. GPS machine learning (ML) techniques to analyze large quantities
organizations serve this diverse population on the front lines, of both historical and current data, extracting meaningful
directly administering essential services that maintain the patterns and insights about audience behaviors and
welfare of communities. The need to cater to constituents preferences. Through examining these patterns, public sector
with varied reading levels, native languages other than English, organizations can better understand the constituents they’re
and a variety of accessibility requirements makes a one-size- serving to develop personalized strategies for better serving
fits-all communications approach insufficient. them. This approach allows for highly-accurate audience
Whether a public sector organization is seeking to segmentation and sets a solid foundation for developing
communicate about a newly developed policy or legislation, compelling content.
increase usage of their services, or send out an emergency
notice, it is critical to know what makes their audiences tick
4
OPPORTUNITY 1 Challenge: Universities across the country are experiencing increased
enrollment pressure in the face of a shifting higher education landscape.
IN ACTION
To combat this pressure and stay competitive, Michigan State University
(MSU) partnered with Deloitte to develop new enrollment strategies that
Michigan State
would balance MSU’s headcount, selectivity, diversity, and net tuition
University uses AI revenue goals. Before they developed these strategies, MSU needed a
deeper understanding of their current student population.
and ML Insights to
Increase Enrollment5 Solution: Deloitte helped to implement the Candidate360TM, 6 solution, which
used AI and ML predictive models to combine US lifestyle data from 250M+
households with MSU data and uncover meaningful behavior patterns and
preferences. The insights — which provided individualized profiles of in-state,
domestic, and international prospects — helped recruiters prioritize limited
time and resources. The data in-hand empowered recruiters and provided
insight into optimal communications channels for reaching students, helping
them increase responsiveness and avoid phone call screening.
Outcome: The Candidate360TM models enabled both recruiters and enrollment
directors to better understand their key geographic, demographic, and academic
student profiles when crafting their strategic and operational plans. In the first
year of utilizing the analytical models, MSU benefited from a 24% increase in out-
of-state student enrollment and a $5M increase in net tuition revenue.
DELOITTE AI CAPABILITY
DELOITTE’S DISCOVER.AI PLATFORM PRODUCES REAL-TIME, DEEP AUDIENCE INSIGHTS
To help public sector organizations track, measure, language processing (NLP), allowing the agency to see what
understand, and improve interactions audiences have with questions their audiences are asking most, and why. This AI
the services they provide, Deloitte has teamed up with technology can also integrate other audience interaction points
Qualtrics7 to provide an experience management platform (e.g., voice, text, survey, email, web, and social media) which
called Discover.ai. The platform, currently used to support would allow the state agency to have a more holistic picture of
a large state public health agency, combines real-time their targets and how to better serve them.
audience insights with powerful AI capabilities. Coupling those
With unparalleled insight into targeted audiences, public
capabilities with expert advisory services creates a powerful
sector organizations can reduce the time and resources
foundation for continuous improvement of the programs,
spent on broad, ineffective communication campaigns.
policies, and applications used by the public and workers.
Insights derived from these AI analytics tools can increase
Consider a state transportation agency, for example, that has the accuracy and relevance of information disseminated,
implemented a new vehicle policy and is receiving hundreds of fostering trust among the public. These insights can be
calls per day from concerned constituents. Audio recordings paired with generative functions of AI to develop meaningful
from more than 100,000+ customer service calls can be fed and receptive content that is personalized to the audience
into the Discover.ai platform, powered by Qualtrics AI. The specifications identified within the AI Analytics tools.
system can transcribe and analyze those calls using natural
5
AI
Developing Hyper-
OPPORTUNITY
2
Personalized Content
TWO-WAY COMMUNICATION IS In this landscape of heightened demand for personalized,
real-time content, the challenge lies in transcending
ESSENTIAL FOR BUILDING TRUST
traditional communication barriers to effectively listen and
IN GOVERNMENT.
cater to individual needs, ensuring meaningful engagement at
That said, engaging with the masses at the 1:1 level presents a scale never before possible.
a unique series of challenges. A recent report8 showed that
82% of survey respondents indicated if government wants
AI CAN HELP HUMANIZE GOVERNMENT SERVICES BY
to earn or keep their trust, governments need to hear the
DEVELOPING INDIVIDUALIZED CONTENT AT SCALE.
public’s concerns and let them ask questions. But this kind
of listening at scale is often resource-prohibitive. On the
Conversational and GenAI9 tools, powered by large language
outreach front, public sector organizations face an increasing
models (LLMs), are a cutting-edge branch of technology that
demand for content. And not just for any old content—but
create new, original content or data by learning from existing
they want increasingly dynamic, personalized, evolving
examples — transforming how we generate ideas, solve
content delivered in real time. A survey1 we conducted with
problems, and create across various fields. Adobe Firefly,
650 communications executives showed that the volume
Google’s Gemini, OpenAI’s ChatGPT and DALL-E, and many
of content that organizations need to meet demand has
others can analyze extensive data on text, images, and videos,
increased by 54% on average in the last year, and that
and create new, human-like content that speaks to a unique
organizations are only able to meet content demands 55%
individual’s needs at a superhuman pace. This technology
of the time on average.
synthesizes information about tone, structure, and visuals
from existing content to produce original, audience-specific
materials at scale.
6
OPPORTUNITY 2 Challenge: Every year, the Colorado Department of Human Services (CDHS)
is inundated with thousands of policy-related questions from case workers
IN ACTION
across the State about their Supplemental Nutrition Assistance Program
(SNAP) and Temporary Assistance for Needy Families (TANF) programs. To
Colorado Uses
answer each appropriately, policy analysts are forced to navigate hundreds,
GenAI to Produce if not thousands, of pages of technical, complex policy and process rules.
Personalized Solution: Deloitte collaborated with CDHS to implement the Program Area
Natural Dialogue Assistant (PANDA). PANDA is a Generative AI powered Policy
Answers to Policy
Engine that makes documents and internal knowledge available through an
Questions
AI-search. Rather than combing through documents, policy analysts can now
type their question into PANDA, which identifies relevant policy and provides
reasoning and references for support.
Outcome: PANDA has allowed policy staff to comb through 500+ pages of
content in an instant, and has equipped them with more tailored, personalized
responses to inquiries. In addition, the solution has significantly reduced
research times — on average it only takes between 20-30 seconds to search
all policy documents, formulate a response, and provide the customer with
references and reasoning.
DELOITTE AI CAPABILITY
DELOITTE’S CREATIVEDGE TOOL USES AI TO GENERATE ON-BRAND, AUDIENCE-TAILORED CONTENT
As GPS organizations continue to face increasing demand for Tools like CreativEdge can empower public sector outreach
more creative assets needed faster, in multiple languages, coordinators and communicators with the level of hyper-
and with decreasing budgets, tools that streamline content personalization required for trust building and effective
development are becoming more and more essential. communications with the public.
CreativEdge10 is designed to address these challenges by These types of AI systems foster two-way communication,
generating audience-specific, brand-compliant content which meets the communication preferences of a digitally
with the click of a button. After collecting a simple set of connected audience and helps humanize public sector
inputs from a public sector user — outreach goals, existing organizations, making them more relatable and accessible
audience insights, and an organization’s brand guidelines — to the populations they serve. In addition, through
CreativEdge develops an editable, one-paragraph description innovative content creation, agencies can bolster their public
of your target audience persona. Using this persona, the tool engagement strategies, ensuring relevance and effectiveness
can then generate a variety of tailored content in the form in a rapidly changing digital landscape. Combining these
of social media posts, printable assets, emails, and even generative AI tools with machine learning algorithms can
several creative brand campaign concepts that can serve ensure that personalized content is shared with audiences
as thought-starters for a public sector outreach team. It effectively and timely.
can even translate this content into 19 different languages.
7
AI
Optimizing
OPPORTUNITY
3
Messaging Delivery
UNDERSTANDING AUDIENCES AND they receive it at lunchtime. Some utilize Instagram to
consume short form video content across a variety of
CRAFTING THE PERFECT MESSAGE
platforms and genres.
IS ONE THING — EFFECTIVELY
DISSEMINATING AND ACTIVATING
AI SOLUTIONS CAN HELP PUBLIC SECTOR ORGANIZATIONS
THOSE MESSAGES OVER MULTIPLE
MORE ACCURATELY REACH THE RIGHT AUDIENCE WITH
CHANNELS IS AN ENTIRELY THE RIGHT MESSAGE AT THE RIGHT TIME.
DIFFERENT BEAST.
ML solutions are uniquely suited to address this challenge by
In 2023, a report11 estimated that as much as $20 billion
optimizing real-time outreach across digital media channels.
of global digital advertising spend was wasted by reaching
ML tools can learn and adapt based on performance of
consumers who weren’t their intended audiences. This figure
content in real-time, enabling public sector organizations
represents more than just a monetary loss — it means
to make data-driven decisions more quickly and efficiently.
that a significant amount of effort was put into outreach
These algorithms can process vast amounts of data from
that did not result in a real impact. Given the roles public
previous campaigns, including user engagement, click-
sector organizations play in stewarding taxpayer dollars, it is
through rates, and demographic information. They can then
important that campaigns target the right segments at the
identify which content performs best on which platforms
right time. Achieving this, however, is no small feat.
and at what times, and then automatically distribute those
Audiences engage with content on different channels in
messages or adjust the allocation of outreach resources on
a variety of mediums, and at different times — some are
different channels.
more likely to open an email and engage with content if
8
OPPORTUNITY 3 Challenge: Due to the difficulty in communicating complex policies to widely diverse
populations, constituents are not always aware of the programs for which they are
IN ACTION
eligible. Not only is it difficult to uniquely identify this underserved population, it can
also be difficult to reach them in a way that resonates.
A State Public
Solution: Deloitte is collaborating with a major state public health agency to enhance
Health Agency
the accuracy and effectiveness of educational outreach by utilizing AI capabilities,
Expands SNAP beginning with the Supplemental Nutrition Assistance Program (SNAP). By integrating
existing data on current SNAP enrollees with insights from HealthPrismTM, 12, a
Access Using AI-
proprietary Deloitte asset containing over 1,700 Social Determinants of Health (SDOH)
Powered Outreach
for more than 250 million adults in the United States, Deloitte can identify all adults
across the state who are predicted to be eligible for SNAP benefits but are not currently
enrolled. With this identified population, Deloitte assisted the state in conducting
hyper-targeted outreach and educating potential enrollees about the program and the
enrollment process. This was achieved by using Salesforce Marketing Cloud (SFMC), an
AI-enabled marketing automation platform, to perform direct-to-consumer outreach.
Outcome: By delivering direct communications tailored to each recipient’s
preferences with personalized information about the SNAP program and how to
check eligibility, the agency was able to successfully reach several underserved
populations, many of whom had children, to educate them on the nutrition assistance
program and how to enroll. To date, roughly 30% of the successfully contacted
individuals took action to either check eligibility or apply for the program.
DELOITTE AI CAPABILITY
DELOITTE’S ALLIANCES ALLOW CLIENTS TO MAXIMIZE IMPACT AND CAMPAIGN EFFECTIVENESS
Deloitte’s significant investment in unique AI capabilities adaptive content creation through AI, while powerful
is bolstered by strategic alliances with over 60 of the customer relationship management (CRM) platforms can use
world’s leading companies13, including many of the major AI AI to re-segment audiences based on demographics and past
innovators. This combination of our own in-house business campaign engagement. Deloitte can then connect clients with
acumen and technologies, paired with industry-leading tools search engine and social media platforms who then can use
from organizations like Salesforce, Google, Amazon, Medallia, AI targeting to pinpoint the precise digital channels and times
Adobe, Sprinklr, and many others, positions Deloitte to deliver to reach underserved communities. After deployment of the
tailored solutions for optimizing messaging delivery in all messaging, our strategic allies with AI-powered social media
forms — not just on a single messaging platform or medium. monitoring tools can analyze sentiment in real-time, allowing
the campaign to course-correct messaging based on
For example, Deloitte can help a public health agency working
public response.
to promote flu vaccinations to leverage a variety of messaging
optimization solutions in their end-to-end outreach process. This collaborative model gives public sector organizations an
Deloitte’s alliance with leading graphic design software unprecedented level of precision and agility, maximizing the
companies ensures clients have visually compelling and impact of every message.
9
Addressing the Risks of AI In Public
Sector Outreach and Engagement
While AI solutions may usher in radical efficiencies and deeper connections with the public, they must be adopted responsibly.
Deloitte is prepared to help agencies navigate the key considerations and risks related to successful adoption of AI-assisted
outreach solutions, including:
Data Security. AI applications often handle sensitive citizen data. Understanding where and how AI tools store and
protect data, and regularly auditing for compliance, can mitigate the risks of data breach or misuse.
Bias and Fairness. AI may inadvertently perpetuate existing discrimination. Identification and correction of biases,
employment of diverse datasets in model training, and continuous algorithm monitoring can promote inclusive
outcomes.
Transparency and Clarity. Advanced technologies often lead to questions about accountability. Choosing AI tools
that offer user-friendly explanations and clearly communicating when and how AI in used communications can build
trust and understanding with stakeholders.
Regulatory Compliance. Use of AI in government communications must conform to policy and regulations. It is
critical that AI deployment is sufficiently informed by legal and compliance expertise, particularly as the law evolves
with novel technologies.
Oversight. Reliance on AI could lead to reduced human oversight. A balanced approach, where AI augments human
expertise, insight, and creativity, is key to developing effective communications that get the best out of both the
workforce and the technology that aids it.
Deloitte’s Trustworthy AI™, 14 framework aims to help public sector organizations address these risks by using ethical safeguards
across seven dimensions related to fairness, transparency, accountability, robustness, privacy, safety, and security. The framework
analyzes these dimensions throughout an AI system’s lifecycle, from the initial design and training to the deployment and ongoing
monitoring, to ensure that bias is minimized across stages.
By acknowledging and addressing these risks, government and public sector organizations can leverage the
transformative potential of AI in a responsible and effective manner, ensuring their outreach and engagement
efforts are citizen-centric and aligned with ethical standards.
10
IN AN INFORMATION -AND By embracing AI carefully and responsibly, GPS organizations
aren’t just adopting new technology; they can lead the way on
CHANNEL- SATURATED
more personalized, insightful, and proactive communications.
ENVIRONMENT, THE CHALLENGE IS
Beyond increasing efficiency, leadership on AI demonstrates
NO LONGER JUST ABOUT GETTING dedication to better understanding and more effectively
serving the needs of citizens. Furthermore, the transformative
A MESSAGE OUT. ENSURING
potential of AI solutions can unlock communications barriers
MESSAGES REACH, INFORM,
and streamline public engagement.
RESONATE, AND INSPIRE IS
It’s clear that AI isn’t just a tool: it’s a formidable ally, capable
CRITICAL TO THE RELATIONSHIP of amplifying and augmenting the impact, reach, and depth of
BETWEEN GOVERNMENT AND THE how we communicate. Embracing this ally means embracing a
future where public sector organizations don’t just speak—they
CONSTITUENTS THEY SERVE.
listen, understand, and respond—with precision and empathy.
ENDNOTES
1) Gen AI powers content marketing advantage for early adopters
2) Deloitte Government Marketing Services
3) U.S. Older Population Grew From 2010 to 2020 at Fastest Rate Since 1880 to 1890
4) Nearly 68 Million People Spoke a Language Other Than English at Home in 2019
5) Deloitte Higher Education Client Success Story — Michigan State University
6) Candidate360™ Higher Education Enrollment Solutions
7) Qualtricsxm | Artificial Intelligence (AI) For Experience Management
8) 2024 Edelman Trust Barometer
9) Designing for the Public Sector with Generative AI | Deloitte US
10) CreativEdge™: A GenAI digital marketing campaign engine
11) ANA Provides “First Look” at In-depth Programmatic Media Transparency Study
12) Deloitte HealthPrism™
13) Deloitte’s Ecosystems & Alliances relationships: We’re Better Together
14) Trustworthy AI™ Bridging the ethics gap surrounding AI
11
GET IN TOUCH
RJ Krawiec
Principal
Deloitte Consulting LLP
[email protected]
Eric Uhlir
Studio Senior Manager
Deloitte Consulting LLP
[email protected]
Evan Tunink
Senior Manager
Deloitte Consulting LLP
[email protected]
Peyton Marion
Senior Consultant
Deloitte Consulting LLP
[email protected]
As used in this document, “Deloitte” means
Deloitte Consulting LLP, a subsidiary of
Deloitte LLP. Please see www.deloitte.com/
us/about for a detailed description of our
legal structure. Certain services may not be
available to attest clients under the rules and
regulations of public accounting.
This publication contains general information
only and Deloitte is not, by means of
this publication, rendering accounting,
business, financial, investment, legal, tax,
or other professional advice or services.
This publication is not a substitute for such
professional advice or services, nor should it
be used as a basis for any decision or action
that may affect your business. Before making
any decision or taking any action that may
affect your business, you should consult a
qualified professional advisor. Deloitte shall
not be responsible for any loss sustained by
any person who relies on this publication.
Copyright © 2025 Deloitte Development LLC.
All rights reserved.
12 |
160 | deloitte | deloitte_ai-adoption-africa-2024.pdf | Source: openai
AI for Inclusive Development in Africa –
Part I: Governance
A I for Inclusive Development in Africa | Governance
Introduction
AI in Africa
2400 AI Companies in Africa,
40% founded since 2017
In South Africa, scientists at the University of
Johannesburg used AI to forecast the peak periods of
COVID-19, helping develop more effective policy
measures.1 In Ghana, StarShea used AI to connect
women farmers globally, increasing their earnings by
50% within six months.2 In Kenya, a groundbreaking
startup, SohpieBot,3 employs AI-driven chatbots to
handle inquiries related to sexual and reproductive
health. These are just some of the many AI use cases
illustrating the potential of AI adoption on the continent
to address critical social, health, and economic
challenges.
Across the continent, public and private sector interest in AI has been growing rapidly,
spurred in part by the capabilities of large language models like ChatGPT. Africa currently
counts over 2,400 AI companies,4 out of which 40% were founded in the last five years.
For African nations to sustain and amplify this growth trajectory, governments and the
private sector must prioritize AI in their investments. This is essential not only for driving
Source: unsplash
economic expansion, but also for accelerating Africa’s progress towards the Sustainable
Development Goals (SDGs), especially considering the recently adopted UN resolution on
AI Governance that aims to promote safe, secure, and trustworthy AI systems.
UN Resolution on AI Governance
AI adoption in Africa does not come without costs or risk. Generative AI and large
language models ingest vast amounts of data, generating concerns around privacy, data
March 21st, 2024: The UN General
security and copyright infringement. Predictive AI models could upend traditional
Assembly adopted a landmark resolution
decision-making and raises ethical questions around biased and inaccurate data.
on the promotion of “safe, secure and
Proponents of AI development in Africa face nascent AI regulations, a large data deficit,
trustworthy” artificial intelligence (AI)
and high capital and operating expenses.
systems that will also benefit sustainable
development for all.
To create strong enabling environments across Africa that can realize AI’s immense
potential, while minimizing risks, we believe that public and private sector decision-
makers should focus on bolstering four pivotal enabling areas: 1) Governance, 2) Data
and Digital Infrastructure, 3) Talent, and 4) Funding.
Today, many African nations lack national strategies, institutions, and regulatory
frameworks that address AI technologies. This governance vacuum creates uncertainty –
stifling investment and hindering innovation in the AI sphere. In this paper we explore
these dimensions of AI governance, including challenges and opportunities in each.
Subsequent papers in this series will focus on the other three enabling areas.
1
AI for Inclusive Development in Africa | Governance
Governance
Trust is pivotal for the successful adoption and cultural
Status of National AI Strategy and
Planning
acceptance of AI in Africa.
•
Countries that have adopted an
It acts as the foundation for the sustainable and beneficial integration of AI into society. AI National Strategy
However, as AI solutions become more widespread on the African continent, a multitude
•
of risks arise that could undermine that trust, such as personal data misuse, inaccuracies Countries in the process of
in outputs of AI models, and systemic biases amplified by AI. To deliver on AI’s potential adopting an AI National Strategy
socioeconomic benefits, uphold human rights, and align with values such as fairness,
•
accountability, equity, transparency, inclusion, and responsible technology use, African
Countries with no national AI
governments and private sector actors must establish a strong governance foundation,
strategy but integrating AI into
including strategic direction, implementing mechanisms, and regulatory and ethical
digital strategy or legal framework.
frameworks for AI adoption.
•
Countries with limited or no
consideration of AI
AI Strategy and Planning
National and regional strategies can accelerate and sustain the adoption of AI, providing
roadmaps to guide its development, implementation and use in a way that respects
societal values and norms and contributes to inclusive growth. Both the United States5
and European Union6 have released AI strategies that set visions, policies, priorities, and
action plans for enabling AI development and commercial uptake and “ensuring that AI
works for the people.” These could be potential models for the African Union and
national governments on the continent.
In 2023, the African Union convened AI experts to draft the African Union Artificial
Intelligence (AU-AI) Continental Strategy for Africa, set to be released in 2024.7 Yet, most
countries in Sub-Saharan Africa have yet to develop national AI strategies or policy plans.
The lack of clear direction hinders collaboration and integration of AI across economic
sectors, impeding coordination, prioritization, and resource allocation of AI efforts within
each country and across the continent.8
Efforts to establish AI strategies in Africa vary, falling into the following categories: 1)
early AI adopters—countries with national AI plans or strategies adopted; 2) countries
with plans or strategies in development; 3) AI integrators – countries incorporating AI
governance into a comprehensive digital strategy or within an existing framework; and 4)
Sources:
non-adopters of AI –countries that have no mention of AI in plans, policies, or strategies
- AI governmental initiatives | Digital Watch Observatory
at the national level. In the first two categories, Egypt, Rwanda, Ghana, Senegal, Tunisia, - National AI Policy Summary II Nw (ictworks.org)
- Le Président sénégalais annonce la finalisation d'une
and Nigeria stand out as the early adopters and have either created AI national strategies
stratégie nationale sur l'IA - AITN (afriqueitnews.com)
or are currently in the process of doing so. - strategie-nationale-d'intelligence-artificielle-et-des-
megadonnees-2023-2027.pdf (gouv.bj)
- AI Readiness Index - Oxford Insights
2
AI for Inclusive Development in Africa | Governance
In 2021, Egypt launched its National AI Strategy and established a National Council
Egypt’s AI Journey
for Artificial Intelligence. This comprehensive strategy serves as a guiding
framework for the responsible and strategic adoption of AI technologies across
various sectors within Egypt. Egypt’s vision includes positioning itself as a thriving 7.7% Increase in Egypt’s GDP is
hub for innovation, drawing in investments, and effectively addressing critical expected from its new AI
societal challenges to stimulate economic growth. The implementation of this National Strategy
strategy is projected to yield a direct impact of $42.7 billion USD by 2030,
equivalent to 7.7% of the nation's GDP.
In the second category, AI integrators, lies South Africa, which lacks a national AI
strategy but has established a Presidential Commission on the Fourth Industrial
Revolution (4IR)9 to develop a strategic plan for the country's 4IR vision, to become a
leader in emerging technologies, notably, Artificial Intelligence, quantum computing,
and smart manufacturing. Similarly, Kenya, which also currently lacks a national AI
strategy, relies on existing laws related to AI and digital technologies as a regulatory
framework.10
Countries in the last group show limited or no consideration of AI. This is often due to
policymakers not perceiving it as a priority and questioning their capability to pursue it.
These countries may be prioritizing other underlying challenges to AI adoption, such as
digital divides, talent gaps, and limited data and digital infrastructure, all of which can
make AI seem like a distant goal.
Other reasons for the slow progress in adopting AI strategies and necessary governing
mechanisms are the limited knowledge and understanding of AI's potential and its
societal implications among the policymakers. This lack of awareness can impede
effective decision-making and the formulation of suitable policies and regulations. As a
result, it can lead to a slower pace of AI adoption and missed opportunities for using AI to
tackle social, economic, and developmental challenges.
Source : freepik
Opportunities for Consideration
African governments must craft national AI strategies to serve as a foundation for
o
effective AI governance. Public officials should consider engaging AI leaders, industry
experts, corporate and international partners to create robust AI strategies that define
objectives and encourage cross-sector collaboration, with clear, measurable roadmaps
for implementation. Strategies and roadmaps should ideally be developed within the
context and aligned to a comprehensive economic growth strategy, guiding long-term
objectives, and maximizing potential. These guiding documents should go beyond
articulating a vision; they should include analysis and decisions to set priorities, optimize
resource allocation, and identify implementation needs. National strategies should align
with and support implementation of the African Union’s forthcoming continental AI
strategy.11
3
AI for Inclusive Development in Africa | Governance
In a 2019 Deloitte article on AI strategy for government leaders, authors
explained that “strategy isn’t just a declaration of intent, but ultimately should
involve a set of choices that articulate where and how AI will be used to create
value, and the resources, governance, and controls needed to do so.” Based on
this premise, Deloitte developed an AI version of its classic strategic choice
cascade framework to reflect the questions and considerations required for AI
adoption (see figure below). To be effective a government AI strategy should
cover five core components – a vision, prioritized focus, a clear definition of
success, capabilities needed, and supporting management systems. This
framework can be applied at the national level, to use AI for improved
government performance or to advance an economic growth agenda, or from a
sectoral lens, for the AI strategy of a specific ministry or public agency.
Source: freepik
Donor partners can provide tailored training and capacity strengthening for
o
policymakers to bridge the knowledge gap. Enhancing domestic AI policymaking
capabilities will empower African countries to autonomously shape AI policies that align
with their distinctive requirements and ambitions. Donors can build on existing models
such as the FAIR Forward program, which initiated peer-learning activities to enhance the
capacity of policymakers from Africa and Asia to respond to the benefits and challenges
of AI. The program was implemented by the Human Sciences Research Council (HSRC)
from South Africa, working with researchers and policy experts from Ghana, Kenya,
Rwanda, South Africa, Uganda, among other countries.12
4
AI for Inclusive Development in Africa | Governance
Public-Private Coordination & Implementation Mechanisms Institutions with public-private
coordination mandate for AI adoption:
There is often a lag and lack of institutional coordination between national AI public
stakeholders and private sector players and initiatives, limiting the potential of such Egypt: Established a National Council for
strategies to be effective in dirving economic growth and social development. To fully Artificial Intelligence to implement the
harness AI's benefits, effective institutional frameworks for AI policy and strategy national AI strategy, which serves as a
implementation and collaboration between the public and private sectors, are essential. guiding framework for the responsible and
strategic adoption of AI technologies
Nigeria, Rwanda, and several other countries are taking a proactive approach on this across various sectors.
front. To promote research and development in emerging technologies, the Nigerian
government established the National Centre for AI and Robotics (NCAIR).13 This Nigeria: The Nigerian government
collaborative endeavor engages government agencies, businesses, and academic established the National Centre for AI and
institutions in a shared mission, to create an African Hub for AI and Robotics. Robotics (NCAIR) to engage government
Furthermore, in May 2023, the Federal Government of Nigeria made a resolute agencies, businesses, and academic
commitment to generate one million jobs in the digital economy by providing accessible institutions to create an African Hub for AI
courses for professionals, in Artificial Intelligence, Cloud computing, Game Programer, E- and Robotics.
commerce, Digital Marketing, etc.14
Rwanda: The AI Hub developed a training
Similarly, Rwanda launched the AI Hub initiative flagship program to focus on building program in collaboration with the Rwanda
vibrant AI ecosystems to support startup researchers and entrepreneurs. The AI Hub Space Agency (RSA), German Aerospace
developed a training program for young professionals in collaboration with the Rwanda Agency (DLR), and the private sector in
Space Agency (RSA), German Aerospace Agency (DLR), and the private sector in targeting targeting machine learning programs for
machine learning programs for earth observation. The program’s goal is to make use of earth observation.
AI and machine learning to leverage geospatial data in the sustainable development of
the country.15 In addition, the Rwanda Information Society Authority (RISA) and the AI
Hub established a natural language processing fellowship with the aim of bolstering the
state’s technical skillset among the local population.
Egypt has also created an AI council to oversee its national AI strategy, and Kenya has
entrusted its national innovation agency16 with developing and implementing IT-related
policies, including those pertaining to AI and data. Meanwhile, Senegal is the only African
country represented in the Global Partnership of Artificial Intelligence (GAPI), a standards
body convening experts from the public and private sectors to develop governance
models and shepherd global innovation.17 These initiatives illustrate the vital role that
public-private cooperation plays in advancing AI-driven economic growth.
Source: freepik
Opportunities for Consideration
Government strategies should empower new or existing institutions or coordination
o
mechanisms to drive implementation. These institutions, whether national AI agencies
or commissions, can carry the mandate of facilitating dialogue, co-creating policies, and
ensuring coherent AI ecosystems. African countries can actively learn from each other
and other contexts to tailor governance mechanisms to their specific needs. In the US, for
example, the government established the National Artificial Intelligence Research
Resource Task Force (NAIRR) to help drive implementation and facilitate dialogue among
stakeholders of AI adoption.18 The aim of the Task Force was to stand up the national
infrastructure for AI development and research. The task force brings together experts
from academia, industry, and government to provide insights and drive implementation.
African governments should actively participate in global AI standards bodies. Global
o
standards bodies are developing and proliferating governance guidance for AI research
5
AI for Inclusive Development in Africa | Governance
and activities—influencing AI product and service delivery models. African governments
should be an active voice at the table, allocating funding and time to these efforts to
influence global decision-making and advancing their own economic and national
security needs. This includes applying for membership, leadership roles in policy
committees and funding African experts at technical committees at these foras for
deeper global engagement.
o Implementation institutions should have a mandate to foster public-private Without the right processes and
partnership. Establishing collaborative governance structures that bridge the gap safeguards in place, the adoption of AI
between the public and private sectors is essential. We’re seeing this in other areas of can exacerbate existing digital divides,
opportunity for digital enablement. For example, in Senegal the National Meteorological including between expat and local
Agency was granted the mandate and liberty to establish PPPs as part of a weather and populations, men and women, urban and
climate data value chain from government to end users. Public agencies charged with rural residents, and along formal
implementing AI strategies should be similarly encouraged and equipped to partner with education level and income lines.
the private sector to strengthen the entire ecosystem.
Regulatory and Ethical Framework
As AI relies heavily on data, including personal and sensitive information, certain laws
and regulations related to data collection, transfer, protection, and cybersecurity have
significant impact on the ethical and safe adoption of AI. Much of the data employed for
AI tools and solutions comes from Global North users. In general, citizens in the Global
North typically have more advanced digital infrastructure. This easy access to digital
technologies leads to greater use of these technologies by Global North users, which in
turn creates greater representation of Global North users in the data within such
technologies. As a result, datasets may reflect the demographics, preferences and
behaviors of population in the Global North, potentially leading to biases when AI
systems are applied at the global level. AI has the potential to further exacerbate the
digital divide between the digitally underservedand the highly digitalized countries and
pose risk for disadvantaged or marginalized groups because these groups may not be
represented in AI training data.19
Similarly, across the African continent and within each country in the region, insufficient
infrastructure and other factors limit access and use of digital tools and technology for
certain groups more than others. Without the right processes and safeguards in place,
the adoption of AI can exacerbate existing digital divides, including between expat and
local populations, men and women, urban and rural residents, and along formal
education level and income lines.
Furthermore, AI as a tool can be co-opted for pernicious uses, such as for promulgating Source: freepik
mis-, dis- , or mal-information, which can have devastating effects on social and political
stability. For example, mis-,dis-, and mal-information have been weaponized to increase
instability in the Sahel region where low literacy rates, existing political tension, and and
uptick in social media use have combined to make an already complex situation even
more challenging.20
Additional challenges and gaps related to AI and data protection, data transfer, and
cybersecurity are significant. According to UNCTAD, most African countries have already
adopted a data protection law21 or are in the process of drafting legislation. However,
due to the rapid evolution of technology, regular updates are necessary. For example,
Tunisia enacted a personal data protection law in 2004. Despite being advanced, this law
6
AI for Inclusive Development in Africa | Governance
faces challenges, including the General Data Protection Regulation for the European
Union22 (GDRP) compliance, effective enforcement, raising awareness among businesses
and the public, and adapting to rapid technological changes.
Protection of inventions also stands as a crucial regulatory challenge for AI providers.
Indeed, AI's unique intellectual property challenges, particularly in algorithm protection
and AI-generated content, are also emerging concerns. Today, in the majority of
countries, AI software is protected only by copyright covering the source code. However,
this protection can be limited as it does not extend to the underlying ideas, methods, or
algorithmic functionalities.
One example of how this governance gap can manifest is in the level of caution used
by AI startups in their data collection. Several interviewed startups admitted that they
have used data from surveillance cameras to create their database and train their
models without the required authorizations from the national data protection
authorities due to a lack clarity and guidance on regulatory requirements. This reveals
a critical need for better regulatory understanding and support systems to ensure that
AI development aligns with legal and ethical standards.
Addressing these risks is essential to creating a safe, secure, inclusive, and transparent
enabling environment for ethical AI adoption and innovation in Africa. In some cases,
rapid adoption of a strict regulatory framework around AI might hamper innovation. For
example, in 2023 tech industry leaders in Kenya raised concerns about the government's
proposed bill for the Kenya Robotics and Artificial Intelligence Society, arguing that such
legislation could potentially stifle innovation, especially considering that the AI sector in
Kenya is still in its nascent stages.23 Conversely, policymakers recognize the importance
Source: freepik
of safeguarding consumers' interests by providing a legal framework for the
establishment and operation of AI technologies.24 This situation highlights the critical The European Union approved the AI Act:
need for a regulatory approach that safeguards users while also fostering technological The AI Act, approved on March 13th, 2024,
advancement and innovation. Recently the European Union approved an AI Act that “aims to protect fundamental rights,
establishes a foundation of safeguareds.25 democracy, the rule of law and
environmental sustainability from high-risk
Opportunities for Consideration AI, while boosting innovation and
establishing Europe as a leader in the field.
Prioritize the establishment of data protection, data transfer, cybersecurity and The regulation establishes obligations for
o
intellectual property regulations that are robust and aligned with international AI based on its potential risks and level of
standards. Adapting these laws is crucial to address the intersecting risks posed by AI, impact.”
such as algorithm protection, ownership and usage rights of AI-generated content and
the use of personal data with the appropriate protection. Data must not only be secured
properly, but ownership rights must also be clearly defined, recognizing both individual
and collective rights. Individuals should be allowed to opt-out or remove access of having
their data used as part of an AI solution. On ownership, resolving conflicts between data
sovereignty and accessibility can be challenging, especially regarding national
governments' control over data, but ethical use would ultimately place power into the
hands of those having their data collected. Reconciling data accessibility with privacy
protection is crucial. While AI innovation relies on diverse datasets, ensuring privacy
rights through measures like anonymization, masking, encryption, or differential privacy
is essential. Striking a balance between accessibility and privacy requires collaboration
among stakeholders and the development of clear ethical guidelines, security controls,
and regulatory frameworks. This not only boosts consumer confidence in African
organizations but also enhances their competitiveness on the international stage.
7
AI for Inclusive Development in Africa | Governance
Conversations about a unified regulatory framework should also be considered for
African nations. Just like the European Union, which last February approved the AI Act to
ensure that AI systems are safe, transparent, and accountable, fostering trust in this
rapidly evolving technology.
Equip local public actors with the knowledge to prevent and redress harm that results
o
from AI implemented with a global but not local perspective in mind. Engaging local
actors within the ecosystems strengthens capacity both to make ecosystem-informed
strategic investments and to develop effective safeguards for AI technology and data.
Local actors may include host country governments, technology companies, digital rights
activists, civil society organizations, local financial institutions, academic institutions, and
regulatory bodies. Local actors impacted by AI should be engaged throughout the design
and implementation processes. This means equipping local actors with the knowledge,
skills, and tools that allow them to analyze and understand when and how the use of AI
Source: freepik
tools might result in unfair or unjust outcomes.
Trustworthy AI™ framework promotes
Improve inclusivity and stakeholder representation in AI design, deployment, the ethical use of AI within organizations.
o
governance, or policymaking, especially for underrepresented or marginalized groups. Trustworthy AI™ requires governance and
Inclusivity and representation of African countries, contexts, and citizens is imperative to regulatory compliance throughout the AI
mitigating potential AI risks and harms on a global level. Within the African context, lifecycle from ideation to design,
inclusion of marginalized groups or those with limited access, such as persons with development, deployment, and machine
disabilities, rural residents, women, and girls,26 can better position AI technologies to learning operations anchored on the
address equity issues rather than exacerbating them. Designing AI using principles of seven dimensions in Deloitte's
gender equity and social inclusion can reduce AI bias27 in recruitment tools that may Trustworthy AI™ framework—transparent
reflect discriminatory hiring practices; in financing tools that determine discriminatory and explainable, fair and impartial, robust
credit scores or loan approvals; or with health tools that influence diagnosis and and reliable, respectful of privacy, safe
subsequent treatment. Knowledge sharing on AI risks and harms with the public enables and secure, and responsible and
active collaboration, learning, and idea sharing to shape an ecosystem where the public accountable.
and those traditionally excluded in other forums can engage on AI-related issues. The
collective perspective creates opportunities for surfacing problems and identifying
current and future risks of AI. This in turn allows for AI use that is more equitable,
inclusive, and rights-respecting; accounts for, and mitigates, potential harms; and is
reflective of a more global reality.
Adopt a human-centered, responsible, and ethical approach to AI. Including
o
considerations such as: 1) participation and inclusion, to engage and empower diverse
and affected stakeholder in the design, development, and governance of AI, so that their
needs, preferences, and values are respected and reflected; 2) accountability and
transparency to establish clear and enforceable rules and standards for the behavior and
performance of AI systems, as well as mechanisms for monitoring, auditing and
redressing any harms or errors; and 3) fairness and justice to enable AI systems to be fair,
equitable, and nondiscriminatory, so they are able to promote the social good and
human rights of all people, especially those who are marginalized, oppressed, and
perhaps lack access to the technology. This is an area of potential support from technical
partners. Tools such as Deloitte’s Trustworthy AI™ framework described below can be
useful in defining parameters and standards across the various dimensions that need to
Source: Deloitte
be considered.
8
AI for Inclusive Development in Africa | Governance
Conclusion
Recognizing the diverse landscape of Africa, each
country presents a unique context for AI adoption, with
varying levels of maturity across the continent.
However, across the board, it is imperative for African governments and businesses to
prioritize developing robust governance and risk mitigation frameworks, including
strategy, institutional implementation capabilities, public-private coordination, and
regulatory and ethical standards, to reap the potential of the technology. Read more
about other critical AI enabling areas in the region as we continue this series on AI
adoption, covering talent, data and digital infrastructure, and funding.
Source: Shutterstock
9
AI for Inclusive Development in Africa | Governance
About the Authors
Courtney Keene is a Senior Manager at Deloitte who brings 15 years of experience working as a trusted strategic advisor on
programming to strengthen local systems and promote inclusive growth. With a focus on the African continent, she has
advised and led programs covering 25 African countries on agriculture, energy, governance, health, and water and
sanitation. Email: [email protected]
Aïcha Mezghani is a Senior Manager at Deloitte Afrique in the International Donor practice, leading several projects on
strengthening entrepreneurial ecosystems and conducting the design and implementation of public policies, promoting
innovation and ICT sectors. Email: [email protected]
Maryam Kyari is a Senior Consultant at Deloitte focused on conducting economic impact studies and devising corporate
social responsibility strategies for clients across energy, real estate, disaster recovery, and education. She aligns the
objectives of clients with regulatory guidelines and standard practices and develops governance structures to manage
investments and grants. Email: [email protected]
Ibrahim Almatri is a Senior Consultant in Deloitte’s Risk and Financial Advisory practice working with state and local entities
in the development and execution of risk management strategies and crisis recovery efforts. Email: [email protected]
Get in Touch
Kathleen O’Dell is a principal with Deloitte's Government & Public Services team and leads Deloitte’s International
Development practice, which includes Deloitte’s work with the U.S. Agency for International Development, Millennium
Challenge Corporation, U.S. Export-Import Bank, and United Nations, among others. Email: [email protected]
Adarsh Desai is a Principal in Strategy and Analytics practice within Deloitte’s Government and Public Services (GPS) group.
Adarsh leads strategy and implementation of Digital, AI, Generative AI, and Data Analytics solutions for International
Development Organizations and US Federal Government Agencies. Email: [email protected]
Mohamed Malouche is a Deloitte Advisory partner and business leader in the International Donors practice, leading work
on public sector modernization, local economic development, energy, entrepreneurship, and innovation across French-
speaking Africa. Email: [email protected]
Carlton Jones is a Deloitte Consulting Leader in Tanzania, the Strategy Leader for East Africa, and the African Lead Client
Service Partner, responsible for the USAID work in economic growth, industry and cluster competitiveness, trade and
investment, and value chain development. Email: [email protected]
Acknowledgements
The authors wish to acknowledge Elizabeth Villarroel, Ramzi Maatoug, Francesca Cavalli, Helena Buckman, Aymen Mtimet,
John Millock, Kwame Antwi, Mohamed Baccar Fayache, and Rali Sloan for their extensive contributions to the development
of this report. We would also like to thank our colleagues Carlton Jones, Kathleen O'Dell, Mohamed Malouche, Shrupti
Shah, Mohamed Sylla, Wessel Oosthuizen and Mulaudzi Rudzani for their insights and guidance.
Finally, this report would not have been possible without the time and invaluable insights shared by startups and SMEs,
entrepreneurs, donors, experts, and support organizations. The authors extend their sincere thanks to Aya ElGebeely
(Talents Arena), Ali Mnif (Digital Africa), Celina Lee (Zindi Africa), Daniel Djaha (Orange), Jethro Datamwin Apeawini (Classic
Data Lab), Olivier Gakwaya (Smart Africa), Sinda Ben Salem (Instadeep), Richard Nii Lante Lawson, Moez Ben Hajhmida
(Fairconnect.ai), Nicolas David (AWS), and Wayan Vota (USAID).
10
AI for Inclusive Development in Africa | Governance
1 “AI Is Here to Stay! How Artificial Intelligence Can Contribute to Economic Growth in Africa”. UNU.
June 23, 2023. Link
2 “AI for Africa’s Socio-Economic Development”. African Union Development Agency AUDA-NEPAD.
Link
3 “SOPHIE BOT Artificial Intelligence”. Link
4 “AI Media Group: 2022 State of AI in Africa report”. Link
5 “Fact Sheet: Biden-Harris Administration Takes New Steps to Advance Responsible Artificial
Intelligence Research, Development, and Deployment”. The White House. Link
6 “European approach to artificial intelligence”. EU. March 6, 2024. Link
7 “Artificial Intelligence is at the core of discuss |
161 | deloitte | deloitte-cn-fsi-ai-in-banking-en-240805.pdf | Changing the game:
the impact of artificial
intelligence on the banking
and capital markets sector
Contents
Overview: where are banks in the AI journey? 01
What impact can AI have on the bottom line and how? 04
How will the sector landscape change and who will be successful? 10
What is special about generative AI and where is this technology heading? 14
How will banks embed AI across the value chain? 17
What risks must be managed and how? 23
What are the key considerations for safe and effective execution? 26
How to get started, scale and drive adoption 28
Contacts 31
Endnotes 33
Changing the game: the impact of artificial intelligence on the banking and capital markets sector
1. Where are banks
in the AI journey?
01
Changing the game: the impact of artificial intelligence on the banking and capital markets sector
1. Where are banks in the AI journey?
Artificial intelligence Artificial Intelligence (AI) is already here and Considering the sector outlook more
shaping the wider world banks operate in. generally, the coming years continue to
will likely determine the
In automotive, Tesla and others delivered include macroeconomic and geopolitical
AI technology for sophisticated driver-assist uncertainty. Any number of unforeseen
banking and capital
functions, with an eventual end goal of events may emerge from an already cloudy
markets sector’s autonomous vehicles operating on public crystal ball. However, in a five-year timeline,
roads.1 The life sciences industry has been we in Deloitte Global Financial Services
winners and losers
realizing value from AI for drug research see AI as the single biggest controllable
and new molecule discovery, as it can draw opportunity for players to improve their
in the coming five
insights from massive data sets faster, competitiveness.
years. The journey process data and automate workflows
more efficiently, and convert insights into AI now allows banks to tackle challenges of
has already started.
actions to improve business performance scale in a way that, previously, would have
– from molecule to market.2 In public required many extra staff. If a particular
safety and security, for example in the function in a bank could be done better or
United Kingdom, London’s Metropolitan faster by adding one hundred extra trained
Police has trialled live facial recognition staff, it’s likely that AI can be transformative
(LFR)3 cameras in specific areas, to for that function. AI offers vast additional
accelerate identification of individuals the operational capacity, at low marginal
police are looking for. Regulating for the cost compared to hiring the equivalent
evolving use of AI is an ongoing challenge processing capacity as staff.
to lawmakers, for example the European
Union’s AI Act is intended to protect health, But more than that, the game in which
safety, fundamental rights, democracy players are competing will likely change.
and the rule of law, and the environment AI is on the threshold of a paradigm
from potential harmful effects – while shift. Through the work we do with
supporting innovation, particularly banks around the world we see leading
among European SMEs (small and innovators already making the step from
medium enterprises).4 AI as an ‘instrument of strategy’ (i.e.,
accelerating delivery of today’s business
Within this evolving societal context, AI is plan) to a ‘determinant of strategy’, where
not new to the banking and capital markets tomorrow’s business is planned around
(B&CM) sector. It has been in production new AI capabilities. JP Morgan Chase, which
for years in specific functions, including topped Evident Insights AI Index (which
algorithmic trading and trade surveillance. benchmarks how ready banks are for the
But the arrival of Generative AI (GenAI) incoming wave of transformation that
marks a new era, exploding the number of AI will bring) for a second year5, sees the
potential use cases and putting benefits in transformational impact that AI can have
the hands of the workforce. and plans to spend $1 billion or more
a year on AI capabilities.6
AI now allows banks to tackle
challenges of scale in a way that,
previously, would have required
many extra staff.
02
Changing the game: the impact of artificial intelligence on the banking and capital markets sector
An important point is that we do not see
FS sector Industry examples of AI-enhanced capabilities
AI displacing humans from the workforce
at large scale. Rather that AI augments
Retail banking NatWest reduced fraud by 6% as a share of UK
the workforce and drastically scales up
Industry (19% to 13%), including a 90% reduction in
processing capacity and quality. The role of
account opening fraud since 2019 which all contributed
the human workforce will naturally shift to a
to reducing operational costs. On the income side they
higher level, with a greater focus on design,
achieved a 5x increase in click-through for personalized
oversight and exceptions management,
lending on customized customer offers. 7
as well as having more bandwidth for the
relationship-based, customer-facing roles
Reduced credit card delinquency by 32% (brighterion
where human emotional intelligence is vital.
by Mastercard).8
Across financial services (FS) sectors,
Corporate and UK banks have been fully automating the loans
we are seeing the green shoots of AI value
transaction banking underwriting process up to US$100,000 (we have
being realized. Bloomberg was among the
seen up to US$250K).9
first to announce training their own model,
with BloombergGPT providing a means for
JPMorgan Chase developed a GenAI model to analyze
users to query and interact with complex
statements from the U.S. Federal Reserve to determine
financial data using natural language.
the nature of policy signals.10
Citigroup uses GenAI to assess the impact of new US
capital rules.11
Goldman Sachs is working on various projects which will
incorporate GenAI into its business practices. Among
An important the most mature of the projects include writing code in
English-language commands, and being able to generate
documentation.12
point is that
Morgan Stanley is using machine learning to identify
we do not see personalized investment ideas and suggest the “Next
Best Action”.13
AI displacing
Investment banking (IB) Projected 27% productivity increase across investment
and capital markets banks and 27%–35% front office employee
humans from productivity by 2026.14
Insurance Underwriting teams at a specialized insurer experienced
the workforce
a 113% productivity increase using generative
AI-supported workflows for underwriting submissions
at large scale. relating to bespoke policies.15
03
Changing the game: the impact of artificial intelligence on the banking and capital markets sector
2. What impact can AI
have on the bottom
line and how?
04
Changing the game: the impact of artificial intelligence on the banking and capital markets sector
2. What impact can AI have on
the bottom line and how?
Successful innovators The recent B&CM industry hype around AI will likely now act as the conduit that
AI could appear as the latest fad – another accelerates business impact and magnifies
can achieve a 5-15%
topic attracting much discussion but not value realization. We look at this more
ultimately leading to sustained operating closely later in this paper.
improvement in cost-
margin uplift. Most banks have invested
income ratio over the in strategic innovations in recent years as Ultimately, the significance of AI to the
cloud, data and digitization technologies sector will be assessed on the extent
next five years.
have advanced. Not all banks have yet that this innovation delivers sustained
achieved material improvement to their operating margin uplift. Here we consider
bottom line from these investments, the “size of the prize” given a typical cost-
particularly where they bolted new systems to-income ratio profile today and expected
and capabilities on to existing technology AI benefit themes.16 We see potential for a
estates, introducing additional cost and 5-7% positive contribution in 2-3 years, and
complexity without decommissioning 10-15% in 5-7 years. This view considers
legacy components. a wide range of banks, and smaller, more
nimble organizations including those with
However, the banks that have learned how currently high cost-income-ratios (CIRs)
to deliver innovation in their organization would find greater opportunity to achieve
will continue to outperform with AI, i.e., the higher end of this 5-15% range of
“the winners will keep on winning”. improvement.17
Successful cloud, data, analytics and
digitization initiatives have provided the
foundational capabilities for AI.
Figure 1. Cost reduction examples
Cost efficiency – examples Typical mid-sized universal bank Income growth – examples
Income = 100
Next generation market analysis/
Trading activities 10 predictive trading algorithms
5–7% uplift on trading income
Fees and
Improved customer retention
commissions 15
1–2% uplift fees/commissions
Workforce acceleration efficiencies
(more from less) across the board Cost = 60
Improved Customer
0–15% total staff cost
Staff Acquisition through
25 hyper-personalised marketing
IT development and
5-10% uplift Interest income and
maintenance acceleration
Technology staff fees/commissions
10–20% of IT staff cost 5–10
Improved credit-risk assessment Credit loss charges 5 Interest income (Net) Tailored loan pricing based
75
leading to fewer impairments Premises and on credit risk assessment
10–-15% saving in impairment charges equipment 10 2–3% increase on net
interest income
Other administrative/
Improved FinCrime/fraud
accounting
detection reduces litigation/redress
20
charges and fraud losses
Source: © 2024 Deloitte research. For information, contact Deloitte Global.
Note: this is an indicative ‘sizing’ view based on a typical cost/income profile in the industry and our ranged estimates of the potential of AI to improve performance in
specific areas. The examples shown are not exhaustive. Broad-brush costs for implementation/operating costs of AI, and for reduction/redeployment of headcount
are considered, while noting these may vary significantly across different organizations. The cost-income profile shown is informed by third party market data from
Refinitiv, Factica, Statista and selected publicly available Bank Annual Reports as available in Q4 2023.
05
Changing the game: the impact of artificial intelligence on the banking and capital markets sector
Where will the benefits come from? However, given AI risks and the evolving regulatory
AI, including GenAI, can bring advantages such as: landscape, AI without appropriate human supervision
may not be suitable for:
• Increased efficiency – automate repetitive tasks, freeing
human resources for more complex, creative or customer • Critical, fast-moving operations where timely human
facing engagement. supervision/intervention is not yet feasible.
• Improved accuracy – process vast amounts of data with • Customer/staff facing activities requiring human
greater precision and fewer errors than humans, leading to emotional intelligence (EQ).
more accurate predictions and outcomes.
• Regulatory-sensitive activities.
• Enhanced personalisation – analyze customer preferences
and behaviours to create tailored experiences, improving We see three key modes for achieving value through AI,
customer engagement. all of which combine AI and human strengths:
• Predict trends – make data driven decisions, detecting trends 1. A focus on productivity through personal agents;
and predicting changes in the market.
2. A focus on improving quality and process performance
• Creativity – new possibilities to create new possibilities for through specialist agents; and,
products, services and business models fostering innovation
and growth. 3. Large scale re-imagining of end-to-end processes
using the multi-modal capabilities of AI.
• Cost savings – streamlining operations, reducing errors, and
enabling better decision-making, AI can help save costs and The persona of core “agent modes” in which humans and AI
allocate resources more effectively. interact to implement the operating improvements that can
deliver financial impact. We stress the point that the benefit
• Protection – improving the effectiveness of financial crime in all three modes comes through combining human and AI
and loss prevention capabilities. strengths, not through large scale replacement of humans
with AI. Institutions should develop and strengthen the
• Accessibility – Make the services more accessible human skills to allow for adoption and value realization.
and affordable.
These modes will be leveraged in creating value across the
financial institution.
06
Changing the game: the impact of artificial intelligence on the banking and capital markets sector
Figure 2. Examples of AI personas
Personal Specialist Transforming
agent focus on agent focus on process focus
productivity improving quality on cost reduction
AI assist AI augmentation AI automation
10-20% potential 20-50% potential 50-80% potential
Executive and specialist roles Those with domain knowledge e.g. Customer facing and support roles
e.g. functional leaders, top levels investment manager, underwriters, e.g. contact centre agents,
of management relationship/account managers central services
Human strengths: Human strengths: Human strengths:
• Emotional intelligence • Relationship management • Problem solving and
• Creativity • Negotiation decision making
• Strategic planning • Domain knowledge • Compassion
• Persuasion and negotiation and experience • AI ethics and regulation
• Motivational leadership • Story-telling and making • AI-Human task management
• Ethical judgement and integrity insights relevant
• Critical thinking Machine strengths:
Machine strengths: • Fraud detection and prevention
• Analyze data and generate Machine strengths: • Data categorization
content • Speed in insight gathering • Quicker processing times
• Schedule meetings • Error checking and • Language translation
• Provide real-time assistance and validation exercises • Voice and text sentiment analysis
suggestions on documents • Trend spotting and
simple graph design
• Trading algorithms
• Predictive analytics
• Routine forecasting
Source: © 2024 Deloitte research. For information, contact Deloitte Global.
07
Changing the game: the impact of artificial intelligence on the banking and capital markets sector
Driving down cost through efficiencies and loss prevention
Most banks are currently building AI business cases around cost reduction, and this is no surprise.18 It is easier to get funding approved
for initiatives which drive out cost. The impact tends to be delivered quicker and benefits tend to be more directly attributable to the
investment made. As AI grows in its ability to take on the increasingly sophisticated tasks that previously required human action, the
opportunity grows for banks to perform a wider scope of activities faster and better, doing more with less.
Key cost reduction themes will likely include:
1. Workforce acceleration 2. Engineering transformation
A “marginal gains” approach to deploying Specifically, to benefit bank’s large
many productivity improvements across technology functions, GenAI can already
Cost
the human workforce. At the most generate and optimize software code,
efficiency
basic level, this will include automation reducing the time to write, while improving
of repetitive tasks such as data entry quality. As many software engineers examples
and analysis, search and query, draft in banking information technology (IT)
production of many varieties of tend to be relatively inexperienced and
operational content (meeting minutes, requiring oversight from senior engineers,
communications) and summarizing large GenAI “co-pilots” have the ability to
3. Loss avoidance
documentation. This is the type of “text and accelerate production releases and make
Risk management, fraud prevention, cyber,
images” productivity support perhaps most maintenance less onerous.
legal and other brand protection functions
associated with GenAI, particularly among
have high potential for improvement
newer users.
through AI. These functions tend to be
AI applied: a Portugal based improved by speeding up processes,
However, we see that the art of the institution has deployed an expanding scope of processes, and
possible is rapidly expanding, with more AI-powered converter tool that providing wider sets of data inputs to
specialist acceleration use cases including converts software code from legacy improve process performance – all of
data governance and management, COBOL-based systems to their which AI readily supports. Specifically,
data quality and remediation, model target Oracle platform to accelerate AI-enhanced credit risk management
development and analytics. a core platform modernization improvements can result in fewer loan
program. The large language impairments and write-off charges. Fraud
Workforce acceleration will likely require model (LLM) based converter prevention and financial crime (FinCrime)
widespread uplift in workforce skills with AI automatically generates functional processes can be accelerated and
in the same way as staff previously became documentation of the legacy expanded using AI to review a wider set of
proficient in typing, spreadsheets and COBOL code and creates a target input data sets to uncover new insights on
calendar management and other functions metadata schema to accelerate the actors and ultimately reduce losses.
which historically were performed by technical specification and build of
specialist resources only. the new data platform.
AI applied: Legal outcomes
A second use case is the ability for prediction. A Middle-East based
AI applied: various proprietary GenAI to consume millions of lines of bank is trialling a GenAI tool based
GenAI tools are being deployed in legacy code that is undocumented, on past contracts and litigation
compliance teams to summarize and rapidly extract business outcomes. The tool examines the
large sets of documentation issued rules/requirements to accelerate contracts and other documentation
by government and regulators.19 modernization. Deloitte practitioners involved in legal disputes and helps
This rapidly makes the key are already leveraging these the legal team better predict likely
takeaways and major insights capabilities to accelerate client’s outcomes of legal matters, as well
available to compliance teams and transformations and modernize as highlight potential risks in new
business staff in frontline roles. our own products internally.20 contracts.21
08
Changing the game: the impact of artificial intelligence on the banking and capital markets sector
Growing revenues through new capabilities and improved retention
While more difficult than cost-cutting, players will likely also invest to grow revenue. Revenue growth is a key challenge for banks due to the
relatively limited number of “opportunities to influence”. Consider supermarkets, which have practically limitless opportunities to influence
consumer purchasing behaviour through ranging, discounts, multi-buy offers and more. Unlike supermarket customers, how often do retail
banking customers re-mortgage, change current account or take on a new loan or credit card?
Conversion rate for any sales campaign is a critical metric for AI to improve. When consumers do switch financial products, pricing/rates
are a key factor in the decision, as is trust, and the quality of relationship that the consumer perceives with the bank – influenced by
service level and relevance of interactions and offers. AI can improve all of these factors, while reducing cost to deliver. We see a number
of key revenue-impacting themes:
2. Customer experience and retention
AI-powered digital agents (e.g., chatbots) AI applied: Advanced chatbot.
Income can reduce customer wait times by Bunq, a Netherlands-based
growth addressing an increasing range of neobank, has recently introduced
complexity of customer requests. While its very own generative AI platform
examples
certain customer journeys (e.g., those called Finn. This innovative platform
associated with large transactions, is designed to impress customers
bereavement etc.) must remain as person- with its exceptional ability to
to-person interactions, the improved provide answers to a wide range
1. New capabilities for growth responsiveness of digital customer service of money-related queries. Finn
We see that banks will invest in revenue- agents can improve customer experience features a chat-style text box that
generating capabilities across business and retention rates. Increasingly, the quality allows users to ask questions
lines, including: of AI interaction with humans will improve or seek advice about their bank
as AI technology develops–adjusting the AI account, spending habits, savings,
a. Insight-driven pricing: real-time agent’s behavior according to the behavior/ and other financial matters. The
customization of pricing (e.g., preferential emotions of the customer. platform is capable of combining
lending rates) to make highly competitive data to provide answers that
offers to target customers based on go beyond simple transactions,
enhanced measurement of their AI applied: Service content such as helping users recall past
credit risk. management. A Netherlands-based experiences like “What was the
institution has implemented a
name of that Indian restaurant I
b. Hyper-personalized marketing: natural language processing (NLP) visited with a friend in London?”23
improved conversion rates based on chatbot to support front-line staff
insightful identification of individual in delivering a more insightful
prospect and customer/client needs, customer experience. The tool
and highly-tailored communication. enables service staff to query
wide datasets in real-time based
c. Next generation trading algorithms: on live customer requests, rapidly
trading income uplift from enhanced returning relevant responses from
market insight and automated trading product catalogue, account fees,
decisions. terms and conditions, policies
etc. The next phase will enable
customers to interact directly with
AI applied: A UK-based universal
the chatbot as a digital agent.
bank has increased click-through
rates on its personal lending offers
by five times, through personalized
offer content and improved target
selection.22
09
Changing the game: the impact of artificial intelligence on the banking and capital markets sector
3. How will the sector
landscape change and
who will be successful?
10
3. How will the sector landscape change and
who will be successful?
The competitive landscape will likely be
redrawn, with sector’s probable winners
and losers determined by the speed and
effectiveness with which their AI initiatives
enable evolution of their business
operations, products and services.
laitnetop
eulaV
Changing the game: the impact of artificial intelligence on the banking and capital markets sector
AI is changing the game As mentioned earlier – for leading
institutions, AI is already making the
– who will be the new
paradigm shift from being an instrument
of strategy, to a determinant of strategy.
winners?
Figure 3. How AI is changing the game
Game 1 Game 2 Game 3
• Same processes – • New processes – • New business –
lower cost same business strategy, segments,
• Step change in • Transform products, service,
efficiency and customer experiences
productivity (cloud, experience, • Distinctive
automation) personalise definition of
products/services purpose /
(digital, data, AI) contribution to
society
• Expanding the Art
of the Possible (AI/
GenAI)
High Changing
the game...
Game 3
Intelligent
banking
Game 2
Digital
banking
Game 1
Classic
banking But how to jump
from Game 1 to
2 to 3?
Low Time High
Source: © 2024 Deloitte research. For information, contact Deloitte Global.
11
Changing the game: the impact of artificial intelligence on the banking and capital markets sector
So, who will be successful? How? Execution is critical
Banks expected to capture the biggest Key technical foundations The players which have realized benefits
benefits from prior waves of technology- • Cloud, where done well, has delivered from prior technical innovations have
enabled innovation (e.g., cloud, digital, data) readily-scalable computing power learned and refined the delivery methods
will continue to outperform in their value and accessible data provisioning, that work in their organization. Typically,
creation from AI. This is because leaders in that abstracted data away from the these have included consideration of:
innovation have already invested in the key complexity of legacy architectures while
organizational enhancements, including reducing total cost ownership of the • Governance – putting in place sufficient
culture, governance, data management IT estate. It also forced banks to learn oversight to adequately assess and
and agile delivery methods, needed to how to assess and manage the risks mitigate the spectrum of risks, without
capitalize on the AI opportunity. associated with introducing third-party unduly constraining delivery;
dependencies to the infrastructure
In many ways, substantial prior supporting core business processes. • Culture – benefits are well
investment in the innovations (cloud, data communicated, business function owners
management etc.) mentioned above has • Automation put in place the governance expect to embrace emerging technology
prepared the ground for value creation and risk management capabilities to to improve process performance;
from AI. All of these investments required oversee automated operations.
considerable capital expenditure that • Idea to value – strong processes
has constrained the bottom-line benefits • Data governance may have been are embedded to generate ideas for
realized to date. However, as above, implemented initially for compliance value delivery from innovation, assess
the organizations that have successfully purposes but has established the feasibility and investment case, rapidly
invested in these ambitious infrastructural organizational accountabilities, policies, deliver the best ideas into production
changes will find AI to be the conduit that quality improvement methods and and scale;
now accelerates the unlocking of value. understanding of organizational data
assets to provide trusted datasets as • Talent – hiring and learning/development
inputs to AI use cases. approaches that build adequate skills and
capacity; and,
• Digital banking has evolved customer
expectations to be more comfortable • Partnerships – engaging with the wider
with self-service, real-time, insight-driven market ecosystem, forming partnerships
and reduce reliance on bank staff for with technology and service providers
many interactions, while streamlining best placed to assist delivery.
key front-to-back processes e.g., client
onboarding, loan fulfilment.
12
Changing the game: the impact of artificial intelligence on the banking and capital markets sector
Realizing the value from AI will take The continuous upskilling of teams who The continuous
more than simply enabling the use these new tools to do more is not
technology. In recent history there a one-time effort, it should be built into
upskilling of
have been great expectations that the talent model and measured. Banks
technology transformation will drive who simply implement AI and GenAI to
significant efficiency gains only to deliver augment existing processes will likely teams who use
underwhelming results. Global chief not see the full value realization and
technology officer of Dell Technologies Inc, could in fact only see increased costs.
these new tools
captures the frustrations of many senior Banks who leverage AI and GenAI to
executives with the sustained investment support continuous transformation and
required: “I must’ve had ten conversations improvement can take the foundational to do more is
last week where CIOs were bemoaning investments already made (e.g., cloud and
that they had run out of money or blown data) and unlock further value.
not a one-time
their [cloud] budget off.24 ”Why could it be
different this time? The past investments Considering these points, the FinTech
(e.g., cloud, automation) have been parts subsector is likely to move quickest, due to effort, it should
of a solution but ultimately have not yet distinct execution advantages. Namely:
delivered transformational bottom-line
be built into the
value. In the case of cloud, organizations • The relative simplicity of their current
may have built the new capabilities but not operating models (considering
yet switched off what these cloud-based products, processes, technology, data talent model
solutions were intended to replace. In and organization) makes them less
the case of automation, it was possible to encumbered by the constraints of legacy
and measured.
automate parts of a process with great systems and processes. They still have
precision but the technology struggled the flexibility to jump straight to newly-
with inferencing and being intuitive; it was conceived processes without lengthy
a brittle solution in areas that required re-engineering of legacy.
elasticity to be effective.
• They typically have a culture tilted to
AI is already interacting with the workforce more rapid growth and innovation – their
in a more natural way and opens the doors greater risk appetite means they will be
for entirely different processes. Solutions willing to push AI capability to customers
for these processes can now be developed and into production processes sooner.
not as 1’s and 0’s but rather with natural But there are risks associated with doing
language providing great flexibility and this, before having the appropriate
speed to solution. guardrails and risk infrastructure in place.
Therefore, true value to banks will be
delivered when costly and long duration
processes are reconceived. As banks
evolve in maturity with AI and GenAI they
will begin to give front line employees
increasing autonomy and improved tooling
that will enable increasing revenue
(see “insight-driven pricing”) while also
reducing non-value add work (e.g., data
entry). But once that tooling is in place and
banks begin to reconceive processes there
must be a focus to continue to redeploy
staff to higher value roles.
13
Changing the game: the impact of artificial intelligence on the banking and capital markets sector
4. What is special about
generative AI and where is
this technology heading?
14
Changing the game: the impact of artificial intelligence on the banking and capital markets sector
4. What is special about generative AI
and where is this technology heading?
GenAI is a branch of AI currently attracting GenAI is about more than just text:
much attention, as it allows for the Gen AI is capable of working with multiple
generation of increasingly sophisticated “modalities” of content, with the ability to
content (e.g., text, code, audio, images, process one modality as input and generate
videos, processes) based on algorithms that another as output (not all combinations
imitate existing content, using statistical shown in figure 4 are currently possible).
predictions learned from large sources. Gen AI is able to produce sophisticated
content output including software code,
The fast-improving apparent quality of PowerPoint presentations and three
this content suggests that GenAI can dimensional (3D) models.
play a large role in business functions
traditionally considered to require
solely human intelligence.
GenAI is predicted to be the start
What is different about GenAI
and why all the excitement?
GenAI rapidly generates sophisticated of a new era for AI. The technology
content, based on vast bodies of source
information, designed to imitate what
will continue to evolve with focus
a skilled human being could produce.
This could be for example summarizing
large volumes of documentation, writing on multi-modal communication
an opinion piece, developing software
code, producing images/video to a and intelligence built into human
given specification, preparing a sales
presentation or defining rules to
interactions.
measure data quality.
Figure 4. AI modalities
Text Text
Code Code
Audio Audio
Image Image
Video Video
Read more about the next six
3D/Specialized 3D/Specialized
modalities in our recent publication
Generative AI Dossier.
Source: © 2024 Deloitte research. For information, contact Deloitte Global.
15
Changing the game: the impact of artificial intelligence on the banking and capital markets sector
The increasing sophistication and GenAI is already being used: According
apparent qu |
162 | deloitte | gr-SEPE-Deloitte-Study-GenAI-eng-2024-noexp.pdf | Federation of Hellenic ICT Enterprises (SEPE)
Gen AI - opportunities and prospects for the Greek economy
December 2023
Executive Summary
Introduction Objective and Results of the present study Alongside the survey, Deloitte assessed the degree of the
impact of Gen AI on the Greek economy and the
In the era of digital transformation both internationally In the above context SEPE, in cooperation with the global
employment of ICT specialists.
and in our Greece, the Information and Communication consulting firm Deloitte, conducted the present study,
Technologies (ICT) sector has experienced significant with the main objective of exploring the opportunities More specifically, with regard to the projected impact of
growth in recent years. and prospects of Gen AI in Greece while in particular it Gen AI on the country's GDP, the analysis concluded that
focuses on Greek entrepreneurship (examples of use its impact is projected to be very significant, with its
This trend is expected to intensify even further in the
cases as well as a primary survey), but also on the cumulative impact being estimated at +5,5% of the
coming years, with the proliferation of Generative
expected impact of this new technology on the Greek country's GDP by 2030 (i.e. €10,7 billion ), which under
Artificial Intelligence (Gen AI), which as a branch of
economy and the employment of ICT specialists. certain conditions can even reach +9,8%. It is noteworthy
Artificial Intelligence has the ability to generate original
that around 50% of this impact is estimated to be
content (such as code, images, video, audio, text and 3D In the context of the study, Deloitte, on behalf of SEPE and
contributed by 5 sectors of the economy: Financial &
models) using big data processing. with the support of the National Documentation Center
Insurance Services, Wholesale Trade, Manufacturing,
(NDC), conducted a survey in the private sector, in which a
Under certain circumstances and conditions, the benefits Services and Information & Communication Services.
large number of companies from both the ICT sector and
that can be achieved for businesses by adopting Gen AI
from other sectors of the economy participated, in order As for the impact of Gen AI on the ICT-specialists gap, it is
solutions are multi-level and relate to both the internal
to record their views regarding, on the one hand, the use also expected to be significant, with the projected gap
operations of corporations (e.g. better decision making,
of Gen AI technologies/solutions and, on the other hand, between supply and demand to increase by ~25.500
cost savings, higher productivity) and extroverted service
their strategic/immediate plans regarding the adoption of specialists, reaching a cumulative total of ~83.000 gap in
provision(e.g. improved customer experience).
this new technology. specialists by 2030*. The implementation of policy
In particular, Gen AI solutions can be deployed across the measures to reduce the ICT skills gap is becoming
As the survey shows, the adoption of Gen AI at Greek
entire spectrum of a company's operations, withmain imperative, with particular importance now being given
enterprises of all sectors of the economy, is still in early
categories of use cases identified relating to customer / to focused and fast-track skills development programs for
stages, although the majority of businesses believe that
public-facing services, content generation, code STEM graduates as well as other academic backgrounds
adopting Gen AI solutions can improve efficiency and
management, knowledge assistance and the extraction of leading to certifications sought by the labor market.
boost their growth. In addition, it was highlighted that
insights from unstructured data.
both the majority of companies in the ICT sector have not
yet adapted their strategy for the integration of Gen AI
solutions.
* The ~83.000 positions are the new estimate of the cumulative ICT
specialist gap until 2030, following last year's study by Deloitte for SEPE,
Copyright 2023Deloitte Business Solutions S.A. whereby the gap was estimated at ~57.500 specia2lists
Table of Contents
Introduction to Generative AI - Gen AI 4
5 Main Categories of Gen AI Use Cases 10
Presentation of results of the SEPE / NDC / Deloitte survey on Gen AI 24
Impact of Gen AI on the Greek economy 34
Conclusions 46
Copyright 2023Deloitte Business Solutions S.A. 3
Introduction to Generative AI - Gen AI
Copyright 2023Deloitte Business Solutions S.A. 4
Generative Artificial Intelligence (Gen AI) | Historical Review
The beginnings of Generative Artificial Intelligence can be traced back to 1943. This technological field developed further at the beginning
of the 21st century, and since 2018 has seen exponential growth, with the largest technology companies entering the field dynamically
First Steps Development Acceleration
(mid-20th century) (early 21st century) (2018 - today)
1943: Development mathematical model of 2003: Development intelligent voice 2018-2019: Open AIreleases "GPT-1", a groundbreaking
neural networks, basis for modern neural assistantson mobile phones advancement forlarge language models (LLM)
networks, by Warren McCulloch and Walter
Pitts 2012-2014:A computer cluster Google
Brain is trained to recognize a cat from
millions of images, using the large-scale
CNNs technique. At the same time, research
is published on new image recognition 2021-2022:DALL-E develops a creation tool image of 12 billion
technique and introduction to Generative parameters that uses only one sentence to create an image and
Adversial Networks (GANs) the Stable Diffusionlaunches an open source model for image
creation
1973:Development of a series of programs
2017: Google releases the first model
known as "AARON" focusing on autonomous
2023: Major technology companies are turning their attention to
Transformer, the foundation of many
art production, by Harold Cohen
Generative AI, such as Adobethrough Firefly, OpenAIthrough
popular AI generation tools today, such as
ChatGTP-4, Metathrough LLaMAandGooglefreeing up public
"Chat GPT"
access to Bard, an AI chatbot
Source: Deloitte Report “Dichotomies”
Copyright 2023Deloitte Business Solutions S.A. 5
Generative Artificial Intelligence (Gen AI) | Definitions and main use
Generative AI is a branch of AI capable of generating original content by performing a learning process, unlike traditional AI, which
does not allow the development of original content
Definition of Generative AI Generative AI (Gen AI) is a branch of AI that can generate
original content, such as: code, images, video, audio, text and
ArtificiaI Intelligence
3D models.
The science of creating intelligent machines, through special computer programs -e.g. personal
digital assistants (e.g. Google Assistant, Apple Siri, Amazon Alexa)
Machine Learning
The above is illustrated in the adjacent "wheel",
The implementation of algorithms that allow computer programs to automatically improve
which categorizes the generated “output” into
through experience -e.g. a movie recommendation system on a streaming platform, based
individual use cases of Gen AI. Until now, the
on consumer preferences creation of this type of content has been carried
out exclusively with human intervention.
Deep Learning
Subfield of machine learning algorithms based on artificial neural networks -e.g.
autonomous vehicles that recognize obstacles, other people and other vehicles on
the road
Generative AI has redesigned the way we communicate, work and innovate, with its
Gen AI adoption expected to open myriad of possibilities that previously seemed unlikely,
A branch of AI that relies on large language models (LLMs) to process large ushering in a new stage of creativity, efficiency and progress.
amounts of data and generate original content.
The diffusion of this technology is extremely fast if we consider that ChatGPThas
recorded1 million users in5 daysfrom the day it was made available to the public
(November 2022), and according to the latest statistics for November 2023, it has
more than 180 million registered users.
1960 1980 2010 2020
Source: Deloitte analysis
Copyright 2023Deloitte Business Solutions S.A. 6
Generative Artificial Intelligence (Gen AI) | Differentiation from Traditional AI
Generative AI is a branch of AI, which however presents important differences from the latter
Traditional Artificial Intelligence (AI) replicates human Generative Artificial Intelligence (Gen AI) performs
cognitive functions (learning, design, creativity) and deep learning, mimics brain function when
focuses on identifying and processing available and processing data and making decisions. Gen AI is
appropriate information in order to extract the based on machines or algorithms that have the
relevantrequired "knowledge" from their composition. ability to createprimary content, e.g. text, images,
audio, video.
Main features of traditional AI:
Main features of Gen AI:
It does not develop primary content
Develops / Generates primary content
Handles certain problems well, for specific business functions
Solves open-ended problems by performing
Solvesproblems defined on the basis of specific rules intelligent, human action
Human supervision and assistance in the learning process is Supports the increase of creativity and the
necessary improvement of the quality of primary ideas
Interprets the information for pattern recognition
Requires limited human supervision and has
Enhancespredictabilityin decision-making autonomous learning potential
Performs a learning process on the basis of existing
information
Source: Deloitte analysis
Copyright 2023Deloitte Business Solutions S.A. 7
Generative Artificial Intelligence (Gen AI) | Benefits for businesses
The adoption of Generative AI can bring users multiple benefits such as better decision making, improved customer experience, higher
productivity, cost savings and improved creativity / innovation
Better decision making
Generative Artificial Intelligence (Gen AI) makes recommendations through big data analysis,
facilitating multi-scenario simulations and exploration of alternative strategies, enhancing and
accelerating the complex decision-making process
Improved customer experience
Generative Artificial Intelligence (Gen AI) helps improve customer loyalty through
personalized service and support
Higher productivity
Through Generative Artificial Intelligence (Gen AI), routine operations (that may make up to
70% of human resources time) are automated, allowing focus on more complex and higher
value-added tasks
Cost savings
Lorem ipsuGmenerative Artificial Intelligence (Gen AI) leads to cost and money savings (≥30%), through
the automation of often repetitive tasks, thus freeing up human labor, while ensuring high
quality
Improved creativity and innovation
Lorem ipsum
Generative Artificial Intelligence (Gen AI), through the analysis of multiple data, can present
alternatives, offering inspiration to boost creativity and help increase the pace of development of new
products or services and bring them to market faster
Source: Deloitte analysis
Copyright 2023Deloitte Business Solutions S.A. 8
Generative Artificial Intelligence (Gen AI) | Critical success factors for adoption
The critical success factors for the integration of Gen AI relate both to how the relevant systems are developed and operated, and how
they are used by the human resources involved
The architecture of the systems that support Gen AI The training of human resources in Generative AI and, by
initiatives is crucial to the outcome and must be extension, in the development and use of these models is
chosen with great care so that the algorithms, a prerequisite for the proper integration of this technology
models and computing infrastructure are in business operations. This training should be given to all
appropriate and contribute to the efficient staff and should be followed by assessments of staff
operation of the whole system. readiness
In the context of ensuring all of the above, it is particularly
The datasets that Gen AI relies on to create content useful for the appropriate preparation of each company,
must be of high quality, multiple, extensive and up-to- through horizontal actions, such as: conducting a relevant
Formulating a well-defined Gen AI policy based on
date, as they are critical to the quality of the feasibility study per use case and corresponding prioritization
the principles of fairness, transparency,
“products” generated, while it is important to ensure of solutions of strategic importance, development of "proofs
accountability and data protection and subsequent
the absence of "biases" of concept" for the technical testing of Gen AI solutions and
compliance with all necessary legislative
gradual adaptation of the models adopted to larger data sets
frameworks should be a priority for businesses.
(scaling).
Source: Deloitte analysis
Copyright 2023Deloitte Business Solutions S.A. 9
5 Main Categories of Gen AI Use Cases
Copyright 2023Deloitte Business Solutions S.A. 10
5 main categories of use cases | Overview
Gen AI can contribute significantly to the development of 5 main categories of use cases that are of major importance for improving the
efficiency of many business processes
Public-facing services
More direct / more effective interaction with target audiences | more personalized user experience by providing customized responses
Content generation
Creating original texts, images, products, etc. Allows businesses to create content quickly and efficiently
Code management
Speed up and improve the code development process, saving time and eliminating human error
Knowledge assistance
Automation of processes for the capture and maintenance of knowledge. Easier navigation of users through the enterprise's
data and knowledge repositories
Extracting insights from unstructured data
Exploitation of information from unstructured data, achieving optimal understanding and utilization of the information containedtherein (e.g.
image and sound analysis)
Source: Deloitte analysis
Copyright 2023Deloitte Business Solutions S.A. 11
5 main categories of use cases | Benefits
Gen AI entails a number of benefits stemming from the 5 main categories of use cases that Gen AI supports
Extracting information
Public-facing services Content generation Code management Knowledge assistance
from unstructured data
Gen AI's processing of
Gen AI-powered analysis of data Gen AI can be used to create Gen AI as a code management Gen AI helps draw conclusions
unstructured data allows
Better decision from customer interactions can content with a more targeted tool can be leveraged when from "complex" information,
companies to make decisions
making contribute to better commercial focus, which helps better creating multidimensional helping to analyze data for
based on more complete
decision making decision making scenarios for decision making better decision making
information bases
Gen AI interacts with its users
Gen AI can contribute to the Gen AI allows the creation of Gen AI helps identify bugs in the Gen AI can analyze unstructured
Improved through natural language
delivery of personalized services original / customized content, code and provides suggestions data such as customer reviews
customer dialogues and can accurately
and therefore an improved adapted to the needs of each for fixes, contributing to a high and extract valuable insights to
experience identify the requested
customer experience client quality "final" output understand their preferences
information
Gen AI can help increase Gen AI provides insights and Gen AI makes knowledge
With the help of Gen AI more Analyses from unstructured data
productivity through faster supports the creative software management easier and faster,
Higher customer requests can be can reveal ways to improve
content development, enabling development process, leading to helping to improve the
productivity supported/processed end-to- business processes, helping to
more customers to be served at higher productivity of ICT productivity of a company's
end increase productivity
the same time specialists employees
The use of Gen AI for public- Gen AI enables businesses to Gen AI helps programmers Gen AI has the ability to analyze Gen AI enables automatic
facing service helps to automate create content quickly and reduce the time they spend on large amounts of information understanding and organization
Cost savings
tasks and therefore, to save efficiently, with less human certain activities such as code and synthesize it automatically of unstructured data, saving
costs intervention correction and very quickly time and human resources
Gen AI can analyze the profile of Gen AI can support the Gen AI helps draw conclusions Gen AI can discover new trends
Improved Content creation helps to quickly
the recipients and suggest more exploration of different, from complex information, from unstructured data,
creativity / generate a wide range of ideas -
creative ways of approaching / innovative approaches to code helping to create innovative supporting the creation of new
innovation vital for innovation
communicating development ideas and solutions innovative products
Copyright 2023Deloitte Business Solutions S.A. 12
5 main categories of use cases | Template for the Analysis
For a better understanding and deeper insight into the main categories of Gen AI use cases, for each of them an analysis is carried out in
6 dimensions: trends, type of extracted original content, points of differentiation from corresponding traditional AI solutions, potential
benefits, main sectors of the economy for application and indicative examples of use cases
For each of the five main categories of use cases, the
following pages provide an overview of the following:
Public-facing services
the trend towards the use of Gen AI
the possible forms of original content that can be
extracted (e.g. image, sound, code, etc.)
Content generation the main points of differentiation from corresponding
traditional AI solutions
the potential benefits that can be achieved
indicative examples of the most important sectors that
Code management are expected to be most applicable
Finally, for each use case category,
examples of use cases are provided
Knowledge assistance
Extracting insights from unstructured data
Copyright 2023Deloitte Business Solutions S.A. 13
5 main categories of use cases | "Public-facing services" - overview
A typical technological solution in the context of public service is chatbots, which can now use generative artificial intelligence to answer questions of the public,
Trend towards the use of solve problems and provide product and/or service recommendations. Gen AI solutions will bring about a significant evolution in public-facing services, as they
Gen AI have the potential for flexibility, offering solutions tailored to the audience they serve (customer experience personalization), even using customer interactions
to provide more comprehensive solutions.
Code code
Elements of "import" Elements of "export"
Inputs / Prompts Output generated
Public-facing Public-facing
services services
Difference with other Today's chatbots have limited service capabilities, as they are based on traditional artificial intelligence (AI) systems andtherefore on predefined dialogues. AI is
technologies and used in public services mainly for automating tasks. The more sophisticated Gen AI can analyze data from customer interactions to suggest solutions, making it
traditional artificial easier for employees in customer support positions to perform their tasks. The high value of such solutions lies in their capabilities to respond to and service a
intelligence high volume of transactions, at high rates and by eliminating waiting times.
Benefits of Gen AI as a public-facing service tool Application to sectors of the economy
Did you know that...
Consumer goods, Retail trade Public Administration
Increase in customer Availability at all Lower call Strengthening Technology, 85%of executives say that
satisfaction hours, in real abandonment personalized Energy
Telecommunications Generative AI will interact
(CSAT Score) time rate service
directly with customers in
Financial Services Health
the next two years without
any human intervention
(Source: IBM)
Education Media
Reduction of Faster response Service in multiple Scalability
operating costs times languages
Copyright 2023Deloitte Business Solutions S.A. 14
5 main categories of use cases | "Public-facing services" - examples of use cases
TypLiocraelm G ipesnum AI use cases for public service in the Public Administration and Consumer Goods Industry are the Digital Public Servant and
Customer Service on demand, respectively
Digital Public Servant Customer service "on demand"
Challenges Challenges
Public administration internationally -including in Greece -is significantly burdened by Many companies operating in the consumer goods industry have already integrated
bureaucracy and the large volume of documents stored in a variety of formats, which certain Artificial Intelligence (AI) capabilities into their systems in order to provide
makes it difficult to quickly access available information. As a result, the quality of automated and quick answers to their customers, should they seek information or
service often falls short of expectations, creating a climate of mistrust among citizens support about a product or service. Such automation, however, has a limited ability to
regarding the functionality and efficiency of public administration bodies. interpret customer questions and respond with absolute efficiency and accuracy.
The "answer" of Gen AI The "answer" of Gen AI
The Digital Public Officer(with the recent example of mAIgov.gr) can provide the
necessary interface between citizens and the services of the Public Administration, An interactive Gen AI "assistant" can foster a new climate of communication and
through the creation of an interaction system that can respond quickly and with high interaction with customers, as it can create personalized conversations during after-sales
quality to requests. support by providing immediate responses, offering relevant solutions and managing
The Digital Public Servant can rapidly identify and summarize information from multiple complaints. As customers can get faster responses to their questions through Gen AI,
sources on a multitude of issues in order to form appropriate responses to the queries of businesses are able to free up human resources to focus on more complex service issues.
requesters, restoring confidence in Public Administration.
Code Critical success factors Code Critical success factors
Ensuring the provision of accurate information/answers Ensuring the provision of accurate and personalized
advice or guidance
Continuous updating and updating of system Enhancing transparency regarding the functionalities
Public-facing information Public-facing of the model
services Services
A priori identification of customer expectations of the
Protecting sensitive data from cyber-attacks
business, for the best possible system response
3D 3D
Copyright 2023Deloitte Business Solutions S.A. 15
5 main categories of use cases | "Content generation" - overview
Existing AI solutions have the potential to focus on data categorization/recognition to support content development processes. The new achievement of Gen AI
Trend towards the use of
solutions is in the direct development of original content, thus enhancing creativity, the development of new ideas, and moreefficient focus and customization
Gen AI
to customer needs.
Code Code
Elements of "import" Elements of "export"
Inputs / Prompts Content Output generated Content
generation generation
Difference with other Gen AI has the ability to create new versions of data in a variety of formats, not just text. This makes it useful for creating marketing materials, original artwork,
technologies and developing video games with dynamic and evolving content, and even creating synthetic data to train other Gen AI models, especially in scenarios where collecting
traditional artificial real data may be difficult or impractical. In addition, by analyzing existing market trends, consumer preferences and historicaldata, Gen AI models can propose
innovative insights that align with current market requirements in order to create new outputs.
intelligence
Benefits of Gen AI as a content generation tool Application to sectors of the economy
Did you know that...
Consumer goods, Retail trade Public Administration
The45% of employees in
Increased Improved user Trend analysis & Efficiency of available
Energy Technology, marketing departments
productivity experience research extraction resources
Telecommunications spend more than 50% of
the time within one
Financial Services Health
working week, for the
creation of content (Source:
Creating original Saving time Enhancing Compliance with Education Media Capterra’s 2022 AI Marketing
content accessibility regulations Survey)
Copyright 2023Deloitte Business Solutions S.A. 16
5 main categories of use cases | "Content generation" - examples of use cases
GenLo rAeIm t eipcshumnology can be applied to many industries, contributing significantly to the creation of content and products that respond to the
needs of each business customer/user
Marketing content assistant Product design assistant
Challenges Challenges
Businesses face a number of challenges when it comes to managing and optimizing
Product development is a time-consuming and demanding process for businesses. The
marketing content. With a large number of websites for their product portfolios,
need to fully understand customers' needs and preferences can be difficult and often
businesses spend a lot of time and resources creating product descriptions for specific
requires extensive research. Moreover, in an environment where competition is fierce,
customer groups, images, videos, etc. A major issue, too, is achieving consistency in
creating products that stand out and offer something unique can be challenging. In
descriptions, iconography, ads and other media. It is therefore imperative to deliver
addition, market needs can change rapidly, and businesses must adapt quickly to remain
personalized customer experiences quickly and in an automated manner, across a
competitive by creating innovative products.
multitude of ecosystems and touchpoints.
The "answer" of Gen AI
The "answer" of Gen AI
Gen AI technology can therefore be used to generate dynamic content (product Gen AI can be applied to a multitude of industries, allowing businesses to innovate and
descriptions, images, videos) based on user data. This dynamic content can be used to offer products that meet modern market needs. Machine learning algorithms can
create personalized ads / experiences and product recommendations, thus helping to analyze large data sets to discover trends, patterns and insights that can help create
increase business revenue / sales, but also to enhance customer / user engagement. products that meet consumer needs. The use of machine learning algorithms,
Creating targeted content for specific user segments also helps save time and costs. therefore, can help to optimize internal production processes, reduce costs or improve
efficiency.
Critical success factors
Critical success factors
Code
Code
Ensuring accuracy and relevance of content produced
Design innovative products that can be manufactured
and comply with the regulatory framework
Ensuring diversity and representativeness to avoid bias in
Content the content produced Content
generation generation
Protecting intellectual property rights when using Gen
Establish strong ethical guidelines regarding the use of
AI in the creative process
3D sensitive data 3D
Copyright 2023Deloitte Business Solutions S.A. 17
5 main categories of use cases | "Code management" - overview
Generative AI can be used in many aspects of software engineering such as managing, developing, completing, debugging, documenting and restructuring of
Trend towards leveraging
code. Images, sounds, texts and code can be fed into the Gen AI model from which, depending on the user's choice, a new form of code is produced in
Gen AI
programming languages such as Python, JavaScript, Java, Verilog, C, C++, TypeScript and more.
Code Code
Elements of "import" Elements of "export"
Inputs / Prompts Code Output generated Code
Management Management
Difference with other A key difference in relation to other technologies such as traditional artificial intelligence (AI) is the possibility of developing new code after the descriptive capture
technologies and of the request. Artificial intelligence is mainly based on "deterministic systems" ("if-then" conditions), which use a set of rules that lead to predetermined results
traditional artificial and are now suitable for generating code for repetitive tasks (e.g. GitHub Copilot, Amazon CodeWhisperer, etc). Therefore, the use of generative artificial
intelligence is suitable for applications where the main prompt is descriptions in natural language.
intelligence
Benefits of Gen AI as a Code Management tool Application to sectors of the economy
Did you know that…
Public Administration
Consumer Goods, Retail
Increased Error detection and Apply code Efficiency of time and Technology, Developers spend ~25-
Energy
productivity prevention standards resources Telecommunications 50% of their time per year
debugging. Gen AI greatly
Financial services Health improves this issue,
creating time for more
creative tasks
Education Media
Flexible decision Documentation Cost Data reliability (Source: Undo.io)
making management
Copyright 2023Deloitte Business Solutions S.A. 18
5 main categories of use cases | "Code management" - use case examples
GenLo rAeIm c iapnsu smignificantly support the overall process of code development by performing functions such as pattern synthesis, testing, and
documentation
Code Support for Developers What do Gen AI applications bring to code development?
Challenges
Through Gen AI, code development as a whole is done without the need for
Code development is a complex process, involving a number of challenges. It requires human intervention, as was required until now. Developer teams provide the
specialized staff and its lack of consistency or inadequacy leads to slow applications and system with descriptions or specifications, with Gen AI developing or suggesting
increased resource usage. The large amount of information and functions present code that meets the requested functionality. In this way, human resources are
significant problems for code review and testing to identify and correct errors. Additional focused on processes to achieve maximum quality and reliability of the models,
code maintenance issues are related to compatibility with other systems, lack of security, while minimizing the possibility of human error.
and lack of documentation.
The "answer" of Gen AI
The testing process, due to the large amount of data that requires testing, has
until now required significant human effort. Gen AI can automatically detect bugs
Using Gen AI to support code serves to offload ICT manpower and focus them on more or predict where they might occur, discover opportunities for optimization, and
complex and higher-value digital transformation tasks. suggest code restructuring points to upgrade its quality, allowing developers to
By using Generative Artificial Intelligence (Gen AI) faster completion of repetitive tasks is engage in the strategic decision-making and solutions they want to "build".
achieved, such as: developing, maintaining, documenting and checking code, adapting
functional code to different environments, data transformation, abstractions, etc.
Code Critical success factors Code documentation is a defining process of the overall process, which until now
has been mostly done manually. Gen AI can generate, without human
Ensuring accuracy and lack of errors intervention, comments / explanations / documentation summaries for specific
functions or even entire user manuals in order to make the code understandable
to others. Also, this technology has the ability to translate code into other
Ensure transparency and explainability of
Code documentation variables and comments programming languages, if there is a need to change or adopt the code in
management another environment.
Protection from cyber security risks
3D
Copyright 2023Deloitte Business Solutions S.A. 19
5 main categories of use cases | "Knowledge assistance" - overview
Gen AI models have access to a range of both structured and unstructured data which they can equally well read, understand, synthesize and extract useful
Trend towards leveraging
information to the u |
163 | deloitte | in-deloitte-pov-safeguarding-gen-ai -with-cybersecurity-measures-noexp.pdf | Safeguarding Generative
Artificial Intelligence (AI) with
cybersecurity measures
Risk insights and building blocks for secure
Generative AI solutions
2024
Safeguarding Generative Artificial Intelligence (AI) with cybersecurity measures
Table of contents
Introduction 02
Evolution of Generative AI 03
What constitutes Generative AI 04
How does Generative AI work? 05
Six categories of Cyber risks with Generative AI 06
Illustrative Cyber risks of Generative AI 07
Industry-wise use cases of Generative AI,
Cyber Risks, and Controls 08
Building blocks for secure Generative AI solutions 10
Way forward 13
Conclusion 14
Connect with us 15
01
Safeguarding Generative Artificial Intelligence (AI) with cybersecurity measures
Introduction
2022 was a watershed in the history of Artificial Intelligence As a result, it will lead to the following outcomes:
(AI). While the journey started in 1932, it has gradually become
1. Improving productivity
all-pervasive − first with digital assistants from various
2. Increased customer satisfactions
multinational technology conglomerates and now with the
3. Propelling Research and Development (R&D)
release of Generative AI. This phase is incredible because
it affects most businesses and personal interactions. While 4. Creating new revenue streams and business models
the world celebrates the coming of age of Generative AI, one
needs to make certain considerations to ensure that the scale While organisations and businesses adopt the Generative
and impact are progressive for individuals, organisations, and AI, Cybersecurity is paramount. Necessary controls should
society alike. be implemented to ensure that investments deliver the right
business results to organisations while maintaining individuals’
Generative AI has many positive implications and could have privacy and confidentiality. Additionally, a lack of adoption
the ability to transform the way we do business. Some of the of the Security by Design (SbD), Privacy by Design (PbD), and
most important aspects include the following: Ethical by Design (EbD) concepts could lead to exposure and
risks to data being used and training of models adopted.
Finally, security technology needs to keep pace with the
development of Generative AI.
Intelligent Information Technology (IT) – This point of view (POV) provides insights into certain
Transforming how IT is structured, how cybersecurity considerations for Generative AI and the
software development is done, and how IT is necessary controls organisations should consider while
enhanced and supported. building these systems.
Intelligent products – Enhancing sensor-
infused products using Generative AI, which
can have huge implications across several
industries.
Intelligent operations – Remodeling
operations with a greater emphasis on
Generative AI-derived inputs that could help
make operations nimble and agile.
02
Safeguarding Generative Artificial Intelligence (AI) with cybersecurity measures
Evolution of Generative AI
Most of the time, Generative AI is considered relatively new. Contrary to the
belief, Generative AI is deep-rooted in history and innovation.
Georges Artsrouni invented a machine
reportedly called the “mechanical brain”
1932
to translate languages on a mechanical Harold Cohen, a painter and professor,
computer encoded onto punch cards.1 collaborated with a programme called
1973 AARON to produce art autonomously.
Yann Lecun, Yoshua Bengio, and Patrick Paintings are done in Cohen’s style.
Haffner demonstrated how Convolutional
1989
Neural Networks (CNNs) can recognise
images. Researchers from the University of
Montreal published “a neural probabilistic
2000 language model”, which suggests a method
to model language using feed-forward
Data scientist Fei- Fei Li set up the ImageNet neural networks.
database that laid the foundation for visual 2006
object recognition.
Apple released Siri, a voice-powered
personal assistant that can generate
2011
responses and take actions in response to
Ian J. Goodfellow and colleagues published voice requests.
the first paper on Generative Adversarial
2014
Networks (GANs) that can determine if an
image is real or fake. Google researchers developed the concept
of transformers in the seminal paper
“Attention is all you need.” The paper
Google researchers implemented 2017
inspired subsequent research into tools that
transformers into BERT, trained on over 3.3
could automatically parse unlabeled text
billion words. It can automatically learn the
into Large Language Models (LLMs).
relationship between words in sentences,
2018
paragraphs, and even books, and predict
Open AI released ChatGPT in November
the meaning of text. Google DeepMind
to provide a chat-based interface to its
researchers developed AlphaFold to predict
2022 GPT 3.5 LLM. It attracted more than 100
protein structures that laid the foundation of
million users in two months, representing
Generative AI.
the fastest-ever consumer adoption of a
service.
Adobe launched Firefly, a family of Generative AI Google released Bard, a Generative AI
models tailor-made for creative professionals, 2023 chatbot built on 137 billion parameters and
with built-in guardrails for safety and copyright embeds Generative AI capabilities into its
standards. workshop products.
1 What is Generative AI? Everything you need to know
03
Safeguarding Generative Artificial Intelligence (AI) with cybersecurity measures
What constitutes Generative AI
Generative AI learns from various inputs to generate its outputs.2
Data can vary from text and photos to videos, audio, and codes. Using these wide varieties of datasets, Generative AI creates novel
outputs. Generative AI models use technological elements, such as LLMs, diffusion networks, GANs, transformers and Variational
Auto Encoder (VAEs), and other novel techniques to identify patterns and structures within existing data to generate new content.
Text
Essays, speech, creating
questions asking
Large Language
Models (LLMs) Photos
New enhancements, such
as photo edits
GANs,
Video
Diffusion
transformers,
New videos, such as reels
networks
and VAEs
Audio
Music, clips, audio, etc.
Novel
techniques
Code
Self-learning models,
prompt engineering, etc.
2 Generative AI – What is it and How Does it Work? (nvidia.com)
04
Safeguarding Generative Artificial Intelligence (AI) with cybersecurity measures
How does Generative AI work?
A simplified view
Cloud and data Generative AI
AI infrastructure Generative AI models
platforms applications
Computer
Data Prediction Models Output
power
Why do these
Powering our Training on the What are Applications − What
applications seem
journey to tomorrow world’s knowledge foundation models? we see
so human?
The scale of compute Foundation models Similar to traditional OpenAI’s GPT-4 and Generative AI
capacity required are trained on AI, foundation NVIDIA’s Megatron applications
to train and process petabytes worth of models predict are two examples of generate content
foundation models global data to shape outputs based on foundation models, across various
necessitates the use understanding, tone, inferences on the specifically Large modalities (e.g., text,
of leading GPUs (e.g., and behaviour while inputs it receives. Language Models image, video, and
A100 NVIDIA) and considering human However, through (LLMs) that use deep audio)
TPUs (e.g., Google communication fine-tuning, prompt learning to process
TPU v4) on scalable styles. engineering, and huge amounts of
infrastructure. adversarial training, data. These models
these models form “memories”
generate outputs on the input
based on their datasets through
understanding tokenisation,
of human thereby shaping
communication. models’ parameters.
Common foundation
model architectures,
such as transformer
and diffusion, drive
modalities for each
model.
05
Safeguarding Generative Artificial Intelligence (AI) with cybersecurity measures
Six categories of Cyber risks with
Generative AI
While there are many Cyber risks related to Generative AI, we have tried to
group them into six categories.
Generative AI model-based risks
Generative AI models are currently being developed by a few organisations. Most others end up using
those models. If not used wisely or ethically, these models can cause potential loss of confidential/
sensitive/copyright information or other intellectual property infringement.
Infrastructure risks
These include risks related to the infrastructure provided to support Generative AI models, applications,
and data. Traditional infrastructure Cyber risks, such as using components with known vulnerabilities,
insecure services, ransomware attacks, and DDoS attacks, are a few examples.
Data risk
While the data discovery and classification themselves have inherent risks, if the correct processes are
not followed, the right controls including those for privacy and confidentiality may not be present.
People risk
People risk is related to ethical use and bias aspects of Generative AI. It is equally important to ensure
that Generative AI systems do not cause harm to end users.
Application/algorithmic risk
These could include inherent algorithmic and coding risks in the applications developed on the
mentioned-above models.
Training and testing risk
These are related to lack of capability to create effective training and testing processes for Generative AI.
06
Safeguarding Generative Artificial Intelligence (AI) with cybersecurity measures
Illustrative Cyber risks of Generative AI
While there may be myriad risks from a cyber perspective, some key risks you
should be aware of while using Generative AI are provided below.
Membership inference Prompt injection
Inferring the presence of specific data points in Prompt Injection manipulates a Generative AI
the training set by querying the AI model and through crafty inputs, causing unintended actions
compromising data privacy. by the Large Language Models (LLMs). Direct
injections overwrite system prompts, while indirect
Insecure output Handling ones manipulate inputs from external sources.
This vulnerability occurs when a Generative AI
output is accepted without scrutiny, exposing Excessive agency
backend systems. Misuse may lead to severe Generative AI-based systems may undertake
consequences, such as XSS, CSRF, SSRF, privilege actions leading to unintended consequences.
escalation, or remote code execution. The issue arises from excessive functionality,
permissions, or autonomy granted to
Model denial of service LLMs-based systems.
Attackers cause resource-heavy operations on
Generative AI, leading to service degradation or Overreliance on Generated AI
high costs. The vulnerability is magnified due to Systems or people overly depending on Generative
the resource-intensive nature of LLMs and the AI without oversight may face misinformation,
unpredictability of user inputs. miscommunication, legal issues, and security
vulnerabilities due to incorrect or inappropriate
Supply chain vulnerabilities content generated by Generative AI.
Generative AI application lifecycle can be
compromised by vulnerable components or Model theft
services, leading to security attacks. Using third- Model Theft could involve unauthorised access,
party datasets, pre-trained models, and plug-ins copying, or exfiltration of proprietary Generative
can increase vulnerabilities. AI models. The impact includes economic losses,
compromised competitive advantage, and potential
Sensitive information disclosure access to sensitive information.
Generative AI may inadvertently reveal confidential
data in its responses, leading to unauthorised data Training data poisoning
access, privacy violations, and security breaches. This occurs when Generative AI training data is
Implementing data sanitisation and strict user tampered, introducing vulnerabilities or biases
policies to mitigate these risks is crucial. that compromise security, effectiveness, or ethical
behaviour. Sources include Common Crawl,
Insecure plug-in design WebText, OpenWebText, and books.3
Generative AI plug-ins can have insecure inputs and
insufficient access control. This lack of application Deepfakes
control makes them easier to exploit and can result Deepfake technology4 has advanced to the
in consequences such as remote code execution. point where it can be used in real-time, enabling
fraudsters to replicate someone’s voice, image, and
movements in a call or virtual meeting.
3 https://owasp.org/www-project-top-10-for-large-language-model-applications/assets/PDF/OWASP-Top-10-for-LLMs-2023-v1_0.pdf
4 https://www.latimes.com/business/technology/story/2023-05-11/realtime-ai-deepfakes-how-to-protect-yourself#:~:text=Cybersecurity%20experts%20
say%20deepfake%20technology,a%20call%20or%20virtual%20meeting.
07
Safeguarding Generative Artificial Intelligence (AI) with cybersecurity measures
Industry-wise use cases of Generative
AI, Cyber Risks, and Controls
Generative AI’s potential extends to almost every industry, as it provides a wide range of automation and enhances creative and
data-driven processes. However, with its various use cases, it also introduces Cyber risks.
A few use cases of Generative AI across key industries, along with the possible Cyber risks and mitigation steps, are provided below:
** This is not an exhaustive list.
Industry Use cases of Description of use case Cyber risk Mitigation/controls
Generative AI
Consumer Blog and social Content generation: Misinformation and fake Fact-checking and
media content Generative AI can be used content: Generative AI can verification: Establish
writing to generate a variety of produce false information, partnerships with
content types, including such as fake news articles, reputable fact-checking
text, photos, and videos. It photos, and videos. This organisations to verify the
can, for example, be used raises the possibility of accuracy of the content
to generate personalised distributing inaccurate generated. Develop
product descriptions, blog information, altering public automated systems
entries, or even fictional opinion, or carrying out to cross-reference the
stories. social engineering attacks. generated information
with reliable sources,
to help identify and flag
potential misinformation.
Government and Social services Generative AI can Bias and discrimination: Transparent decision-
Public Services and welfare assist in personalised Generative AI models making: Enhance
service delivery in trained on biased or the transparency of
healthcare, social incomplete data may Generative AI systems
welfare, or education. It inadvertently perpetuate by explaining decision-
can analyse individual biases or discrimination making processes. Use
data to recommend in social service delivery. techniques such as
suitable programmes, This could lead to unfair or Explainable AI (XAI) to
interventions, or support discriminatory outcomes, make the generated
services based on specific disadvantaging certain outcomes more
needs. individuals or groups based understandable to
on their demographic citizens, fostering trust
or socio-economic and accountability.
characteristics.
Energy, Resources, Energy Generative AI models Manipulation of Implement robust data
and Industrials demand can analyse historical forecasting data: security measures,
forecasting energy consumption Adversaries may attempt to including encryption,
data, weather patterns, manipulate or tamper with access controls, and
economic indicators, and the data used for energy secure storage to protect
other relevant factors, to demand forecasting. By the confidentiality and
help forecast future energy injecting false or misleading integrity of the data
demand. Accurate demand information into the dataset, used for forecasting.
forecasting helps utilities they could manipulate the Use anomaly detection
and energy providers forecast demand, potentially and outlier analysis
optimise resource allocation, leading to inefficient techniques to identify and
plan for peak demand resource allocation, financial mitigate potential data
periods, and enhance loss, or disruptions in energy manipulation attempts.
energy distribution.5 supply.
5 How generative AI can boost productivity in enterprises and industries
08
Safeguarding Generative Artificial Intelligence (AI) with cybersecurity measures
Industry Use cases of Description of use case Cyber risk Mitigation/controls
Generative AI
Financial Services Financial By learning from historical Model poisoning Regular model auditing
forecasting financial data, Generative threat actors may and monitoring of
AI models can capture manipulate the training adversarial activities are
complex patterns and process of Generative a few mitigating controls
relationships in the data, AI models by injecting to help combat the
enabling them to make malicious data or disturbing “poisoning of the data”.
predictive analytics about the training data to
future trends, asset prices, undermine the accuracy of
and economic indicators.6 forecasts.
Technology, User interface Generative AI can help SSRF vulnerabilities Rigorous input validation
Media, and design in User Interface (UI) allow the exploitation of and regular audit
Telecommunications design by providing Generative AI models by network/ application
automated suggestions performing unintended security.
for layouts, colour requests or accessing
schemes, and component restricted resources, such
placement based on as Application Programming
user requirements or Interfaces (APIs) or internal
predefined templates. This services that may lead to
can help developers in wrong designs.
rapid prototyping.7
Life Sciences and Drug discovery Generative AI can be Intellectual property Intellectual risks in
Health Care sector used to streamline drug theft: Generative AI models Generative AI can be
discovery and development are trained on extensive mitigated using multiple
by identifying potential datasets, which may include strategies together, such
drug candidates and proprietary or patented as encryption,
testing their effectiveness information. Unauthorised secure data hosting,
before moving them for access to these models or access controls,
other trials.8 their outputs could result water-marking, digital
in intellectual property signatures, or content
theft, where competitors fingerprinting.
could access confidential
drug discovery processes,
formulas, or compounds.
6 Generative AI in the Banking and Finance Industry
7 Generative AI: The Next Frontier in Telecom Innovation
8 Generative AI Healthcare Industry: Benefits, Challenges, Potentials
09
Safeguarding Generative Artificial Intelligence (AI) with cybersecurity measures
Building blocks for secure Generative
AI solutions
Business values layer
Growth and Operational Risk
Compliance
innovation efficiency management
Governance layer
Strategy and Risk and compliance Training and Policies and
roadmap management awareness standards
Organisation Integration with
and operating business and IT
model Processes
Five pillars of responsible Generative AI adoption for a secure AI ecosystem
Adoption Maintenance Scaling Customisation Decommissioning
Policy and principles Continuous evaluation Planning Impact assessment Decommissioning
Develop policies and and monitoring Create a scalability plan Conduct impact policies
principles focusing Regularly assess fairness, that outlines the steps, assessments to Create clear protocols
on transparency, bias, accuracy, and benchmarks, and risk evaluate the potential and procedures for
accountability, and impact on users. mitigation measures impacts, risks, and shutting down or
safety for developing/ to scale Generative AI considerations decommissioning
Infrastructure
buying Generative AI systems. associated with Generative AI systems.
monitoring and
technologies. customising Generative Ensure to address data
security Model robustness
AI systems. retention issues and
Governance Set up processes Verify the robustness
privacy problems.
Establish a governance for infrastructure and generalisation Transparency
programme to monitoring, to help capabilities of Maintain transparency Safeguard against
manage Generative AI address vulnerabilities Generative AI models while customising the malicious use
technology. and incorporate the most during scaling. Generative AI systems, of Generative AI
recent developments in to enable users to trust technologies, such
Risk assessment Risk assessment
Generative AI security. the customised system. as models, data,
Conduct risk Conduct a risk
and associated
assessments to identify Awareness assessment to identify
infrastructure after
potentials risks, safety Encourage the potential risks, such
termination.
and bias issues. responsible use and as data breaches,
discourage malicious or or unintended
Update your policy
unethical applications of consequences, as a
regularly, with
Generative AI. result of scaling.
mitigation controls.
Risk control layer
Confidentiality Integrity Availability Authenticity Authorisation Privacy
Regulatory compliance
10
Safeguarding Generative Artificial Intelligence (AI) with cybersecurity measures
The Generative AI risk management framework rests upon a solid foundation comprising four key layers:
Business values layer: It evaluates potential risks and benefits from AI implementation, aligning
projects with overarching strategic objectives, financial robustness, reputation management, and
competitive edge.
Governance layer: Governance of Generative AI involves managing and overseeing its application
across people, processes, and technology to ensure its responsible, secure, and ethical use.
Effective governance of Generative AI requires a multidisciplinary approach involving collaboration
between different teams and stakeholders. It should be an ongoing process that evolves with
advancements in Generative AI technology and dynamic in nature to keep up with societal norms
and regulation changes.
Five pillars: It encompass adoption, maintenance, scaling, customisation, and decommissioning,
offering a comprehensive roadmap for navigating the complete Generative AI lifecycle and
proactively identifying and mitigating risks at each stage.
Risk control layer: The culminating risk control layer bolsters the framework with its paramount
role in ensuring that AI technologies harmonise with data security, privacy imperatives, and
regulatory compliance, extending from established principles like the CIA triad to encompass the
full spectrum of privacy considerations and adherence to pertinent regulations.
11
Safeguarding Generative Artificial Intelligence (AI) with cybersecurity measures
12
Safeguarding Generative Artificial Intelligence (AI) with cybersecurity measures
Way forward
Some key cyber questions to help you assess your organisation’s readiness
for a secure, private and ethical use of Generative AI. Questions we want to
leave you with:
01
What are your key business use cases to make this programme a success?
02
Do you have the right cyber investments (tools, technologies, processes, and skillset) in your strategic roadmap?
03
Is there a plan to ensure that your Generative AI tools do not threaten your organisation’s end users and
customers?
04
How do you ensure your Generative AI tools are not using your sensitive data for training?
05
Do you have a process in place to ensure sensitive data is not used without the right controls?
06
Do you have policies that ensure the security of your Generative AI models?
07
Do you have a Security Operations Center (SOC) to monitor threats in your Generative AI landscape?
08
How does your business ensure that only authorised users can access Generative AI tools, models,
infrastructure, and data?
09
Will your Generative AI be used or built by a third party, and do you need to re-assess and re-look at your current
third-party risk programmes?
10
Are you prepared to use Generative AI with your organisation’s privacy and confidential controls (including
consent mechanism and data sanity)?
13
Safeguarding Generative Artificial Intelligence (AI) with cybersecurity measures
Conclusion
Generative AI has immense possibilities regarding content, These risks cut across the ecosystem’s foundation, bringing
which can help reduce the effort and bring out efficiencies in the in human capital, technology, and industry processes. While
system. Generative AI has applications across the ecosystem, the potential for Generative AI is undeniable, what will make
affecting individuals, organisations, and society alike. it a transformational force is balancing the risk and bringing
in the right controls for the global scale of adoption.
While we celebrate this quantum leap in technological
advancement, similar to any technology, the cybersecurity Generative AI will create immense growth opportunities in
perspective that needs to be considered would enhance key areas, such as intelligent IT, products, and operations.
the scale and application of Generative AI. The risks largely The next decade will be when AI will become mainstream and
lie in key areas, i.e. Generative AI models, applications, further enhance human potential and growth.
infrastructure, people, data, and the training and testing
methodologies.
14
Safeguarding Generative Artificial Intelligence (AI) with cybersecurity measures
Connect with us
Sathish Gopalaiah Deepa Seshadri
President, T&T, Deloitte India Partner, Deloitte India
[email protected] [email protected]
Gaurav Shukla Praveen Sasidharan
Partner, Deloitte India Partner, Deloitte India
[email protected] [email protected]
Vikram Venkateswaran
Partner, Deloitte India
[email protected]
Contributors
David George
Rajat Kothari
Vivekchandran N V
Titas Nath
15
Safeguarding Generative Artificial Intelligence (AI) with cybersecurity measures
16
Deloitte refers to one or more of Deloitte Touche Tohmatsu Limited (“DTTL”), its global
network of member firms, and their related entities (collectively, the “Deloitte organization”).
DTTL (also referred to as “Deloitte Global”) and each of its member firms and related entities
are legally separate and independent entities, which cannot obligate or bind each other in
respect of third parties. DTTL and each DTTL member firm and related entity is liable only for
its own acts and omissions, and not those of each other. DTTL does not provide services to
clients. Please see www.deloitte.com/about to learn more.
Deloitte Asia Pacific Limited is a company limited by guarantee and a member firm of DTTL.
Members of Deloitte Asia Pacific Limited and their related entities, each of which is a separate
and independent legal entity, provide services from more than 100 cities across the region,
including Auckland, Bangkok, Beijing, Bengaluru, Hanoi, Hong Kong, Jakarta, Kuala Lumpur,
Manila, Melbourne, Mumbai, New Delhi, Osaka, Seoul, Shanghai, Singapore, Sydney, Taipei
and Tokyo.
This communication contains general information only, and none of DTTL, its global network
of member firms or their related entities is, by means of this communication, rendering
professional advice or services. Before making any decision or taking any action that may
affect your finances or your business, you should consult a qualified professional adviser.
No representations, warranties or undertakings (express or implied) are given as to the
accuracy or completeness of the information in this communication, and none of DTTL, its
member firms, related entities, employees or agents shall be liable or responsible for any loss
or damage whatsoever arising directly or indirectly in connection with any person relying on
this communication.
© 2024 Deloitte Touche Tohmatsu India LLP. Member of Deloitte Touche Tohmatsu Limited |
164 | deloitte | in-strategy-a-gcc-leaders-guide-for-driving-gen-ai-adoption-single-page-web-version-v5-noexp.pdf | A GCC leader’s guide
for driving Generative
AI adoption
December 2024
A GCC leader’s guide for driving Generative AI adoption
ii
A GCC leader’s guide for driving Generative AI adoption
Table of contents
1. Foreword 02
2. The role of Global Capability Centres (GCCs) in harnessing
GenAI capabilities 04
2.1 Accelerating GenAI adoption 05
3. How can GCCs gauge their GenAI adoption readiness: A strategic
assessment framework 09
4. How can GCCs identify, qualify and prioritise use cases? 13
4.1 Identification of use cases 13
4.2 Use case qualification 16
4.3 Prioritisation of use cases 19
5. How can GCCs implement GenAI use cases? 22
5.1 Ensuring readiness for successful deployment 24
5.2 Achieving success in scaling GenAI solutions 26
6. The emergence of AI agents and multi-agents 32
7. Key considerations for GCC leaders 34
7.1 Solution trustworthiness 34
7.2 Other considerations 36
8. Conclusion 37
9. Connect with us 38
01
A GCC leader’s guide for driving Generative AI adoption
Foreword
The transformative power of Generative AI GenAI's full potential while mitigating potential
(GenAI) is undeniable, and its potential to reshape challenges.
industries and business operations is significant.
As the GCC landscape evolves, it is clear that these As Deloitte India’s GCC and AI leaders, we are
centres are not just passive observers but active excited to witness the role GCCs will play in
catalysts in this AI-driven transformation. The shaping the future of GenAI. This thought paper
combination of strategic alignment, technological reflects our commitment to empowering GCCs
prowess, data-driven insights and a culture of with the knowledge they need to initiate and lead
innovation positions GCCs well to adopt GenAI. this transformative journey. We believe that by
embracing GenAI and using the insights within this
This report is a resource for GCC leaders to report, GCCs can unlock value, drive innovation and
assess readiness for GenAI adoption. It provides achieve success in the AI-powered era.
an approach for identifying and prioritising use
cases, a high-level approach to implementation
and metrics to measure success. Addressing key
considerations and risks helps GCCs harness
Rohan Lobo Anjani Kumar
Partner, GCC Leader Partner, Artificial
Technology & Intelligence & Data
Transformation Technology &
Deloitte India Transformation
Deloitte India
02
AA GGCCCC lleeaaddeerr’’ss gguuiiddee ffoorr ddrriivviinngg GGeenneerraattiivvee AAII aaddooppttiioonn
Generative AI is the single most
significant platform transition
in computing history. In the
past 40 years, nothing has
been this big. It is bigger than
PC, it is bigger than mobile, and
it is going to be bigger than the
internet by far.1
- Jensen Huang, CEO, NVIDIA
1. The Practical Impact Of AI For The Masses, Forbes, 28 November 2023
03
A GCC leader’s guide for driving Generative AI adoption
The role of Global Capability Centres
(GCCs) in harnessing GenAI capabilities
GCCs can be at the forefront of driving
enterprise-wide adoption of GenAI. Their Text
strategic role allows them to effectively
Creative
pilot and scale AI initiatives, unlocking new Chatbots Translation
writing
opportunities for their parent organisations.
With their robust technological ecosystems
Image
and unwavering commitment to innovation,
GCCs are poised to lead this transformation 2D and 3D Image Product
and not just participate. Images enhancement simulation
GCCs are strategic partners driving innovation
Code
and digital transformation for their parent
organisations. They use advanced tools, agile Code Code Bug
methodologies and external partnerships to generation compilation fixing
enhance efficiency and customer/employee
experiences. Initiatives such as innovation labs
Audio
and hackathons keep them at the forefront
of technology. While GCCs have long led
Text-to-Speech Music Voice
technology transformation, GenAI presents a
Generator composition assistants
new frontier, offering incremental digitisation and
transformative opportunities for new services and
business models. Video
2D and 3D Video Video
GenAI can be applied across various modalities,
Videos simulation processing
offering unique capabilities for automating and
enhancing business operations. The primary
modalities include text, image, code, audio and
video. The illustration below demonstrates
The ability of GCCs to integrate GenAI
GenAI’s significant potential to transform business
into the fabric of their service delivery
processes and create new opportunities for
will be a game-changer, enabling
innovation at the enterprise scale. GCCs are the
them to redefine processes and
perfect testbed for these capabilities and will play
improve efficiency, establishing them
a key role in the widespread adoption of this new
as leaders in the global enterprise
technology.
ecosystem.
– Yatin Patil, Partner, Leader - Enterprise
Technology and Performance, Deloitte
South Asia
04
A GCC leader’s guide for driving Generative AI adoption
Accelerating GenAI adoption
GCCs possess inherent strengths and capabilities, making them the perfect leaders for their parent
organisation's GenAI adoption initiatives.
Strategic alignment
By proactively collaborating with the parent organisation’s
leadership, GCCs can define a clear GenAI strategy that
complements the broader business objectives. This enables
GCCs to integrate their operations seamlessly into their global
strategy, enhancing the organisation's ability to innovate and
operate efficiently.
Tech-enabled ecosystem
GCCs have a tech-enabled ecosystem that enhances
their ability to harness GenAI. Their robust
infrastructure, extensive network of technology
vendors, access to digital talent, thriving ER&D
community, and mature tech start-up ecosystem create
fertile ground for the rapid adoption and development
of GenAI solutions.
Data access and management
As key data custodians for their parent organisations, GCCs have
extensive experience in managing vast, cross-functional datasets.
Their adoption of a Centers of Excellence (COE) approach ensures
robust data governance and infrastructure. The ability to gather,
clean and maintain high-quality datasets, along with their
proficiency in automation and custom workflows, makes GenAI a
natural extension, enabling insights and innovative operations.
05
A GCC leader’s guide for driving Generative AI adoption
Talent availability
GCCs have expertise in AI/ML, product engineering and
analytics essential for developing, deploying and maintaining
GenAI solutions. This is complemented by business/domain
knowledge and process ownership, enabling them to provide
a well-rounded solution to business problems using GenAI.
The opportunity to work on cutting-edge global projects and
the learning thereof enable GCCs to attract and retain
AI talent necessary to drive innovation.
Culture of experimentation
A culture of experimentation is key for GCCs, enabling
stakeholders to explore and pilot new GenAI applications
across business functions. This approach allows GCCs
to rapidly test, refine and implement AI solutions,
driving meaningful innovation. The parent organisation
also plays a vital role by creating a supportive and
empowering environment that encourages GCCs to
experiment freely, gain insights and apply new learnings.
Scalability and flexibility
Using advanced infrastructure and methodologies, GCCs ensure
GenAI solutions remain scalable and flexible, adapting efficiently
to project demands. Cloud-based platforms allow GCCs to scale
resources as needed. Their experience in Agile and DevOps
practices enables rapid development and iterative improvement
of AI models. Such enablers allow GCCs to deploy modular,
maintainable AI solutions capable of handling varying workloads.
06
AA GGCCCC lleeaaddeerr’’ss gguuiiddee ffoorr ddrriivviinngg GGeenneerraattiivvee AAII aaddooppttiioonn
Several GCCs have successfully adopted and implemented GenAI across various use cases. The following
is a selection of Deloitte case studies that illustrate how these GCCs have applied GenAI effectively:
Context Solution Impact
A British-Dutch multinational consumer goods company
GenAI bot for Employees Implemented • Enhanced employee
enhancing experienced delays voice-enabled AI experience through 24/7
employee and inefficiencies bots integrated support.
experience in resolving payroll with ServiceNow • Achieved a 65 percent
and HR queries, to efficiently handle success rate and lowered
such as taxation and diverse payroll and ticket volumes by 5 percent.
benefits issues. HR-related inquiries. • Reduced dependency on
human intervention through
self-service options.
An American multinational technology company
GenAI-enabled Various customer Automated template • Reduce effort and time in
customer scenarios, such as generation using generating new templates.
communications declined payments GenAI to craft • Improve operational
or subscription personalised messages efficiency.
renewals, based on customer • Facilitate faster response
required tailored interactions. times to customer inquiries.
communications.
Manual drafting was
time-consuming and
inefficient.
A German luxury automotive company
Automated Instructions drafting Developed an AI-based • Reduce effort and time in
generation was manually system to automate generating instruction.
of assembly done, requiring the generation • Improve operational
instructions significant effort of assembly efficiency.
to prepare part- instructions based on • Reduce manual effort for
specific assembly car model BOM data. new part assembly.
instructions.
07
A GCC leader’s guide for driving Generative AI adoption
The nexus of enterprise systems, GenAI and GCCs for
enterprise transformation
Integrating GenAI into enterprise systems, unconventional GenAI experiences, enabling
especially those adopted by GCCs such as SAP, productivity improvements, increased
Oracle and ServiceNow, presents a compelling operations agility and better employee/
opportunity for GCCs to transform enterprise customer experiences for IT service and
operations. Using AI’s capabilities within these operations management, customer service
core systems, GCCs can automate routine tasks, management, HR services, portfolio
extract valuable insights from vast datasets and management, etc.
enable intelligent decision-making. The ability of
GenAI to understand natural language queries Deloitte’s Ascend5 platform for ERP systems
and generate contextually relevant responses can incorporates GenAI capabilities and process
significantly enhance user experiences within their accelerators to support clients' transformation
environments: initiatives. These advancements cater to various
use cases embedded with enterprise systems,
• Per SAP, SAP Joule2 has been integrated
enabling automation across critical functions such
across various SAP applications, including
as autonomous coding, configuration, design,
HR, finance, supply chain, procurement,
testing and project management. This integration
customer experience and into the SAP Business
streamlines operations and accelerates digital
Technology Platform. Joule aims to enhance
transformation for clients, positioning them to
user interaction by providing seamless
achieve greater efficiency and innovation in their
navigation, rapid information retrieval and
ERP-driven processes.
efficient execution of business tasks. It also
offers proactive recommendations and even AI-
assisted code generation.
• Oracle cites its OCI Generative AI Service,3 which
incorporates large language models into its
GCCs are experiencing rapid growth, becoming
modules. It supports use cases, such as writing
mature, efficient and innovative. As they continue
assistance, summarisation, data analysis and
this journey, GenAI presents a unique opportunity
interactive chat, helping businesses automate
that they must seize, lead and own.
and enhance various operations across their
– Deepak Mowdhgalya, Partner, Leader, Finance
systems.
Transformation, Deloitte India
• According to ServiceNow, GenAI has been
integrated into the workflows of its Now
Platform,4 called Now Assist. It provides
2. SAP Joule, SAP
3. Oracle Generative AI Service, Oracle
4. “Put Generative AI to work with Now Assist,” Service Now
5. Deloitte’s Ascend™, Deloitte
08
A GCC leader’s guide for driving Generative AI adoption
How can GCCs gauge their GenAI
adoption readiness: A strategic
assessment framework
As GCCs plan to use GenAI for business transformation, they must assess their readiness to implement
and adopt this technology effectively, focusing on two key dimensions:
Ecosystem enablers: Strategic factors Capabilities: Organisational and technical
that support GCCs’ overall readiness and elements that ensure the GCCs are
alignment with the parent organisation’s equipped to develop, deploy and sustain
objectives, focusing on fostering innovation GenAI solutions.
and ensuring leadership buy-in.
Ecosystem enablers Capabilities
1. Strategic alignment: Alignment of a GCC with 1. Technology infrastructure: Technology
the parent organisation’s goals and objectives, capabilities include computing power, scalable
demonstrating its ability to deliver strategic storage, advanced AI/ML tools, frameworks and
business outcomes and support the parent libraries, and networking to support end-to-
organisation in pursuit of its goals. end solutions.
2. S ervices/Processes delivered: The range 2. T alent pool: Expertise in AI, ML, data science
and depth of services and processes delivered and software development for strategising and
indicate the level of collaboration and demand implementing GenAI solutions at scale.
for GenAI use cases.
3. D ata management capability: Effective
3. L eadership buy-in: Align with the business/ data storage, processing and management
functional and regional leaders to obtain capabilities within GCC.
resources and sponsorship to drive GenAI
4. Change management and communications:
initiatives.
Effectively drive awareness of GenAI solutions
4. Culture of innovation: The extent to which a and ensure employee readiness through
GCC fosters an environment that encourages knowledge management initiatives and
experimentation, innovation and adoption of communication.
new technologies and methodologies.
5. R isk, compliance and security: Established
governance structures for data privacy and
processes to mitigate hallucinations and
unethical responses.
6. P artnerships: Third-party partnerships with
industry players, hyperscalers, academia,
research institutions or start-ups to enhance
GenAI capabilities.
09
A GCC leader’s guide for driving Generative AI adoption
Figure 1: GenAI adoption readiness assessment framework
High
High
Initiate groundwork: The GCC is in a nascent
stage and should work towards building the
ecosystem enablers and capabilities to deliver
GenAI initiatives. It must start by aligning with
the parent on how its contributions will enable
it to achieve the goals and objectives. GCCs
must seek sponsorship from leaders and strive
to build a pipeline for GenAI use cases. Based
on the role alignment with the parent, GCCs
must build/enhance capabilities to meet the
desired objectives.
srelbane
metsysocE
A GCC leader’s guide for driving Generative AI adoption
A quick mapping of GCC’s abilities across the two dimensions would reveal its readiness quotient to
successfully undertake the GenAI journey. This assessment identifies GCC's current positioning and
highlights the key focus areas.
Favourably placed: The GCC is well-positioned
Build capability: While there is clarity and
to implement GenAI initiatives effectively as
alignment on GCC contributions to the parent
it has clarity on its role and strategy in line
organisation and express support from the
with that of the parent organisation. The
leadership for driving GenAI initiatives, the GCC
GCC collaborates seamlessly with the parent
must look to ramp up capabilities across talent,
company and ensures there is sponsorship and
technology, etc., to successfully deliver the
demand from the parent leadership to drive
GenAI initiatives.
the GenAI initiatives. The GCC has also built
capabilities to deliver on the GenAI agenda.
Build Favourably
capability placed
Initiate Re-evaluate
groundwork strategy
Low
Capabilities
Re-evaluate strategy: While there could be
capabilities to deliver on GenAI, re-engage with
the parent and align on the strategy for the
GCC in driving GenAI initiatives. The GCC should
ensure clarity on how its initiatives will contribute
to the overall goals of the parent organisation
and seek leadership buy-in to ensure continuous
demand for GenAI driven from the GCC. As the
GCC has developed certain GenAI capabilities,
it should reassess, reorganise and redeploy its
capabilities in line with the strategy defined for
the GCC.
10
A GCC leader’s guide for driving Generative AI adoption
2. R e-evaluate strategy: While there could be capabilities to deliver on GenAI, re-engage with the parent and align on the strategy
for the GCC in driving GenAI initiatives. The GCC should ensure clarity on how its initiatives will contribute to the overall goals of
the parent organisation and seek leadership buy-in to ensure continuous demand for GenAI driven from the GCC. As the GCC has
developed certain GenAI capabilities, it should reassess, reorganise and redeploy its capabilities in line with the strategy defined
for the GCC.
Figure 2: Assessment criteria for parameters to evaluate GCC readiness
Ecosystem enablers
Low Parameter/Dimension High
Limited alignment with parent GCC goals and objectives,
organisation on goals and Strategic and its operations, methods
objectives, and its operations, alignment and practices are aligned and
methods and practices operating in unison with that of
the parent organisation
GCC supports a limited number GCC supports multiple functions
Services/
of functions and processes and a wide array of processes
Processes
delivered and sub-processes
Leaders not forthcoming in Leaders actively support shared
supporting and committing goals and vision and are willing
Leadership
resources for GCCs to undertake to contribute to success by
buy-in
new and bold initiatives committing effort and budget for
new and bold initiatives
Risk-averse mindset with a focus Strong focus on innovation with
on maintaining the status quo; Culture of consistent support, fostering
minimal focus on innovation innovation creativity and proactive adoption
of new technologies across GCC
11
A GCC leader’s guide for driving Generative AI adoption
Parameter/Dimension Low High
Strategic alignment Limited alignment with parent
organisation on goals and objectives, and
its operations, methods and practices
Services/Processes delivered GCC supports a limited number of GCC supports multiple functions and
functions and processes a wide array of processes and sub-
processes
Leadership buy-in Leaders not forthcoming in supporting Leaders actively support shared goals
Capabilities
and committing resources for GCCs to and vision and are willing to contribute
undertake new and bold initiatives to success by committing effort and
budget for new and bold initiatives
Low Parameter/Dimension High
Culture of innovation Risk-averse mindset with a focus on Strong focus on innovation with
maintaining the status quo; minimal focus consistent support, fostering creativity
on innovation and proactive adoption of new Basic infrastructure is limited Full stack infrastructure with
technologies across GCC Technology
by computing power, storage greater levels of computing
Technology infrastructure Basic infrastructure is limited by Full stack infrastructure with greater infrastructure
capacity, AI frameworks and power and storage, and access
computing power, storage capacity, AI levels of computing power and storage,
libraries, etc. to AI frameworks and libraries
frameworks and libraries, etc. and access to AI frameworks and
libraries
Talent pool Talent with skillsets in traditional tools and Availability of talent pool with
Talent with skillsets in traditional Availability of talent pool with
technologies; talent not geared for driving proficiency in new and advanced
tools and technologies; talent proficiency in new and advanced
new and cutting-edge technologies technologies, including AI/ML, analytics Talent pool
and data science not geared for driving new and technologies, including AI/ML,
Data management capability Inconsistent data governance, limited Robust data governance, advanced cutting-edge technologies analytics and data science
data handling and poor data quality data processing capabilities and high
data quality
Change management and Ad-hoc change management with limited Proactive change management with Inconsistent data governance, Data Robust data governance,
communications communication, training and employee effective communication, employee limited data handling and poor management advanced data processing
engagement involvement and comprehensive
data quality capability capabilities and high data quality
training and support
Risk, compliance and security Inadequate control and monitoring Actively enforced data regulation
mechanisms to comply with data privacy guidelines and controls with strong
Ad-hoc change management Proactive change management
regulations, relying only on basic data data encryption, access controls,
encryption and access controls network security measures, etc. with limited communication, Change with effective communication,
Partnerships Vendor relationships with the ecosystem Vendor relationships are long- training and employee management and employee involvement and
players are new, transactional and at a term, strategic and mature; hold engagement communications comprehensive training and
nascent stage; require significant effort to considerable bargaining power
support
gain bargaining power
Inadequate control and Actively enforced data
monitoring mechanisms to Risk, compliance regulation guidelines and
comply with data privacy and security controls with strong data
regulations, relying only on encryption, access controls,
basic data encryption and network security measures, etc.
access controls
Vendor relationships with the Vendor relationships are long-
ecosystem players are new, term, strategic and mature;
transactional and at a nascent Partnerships hold considerable bargaining
stage; require significant effort power
to gain bargaining power
After establishing the readiness levels of GCCs to adopt GenAI, the next step is to identify impactful use
cases for implementation. The next section will explore the methodology for identifying and evaluating
potential use cases.
12
A GCC leader’s guide for driving Generative AI adoption
How can GCCs identify, qualify and
prioritise use cases?
GCCs should adopt a systematic approach to identifying, qualifying and prioritising GenAI use
cases. This approach should be focused on achieving end-user adoption and ensuring alignment
with business goals and feasibility for maximum impact.
Identification Qualification Prioritisation
Analyse existing service Qualify the identified Identify the priority
catalogues or process use cases by evaluating order for the qualified
taxonomy to identify use key factors such as data use cases by evaluating
cases that use GenAI availability, technical value/benefits vis-à-vis
capabilities both within feasibility, financial the effort required.
the GCC and in the viability and associated
broader organisation. risks.
Identification of use cases
As part of the first step, GCCs need to identify potential use cases that can significantly benefit from
this technology by analysing existing service catalogues and process taxonomies and collaborating
with process leads/owners. During this assessment, it is important to identify the end user and the
benefits they realise. This process is guided by the following key questions, each aligned with different
applications of GenAI:
13
AA GGCCCC lleeaaddeerr’’ss gguuiiddee ffoorr ddrriivviinngg GGeenneerraattiivvee AAII aaddooppttiioonn
Is the content being
summarised?
Condense large volumes of text
Is there any content being or data into concise and coherent Are there any conversations
generated? summaries, highlighting the most involved?
Original text, images, music critical information. Engages in human-like dialogue,
or other media created understanding and responding
E.g., Meeting minutes, workshop
to queries and maintaining
from scratch based on given
summary and document summary.
context over multiple
parameters or prompts.
interactions.
E.g., Report writing, email
E.g., FAQs, chatbots, internal
drafting, image banners and video
helpdesk support and non-textual
generation.
help.
GenAI
Is the content being
improves upon rule-
personalised?
based traditional AI
Tailor content,
by offering greater
recommendations,
creativity, contextual
or interactions to
understanding and
individual users based
nuanced output.
on their preferences,
behaviours and needs.
E.g., Personalised product
recommendations,
adaptive learning paths
and tailored marketing
content.
Is the content being analysed?
Is the content being Examine and interpret data, Is the content being
translated? text or other inputs to identify transformed?
Convert text or speech from patterns, insights and actionable Convert existing content into
one language to another while information. different formats or styles.
maintaining the original meaning
E.g., Forecasting (demand, supply, E.g., Text to code, style transfer
and context.
price), risk identification and and personalisation, text to table.
E.g., French to English, English to feedback sentiment analysis.
Hindi.
14
A GCC leader’s guide for driving Generative AI adoption
FINANCE
Procurement and
L2
purchasing
Supplier selection, Purchase order Document receipt, Approval routing
L3 onboarding and creation and data extraction and database
management delivery management and validation updates
15
seitivitcA
sesac
esU
IAneG
seitilibapaC
Figure 3: Methodology illustrating identification of use cases of two processes
ILLUSTRATIVE
L1
Invoice processing
and management
• Identify and • Create and approve • Receive invoices • Match invoices
evaluate potential purchase orders through channels to appropriate
suppliers • Track delivery of such as email, post approvers based
• Negotiate terms goods and services and fax; scan and on a predefined
and conditions • Address digitise physical approval hierarchy
• Finalise discrepancies, invoices • Update the
agreements and handle returns and • Extract key data company's
onboard suppliers manage supplier and convert various accounting system
performance inputs into one and archive
standard format paid invoices for
• Validate extracted auditing and future
data against reference
purchase
orders; identify
discrepancies, if any
• Supplier risk • Purchase order • Format • Smart approval
report and drafting standardisation routing
evaluation • Delivery forecasting • Error detection • Order and
summary • Personalised and discrepancy delivery query
generation supplier report generation bot for real-time
• Contract drafting communications updates
drafting
Content Content Content Content
generation generation summarisation analysis
Content analysis Content analysis Content analysis
Conversation
Content Content Content
summarisation personalisation transformation
By replicating the use case identification approach across processes and functions within the
organisation, a comprehensive list of potential use cases can be derived. This enables GCCs to
systematically uncover and harness the full spectrum of opportunities presented by GenAI.
AA GGCCCC lleeaaddeerr’’ss gguuiiddee ffoorr ddrriivviinngg GGeenneerraattiivvee AAII aaddooppttiioonn
Use case qualification
GCCs must next evaluate their list of identified use cases for GenAI to ensure they are feasible,
valuable and aligned with organisational goals. The following considerations outline the key
aspects necessary for an in-depth evaluation:
Fact ors to consider for qualification of use cases
Technical
feasibility
Data
availability
Infrastructure requirements
• A re robust API integration capabilities available?
• A re development frameworks such as ML and
NLP libraries available?
Requirement • A re other supporting hardware/software
Does the use case require domain-specific data for infrastructure available?
training the model?
Vendor availability
Quantity
Can use cases or clusters of use cases (with the
Is data sufficient for model building? Are there
same capability) be implemented using common
significant gaps or any missing data points?
vendors available in the market?
Quality
Is the data good enough for contextual
Reusability
understanding? Are multiple data sources available
Can the solution be reused or adapted for other
to enrich and diversify the dataset and remove
use cases requiring similar capability?
bias?
Validation
Are evaluation metrics in place to ensure GenAI
output accuracy and reliability? Is there a process
for identifying and addressing biases and for
comparing outputs to desired outcomes?
16
AA GGCCCC lleeaaddeerr’’ss gguuiiddee ffoorr ddrriivviinngg GGeenneerraattiivvee AAII aaddooppttiioonn
Financial
feasibility
Risk,
compliance
and
regulatory
Cost Privacy and security
• W hat are the costs associated with infrastructure Are there data privacy or security concerns?
setup, maintenance, hiring and upskilling?
Ethical concerns
• W hat are the vendor partnerships, licensing fees, Could the use case trigger unethical responses?
integration downtime costs and ongoing support
Safety check
costs?
Can biases, errors and hallucinations be corrected?
Regulatory
Benefits
Are there regulatory requirements or compliance
• W hat tangible and intangible benefits can be
issues that must be addressed?
achieved, such as direct cost savings, operational
efficiency and resource optimisation? Use case exposure
Are the use cases being implemented for external
• H ow does it enhance quality, reduce errors/risks
customers with high exposure and risk compared
and improve overall performance?
with internal customers?
17
A GCC leader’s guide for driving Generative AI adoption
Figure 4: Methodology illustrating qualification of use cases of two processes
18
ECNANIF
The application of the use case qualification framework is illustrated through two finance processes.
This example demonstrates how use cases can be thoroughly vetted, ensuring only the most viable and
impactful ones are selected for implementation.
Content
personalisation
By performing these checks for each use case, a subset of qualified use cases can be derived. This
serves as a foundation for further prioritisation, ensuring that only the most valuable and practical use
cases are qualified.
gnisahcrup
dna
tnemerucorP
tnemeganam
dna
gnissecorp
eciovnI
L1 L2 L3 Use cases GenAI Data Technical Financial Risk, Qualification
capability availability feasibility feasibility compliance
and
regulatory
Supplier Supplier risk
selection, report and
onboarding evaluation
and summary
management generation
Contract drafting
Purchase Purchase order
order creation and order
and delivery confirmation
management drafting
Delivery
forecasting
Personalised
supplier
communications
drafting
Document Format
receipt, data standardisation
extraction
Error detection
and validation
and discrepancy
report generation
Smart approval
Approval
routing
routing and
database Order delivery
updates query bot for real
time updates
Content Content Content Content Content
Conversation
generation analysis summarisation transformation personalisation
A GCC leader’s guide for driving Generative AI adoption
Prioritisation of use cases
After qualifying use cases, GCCs must prioritise them by evaluating the expected benefits and effort
required. A framework provided below supports use case prioritisation across two dimensions: Benefits
and effort.
High
Low-hanging High-impact
fruits investments
Incremental
Resource drains
gains
Low
Low High
19
stfieneB
Effort
Benefits
Financial benefit: The net financial benefit enabled through a business case considering key metrics,
including run costs, Total Cost of Ownership (TCO) and payback period.
Strategic alignment: The extent to which the use case aligns with the organisation’s strategic
objectives.
Scalability and reusability: The ability of a solution to scale or use across multiple use cases (cross-
functional potential).
Non-financial benefit: The benefits such as improvement in productivity, agility, customer satisfaction
or employee experience.
Effort
Time: The time required to realise benefits.
Talent: Resources with multiple skillsets deployed across lifecycle.
Budget: Estimated budget, including development and run cost |
167 | deloitte | Generative AI in Europe _ Deloitte Insights.pdf | Now decides next. Is Europe ready for
generative AI?
Opportunities and hurdles: Europe's path in the Generative AI era
ARTICLE • 12-MIN READ • 18 JANUARY 2024
Since its debut in 2022, ChatGPT has rapidly seized the attention of businesses and societies worldwide, prompting organisations to rethink their
practices and strategies around tech and talent. Yet, as revealed in Deloitte’s The state of generative AI in the enterprise, regional disparities exist in the
adoption and readiness for such generative artificial intelligence tools as ChatGPT and Bard.
Factors such as investment levels, regulatory environments, risk appetite and talent availability vary significantly around the world, influencing
organisations’ ability to unlock the potential of generative AI. Europe, in particular, has the potential for growth in organisational preparedness,
adoption of generative AI tools and applications, risk management of generative AI and talent-related strategies. This article focuses on the
opportunities and challenges affecting Europe’s AI landscape, including labour shortages, skills gaps and stricter regulations.
Understanding generative AI: Deloitte's global research methodology
From 12 October to 5 December 2023, Deloitte surveyed over 2,800 global leaders (directors and above) to understand their views on generative AI. Participants were
required to have at least one working implementation of AI and a pilot of gen AI. The survey included respondents from the Americas (56%), Europe (27%) and Asia-
Pacific (17%).
There were 756 European business leaders from various countries and industries, with most representing organisations earning over US$1 billion annually. All
respondents have roles in their organisation’s AI and data science strategy decisions, investments, implementation approach and value measurement.
Generative AI: Transforming content generation, search and conversational
interfaces
Generative AI, a specific type of AI known for creating human-like outputs,1 is used to develop content across various formats like text, computer code,
audio and/or visual output.2 The most common applications reported by survey respondents globally included content generation, search/knowledge
management, virtual assistants/conversational chatbots and content summarisation. In terms of integrated generative AI resources, the top categories
are productivity applications, enterprise platforms, publicly available large language models (LLMs) and code generators.
The state of generative AI in the
enterprise
Read more
European perspective on generative AI
Balancing caution with opportunity in adoption of generative AI
While there are broad similarities in use cases globally, European leaders show less interest and attention towards generative AI than their counterparts
in the Americas and Asia-Pacific regions (figure 1). In line with this lower level of engagement, a significant portion of European respondents (over
20%) believe their industry and their own organisations are paying ‘too little attention’ to generative AI’s potential and implications.
This could relate to less perceived pressure for European respondents to adopt generative AI, with only 26% reporting significant pressure compared to
higher percentages in the Americas and Asia-Pacific. Additionally, they anticipate a more extended time frame for AI to significantly transform their
organisations, with a higher proportion of European leaders believing it will take more than three years, and only 9% currently seeing transformative
effects take shape. This contrasts with higher percentages in other regions where users believe in AI’s immediate transformative impact (figure 2).
Research from The Deloitte Global Boardroom Program found that almost half (48%) of European leadership teams and board members identified
their inability to show how technology enables growth as their biggest challenge when assessing the value of digital transformations.3 This reflects a
broader technology-literacy predicament for European executives resulting in their belief that their organisations are not ready for generative AI.
When asked about the emotions leaders associate with generative AI, excitement and fascination are common responses across regions. Still, European
leaders report notably lower trust concerning the technology. This mistrust may stem from cultural differences and concerns about AI-associated risks
like biases and copyright issues.4 European companies are focused on developing this new technology responsibly and ensuring its trustworthiness.
They aim to balance the potential and advantages of generative AI with the need for it to be regulated. This means ensuring that AI systems are fair,
impartial and accountable. They also want AI to be responsible, robust and dependable, while being safe and secure and protecting privacy and
confidentiality. Emphasising ethical AI practices could help organisations avoid reputational risk and enhance trust among customers and employees.
High expectations for productivity amidst slow adoption of generative AI tools
European leaders in our study highlight efficiency, productivity, cost reduction, innovation and growth improvements as the benefits of generative AI,
which mirrors global findings. These results are also consistent with previous reports such as the autumn 2023 edition of Deloitte’s European CFO
Survey.5 A significant 91% of European respondents expect generative AI to increase productivity, aligning with global results. This is particularly
significant for Europe, given the region’s recent productivity challenges, as highlighted by Deloitte Germany’s research into the economic effects of a
shrinking workforce.6
Despite such acknowledged benefits, European leaders face implementation challenges. Lower interest levels, trust gaps, slow implementation of
governing regulation and expectations of longer timelines for generative AI–driven change hinder organisational investment in and readiness for these
technologies. Compared to other regions, European leaders report less preparedness for adopting generative AI in business areas like risk management,
strategy, talent development and technology infrastructure.
Similarly, generative AI adoption in Europe is lower across all business functions compared to other regions (figure 3). Alongside regulatory
considerations, this may stem from Europe’s challenging economic conditions and ongoing geopolitical tensions impacting interest and slow adoption.
The survey took place against a backdrop of a US economy that had outperformed expectations and in which growth had accelerated. In contrast,
European growth had slowed sharply and Germany, although not the euro area as a whole, had fallen into recession. The US has also enacted policies
to enhance economic competitiveness, such as the Inflation Reduction Act and the CHIPS and Science Act. However, this does not necessarily explain
the higher levels of adoption in the Americas as the NextGenerationEU programme could provide similar incentives for European organisations to
adopt generative AI.7
Lower levels of generative AI adoption are certainly a result of European companies operating in a more complex and regulated environment than their
counterparts in the Americas and Asia-Pacific regions. In December 2023, the EU provisionally agreed on the EU AI Act, its landmark, world-first AI
regulation, which will introduce a comprehensive, legally binding, cross-sectoral framework for the technology to regulate its use and development.
Using a risk-based but prescriptive approach, the law will regulate AI, including generative AI, based on the potential risks of specific models or
applications. Certain AI use cases, such as behavioural manipulation, will be banned altogether. For AI systems and models deemed high-risk,
organisations providing or deploying them will be subject to stringent requirements, including pre-deployment fundamental rights impact assessments,
pre-market conformity assessments and transparency obligations, to name but a few.8 While the compliance implications are likely to be substantial,
the Act will also bring more accountability and fairer distribution of responsibilities across the AI value chain, as well as increased consistency across
sectors.
The Act will also have global implications, as it will apply to any AI providers or deployers whose systems are marketed or affect individuals residing in
the EU, regardless of their location. The final legal text, expected in early 2024, will give organisations further details to fully assess the Act's
operational and strategic impacts.9 It will be interesting to observe whether further clarity on the EU regulations will speed up the pace of
implementation of generative AI in Europe.
Walking the tightrope: As generative AI meets EU regulation, pragmatism is likely
Read more from TMT Predictions 2024
While there’s an expectation of comparable increased investment across the Americas, Asia-Pacific and Europe, European organisations in our survey
reported allocating less budget to generative AI than their peers in other regions. The wait for the final legal text of the EU AI Act may account for the
reluctance of European executives to move forward with investment as they wait to understand the regulatory trickledown of what the Act means for
them in practice. Further, Europe does not have the same legacy of investing in digital transformation and disruptive technologies (figures 4 and 5).
Historically, most external private investments in such technologies have been concentrated in the Americas, with the leading creators of generative AI
and the most notable LLMs in the world being based mainly in the US.10
Globally, our report shows that leaders tend to prefer buying over building generative AI tools, a trend particularly noticeable in Europe, where 37%
acknowledge this as their go-to strategy. In the Americas, it is 33%, and in Asia-Pacific 32%. This strategy is cost-effective for routine activities but
offers limited control and lacks a disruptive competitive advantage.11 However, this may not be a choice for many European organisations, who likely
do not have the resources to create and experiment with LLMs and lack access to the high-specification hardware needed to train models. It has been
widely reported that the graphic-processing units needed, for example, Nvidia A100/H100, have been stockpiled by various entities, especially in
Asia.12
Talent strategies
European organisations are less active in reskilling workers, educating their workforce and recruiting technical talent (figure 6). The latter is partially
due to the region’s more limited talent pool and existing skills shortages.13 More than a third of the EU’s labour force lacks necessary digital skills,14
and the UK seems to be in a similar position.15 These talent shortages, combined with modest efforts in educating and reskilling workers, are hindering
Europe’s ability to leverage the benefits of generative AI fully.
Europe’s cautious approach to reskilling its workforce may be influenced by its strong labour protection laws and high unionisation rates. In the case of
generative AI leading to job displacement, European businesses may perceive the immediate benefits of generative AI, like cost savings and productivity
gains, as less substantial compared to regions with less stringent labour laws.16 Additionally, robust labour protection and trade unions require
European companies to adopt a more deliberate approach when implementing technologies that could displace jobs as it can involve complex legal
considerations.
Yet there is also the possibility of generative AI leading to job augmentation, rather than job displacement via automation. A recent Deloitte report on
generative AI and the future of work17 suggests “there is a growing sense that generative AI will augment the human workforce rather than replace it.”
In other words, generative AI can enhance the workforce experience by eliminating routine tasks, allowing employees to focus on more meaningful
work and increasing employee job satisfaction and performance in the process.
As such, these rather limited efforts around talent might have adverse implications. The general-purpose nature of generative AI means that the demand
for skilled labour could increase across a broad range of occupations and industries. In addition, in countries with ageing workforces or declining
working-age populations, there’s often an increased drive towards automation to compensate for labour shortages.18 Firms in regions with a declining
number of middle-aged workers have historically turned to automation to make up for this demographic shortfall. With many European countries
dependent on declining working-age populations, the likelihood of widespread generative AI adoption increases.19
Completing such a transition means an increased demand for skilled workers at a time when demographic trends mean companies will be competing
for an ever-shrinking labour pool.20 This makes the lack of transparency of European businesses and reluctance to actively educate their workforce
about AI’s capabilities, benefits and value puzzling. Still, organisations will only realise generative AI’s potential with the understanding and acceptance
of employees. In particular, their fears about automation and job displacement need to be addressed.
Many European respondents in our study believe it will take up to two years to adjust their talent strategies for generative AI, with fewer feeling an
immediate need for change than counterparts in the Americas or Asia-Pacific. This may indicate a more cautious approach to organisational change
amid ongoing considerations of the technology’s risks, or it may simply be as a result of not yet knowing what the workforce implications will be as
this technology rolls out. Will generative AI replace jobs or make jobs easier and more enjoyable? Whether it plays more of a role in enhancing the
employee experience and enabling people to be more productive at work or taking over entire tasks and roles is yet to be determined as the potential of
this technology is explored.
Talent and skill gaps: Europe’s main challenge to maximising generative AI's potential
Across all regions, the technical talent shortage is a critical barrier to developing and deploying generative AI, with nearly 40% of European leaders
selecting this as a key obstacle This is consistent with previous Deloitte analyses that identify talent resources and capabilities as the main challenge in
Europe.21 European leaders also cite a lack of an adoption strategy and regulatory compliance concerns more than leaders in other regions (figure 7).
This is even though European organisations have less difficulty identifying use cases than peers in different regions.
Concerns common across regions include intellectual property issues, regulatory compliance, a lack of confidence in AI results, transparency, data
privacy and data misuse. European respondents more frequently see risk management as a barrier to implementing generative AI and are less convinced
about their organisation’s efforts in governing AI adoption and mitigating potential risks. Effective governance of generative AI is likely to be an
essential precursor to its scalable adoption across European organisations.
Respondents were also asked about strategies for managing generative AI risks. Top actions include monitoring and regulatory compliance, governance
frameworks and internal audits. European respondents particularly emphasised regulatory compliance as important, tying back to the need for a clearer
understanding of how the EU AI Act will impact organisations in practice.
Moreover, with generative AI, risk and regulation are no longer an exercise in technology management. Instead, when considered equally to other
strategic levers they can realise significant value. The relative novelty of LLMs in business applications can be a challenge, and the risks of LLMs are
dynamic and may change depending on their interactions with the user. However, development of guardrails, alongside proportionate deployment of
testing, controls and monitoring mechanisms can empower organisations to use generative AI safely and confidently.22
Generative AI: A strategic imperative for European businesses
This analysis shows that European leaders should prioritise preparing their organisations and workforce for the disruptive potential of generative AI.
Recent Deloitte reports indicate that, although generative AI is a new technology requiring time for adoption and benefits realisation, aligning it with
an organisational strategy is critical.23
Europe’s cautious approach to this emerging technology, characterised by a wait-and-see attitude, contrasts with the more proactive stances reported in
the Americas and Asia-Pacific regions. This difference in approach could see Europe lag in exploring the potential for generative AI, but it could also
result in a more responsible deployment environment that considers new responsibilities that are created when technologies are invented.
Balancing the need for trust with the urgency to remain competitive in the global market is critical. This involves taking a multi-disciplinary approach
to develop generative AI transformation strategies from the outset, and not just considering the technology potential itself. By approaching technology
investment responsibly, while also investing in the necessary training and development of the workforce, European organisations can better position
themselves to capitalise on the enormous benefits of generative AI, such as increased efficiency, innovation and competitive advantage.
Let’s make this work.
Change your Analytics and performance cookie settings to access this feature.
BY Stacey Winters Richard Horton
United Kingdom United Kingdom
Roxana Corduneanu
United Kingdom
Endnotes
1. Deloitte, “Deloitte AI Institute UK,” accessed 11 January 2024.
View in Article
2. Ibid.
View in Article
3. Dan Konigsburg, William Touche, and Jo Iwasaki, Digital frontier: A technology deficit in the boardroom, Deloitte Insights, 13 June 2022.
View in Article
4. Caroline Atkinson, Europe and technology, Hoover Institution, 4 February 2019; Lukas Kruger and Michelle Seng Ah Lee, “Risks and ethical
considerations of generative AI,” blog, Deloitte, 6 June 2023.
View in Article
5. Jose Manuel Dominguez Carravilla, Richard Muschamp, Rolf Epstein, Dr. Pauliina Sandqvist, and Ram Krishna Sahu, European CFO Outlook
—Autumn edition, Deloitte Insights, accessed 11 January 2024.
View in Article
6. Deloitte Insights Magazine, To help bolster aging economies, boost workforce participation, data point, accessed 11 January 2024.
View in Article
7. Stefano Alfonso, Hilde Van de Velde, Miguel Eiras Antunes, Luca Bonacina, and Carlos Bofill, Futureproofing Europe: How the
NextGenerationEU programme is inspiring companies to transform, Deloitte Insights, 24 July 2023.
View in Article
8. Providers of general-purpose AI models and systems will be subject to specific requirements, based on the level of risk their products pose.
View in Article
9. Valeria Gallo and Suchitra Nair, “The EU AI Act: The finish line is in sight,” blog, Deloitte, 13 December 2023.
View in Article
10. Michael Chui, Eric Hazan, Roger Roberts, Alex Singla, Kate Smaje, Alex Sukharevsky, Lareina Yee, Rodney Zemmel, The economic potential of
generative AI: The next productivity frontier, McKinsey & Company, accessed 11 January 2024.
View in Article
11. Forbes, “Should you build or buy your AI?,” 22 May 2019.
View in Article
12. Qianer Liu and Hannah Murphy, “China’s internet giants order $5bn of Nvidia chips to power AI ambitions,” Financial Times, 10 August 2023.
View in Article
13. Martin Arnold and Valentina Romei, “Eurozone jobless rate hits record low of 7% as worker shortages spread,” Financial Times, 1 February
2022.
View in Article
14. European Union, “Plugging the digital skills gap,” accessed 11 January 2024.
View in Article
15. Jo Thornhill and Kevin Pratt, IT skills gap report 2023, Forbes, 13 September 2023.
View in Article
16. Ira Kalish and Michael Wolf, Generative AI and the labor market: A case for techno-optimism, Deloitte Insights, accessed 11 January 2024.
View in Article
17. Nicole Scoble-Williams, Diane Sinti, Jodi Baker Calamai, Bjorn Bringmann, Laura Shact, Greg Vert, Tara Murphy, and Sue Cantrell, Generative
AI and the future of work: The potential? Boundless, Deloitte AI Institute, accessed 11 January 2024.
View in Article
18. Daron Acemoğlu and Pascual Restrepo, Demographics and automation, MIT, accessed 11 January 2024.
View in Article
19. Kalish and Wolf, Generative AI and the labor market.
View in Article
20. Ibid.
View in Article
21. Carravilla, Muschamp, Epstein, Sandqvist, and Sahu, European CFO Outlook—Autumn edition.
View in Article
22. Deloitte, “Embedding controls and risk mitigations throughout the generative AI development lifecycle,” blog, accessed 11 January 2024.
View in Article
23. Gregory Dost and Diana Kearns-Manolatos, “Unleashing value from digital transformation: Paths and pitfalls,” blog, 14 February 2023; Brenna
Sniderman, Diana Kearns-Manolatos, and Nitin Mittal, Generating value from generative AI, Deloitte Insights, accessed 11 January 2024.
View in Article
Acknowledgments
The authors would like to thank Nancy El-Aroussy, Ralf Esser, Valeria Gallo, Ira Kalish, Paul Lee, Lucia Lucchini, Costi Perricos, Pauliina Sandqvist,
Michelle Seng Ah Lee, Sulabh Soral, Ben Stanton, Ian Stewart and Michael Wolf for their insights and contributions to this piece.
Cover image by: Mark Milward |
168 | deloitte | in-ra-ai-risk-management-noexp.pdf | AI Risk Management
Risk mitigation "now" and
strategic insights "next"
March 2024
AI Risk Management | Risk mitigation "now" and strategic insights "next"
2
AI Risk Management | Risk mitigation "now" and strategic insights "next"
Table of contents
Introduction 04
Enhancing trustworthiness at every stage of
the AI lifecycle 04
A layered approach to building a trustworthy AI 05
AI risk universe—Illustrative 07
Deloitte’s Trustworthy AI framework 11
Need for governance structure across the AI lifecycle 16
Way forward 17
3
AI Risk Management | Risk mitigation "now" and strategic insights "next"
Introduction
In today’s growing market, Artificial Intelligence (AI) is an imperative for various industries. Organisations are exploring the use of AI for
several solutions, including automation, to deliver value and bring efficiency to operations. If companies are relying heavily on AI, they
need to ensure ethical assurance and trustworthiness to make their AI systems dependable.
A solid framework can help organisations navigate this journey and gain confidence against various regulatory requirements as the AI
landscape evolves.
Enhancing trustworthiness at every stage of the AI lifecycle
Ideation and design
Accounting for applicable regulations for the business/industry and local or target geography to ensure
compliance by design from the initial stages of the AI project.
Model requirement
Establishing clear and comprehensive requirements during the model requirement stage of the AI lifecycle to
ensure successful development and deployment; data scientists and engineers could be involved proactively
during this stage to minimise the risk of project failures, costly rework, and potential ethical or legal issues.
Model development
Forming standards and best practice guidelines for developers, ensuring their technologies adhere to
compliance requirements at every stage of the AI lifecycle.
Model deployment
Demanding concrete and trustworthy demonstration from developers and/or vendors, ensuring their AI
systems adhere to relevant ethical, legal, and technical standards.
Data requirement
Certifying prerequisites in available data for AI models, such as adequacy, representativeness, and high quality
to prevent bias, discrimination, and unreliable results.
4
AI Risk ManaTgheem ‘wehnat t|, wRihsky ,m aintidga htioown ’" onfo twh"e a UndP Is ptraaytemgeicn itnss iegchotss y"snteexmt"
Data cleansing
Conducting data cleansing (error detection, standardisation, and normalisation) to eliminate errors, ensure
consistency, and optimise model performance in AI projects.
Data labelling
Ensuring accurate and detailed labels with bias mitigation to avoid errors and manage data effectively.
Model training and testing
Conduct thorough validation and testing of the model using diverse datasets, including both training and
validation data. Perform sensitivity analysis and stress testing to assess the robustness and reliability of the
model under different scenarios. Use adversarial testing to identify vulnerabilities and potential security
risks, such as adversarial attacks.
Model monitoring
Conducting continuous performance tracking, data drift detection, model retraining, maintaining
transparency, and confirming compliance with regulations and ethical standards to maintain model reliability
and accountability in decision-making.
A layered approach to building a trustworthy AI
To achieve a strong AI governance and risk management, it is crucial to establish multiple security layers when deploying AI programs.
The three Lines of Defence (3LoD) model is a fundamental framework that delineates three integral layers of defence, each with unique
responsibilities and accountabilities. At the core of this framework lies the pivotal role of personas, seamlessly integrated across these
lines of defence.
Through this process, organisations establish a resilient AI governance structure and foster transparency, accountability, and risk
mitigation throughout the AI lifecycle.
5
AI Risk Management | Risk mitigation "now" and strategic insights "next"
Lines of defence
Teams responsible:
• Independent AI assurance and audit team
• Internal auditors
Third
• Ethical AI review board
line of
defence
Teams responsible:
• AI governance team
• Compliance and ethics team
Second line of • Risk and compliance function
• Data privacy officer
defence
• Cyber security experts
Teams responsible:
• Business unit owners
First line of
• AI developers
defence • AI/ML engineers
Assurance checks
First line Second line Third line
• Enable increased first line of defence • Setting AI risk appetite • Enhance transparency and
testing by model owners through stress accountability with internal
• Identifying KRIs
testing and continuous testing audits by sharing model data
• Including forward-thinking risk and enabling audit trails, etc.
• Automating model validation and
taxonomies
monitoring
• Review model docs via governance
• Define model parameters and refine model
dashboards
development processes
• Establish AI risk strategy
• Enable real-time issue alerts
6
6
AI Risk ManaTgheem ‘wehnat t|, wRihsky ,m aintidga htioown ’" onfo twh"e a UndP Is ptraaytemgeicn itnss iegchotss y"snteexmt"
AI risk universe—Illustrative
Awareness of the following risks in the AI development lifecycle is crucial for promoting responsible design, ensuring ethical
implementation, and fostering sustainable technological advancement.
Strategic
Risk category description
Risk of AI strategy/leadership not aligned to organisational/business objectives/leadership
Individual risks
Model requirements:
• AI strategy not coordinated with company strategies/value systems/risk appetite leads to
ineffective or even malicious/unethical models
Financial
Risk category description
Risk of inadequate and incorrect decisions/recommendations due to poor AI models, resulting in
direct and indirect losses or threats to the organisation, customer, brand, and reputation
Individual risks
Model evaluation:
• Financial losses, wastage of resources, and reputational losses because of wrong AI models
Data
Risk category description
Risk of unavailability of accurate, labelled, relevant, and unbiased data to develop, train, and deploy
models that meet its intended purposes
Individual risks
Data labelling:
• Inaccurate models from mismatched tests, production data, and improper data tagging
Data collection:
• Risk of biased or insufficient data for model development data cleaning
• Unauthorised access disintegrates solution alignment with business goals
Data labelling:
• Test data different from production data can result in inaccurate models, while inadequate data
tagging based on sensitivity can result in inappropriate safeguards.
7
TAhI eR i‘swkh Mata, nwahgye,m aenndt h |o Rwis’ ko fm tihtieg aUtiPoIn p "anyomwe" nantsd e sctorastyesgtiec minsights "next"
Technology
Risk category description
Risks associated with the technology used regarding auditability, scalability, and monitoring
Individual risks
Model monitoring:
• Tech constraints limit auditability and audit logs, hindering transparency
• Lack of monitoring and feedback loops delay corrections for model discrepancies
Model deployment:
• Single points of failure in deployment without redundancy and inflexible technology limit
scalability as the organisation grows.
Algorithmic
Risk category description
Risks associated with the algorithms leading to incorrect/inconsistent/biased/unethical decisions and
financial and reputational implications.
Individual risks
• Model training: Biased data begets biased and unreliable AI models
• Model evaluation: Inadequate risk-based stress testing and documentation can harm models
• Model deployment: Insecure coding and design flaws invite vulnerabilities
• Model monitoring: Absence of mechanisms for monitoring changing environments
Cyber ( including Data Privacy)
Risk category description
Risk of not identifying, labelling, storing, and securing Personally Identifiable Information (PII)
resulting in data privacy breaches, leading to reputational backlashes and regulatory repercussions.
Individual risks
Privacy: (Data labelling and data collection)
• Insufficiently secured data in AI models, lack of opt-in/opt-out options, and unauthorised data
use infringe on privacy rights.
8
8
AI Risk ManaTgheem ‘wehnat t|, wRihsky ,m aintidga htioown ’" onfo twh"e a UndP Is ptraaytemgeicn itnss iegchotss y"snteexmt"
Cyber
Risk category description
Lack of adequate access controls in place to safeguard infrastructure, application, model,
and underlying code
Individual risks
Infrastructure:
• Risks pertaining to the underlying technology and resources that support the AI system. This
includes servers, networks, databases, and cloud services.
Application:
• Risks involving issues related to the AI application's functionality, usability, and integration
Model:
• Risks focussing on the AI model’s performance, interpretability, and generalisation capabilities
Underlying code:
• Risks involving challenges related to the quality, security, and documentation of the AI
system's codebase
People
Risk category description
Risk of unavailability of skilled people at each stage of the AI lifecycle and lack of clear segregation
of roles and responsibilities in terms of human-machine interface.
Individual risks
Talent:
• Risk on the company's talent culture (skills atrophy) due to AI implementation may lead to
employee resentment.
Governance:
• Insufficient AI skills
• Unclear roles and unapproved developments
• Missing human-machine interaction guidance (Override)
• Expertise loss risk
• Diversity prevents bias
Regulator
Risk category description
Risk of not catering to geographical or sectoral regulatory and compliance requirements with respect
to AI models, resulting in litigations, fines, and regulatory scrutiny.
Individual risks
Model evaluation:
• Lack of clarity on regulations and its changes around privacy and data security leads to the
creation of ambiguous models, financial penalties and regulatory scrutiny.
Model monitoring:
• Risks such as social engineering and privacy invasion without AI regulation
• Neglecting compliance may result in penalties and business continuity risks
9
TAhI eR i‘swkh Mata, nwahgye,m aenndt h |o Rwis’ ko fm tihtieg aUtiPoIn p "anyomwe" nantsd e sctorastyesgtiec minsights "next"
Third/Fourth-party
Risk category description
Risks arising due to the involvement of third/fourth parties in the AI deployment lifecycle
may lead to technology dependency and intellectual property loss.
Individual risks
• Unclear vendor roles hinder ownership
• Vague contract terms challenge risk management
• Inadequate security controls risk fines and reputation damage
Societal
Risk category description
Risk of incorrect, inconsistent, biased decisions and recommendations made by AI model
leading to issues, such as loss of jobs and exclusion of services causing socio-economic disparity.
Individual risks
• A lack of societal expectation management erodes trust in AI adoption.
• Non-transparent AI models contribute to societal bias and exclusion.
An independent assessor should address various risks associated with AI models, as meeting regulatory requirements will bolster the
entity’s trust:
Independent assurance:
To establish confidence and trust in AI systems, it is necessary to demand well-defined, consensus-
driven standards and credible evidence from developers, vendors, and executives. This evidence
should demonstrate the validity and suitability of the assurance for a specific use case. This can be an
internal and/or external assurance team (auditors, certification bodies, etc.)
Regulations and standards
compliance:
Seeking assurance involves the
essential reliability of AI systems
falling under their regulatory purview,
ensuring compliance with regulations
and best practice guidelines.
The control frameworks developed by
the organisation can use the existing
frameworks, such as ISO 27001,
ISO 42001, COBIT, GDPR, Fairness
Accountability and Transparency
in Machine Learning (FAT ML), and
implementation guidelines, along with
best practices, such as NIST SP 800,
NIST AI Risk Management Framework,
CIS Controls, and OWASP.
10
10
AI Risk ManaTgheem ‘wehnat t|, wRihsky ,m aintidga htioown ’" onfo twh"e a UndP Is ptraaytemgeicn itnss iegchotss y"snteexmt"
Deloitte’s Trustworthy AITM framework
Governments, industries, and various other groups have struggled to set up an AI framework due to the challenging AI evolution across
industries. To bridge the gap, we have developed a Trustworthy AI framework, putting trust at the centre of everything we do.
This helps organisations set up governance structures for AI programmes and meet regulatory compliance throughout the AI lifecycle
from ideation to design, development, deployment, and Machine Learning Operations (MLOps) to empower employees, businesses,
customers, and industries.
11
TAhI eR i‘swkh Mata, nwahgye,m aenndt h |o Rwis’ ko fm tihtieg aUtiPoIn p "anyomwe" nantsd e sctorastyesgtiec minsights "next"
This trustworthy framework is based on the following seven dimensions
Transparent and explainable Fair and impartial Robust and reliable
AI models enable users to AI models prioritise inclusive AI models produce
make decisions that are easy design, promoting equitable consistent and accurate
to understand, auditable, application, access, and outputs, withstand errors,
and open to inspection. outcomes. An impartiality and recover quickly from
This involves assessing assessment examines system unforeseen disruptions and
system complexity, training design to ensure fairness, misuse. AI models must
methods, and efforts to by considering bias and maintain robustness and
enhance comprehension. cultural context. An integral reliability throughout their
It also examines how the part of this is to provide entire lifecycle. They should
system communicates comprehensive support for operate suitably in various
results, reasoning, displaced workers. Ongoing conditions, including
involvement in outcomes, user bias training and diverse normal, foreseeable, and
and avenues for recourse to fairness testing are conducted adverse scenarios.
users and data subjects. to address potential biases
using various definitions.
Private Safe and secure Accountable
AI models help respect user AI models are protected Policies dictate
privacy by limiting data use from risks that may responsibility for
to its intended purpose cause individual and/ AI-related decisions.
and duration. They provide or collective physical, Accountability is gauged
opt-in/out options for emotional, environmental, by transparent supervision
data sharing and evaluate and/or digital harm. of AI model creation and
transparency in user deployment. This ensures
communication regarding clarity and prevents
data policies, system risks, manipulation, with effective
Responsible
testing outcomes, and communication of system
appropriate use. They functions and limitations.
also scrutinise privacy by It includes validating
detailing sensitive data documented design
types used and strategies decisions, system failure
for data protection during reviews, and scenario
AI models are created
training and deployment. planning by the AI team.
and operated in a socially
responsible manner. They
put an organisational
structure in place that can
help determine who is
responsible for the output of
AI system decisions.
12
12
AI Risk ManaTgheem ‘wehnat t|, wRihsky ,m aintidga htioown ’" onfo twh"e a UndP Is ptraaytemgeicn itnss iegchotss y"snteexmt"
Enhancing reliability throughout the AI lifecycle
We explore stage-specific techniques to bolster reliability, linking each stage to its Trustworthy AI element, key stakeholders, guiding
principles, and crucial audit points to consider.
Trustworthy AI
AI lifecycle stage Associated persona Principles Audit focus
element
Ideation and design • Transparent and • AI architect • Traceability and • Assess traceability
explainability of and explainability
explainable
• AI developers significant decisions implementation
• Safe and secure taken by the system
• Review algorithm
• Usage of the simplest simplicity and
algorithm that meets decision override
performance goals mechanisms
• Ability to override • Verify security
the AI system's measures and
decision by third/fourth-party
designated people controls
• Security of users'
data
• Following secure
coding and security-
by-design practices
• Ensuring that
third/fourth-party
stakeholders
implement all the
necessary security
controls
• Alignment to the • Scrutinise
Model requirement • Robust and reliable • Business unit
principles of both alignment with
owners
• Accountable organisation and responsible AI
• AI/ML engineers responsible AI principles
• Reproducibility • Validate
and consistency of reproducibility and
outcomes grievance handling
• Implementation • Review human
of appropriate supervisory
grievance redressal control
and compensation implementation
mechanisms
• Quality assurance—
Human supervisory
control wherever
possible
13
TAhI eR i‘swkh Mata, nwahgye,m aenndt h |o Rwis’ ko fm tihtieg aUtiPoIn p "anyomwe" nantsd e sctorastyesgtiec minsights "next"
Trustworthy AI
AI lifecycle stage Associated persona Principles Audit focus
element
Data cleansing • Fair and impartial • AI governance • Ensuring system • Assess fairness
fairness and data quality
team
• Private maintenance
• Minimisation of the
• Data privacy
officers use of sensitive data • Review procedures
for sensitive data
• Usage of
handling
representative
datasets
• Ensuring the quality
and correctness of
data annotations
Data labelling • Fair and impartial • AI governance • Setting clear goals • Evaluate diversity
for diversity and and bias mitigation
team
• Private inclusion
• Data privacy • Review testing
• Countering various procedures with
officer
sources of bias diverse user groups
• Testing the AI
system with diverse
user groups
Model training • Robust and reliable • AI/ML engineers • Quality Assurance • Validate quality
assurance
• Monitor the feedback
• Risk and and feedback
to the system
compliance monitoring
functions • Implementation of
• Review failover
failover mechanisms
mechanisms and
• Optimisation of the stress testing
model’s inference implementation
speed
• Review the
• Proper integration documentation of
with data sources and the training process
other AI systems for transparency
and reproducibility
• Implementation of ML
Ops • Verify the
adherence to legal
• Usage of risk-based
and compliance
stress testing
requirements
techniques
during the model
training
14
14
AI Risk ManaTgheem ‘wehnat t|, wRihsky ,m aintidga htioown ’" onfo twh"e a UndP Is ptraaytemgeicn itnss iegchotss y"snteexmt"
Trustworthy AI
AI lifecycle stage Associated persona Principles Audit focus
element
Model deployment • Safe and secure • AI developers • Security of users’ data • Review security
protocols and data
• Adequate controls to
• Cybersecurity safety measures
prevent the possibility
experts
of a malicious attack • Validate measures
for preventing
• Ensuring the safety
attacks
and security of all the
stakeholders • Assess on-device
processing
• Usage of on-device
implementation
processing whenever
possible
Model monitoring • Robust and reliable • AI/ML engineers • Live monitoring in • Assess the efficacy
production to ensure of live monitoring
• Independent AI that the AI system is and diagnostic
assurance and operational capabilities
audit team
• Ability to trace, • Verify the
diagnose and existence and
rollback, if necessary, effectiveness of
in case of a failure disaster recovery
and business
• Disaster recovery
continuity plans
and business
continuity plans
• Resiliency of AI
systems
15
TAhI eR i‘swkh Mata, nwahgye,m aenndt h |o Rwis’ ko fm tihtieg aUtiPoIn p "anyomwe" nantsd e sctorastyesgtiec minsights "next"
Need for governance structure across the AI lifecycle
To ensure AI development and deployments, it is essential to follow the ethical principles defined by the enterprise AI policy. A
governance structure at various levels ensures that AI systems are developed, deployed, and maintained responsibly, ethically, and
transparently. Following is a basic outline of an AI governance structure:
AI tracking Risk assessment and
Alignment with the Trustworthy AI measurement
framework, establishing clear role Methods to measure and
assignments and responsibilities, and understand AI risks using
outlining essential life-cycle criteria. This numbers and information, e.g.,
helps maintain uniformity across the metrics design and monitoring,
organisation, along with an updated risk categorisation, meaningful
inventory that includes attributes for the reporting, and analytics.
risk management programme.
Lifecycle standards Regulatory
Well-defined rules, tools, and Ability to adjust according to various
technology are needed to implement regulations set by different regulators in
the AI policy at every stage. They can different countries/regions. When it makes
be changed to fit different situations, sense, these adjustments should be added
such as using AI from other sources gradually to the existing programmes to
or creating new AI. This way, distinct manage risks related to models, data,
functions of the company can adjust cybersecurity, and legal matters.
requirements as needed.
In India, we do not have any regulations on AI for the development, classification, and use of non-personal and personal data in the
public domain.
In the recent B20 summit (G20 Business Forum) in India, the B20 task force recommended setting up a regulatory framework for
responsible AI, and the Indian government called for a global AI framework to promote the ethical development of AI.
Below are a few key considerations for setting up an effective governance structure for AI that could mobilise the people for
AI governance.
• Define goals and articulate objectives.
• Set up an ethics statement.
• Establish guardrails to guide, monitor, and assess AI solutions. For example, embedded controls in the AI model could prevent
specific actions from being completed.
• Define roles and responsibilities for the people responsible for the governance, development, deployment, management, and
monitoring.
• Set up an inventory of AI models and procedures for tracking and maintaining AI implementations.
• Create role-specific upskilling of stakeholders and employees to guide on AI solutions and their responsible development and
deployment.
• Define or optimise the existing data governance for the data.
• Develop KPIs to evaluate the AI models' performance.
16
16
AI Risk ManaTgheem ‘wehnat t|, wRihsky ,m aintidga htioown ’" onfo twh"e a UndP Is ptraaytemgeicn itnss iegchotss y"snteexmt"
Way forward
Maintaining trust in AI necessitates continuous monitoring of AI models to ensure they function as intended and align with trust criteria.
This is particularly challenging with opaque AI models.
Adequate awareness of AI Risk Management across the entire AI lifecycle and relevant stakeholders along with leveraging AI Risk
Management solutions to assess and validate model performance can restore balance in transparency and accuracy.
Beyond model evaluation, AI data management, privacy, cybersecurity, and post-deployment monitoring also benefit from such
solutions. These tech-enabled assessments enhance AI evaluations, fostering better governance and understanding of model
performance for comprehensive AI management.
17
TAhI eR i‘swkh Mata, nwahgye,m aenndt h |o Rwis’ ko fm tihtieg aUtiPoIn p "anyomwe" nantsd e sctorastyesgtiec minsights "next"
Connect with us
Anthony Crasto Peeyush Vaish
President, Risk Advisory Partner, Risk Advisory
Deloitte India Deloitte India
[email protected] [email protected]
Nitin Naredi Samanth Aswani
Partner, Risk Advisory Partner, Risk Advisory
Deloitte India Deloitte India
[email protected] [email protected]
Key contributors
Manish Dayma Adarsh Mishra
Bharath Yellapu Sachin Arora
Acknowledgment
Akshay Dalvi Neha Kumari
18
18
19
Deloitte refers to one or more of Deloitte Touche Tohmatsu Limited (“DTTL”),
its global network of member firms, and their related entities (collectively, the
“Deloitte organization”). DTTL (also referred to as “Deloitte Global”) and each
of its member firms and related entities are legally separate and independent
entities, which cannot obligate or bind each other in respect of third parties.
DTTL and each DTTL member firm and related entity is liable only for its own
acts and omissions, and not those of each other. DTTL does not provide
services to clients. Please see www.deloitte.com/about to learn more.
Deloitte Asia Pacific Limited is a company limited by guarantee and a member
firm of DTTL. Members of Deloitte Asia Pacific Limited and their related
entities, each of which is a separate and independent legal entity, provide
services from more than 100 cities across the region, including Auckland,
Bangkok, Beijing, Bengaluru, Hanoi, Hong Kong, Jakarta, Kuala Lumpur, Manila,
Melbourne, Mumbai, New Delhi, Osaka, Seoul, Shanghai, Singapore, Sydney,
Taipei and Tokyo.
This communication contains general information only, and none of DTTL, its
global network of member firms or their related entities is, by means of this
communication, rendering professional advice or services. Before making any
decision or taking any action that may affect your finances or your business,
you should consult a qualified professional adviser.
No representations, warranties or undertakings (express or implied) are given
as to the accuracy or completeness of the information in this communication,
and none of DTTL, its member firms, related entities, employees or agents shall
be liable or responsible for any loss or damage whatsoever arising directly or
indirectly in connection with any person relying on this communication.
© 2024 Deloitte Touche Tohmatsu India LLP. Member of Deloitte Touche
Tohmatsu Limited |
170 | deloitte | DI_CGI-state-of-ai-for-gov.pdf | A report from the
Deloitte AI Institute for Government and
the Deloitte Center for Government Insights
Scaling AI in government
How to reach the heights of enterprisewide adoption of AI
About the Deloitte AI Institute
The Deloitte AI Institute helps organizations connect all the different dimensions of the robust, highly
dynamic, and rapidly evolving AI ecosystem. The AI Institute leads conversations on applied AI
innovation across industries, using cutting-edge insights to promote human-machine collaboration in
the Age of WithTM.
The Deloitte AI Institute aims to promote dialogue about and development of artificial intelligence,
stimulate innovation, and examine challenges to AI implementation and ways to address them. The AI
Institute collaborates with an ecosystem composed of academic research groups, start-ups,
entrepreneurs, innovators, mature AI product leaders, and AI visionaries to explore key areas of
artificial intelligence including risks, policies, ethics, future of work and talent, and applied AI use cases.
Combined with Deloitte’s deep knowledge and experience in artificial intelligence applications, the
institute helps make sense of this complex ecosystem and, as a result, delivers impactful perspectives
to help organizations succeed by making informed AI decisions.
No matter what stage of the AI journey you’re in—whether you’re a board member or C-suite leader
driving strategy for your organization, or a hands-on data scientist bringing an AI strategy to life—the
Deloitte AI institute can help you learn more about how enterprises across the world are leveraging AI
for a competitive advantage. Visit us at the Deloitte AI Institute for the full body of our work, subscribe
to our podcasts and newsletter, and join us at our meetups and live events. Let’s explore the future of
AI together. Learn more.
About the Deloitte Center for Government Insights
The Deloitte Center for Government Insights shares inspiring stories of government innovation, looking
at what’s behind the adoption of new technologies and management practices. We produce cutting-
edge research that guides public officials without burying them in jargon and minutiae, crystalizing
essential insights in an easy-to-absorb format. Through research, forums, and immersive workshops,
our goal is to provide public officials, policy professionals, and members of the media with fresh
insights that advance an understanding of what is possible in government transformation.
Connect
To learn more please visit www.deloitte.com/us/cir.
Contents
The power of AI at scale 2
Government organizations are making a strong start on AI 3
Progress can be stalled by an overreliance on pilots 5
Following the path of the trailblazers 9
Appendix—Respondent profile 14
Scaling AI in government
The power of AI at scale
ONE OF THE few bright spots to emerge primed to play an important role in the future of
from the difficult period of the COVID-19 government. To reap the transformative benefits of
pandemic has been the rapid development AI, the technology needs to be scaled and our
of an entirely new class of drug: the messenger global survey of 500 government leaders shows
RNA–based vaccine. While research into mRNA three key findings for organizations looking to
vaccines was not new, the pace with which multiple adopt AI at scale:
companies were able to use that approach to tackle
a new pathogen opens new doors into treating • Government organizations have made a strong
everything from other viruses to cancer. These start in exploring a wide variety of AI proofs
vaccines were not just the product of human genius of concept.
and resources; artificial intelligence (AI) also
played a key role. • The transformational benefits of AI require
adoption of AI at scales much larger than proofs
AI helped to identify potential molecular “targets” of concept.
on the virus where vaccines might act.1 As
researchers homed in on mRNA as a tool, AI • To move from pilots to at-scale AI,
helped to optimize the mRNA sequences for organizations need to not just adopt the
efficacy and ease of manufacture.2 Once vaccines technology, but to adapt their organizations
were developed, AI continued to help by predicting across six key dimensions.
the spread of the virus to help with testing.3 The
story of mRNA vaccines is a success story of Those organizational changes will help to drive AI
collaboration between government and industry from the fringes of an organization into the heart
that shows the world-transforming power of AI of the mission. There AI can bring its
when used at scale. transformational power to bear to improve the
lives of citizens.
Given the important mission and large data stores
in government organizations at every level, AI is
2
How to reach the heights of enterprisewide adoption of AI
Government organizations are
making a strong start on AI
THE TRANSFORMATIONAL POTENTIAL of AI With enthusiasm and a growing pool of resources,
is not lost on organizations at every level of many government organizations have launched
government. For example, in our recent pilots to explore how AI can help their
survey of government leaders, respondents at the organizations. Government organizations are
national, state, and local level all saw AI as exploring a range of AI use cases from speech
important to future mission outcomes (figure 1). recognition to predictive maintenance.
The fact that governments are serious about AI Government sectors such as defense and health
adoption is also reflected in the increasing share of that have a long history of AI experimentation are
AI investments—84% of agencies believe their AI among the leaders in fields such as responsible AI
investments will increase by 6% or more in the and data-sharing. For example, more than 35
next fiscal year.4 With budget analysis showing that countries have released AI strategies that include a
US Federal funding for AI research and focus on responsible AI—a finding backed by our
development alone is expected to have already respondents. Eighty-five percent of surveyed
grown by nearly 50% to more than US$6 billion in government executives indicated their organization
FY 2021, government leaders are clearly bullish on had an enterprisewide AI strategy.6
AI.5 As a result, they are making significant
investments and exploring new AI projects.
FIGURE 1
AI is important for mission outcomes across all levels of government over the
next five years
AI important for mission AI not important for mission
3%
8%
16%
Federal State Local
92% 97% 84%
Source: Deloitte's State of AI survey.
Deloitte Insights | deloitte.com/insights
3
Scaling AI in government
However, there is also a weakness in this pattern of (figure 2). This means that despite the significant
pursuing AI. While government respondents are effort and attention that government
exploring a wide range of AI use cases, they are organizations are paying to AI, most projects
fully deploying only a small fraction of them remain at the pilot scale.
FIGURE 2
Government organizations are pursuing a wide range of AI techniques, but most
of those efforts are developing rather than fully deployed
Percentage of respondents who are developing/have deployed each use case
Deployed Developing
Recommendations
31%
48%
Predictive maintenance
31%
4%7
Computer vision
29%
46%
NLP/NLG
26%
49%
Speech recognition
32%
41%
Biometrics
30%
43%
Pattern/anomaly detection
31%
41%
RPA
24%
47%
Sentiment detection
23%
45%
Intelligent robotics
25%
40%
Source: Deloitte's State of AI survey.
Deloitte Insights | deloitte.com/insights
4
How to reach the heights of enterprisewide adoption of AI
Progress can be stalled by
an overreliance on pilots
WHILE PILOTS PLAY a critical role in developing AI mostly through pilots and
developing successful AI, an exploration may be holding back
overreliance on them can be detrimental. further development.
To try to understand how, we analyzed the
reported capabilities and actions of respondent If organizations only pursue pilots, it can create a
organizations to evaluate how prepared they were sense of overconfidence. As small-scale pilot
for AI at scale. The analysis showed that while projects succeed, organizations may mistakenly
there are a significant number of mature think that they have all the capabilities they need
government organizations blazing a trail in AI to tackle AI at scale. We observed signs of this
(28%), a near majority are still beginners (48%, see overconfidence in our survey results. Seventy-three
figure 3). Being a beginner in AI is not necessarily a percent of government respondents believe that
problem. Even most trailblazing organizations they are ahead of the private sector in AI
were beginners at one point. The problem many capabilities. And as if to reinforce the optimism
governments face is that their pattern of
FIGURE 3
Despite experience with pilots, most government organizations are still
beginners in the journey to AI at scale
Trailblazers
High strategy/governance 28% High capabilities
(Three AI readiness
dimensions:
strategy, process, ethics)
Strategists Techies
17% 8%
Low capabilities Beginners Low strategy/governance
48%
(Three AI readiness dimensions:
data, technology and
platforms, people)
Source: Deloitte's State of AI survey.
Deloitte Insights | deloitte.com/insights
5
Scaling AI in government
bias, 80% believed they are also ahead of their CDO of a large US city describes initially being
public sector peers. surprised at the slow pace of AI development
among peers in the private sector. Only later did
the CDO begin to realize that the slower pace may
Pilot purgatory
be needed to tackle larger AI projects. The limited
scope of pilots may make them easier to pursue
The problem is that AI at scale requires different more quickly, but for larger-scale projects it takes
organizational capabilities than pilots or proofs of time to make sure that the right data is gathered,
concept. Pilots are typically smaller and narrower the appropriate use case is chosen, and costly
in focus than full-scale AI efforts. As a result, pilots mistakes are not made while developing
can often make use of different technologies and technological architecture. For those just starting
data sources than would be required for full-scale out in their AI journey, it can seem counterintuitive
use. They may not need to meet as rigorous that slowing down the process may be a way to
security and privacy requirements. Further, the achieving AI at scale quickly. Slow is smooth;
smaller scope of pilots means that they touch fewer smooth is fast.7
parts of an organization so that change
management is less of a factor in their success. In short, organizations that have only
experimented with pilot-scale AI cannot make it to
For these reasons, development of AI at scale just the heights of at-scale AI simply by doing more of
looks different than pilots. For example, a former what they are doing. Without intentional action to
acquire the organizational capabilities needed for
at-scale AI, organizations can easily become stuck
in “pilot purgatory” continually cycling through
promising AI pilots but never realizing the
transformational benefit that AI promises for their
core mission.
Adapt, don’t just adopt
The good news is that government leaders appear
to be increasingly aware of the gap between pilots
and at-scale AI. The respondents of our survey
again and again highlighted the gap between their
goals for AI and where they currently assessed
their AI capabilities (figure 4).
The US Department of Defense (DoD) is just one
example of the path leading government
organizations are taking to scale AI. In its 2018 AI
strategy, DoD outlined that, “The DoD will identify
and implement new organizational approaches,
establish key AI building blocks and standards,
develop and attract AI talent, and introduce new
operational models that will enable DoD to take
6
How to reach the heights of enterprisewide adoption of AI
FIGURE 4
Government leaders are aware of the gap between current and desired state
of AI capabilities
Strategy Ethics Process
91% No.1
91% of agencies believe AI the No.1 goal for AI reported by
...but
will be important to deliver respondents is “making internal
mission outcomes over the processes efficient” and not
next five years applying AI to the mission
72% 44%
72% of respondents say ...but 44% of respondents also
their organization is say that AI has negatively
prepared to deal with impacted the reputation of
issues related to ethical AI their organization
Top 3 33%
”Documented and enforced AI ...but only 33% of organizations
ops and governance follow documented
procedures” is among the top 3 MLOps procedures when
Strategy and governance
critical factors for successful developing an AI solution
AI implementations
State of AI in government
People Data Technology and platforms
78% 50%
78% of agencies say they 50% of respondents also cite a
...but
have sufficient skills to lack of skills as a major barrier
implement AI initiatives to taking advantage of AI
89% 41%
89% of public sector ...but only 41% of agencies consider
organizations say they public cloud as a data platform
have access to all neces- for various applications
sary data for AI
No.1 28%
”Lack of technology supporting ...but only 28% of respondents fully
AI” is the No.1 barrier for scale AI applications
agencies to take advantage of
Capabilities
AI
Source: Deloitte's State of AI survey.
Deloitte Insights | deloitte.com/insights
7
Scaling AI in government
advantage of AI systematically at enterprise • People. Agencies may face challenges around
scale.” 8 Since then, DoD has established the Joint accessing and recruiting necessary technical
Artificial Intelligence Center (JAIC) to better skills, as well as helping existing employees
govern AI use cases, set up the develop and deploy AI skills.
Joint Common Foundation (JCF) to provide ready-
to-use tools to experiment and scale AI use cases, • Process. AI can be a powerful new tool, but
and started offering new AI career paths to attract simply embedding it within existing business
and retain talent.9 processes designed for older tools will limit
its benefits.
While organizations like the US DoD are at the
forefront of AI in government, other organizations • Data. AI is only as good as the data upon which
may find it hard to replicate organizational change it is built, and its appetite for data is voracious.
of that nature. It is comparatively easy to adopt a
technology and graft it onto existing organizational • Ethics. While any technology’s deployment
structure and business processes. But it is much should be ethical, AI brings issues such as
harder to adapt the organization to allow it to take transparency, privacy, and bias into
full advantage of a new technology. To build the particular focus.
organizational capabilities needed for AI at scale,
organizations need to adapt their: • Technology and platforms. A variety of
models for pursuing AI exist that vary in terms
• Strategy. Because AI is a transformative of platforms and ownership of technology (e.g.,
technology, alignment on direction and level of internal or in partnership), but, in all cases, AI
ambition is crucial. requires a coherent approach that considers
future requirements as AI scales within the
organization and its usage evolves.
8
How to reach the heights of enterprisewide adoption of AI
Following the path of
the trailblazers
TRAILBLAZING GOVERNMENT
Organizations where senior leaders
ORGANIZATIONS such as many in the
communicate a clear vision for AI
defense and health sectors have already
are 50% more likely to achieve their
charted the way toward developing these six
organizational capabilities. Following their lead desired outcomes with AI.
can help other organizations to iteratively build
capabilities across those six dimensions and realize Drive AI into the heart of the mission: AI
the transformational benefits of AI at scale. should be about doing more and doing better.
However, our analysis found that organizations
that are just beginning their AI journey are more
Strategy
likely to use AI merely to improve internal
efficiency. As organizations gain experience and
Senior leaders should ensure AI strategy become more mature, they are more likely to use
supports the mission: The focus of an AI for mission-focused goals such as improving
organization’s AI strategy should not be merely to collaboration or creating new programs. In one
deploy AI for its own sake but rather should focus large-scale example, Singapore created a US$73
on how AI can be an enabler to deliver the million AI-enabled digital twin of the city, not to
organization’s mission outcomes. This means that make government more efficient, but to model
an organization’s AI strategy cannot be a product decision-making, experiment with service
purely of IT or technical teams but must be driven provision, and address some of the most pressing
by senior leaders. Our survey found that challenges facing the country.12
organizations where senior leaders communicate a
clear vision for AI are 50% more likely to achieve
As organizations gain experience
their desired outcomes with AI.10 In the early
and become more mature, they are
2010s, Jeff Bezos mandated that every leader
across Amazon develop a plan for how to use AI in more likely to use AI for mission-
their division. That mandate was instrumental in focused goals.
Amazon’s rise to become an AI leader today.11
9
Scaling AI in government
People Process
Balance outside hiring with reskilling: Our Reimagine processes and career paths: For
survey found that 69% of respondents would prefer government to truly revolutionize the lives of
to bring in new hires with required skill sets. Given citizens using AI, it will have to revolutionize the
the widespread shortage of AI talent,13 agencies way AI is deployed in its business processes and
should balance outside hiring with reskilling their workflows. After all, you cannot deliver new results
existing workforce. For example, both Denver and with old processes. Organizations that have
San Francisco city governments have established significantly changed workflows are 36% more
data academies to help train city workers and likely to achieve desired outcomes from their AI
others in the basic skills needed to harness AI.14 projects.17 Introducing new processes can also help
The National Security Commission on Artificial organizations create new career paths for workers
Intelligence (NCSAI) goes a step further, calling for who work with this technology, which can be a
establishing a digital service academy, modeled critical enabler to success.18 We found that agencies
after US service academies, to produce a trained that added new AI roles are 60% more likely to
workforce that caters to all federal agencies.15 achieve desired outcomes.19 While adding new
roles can help organizations, those benefits may be
Building technical skills is a clear benefit to temporary unless organizations can provide new
technical staff but can also help the wider career pathways for talent to grow and develop.
organization. Government will always need AI This is exactly what the Australian Public Service
specialists, but to adopt AI at scale, it should also and the country’s Digital Transformation Agency
improve data literacy for the workers who must collaborated on, defining over 150 new digital roles,
buy AI tools and services or use AI to deliver and creating the APS Career Pathfinder tool to help
services to citizens. For example, Abu Dhabi has people in those roles explore digital career options
created AI training workshops to help government in government.20
employees understand AI’s benefits and make
better decisions around its utility.16
Organizations that have significantly
changed workflows and added
new AI roles are 36% and 60%,
respectively, more likely to achieve
desired outcomes from their AI
projects.
10
How to reach the heights of enterprisewide adoption of AI
Data
Documenting and enforcing MLOps
makes organizations twice as likely
Identify relevant data and determine its
to achieve goals and three times
accessibility: Agencies that have access to the
more likely to be prepared for AI
necessary data are twice as likely to exceed
risks.
expectations in their AI initiatives.21 To make the
best use of AI, agencies need to identify relevant
Prioritize change management. If AI is to be
datasets and develop platforms to access that data.
successful, it will, by definition, be disruptive for
For instance, the US Air Force has adopted the
government organizations. AI can change not only
VAULT data platform which gives airmen access to
how processes are done, but even what services
the cloud-based data and tools they need to use AI
government delivers to its citizens. Our analysis
to improve readiness and mission success.22
indicates organizations that invest in change
management are 48% more likely to report that
Agencies that have access to the AI initiatives exceed expectations.25 However, the
more significant the change brought by AI, the
necessary data are twice as likely
more difficult it can be. Governments should
to exceed expectations in their AI
use the principles of behavioral economics to
initiatives.
understand the human impact of transformations
and how to provide appropriate support to
encourage change.26
Ethics
Organizations that invest in change
Document and enforce MLOps: Developing
management are 48% more likely
and deploying AI is not without ethical risks. That
to report that AI initiatives exceed
is why having clear documentation and enforceable
processes is important to having trustworthy and expectations.
transparent AI. This is where MLOps—the set of
automated pipelines, processes, and tools that
Technology and platforms
streamline all steps of AI model construction—can
help. After all, it is difficult to address ethical issues
with a model unless you know how that model was Build a diverse ecosystem: Every government
built and operated. In fact, our survey found that agency does not need to solve every problem itself.
documenting and enforcing MLOps makes From chatbots to speech-to-text, many solutions to
organizations twice as likely to achieve goals and technical problems already exist. Tapping into
three times more likely to be prepared for AI risks.23 other entities that have existing technical solutions
Organizations like the Internal Revenue Service or solved organizational challenges can accelerate
(IRS) have discovered that scaling AI beyond the progress toward AI at scale. In fact, our survey
pilot stage across the agency requires adopting found that continually cultivating a wide range of
different and rigorous processes for creating and relationships with industry, academia, and other
managing AI models.24 agencies dramatically improves the likelihood that
11
Scaling AI in government
an organization has what it needs to scale AI on where government needs help (figure 6).
(figure 5). As Eileen Vidrine, chief data officer at Partners don’t always need to be organizations at
the US Air Force says: “It’s really about working all. The City of LA’s Data Angels program brought
together, building collaborative, trusted volunteer data scientists into government on a
FIGURE 5
Agencies with diverse ecosystems are more likely to have what they need to
achieve their goals for AI
Diverse ecosystem Narrow ecosystem
81%
Sufficient skills
64%
Use AI to improve both front-end 69%
and back-end operations 36%
AI initiatives exceeded 39%
expectations 8%
Achieve outcomes set 37%
for AI initiatives 22%
Percentage of 33%
trailblazer organizations 9%
Source: Deloitte's State of AI survey.
Deloitte Insights | deloitte.com/insights
partnerships. It needs to be part of the part-time basis to help with a variety of tasks.27
conversation at the beginning and through the The program tapped into private sector data
whole life cycle about trying to optimize specialists who wanted to help the community
interoperability and avoiding what I would call while still retaining their jobs, bringing some of the
‘vendor lock’ as much as possible.” top data talent into public service with little cost to
the government.
Find partners that complement your need:
Find partners that provide the capabilities your AI is the future. Government leaders clearly
particular agency lacks. These partners should be a understand this. But getting to that future can be
wide variety of different organizations depending more difficult and more rewarding than it may
12
How to reach the heights of enterprisewide adoption of AI
FIGURE 6
Governments are partnering with a variety of players depending on their
unique needs
IT analyst
49%
Professional services/consulting
37%
Cloud vendors/hyperscalers
37%
Traditional IT firms
30%
Startups/boutique software providers
25%
Source: Deloitte's State of AI survey.
Deloitte Insights | deloitte.com/insights
seem at the start. Taking a realistic view of the right strategies, and the right governance to make
challenges inherent in developing AI at scale can sure that the AI of the future serves the citizens of
help government develop the right capabilities, the the future.
13
Scaling AI in government
Appendix—Respondent profile
FIGURE 7
Appendix: Respondent profile
SECTOR TITLE
9%
16%
9%
32% 31%
9%
13%
8%
12% 19% 19%
23%
Federal/central government—Civilian C-suite
Federal/central government—Defense Deputy or other top-level executive but below C-suite
Federal/central government—Health Deputy secretary/deputy agency head
Higher education Local government Manager level
State/provincial government Secretary/undersecretary/agency head Others
REGION
12%
19%
69%
Americas Europe APAC
Note: N = 517 respondents.
Source: Deloitte's State of AI survey.
Deloitte Insights | deloitte.com/insights
14
How to reach the heights of enterprisewide adoption of AI
Endnotes
1. Gunjan Arora et al., “Artificial intelligence in surveillance, diagnosis, drug discovery and vaccine development
against COVID-19,” Pathogens 10, 8 (2021): p. 1048.
2. MIT Sloan Management Review, “AI and the COVID-19 vaccine: Moderna’s Dave Johnson,” Me, Myself, and AI,
podcast, July 13, 2021.
3. Terri Park, “Behind COVID-19 vaccine development,” MIT News, May 18, 2021.
4. Based on analysis of government respondents of 2021 State of AI Survey.
5. Jon Harper, “Federal AI spending to top US$6 billion,” National Defense Magazine, February 10, 2021.
6. Future of Life Institute, “National and international AI strategies,” accessed December 6, 2021.
7. Phone interview with Sari Ladin-Sienne on September 15, 2021.
8. William D. Eggers et al., Crafting an AI strategy for government leaders: Does your agency have a holistic
AI strategy?, Deloitte Insights, December 4, 2019.
9. Ibid.
10. Based on analysis of government respondents of 2021 State of AI Survey. Respondents whose senior leaders
communicated a clear AI vision (78%) were 26 percentage points more likely in achieving desired AI outcomes
than respondents whose senior leaders did not communicate a clear AI vision (52%).
11. Steven Levy, “Inside Amazon’s artificial intelligence flywheel,” Wired, February 1, 2018.
12. Joe Mariani, Adam Routh, and Allan V. Cook, Convergence of technology in government: Power of AI, digital
reality, and digital twin, Deloitte Insights, March 11, 2020.
13. Joe McKendrick, “Artificial intelligence skills shortages reemerge from hiatus,” ZDNet, October 22, 2020.
14. Brian Elms, Peak Performance: How Denver’s Peak Academy is saving millions of dollars, boosting morale and just
maybe changing the world. (And how you can, too!), Washington DC: Governing, 2016, Data@SF, “Data Academy,”
accessed October 8, 2021.
15. National Security Commission on Artificial Intelligence, “Chapter 6: Technical talent in government,” accessed
October 8, 2021.
16. Eggers et al., Crafting an AI strategy for government leaders.
17. Based on analysis of government respondents of 2021 State of AI Survey. Respondents whose organizations
have significantly changed workflows (76%) were 20 percentage points more likely in achieving desired AI
outcomes than other respondents (56%).
1188.. NSCAI, “Chapter 6: Technical talent in government”; NSCAI, Final report, March 2021.
19. Based on analysis of government respondents of 2021 State of AI Survey. Respondents in whose organizations
new AI job roles/ functions were created (78%) were 28 percentage points more likely in achieving desired AI
outcomes than respondents in whose organizations no new AI roles were created (46%).
20. Digital Transformation Agency, “New APS career pathfinder tool,” October 19, 2020.
21. Based on analysis of government respondents of 2021 State of AI Survey.
15
Scaling AI in government
22. Eileen Vidrin, “Air Force CDO: Flying High With AI,” CIO Journal, August 20 2021; Secretary of the Air Force Public
Affairs, “Chief Data Office announces capabilities for the VAULT data platform,” US Air Force, October 11, 2019.
23. Based on analysis of government respondents of 2021 State of AI Survey.
24. Austin Price, Ashley Prusak, and Maria Wright, Tech Trends 2021: Peering through the lens of government,
Deloitte Insights, accessed October 8, 2021.
25. Based on analysis of government respondents of 2021 State of AI Survey.
26. William D. Eggers et al., Behavior-first government transformation: Putting the people before the process, Deloitte
Insights, August 25, 2020.
27. Harsha Mallajosyula, “Get to know a data angel,” Medium, February 15, 2019.
Acknowledgments
The authors would like to thank Thirumalai Kannan for his invaluable quantitative analysis of the
survey data. Kannan’s expertise was central to teasing out the signal from the noise in the data.
16
How to reach the heights of enterprisewide adoption of AI
About the authors
Edward Van Buren | [email protected]
Edward Van Buren is the Strategic Growth leader—Artificial Intelligence (AI) for Deloitte Consulting
LLP’s Government & Public Services (GPS) Industry and the executive director of the Deloitte AI Institute
for Government. He works with technology companies and other strategic partners to develop
solutions harnessing the power of AI/ML for federal, state, local, and higher education clients. Van
Buren has more than 25 years of experience in consulting and the public sector. He has served diverse
clients such as the United States Postal Service, Internal Revenue Service, Office of Performance
Management, and the United States military, helping them transform their organizations to better
execute missions and utilize technologies.
William Eggers | [email protected]
William Eggers is the executive director of Deloitte’s Center for Government Insights, where he is
responsible for the firm’s public sector thought leadership. His most recent book is Delivering on Digital:
The Innovators and Technologies That Are Transforming Government (Deloitte University Press, 2016). His
other books include The Solution Revolution, the Washington Post bestseller If We Can Put a Man on the
Moon, and Governing by Network. He coined the term “Government 2.0” in a book by the same name.
His commentary has appeared in dozens of major media outlets including New York Times, Wall Street
Journal, and Washington Post.
Tasha Austin | [email protected]
Tasha Austin is a Principal in Deloitte’s Risk and Financial Advisory business and has more than 22
years of professional services experience involving commercial and federal financial statement audits,
fraud, dispute analysis and investigations, artificial intelligence and advanced data analytics. Tasha
serves as the Director of Deloitte’s Artificial Intelligence Institute for Government and is a leader in
Deloitte’s Artificial Intelligence and data analytics offering where she focuses on amplifying Deloitte’s
capabilities and services in key areas such as trustworthy/ethical AI, provides insight-driven solutions to
her clients, and is responsible for elevating Deloitte’s thought leadership and digital presence in AI to
the federal market. Tasha also leads Deloitte’s strategic firm-wide engagement initiatives with HBCUs.
She has a passion for bridging the data analytics and digital divide in under-resourced communities
and working with non-profit organizations to deliver and scale solutions that help advance equity and
promote social justice.
Joe Mariani | [email protected]
Joe Mariani is a research senior manager with Deloitte’s Center for Government Insights. His research
focuses on innovation and technology adoption for government organizations. His previous work
includes experience as a consultant to the defense and intelligence industries, high school science
teacher, and Marine Corps intelligence officer |
171 | deloitte | Deloitte-Insights-Magazine-Issue-33.pdf | Contents Deloitte INSIGHTS
Magazine 33
04 On the web / 07 Editor’s letter / 08 Contributors / 90 The end note
1. Data points
Bite-size insights from Deloitte research
18 Gen AI investments increasingly extend beyond the AI itself
19 While business leaders look inward for AI’s impact, tech leaders look outward
20 Few AI regulations across the globe address the outcomes rather than the tech
21 European organizations’ gen AI preparedness has increased, but few feel ready for the associated risks
22 A burgeoning ‘AI-generated’ market: Insurance safeguards against AI risk
23 More hands-on gen AI experience increases optimism—and caution—for millennials and Gen Z
24 Many tech leaders’ influence in the C-suite is growing, new Deloitte research suggests
25 Are new generative AI features in software a monetizable enhancement or table stakes?
26 A snapshot of AI adoption: Italy’s design sector
27 More US consumers think AI-generated health information should be left to the experts
2. Perspectives 3. Features
30 Better questions about generative AI 52 Predicting the unpredictable: Exploring how
Four scholars share critical questions leaders should ask technology could change the future of work
about generative AI, from concerns about bias to existential What does the future hold for worker and AI collaboration?
considerations about human values It depends less on the tech and more on the decisions we make
along the way.
34 Generative AI and the labor market:
A case for techno-optimism 60 Generative AI in Asia Pacific: Young employees
Generative AI can boost productivity and enhance the lead as employers play catch-up
labor market, yet it remains to be seen if everyone can reap A survey of more than 11,900 employees and students across
its many benefits the region finds that gen AI is already affecting 11 billion work
hours per week, but many employers likely aren’t optimizing
38 The more AI-enabled work becomes, that impact
the more important human imagination is
One of the most valuable skills you need to succeed in an 70 Designing for growth in the C-suite
AI-enabled working world you likely learned in kindergarten An analysis of over 46,000 job postings reveals the most
in-demand skills for C-suite roles like CFOs, COOs, and other
42 The democratization of deepfake technology brings executive leaders
new perils for business
A chief executive of a deepfake detection platform company and 78 Generative AI and government work:
Deloitte US’s chief futurist explore the growing deepfake risks, An analysis of 19,000 tasks
as well as mitigation strategies that can help organizations fight Deloitte US’s analysis reveals three criteria that can help
AI-enabled fraud with AI determine which tasks could be assigned to generative AI tools
and when different occupations could feel pressure to adopt them
46 Organizations talk about equity in AI, but are they
following through?
Diversity, equity, and inclusion leaders are in a unique position to
advocate for AI that works for everyone. Here’s where they have
opportunities to lead at the intersection of AI and DEI.
Issue 33 1
Masthead
Deloitte Insights Magazine
EXECUTIVE ADVISOR EDITORIAL
Rod Sides Aditi Rao (team lead, US and India)
Annalyn Kurtz (team lead, US and global)
PUBLISHER Richard Horton (team lead, Europe)
Jeff Pundyk Jennifer Wright (team lead, Asia Pacific)
Andy Bayiates
EDITOR IN CHIEF Rupesh Bhat
Elisabeth Sullivan Cintia Cheong
Corrie Commisso
ART DIRECTOR Pubali Dey
Matt Lennert Karen Edelman
Abrar Khan
CREATIVE Rebecca Knutsen
Sylvia Yoon Chang (team lead) Kavita Majumdar
Jaime Austin Debashree Mandal
Manya Kuzemchenko Sanjukta Mukherjee
Natalie Pfaff Elizabeth Payes
Molly Piersol Arpan Kumar Saha
Sofia Sergi Sara Sikora
Jim Slatton Rithu Thomas
Sonya Vasilieff
Harry Wedel PUBLISHING OPERATIONS
Alexis Werbeck Stacy Wagner-Kinnear
USER EXPERIENCE RESEARCH AND DESIGN PRODUCTION
Denise Weiss (team lead) Blythe Hurley (team lead) CONTACT
Email: [email protected]
Danielle Johnson Hannah Bachman
www.linkedin.com/company/deloitte-insights
Joanie Pearson Prodyut Borah
Sanaa Saifi Preetha Devan
Aparna Prusty
WEB PRODUCTION Shambhavi Shah Unlimited insights. One app.
Melissa O’Brien (team lead) Download the Deloitte Insights app
MULTIMEDIA www.deloitte.com/insights/app
WEB DEVELOPMENT Sarah Jersild
Sourabh Yaduvanshi (team lead)
Megha Priya AUDIENCE INSIGHTS AND
Nitin Gaurav Singh PRODUCT MANAGEMENT
Naveen Bhusare Amy Bergstrom (team lead)
Supreetha R Hannah Rapp Collins
Nishant Raghuwanshi
DIGITAL OPERATIONS
Ekta Dubey AUDIENCE DEVELOPMENT
Turner Roach (team lead)
Deloitte refers to one or more of Deloitte Touche Tohmatsu Limited,
KNOWLEDGE SERVICES Atira Anderson
a UK private company limited by guarantee (“DTTL”), its network of
Joy Kishta (team lead) Pooja Boopathy
member firms, and their related entities. DTTL and each of its
Pruthvi C. Kelly Cherry member firms are legally separate and independent entities. DTTL
Pallavi V Nair Maria Martin Cirujano (also referred to as “Deloitte Global”) does not provide services to
Abhijit Sahu Nikita Garia clients. In the United States, Deloitte refers to one or more of the US
member firms of DTTL, their related entities that operate using the
Roshni Thawani
“Deloitte” name in the United States and their respective affiliates.
Certain services may not be available to attest clients under the
rules and regulations of public accounting.
This publication contains general information only and Deloitte is not, by means of this publication, rendering accounting, business, Please see www.deloitte.com/about to learn more about our global
financial, investment, legal, tax, or other professional advice or services. This publication is not a substitute for such professional advice network of member firms.
or services, nor should it be used as a basis for any decision or action that may affect your business. Before making any decision or taking
any action that may affect your business, you should consult a qualified professional advisor. Deloitte shall not be responsible for any loss Copyright © 2024 Deloitte Development LLC.
sustained by any person who relies on this publication. All rights reserved.
2 Deloitte Insights Magazine
On the web
What Deloitte Insights readers are reading
The important role of leaders in advancing human sustainability
How can organizations work better for working women?
Navigating the tech talent shortage
Gen Zs and millennials find reasons for optimism despite
difficult realities
Households transforming the grid: Distributed energy resources
are key to affordable clean power
Lessons for middle-market tech executives to consider from
their fast-growing peers
www.deloitteinsights.com
4 Deloitte Insights Magazine
Unlimited insights.
One app.
Get personalized industry insights
Stay up to date with industry news
Share and save your favorite articles
Listen to podcasts on the go
Be the first to access the latest issue of
Deloitte Insights Magazine
DOWNLOAD TODAY
SCAN QR CODE
Data-driven discovery at your fingertips
The Deloitte Insights app is designed to help future-
focused leaders navigate what’s next by providing
on-demand access to the latest insights and analysis.
Copyright © 2024 Deloitte Development LLC.
All rights reserved.
Deloitte Insights Magazine
EDITOR’S LETTER
Advancing the AI conversation
Over the past two decades, and then seemingly overnight, artificial intelligence has gone
from a fringe technology to what many consider to be must-have, market-making and
-shaping tech. And with each passing day, the AI conversation is evolving in real time, spurred
on, of course, by all things generative-AI–related and the more readily apparent impact AI could
have on organizations, industries, and economies.
In this issue, we’re featuring some of Deloitte’s latest proprietary research and insights to
help move the AI conversation forward, offering fresh perspectives and foresight on what those
organizational and economic impacts might be.
For instance, Deloitte researchers have sliced and diced a data set from a proprietary survey
of nearly 2,800 board members, C-suite executives, and other senior leaders in 14 countries for
insights into how generative AI budgets are being spent (page 18), what it takes to scale from
gen AI pilots to full implementation (page 90), and whether organizations feel prepared for
any risk and governance issues associated with the technology (page 21).
We look at which success metrics business and tech leaders turn to when determining
the impact of their AI investments (page 19), and which kinds of AI regulations could be
most effective for the ever-evolving technology, safeguarding the public while not hindering
innovation (page 20).
And we examine AI’s potential impact on work and the workforce from several angles—
considering gen AI’s potential impact on productivity and labor demand and, therefore,
economic outcomes (page 34); tracking the trend in C-suite roles requiring more data and
analytics skills (page 70); making the case for the key capability people could need as work
becomes more AI-enabled (page 38); and discussing why new prediction models might be
needed to determine how AI and other tech could change the future of work (page 52).
And we’re just scratching the surface. Deloitte is building a rich and diverse portfolio
of AI-related business research and insights—the kind of trustworthy, deeply researched
information that your AI-enabled searches rely on. Check out www.deloitteinsights.com for
lots more where this came from.
Best,
Elisabeth Sullivan
Editor in chief, Deloitte Insights
[email protected]
Issue 33 7
Contributors
Tasha Austin Andrew Blau
[email protected] [email protected]
Tasha Austin is a principal in Deloitte US’s government and public services Andrew Blau leads Deloitte US’s strategic futures practice, helping clients
practice whose experience includes commercial and federal financial statement think creatively and strategically about how they can successfully compete
audits, fraud, dispute analysis and investigations, artificial intelligence, and despite macro and market uncertainties in the world around them. He also
advanced data analytics. Austin helps lead the practice’s AI and data analytics leads eminence and insights for Deloitte Consulting, developing Deloitte
offering, helping clients with their financial management transformation. US’s perspectives on cross-cutting issues shaping organizations and markets.
Sonia Breeze Susan Cantrell
[email protected] [email protected]
Sonia Breeze leads Deloitte New Zealand’s human capital consulting prac- Susan Cantrell is vice president of products and workforce strategies at
tice and is also the internal talent partner. She’s committed to enabling Deloitte Consulting LLP, and a frequent speaker on human capital and
organizations to maximize the potential of their people. She leverages 25 the future of work. She is coauthor of the Harvard Business Press book
years of experience in human capital consulting and health care, and as a Workforce of One and has also been published widely in publications
CPO, to advise on and implement the people-oriented aspects of change including Harvard Business Review, The Wall Street Journal, and MIT Sloan
including technology enablement. Management Review.
Corrie Commisso Peter Evans-Greenwood
[email protected] [email protected]
Corrie Commisso is a senior editor at Deloitte Insights leading global content Peter Evans-Greenwood is an independent advisor and consultant, a senior
strategy for human capital and the future of work. She has more than 20 years fellow at the Australian Davos Connection, and a former fellow at the
of experience in writing, editing, and creative direction, and holds degrees in Deloitte Centre for the Edge in Australia. With over 20 years of experience
journalism and library and information science—a combination she credits at the intersection of business and technology, he combines systematic and
for her ability to tell deep and engaging stories. integrative thinking to help organizations make informed decisions in an
uncertain business environment.
Robert Hillard Susan C. Hogan
[email protected] [email protected]
Robert Hillard leads consulting across Asia Pacific, including in China, Japan, Susan C. Hogan is the global leader of Deloitte’s finance and performance
Australia, India, Singapore, Malaysia, Indonesia, Thailand, Vietnam, New practice, and the leader of Deloitte US’s finance transformation practice.
Zealand, and South Korea. He has previously served as a member of Deloitte’s Hogan has nearly 30 years of transformation consulting experience, primarily
global board (Deloitte Touche Tohmatsu Limited), CTO for Deloitte Asia in finance and global business services.
Pacific, and the chief strategy and innovation officer for Deloitte Australia.
8 Deloitte Insights Magazine
Stuart Johnston Ira Kalish
[email protected] [email protected]
Stuart Johnston is the Deloitte Global clients leader and a member of the global Ira Kalish is the chief global economist at Deloitte Touche Tohmatsu Ltd.
clients and industries executive. Based in Australia, he also serves as a lead client He is a specialist in global economic issues as well as the effects of economic,
partner and advisory partner, focusing on enhancing relationships with board demographic, and social trends on the global business environment. His
members and executives, and connecting Deloitte member firms globally. insights have appeared in publications including The Wall Street Journal, The
Economist, and The Financial Times.
Thirumalai Kannan Pankaj Kishnani
[email protected] [email protected]
Thirumalai Kannan is a researcher with the Deloitte Center for Government Pankaj Kishnani is a research manager with the Deloitte Center for Government
Insights. His research focuses on the quantitative analysis of cross-cutting Insights. His research focuses on identifying trends in emerging technolo-
issues in government, including emerging technologies like AI and advancing gies in the public sector. He closely tracks digital government transformation,
citizen trust. the regulation of emerging technologies, the role of government in catalyzing
innovations, and citizen-centric service delivery models.
Annalyn Kurtz Chris Lewin
[email protected] [email protected]
Annalyn Kurtz is the executive editor of Deloitte Insights. Prior to her role at Chris Lewin is a partner in the consulting practice at Deloitte Australia with a
Deloitte, she worked as a business journalist, newsroom leader, and educator. specialization in automation. He has extensive data and analytics experience
She has written about business and economics topics for CNN Business, driving IT transformation programs.
Fortune, and The New York Times, among others.
David Mallon Joe Mariani
[email protected] [email protected]
David Mallon is chief analyst and market leader for Deloitte US’s Joe Mariani is a senior research manager with Deloitte’s Center for Government
Insights2Action team, helping clients sense, analyze, and act at the ever- Insights. His research focuses on innovation and technology adoption for both
shifting intersection of work, workforce, workplace, and industry. He national security organizations and commercial businesses. His previous work
brings more than 20 years of experience in human capital, with expertise includes experience as a consultant to the defense and intelligence industries,
in organization design, organizational culture, HR, talent, learning, and a high school science teacher, and a US Marine Corps intelligence officer.
performance.
Issue 33 9
Timothy Murphy Kellie Nuttall
[email protected] [email protected]
Tim Murphy is a senior manager in Deloitte’s Center for Integrated Research. Kellie Nuttall is lead partner for strategy and business design and the
As a researcher and analytical scientist, he focuses on understanding how AI Institute at Deloitte Australia. Nuttall specializes in AI strategy and
organizations undergo large-scale transformations that grow the business, transformation initiatives, working with senior executives to build their AI
bolster operations, and make the enterprise more resilient against external fluency and their understanding of AI technologies. Prior to joining Deloitte,
shocks and disruptions. Nuttall built analytics capabilities within both government and private-sector
organizations.
John O’Mahony Julian Sanders
[email protected] [email protected]
John O’Mahony is a lead partner at Deloitte Australia with over 20 years Julian Sanders is a research lead in Deloitte’s DEI Institute with multi-industry
of professional economics experience. His skills and expertise include experience in education, public policy, program management, and research.
economic impact analyses, economic modelling, and economic policy. His In his current role, he manages and activates diversity, equity, and inclusion
specific industries of interest include telecommunications, media, technology, research, contributing to the institute’s thought leadership.
infrastructure, retail, housing, and manufacturing.
Nic Scoble-Williams Brenna Sniderman
[email protected] [email protected]
Nic Scoble-Williams is a partner at Deloitte Tohmatsu Consulting LLC and Brenna Sniderman leads the Center for Integrated Research, where she
the Asia Pacific leader for the future of work. Based in Japan, and with more oversees cross-industry thought leadership for Deloitte. In this capacity,
than 20 years of cross-industry experience in IT services, talent strategy and Sniderman leads a team of researchers focused on global shifts in digital
advisory, and mergers and acquisitions, Scoble-Williams works with businesses transformation, innovation and growth, climate, and the future of work—
and governments to embed a “future of work” vision into enterprise how organizations can operate and strategize in an age of digital, cultural,
transformation strategies. environmental, and workplace transformation.
Peter Williams Michael Wolf
[email protected] [email protected]
Peter Williams is a retired partner for Deloitte Australia. He served as the chief Michael Wolf is a global economist at Deloitte Touche Tohmatsu Ltd. He
edge officer at Deloitte’s Center for the Edge Australia and the chairman of began his career as an economist at the US Labor Department and has since
Deloitte Australia’s Innovation Council, and was a founder of Deloitte Digital. held economist positions at Moody’s Analytics, Wells Fargo Securities, and
PwC. His insights have been featured in media outlets including The Wall
Street Journal and National Public Radio.
10 Deloitte Insights Magazine
Artists
Jaime Austin Bose Collins
Jaime Austin is an art director at Deloitte Insights, a professional circle Bose Collins is a London-based design agency established in 1994, known for
designer, a design process expert, and an Excel enthusiast. Her design its use of state-of-the-art tools. Its capabilities extend from film direction,
approach is anchored in strategic problem-solving, where she leverages animation, and sound design, to computer graphics, 3D modeling, and AI
creativity and logic to craft beautiful artwork with meaningful narratives. whispering.
Sylvia Chang Matt Lennert
Sylvia Chang is Deloitte Insights’ creative director. She is a hunter and collector Matt Lennert led creative for Deloitte Insights and was the art director for
of trends, and serves as a repository of inspiration for the creative team. In Deloitte Insights Magazine. His work with artists and data visualization
a world full of noise and detritus, she’s able to see patterns. She treasures designers over the last two decades visually brought the stories to life. This
her Sundays jumping on the trampoline with her daughters at her home in issue marks the 33rd, and last, issue that he produced. Lennert and his wife
Connecticut. have retired and are off traveling the world.
Natalie Pfaff Molly Piersol
Natalie Pfaff is a senior graphic and data visualization designer at Deloitte Molly Piersol leads data visualization design at Deloitte Insights and is
Insights, passionate about crafting compelling visual stories. She believes that the designer for Deloitte Insights Magazine. She believes raw data creates
great design begins with a strong brand foundation and leverages it to push beautiful art and works to further expand the stories that live between the
creative boundaries. At home in Wisconsin, she cherishes everyday moments lines of figures. Piersol is a Virginia transplant to the Seattle area and her roots
with her husband and daughter. have grown deep enough that she’ll never go back.
Sofia Sergi Jim Slatton
Sofia Sergi is a Deloitte Insights senior graphic designer from New York. Her Jim Slatton is a Deloitte Insights designer and illustrator from Asheville, N.C.
passion for painting and drawing at a young age evolved into a love for design His graphic style is rooted in decades of branding work and a love of mid-
and storytelling. She draws inspiration from merging traditional art forms with century minimalism. He works with custom iconography and typography,
contemporary aesthetics to tell engaging and meaningful stories. photo collage, and data to distill complex information into simple visual stories.
Issue 33 11
Sonya Vasilieff Harry Wedel
Sonya Vasilieff is an art director at Deloitte Insights. She works in several Harry Wedel is a senior data visualization designer at Deloitte Insights. With
mediums including graphic design and illustration. Vasilieff is a native of a background in scientific research, he loves to find innovative ways to display
Seattle, which is as rare as the razor clams she digs for every year on her complex data through engaging information design. Based in New York,
beloved gray and rainy Washington coast. he credits his interests in music and photography for his passion for using
emotion in art to bring people together.
Alexis Werbeck
Alexis Werbeck is an art director and visual storyteller at Deloitte Insights.
Known for her bold, color-driven style, she loves to push creative boundaries
but will get irked if a single pixel is out of place. Although her dream of
becoming a pop star never transpired, you can find her singing karaoke and
rocking at least one article of clothing with rhinestones.
12 Deloitte Insights Magazine
In this era of disruption,
you need practical foresight,
fresh insights, and trustworthy
data to help make your
organization more resilient
and better prepared for new
opportunities.
From investigating current
trends to offering cutting-
edge solutions for your most
complex business challenges,
our teams of researchers, data
scientists, and multimedia
storytellers bring clarity to an
uncertain world.
We are Deloitte Research and Insights
•
Center for Energy & Industrials
•
Center for Financial Services
•
Center for Government Insights
•
Center for Health Solutions
•
Center for Integrated Research
•
Center for Machine Intelligence and Data Science
•
Center for Technology, Media & Telecommunications
•
Consumer Industry Center
•
Global Economist Network
•
Deloitte Insights
Get informed. Get inspired.
www.deloitteinsights.com
Future-focused
insights delivered
Elevate your email with Deloitte Insights.
Subscribe now and transform your inbox
into your destination for the freshest
business insights.
• Weekly alerts on key business topics
• Monthly recap of our top publications
• Access to exclusive Deloitte Insights Magazine content
• Quarterly picks recommended by readers like you
www.deloitte.com/insights/subscriptions
Copyright © 2024 Deloitte Development LLC.
Subscribe now
All rights reserved.
1
Gen AI investments increasingly extend
beyond the AI itself
Deloitte’s State of Generative AI quarterly survey explores where industry leaders are directing their
gen-AI–related funding
Generative AI’s near- and long-term success prioritization of those investments varies by indus- seem particularly focused on increasing their cyber-
hinges on continued co-investment in the try, the anticipated investments suggest that tech- security spending to support gen AI initiatives.
wider technology ecosystem, and recent Deloitte nology budgets may need to increase across the On average, 53% of respondents expect invest-
research signals that many early adopters are plan- board to take advantage of gen AI’s promise. ments in traditional AI and machine learning to
ning their AI-related investments accordingly. Strong data hygiene is a prerequisite for success- increase alongside investments in gen AI, sug-
According to the third installment of Deloitte’s ful AI and gen AI strategies, and 70% of leaders in gesting that those spending more on both will
quarterly State of Generative AI in the Enterprise the Deloitte State of Gen AI study are investing in be looking to combine predictive and generative
survey, which was fielded in May and June 2024, data management capabilities. Meanwhile, 73% of capabilities in powerful applications.
and gathered responses from nearly 2,800 lead- respondents expect their investment in cloud con-
ers whose organizations are further along in their sumption to increase along with investments in gen
adoption and implementation of gen AI solutions, AI. And, while cybersecurity capabilities are seeing Research and analysis by the Deloitte Center for
leaders across industries expect to be making criti- high co-investment levels from respondents across Integrated Research
cal investments in both gen AI and the intertwined all industries in the survey, averaging 75%, three
and AI-enabling capabilities of data management, industries—financial services; energy, resources, Read the full report at
cloud consumption, and cybersecurity. While the and industrials; and government institutions— www.deloitte.com/us/state-of-gen-ai
18 Deloitte Insights Magazine
losreiP
ylloM
yb
cihpargofnI
DATA POINTS
Q: “To what extent are technology investments in the following areas impacted
as a result of your organization’s enterprisewide generative AI strategy?”
Percentage of respondents who selected “increasing” or “significantly increasing”
100%
Data management
70% average
Traditional AI and
machine learning
capabilities
Cloud consumption 53% average
73% average
Communication
networks
42% average
Cybersecurity
75% average
Hardware
35% average
Note: Deloitte’s AI Institute and Center for Technology, Media & Telecommunications also contributed to this data collection and analysis.
Source: Deloitte Center for Integrated Research’s analysis of data from the Deloitte State of Generative AI wave 3 survey of 2,770 artificial
intelligence leaders, fielded in May and June 2024. These organizations should be considered more advanced users of artificial intelligence.
While
business
leaders look
inward for
AI’s impact,
tech leaders
look outward
A Deloitte Global study
examines the differences in
metrics used by organizations’
leaders to determine the success
of their AI investments
Strategies are being determined. Experimen-
tation is running rampant. Proofs of concept
abound. As generative AI quickly gains a foothold
across organizations and industries, there’s lit-
tle consensus yet about how best to determine its
impact—and whether C-level executives will reach
consensus, themselves. There are clues, however,
in how business and technology leaders meas-
ure value for traditional artificial intelligence, the
larger class of AI investments such as machine
learning, deep learning, and conversational AI for
which executives have established measurement
behaviors and preferences.
Using data from a global survey of 1,600 busi-
ness and technology leaders across 14 countries
conducted in February 2023,1 the Deloitte Center
for Integrated Research analyzed how technol-
ogy leaders and business leaders prioritize the key
performance indicators commonly associated with
digital investments when assessing the impact of
their organizations’ AI capabilities. The results
of this assessment proved to be counterintuitive:
Interestingly, while business leaders who partic-
ipated in the survey reported that they’re more
focused on AI’s process-related benefits within
their organizations, tech leader respondents said
they’re more often looking outward—at KPIs asso-
ciated with sales and customer satisfaction.
According to the survey, technology leaders
are 12 percentage points more likely than business
Issue 33 19
ledeW
yrraH
yb
cihpargofnI
KPIs for traditional Al, showing misalignment greater than or equal to 7 percentage points
between business and tech leaders
Tech leaders Business leaders Difference between responses
80% Overall
utilization
of KPI
10
9 9
7
10
12
10
40%
7
8
0%
Sales Sales of Net Intangible Share price Procurement Employee Employee Process
through new promoter assets as a volatility value for development utilization effectiveness
new digital score percentage money rate
digital products of long-term
platform assets
Used more by tech leaders Used more by business leaders
Notes: 1) N = 1,600; 2) Out of 1,204 respondents for traditional Al, 1,180 are technology and business leaders. The remaining are categorized
under “other.” 3) Business roles include administration, finance, human resources, marketing, operations, procurement, risk/compliance,
sales, strategy. Tech/transformation roles include digital, R&D, technology/IT, transformation.
Source: Deloitte Center for Integrated Research survey of global tech value leaders, conducted in February 2023.
leaders to be using the sales of new digital products Leadership’s alignment on AI success met-
as a KPI and 7 percentage points more likely to be rics could be less critical during an organization’s
focused on sales through new digital platforms, for experimentation or initial adoption phase, but it
instance. They also use net promoter scores and could, of course, become increasingly important
intangible assets more than business leaders.2 as the organization works to assess the technolo-
When it comes to all forms of AI, business and gy’s current and potential impact, and makes the
tech leaders alike might collectively be missing case for continued investment.
opportunities to consider innovation measures
and long-term value creation, the survey findings
suggest. Among those leaders who measure tradi- Research and analysis by the Deloitte Center for
tional AI, only about 30% use innovation-oriented Integrated Research
KPIs like the tech’s effect on an organization’s tol-
erance for experimentation or intelligent failure, Read the full report at
or the number of agile pods or teams.3 www.deloitte.com/insights/measuring-ai
DATA POINTS
Few AI regulations across the globe address
the outcomes rather than the tech
Outcome-based and risk-weighted regulations are an underused tool that can both protect the public
interest and encourage innovation, a Deloitte US analysis shows
When it comes to fast-moving technologies regulatory sandboxes that allow for prototyping risk-weighted, and no regulations included in the
like artificial intelligence, how can govern- and testing new methods; and collaborative regula- data set were both.
ments strike the balance between enabling innova- tion, which seeks alignment and engagement across This isn’t to say that outcome-based and risk-
tion and protecting the public interest? Innovation national and international players within the ecosys- weighted regulations don’t exist. They likely con-
and regulation tend to operate on two different tem. Second, the research center outlines principles stitute part of the regulatory structures of the 69
time frames, which can cause problems when gov- related to the regulations’ focus: outcome-based countries included in the analysis, according to
ernments are working to regulate rapidly evolving regulation, which focuses on the results rather than the researchers. It’s just that those regulations
technology like AI. And consider AI’s complexity the processes; and risk-weighted regulation, which aren’t considered “AI regulations,” so there’s an
and diversity: From computer vision finding pot- proposes a shift from one-size-fits-all regulation to opportunity for many governments’ AI-adjacent
holes in roads to generative pretrained transform- a data-driven, segmented approach. regulations to become more explicit. And these
ers answering people’s tax questions and more, it Outcome-based and risk-weighted regulations clarifications don’t just protect the public. They
could be a formidable challenge to find a single set can be powerful tools for regu |
172 | deloitte | guidance-eo-cheat-sheet-2024.pdf | A CDAO Perspective:
Safe, Secure, and Trustworthy AI
Updated: September 2024
On October 30, 2023, President Joe Biden signed an Executive Order (EO) on the Safe, Secure, and Trustworthy Development and Use of
Artificial Intelligence. The EO built on the administration’s work that led to voluntary commitments from leading companies to drive safe,
secure, and trustworthy development of AI and directed federal agencies to take certain actions to incorporate and govern AI in their
missions. Federal agencies have reported completing all the 150-day actions required by the EO since completing the 90-day actions. On
March 28, 2024, the Office of Management and Budget (OMB) produced its first government-wide policy for managing AI risks and
harnessing AI’s benefits. For key tenants, refer to the EO Fact Sheet and the OMB Guidance Fact Sheet. On August 14, 2024, final
guidance was released requiring agencies to submit their AI use case inventories to OMB for public reporting.
Government Chief Data Officers (CDO) and data leaders play a The following key themes for CDOs exist in an evolving
critical role preparing their organizations for the safe, secure, and landscape of federal guidance on AI adoption and do not capture
trustworthy application for AI. Government CDOs may face the extent of the data-specific requirements within the EO and
challenges and opportunities to improve their organization’s accompanying OMB Guidance.
data and operations for enhanced, equitable, and innovative
applications of AI.
DESIGNATING A CHIEF ARTIFICIAL INTELLIGENCE OFFICER (CAIO)
OMB POLICY REVIEW WHAT TO DO NEXT
“Within 60 days of the issuance of the The formal introduction of the CAIO prompts an immediate step to
October 2023 memorandum, the head ensure the careful coordination between the responsibilities
of each agency must designate a designated for CDOs, CDAOs and those required for CAIOs and to
Chief Artificial Intelligence Officer make the proper designations for the relevant department. This will
(CAIO).” (OMB M-24-10 Section 3.b.i) differ based on the agency’s current designations and
considerations and may involve designating the existing CDO as the
The primary responsibility for CAIOs
CAIO.
is “coordinating their agency’s use of
AI, promoting AI innovation, managing Expanded or new notable CAIO responsibilities include a) ensure AI
risks from the us of AI, and carrying code and data used for AI are “appropriately inventoried, shared,
out the agency responsibilities…” (OMB and released in agency code and data repositories”, b) developing
M-24-10 Section 3.b.ii) AI risk management guidance, c) instituting governance to remain
compliant, working on resourcing requirements and
recommending investment areas to build enterprise capacity, and
d) sharing relevant information with agency officials involved in
agency AI policymaking initiatives.
Action: Evaluate existing responsibilities and coordinate the role of
CDOs, CDAOs, and CAIOs to use, promote, and manage risks for the
agency’s use of AI
HIRING AI TALENT AND PROVIDING AI TRAINING
POLICY REVIEW WHAT TO DO NEXT
“…agencies are strongly encouraged to CDOs can take advantage of this focus on hiring AI talent and AI-
prioritize recruiting, hiring, developing, enabling talent to energize existing efforts towards building data
and retaining talent in AI and AI- fluency and upskilling their workforce for the use of AI. Data fluency
enabling roles to increase enterprise is a foundational element to prepare an organization’s workforce
capacity for responsible AI innovation.” and data for the effective use of AI solutions. CDOs can use OMB’s
(OMB M-24-10 4.c) forthcoming AI and Tech Hiring Playbook as a resource and
coordinate with the newly required AI Talent lead for the AI Talent
Task Force.
Action: CDOs can focus on incorporating AI fluency into existing
data literacy training programs to upskill their workforce and
prepare to engage new AI talent.
A CDAO Perspective:
Safe, Secure, and Trustworthy AI
CONNECTING NEW AI STRATEGIES TO DATA STRATEGY
POLICY REVIEW WHAT TO DO NEXT
“Within 365 days of the issuance of the To develop AI strategies and pursue high-impact AI use cases, CDOs
October 2023 memorandum, each should review existing data strategies with a focus on data
CFO Act agency must develop and infrastructure for AI and workforce readiness. Strategies should
release publicly…a strategy for connect on the agency’s top opportunities for AI, plans to increase
identifying and removing barriers to AI capacities and AI maturity, improvements for practitioner AI and
the responsible use of AI…” (OMB M- data literacy, and effective governance of AI usage.
24-10 Section 4.a).
To advance responsible AI innovation, the EO and accompanying
“Any data used to help develop, test, OMB guidance focus on several data actions CDOs can take to
or maintain AI applications, regardless remove barriers, including developing adequate infrastructure
of source, should be assessed for and curated agency data sets, maximizing access to internal
quality, representativeness, and data, and encouraging public access datasets.
bias.” (OMB M-24-10 Section 4.a.ii)
Action: Assess current data strategies and implementation efforts
to identify AI strategies and AI use cases, with attention to
organizational AI maturity, data literacy, and governance.
MANAGING RISKS FOR RIGHTS-IMPACTING AND SAFETY-IMPACTING AI
POLICY REVIEW WHAT TO DO NEXT
“Within 60 days of the issuance of the CDOs may participate in AI Governance Boards (with senior officials
October 2023 memorandum, each to govern the use of AI) as a representative responsible for data’s
CFO Act agency must convene an role as a key enabler and risk factor in AI adoption.
agency AI Governance board” (OMB
Agencies must review current or planned AI use to assess whether
M-24-10 Section 3.a.ii)
it meets the definition of safety or rights-impacting AI, part of which
“…all agencies are required to asks if the AI output “serves as a principal basis for a decision or
implement minimum practices…to action.” While is ultimately the determination of the CAIO, the CDO
manage risks from safety-impacting may support determining which AI is safety-impacting or rights-
AI and rights impacting AI.” (OMB M- impacting.
24-10 Section 5)
By December 1, 2024, Organizations will be required to document
This includes specific actions, for data assessments and other data-related activities in an AI Impact
example on terminating non- Assessment, required to be updated and leveraged throughout
compliant AI, determining which AI is the AI’s lifecycle. As part of the minimum practices prescribed for
safety-impacting or rights impacting, safety-impacting or rights-impacting AI, policy continues to focus on
and minimum practices for safety or properly documenting the agency’s data use for AI, the use of AI
rights-impacting AI. adequately representing communities and including activities
such as monitoring for improper bias and AI-enabled
“Agencies must assess the quality of
discrimination.
the data used in the AI’s design,
development, training, testing, and Action: CDOs can recognize their critical role in the AI impact
operation and its fitness to the AI’s assessment, ensuring the organization is prepared with the
intended purpose.” (OMB M-24-10 appropriate data infrastructure and data quality needed for
Section 5.c.iv.A.3) trustworthy AI.
FEDERAL CDO INSIGHTS
For or agencies that are already utilizing AI as an accelerator or preparing for its use, CDOs play a unique role across
innovating with AI and managing the potential barriers and risks tied to the safe, secure, equitable and trustworthy
utilization of AI. For more insights on the CDO role and CDO community needs, check out “The Mission-Driven CDO:
Insights from the 2023 Survey of Federal Chief Data Officers”.
A CDAO Perspective:
Safe, Secure, and Trustworthy AI
PUBLIC REPORTING OF AI USE CASES
POLICY REVIEW WHAT TO DO NEXT
“By December 16, 2024, each agency To support transparency, agencies must send an inventory of their
(except for the Department of Defense use cases to OMB. These use cases will be reported publicly.
and agencies in the Intelligence
Exceptions include research and development use cases, one-time
Community) must:
stand-alone use cases, use cases by the Department of Defense,
1. “Annually submit an inventory of use cases within a National Security System or within the
its AI use cases to OMB…” intelligence community, and use cases for which public reporting is
inconsistent with federal law or government wide policy.
2. “Subsequently post a
consolidated, machine-readable CSV of Action: For each qualifying use case in the inventory, complete a
all publicly releasable use cases on form at https://collect.omb.gov/site/212/home-page. Next post a
their agency’s website …”(EO 14110, machine-readable CSV of all publicly releasable use cases on their
2.a.1 and 2a.2) agency’s website at [agency.gov]/ai.
For non-reportable use cases, agencies must keep an inventory and
report aggregate metrics to https://collect.omb.gov/site/212/home-
page.
FEDERAL CDO INSIGHTS
For or agencies that are already utilizing AI as an accelerator or preparing for its use, CDOs play a unique role across
innovating with AI and managing the potential barriers and risks tied to the safe, secure, equitable and trustworthy
utilization of AI. For more insights on the CDO role and CDO community needs, check out “The Mission-Driven CDO:
Insights from the 2023 Survey of Federal Chief Data Officers”.
A CDAO Perspective:
Safe, Secure, and Trustworthy AI
CONNECT & INNOVATE ADDITIONAL SOLUTIONS
TOGETHER AND ACCELERATORS
CDAO Services Government AI Use Case Dossier
Support Chief Data Officers and other data leaders to See what’s working for other agencies and
enable and improve data-driven organizations through consider the ways AI can advance your
services like data governance, literacy, and strategy. mission with the Government and Public
Services Sector AI Use Case Dossier.
Identify: Data requirements, data sources,
insights, and value CAIO Transition Lab
Refine the CAIO role requirements, establish
Discover & Prep: Assess Risks, and
effective governance approaches, and create a
develop the Data and Insights Strategy
transition plan to empower a new CAIO through
Design & Build: Manage the data value a specialist guided experience.
chain; procure, ingest, and store
Trustworthy AI™
Launch & Integrate: Develop delivery
Understand seven key areas of risk for AI and
models, governance, and operations
keep your use of AI safe and ethical with Deloitte’s
Monitor & Mature: Support, enhance, Trustworthy AI ™ framework in line with NIST.
and scale data and insights services
AI and Data Strategy Services
Align on an organizational vision for AI, prioritize
Suite of CDO, Data, and AI Labs
AI use cases, and make strategic choices about
A one-day experience designed to—establish a
where to invest in AI, accelerated by Playbooks and
common understanding of the aspirations and
immersive Labs guided by experienced facilitators.
challenges of the CDAO’s team and key
stakeholders and develop a 180-day plan to drive AI Readiness & Management Toolkit
the CDAO’s priorities.
Apply our framework and tools to assess current
state, define future state, and chart a path
forward to build AI maturity across workforce,
CDO Playbook
data, and technology.
See the most recent thought leadership of CDOs in
the government based on trends and understanding GovConnect AI Ready Data Foundation
for AI priorities, AI Strategies and implementation of Suite of services designed to assist government
operating models agencies in building and managing modern,
cloud integrated data ecosystems, enabling the
delivery of AI at scale.
Contacts
Deloitte supports many Federal clients in the data and AI space. With best-in-class
AI advice and capabilities, We can help at each stage of the race, providing Chief
Data Officers with the CDAO Services they need to navigate changing regulation Ed Van Buren Adita Karkera
and the safe, secure, and trustworthy application of AI. Reach out for a consultation Principal, Deloitte Managing Director, Deloitte
Government and Public Services Government and Public Services
or to ask about our approach to the new executive branch AI guidance.
AI Strategic Growth Offering Leader Chief Data Officer
[email protected] [email protected]
As used in this document, “Deloitte” means Deloitte Consulting LLP, a subsidiary of Deloitte LLP. Please see http://www.deloitte.com/us/about for a detailed description
of our legal structure. Certain services may not be available to attest clients under the rules and regulations of public accounting. This publication contains general
information only and Deloitte is not, by means of this publication, rendering accounting, business, financial, investment, legal, tax, or other professional advice or services.
This publication is not a substitute for such professional advice or services, nor should it be used as a basis for any decision or action that may affect your business. Before
making any decision or taking any action that may affect your business, you should consult a qualified professional advisor. Deloitte shall not be responsible for any loss
sustained by any person who relies on this publication. Copyright © 2024 Deloitte Development LLC. All rights reserved |
173 | deloitte | DI_CIR_State-of-AI-4th-edition.pdf | A report from the
Deloitte AI Institute and the
Deloitte Center for Integrated Research
Becoming an AI-fueled
organization
Deloitte’s State of AI in the Enterprise, 4th Edition
About the Deloitte AI Institute
The Deloitte AI Institute helps organizations connect all the different dimensions of the robust, highly
dynamic, and rapidly evolving AI ecosystem. The AI Institute leads conversations on applied AI
innovation across industries, using cutting-edge insights to promote human-machine collaboration in
the Age of WithTM.
The Deloitte AI Institute aims to promote dialogue about and development of artificial intelligence,
stimulate innovation, and examine challenges to AI implementation and ways to address them. The AI
Institute collaborates with an ecosystem composed of academic research groups, start-ups,
entrepreneurs, innovators, mature AI product leaders, and AI visionaries to explore key areas of
artificial intelligence including risks, policies, ethics, future of work and talent, and applied AI use cases.
Combined with Deloitte’s deep knowledge and experience in artificial intelligence applications, the
institute helps make sense of this complex ecosystem and, as a result, delivers impactful perspectives
to help organizations succeed by making informed AI decisions.
No matter what stage of the AI journey you’re in—whether you’re a board member or C-suite leader
driving strategy for your organization, or a hands-on data scientist bringing an AI strategy to life—the
Deloitte AI institute can help you learn more about how enterprises across the world are leveraging AI
for a competitive advantage. Visit us at the Deloitte AI Institute for the full body of our work, subscribe
to our podcasts and newsletter, and join us at our meetups and live events. Let’s explore the future of
AI together. L earn more.
About the Deloitte Center for Integrated Research
The Deloitte Center for Integrated Research (CIR) offers rigorously researched and data-driven
perspectives on critical issues affecting businesses today. We sit at the center of Deloitte’s industry and
functional expertise, combining the leading insights from across our firm to help leaders confidently
compete in today’s ever-changing marketplace.
Connect
To learn more please visit www.deloitte.com/us/cir.
Contents
Foreword: Becoming an AI-fueled organization 2
Executive Summary: Constantly transforming, never fully transformed 3
Strategy: What should your north star be? 7
Operations: How do you bring transformation into everyday work? 11
Culture & Change Management: Why is valuable change so elusive? 14
Ecosystems: How should you orchestrate your partnerships? 18
Our AI-fueled future: The pathway is clear 20
Methodology 21
Endnotes 23
Becoming an AI-fueled organization
Foreword: Becoming an
AI-fueled organization
RAPIDLY TRANSFORMING, BUT not fully Algorithms can independently balance financial
transformed—this is our overarching portfolios. Support centers often know customers’
conclusion on the market, based on the problems before they call.2 And these are still
fourth edition of our State of AI in the Enterprise early days.
global survey. Very few organizations can claim to
be completely AI-fueled, but a significant and In combination, these developments help enable
growing percentage are starting to display the businesses to increasingly liberate themselves from
behaviors that can get them there. the time constraints of human rhythms. Core
business operations can meet customer needs at a
To us, this is exciting and has reinforced our belief faster pace, while freeing up time and energy for
that now is one of the most opportunity-rich the workforce to use new tools to discover
periods in the history of AI technology. innovative avenues for value creation. Conversely,
Organizations are swiftly building capabilities and for those organizations lagging in AI capability
reaching enterprise scale: In fact, more than a development, it could pose an ever-increasing risk
quarter of our survey respondents have reached to their competitive viability in the not-too-distant
full-scale deployment of five or more types of AI future. The massive global disruptions over the last
applications within their organization.1 This year have only accelerated these trends beyond our
widespread enterprise experimentation has set a most aggressive predictions.3
solid foundation for many, making way for what
we believe could be a bumper crop of meaningful Fortunately, we’ve learned in recent years about
advancements and impact over the next few years. which practices can accelerate transformation, and
This is especially true for those who are already this knowledge can help fast-track outcomes. The
beginning to use AI to solve some of their most findings in this report aim to support organizations
business-critical and challenging problems. in navigating through the growing pains, in
whichever stage they may find themselves on the
Within just the last 18 months, AI capabilities have journey to becoming an AI-fueled organization.
advanced considerably, maturing from what was
often experienced as a bothersome critic—telling The Age of With is no longer on its way—it has
workers what to do or pointing out their mistakes— arrived. We hope you’ll join us as this story
to more frequently serving as a copilot, continues to unfold.
independently executing on insights and trends — Jason Girzadas, managing principal of
surfaced through the power and speed of cloud- Businesses, Global, and Strategic Services,
Deloitte LLP
based data hosting and computation. Today, some
supply chains are managing themselves.
2
Deloitte’s State of AI in the Enterprise, 4th Edition
Executive summary:
Constantly transforming,
never fully transformed
SINCE 2017, DELOITTE has documented the To learn how organizations across the globe are
increasing adoption of AI across the progressing toward this vision, we surveyed 2,875
enterprise. The third edition, published in executives from 11 top economies who have
2020, declared that we had entered the “age of purview into AI strategies and investments within
pervasive AI.”4 Pervasive AI adoption, however, their organizations. We asked them about a wide
does not mean that AI is being used to its full variety of behaviors—from their overarching AI
potential. And so, with the fourth edition of our strategy and leadership, to their technology and
global State of AI in the Enterprise report, we data approaches, and how they are helping their
explored the deeper transformations happening workforce to operationalize AI. Then, to
inside organizations that are using AI to drive understand which behaviors lead to the greatest
value. In other words, we wanted to know: outcomes, we analyzed the survey responses based
What are today’s most AI-fueled on how many types of AI applications a company
organizations doing differently to has deployed full-scale and the number of
drive success? outcomes achieved to a high degree (figure 1).
AI-fueled organizations leverage data as an asset to
deploy and scale AI systematically across all types
of core business processes in a human-centered
way. They use the power of rapid, data-driven
decision-making to enhance workforce and
customer experiences to achieve competitive
advantage and continuously innovate.5
“Becoming an AI-fueled
organization is to understand
that the transformation process
is never complete, but rather a
journey of continuous learning
and improvement.”
— Nitin Mittal, Deloitte AI coleader,
principal, Deloitte Consulting LLP
3
FIGURE 1
An organization’s AI maturity can be profiled based on the number of
applications deployed and outcomes achieved
AI application types fully
deployed
None 1–3 4–7 >7
"Pathseekers" "Transformers"
Low deployed/high High deployed/high
achieving: 26% achieving: 28%
N=753 N=794
Outcomes
achieved
(high degree)
"Starters" "Underachievers"
Low deployed/low High deployed/low
achieving: 29% achieving: 17%
N=832 N=496
4
7>
7–4
3–1
enoN
Becoming an AI-fueled organization
Transformers Pathseekers Underachievers Starters Total
6.8
6.2
5.9
5.5
4.0
3.5
1.9
1.6 1.4
1.0
Average number of AI apps in full deployment Average number of outcomes achieved to a high
degree
Source: The State of AI 4th Edition data analysis.
Deloitte Insights | deloitte.com/insights
Deloitte’s State of AI in the Enterprise, 4th Edition
This analysis revealed four key profiles: strong outcomes became evident. They fell into the
following categories: Strategy, Operations,
• Transformers (High outcome and high
Culture and change management, and
deployed—28% of survey respondents):
Ecosystems.
Transforming but not fully transformed, this
group has identified and largely adopted
Analysis of survey data and executive interviews
leading practices associated with the strongest
AI outcomes. They average 5.9 out of 10 revealed that success is built upon the foundation
possible full-scale deployments of different of a clear strategy that is communicated and
types of AI applications, and 6.8 out of 17 incentivized from the highest leadership—but that
possible outcomes achieved to a high degree. is not enough. With that clear strategy in place, two
They are the market leaders on their way to inter-related leading practices typically work
becoming AI-fueled organizations. together to support AI adoption and scale across
the enterprise: operations and culture plus change
• Pathseekers (High outcome and low
deployed—26% of survey respondents): management. And finally, the support of a robust
Pathseekers have adopted capabilities and set of ecosystem partners was shown to provide the
behaviors that are leading to success, but on technical foundations and outside perspectives
fewer initiatives. They are making moves but needed to deliver and perpetually innovate at scale.
have not scaled to the same degree as
Transformers. They average 1.9 out of 10 Our analysis also revealed not just what those
possible full-scale deployments of different
leading practices were, but how much of an effect
types of AI applications, and 6.2 out of 17
they had on organizational achievement:
possible outcomes achieved to a high degree.
• Strategy leading practice: AI-fueled
• Underachievers (Low outcome and high
organizations view AI as a key element of
deployed—17% of survey respondents): A
business differentiation and success, and
significant amount of development and
they set an enterprisewide strategy that
deployment activity characterizes this group;
is championed from the top. Organizations
however, they haven’t adopted enough leading
with an enterprisewide strategy and
practices to help them effectively achieve
leadership who communicate a bold vision are
meaningful outcomes. They average 5.5 out of
1.7 times more likely to achieve outcomes to a
10 possible full-scale deployments of different
high degree.
types of AI applications, and 1.4 out of 17
possible outcomes achieved to a high degree. • Operations leading practice: AI-fueled
organizations establish new operating
• Starters (Low outcome and low
models and processes that drive
deployed—29% of survey respondents):
sustained quality, innovation, and value
Getting a late start in building AI capabilities
creation. Organizations that document and
seems to characterize this group. They are the
enforce MLOps processes are approximately
least likely to demonstrate leading practice
two times as likely to achieve their goals to a
behaviors. They average 1.6 out of 10 possible
high degree. They are also about two times as
full-scale deployments of different types of AI
likely to report being extremely prepared for
applications, and 1.0 out of 17 possible
risks associated with AI and nearly two times
outcomes achieved to a high degree.
as confident that they can deploy AI initiatives
in a trustworthy way.
By analyzing these groups—the Transformers in
particular—the behaviors most associated with
5
Becoming an AI-fueled organization
• Culture and change management leading “By embracing AI strategically
practice: AI-fueled organizations
and challenging orthodoxies,
nurture a trusting, agile, data-fluent
culture and invest in change organizations can define a
management to support new ways of road map for adoption, quality
working. Organizations that invest in change
delivery, and scale to create
management to a high degree are 1.6 times as
likely to report that AI initiatives exceed or unlock value faster than
expectations and more than 1.5 times as likely
ever before.”
to achieve their desired goals, compared to
the rest. — Irfan Saif, Deloitte AI coleader, principal,
Deloitte & Touche LLP
• Ecosystems leading practice: AI-fueled
organizations orchestrate dynamic
ecosystems that help build and protect
competitive differentiation. Overall, In the following report, we explore each leading
organizations with more diverse ecosystems practice in detail, sharing critical and often
were 1.4 times as likely to use AI in a way that overlooked actions that leaders can take to avoid
differentiates them from their competitors. pitfalls on their transformation journey.
6
Deloitte’s State of AI in the Enterprise, 4th Edition
Strategy: What should
your north star be?
CORE LEADING PRACTICE: SET A CLEAR ENTERPRISEWIDE STRATEGY AT THE TOP THAT
ENABLES LEADERS TO HARNESS AI CAPABILITIES TO DRIVE NEW OPPORTUNITIES AND
COMPETITIVE ADVANTAGE.
Key findings:
• Set and communicate a bold vision. Organizations with an enterprisewide strategy and leaders who
communicate a bold vision are 1.7 times as likely to achieve outcomes to a high degree.
• Look for ways AI can help achieve a differentiated strategy. Only 38% of respondents believe their
use of AI differentiates them from competitors.
• Communicate your strategy transparently. Tell your workforce and the market about your strategy
and the implications and trade-offs along the way.
Pitfalls to avoid:
• Don’t ask data scientists or IT to drive your AI strategy. Senior business leaders should drive it based
on the core business strategy in partnership with data scientists.
• Don’t overindex on efficiency goals. Balance efficiency targets with growth- and
innovation-oriented goals.
ONE OF THE most frequently cited leading Lost in AI use cases: Leaders
practices for AI transformation is the need can forget to put their
for a bold, enterprisewide strategy that is business strategy first
set and championed by an organization’s highest
leadership. Our research confirmed this: To many leaders, it comes as a surprise to learn
Transformers are more than three times as likely to that the investment needed to develop AI solutions
have an enterprisewide strategy in place, and well cannot realize a return through the deployment of
over twice as likely as Starters to report their single, disconnected use cases, or even a handful.6
leaders communicate a vision for AI. However, This is why it’s so important to have an AI strategy
only 40% of our total survey respondents that is connected and coordinated across the
completely agreed that their company has one in enterprise, in tight alignment with the overarching
place. Meanwhile, even though a significant business strategy. All too often, however, business
majority (66%) of respondents view AI as critical to leaders get the planning process out of order,
success, only 38% believe their use of AI focusing too much on use cases or abdicating
differentiates them from competitors. What should leadership of the AI strategy to IT or data sciences.
organizations do differently to strengthen This can be a slippery slope, diminishing the
their approach? organization’s ability to use AI to create new ways
7
Becoming an AI-fueled organization
FIGURE 2
Leading AI strategy practices
Percentage of respondents who selected "completely agree" or "very important" to
these statements about strategy
Transformers Pathseekers Underachievers Starters Total
79%
68% 69% 66%
60%
55% 57%
48% 49%
44% 45%
38% 40% 40%
33% 33%
30%
22% 24%
19%
My company's use of My company has an My senior leaders Our AI initiatives are
AI differentiates us enterprisewide AI communicate a vision for important to our
from our competitors strategy AI that will significantly remaining competitive
change how we operate over the next five years
Source: State of AI 4th edition data analysis.
Deloitte Insights | deloitte.com/insights
of competing for customers, launching products, same key performance indicators (KPIs) that have
accelerating time-to-market, securing supply been crafted to incentivize and grow
chains, and beyond. competitive advantage.
In a now famous example from the early
To many leaders, it comes as a surprise
2010s, Jeff Bezos mandated that every
to learn that the investment needed
leader across Amazon plan for how they
to develop AI solutions cannot realize would use AI and machine learning (ML) to
a return through the deployment of help the company compete and win. This
imperative drove unparalleled innovation
single, disconnected use cases, or even
and was cited as the catalyst for the
a handful.
Amazon’s rise to become an AI leader
today.7 Many of the strongest AI strategies
start in this same way: by pushing clear
The strongest AI strategies tend to begin without objectives down to business leadership, so they can
ever mentioning AI. Instead, they should begin identify gaps and opportunities within their
with the organization’s north star: the core divisions and work backward from there to
business strategy. From there, the process requires apply AI as a solution.
tight collaboration with engaged leaders across all
business divisions and the focus of workers at all These local plans should then be brought back to
levels. Ultimately, AI strategy should function as the top, so that mutual goals and initiatives can be
the fuel to the business strategy, aligning to the aligned and unified with the core business strategy.
8
Deloitte’s State of AI in the Enterprise, 4th Edition
This step is often critical: It’s only when AI has “You have to go both for impact and build the
been integrated and proliferated throughout the foundations in parallel, and that is the most
enterprise that it can deliver the combination of challenging part,” advises Najat Khan, PhD, chief
efficiency- and value-creating outcomes needed to data science officer and global head of strategy &
fuel ongoing returns. operations for Janssen Research & Development.
“You have to pick the right questions, and have
what I call a diversified portfolio of questions to
Balance your goals:
drive impact, ensuring that you can demonstrate
Overindexing on efficiency can early value to build momentum for achieving
lead to missed opportunities longer-term, sustainable impact.”
It’s through the combination of both efficiency- and AI-fueled organizations can create durable
value-creation targets that organizations typically competitive advantage when the CEO and C-suite
achieve the most success. “When digitally collectively harness data, advanced analytics, and
transforming a company, you want greater degrees AI to shape strategic possibilities for both the near
of efficiency,” remarks Rajeev Ronanki, SVP and and long term in support of their
chief digital officer at Anthem. “But there is a corporate strategy.
second order of business: What new business
opportunities, what capabilities does AI open up
Communicate the
that allow for servicing adjacent or maybe entirely
new areas?” vision: Publicly signaling
transformation can
Our survey results reinforced this, demonstrating
build market value
that lower-achieving organizations (Starters and
Underachievers) tended to focus more on efficiency Chief executives of high-achieving organizations
or “cost out” goals, while high-achieving typically serve as the AI communicator-in-chief.
organizations (Transformers and Pathseekers) According to our survey data, those organizations
were more likely to emphasize growth-oriented that communicate a clear vision are 1.5 times as
goals, such as: improving customer satisfaction, likely to achieve desired outcomes compared to
creating new products and offers, and entering new those who do not. The most effective leaders tend
markets. In other words, high-achieving to use their platform not only to communicate and
organizations are more likely to maintain an eye champion their plans; they also clarify the
toward the art of the possible and a growth implications and trade-offs required along the way.
mindset, which allow them to take advantage of This is often essential for maintaining focus and
opportunities often missed by those who overindex ensuring that decisions made at all levels of the
on efficiency or supporting business as usual. organization remain aligned to the vision.
Leaders should also remember that value
“Envision what is possible in your
can be created by influencing perceptions
business, whether it’s been done before of the market and investors.
Communicating the company’s vision
or never been done before.”
publicly can amplify success, signaling to
— Michelle Lee, VP of Amazon Machine Learning
capital markets and the competitive talent
Solutions Lab
market that an organization is investing in
a bold and exciting future.8 If it’s not
9
Becoming an AI-fueled organization
important enough to merit such a forceful signal technology developments. As the organization’s
toward change, it’s highly likely that the core business strategy and AI capabilities mature
gravitational pull toward the status quo could over time, leaders should continually sharpen their
dampen outcomes for even the strongest strategy. goals, moving beyond staying competitive to
increasingly using AI and ML as competitive
differentiators.
Remain dynamic: Perpetually
iterate your AI strategy For more AI strategy recommendations:
Finally, developing an enterprisewide AI strategy • An innovation strategy powered by tech
that’s set up to fuel a differentiating core business
strategy is not a one-and-done exercise. • The AI Dossier: Top uses for AI in every major
Organizations should develop dynamic ways of industry — now and in the future
assessing their strategy to ensure it remains
responsive to ever-changing market and • A new language for digital transformation
10
Deloitte’s State of AI in the Enterprise, 4th Edition
Operations: How do you
bring transformation
into everyday work?
CORE LEADING PRACTICE DRIVE ONGOING QUALITY, INNOVATION, AND VALUE CREATION
THROUGH NEW OPERATING MODELS, ROLES, AND PROCESSES.
Key findings:
• Reimagine business workflows and roles. Organizations that have undergone significant changes to
workflows or added new roles are more than 1.5 times as likely to achieve outcomes to a high degree.
• No excuses: Document and follow MLOps. Organizations that document and follow MLOps processes
are twice as likely to achieve their goals to a high degree. They are also approximately two times as likely
to report being extremely prepared for risks associated with AI and nearly two times as confident that
they can deploy AI initiatives in a trustworthy way.
Pitfalls to avoid:
• Business leaders should allocate more time to solution design. Effectively redesigning processes and
how AI tools fit into workflows requires thoughtful attention.
• Don’t underestimate the unique maintenance needs of AI solutions. Establish and document robust
MLOps procedures to ensure ongoing quality and ethical delivery.
TECHNOLOGY CANNOT DELIVER documenting AI life cycle publication strategies,
transformative results unless organizations and updating workflows, roles, and team structures
reimagine how work gets done. Most leaders across the business.
today understand this intellectually; however,
survey results show a disconnect in putting it into To ensure quality AI solution development,
action: Across a variety of operational activities— enterprise adoption, and the most successful
both on the business side and within IT or data outcomes, organizations should rethink their
science teams—only about one-third of those operations from two key perspectives: across the
surveyed report that they have adopted leading business workflow, and within their IT and data
operational practices for AI. This includes adhering science team processes.
to a well-calibrated MLOps framework,
11
Becoming an AI-fueled organization
FIGURE 3
Leading AI operations practices
Percentage of respondents who selected "completely agree" to these statements
about operations
Transformers Pathseekers Underachievers Starters Total
56%
53%
49% 50%
44%
41% 42% 41%
38% 38%
32% 33% 31% 30%
24% 24% 25%
20%
15%
13%
My functional group My functional group My functional group has My functional group
follows a documented follows documented undergone significant changes has created new AI job
AI model life cycle MLOps procedures in how we create teams and roles/functions to
publication strategy when developing an manage workflows to take maximize AI
AI solution advantage of new technologies advancements
in the last five years
Source: The State of AI 4th Edition data analysis.
Deloitte Insights | deloitte.com/insights
When this happens, Michelle Lee, VP of Amazon
A call for leaders:
Machine Learning Solutions Lab, observes, “They
Business stakeholders then experience organizational inertia, because
should take ownership either the use case being addressed wasn’t
important enough, or there is an unwillingness to
of AI-fueled solutions
adopt a new and an unproven method.”
A successful AI solution should be conceived and
designed to fit within a new workflow created to Dr. Tian He, vice president and the head of JD
improve value delivery. To do this effectively, Logistics AI and Data Science, underscores that
business stakeholders should take a lead role, but “Most people learning especially machine learning
unfortunately, many misunderstand how to do this and deep learning just came out of school, and they
effectively. know the AI skills ... They’re technicians. But you
need to understand the business.”
This causes them to allocate too little
“We’ve seen a lot of AI projects where
time to rethink the broader operational
shifts needed to support successful people have implemented amazing
adoption of a value-creating solution.
models, but they’ve never seen the light
All too often, AI and ML development
of day because the business rejects the
teams are put in charge without a clear
view into the business processes they process changes that go along with it.”
are tasked with transforming.
— Rajen Sheth, VP of Google Cloud AI and Industry
Solutions
12
Deloitte’s State of AI in the Enterprise, 4th Edition
Only with an engaged partnership between the and operational managers is important to align the
business and AI and ML development teams, can a necessary processes for AI and ML to take hold.
new way of working emerge. Even when business
leaders understand their role, a lack of AI fluency While developing these processes is generally the
can inhibit their ability to collaborate effectively responsibility of IT and data science leadership, all
with the AI and ML development teams. Some stakeholders and senior leaders should be
organizations have found success in creating new concerned that these processes and standards are
roles to help translate between business in place and observed across the organization.
stakeholders and model development teams. In They are key to ensuring the ongoing quality of
these circumstances, an individual well-versed in models that are fueling critical business processes.
both business and analytics can serve as the bridge Data from our survey bear out just how important:
between overarching business strategic goals and Organizations that strongly agreed that MLOps
AI technical requirements.9 Our survey processes were followed were twice as likely to
demonstrates that efforts in creating new roles like achieve their goals, compared to the rest.
this can pay off. High-achieving surveyed Furthermore, these surveyed organizations were
organizations (Transformers and Pathseekers) also approximately two times more likely to report
were significantly more likely to create new roles feeling extremely prepared for risks associated
and functions to maximize AI advancements. with AI, and nearly twice as confident that they can
deploy AI initiatives in an ethical, trustworthy way.
MLOps: New capabilities
require new processes Rethinking ops: A catalyst
for AI transformation
In the early days of enterprise AI, initiatives took
place within localized teams and were contained Establishing the appropriate structures, roles, and
within business divisions. Models were frequently working relationships across an organization can
built on data scientists’ desktops and required be one of the most important steps in bringing an
relatively simple and smooth processes to AI transformation to life: “If I were to give one
maintain.10 Today, models are being deployed in piece of advice to a C-suite–level person looking at
the cloud and running mission-critical workloads. how to get this right in their organization, I would
As organizations reach this scale, the level and say, ‘Look at the organizational structure, because
complexity involved in perpetually developing, that can really facilitate the change,’” advises Phil
training, testing, deploying, monitoring, and Thomas, executive vice president of Customer
maintaining models have caught many Insights Data & Analytics at Scotiabank. “That for
organizations by surprise: Only 33% of all survey us was a massive accelerant in our journey—getting
respondents completely agree they have MLOps the org structure right and creating a culture of
processes in place. being a data-driven organization that’s accepting of
the use of AI.”
Not all data scientists are skilled in taking on an
engineering or operational mindset to manage this For more AI operations recommendations:
at scale. This is why a strong collaboration across
• ML Oops to MLOps
data scientists, engineering, application developers,
• Taking AI to the Next Level
13
Becoming an AI-fueled organization
Culture and change
management: Why is
valuable change so elusive?
CORE LEADING PRACTICE BUILD A TRUSTING, AGILE, DATA-FLUENT CULTURE AND INVEST IN
CHANGE MANAGEMENT TO SUPPORT NEW WAYS OF WORKING.
Key findings:
• Data fluency pays off. High-achieving organizations (Transformers and Pathseekers) are nearly three
times as likely to trust AI more than their own intuition, compared to low-achieving organizations
(Starters and Underachievers).
• Prioritize change management. Organizations that invest in change management to a high degree are
1.6 times more likely to report that AI initiatives exceed expectations and more than 1.5 times as likely to
achieve their desired goals.
• Fear can be an indicator of positive change if paired with supportive culture and
change management.
Pitfalls to avoid:
• Don’t take a one-size-fits-all approach to change management. Tailor your efforts to key audiences
and ensure a variety of resources are available to support new behaviors.
• Don’t expect change management to fix a poorly designed transformation. Thoughtfully designing
a new solution from the beginning can set the foundation for positive change.
OVER THE PAST few decades, the pace of Through interviews and survey data analysis, we
business and technology change has found the organizations with the strongest AI
quickened, requiring workers to adapt, outcomes tend to display some common
perpetually learn new skills, and make decisions characteristics, including high levels of
amid growing ambiguity. For many organizations, organizational trust, data fluency, and agility. And
these shifts have challenged a critical facet within to get there, investment in change management
their organization: their culture. has been key to successful AI transformation:
Executive interviewees repeatedly emphasized how
“Not to say that technical model
the cultural characteristics of their organizations
building is easy, but the biggest
either facilitate or hinder their AI-transformation
efforts. This aligned with a 2019 Deloitte Survey challenge is culture change.”
that found that organizations with the most data-
— Phil Thomas, executive vice president
driven cultures were twice as likely to significantly
of Customer Insights Data & Analytics at
exceed business goals.11 Scotiabank
14
Deloitte’s State of AI in the Enterprise, 4th Edition
Organizations that invest in change management machines replacing humans. But high-achievers
are 1.6 times as likely to report that AI initiatives also reported little desire to reduce employee
exceed expectations and more than 1.5 times as headcount as well as high investment in training
likely to achieve outcomes than those that don’t. A and change management. Wh |
178 | deloitte | DI_State-of-AI-in-the-enterprise-2nd-ed.pdf | State of AI in the Enterprise, 2nd Edition
Early adopters combine bullish enthusiasm with strategic investments
Early adopters combine bullish enthusiasm with strategic investments
Contents
Executive summary | 2
Activity, investment, and positive results | 3
To maximize value, early adopters should become
risk and change management experts | 9
Early adopters want more talent, and need a
better mix of it | 14
Enthusiastic early adopters can take the next step
by getting serious | 17
Endnotes | 21
1
State of AI in the Enterprise, 2nd Edition
Executive summary
FOR THE SECOND straight year, Deloitte risks. Project selection and managing return on
surveyed executives knowledgeable about cog- investment are also critical.
nitive technologies and artificial intelligence,1 3. Early adopters need the right mix of
representing companies that are testing and imple- talent—not just technical skills—to ac-
menting them today. We found that these early celerate their progress. They are short of
adopters2 remain bullish on cognitive technologies’ AI researchers and programmers but also need
value. As in last year’s survey, the level of support business leaders who can select the best use
for AI is truly extraordinary. Our analysis uncov- cases. To garner this talent, they are training
ered three main findings: their current workforce, but many feel the need
to replace existing workers with new people.
1. Early adopters are ramping up their AI Early adopters also may need a strategic ap-
investments, launching more initiatives, proach to talent that automates what machines
and getting positive returns. Cloud-based do best, while still capitalizing on human judg-
cognitive services are increasing adoption by ment and creativity.
reducing the investment and expertise required
to get started. These findings illustrate that cognitive tech-
2. Companies should improve risk and nologies hold enticing promise, some of which is
change management. This includes reducing being fulfilled today. However, AI technologies may
cybersecurity vulnerabilities—which can slow or deliver their best returns when companies balance
even stop AI initiatives—and managing ethical excitement over their potential with the ability to
execute.
METHODOLOGY
To obtain a cross-industry view of how organizations are adopting and benefiting from cognitive
computing/AI, Deloitte surveyed 1,100 IT and line-of-business executives from US-based companies
in Q3 2018. All respondents were required to be knowledgeable about their company’s use of
cognitive technologies/artificial intelligence, and 90 percent have direct involvement with their
company’s AI strategy, spending, implementation, and/or decision-making. The respondents
represent 10 industries, with 17 percent coming from the technology industry. Fifty-four percent are
line-of-business executives, with the rest IT executives. Sixty-four percent are C-level executives—
including CEOs, presidents, and owners (30 percent), along with CIOs and CTOs (27 percent)—and 36
percent are executives below the C-level.3
22
Early adopters combine bullish enthusiasm with strategic investments
Activity, investment, and
positive results
AY EAR LATER, AND the thrill isn’t gone. In
Deloitte’s 2017 cognitive survey, we were
struck by early adopters’ enthusiasm for
cognitive technologies.4 That excitement owed
much to the returns they said cognitive technolo-
gies were generating: 83 percent stated they were
seeing either “moderate” or “substantial” benefits.
Respondents also said they expected that cognitive
technologies would change both their companies
and their industries rapidly. In 2018, respondents
remain enthusiastic about the value cognitive tech- layers of abstract variables. Deep learning
nologies bring. Their companies are investing in models are excellent for image and speech recog-
foundational cognitive capabilities, and using them nition but are difficult or impossible for humans
with more skill. to interpret. New technologies are making it
easier for companies to launch deep-learning
projects, and adoption is increasing. Among
Higher adoption, multiple
our respondents, 50 percent said they use deep
options learning, a 16 point increase from 2017—the
largest jump among all cognitive technologies.
Compared with their counterparts in typical • Natural language processing is the ability to
companies,5 our early-adopter respondents have extract or generate meaning and intent from text
high—and growing—penetration rates of key cogni- in a readable, stylistically natural, and grammat-
tive technologies: ically correct form. NLP powers the voice-based
interface for virtual assistants and chatbots, and
• Machine learning is the ability of statistical the technology is increasingly used to query data
models to develop capabilities and improve sets as well.6 Sixty-two percent of respondents
their performance over time without the need to have adopted NLP, up from 53 percent last year.
follow explicitly programmed instructions. Most • Computer vision is the ability to extract
cognitive technologies are based on machine meaning and intent out of visual elements,
learning and its more complex progeny, deep whether characters (in the case of document
learning. That includes computer vision and digitization) or the categorization of content
natural language processing (NLP). Machine- in images such as faces, objects, scenes, and
learning adoption was already high at 58 percent activities. The technology behind facial recogni-
in 2017, and it grew by 5 percentage points in tion—computer vision—is a part of consumers’
2018. everyday lives. For example, some mobile
• Deep learning is a complex form of machine phones permit their owners to log in simply by
learning involving neural networks, with many
33
State of AI in the Enterprise, 2nd Edition
looking at them, via facial recognition.7 Com- finance, video analysis in brand management, and
puter vision technology “drives” driverless cars trouble ticket analysis in customer service. The
and animates cashier-less Amazon Go stores.8 need for companies to develop bespoke cognitive
Computer vision has also gone mainstream with initiatives will likely decline as similar services
our survey respondents, 57 percent of whom say enter the market.
their company uses it today. Off-the-shelf can go only so far, however. Many
companies will likely need to develop customized
What’s behind the growth of cognitive technolo- solutions to meet their lofty expectations for cog-
gies among early adopters,9 especially the popularity nitive technologies. Here, too, there are tools to
of sophisticated technologies such as deep learning? accelerate adoption. Many of the big cloud providers
One answer is investment. Thirty-seven percent of offer AI through an as-a-service model: Instead of
respondents say their companies have invested having to build their own infrastructure and train
US$5 million or more in cognitive technologies. algorithms, companies can tap into the technologies
Another reason is that companies have more ways
FIGURE 1
to acquire cognitive capabilities, and they are taking
Enterprise software represents the
advantage. Nearly 60 percent are taking what is
most popular—and easiest—path to AI
perhaps the easiest path:10 using enterprise soft-
ware with AI “baked in” (see figure 1). Respondents who report their company
uses this method of acquiring/developing AI
More respondents gain cognitive capabilities
through enterprise software, such as CRM or ERP
systems, than by any other method. These systems
Enterprise
59%
have the advantage of access to immense data sets software with AI
(often their own customers’ data), and can often be
used “out of the box” by employees with no special- 59%
Codevelopment
ized knowledge. 53%
with partners
The cognitive tools available through enterprise
59%
software are often focused on specific, job-related
tasks. While this can make them less flexible, they
Cloud-based AI 49%
may be impactful nonetheless. For example, Sales-
90%
force Einstein can help sales reps determine which
leads are most likely to convert to sales, and the 59%
Open-source
optimal time of day to contact those prospects. development tools 49%
Moreover, vendors continually develop advanced 59%
tools, which are gradually integrated into the soft-
ware. Salesforce recently developed an advanced Automated
46%
machine learning
NLP model for handling multiple use cases that
typically require different models.11
59%
The “easy path” will likely become even more
Data science
44%
attractive as software vendors and cloud providers modeling tools
develop AI offerings tailored to business functions. 59%
Google recently announced a set of prepackaged AI
services aimed at contact centers and HR depart- Crowdsourced 39%
development
ments.12 SAP’s AI capabilities, which it collectively
calls “Leonardo Machine Learning,” also include
specific solutions such as cash management in Source: Deloitte State of AI in the Enterprise, 2nd Edition, 2018.
44
Early adopters combine bullish enthusiasm with strategic investments
Cognitive technologies are
they need right away, and pay only for what they use.
a necessity, not an option
According to a recent Deloitte study, 39 percent of
companies prefer to acquire advanced technologies
such as AI through cloud-based services, versus Many early adopters are investing in cognitive
15 percent that prefer an on-premise solution.13 technologies to improve their competitiveness.
Indeed, the appeal of the AI-as-a-service model is Sixty-three percent of surveyed executives said their
reflected in its annual global growth rate, which is AI initiatives are needed to catch up with their rivals
estimated at a remarkable 48.2 percent.14 or, at best, to open a narrow lead (see figure 2).
Cloud-based deep-learning services can give And the linkage between adept application of AI
companies access to immense—and previously and competitive advantage appears to be growing
costly—computing power necessary to extract stronger. Eleven percent said that adopting AI is
insights from unstructured data. They can also of “critical strategic importance” today, but 42
manage large data sets and accelerate app develop- percent believe it will be critical two years from
ment with pretrained models. now. This is a small window for companies to hone
While there are myriad ways for companies to their AI strategies and skills, and they believe their
access ready-made AI or develop their own, many success depends upon getting it right. Executives
also seek outside expertise. Fifty-three percent of are becoming more realistic about the time this will
respondents codevelop cognitive technologies with require, however. In our 2018 survey, 56 percent
partners, and nearly 40 percent use crowdsourcing of respondents said cognitive technologies would
communities such as GitHub. transform their companies within three years, down
Through cloud services and enterprise soft- from 76 percent last year. The same was true of in-
ware, companies can try cognitive technologies dustrywide transformation: 37 percent of our 2018
and even deploy them widely, with low initial cost respondents think it will happen within three years,
and minimal risk. The growing number of cloud- 20 points lower than in 2017. We believe executives
based options may explain the spike in pilots and are acknowledging the complexity of using cogni-
implementations between 2017 and 2018. Fifty-five tive technologies to drive change across lines of
percent of executives say their companies have business, without despairing of attaining that goal.
launched six or more pilots (up from 35 percent in
2017), and nearly the same percentage (58 percent)
Earning while they learn
claim that they have undertaken six or more full
implementations (up from 32 percent). Many companies’ AI goals extend well beyond
ROI. Positive ROI, however, can build momentum
for future investment and generate support for ex-
FIGURE 2
AI helps organizations keep up with the (Dow) Joneses
Relative to competitors, respondents say their company’s adoption of AI has allowed them to . . .
16% 20% 27% 28% 9%
Catch up Stay on par Edge slightly ahead Widen a lead Leapfrog ahead
Source: Deloitte State of AI in the Enterprise, 2nd Edition, 2018.
Deloitte Insights | deloitte.com/insights
55
0 20 40 60 80 100
25%
27%
22%
24%
25%
24%
ecutive champions of AI, and the technologies seem tions and delivering superior customer experience.
to be delivering. In our survey, 82 percent said they Netflix found that if customers search for a movie
have gained a financial return from their AI invest- for more than 90 seconds, they give up. By using AI
ments. For companies across all industries, the to improve search results, Netflix prevents frustra-
median return on investment from cognitive tech- tion and customer churn, saving US$1 billion a year
nologies is 17 percent. Some are more adept than in potential lost revenue.17
others at turning investment into financial benefits Robust returns are not limited to tech companies.
(see figure 3). Both established manufacturers and innovative
While these returns are estimates based on self- startups are using AI to make manufacturing more
reported data, they show that executives across efficient. For example, industrial firms, such as GE
industries feel they’re getting value from cogni- and Siemens, are taking advantage of the data in
tive technologies. Tech companies are spending “digital twins” of their machines to identify trends
significantly on cognitive, and getting a strong and anomalies, and to predict failures.18
return. They are also the driving force behind cog- Companies such as these are using AI to improve
nitive technologies, developing them for a market business processes, which are prominent benefits
already estimated at US$19.1 billion globally.15 companies seek. In fact, our survey findings suggest
This includes giants such as Google, Microsoft, and that companies are placing increased emphasis on
Facebook, and literally thousands of startups.16 AI internal operations (see figure 4).
has also generated returns by improving opera-
Low AI investment/high returns High AI investment/high returns
Industrial products and services
Technology/media and
entertainment/telecommunications
Professional services
Financial services and insurance
Consumer products
Government/public sector
(including education)
Life sciences and health care
Low AI investment/low returns High AI investment/low returns
6
tnemtsevni
no
nruteR
State of AI in the Enterprise, 2nd Edition
FIGURE 3
Everyone’s winning, but some industries are winning bigger
AI investment and ROI: Relative landscape of industries
22%
20%
18%
16%
14%
12%
10%
Lower investment Median Higher investment
Note: The dotted lines in the graph represent the median ROI and median AI investment for all respondents, cross-industry.
Source: Deloitte State of AI in the Enterprise, 2nd Edition, 2018.
Deloitte Insights | deloitte.com/insights
6
Early adopters combine bullish enthusiasm with strategic investments
FIGURE 4
AI’s leading benefits are enhanced products and processes—
and better decisions
Respondents rating each a top-three AI benefit for their company
2017 2018
Enhance current products
51%
44%
Optimize internal operations
36%
42%
Make better decisions
35%
35%
Optimize external operations
30%
31%
55%
Free workers to be more creative
36%
31%
Create new products
32%
27%
Capture and apply scarce knowledge
25%
27%
Reduce headcount through automation
22%
24%
Pursue new markets
25%
24%
Source: Deloitte State of AI in the Enterprise, 2nd Edition, 2018.
Deloitte Insights | deloitte.com/insights
This shift toward internal operations has been Health care and life sciences companies are in-
accompanied by a somewhat reduced emphasis on vesting in AI but, according to our data, have less to
integrating AI into existing products and services, show for it. Certainly, some health care “big bang”
although that remains the most popular objective. projects have disappointed thus far. However, ad-
In fact, operational change is often required before vances in fields as diverse as radiology and hospital
such integration can take place. Our respondents claims management show that AI offers substantial
may be realizing that they should make operational potential for value in health care,19 despite some
changes first. high-profile stumbles. For example, in a recent
77
State of AI in the Enterprise, 2nd Edition
study, deep-learning neural networks identified Earlier, we noted that eight in 10 surveyed exec-
breast cancer tumors with 100 percent accuracy utives claim positive ROI from their companies’ AI
by analyzing pathology images.20 Such advances, efforts. However, we should view ROI claims with
however, are thus far only in the lab and will take a bit of caution: Less than 50 percent of surveyed
time before entering clinical practice. companies measure key performance indicators
necessary for gauging financial returns accurately.
These indicators include critical elements such as
(Mostly) rational exuberance
project budget/cost, ROI, and targets for produc-
Despite the hype AI generates, many executives tivity, cost savings, revenue, and customers (such
are excited—not wallowing in a trough of disil- as satisfaction and retention). This lack of mea-
lusionment. That’s translating into investment. surement gets to the heart of a significant problem
Eighty-eight percent of companies surveyed plan with cognitive implementations: They are often not
to increase spending on cognitive technologies managed with the same rigor that companies use
in the coming year; 54 percent say they will boost with more mature technologies.
spending by 10 percent or more.
8
Early adopters combine bullish enthusiasm with strategic investments
To maximize value, early
adopters should become risk
and change management
experts
BUSINESS AND TECHNOLOGY leaders con- low levels of experience with them, it’s unsurprising
front an array of challenges as they seek to that this was the most-cited challenge. Integration
create business value with artificial intel- into the business is a challenge for technologies
ligence. Many respondents cited implementation, in general, but it may be particularly problematic
integration into roles and functions, and measuring with AI given the impact it can have on knowledge-
and proving the business value of AI solutions as worker tasks and skills.
top challenges of AI initiatives (see figure 5). Imple- Companies sometimes struggle in AI projects
mentation can be a challenge with any technology, to navigate the “last mile” of behavior change.21 An
but given the relative newness of AI tools and the example we have seen is an organization that built a
FIGURE 5
Many early adopters struggle with the basics
Top challenges for AI initiatives: Ranked 1–3, where 1 is greatest challenge
Ranked 1 Ranked 2 Ranked 3 Ranked top three
Implementation challenges 13% 14% 12% 39%
Integrating AI into the company’s
14% 13% 12% 39%
roles and functions
Data issues (e.g., data privacy,
16% 13% 10% 39%
accessing and integrating data)
Cost of AI technologies/
13% 12% 11% 36%
solution development
Lack of skills 11% 10% 10% 31%
Challenges in measuring and
10% 11% 9% 30%
proving business value
Source: Deloitte State of AI in the Enterprise, 2nd Edition, 2018.
Deloitte Insights | deloitte.com/insights
99
State of AI in the Enterprise, 2nd Edition
machine-learning system to support the sales team system. (It is possible to automate this analysis, but
by predicting which prospects were likely to convert that would be an AI project of its own.)
and which customers were likely to churn. Though Getting the data required for an AI project,
the system worked as planned, the sales team was preparing it for analysis, protecting privacy, and
initially unprepared to accept its recommendations. ensuring security can be time-consuming and costly
The team had not been closely involved in the devel- for companies. Adding to the challenge is that data—
at least some of it— is
Twenty-three percent of respondents often needed before it is
even possible to conduct
ranked “cybersecurity vulnerabilities” as
a proof of concept. We
have seen companies
their No. 1 overall AI/cognitive concern.
that, because they had
not fully considered the
opment of the solution and neither understood nor difficulty of obtaining the data they need, decided to
trusted the results it produced. One way to avoid shelve projects and disband teams until they were
this problem is to involve business owners closely able to lay the proper data foundation.
throughout the development process so they can Some organizations also struggle to articulate
better understand what is being delivered. a business case or to define success for AI projects.
Anyone following business news about AI knows This may be because AI is viewed as experimental.
about the critical role played by data. Survey re- Sometimes it is because machine learning—one of
spondents consider “data issues” as one of the top the most widely used AI technologies—is inherently
challenges for their companies’ AI initiatives. There probabilistic, meaning that a new system’s ultimate
are numerous reasons for this. Some AI systems, performance can be difficult to estimate precisely.
such as virtual assistants to enable customer self- And sometimes it’s because the group charged with
service, require data from multiple systems that developing an AI solution is unaccustomed to devel-
may never have been integrated before. Customer oping business cases to justify its work.
information may reside in one system, financial
data in another, and virtual assistant training and
Managing risks of AI
configuration data in a third. AI creates a need for
data integration that a company may have managed It is a fact of life that novel situations often
to avoid until now. This can be especially chal- present new risks. The same is true of emerging
lenging in a company that has grown by acquisition technologies such as AI. Executives are concerned
and maintains multiple, unintegrated systems of about a host of risks associated with AI technologies
diverse vintages. (see figure 6). Some of the risks are typical of those
Another challenge for companies is that the type associated with any information technology; others
of data required for some AI projects is different are as unique as AI technology itself.
from the data with which they’re accustomed to
working. For example, some solutions depend on CYBER RISK
access to significant amounts of unstructured data Chief among the AI risks that concern executives
that may have been retained for record-keeping are cyber risks, which ranked as a top-three concern
but was never intended for analysis. In one virtual for half of our survey respondents (see figure 6).
assistant project we know of, the team needed to In fact, 23 percent of respondents ranked “cyber-
review thousands of recorded phone calls to identify security vulnerabilities” as their No. 1 overall AI/
common themes with which to derive rules for the cognitive concern. This apprehension is probably
1100
Early adopters combine bullish enthusiasm with strategic investments
FIGURE 6
Cybersecurity heads the lists of AI-related concerns
Potential AI risks of top concern to companies: Ranked 1–3, where 1 is greatest concern
Ranked 1 Ranked 2 Ranked 3 Ranked top three
Cybersecurity vulnerabilities
23% 15% 13% 51%
of AI
Making the wrong strategic
16% 13% 14% 43%
decisions based on AI
Legal responsibility for decisions/
11% 15% 13% 39%
actions made by AI systems
Failure of AI system in a mission-
13% 14% 12% 39%
critical or life-or-death context
Regulatory noncompliance risk 12% 15% 10% 37%
Erosion of customer trust from 11% 11% 11% 33%
AI failures
10% 12% 10% 32%
Ethical risks of AI
Source: Deloitte State of AI in the Enterprise, 2nd Edition, 2018.
Deloitte Insights | deloitte.com/insights
well placed: While any new technology has certain AI has also been used recently to create fake
vulnerabilities, the cyber-related liabilities sur- photos and videos of celebrities and politicians.
facing for certain AI technologies seem particularly While there are also techniques for identifying fakes,
vexing. it appears that technologies may fuel an arms race
Researchers have discovered that some ma- of fake image development versus detection. Given
chine-learning models have difficulty detecting the prominence of AI-based image recognition, this
adversarial input—that is, data constructed specifi- area is likely to be a cyber-risk battleground in the
cally to deceive the model. This is how one research future.
team fooled a vision algorithm into classifying as a There is evidence that cyber-risk concerns are
computer what appeared to be a picture of a cat.22 slowing or pausing AI projects at some compa-
The process of training machine-learning models nies. In addition, one in five respondents said they
can itself be manipulated with adversarial data. decided not to launch AI initiatives due to cyberse-
By intentionally feeding incorrect data into a self- curity worries (see figure 7).
learning facial recognition algorithm, for instance, Executives are commonly concerned about the
attackers can impersonate victims via biometric safety and reliability of AI systems as well. Forty-
authentication systems.23 In some cases, machine- three percent of respondents rated “making the
learning technology may expose a company to the wrong strategic decisions based on AI/cognitive rec-
risk of intellectual property theft. By automatically ommendations” as a top-three concern (see figure
generating large numbers of interactions with a 6). Nearly as many cited failure of an AI system in
machine-learning-based system and analyzing the a mission-critical or life-or-death situation. Placing
patterns of responses it generates, hackers could re- strategic decisions or mission-critical actions en-
verse-engineer the model or the training data itself. tirely in the hands of an AI system would certainly
1111
State of AI in the Enterprise, 2nd Edition
FIGURE 7
Cybersecurity threats are giving some companies pause
Effect of cybersecurity concerns on companies
Moved ahead with AI initiatives despite cybersecurity concerns
36%
Experienced a cybersecurity breach relating to AI initiatives within the last two years
32%
Slowed an AI initiative in order to address cybersecurity concerns
30%
Decided not to start an AI initiative due to cybersecurity concerns
20%
Canceled or halted an in-progress AI initiative due to cybersecurity concerns
16%
Source: Deloitte State of AI in the Enterprise, 2nd Edition, 2018.
Deloitte Insights | deloitte.com/insights
entail special risks. Entrusting AI systems with such and operational risks associated with these systems.
responsibilities remains rare today, however. A Complicating matters are questions surrounding
prominent exception is the use of AI in autonomous who can be held liable in the event of an AI-related
vehicles: The technology has been implicated in crime or mishap. How liability is assigned in these
several accidents, some fatal, during testing.24 cases is a topic of ongoing discussion.26
Another element of cyber risk that companies Two themes are particularly salient when
should consider is how much data—and what it comes to AI and regulatory risk: privacy and
kind of data—they are willing to put into public explainability. Because data is so critical to AI, com-
cloud environments, allowing them to use cogni- panies seeking to apply the technology are often
tive technologies to analyze much larger data sets hungry for the stuff. Privacy regulations governing
than private clouds. Analysis of sensitive customer personal data may dampen their appetite, though:
and financial data can yield valuable insights, but The General Data Protection Regulation (GDPR),
companies should weigh the perceived risks with which has recently come into force in Europe, sets
the benefits. A recent Deloitte study found that the privacy rules that require careful implementation.
more experience organizations have with cloud GDPR also mandates that companies using per-
computing, the more comfortable they are putting sonal data to make automated decisions affecting
sensitive data into public clouds.25 people must be able to explain the logic behind the
decision-making process.27 Guidance published by
LEGAL AND REGULATORY RISKS the US Federal Reserve (SR 11-7) affects US banking
Products and systems of all types, including IT similarly: It requires that the behavior of computer
systems, present a range of legal and regulatory risks. models be explained.28 What makes these regula-
As a result, it is unsurprising that four in 10 survey tions challenging for some AI adopters is the growing
respondents indicate a high degree of concern complexity of machine learning and the increasing
about the legal and regulatory risks associated with popularity of deep-learning neural networks, which
AI systems. Because not all methods of validating can behave like black boxes, often generating highly
AI systems’ accuracy and performance are reliable, accurate results without an explanation of how
companies will need to manage the legal, regulatory, these results were computed. Many tech companies
1122
Early adopters combine bullish enthusiasm with strategic investments
and government agencies are pouring resources Some of the ethical risks that resonated with
into improving the “explainability” of deep-learning our respondents are linked to the aforementioned
neural networks.29 cyber-safety and regulatory issues: unintended
consequences, misuse of personal data, and lack of
ETHICS AND REPUTATION explanation for AI-powered decisions. But there is
For most of our respondents, ethical risks are one concern that has achieved special prominence
not a top-of-mind information technology concern. in recent years, and ranked second among our re-
And while ethical risks ranked at the bottom of risk spondents’ ranking of ethical risks: bias.
concerns in our survey, about a third of executives Today, algorithms are commonly used to help
did cite them as a top concern. make many important decisions, such as granting
In a deeper look at potential ethical risks, sur- credit, detecting crime, and assigning punishment.
veyed executives revealed a wide range of concerns. Biased algorithms, or machine-learning models
At the top of the list is AI’s power to help create or trained on biased data, can generate discriminatory
spread false information. This may be due to the or offensive results. For example, one study found
attention that social-media-driven “fake news” re- that ads for high-paying jobs were shown more
ceived in the 2016 US elections. often to men than to women.30
1133
State of AI in the Enterprise, 2nd Edition
Early adopters want more
talent, and need a better
mix of it
DO EARLY ADOPTERS have the talent to need more skilled people. Thirty percent said they
develop and deploy cognitive solutions? face a major (23 percent) or extreme (7 percent)
The overall survey results suggest both a skills gap. Another 39 percent said their gap is
considerable amount of talent already and a strong “moderate.” Interestingly, the most advanced com-
demand for more. “Lack of AI/cognitive skills” was panies in our survey feel the skills gap acutely.31 The
a top-three concern for 31 percent of respondents— limitations of their technical skills may be exposed
below such issues as implementation, integration, as they launch more AI solutions, and as those solu-
and data. A skills shortage was identified as the tions increase in complexity and scale.
biggest challenge in moving from prototypes to Some skills are needed more than others (see
full production deployments for only 8 percent figure 8).32 Respondents report the highest level of
of respondents. need for AI researchers to invent new kinds of AI al-
Companies generally feel that they have sub- gorithms and systems. This suggests an aggressive
stantial AI capabilities. About four in 10 executives level of ambition for the technology. In addition, 28
report their companies have a high level of sophis- percent said they need AI software developers, 24
tication in managing and maintaining AI solutions, percent need data scientists, and roughly similar
selecting AI technologies and technology suppliers, percentages need user-experience designers,
integrating AI technology into the existing IT en- change-management experts, project managers,
vironment, identifying valuable applications of AI, business leaders, and subject-matter experts. Sixty-
building AI solutions, and hiring and managing one percent a |
179 | deloitte | DI_government-execs-on-AI.pdf | FEATURE
Government executives on AI
Surveying how the public sector is approaching an
AI-enabled future
William Eggers, Sushumna Agarwal, and Mahesh Kelkar
THE DELOITTE CENTER FOR GOVERNMENT INSIGHTS
Government executives on AI: Surveying how the public sector is approaching an AI-enabled future
As government uses artificial intelligence more, how can the experience of
early adopters guide other public sector organizations?
AS ONE OF the hottest technologies of recent exponential growth in processing power seem to
years, artificial intelligence (AI) has started finally be fast-tracking AI into the mainstream.
penetrating both the US public and the
private sectors—though to differing degrees. While This growing usage is reflected in the AI initiatives
the private sector seems bullish on AI, the public being undertaken by public sector organizations
sector’s approach appears tempered with more across levels. For instance, in February 2019, US
caution—a Deloitte survey of select early adopters President Donald Trump signed an executive order
of AI shows high concern around the potential risks to create the American AI Initiative, which aims to
of AI among public sector organizations (see the prioritize and guide AI development in the United
sidebar “About the survey”). The findings in this States.2 This builds on other federal AI initiatives,
study show the approaches and experiences of such as the Select Committee on AI.3 At the state
these early adopters of AI in the public sector. They level, the government of New Jersey has set up an
give a peek into how public sector organizations are innovation training platform to educate
approaching AI; and how the approaches, in many government workers about new technologies such
cases, differ from those of their private as AI and blockchain.4
sector counterparts.
As we have documented in previous studies,5 the
AI is not completely new to the public sector. The number of AI use cases in the public sector has
first AI contract was awarded in 1985 by the US increased manifold. As AI usage in the public
Social Security Administration,1 but the technology sector continues to grow, we sought to answer
still wasn’t advanced enough to become common in questions such as, how do early adopters in the
the following decades. Now, the growing ubiquity public sector perceive AI? What approaches are
of digital technologies, advances in the ability to these early adopters pursuing? Do these
store and analyze massive amounts of data, and approaches differ from those of the private sector?
ABOUT THE SURVEY
To gain insights into the experiences of early users of AI, Deloitte surveyed 1,100 executives from
US-based organizations across 10 industries currently using AI, in the third quarter of 2018. About 10
percent of the respondents were from the federal government, state government, higher education,
defense, international donor organizations, public health and social services, public transportation,
and security and justice—a collection of entities we refer to as “public sector.” This sample allowed
us to examine how the AI approach of early adopters in the public sector compares with that of
private industry.
The survey required the respondent’s organization to be using at least one AI technology and to have
built (or be building) at least one AI prototype system or full implementation/production system.
Also, respondents were required to be knowledgeable about their organization’s use of AI.
2
Government executives on AI: Surveying how the public sector is approaching an AI-enabled future
WHAT IS ARTIFICIAL INTELLIGENCE? This caution has led to new ways of developing AI
solutions in the public sector, such as prototyping
AI technologies are defined as those that
AI projects in controlled environments and
can perform or augment tasks, better
codeveloping AI solutions with partners.
inform decisions, and accomplish objectives
that have traditionally required human
intelligence, such as planning, reasoning
Early adopters believe AI can
using partial or uncertain information, and
learning.6 AI technologies include robotic be critical to organizational
process automation, natural language success
processing, machine learning, computer
vision, speech recognition, deep learning, Many early adopters in the public sector expect AI
and intelligent robotics.
technologies to become increasingly important in
the coming years. About 57 percent of early
adopters surveyed believe that AI is “very” or
The survey results reveal that the public sector “critically” important to their organization’s
early adopter respondents generally feel positive success today, and 74 percent of respondents
about their early AI experiences. They are using AI believe it will be in the next two years (figure 1).
to augment human capabilities, generating demand
for newer skills. However, many still lag other Some of the most popular AI uses cases in the
industries due to reasons such as lack of public sector focus on quality control issues
investment and skilled talent. Also, they tend to be (detecting defects and finding errors in software
understandably more cautious than other code), workforce management (recruiting and
industries due to ethical risks associated with AI. training), and cybersecurity (figure 2).
FIGURE 1
Early adopters in the public sector believe AI will become increasingly
important to their organization’s success
How would you rate the strategic importance of adopting/using AI/cognitive to your organization’s
overall success?
Critically important Very important None/minimal/somewhat important
NOW
Public sector
11% 46% 43%
Private sector 11% 60% 30%
IN TWO YEARS
Public sector 31% 43% 25%
Private sector 43% 40% 17%
Source: Deloitte analysis of the State of AI in the Enterprise Survey.
Deloitte Insights | deloitte.com/insights
3
Government executives on AI: Surveying how the public sector is approaching an AI-enabled future
FIGURE 2
The top AI uses cases in the public sector are quality control, workforce
management, cybersecurity
In which of the following areas is your organization using AI/cognitive technologies?
Quality control (e.g., detecting defects, finding errors in software code)
47%
Workforce management (e.g., recruiting and training)
38%
Cybersecurity
38%
IT automation (e.g., network and cloud management)
35%
Predictive analytics (e.g., predicting and preventing downtime, predicting medical outcomes)
35%
Risk management (e.g., detecting and preventing fraud)
31%
Customer service (e.g., chatbots and virtual assistants)
31%
Decision support (e.g., diagnosis)
30%
Tax, audit, and compliance (e.g., anomaly detection, document discovery)
28%
Connected equipment, devices, products (e.g., devices that learn user preferences, self-driving cars)
25%
Source: Deloitte analysis of the State of AI in the Enterprise Survey.
Deloitte Insights | deloitte.com/insights
Early adopters see AI as a Innovation Center, or iCenter, uses chatbots to aid
internal IT help desk personnel. The iCenter found
way to augment human
that 80 to 90 percent of the IT help desk tickets
capabilities
were for password resets. By leveraging chatbots
for such routine requests, the iCenter is allowing
The survey results suggest that AI is primarily
workers to deal with more complex issues.7
being used to make the work of humans more
effective rather than automate it altogether. As AI gets integrated into public sector
Freeing up workers to be more creative by organizations and routine tasks are automated,
automating tasks has been identified among the workers will need to learn to work with these
top three benefits of AI by early adopters surveyed,
technologies or will need to perform new and
while reducing headcount through automation is
different work. As many as 76 percent of early
near the bottom (figure 3).
adopter respondents in the public sector said
One area where AI is being used in many human workers and AI will augment each other to
governments to free workers from repetitive tasks produce new ways of working (figure 4).
is customer service chatbots. North Carolina’s
4
Government executives on AI: Surveying how the public sector is approaching an AI-enabled future
FIGURE 3
Freeing workers to be more creative is one of the top benefits of AI
What do you view as the primary benefit of AI/cognitive technology for your organization?
(Percent of respondents rating each category as one of the top three benefits of AI.)
Public sector Private sector
Enhances features, functions, and/or performance of our products and services
47%
43%
Optimizes internal business operations
41%
42%
Frees up workers to be more creative by automating tasks
35%
31%
Helps us make better decisions
34%
35%
Captures and applies scarce knowledge where needed
28%
27%
Reduces headcount through automation
21%
24%
Creates new products
20%
28%
Source: Deloitte analysis of the State of AI in the Enterprise Survey.
Deloitte Insights | deloitte.com/insights
The use of AI is already generating demand for new adopters. (See the sidebar “Starter, skilled, and
skills in the public sector, beyond technology and seasoned adopters of AI” to learn more about this
technical skills. While 34 percent of early adopters classification.)
surveyed are looking for software developers and
23 percent for data scientists, a sizable 30 percent
cite the need for business leaders, and 23 percent The use of AI is already
for change management experts (figure 5).
generating demand for
new skills in the public
The public sector lags other
sector beyond technology
industries in AI adoption
and technical skills.
Compared with other industries, the public sector
has the highest proportion of “starters,” those at
the beginning of their AI journey, and the lowest
proportion of “seasoned,” or experienced, AI
5
Government executives on AI: Surveying how the public sector is approaching an AI-enabled future
FIGURE 4
Early adopters in the public sector believe that AI will augment human labor
What do you view as the primary benefit of AI/cognitive technology for your organization?
Public sector Private sector
PERCENT SAYING THEY AGREE/STRONGLY AGREE
AI empowers people at our organization to make better decisions
68%
80%
Human workers and AI technologies will augment each other to produce new ways of working
76%
78%
Source: Deloitte analysis of the State of AI in the Enterprise Survey.
Deloitte Insights | deloitte.com/insights
FIGURE 5
Early adopters are seeking new skills to work with AI technologies
What kinds of skills/capabilities are most needed to fill your organization’s skills gap?
Software developers
34%
Business leaders
30%
Project managers
24%
Change management/transformation experts
23%
Data scientists
23%
AI researchers
23%
Subject matter experts
21%
User experience designers
15%
Source: Deloitte analysis of the State of AI in the Enterprise Survey.
Deloitte Insights | deloitte.com/insights
6
Government executives on AI: Surveying how the public sector is approaching an AI-enabled future
STARTER, SKILLED, AND SEASONED ADOPTERS OF AI
Some adopters of AI are further along in their efforts than others. To aid our comparison, we
identified three distinct segments at different levels of maturity. The “seasoned” (24 percent of all
respondents) is the most experienced cohort, at the leading edge of AI adoption maturity. They
have undertaken a number of AI production deployments; they also report that they’ve developed
a high level of expertise in selecting AI technologies and suppliers, identifying use cases, building
and managing AI solutions, integrating AI into their IT environment and business processes, and
hiring and managing AI technical staff. In the middle is the “skilled” cohort (45 percent). They have
launched multiple AI production systems but are not yet as AI-mature as the seasoned adopters.
They lag in their number of AI implementations, level of AI expertise, or both. At the low end of the
spectrum are “starters” (31 percent), which are just dipping their toes into AI adoption and have not
yet developed solid proficiency in building, integrating, and managing AI solutions.8
Only 14 percent of public sector adopters surveyed The public sector is also behind other industries
are classified as seasoned, whereas 45 percent are in integrating AI technology into business
still classified as starters. Meanwhile, in industries processes and the IT environment, and finding the
leading the way in AI such as financial services and right use cases for AI (figure 7). The one area in
technology, media, and telecommunications, which the public sector is on par with other
around 30 percent of respondents are classified as industries is selecting AI technologies and
seasoned AI adopters (figure 6). However, some technology suppliers, with about 44 percent of
pockets in the public sector, such as defense and respondents from both sectors saying they were
national security, are outliers, since they have been mature in this area (figure 7).
developing and using AI for many years.
7
Government executives on AI: Surveying how the public sector is approaching an AI-enabled future
FIGURE 6
Public sector is at the lower end of the AI maturity curve
Starters Skilled Seasoned
Overall
31% 45% 24%
Financial services and insurance
27% 42% 31%
Technology, media, and telecommunications
29% 41% 30%
Energy, resources, and utilities (including oil, gas, chemicals)
29% 41% 29%
Professional services
33% 41% 26%
Industrial products and services (aerospace, construction, industrial manufacturing)
37% 40% 22%
Consumer products
25% 57% 18%
Life sciences and health care
32% 52% 15%
Government/public sector
45% 41% 14%
Source: Deloitte analysis of the State of AI in the Enterprise Survey.
Deloitte Insights | deloitte.com/insights
FIGURE 7
The public sector’s AI activity is less sophisticated than that of the private sector
How would you characterize the level of sophistication in your organization when it comes to the
following tasks?
Public sector Private sector
PERCENT SELECTING HIGH SOPHISTICATION
Selecting AI technologies and technology suppliers
44%
44%
Identifying valuable applications for AI
30%
40%
Integrating AI technology into our existing IT environment
28%
40%
Integrating AI technology into business processes
24%
41%
Source: Deloitte analysis of the State of AI in the Enterprise Survey.
Deloitte Insights | deloitte.com/insights
8
Government executives on AI: Surveying how the public sector is approaching an AI-enabled future
Lack of investment and a skills operations, with the remaining amount going into
business enhancements (27 percent) and
gap could hinder adoption of
innovation and growth (26 percent).10
AI
Nevertheless, most public sector respondents plan
The survey shows that, among early adopters
to increase investments in the future. Some
across industries, the public sector has both the
40 percent of respondents said their organization
lowest level of AI investments and lower return on
plans to increase investment by more than
investments from their AI initiatives (figure 8).
10 percent, and only 4 percent said their
The low returns on AI investments could be due to
a focus on improving citizen services rather than
The one area in which
cost savings. The low investment itself could be
attributed to the high maintenance costs of
the public sector is on par
government legacy systems. In 2018, the US
government allocated 78.5 percent of its US$95.7 with other industries is
billion IT budget to operating and maintaining
selecting AI technologies
legacy systems.9 At the same time, CIOs of the most
innovative private sector organizations allocated a
and technology suppliers.
little less than half of their budgets (47 percent) to
FIGURE 8
Of all the sectors, the public sector invests the least in AI
Based on self-reported data
Source: Deloitte analysis of the State of AI in the Enterprise Survey.
Deloitte Insights | deloitte.com/insights
9
Government executives on AI: Surveying how the public sector is approaching an AI-enabled future
organization plans to reduce investment, implying Another factor seeming to hold back adoption in
an increased investment focus on AI technologies the public sector is the lack of needed skills. As
(figure 9). However, these numbers are still lower much as 71 percent of respondents cited the skills
than in the private sector, as 55 percent of private gap as a barrier—ranging from moderate to
sector respondents said their organization plans to extreme—to advancing AI projects in their
increase investment by more than 10 percent, and organization.
only 1 percent said their organization plans to
decrease investment.
FIGURE 9
Early adopters in the public sector have been increasing AI investments and plan
to continue doing so in the next fiscal year
How does your organization’s AI/cognitive investment in the current fiscal year compare with the
previous fiscal year’s investment?
Public sector Private sector
51%
44%
39%
36%
18%
8%
2% 2%
Decreased Stayed the +1–9% Increased by
investment same more than 10%
Thinking ahead, how do you expect your organization’s investment in AI/cognitive to change in
the next fiscal year?
55%
40% 40%
34%
16%
10%
4%
1%
Decreased Stayed the +1–9% Increased by
investment same more than 10%
Source: Deloitte analysis of the State of AI in the Enterprise Survey.
Deloitte Insights | deloitte.com/insights
10
Government executives on AI: Surveying how the public sector is approaching an AI-enabled future
Cybersecurity and ethical risks Early adopters are trying
are a major concern to find ways to balance AI
benefits and risks
About 46 percent of respondents from the public
sector said they have “major” or “extreme” While public sector early adopter respondents are
concerns about the potential risks associated with enthusiastic about the potential of AI technologies,
AI (figure 10). they are reportedly constrained by resource
shortages and are concerned about the risks
Of the different types of risk, ethical risk is the associated with these technologies.
second-highest concern cited by public sector early
adopters surveyed, but the lowest-ranked concern However, public sector organizations are making
for early adopters in other industries (figure 11). efforts to overcome these hurdles and move ahead
Further, recent media reports highlight how on their AI journey. Such efforts include:
government concerns around ethical risks of AI are
slowing and, in some cases, even halting the use of • Codevelopment with partners: Our survey
some AI technologies. San Francisco, for example, found public sector early adopters to be more
was the first major US city to block the use of facial inclined than the private sector to codevelop AI
recognition technology. The decision was rooted in solutions with partners (figure 12)—possibly
concerns about the invasion of citizen privacy as due to the advantages of tapping into market
well as potential racial bias.11 expertise and bringing in new capabilities to
mitigate technology risks. In fact, the
The concerns around cybersecurity vulnerabilities Interagency Select Committee on Artificial
are not surprising, considering the increasing Intelligence, which advises the White House on
number of cyberattacks on government systems. In AI research and development priorities,
2017 alone, federal civilian agencies reported more proposes this approach.13 Working with
than 35,000 security incidents.12 industry partners may require capabilities
within government for auditing algorithms.
FIGURE 10
Early adopters in the public sector are concerned about the potential risks
associated with AI
Overall, how concerned is your organization about the potential risks associated with your
AI/cognitive initiatives (e.g., cybersecurity, ethical, or legal risks)?
Public sector Private sector
MAJOR/EXTREME CONCERN ABOUT AI RISKS
46%
46%
Source: Deloitte analysis of the State of AI in the Enterprise Survey.
Deloitte Insights | deloitte.com/insights
11
Government executives on AI: Surveying how the public sector is approaching an AI-enabled future
FIGURE 11
Public sector early adopters are more concerned about ethical risks than any
other industry
Which of the following risks of AI/cognitive is your organization most concerned about?
Public sector Private sector
PERCENT RATING EACH CATEGORY A TOP-THREE AI RISK
Cybersecurity vulnerabilities of AI
47%
51%
Ethical risks of AI
44%
31%
Making the wrong decisions based on AI recommendations
43%
43%
Legal responsibility for decisions/actions made by AI systems
38%
39%
Failure of AI systems in a mission-critical for life-and-death context
35%
38%
Erosion of customer trust from AI failures
34%
33%
Regulatory noncompliance risk
31%
37%
Source: Deloitte analysis of the State of AI in the Enterprise Survey.
Deloitte Insights | deloitte.com/insights
• Testing new approaches: To foster sector early adopter respondents said that their
innovation, many early adopters surveyed in the organization had tested more than five
public sector are experimenting with new prototypes to date. In the US Department of
technologies and building prototypes. Defense’s AI strategy, released in February
Prototyping can help early adopters assess the 2019, prototyping is listed as one of the
vulnerabilities and test the impact of AI techniques to enhance the department’s
solutions in a controlled environment before AI capabilities.14
they are scaled. Around 41 percent of public
12
Government executives on AI: Surveying how the public sector is approaching an AI-enabled future
FIGURE 12
Early adopters in the public sector prefer codeveloping AI solutions with partners
Indicate whether your organization is already using or planning to use each of the following
ways of acquiring or developing AI/cognitive technologies
Public sector Private sector
Codevelopment with partners (e.g., IT and professional services firms)
62%
52%
Enterprise software with integrated AI
55%
60%
AI-as-a-service
55%
49%
Open source AI development tools
45%
49%
Data science modeling tools
39%
45%
Automated machine learning
35%
47%
Crowdsourced development communities
30%
40%
Source: Deloitte analysis of the State of AI in the Enterprise Survey.
Deloitte Insights | deloitte.com/insights
Learning from early adopters
should learn from the experiences of early adopters
within the government, as well as their private
The road to full-scale AI implementation may be a sector counterparts, to identify the use cases that
long one for many public sector agencies, but can be applied to their agencies and discover
pilots, experiments, and AI initiatives in different proven techniques to overcome challenges. With
pockets of government continue to grow. As more these learnings, the public sector can move up the
public sector agencies begin their AI journey, they AI adoption curve.
13
Government executives on AI: Surveying how the public sector is approaching an AI-enabled future
Endnotes
1. The Pulse, “Artificial intelligence’s impact on government contracting,” January 24, 2018.
2. James Vincent, “Trump signs executive order to spur US investment in artificial intelligence,” The Verge,
February 11, 2019.
3. The White House Office of Science and Technology Policy, “Summary of the 2018 White House Summit on
Artificial Intelligence for American Industry,” May 10, 2018.
4. Katya Schwenk, “New Jersey touts ‘first’ innovation training platform for state government,” Statescoop, June
18, 2019.
5. William D. Eggers, David Schatsky, and Peter Viechnicki, AI-augmented government: Using cognitive technologies to
redesign public sector work, Deloitte University Press, April 26, 2017; William D. Eggers, Neha Malik, Matt Gracie,
Using AI to unleash the power of unstructured government data, Deloitte Insights, January 16, 2019.
6. Deloitte Insights, Cognitive technologies: A technical primer, February 6, 2019.
7. Kevin C. Desouza and Rashmi Krishnamurthy, “Chatbots move public sector toward artificial intelligence,”
Brookings, June 2, 2017.
8. Jeff Loucks et al., Future in the balance? How countries are pursuing an AI advantage, Deloitte Insights, 2019.
9. David Wennergren et al., “Accelerating IT modernization in government,” Wall Street Journal, October 2, 2018.
10. Bill Briggs et al., Follow the money: 2018 global CIO survey, chapter 3, Deloitte Insights, August 8, 2018.
11. Kate Conger, Richard Fausset, and Serge F. Kovaleski, “San Francisco bans facial recognition technology,” New
York Times, May 14, 2019.
12. US Government Accountability Office, “Cybersecurity challenges facing the nation–high risk issue,” accessed
August 27, 2019.
13. Aaron Boyd, “Here’s what the White House’s AI Committee will focus on,” Nextgov, June 28, 2018.
14. Lauren C. Williams, “Pentagon outlines AI strategy,” GCN, February 13, 2019.
14
Government executives on AI: Surveying how the public sector is approaching an AI-enabled future
Acknowledgments
A special thanks to Susanne Hupfer from the Deloitte Center for Technology, Media &
Telecommunications for her support in analyzing the survey data and extracting insights for this study.
The authors would also like to thank Melissa Smith, David Noone, and Joe Mariani for reviews at
critical junctures and contributing their ideas and insights to this project.
About the authors
William Eggers | [email protected]
William Eggers is the executive director of Deloitte’s Center for Government Insights, where he is
responsible for the firm’s public sector thought leadership. His most recent book is Delivering on Digital:
The Innovators and Technologies that Are Transforming Government (Deloitte University Press, 2016). His
other books include The Solution Revolution, the Washington Post best-seller If We Can Put a Man on the
Moon, and Governing by Network. He coined the term Government 2.0 in a book by the same name. His
commentary has appeared in dozens of major media outlets including the New York Times, the Wall
Street Journal, and the Washington Post. He can be reached at [email protected] or on Twitter
@wdeggers. He is based in Rosslyn, Virginia.
Sushumna Agarwal | [email protected]
Sushumna Agarwal is a senior analyst with the Deloitte Center for Government Insights, Deloitte
Services LP. She researches workforce issues at the federal, state, and local government level. Her
primary focus is on applying quantitative techniques to enable data-driven research insights.
Mahesh Kelkar | [email protected]
Mahesh Kelkar of Deloitte Services LP is a research manager with the Deloitte Center for Government
Insights. He closely tracks the federal and state government sectors, and focuses on conducting
in-depth research on the intersection of technology with government operations, policy, and
decision-making.
15
Government executives on AI: Surveying how the public sector is approaching an AI-enabled future
Contact us
Our insights can help you take advantage of change. If you’re looking for fresh ideas to address your
challenges, we should talk.
Practice contact
Thomas Beyer
Principal | Deloitte Consulting LLP | + 1 619 237 6659 | [email protected]
Thomas Beyer, principal, leads Deloitte Consulting’s Government and Public Services (GPS) Analytics &
Cognitive offering.
The Deloitte Center for Government Insights
William Eggers
Executive director | The Deloitte Center for Government Insights | Deloitte Services LP
+ 1 202 246 9684 | [email protected]
William Eggers is the executive director of Deloitte’s Center for Government Insights, where he is
responsible for the firm’s public sector thought leadership.
16
About the Deloitte Center for Government Insights
The Deloitte Center for Government Insights shares inspiring stories of government innovation, looking
at what’s behind the adoption of new technologies and management practices. We produce cutting-
edge research that guides public officials without burying them in jargon and minutiae, crystalizing
essential insights in an easy-to-absorb format. Through research, forums, and immersive workshops,
our goal is to provide public officials, policy professionals, and members of the media with fresh
insights that advance an understanding of what is possible in government transformation.
Deloitte’s “Cognitive Advantage” is a set of offerings designed to help organizations transform
decision-making, business processes, and interactions through the use of insights, automation,
and engagement capabilities. Cognitive Advantage is tailored to the federal government and
powered by our cognitive platform. Cognitive Advantage encompasses technologies capable of
mimicking, augmenting, and in some cases exceeding human capabilities. With this capability,
government clients can improve operational efficiencies, enhance citizen and end-user
experience, and provide workers with tools to enhance judgment, accuracy, and speed.
Sign up for Deloitte Insights updates at www.deloitte.com/insights.
Follow @DeloitteInsight
Deloitte Insights contributors
Editorial: Aditi Rao, Blythe Hurley, Anya George Tharakan, and Rupesh Bhat
Creative: Sonya Vasilieff
Promotion: Alexandra Kawecki
Cover artwork: Traci Daberko
About Deloitte Insights
Deloitte Insights publishes original articles, reports and periodicals that provide insights for businesses, the public sector and
NGOs. Our goal is to draw upon research and experience from throughout our professional services organization, and that of
coauthors in academia and business, to advance the conversation on a broad spectrum of topics of interest to executives and
government leaders.
Deloitte Insights is an imprint of Deloitte Development LLC.
About this publication
This publication contains general information only, and none of Deloitte Touche Tohmatsu Limited, its member firms, or its
and their affiliates are, by means of this publication, rendering accounting, business, financial, investment, legal, tax, or other
professional advice or services. This publication is not a substitute for such professional advice or services, nor should it be
used as a basis for any decision or action that may affect your finances or your business. Before making any decision or taking
any action that may affect your finances or your business, you should consult a qualified professional adviser.
None of Deloitte Touche Tohmatsu Limited, its member firms, or its and their respective affiliates shall be responsible for any
loss whatsoever sustained by any person who relies on this publication.
About Deloitte
Deloitte refers to one or more of Deloitte Touche Tohmatsu Limited, a UK private company limited by guarantee (“DTTL”), its
network of member firms, and their related entities. DTTL and each of its member firms are legally separate and independent
entities. DTTL (also referred to as “Deloitte Global”) does not provide services to clients. In the United States, Deloitte refers to
one or more of the US member firms of DTTL, their related entities that operate using the “Deloitte” name in the United States
and their respective affiliates. Certain services may not be available to attest clients under the rules and regulations of public
accounting. Please see www.deloitte.com/about to learn more about our global network of member firms.
Copyright © 2019 Deloitte Development LLC. All rights reserved.
Member of Deloitte Touche Tohmatsu Limited |
180 | deloitte | gps-hc-genai-powered-hr-final.pdf | Power Human Resource
Service Delivery with AI
Public sector organizations and their HR functions are at the forefront
of creating opportunities to “mitigate the harms and maximize the
benefits of AI for workers,” as President Biden directed in his 2023
Executive Order on Safe, Secure, and Trustworthy AI. HR functions of
federal, state, and local government agencies have an opportunity to
leverage the power of AI to streamline their operations and enhance
the overall employee experience. By automating repetitive and time-
consuming tasks, HR professionals can free up their time to focus on
more strategic initiatives. Additionally, AI can enable data-driven
decision making, which can lead to better outcomes for the
organization and its employees.
Why now?
or 20,000 tasks performed by the federal of HR leaders believe they will be behind their peers
33% 76%
government can be completed by GenAI if they don’t implement AI solutions in 12-24 months
(HRD, 2023). (Gartner, 2023).
of organizations that have adopted AI for HR
43% 30% is saved by companies on their cost-per-hire through
are using AI for employee learning and
AI recruitment augmentation (Deloitte, 2023).
development (SHRM, 2024).
Integrating AI Across Key Human Resource Processes at your Organization
Leveraging AI to augment HR capabilities enables public sector organizations to address current workforce challenges and
make data-driven decisions that support their mission goals.
Where GenAI is Most Powerful in your HR Processes
Data-Driven Strategic Career &
Requisition Performance Learning &
Workforce Outreach & Succession
Creation Management Development
Planning Recruitment Planning
Infer skills of the existing Generate Position Identify new talent pools Detect and eliminate Provide AI-informed Analyze employee
workforce with future Descriptions, job (both active and passive unconscious bias from career recommendations performance data,
organizational needs for analyses, and job candidates) to optimize the performance including training path identify trends, and
targeted upskilling or announcements with the recruiting time and evaluation process to navigation for employees predict potential skill
recruiting efforts. use of an intelligent pinpoint prospects boost goal quality and and individual shortages using AI
assistant using existing inclined to accept an encourage achievement. development plan algorithms to cultivate
position management organization’s offer. suggestions. targeted training
data and light input from programs that address
hiring managers. specific skill gaps/org
priorities.
HR POLICY BOT
Provide case deflection, resolving customer questions at Tier 0 to reduce research and resolution times for HR.
HR DATA QUALITY & EMPLOYMENT MILESTONE PROCESSING
Provide insight into operational efficiencies through data to promote HR data quality while reducing burden on HR Quality Assurance teams; Automate HR tasks at
key milestones from offer letter generation to offboarding activities to improve the HR and employee experience.
Where Do I Start? The AI-Powered HR Maturity Curve
As organizations incorporate AI into their processes and applications, they progress along the maturity curve. Leaders
may want to consider the following HR initiatives, starting with HR automation and progressing to AI-driven strategies to
further their AI maturity goals.
Begin with establishing HR
data foundation, enabling AI-
driven data insights and HR
strategy
DEVELOP AN AI-
INFORMED HR STRATEGY
ESTABLISH in coordination with the IMPLEMENT AI-
organization’sn AI strategy
ESSENTIAL HR DATA EXPAND AI USE CASES INFORMED HR
FOUNDATION AUTOMATE HR TASKS across the talent lifecycle STRATEGY
for more accurate and identify AI use cases PILOT HR AI USE CASES informing the
outcomes across the talent lifecycle organization’s HR strategy
CREATE AI-INFORMED
HR AND DATA
GOVERNANCE
for strategy implementation
FOUNDATIONAL TRANSFORMATIVE INNOVATIVE EXPONENTIAL
How Deloitte Can Help Drive & Enable AI-powered HR Adoption:
Work with technical teams to create clean, structured, and consistent data that can be
1 Establish HR Data Foundation easily processed by AI systems while incorporating clear data governance and
management procedures to maintain data quality and integrity.
Collaborate with HR and organizational leaders to develop the roadmap to integrating AI
Develop AI-Powered HR
2 into your HR strategy and operations to prioritize AI use cases that can make immediate
Transformation Roadmap
impact and accelerate organizational success.
Implement Implement priority AI-use cases, start AI integration, and enhance your HR function with
3
feedback loops to integrate AI capabilities throughout talent lifecycle and HR processes.
AI-Use Cases
Ready your HR workforce for AI implementation by promoting AI literacy, upskilling and
4 Promote Workforce Readiness reskilling employees, helping employees adapt to new roles and skills, and promoting
continuous improvement.
BOTTOM LINE:
Leverage Gen AI to Streamline HR
Service Delivery and Enhance the
Employee Experience
AI provides the opportunity to think creatively about the tasks we ask
HR professionals to complete, versus the routine tasks AI can complete Kate Reilly Erin Schneider Lucy Melvin
Principal Managing Director Principal
to save HR (and their customers) time and energy. Build your strategy
Deloitte Consulting LLP Deloitte Consulting LLP Deloitte Consulting LLP
to enable AI-Powered HR today. [email protected] [email protected] [email protected]
About Deloitte
Deloitte refers to one or more of Deloitte Touche Tohmatsu Limited, a UK private company limited by guarantee (“DTTL”), its network of member firms, and their related entities. DTTL and each of its member firms are legally separate
and independent entities. DTTL (also referred to as “Deloitte Global”) does not provide services to clients. In the United States, Deloitte refers to one or more of the US member firms of DTTL, their related entities that operate using the
“Deloitte” name in the United States and their respective affiliates. Certain services may not be available to attest clients under the rules and regulations of public accounting. Please see www.deloitte.com/aboutto learn more about
our global network of member firms.
Copyright © 2024 Deloitte Development LLC. All rights reserved. |
181 | deloitte | ai-readiness-and-management-framework.pdf | AI Readiness & Management
Framework (aiRMF)
Navigating your AI journey
To implement artificial intelligence (AI) at scale, organizations need to build AI
maturity across the enterprise. Deloitte’s AI Readiness & Management
Framework integrates 10 capability areas to achieve enterprise AI readiness
and maturity. Deloitte partners with your organization to assess where you
are on your AI journey, define your target outcomes, and chart a path forward
to achieve your business and mission needs.
About Deloitte: As used in this document, a Deloitte means Deloitte Consulting LLP, a subsidiary of Deloitte LLP. Please see www.deloitte.com/us/about for a 1
detailed description of our legal structure. Certain services may not be available to attest clients under the rules and regulations of public accounting. Copyright © 2024 Deloitte Development LLC. All rights reserved.
Deloitte’s Integrated Approach to AI
Deloitte’s AI Readiness and Management Framework (aiRMF) is applied across three core functions: 1) Setting the Direction, 2) Building Core Capabilities to
Deliver AI Value, and 3) Managing AI Holistically.
Determine where and how AI can improve an organization’s operations and achieve mission/business needs
AI Exploration AI Strategy & Governance
Identify AI Opportunities & Use Cases Define Vision and Establish Governance
Develop foundational capabilities across data, technology, and people to enable AI solutions and deliver value
Data Technology People
Customer & User
Trustworthiness,
Experience* Data Readiness AI Apps and Solutions AI Enabled Workforce Security, & Risk*
Apply Customer-Centric Provide the Data Foundation Develop AI Solutions Prepare the Workforce Mitigate Risk and Instill
Design & Delivery
Confidence
AI Infrastructure & Platforms
Provide Technical Foundation
Continuously maintain, manage, and build upon AI capabilities
AI Delivery & Operations AI Sourcing Management
Scale, Maintain, and Operate AI Solutions Streamline Procurement
*Trustworthiness, Security, & Risk and Customer & User Experience are core to all AI capability areas and should be considered throughout the AI Journey
About Deloitte: As used in this document, a Deloitte means Deloitte Consulting LLP, a subsidiary of Deloitte LLP. Please see www.deloitte.com/us/about for a
detailed description of our legal structure. Certain services may not be available to attest clients under the rules and regulations of public accounting. Copyright © 2024 Deloitte Development LLC. All rights reserved.
aiRMF Capability Area Descriptions
AI Exploration AI Strategy & Governance
Defined mission and business needs, pin-pointed AI- Form a demand-driven strategy factoring the capabilities
enabled solutions, and discrete use cases outlining how AI necessary to implement AI technology responsibly,
could be applied throughout the organization to achieve securely, and consistently across the enterprise through
desired outcomes. plans, policies, procedures, and program alignment.
Trustworthiness, Security, & Risk Data Readiness
Mitigate risks and comply with AI regulations to Provide the foundation for accurate and impactful AI
create trust and confidence in the technology, solutions using high-quality, accessible, and labeled
while maintaining cybersecurity, the protection of data understood and trusted across the enterprise.
information, and the ethical use of data.
AI Delivery & Operations AI Infrastructure & Platforms
aiRMF
Scale and maintain AI solutions Implement a scalable architecture with the
and processes reliably and platform and tools needed to provide the speed,
CAPABILITY AREAS
efficiently in production. capacity, and processing power you need to
sustain AI-enabled solutions.
Customer & User Experience AI Apps & Solutions
Drive a human-centered AI experience and Implement AI software, models, and products across the
improve the adoption and value of AI organization to modernize, improve performance,
solutions with human-centered design and reduce total cost of ownership, and accelerate decision-
UI/UX techniques. making and workflows for mission-critical challenges.
AI Sourcing Management AI Enabled Workforce
Develop a sourcing strategy for effective procurement, Prepare your workforce to integrate AI into their
oversight, and management of vendor-provided AI solutions operational processes and determine the talent and skills
and services to meet your outcomes and advance your mission, they need to provide AI oversight and use it responsibly.
operations, and technology objectives.
About Deloitte: As used in this document, a Deloitte means Deloitte Consulting LLP, a subsidiary of Deloitte LLP. Please see www.deloitte.com/us/about for a
detailed description of our legal structure. Certain services may not be available to attest clients under the rules and regulations of public accounting. Copyright © 2024 Deloitte Development LLC. All rights reserved.
Contact the aiRMF Team for More Information
Aman Vij Leanna Pomponio Leigh Bechet Jordan Aulen
Principal Senior Manager Senior Consultant Consultant
[email protected] [email protected] [email protected] [email protected]
About Deloitte: As used in this document, a Deloitte means Deloitte Consulting LLP, a subsidiary of Deloitte LLP. Please see www.deloitte.com/us/about for a
detailed description of our legal structure. Certain services may not be available to attest clients under the rules and regulations of public accounting. Copyright © 2024 Deloitte Development LLC. All rights reserved. |
182 | deloitte | deloitte-cn-trustworthy-ai-report-en-250113.pdf | AI at a crossroads
Building trust as the path to scale
Deloitte Asia Pacific | AI Institute
AI at a crossroads | Building trust as the path to scale AI at a crossroads | Building trust as the path to scale
Contents
Report overview 4
01 Navigating the risks from rapid AI adoption 6
02 What does good AI governance look like? 8
03 AI Governance across Asia Pacific 12
04 The dividends from good AI governance 22
05 Building the foundations of trustworthy AI 25
Appendices 29
AI at a crossroads | Building trust as the path to scale AI at a crossroads | Building trust as the path to scale
AI at a crossroads:
Building trust as the path to scale
Report overview
As senior leaders move from experimenting to rolling Concerningly, more than half of technology workers
out AI solutions, a number of key risks – such as security do not believe their workplace can address AI related
vulnerabilities, privacy and legal risk – are experienced risks. To understand how effective AI governance can
by the organisation. While AI solutions offer powerful help to address these risks and unlock the potential
This report was co-developed by Deloitte Access Economics and productivity tools, they can lead to data breaches, of AI, Deloitte has surveyed nearly 900 senior leaders
loss of reputation and business and regulatory fines from 13 locations across the Asia Pacific region in one
the Deloitte AI Institute to provide insights to Asia Pacific C-suite
if the risks of these tools are not managed properly. of the most comprehensive stocktakes of AI governance
executives and tech leaders, on how they can improve their
maturity levels to date.
governance structures and organisation settings to develop
more trustworthy AI solutions.
There is a rising number of incidents from using AI across all industries
Over a quarter of organisations have experienced an increase of incidents related to AI in the past financial year.
Deloitte has created a Trustworthy AI Framework that
identifies seven dimensions necessary for organisations
Increase in incidents recorded in the past financial year, by industry
to have trust in their AI solutions – transparent and
explainable, fair and impartial, robust and reliable, 1
TRANSPARENT AND EXPLAINABLE
respectful of privacy, safe and secure, responsible
28% 31% 24% 42%
and accountable.
2
FAIR AND IMPARTIAL
But what needs to be in place for organisations
to achieve trustworthy AI? Good AI governance. Government and Life sciences and Technology sector Financial sector
public service health care
3
ROBUST AND RELIABLE
For C-suite executives and board members, activating
and supporting effective AI governance practices can be Good governance also leads to greater AI adoption and financial returns
challenging amidst competing priorities. To help address 4
RESPECTFUL OF PRIVACY
this ambiguity, we’ve developed an AI Governance
Maturity Index to identify what good AI governance looks
like in practice. This index contains a set of criteria to 5
SAFE AND SECURE
assess AI governance within an organisation and was 28% more staff 3x more likely 4.6 percentage points 45% of senior leaders
using AI solutions across to be using AI solutions higher in revenue believe good governance
applied to the responses of nearly 900 surveyed senior the business in areas such as R&D, growth from AI improves
leaders from Australia, China, India, Indonesia, Japan, 6 RESPONSIBLE operations and production, solutions reputation
Malaysia, New Zealand, Philippines, Singapore, South and customer service, among customers
marketing and sales
Korea, Taiwan (China), Thailand, and Vietnam.
A range of industries, organisation sizes and public 7 ACCOUNTABLE
sector organisations were included in the responses. Yet more than 90% of organisations can improve AI governance
The survey questions aimed to understand the maturity
level of AI governance across organisations, identify key Deloitte’s Governance Maturity Distribution of AI Trustworthy Index across Asia Pacific
Index uses 12 indicators to
enablers of effective AI governance and assess
the benefits to organisations from having these assess AI governance across 17% 74% 9%
organisations.
arrangements in place.
Basic In progress Ready
Actions to build Trustworthy AI
1 2 3 4
Prioritise AI governance to Understand and Build risk managers, Communicate and ensure AI
realise the returns from AI leverage the broader AI not risk avoiders transformation readiness
supply chain
4
AI at a crossroads | Building trust as the path to scale AI at a crossroads | Building trust as the path to scale
01 Figure 1
Top concerns about potential
Navigating the risks from rapid AI adoption
risks associated with using AI
The adoption of AI across the Asia Pacific region Security vulnerabilities can arise from AI solutions or
is transforming the business landscape. The rapid the vast amount of data used by the solutions, which Security vulnerabilities
86%
emergence of generative AI (GenAI) has only accelerated can become targets for theft or data breaches, and
this process, with investment in AI across the Asia can result in significant costs. The global average cost
Surveillance
83%
Pacific region expected to grow fivefold by the end of a data breach reached nearly $5 million USD in 2024,
of the decade, reaching $117 billion USD by 2030.1 a 10% increase from the previous year.4 Of course, for
Privacy
GenAI has quickly become the region’s fastest-growing large organisations, this cost can be significantly higher. 83%
enterprise technology.
There are also broader costs that are difficult to quantify,
Legal risk and copyright infringement
Behind the rapid pace of adoption are employees, such as damage to brand and loss of customers. The 80%
who often outpace their leaders. A previous Deloitte erosion of consumer confidence and the negative impact
study on Generation AI found that more than two in on brand reputation can have long-lasting effects, making Regulatory uncertainty
79%
five employees were already using generative AI at it crucial for businesses to manage AI and cybersecurity
work, with young employees leading the way.2 effectively. At the same time, there is a strong consumer
Reliability and errors
preference for businesses that use AI in a way that aligns
78%
This pace and scale of AI adoption means leaders with their ethical standards, such as transparency when
are encountering AI related risks in real time as AI is used. Research indicates that 62% of consumers Malicious content
78%
they experiment and roll out the technology. place higher trust in companies whose AI interactions
Our survey of nearly 900 senior leaders reveals that they perceive as ethical, and 53% are willing to pay
Regulatory burden
risks related to security vulnerability (86%), surveillance a premium for such products and services.5 76%
(83%) and privacy (83%) are the most common concerns
for senior leaders when using AI (Figure 1). These Organisations must also ensure that their use of AI Accountability
75%
risks have become even more pronounced since the is compliant with evolving legislative and regulatory
advent of GenAI, which has seen a step change in the requirements, which was a shared theme among
Transparency/explainability
capabilities of the technology alongside more user- the most common risks identified by senior leaders. 73%
friendly interfaces that have broadened the number While there has been a focus on developing and
of people who can use these powerful tools. enacting regulations and legislation across Asia Pacific Responsibility
73%
governments, these existing regulatory requirements
are usually a minimum standard for organisations to
Bias and discrimination
meet rather than comprehensive best practices. As a 71%
“Over half of technology workers
result, senior leaders must develop, adopt and enforce
believe their workplace does not have
organisational trustworthiness standards for AI solutions Job displacement
the appropriate settings to identify or
70%
and systems.6
address AI-related risks according
to a Deloitte study.”3 Addressing AI-related risks is essential: without
Source: Deloitte Trustworthy AI survey (2024)
proper management, these risks could lead to strained
customer relationships, regulatory penalties or public
backlash. Furthermore, fear of these risks can also deter
organisations from using AI. The State of AI Enterprise
survey found that three out of the four biggest challenges
to developing and using AI tools are risk, regulation and
governance issues.7 This highlights the importance of
effective AI governance for managing the ethical and
operational risks associated with AI and fully leveraging
this technology.
6 7
AI at a crossroads | Building trust as the path to scale AI at a crossroads | Building trust as the path to scale
02 Developing trustworthy AI solutions that meet AI governance can often feel elusive with constantly
these seven criteria does not happen automatically. shifting goalposts. To assist organisations to take
Organisations must have robust AI governance to practical steps to achieving trustworthy AI, we have
What does good AI governance look like?
provide the structure that ensures AI solutions align created an AI Governance Maturity Index.
with these principles.
This Index, based on 12 key indicators across five
At its core, good AI governance is required at all pillars (organisational structure, policy and principles,
Developing trustworthy AI solutions is essential for Deloitte has developed a Trustworthy AI Framework that
stages of the technology lifecycle and is embedded procedures and controls, people and skills and
senior leaders to successfully navigate the risks of outlines seven key dimensions that are necessary to build
across technology, processes, and employee monitoring, reporting and evaluation), assesses an
rapid AI adoption and fully embrace and integrate this trust in AI solutions – 1) transparent and explainable, 2) fair
training. Governance arrangements require tailoring organisation’s AI governance maturity (Table 1). Based
transformative technology. Trustworthy AI provides and impartial, 3) robust and reliable, 4) respectful of privacy,
to the sophistication of AI solutions used, location on these indicators, we categorise organisations as
a level of certainty that the technology is ethical, lawful 5) safe and secure, 6) responsible, and 7) accountable
and industry-specific regulations, and internal ‘Basic’, ‘In progress’ or ‘Ready’ in terms of their AI
and technically robust and provides confidence for senior (Figure 2). This framework and criteria should be applied to
organisational policy and standards. governance maturity. Further details about the Index
leaders to use AI solutions throughout their organisation. AI solutions from ideation through to design, development,
and the underlying questions are available in Appendix B.
procurement and deployment.
Figure 2: Deloitte Trustworthy AI framework
Table 1: Deloitte AI Governance Maturity Index
Reskilling and education
Pillars BASIC IN PROGRESS READY
S UA sF eE
r
frA
i
eN D U sS
e
rE pC
r
o
tU
e
ctR
i o
nE In v
u ln
e
r
a
b le
Autonomous Confidential DiscrP etiR onIV
al
nA seT nE
sual T
R A N
O str rg ua cn tuis ra etional L r foea rsc pk Ao Io n gf s or i vo b el ie l ris nty aa a nn s cd s
e
i .gned I a
i
fnd on de rd n
i Av
rt Iiei dfi gs
u
oe p ad vo
l
e
sns
r
o as nim
n
ab
d
nie l i
c
gtr ei ro e .ol se
u
fs po sr B d
r
t
o
goe
oreo
s
g
vfi ma
p
a
enr
o
a
n
rd e
nnn
i
d sa
aas
a,c
i
n
gb
twc
e
i
co
i
oi
l
emt iu
nt
.h in
ee
wrt nsoa
i
t
dalb e
t
esi osl si
A
t
i
sa gy Iun
n
pd
e
p
d
o rt
AND
RELIABLE
ConPr se isd ti ect na tble
n
dly
Co
Ju s Iti nfi tea rbl pe retable
S
P
A R E N T A
N
D
E
P po ril nic cy ip a ln esd N o Ar Io gp A orI i v np ec ro i npli ac le ny s ci n t eo .p gla uc ide e B w gua it is h dic eg o e Ar n I d e gr ora i vcf e t p r p r nio n al ncic i cpy el .ein s p tola ce R b t ua nyo i lb iw o qu r ues el elt d- cp d t oo e o nfil i o tc n ery e g x, d a tg
.
nr po ir su i ann tc id oipe nld e ’s si n
T Deloitte's X P
OBUS Accurate
Trustworthy AI
Auditable L
A
IN
G
o v
R Adaptable FrameworkTM Visible ELBA P anro dc ce od nu tr re os
l s
N coo
n
r ti rs ok
l
sp fr oo rc e dd eu vere los
p
o mr
ent,
R cois nk
t
rp or lo sc ue nd du ere
r
s
d
a en ved l/ oo pr
m ent
E px rois ct ein dg
u
s rey ss t ae nm
d
/o of
r
r isk
e
r deployment or use of AI for development, controls sufficient to guide
n a n c e
a n d
c o n tr ol s A C C O U N T A
BLA Enswe Rr ea sb olle
vabl Oe
wnership Humane Common/ Social goody tilib a n ia
ts
u
Sd e s u c o f g n id d a e u
la
V
U n bi a s e
dI ncl
usE iq
v
euA itc ac be less Fi Ab
I
ul
R
l
e
aA tN orD I
y
M cP oAR mTI plA iL ance and
policy
P aneo dp sl ke i l ls
s
N s
rety
o a
ss
ff
pt re
e o
tm
s noo s
s
su
i.
bur lpc ye p .s o o rtr utr sa ei n oi fn Ag I for
d o
R d
tofe
ee
p A
vs uo
el sIo
lu
es oy y
r
Apm cs
e
Iet e e
rds
en m
c f
st
ou
pso rr. or
r e
neu
mn
ss ite
p bl y l l o y b .ye ei en sg
d
R g t t
ro
or
eure
e a s
iv
s
eu
id
pnoe ms
e
oil ue no
l npir
gnp o
sc
lo,em f
ie b a
ysA
s
le
er ,
yI
f
en
e i o
.s
n
st ary c,
v
ts uld
ou
at se e
id e
ulp m
ai
snl
a
bo es
g n l
.y
e
Adm
I
ent
R
eg Monitoring, No mechanism for Mechanism and tools for Existing mechanism and
RESPONSIBLE reporting and monitoring or reporting on monitoring or reporting tools for monitoring or
evaluation AI systems in operation. on AI systems in operation reporting on AI systems
under development. in operation.
Source: Deloitte (2024)
Source: Deloitte (2024)
8 9
AI at a crossroads | Building trust as the path to scale
CASE STUDY
The figure below depicts how each of the pillars in the Empowering the future:
Deloitte AI Governance Maturity Index is a foundational
element that can enable an organisation to achieve
Energy Queensland’s commitment to
trustworthy AI. Furthermore, the Index identifies the
practical arrangements and activities that an organisation
responsible AI and sustainable innovation
should undertake to achieve the seven dimensions
highlighted in the Trustworthy AI Framework.
Energy Queensland is Australia’s largest, wholly government-owned electricity
Figure 3: the Deloitte AU Governance Maturity Index company, servicing over 2.3 million customers and employing more than 9,300
people across its distribution, retail, and integrated energy solutions businesses.
Sharyn Scriven, CIO Energy Queensland expresses that incrementally’, according to Josh. This has involved
“AI is a game changer and as it matures will help aid trialling enterprise tools and building AI platform
Achieving
our business and people to achieve our vision and services to initially support corporate users with
Trustworthy AI 2032 Corporate Strategy.” heavy documentation, meetings and emails.
requires each organisation Josh Gow, General Manager of Customer and Emerging Effective and responsible use of AI requires team
Platforms, recognises that integrating AI is an important members with the right capabilities alongside
to develop:
focus area for Energy Queensland to drive operational powerful AI solutions. For this reason, ‘control group
excellence and enhance customer experience, releases’ are being conducted and reviewed, where
supporting the organisation’s ambitious strategy. While employees in different roles participate in a controlled
Energy Queensland has been using AI for several years, release, education and training program before further
there has been a shift from niche specialised use cases deployment.
The five pillars of the AI Governance Maturity Index
to broader use case evaluation and deployment.
“Ensuring we capture the value, opportunity and
Drafting an AI policy has been essential for Energy continue to manage the risk that AI will bring with
Queensland to ensure the right policies and settings further adoption is critical. It’s a matter of when,
are in place before introducing new AI solutions. This not if, AI will be in broader use across many more
has involved developing an AI Policy and a roadmap technologies. Not everyone will get the same AI and
for use case rollout across the organisation, along it may also be ‘under the hood’. We need to tailor
with necessary actions to establish appropriate how AI will aid our company to ensure it is effective,
Organisational Policy and Procedures People Monitoring,
guardrails. To ensure the AI policy adhered to industry responsible, and valuable.”
structure principles and controls and skills reporting and
best practices and was implemented correctly, Energy
evaluation
Queensland had the AI policy independently reviewed
Source: Deloitte (2024) by an external organisation, as well as internally.
Josh explains:
Key features to ensure trustworthy AI
“Our AI policy is under continued review, as a living,
There is no one-size-fits-all approach to AI governance. It should also be noted that higher levels of AI Governance
breathing document, given the rapidly changing
The specific governance structures will vary depending Maturity do not automatically lead to trustworthy AI
environment of AI and maturing industry standards
AI policy
on the industry, regulatory environment, AI ambition outcomes. If governance procedures are in place but are
and guidelines. Our monthly AI steering committee
and type of AI solutions being adopted. For instance, not effectively implemented, understood by staff or well-
includes senior executives who regularly discuss
AI steering committee
an AI-powered chatbot providing employees with tailored to the business context and strategy, trustworthy
the progress, risks and opportunities of AI.”
information about HR policies will require different AI outcomes may not be achieved. Effective AI governance
control processes compared to a bank’s AI-driven is different for every organisation. For this reason, it is Testing and piloting AI use cases before full Piloting and trialling
credit application solution that interfaces directly important for organisations to continuously evaluate and implementation is an important feature of Energy AI programs internally
with customers. Comparing common features of refine their AI governance framework to ensure that it is Queensland’s approach to AI. Trialling AI through
AI governance can help organisations identify areas right-sized to their unique needs and evolving regulatory internal use cases has been a strategic choice to Training programs
for improvement in their governance standards. requirements. create an environment where it has been ‘test and
learn focused to further evaluate risk and opportunity
10 1111
AI at a crossroads | Building trust as the path to scale AI at a crossroads | Building trust as the path to scale
03
PILLAR 1
AI governance across Asia Pacific Organisational structure
Having clearly identified roles within an organisation How organisations structure the teams responsible
Fewer than one in ten organisations across the Chart 1: Distribution of AI Trustworthy Index that are accountable for managing AI standards helps for ethical, legal and regulatory compliance related to
Asia Pacific have the governance structures necessary across Asia Pacific
to ensure any emerging AI-related issues are addressed AI may vary. Just over a quarter (28%) of organisations
to achieve trustworthy AI. Using our AI Governance
appropriately. For most organisations surveyed, this have a centralised ethics and risk team to monitor trends
Maturity Index, we classify 91% of organisations as 17% 74% 9%
responsibility lies with senior leadership, with 91% and detect risks related to AI use, while the majority (61%)
having ‘Basic’ or ‘In progress’ AI Governance structures
of organisations having a board member or C-suite of organisations have dedicated professionals working in
Basic In progress Ready
in place, highlighting substantial room for improvement
executive explicitly responsible. A further 7% nominated all or some departments or teams (Chart 3). The remaining
in AI governance (Chart 1). Source: Deloitte Trustworthy AI survey (2024)
a non-executive AI lead as responsible for managing risks organisations have either some teams with dedicated
and standards, while less than 2% of respondents were professionals or no dedicated roles for AI use.
Examining the five pillars of the AI Governance
not able to identify anyone primarily responsible in their
Maturity Index, organisations across Asia Pacific have Chart 2: Distribution of Trustworthy AI Index across pillars
organisation. More important than the structure of the team is having
the greatest opportunity for improvement in policies
clear responsibility and accountability for AI standards,
Organisational structure
and principles as well as procedures and controls.
yet this is less common in smaller organisations. For
Currently, 31% and 23% of organisations, respectively, 9% 73% 18%
organisations with more than 1,000 employees, only 3%
are categorised at ‘Basic’ levels in these two pillars.
have no dedicated AI risk roles, compared to 23% of those
In contrast, organisations performed better in the Policy and principles with fewer than 100 employees.
organisational structure and monitoring and evaluation
31% 56% 13%
pillars, with more than 90% achieving at least
‘In Progress’ status.
Procedures and controls
Achieving a ‘Ready’ status for the AI Governance Chart 3: Structure of team responsible for ethical, legal and regulatory compliance related to AI
23% 66% 10%
Maturity Index overall requires high performance
across all five pillars. While nearly one in five
organisations achieved a ‘Ready’ status in one of the People and skills
pillars, only half that shared achieved ‘Ready’ for their 28% 31% 29% 11%
22% 64% 14%
AI governance overall. This highlights the need to
consider AI governance in a holistic sense to develop A centralised team Every team / department Some teams / No dedicated roles
Monitoring, reporting and evaluation working across the has dedicated departments have for AI use
the conditions required for trustworthy AI.
organisation professionals dedicated
6% 77% 18% professionals
Basic In progress Ready
Source: Deloitte Trustworthy AI survey (2024)
Source: Deloitte Trustworthy AI survey (2024)
Addressing the overconfidence bias
Leaders may overestimate the maturity of AI Governance. Deloitte’s State of Generative AI in the Enterprise survey
found that 23% of organisational leaders rated their risk management procedures and governance as highly
prepared. However, this more detailed study, exploring the underlying structure of AI governance revealed only 9%
had actually achieved a ‘Ready’ level of governance.8 While the specific questions and sample differ, the extent of the
variation in these studies suggests that senior leaders need to have a detailed understanding of their AI governance
maturity. This is pertinent as overconfidence can represent a barrier to improving AI governance; if leaders believe
they have sufficient settings in place to manage AI risks, they are less likely to explore how they can improve.
12 13
AI at a crossroads | Building trust as the path to scale AI at a crossroads | Building trust as the path to scale
PILLAR 2 PILLAR 4
Policy and principles People and skills
Clear, broadly understood policies and principles are Chart 4: Implementation of trustworthy AI policies Employees play a crucial role in ensuring trustworthy Chart 6: Resources available to employees
a fundamental prerequisite for effective AI governance. AI. Yet, this remains a challenge for many organisations, to support them using AI
This AI policy differs from an AI strategy, with the latter Incident response and remediation plans where only 56% of employees, on average, have the skills 58%
including broader elements such as ambitions related and capabilities to use AI responsibly. 55% 55%
15% 38% 48% 52%
to AI and key metrics to measure progress. While most 49%
organisations across Asia Pacific have an AI strategy Training can be a powerful tool to bridge this gap.
Ethical guidelines and principles
in place, many are missing key elements of good Organisations that provide AI training see a 27% higher
governance in their AI policy. More than half of AI policies 11% 34% 55% share of employees equipped to use AI safely compared 36%
34%
lack timelines for implementing AI governance goals or to those that don’t – though just 52% of organisations 31% 32% 32%
contain ethical guidelines and principles related to AI. surveyed currently offer such programs. That said, 72%
AI policy for safe and responsible use of AI in the organisation
of organisations that currently don’t offer training are
Including these governance features in an AI policy is 8% 25% 67% actively developing programs for their teams.
key for employees to see the value. Among organisations
with an organisation-wide AI strategy, 30% report that The majority of organisations do offer guidelines
No plans/ Future plans Currently
not all employees see the strategy’s value. Where the Unsure implemented on responsible AI use, and 55% encourage on-the-
AI policy includes monitoring or auditing, i.e. having job learning and experimentation and slightly fewer
a defined risk appetite, response and remediation plan Source: Deloitte Trustworthy AI survey (2024) organisations have an advisory service or body for Guidelines on Advisory Training on AI security Encouraged
integrated with broader organisation policies, employees employees (49%). Private sector organisations lead in appropriate service for appropriate and privacy on-the-job
AI use employees AI use measures learning
are more likely to see the value in the strategy. offering AI use guidelines and training, whereas public
sector organisations are more likely to focus on security In development Currently available
measures and encourage on-the-job learning.
Source: Deloitte Trustworthy AI survey (2024)
PILLAR 3 PILLAR 5
Procedures and controls Monitoring, reporting and evaluation
The third pillar explores day-to-day practices for A key element of effective AI governance is a system for Having AI governance systems that are responsive
managing AI-related risks and standards in an employees to report queries or incidents related to AI use to changing requirements and emerging issues is Chart 7: Frequency of evaluating AI systems against
internal organisation standards
organisation. This includes an assessment procedure to in the workplace. Yet, two in five organisations lack such a critical to ensuring organisations can respond to risks
identify and manage AI-related risks, a comprehensive reporting mechanism. Organisations with formal reporting and incidents as they emerge. Overall, organisations
Less than
Unsure/Never
inventory of AI solutions used, and control frameworks systems see five times more queries and twice as many performed relatively well in this pillar, with the equal once a year
2%
that mitigate risks associated with the use of an AI reported incidents – indicating that those without these highest share (18%) achieving ‘Ready’ status. The majority 3%
At least yearly
solution. With the fewest organisations categorised as systems may be blind to emerging risks associated with (85%) of organisations evaluated their AI governance Ongoing or
‘Ready’ for this pillar, progress in this area will be key for AI. This issue is only growing more urgent, especially in against internal standards at least every six months real-time
10%
improving trustworthy AI performance across the region. Asia Pacific, where the number of queries and incidents (i.e. those evaluating at least every six months, three 47%
continues to rise (see Chart 5). months or in real-time). Monitoring and evaluating
whether AI governance is complying with any changes in
regulatory requirements is another element of this pillar. 27%
Chart 5: Change in the number of incidents related to AI in FY24 compared to FY23
Nearly three-quarters of organisations review legal and
regulatory requirements at least every six months.
30%
At least every
27% 35% 32%
six months
At least every
Increased Remained about Decreased
the same few months
Source: Deloitte Trustworthy AI survey (2024)
Source: Deloitte Trustworthy AI survey (2024) Note: excludes ‘unsure’ answers (6)
14 15
AI at a crossroads | Building trust as the path to scale AI at a crossroads | Building trust as the path to scale
CASE STUDY
Navigating innovation and governance:
How does trustworthy AI
Dai-Ichi Life Holdings’ approach
compare across industries?
to responsible AI
Dai-Ichi Life Holdings, Inc. is a leading insurance group, with 122 years
The results for the AI Governance Maturity Index and individual
of experience in providing life insurance and investment products across pillars vary by industry. We find that organisations within
the Asia Pacific region and to the global market. technology, financial services and professional services more
generally have the highest share of organisations that are
‘Ready’ for trustworthy AI. Meanwhile, public sector and life
Figen Ulgen is the Chief Data and AI Officer for Dai-Ichi Dai-Ichi Life Holdings views effective AI governance science and healthcare organisations have a lower share.
Life Holdings and oversees the organisation’s strategy as a collective responsibility. Notably, at the Dai-Ichi A high-level summary of four key industries is over the
for Artificial Intelligence and Data. Delivering high Global Data and AI Synergy Leadership forum, where following pages. A similar summary for key geographies
quality customer service and building a strong, leaders in the organisation and member companies across Asia Pacific is available in Appendix D.
trusting relationship with clientele are key values meet, AI governance was the voted to be the topic to work
for Dai-Ichi Life Holdings. Implementing AI solutions, on together. Furthermore, responsible AI is regarded as
in a responsible and ethical way, is key to achieving empowering for all stakeholders involved. The business
these goals. Dr. Ulgen explains: can comfortably roll out new AI solutions and internal
users can safely explore AI knowing that guards are in
“Our AI solutions need to be designed and place and flags will be raised if necessary. Importantly,
implemented in a way that reinforce our company’s Dr. Ulgen emphasised that processes should include
value around serving our customer. This requires how incidents are handled, if, and when they do occur.
time and patience to make sure our systems are
acting in the ways we expect. We know this is going “If we have the right framework and processes
to be a marathon, not a sprint.” in place, our staff don’t have to carry the burden
or feeling that they are taking a risk. They feel
Dai-Ichi Life Holdings is currently exploring the empowered to use the solutions we have designed
possibilities for innovation using generative AI, with confidence and knowing when to query a result.”
through digital agents, which are digital avatars
including chatbot capability. Accompanying Dai-Ichi Dr. Ulgen also highlighted organisational culture, and
Life Insurance agents to customer meetings, these specifically empathy, as foundational for delivering
digital agents help with note taking, extracting high quality customer service, which extends to the
relevant documents to the customer’s questions implementation of ethical AI. In the case of life insurance,
and later summarising the conversations. Dai-Ichi this looks like understanding that there is a “high level
Life Holdings is undertaking ongoing and long- of ethical responsibility toward the customer”.
term testing to ensure that the digital agents are
implemented in a responsible way. In fact, the digital
agents have been tested in multiple sales offices
for almost a year, with hundreds of sales agents.
Key features to ensure trustworthy AI
Additionally, critical to ensuring accuracy of answers,
for Dai-Ichi, is maintaining a ‘human touch’, with
every piece of information created by a digital agent
Long term approach
checked by employees.
Human touch
A collaborative effort
1166 17
AI at a crossroads | Building trust as the path to scale
Spotlight on Spotlight on
Financial services industry Technology industry
Being a knowledge and data intensive industry, Our AI Governance Maturity Index shows the financial The technology industsry is at the forefront of AI Based on the Deloitte Generation AI report, technology
financial services have been leading adopters services industry has higher levels compared with other disruption and a key enabler of developing AI solutions employees lead in adoption of GenAI into their workflow,
of digital innovation. The relatively higher levels industries. Demand for financial services is growing, for other industries. As long-time users of |
183 | deloitte | gen-ai-multi-agents-pov-2.pdf | The cognitive leap
How to reimagine work
with AI agents
December 2024
The cognitive leap | How to reimagine work with AI agents
Content
Key takeaways
• Multiagent AI systems can help transform traditional, rules-based
business and IT processes into adaptive, cognitive processes.
• Organizations should leverage key principles of AI agent and multiagent AI
system design and management, which borrow from tenets of composable design,
microservices architecture, and human resources deployment and teaming.
• The ability to scale AI agents and multiagent frameworks across a range of use
cases depends on developing a comprehensive reference architecture populated
with reusable core components.
• A systematic approach can make the difference between incremental,
isolated improvements and exponential enterprise transformation.
Vaulting ahead on the path to GenAI value 3
How agents deliver a cognitive advantage: 5
Principles of AI agent design and management
Adaptive processes for innovative outcomes: 8
Principles of multiagent AI system design and management
Expanding and scaling multiagent AI systems: 10
A reference architecture for agent-powered transformation
Multiagent AI systems in action: 12
An example use case for transforming traditional IT support processes
From generating to innovating: 14
Key considerations on the path to AI agent-enabled transformation
Making the cognitive leap 16
Get in touch & Endnotes 17
2
The cognitive leap | How to reimagine work with AI agents
Vaulting ahead on the
path to GenAI value
Everyone remembers that pivotal moment when we first saw
what large language models (LLMs) and Generative AI (GenAI)
Business executives say
could accomplish. Suddenly, the long-discussed theory of
conversational, intuitive, creative AI became a reality, right there deeply embedding GenAI
at our fingertips. Adoption of GenAI surged across industries:
into business functions and
By the end of 2023 most companies had embraced GenAI
solutions.2 By midyear 2024, 67% of companies using GenAI processes is the No. 1 way to
said they were increasing investments after seeing strong
drive value from the technology.1
results from the technology.3
But as companies dove into testing GenAI’s potential, many
came to recognize the limitations of standalone GenAI models.
Context and reasoning limitations of typical LLMs can make it
difficult to apply GenAI to complex, multistep workflows. As with
traditional AI, hallucination and bias can create significant barriers
to trust. And the creative outputs for which GenAI is celebrated
require continuous human monitoring for quality and accuracy.
For these and other reasons, early GenAI use cases were
mostly limited to isolated or narrowly defined tasks within larger
workflows. For example, a wealth management adviser may
quickly produce a meeting recap using a standalone, LLM-based
solution. But extracting rich post-meeting analytics based on
different information categories discussed in the meeting (e.g.,
client profile, client goals, retirement information, etc.) remained
too complex to achieve with a standalone GenAI solution.
AI agents and multiagent AI systems are helping
organizations hurdle these limitations and make
the cognitive leap into a new paradigm of business
process transformation and innovation.
AI agents enable organizations to tackle significantly more
complex tasks with GenAI across an expanded range of
processes and use cases. When AI agents work together in a
system, they can help collaboratively reason, plan, design and
execute novel workflows that amplify speed, differentiation
and efficiency across the enterprise.
In this paper we outline key design principles and a reference
architecture for scaling AI agent use cases that can help your
business seize the potential of AI agents now.
3
The cognitive leap | How to reimagine work with AI agents
“Each mind is made of many smaller
processes. These we’ll call agents.
Each mental agent by itself can only
do some simple thing that needs
no mind or thought at all. Yet when
we join these agents in societies—in
certain very special ways—this leads
to true intelligence.”
—Marvin Minsky, The Society of Mind4
4
The cognitive leap | How to reimagine work with AI agents
How agents deliver a
cognitive advantage
Determining the most appropriate roles and uses for AI agents Language, planning, reasoning, reflection, and the ability to use
begins with adopting a shared, enterprisewide understanding of tools, data and memory: These attributes are central to how AI
what they are and how they can fit into your organization. agents work and demonstrate cognitive abilities as well.
AI agents are reasoning engines that can understand context, In the realm of business, AI agents and human workers have other
plan workflows, connect to external tools and data, and execute broad similarities. Both must be carefully selected, well trained and
actions to achieve a defined goal. They do so by echoing some of well equipped to perform their jobs. And both should be smartly
the key qualities and advantages that have helped humans survive deployed and consistently managed in ways that help ensure
and flourish. efficient, value-adding performance.
As people, we can understand language and creatively articulate Not surprisingly then, our recommended principles of AI
responses. By employing specialized tools, we can amplify our agent design and management echo familiar themes from
physical and mental capabilities. By learning and remembering organizational design and human resource management.
information, we avoid mistakes and improve on what we’ve (Please see next page.)
already accomplished.
5
The cognitive leap | How to reimagine work with AI agents
Principles of AI agent design and management
• Domain-driven approach: Every area of expertise and function of your business
utilizes different processes, data and tools. While some AI agents may be able to serve
multiple domains and processes, most should be sourced and/or designed based on
specific domain requirements. To achieve this, each domain of your business should be
analyzed, subdomains and processes identified, and agents assigned based on specific
roles within the domain.
• Role-based design: Agents should be designed to perform roles rather than specific
tasks, grouping similar activities to avoid confusion and ensure efficient operation.
This approach—which aligns with the “single responsibility principle”5—can help your
organization reduce AI agent overlap and unnecessary technology complexity. It also
can help enable reusability of agents across systems and domains.
• Right balance: Related to the principle of role-based design, it is important to find
the proper balance between the number and the scope of responsibilities of individual
AI agents. Too many agents with too few responsibilities can result in unnecessary costs
as well as challenges related to consistent governance, maintenance, monitoring and
upgrades. Too few agents with too many responsibilities can result in bottlenecks and
poor performance.
• Controlled access to data, skills and tools: You wouldn’t give every employee in
your enterprise access to every application or data resource in your business. Similarly,
the tools, data and skills made available to a given AI agent should be limited to those
that are essential to its role. These constraints help reduce risk and improve outputs
from the agent. If an agent’s role requires more than five tools, consider how you might
separate its responsibilities across two or more agents.
• Reflective cycle: Agents—like people—get better and better when given an
opportunity to reflect on their own performance or receive constructive criticism.
That’s why it’s important to design a self-reflective pattern in which agents critically
evaluate their own output by referring to past examples or testing the results of its
output. Agents also receive feedback from other agents and humans. This combination
of self-assessment and external input creates a continuous loop of learning and
improvement that helps ensure compliance with quality, brand and risk standards.
6
The cognitive leap | How to reimagine work with AI agents
“Synergy (is) the bonus that
is achieved when things work
together harmoniously.”
—Mark Twain
7
The cognitive leap | How to reimagine work with AI agents
Adaptive processes for
innovative outcomes
The achievements of remarkable individuals—from Aristotle to Multiagent AI systems have the potential to impact every
Simone Biles—are often treated as proof of our boundless human layer of enterprise architecture—not just automating
potential. But as any leader today knows, individual strengths are existing processes and tasks, but also reinventing them.
no match for team synergy. Organized and managed well, teamwork By engaging with users and within workflows semantically rather
leverages and amplifies the strengths of each individual—making it than syntactically, AI agents can comprehend emerging needs
possible to achieve goals that no person could do alone. and address them in novel ways that obviate traditional, rules-
based processes. By continuously self-monitoring, multiagent AI
As with people, so too with AI agents. Research has shown that systems can improve their outputs in near real time. Meantime,
AI agents working together are more effective than individual the shared persistent state of AI agents in a system enables them
agents.6,7 By leveraging an “agency” of role-specific AI agents, to collaborate and coordinate activities in ways that continuously
multiagent AI systems can understand requests, plan workflows, streamline efficiency.
delegate and coordinate agent responsibilities, streamline actions,
collaborate with humans, and ultimately validate and improve The principles of agent design discussed in the previous section
outputs. Processes that were considered too complex for typical become especially important in this context. For example, dynamic
language models can be automated at scale—securely and workflow planning and task decomposition in a multiagent AI
efficiently. Projects that once took weeks can be completed in a system are critical to effectively automating and reinventing
small fraction of that time. Human workers who previously spent end-to-end processes—and are dependent on the right balance
precious hours performing routine, repetitive tasks can instead of domain-specific, role-based agents to perform each task.
focus on higher-level, higher-value activities. By providing each agent with controlled access to data, skills and
tools—and by providing checks and balances throughout the whole
So, while standalone AI agents can help accelerate the completion system—redundancies can be avoided and quality improved.
of individual tasks, multiagent AI systems can open new realms of
business process automation, speed and reliability. Agents within a When designing multiagent AI systems, we recommend a set of
system can interact and collaborate in various deployment patterns, principles to help ensure that these systems are robust, reliable
depending on the specific needs and complexity of the process. and trustworthy. (Please see next page.)
8
The cognitive leap | How to reimagine work with AI agents
Principles of multiagent AI system design and management
• Understandable and explainable systems: Good business leaders explain and justify their
decisions, and AI systems should do the same. The actions of your multiagent AI systems need to be
explainable, particularly in tasks related to perception and classification. Systems should be designed
to document each agent’s chain of thought,8 and not just the final output. (Think of it as “showing your
work” in math class.) Clarity and interpretability will help minimize biases originating from their design
or datasets.
• Composable design: Multiagent solutions should be designed with composability in mind.
A composable design can allow organizations to bring best-of-breed components together in a
microservices architecture to develop optimized and efficient multiagent systems. By orchestrating
custom and third-party agents that include different programming languages and agent frameworks,
your organization can design more complex agentic patterns that integrate with multiple internal
and external systems.
• Human in the loop: AI agents shouldn’t be solely responsible for critiquing their own or other
agents’ outputs. Knowledgeable humans must be essential parts of AI systems as a safeguard
against potential errors or biases. This isn’t just common sense; it’s a regulatory mandate in some
industries and/or US states. California, for example, recently required that AI-generated health
care-related decisions must be reviewed by a human before being shared with consumers.9
• Dynamic data patterns: In designing multiagent AI systems, data should be able to flow in
two distinct patterns: data to the agent and agent to the data. In the data-to-the-agent pattern,
unstructured data is typically captured into a vector or graph database. It’s important to include
not only the data itself but its hierarchy relevant to the specific use case. This enables agents to
apply the data appropriately within various contexts. In the agent-to-the-data pattern, the agent
uses suitable tools built into the model (such as search tools or API specifications) to determine
how to retrieve relevant structured data for the task at hand.
• Ecosystem integration: A multiagent AI system often needs to integrate with various existing
applications or processes to achieve its intended goals. Therefore, the design of these systems
should consider integration patterns with ecosystem processes and applications. Some integrations
may be achieved via application programming interfaces (APIs), while others may be event-driven.
For example, a multiagent system for post-meeting analytics may need to integrate with a CRM
platform through an API to upload client profiles or other information discussed during the meeting.
• Continuous improvement and adaptation: Performance improvement must be built into
the “DNA” of multiagent AI systems. Systems should be designed to learn from prior interactions
and evolve in response to new data and changing conditions. This capability can be implemented
through agent and workflow memory, which stores past interactions and workflow executions.
The stored information can later be leveraged to enhance future executions.
• Ethical considerations: The same ethical principles you apply to human capital decisions,
such as impact, justice and autonomy, should guide the design and deployment of multiagent AI
systems. In addition to prioritizing explainability, your organization should regularly assess AI system
outputs to ensure they contribute positively to society and avoid causing harm.
9
The cognitive leap | How to reimagine work with AI agents
Expanding and scaling
multiagent AI systems
Imagine you’re the chief transformation officer at a global
financial services company. You understand the principles of AI
agent and multiagent AI system design. You see the potential in
this next evolution of Generative AI technology everywhere in
your organization.
But where to apply it?
A multiagent AI system could help your HR team identify, recruit
and onboard talent by analyzing mountains of resumes against
job requirements, intelligently assessing candidates based on
skills and experience, even conducting initial screening interviews.
The benefits seem obvious: greater scalability and efficiency,
improved candidate matching, less bias …
Then again, AI agents could transform efficiency in your call
center by enabling plain-language conversations between clients
and chatbots. This could help digital self-service feel more like
old-fashioned client service—while your human support reps are
freed to focus on more sensitive, higher-value interactions.
Or maybe the place to focus is in improving personalization in
financial advisory services? Or in automating financial reports?
The list goes on—across every domain of the enterprise.
Thanks to the innate flexibility and scalability of multiagent AI
systems, your organization doesn’t have to limit its focus. While
it is true that no organization possesses the financial, talent or
technological resources to design and deploy bespoke multiagent
AI systems for every possible domain or use case—no longer are
these resources requisite to success.
The key is to treat a multiagent AI system as an
ecosystem of capabilities instead of solutions and to
develop a reference architecture that can support both
business and technical delivery processes. This approach
can allow your organization to more rapidly scale, expand and
reuse AI agents and multiagent frameworks across a range of use
cases—while also streamlining governance, monitoring, operation
and improvement of agentic outputs.
The essential layers of a reference architecture are shown in the
illustration on the next page. Each layer within the architecture
is loosely coupled with—but independent of—other layers.
Similarly, each component within a given layer can be leveraged
independently. This makes it possible to adapt, connect and
apply best-fit solutions for any use case that arises.
10
The cognitive leap | How to reimagine work with AI agents
A reference architecture for agent-powered transformation
Interaction layer
Purpose: Allow users, processes and Example elements for a financial services company:
existing applications to collaborate with
multiagent AI systems.
Actions for success: Develop defensive user
Mobile banking CRM Conversational IT support
interfaces that can anticipate and mitigate
app system IVR system portal
potential user errors or misuse, while guiding the
multiagent system(s) to respond contextually.
Workflow layer
Purpose: Ensure controlled flow engineering Example elements for a financial services company:
to help agents interact with each other
efficiently and in a more deterministic manner.
Actions for success: Implement value-stream
Know Risk Financial Software
analysis to monitor efficiency and effectiveness your control planning incident
of workflows. Identify governance guardrails customer testing workflow support
and touch points for human monitoring workflow workflow workflow
(“human in the loop”) to help reduce risks.
Human in Human in Human in Human in
Infuse long-term memory into workflows.
the loop the loop the loop the loop
Agents layer
Purpose: Create, manage, deploy Example elements for a financial services company:
and optimize role-specific AI agents.
Actions for success: Focus on MODEL GARDEN AGENT FACTORY TOOLS
industrializing the creation of role-
Multimodal Data retrieval
specific agents to accelerate speed Search engines
commercial LLM agent
to value.
Multimodal Recommendation Financial analysis
Each agent should be equipped with: open-source LLM agent tool
• A fit-for-use language model Fine-tuned Incident classification Code interpreter
model agent
• Tools that augment language model
capabilities with skills to perform Domain-skilled Incident analysis
specific tasks/roles SLMs agent DATA SOURCES
• Approved sources of authoritative data Incident resolution
Customer 360 record
agent
• Memory of past tasks to help improve
PROMPT REGISTRY
performance of new tasks Quality assurance Financial markets
Prompt agent data
• Access to effective prompts for engaging templates
with other agents and/or humans in a Agents from Incident history
given workflow Prompt third-party vendors
versioning
MEMORY
Prompt
testing Short-term
(current session)
Prompt
access management Long-term
(past sessions)
Agent operations layers
Purpose: Monitor outputs and metrics to help Example elements for a financial services company:
ensure agents are functioning as expected.
Actions for success: Implement
instrumentation and telemetry, along
Operational Qualitative Thought
with logs, traces and metrics, to gather data
metrics metrics metrics
about system activities. Activate alerts and
dashboards to simplify performance
monitoring against service-level objectives.
11
The cognitive leap | How to reimagine work with AI agents
Multiagent AI systems
in action
Continuing our exploration of the reference architecture layers and elements
that contribute to effective, efficient and scalable multiagent AI systems, let’s look
more specifically at an IT operations process—specifically, a support scenario for
a business software application.
Traditionally, this process involves multiple support team interventions and touch
points for the business user. The diagram below illustrates this resource-intensive,
inefficient and often time-consuming workflow.
Service desk rep (L1) gathers details Support analyst (L2) is assigned
from the business user and attempts to and then reaches out to the business
find a solution by searching knowledge user to collect details, analyze the issue
resources. If an existing solution is not and try to fix it. If the issue remains
available, the issue is escalated to the unresolved or may affect other users,
“I’m having a appropriate support specialist. it is assigned to L3.
problem with
a software app
that’s important 1 2
to my work.” Business user has to take time Business user often has to repeat the
to engage in a dialogue with L1. same information already provided to L1.
1 1 2
Business 3 Service desk rep Support analyst
user (L1) (L2)
2
Support technician (L3)
conducts a root cause
analysis to identify a
permanent fix.
3
3
Business user may be
engaged again to provide
Release Release manager is engaged to Support technician more information or test
manager plan deployment of the application (L3) potential solutions.
change to production so the issue
does not recur.
12
The cognitive leap | How to reimagine work with AI agents
Traditional L1 and L2 IT support workflows are primed for transformation through
multiagent AI system solutions. By leveraging an AI agent-enabled process, the user is
continuously updated—but can be much less actively engaged. Support personnel are engaged only
to monitor, review and approve rather than find and implement most solutions. This frees the human
support personnel to focus on the most complex and business-critical resolution of select issues.
And it frees business users to get back to the important work of generating enterprise value.
Here’s how it can work.
(This example shows one variation of IT support for illustrative purposes.
The most appropriate solution for your business may differ.)
REFERENCE ARCHITECTURE
“I’m having a problem with a software “Tell me more about LAYERS
app that’s important to my work.” your problem.”
Interaction
Workflow
Agents
Business A business user files a support ticket IT support
user through the enterprise IT support portal. portal Agent operations
The software incident support workflow
is triggered to resolve the incident.
Software incident Human
support workflow in the loop
The solution identified by the agents
is handed over to a “human in the
The workflow orchestrates the agents loop” to verify and execute the
for the resolution of the incident. resolution. This helps ensure that
human knowledge and judgment
remains part of the solution.
Incident An incident classification agent Software A specialized software incident analysis agent Software The solution
classification identifies the type of issue and incident reviews the ticket against existing data resources incident is implemented
agent engages the appropriate software analysis (knowledge base articles, SOPs, etc.). If a potential resolution for the business
agent agent
incident analysis agent. solution has already been developed the ticket passes user—who
to a software incident resolution agent, which has been
The incident classification agent’s either validates the solution or sends it back to the updated on
role fulfills typical L1 support. analysis agent for more information or other solutions. progress/status
throughout
The agents in this workflow fulfill typical L2 support. the process.
If no existing solution is found, the incident is elevated
to L3 (human) support.
As the workflow is executing, the traces and spans from the
agent interactions are continuously logged, processed and
aggregated through telemetry. This provides key operational and
response metrics for appraising performance of the workflow
“Back in
and each individual agent in the workflow. business!”
Operational Qualitative Thought
metrics metrics metrics Business
user
13
The cognitive leap | How to reimagine work with AI agents
From generating to innovating:
Key considerations on the path to
AI agent-enabled transformation
Every promising technology innovation comes with its own set of challenges. Multiagent AI systems are no
exception. Strategically, organizations need to identify priority areas and use cases where AI agents can have
the most rapid and valuable impact. Implications around change management also come into play, from
training employees in new skills to modifying existing processes. At Deloitte we’ve gleaned valuable lessons
that can help you realize the full value potential of this technology innovation.
As you explore the potential for multiagent AI systems for your organization, these considerations can help
provide a valuable head start.
1 4
Starting smartly Evaluating technologies
With so many potential use cases for multiagent AI There are numerous technology choices related to
systems, it’s important to be strategic about where each layer of the agentic architecture. To simplify the
to begin and how to move forward. Executive sponsorship process of selecting the right technology stack and agent
and appetite, rigorous cost/benefit analysis, and a development tool kit(s), consider leveraging an evaluation
clear understanding of the state of your underlying data framework that helps to objectively score the choices at
fabric form the foundation of use case prioritization and each layer to baseline the right-fit technology stack of
planning. To accelerate return on investment, proactive the agentic architecture.
and thorough change management should be a part of any
agent-powered transformation initiative, with an emphasis
on building trust across your organization and among your
stakeholders as new solutions are rolled out.
2
Pinpointing the right data, in the right context
Data forms the backbone of any agentic architecture.
For every use case, it’s essential to not only identify the
authoritative source of data that the agents will use but
also ensure that agents can evaluate the appropriate
context for that data. This is where knowledge engineering
comes into play: By organizing data (i.e., knowledge sources)
into a classification system or taxonomy, you make it easier
for agents to navigate and retrieve the right data.
3
Tapping talent
Your system’s design and development will require data
engineering, business process engineering, machine
learning and application architecture knowledge—in other
words, some of the most high-demand skills in today’s talent
market. Accessing the necessary human expertise typically
involves a combination of workforce upskilling and hiring,
combined with strategic outsourcing to fill the roles that
will be needed to support agentic AI transformation.
14
The cognitive leap | How to reimagine work with AI agents
5
Decomposing processes
Reimagining an existing process or developing new agent-
based workflows means breaking the overall process into
smaller, more manageable subprocesses. By decomposing
the process based on roles, each agent can specialize in a
clear set of tasks, ensuring there are no overlapping
responsibilities. To achieve this, consider using domain-driven
design principles in which the boundaries for each subprocess
are defined by and align with the organization’s domain and
team structure. This approach not only defines clear task
boundaries but helps pinpoint the right number of agents to
accomplish the overall process.
6
Scaling multiagent AI system impact with
sound reference architecture
A thoughtfully designed reference architecture allows your
organization to scale multiagent systems across a wide
range of use cases in trustworthy and transparent ways.
By embedding best practices and reusable components, this
approach establishes a standard and repeatable process for
design, deployment and continuous improvement. This not
only ensures interoperability and reduces redundancy but
also enables rapid adaptation and integration of best-fit
agents for any emerging use case. It also provides a solid
and ethical foundation for governance and optimization,
ensuring that the multiagent AI systems remain aligned
with enterprise goals and can evolve in response to
changing needs and technological advancements.
To design a reference architecture appropriate for
your whole organization, we recommend taking into
account industry best practices, market and customer
expectations, and the technology, process and data
realities of your own enterprise.
7
Embedding sound governance
It is very important to ensure that multiagent AI
systems, once deployed in production, consistently
generate quality outputs that do not introduce enterprise
risk. Continuous monitoring and analysis of system outputs
is critical to enabling timely identification of any potential
anomalies or inaccuracies. It’s important therefore to
ensure that every multiagent AI system be smartly
developed in ways that ensure multiple “checkpoints”
within the workflow—and that checks and balances are
engineered into each individual agent.
15
The cognitive leap | How to reimagine work with AI agents
Making the
cognitive leap
The rapid evolution of multiagent AI systems is transforming how
organizations address challenges and streamline processes.
This space is rapidly evolving as commercially available language
models, frameworks and agents continue to improve. Organizations
that adopt a systematic approach to multiagent AI system design
and management will be well positioned to scale these systems
effectively. Rather than limiting AI agent deployment to isolated
business processes, a comprehensive approach allows for the
expansion of AI capabilities across various use cases and domains.
By anchoring in the foundational principles we have outlined—
and by leveraging a robust reference architecture that enables
reuse and rapid adaptation of core components—organizations
can maximize the potential usage and scale of multiagent AI
systems. This approach helps empower organizations to derive
more value from their AI investments, putting them not just at
the forefront of technological advancement but giving them a
competitive advantage.
16
The cognitive leap | How to reimagine work with AI agents
Get in touch Endnotes
Prakul Sharma 1. Deborshi Dutt, Beena Ammanath, Costi Perricos and Brenna
Principal, Sniderman, Now decides next: Moving from potential to
AI & Data performance, Deloitte, August 2024, p. 10, https://www2.
Deloitte Consulting LLP deloitte.com/content/dam/Deloitte/us/Documents/consulting/
[email protected]
us-state-of-gen-ai-q3.pdf, accessed December 3, 2024.
2. Benjamin Finzi, Brett Weinberg and Elizabeth Molacek, Winter
Sanghamitra Pati
2024, Fortune/Deloitte CEO Survey, Deloitte, 2024, p. 11,
Managing Director,
https://www2.deloitte.com/content/dam/Deloitte/us/
US India AI Leader
Documents/us-winter-2024-fortune-deloitte-ceo-survey.pdf,
Deloitte Consulting LLP
[email protected] accessed December 3, 2024.
3. Dutt et al, Now decides next: Moving from potential to
performance, p. 8.
Abdi Goodarzi
Principal, 4. Marvin Minsky, The Society of Mind, New York: Simon & Schuster,
GenAI Innovation Leader March 15, 1988, ISBN 0-671-60740-5.
Deloitte Consulting LLP
5. Robert C. Martin, Agile Software Development: Principles,
[email protected]
Patterns, and Practices, Prentice Hall, 2003, p. 95. ISBN 978-
0135974445.
Vivek Kulkarni
6. KaShun Shum, Shizhe Diao and Tong Zhang, Automatic Prompt
Managing Director,
Augmentation and Selection with Chain-of-Thought from Labeled
AI Transformation
Data, Cornell University, February 27, 2024, https://arxiv.org/
Deloitte LLP
abs/2302.12822, accessed September 16, 2024.
[email protected]
7. Boshi Wang, Sewon Min, Xiang Deng, Jiaming Shen, You Wu,
Luke Zettlemoyer and Huan Sun, Towards Understanding
Ed Van Buren
Chain-of-Thought Prompting: An Empirical Study of What
Principal,
Matters, Cornell University, June 1, 2023, https://arxiv.org/
GPS Applied AI Leader
pdf/2212.10001, accessed September 16, 2024.
Deloitte Consulting LLP
[email protected]
8. Wang et al, Towards Understanding Chain-of-Thought Prompting.
9. California Legislative Informati |
184 | deloitte | the-state-generative-ai-enterprise.pdf | Now decides next:
Insights from the leading
edge of generative AI adoption
Deloitte’s State of Generative AI in the Enterprise
Quarter one report
January 2024
Table of contents
Foreword
Introduction
Now: Key findings
1 Excitement about generative AI remains 4 Current generative AI efforts remain more
high, and transformative impacts are focused on efficiency, productivity and cost
expected in the next three years. reduction than on innovation and growth.
2 M any leaders are confident about their 5 Most organizations are still primarily relying
organization’s generative AI expertise. on off-the-shelf generative AI solutions.
3 Organizations that report very high 6 Talent, governance and risk are critical areas
expertise in generative AI tend to feel more where generative AI preparedness is lacking.
positive about it—but also more pressured
7 Leaders see significant societal impacts on
and threatened.
the horizon.
8 Leaders are looking for more regulation and
collaboration globally.
Next: Looking ahead
Authorship & Acknowledgments
About the Deloitte AI Institute
About the Deloitte Center for Integrated Research
About the Deloitte Center for Technology, Media & Telecommunications
Methodology
22
Foreword
Now decides next
The arrival of generative AI heralds disruption and From these wave one insights, we can gain a clearer
opportunity across industries. Organizations are picture of how leaders are using generative AI,
exploring how generative AI can be used to unlock challenges, and lessons learned thus far. This helps
business value, supercharge efficiency and productivity, reveal some of the essential questions leaders should
and open the door to entirely new products, services be asking now and actions they should be taking to
and business models. As business leaders contend prepare their enterprise for what comes next.
with this new technology and make decisions about the
There is still much to discover with generative AI.
future of the enterprise with generative AI, it is helpful
As it matures and is deployed at scale for a litany of
to keep one’s finger on the pulse of adoption.
applications, new questions and challenges will become
To that end, The State of Generative AI in the Enterprise: clearer. Our quarterly reports will be available to help
Now decides next, captures the sentiments of 2,835 you make sense of this fast-moving space, consider
business and technology leaders involved in piloting or practical guidance based on what we have learned,
implementing generative AI in their organizations. In this and take a forward-looking view in your business
inaugural release of the quarterly report series, leaders future with generative AI.
indicated persistent excitement for using generative
Learn more about the series and sign up for updates at
AI and many expect substantial transformative
deloitte.com/us/state-of-generative-ai.
impacts in the short term. Yet, they also acknowledged
uncertainty about generative AI’s potential implications
Deborshi Dutt, Beena Ammanath, Costi Perricos and
on workforces and society as the technology is
Brenna Sniderman
widely scaled, calling for greater investment in talent,
governance and global collaboration.
3
Introduction
Now decides next: Insights from the leading edge of generative AI adoption
Will generative AI (gen AI) be the greatest, most impactful technology innovation in Generative AI seems to be following the same pattern, only much, much faster. ChatGPT
history? Will it completely transform how humans live and work? Or will it turn out to was publicly released on November 30, 2022, largely as a technology demonstration.
be just another technology du jour that promised revolutionary change but ultimately Two months later, it had already attracted an estimated 100 million active users—
delivered only incremental improvement? Right now, we can’t be certain. making it the fastest-growing consumer application in history.1
What we do know is that many breakthrough technologies of the past have followed Since then, generative AI has continued to advance by leaps and bounds and many new
a common adoption pattern: initial awareness; excitement that led to hype; mild tools and use cases have emerged—providing a powerful glimpse at the technology’s
disappointment as hype met reality; and then explosive growth once the technology vast potential to transform how people live and work.
reached critical mass and proved its worth.
4
Introduction
Insights from the leading edge (cont.)
About The State of
Generative AI in
During this frenzied period of generative AI advancement To help make smart decisions, leaders need objective, timely
and adoption, leaders in business, technology and information about current generative AI developments— the Enterprise
the public sector are under tremendous pressure to and where things are headed. Which is why Deloitte is
To help leaders in business, technology and the
understand generative AI—and to figure out how to harness conducting this ongoing quarterly survey. Our goal is to take public sector track the rapid pace of generative
AI change and adoption, Deloitte is conducting a
its capabilities most effectively (or at least avoid being the pulse of generative AI adoption, offer a view of what’s
series of quarterly surveys. The series is based
disrupted). They also sense that now decides next; that their happening, track evolving attitudes and activities, and deliver on Deloitte’s State of AI in the Enterprise reports,
which have been released annually five years
decisions and actions today will significantly affect how practical, actionable insights that can help leaders like you
running. The wave one survey was fielded to more
generative AI unfolds in the future, for better or worse. make informed and confident decisions about AI, strategy,
than 2,800 director- to C-suite-level respondents
across six industries and 16 countries between
investment and deployment.
It’s been said that people tend to overestimate the effect of October and December 2023. Industries included:
Consumer; Energy, Resources & Industrials;
a technology in the short run and underestimate its effect in In this report, we examine our first quarterly survey findings
Financial Services; Life Sciences & Health Care;
the long run. This phenomenon has occurred many times in in detail, supported by insights from Deloitte’s AI-related Technology, Media & Telecom; and Government
& Public Services. Learn more at deloitte.com/us/
the past and could very well happen again with generative AI. work with organizations across every major industry and
state-of-generative-ai.
Note here that given generative AI’s dizzying pace of change, many geographic regions. We also offer a forward-looking
the gap between the short run and long run might be view to help you decide what generative AI actions may make
measured in days, weeks or months—not years or decades. sense for your own organization and situation.
All statistics noted in this report and its graphics are derived from Deloitte’s
first quarterly survey, conducted October – December 2023; The State of
Generative AI in the Enterprise: Now decides next, a report series. N (Total
leader survey responses) = 2,835
Generative AI is an area of artificial intelligence and refers to AI that in
response to a query can create text, images, video and other assets.
Generative AI systems can interact with humans and are often built using
large language models (LLMs). Also referred to as “gen AI.”
5
Now: Key findings
This first pulse of our generative AI quarterly surveys was
completed in December 2023, and included more than
2,800 AI-savvy business and technology leaders directly
involved in piloting or implementing gen AI at major
organizations around the world. Here’s what they had to
say about sentiment, use cases, challenges and more.
66
Generative AI elicits a range
Now: Key findings
of strong emotions
1 Excitement about generative AI remains high, and
transformative impacts are expected in the next three years.
62% Excitement
Nearly two-thirds (62%) of the business and technology leaders surveyed reported excitement as a top
Fascination 46%
sentiment with regard to generative AI; however, that excitement was tinged with uncertainty (30%) (figure 1).
The vast majority of respondents (79%) said they expect generative AI to drive substantial transformation
30% Uncertaintly
within their organization and industry over the next three years—with nearly a third expecting substantial
transformation to occur now (14%) or in less than one year (17%) (figure 2). Trust 17%
The survey results suggest that many AI-fueled organizations are on the verge of scaling up their efforts
16% Surprise
and embracing generative AI in a more substantial way. This aligns with what we’re seeing in the marketplace,
where organizations around the world are racing to move from experimentation and proofs-of-concept
Anxiety 10%
to larger-scale deployments across a variety of use cases and data types—pursuing both speed and
value capture while managing potential downside risks and societal impacts. 8% Confusion
In future surveys, we will be closely monitoring progress in this area—particularly with regard to
Fear 6%
organizations’ expertise, capabilities, tangible outcomes, and responses to rapidly emerging advances in
generative AI technology. 4% Exhaustion
Anger
1%
31%
of the leaders we surveyed expect substantial transformation
Figure 1
in less than one year; 48% expect it in one to three years.
Q: Thinking about generative AI, what emotions do you feel most about the technology?
(Oct./Dec. 2023) N (Total) = 2,835
77
Now: Key findings
When is generative AI likely
to transform your organization?
1%
14%
Never
Now
20%
17%
Beyond
three years
Less than
one year
48%
In one to
three years
Figure 2
Q: When is generative AI likely to substantially transform your organization and your industry, if at all?
(Oct./Dec. 2023) N (Total) = 2,835
8
44% rate their organization’s generative AI expertise as
Now: Key findings high or very high, but is such expertise even possible
given the pace of the technology’s advancement?
2 Many leaders are confident about their organization’s
generative AI expertise.
Self-assessed expertise with
A large percentage of our survey respondents (44%) said they believe their organizations currently have generative AI runs high
high (35%) or very high (9%) levels of expertise with generative AI. This result is somewhat surprising given
how rapidly generative AI is evolving (figure 3).
1%
But within the specific context of our survey, high levels of confidence seem entirely reasonable since
No expertise
9%
we deliberately chose experienced leaders with direct involvement in AI initiatives at large organizations 10%
Very high
already piloting or implementing generative AI solutions. However, given how rapidly the field is unfolding, it Little
expertise
expertise
may be worth questioning the extent to which any leader should feel highly confident in their organization’s
expertise and preparedness. In fact, even today’s foremost AI experts who are personally developing
generative AI technologies at times seem genuinely surprised by their own creations’ capabilities.2
35%
High
Do some leaders consider their organizations to have high expertise based largely on the knowledge
expertise
and experience gained from small-scale pilots with a small number of generative AI tools? If so, leaders
45%
and organizations might actually become less confident over time as they gain experience with the larger
Some
challenges of deploying generative AI at scale. In other words, the more they know, the more they might
expertise
realize how much they don’t know. This is a trend we’ve seen time and again with other technological
advancements, and one we’ll be watching closely in our future surveys.
Figure 3
Q: How would you assess your organization’s current level of overall expertise
regarding generative AI?
(Oct./Dec. 2023) N (Total) = 2,835
9
Expertise with generative AI drives
attitudes toward adoption
Now: Key findings
3 Organizations that report very high expertise in generative
Very high Some
expertise expertise
AI tend to feel more positive about it—but also more
Trust prevails Rank trust
39% 9%
among top
over uncertainty
pressured and threatened.
emotions felt
11% 38% Rank uncertainty
among top
Relative to other respondents, leaders who rated their organization’s overall generative AI expertise as “very
emotions felt
high” tended to feel much more positive about the technology; however, they also feel more pressure to
adopt it—and see it as more of a threat to their business and operating models (figure 4).
Analysis showed this group using more modalities, deploying generative AI across more enterprise functions, Broad interest 78% 38% Say employees
show high interest
sparks
and pursuing more use cases. As you can see in the figure 4, leaders who reported very high levels of
in gen AI
transformation
expertise were also more likely to report higher levels of trust and lower levels of uncertainty. They also
31% 9% Say gen AI
tended to show broader interest in generative AI and expected faster transformation for their organizations.
is already
transformative
At the same time, these respondents’ greater understanding of generative AI appears to be shaping their
perspective on potential impacts—positive and negative. Many reported they viewed widespread adoption of
the technology as a threat to how their organizations operate and conduct business, amplifying the pressure
Widespread 33% 16% Feel widespread
and urgency they felt to adopt generative AI and scale it.
adoption is a
adoption
threat to business
generates
pressure
44% 25% Feel greater
Leaders of organizations with very high expertise are more likely to pressure to
adopt gen AI
view generative AI as a threat to their business and operating models.
Figure 4
(Oct./Dec. 2023) N (Total) = 2,835,
N (Very high) = 267; N (Some) = 1,273
10
Key benefits organizations hope
to achieve with generative AI
Now: Key findings
4 Current generative AI efforts remain more focused
Improve 56%
on efficiency, productivity and cost reduction than on
efficiency and
productivity
innovation and growth. 35% Reduce costs
Improve existing 29%
The majority of organizations surveyed are currently targeting tactical benefits such as improving
products and
services
efficiency / productivity (56%) and/or reducing costs (35%). Also, 91% said they expect generative AI to
29% Encourage
innovation
improve their organization’s productivity, and 27% expect productivity to increase significantly. A smaller
and growth
percentage of organizations reported targeting strategic benefits such as innovation and growth (29%)
Shift workers from 26%
(figure 5). lower to higher
value tasks
26% Increase speed
This is consistent with past technology adoption patterns. Initially, most organizations logically focus on
and/or ease of
developing new
incrementally improving their existing processes and capabilities—capturing value from low-hanging fruit
systems / software
Increase 25%
while building knowledge, experience and confidence with the new technology. Later, they expand or shift
revenue
their focus to improvements that are more innovative, strategic and transformational—using the new
technology to drive growth and competitive differentiation and advantage through capabilities that simply
23% Enhance
relationships
weren’t possible before.
with clients /
customers
Surveyed leaders that cited higher levels of AI expertise show earlier signs of moving up this curve. They Uncover new 19%
ideas and
are more focused on uncovering new ideas and insights (23% vs. 19% for the overall respondent pool),
insights
with less emphasis on efficiency and productivity (44% vs. 61% for the overall respondent pool) and cost
18% Detect fraud and
manage risk
reduction (26% vs. 38% for the overall respondent pool)—although those tactical benefits continue to be Figure 5
Q: What are the key benefits you hope to achieve through your generative AI efforts?
(Oct./Dec. 2023) N (Total) = 2,835
11
Now: Key findings
their bigger focus. In addition, nearly three-quarters of organizations that cited very high generative
AI expertise had already begun integrating the technology into their product development and R&D
activities, which are key drivers of innovation and growth.
As more organizations gain expertise and experience with generative AI, will they reinvest their dividends
from improving efficiency and productivity toward pursuing more strategic benefits such as innovation
and growth? Or will they use those dividends in other ways? This is another area we’ll be monitoring
closely in future pulse surveys.
Certainly, productivity and efficiency can be transformational, especially given the massive scale
generative AI has the potential to enable. However, the greatest value and strategic differentiation will
likely come from using the technology to innovate. First, by helping to generate new products, services
and capabilities that wouldn’t be possible otherwise. And, second, by enabling new business models and
ways of working across an enterprise.
In addition, organizations that cited very high generative AI expertise were already taking a much more
comprehensive approach than average, with significantly higher adoption levels across a broad range
of functional areas. In specific areas such as HR, and legal, risk and compliance, those organizations’
generative AI adoption rates were nearly three times higher than for the total respondent pool (figure 6).
91%
of all organizations expect their productivity to increase due
to generative AI.
12
Now: Key findings % of those who are using generative AI
Total Little expertise Some expertise High expertise Very high expertise
in a limited or at-scale implementation
Level of generative AI
adoption
IT / cybersecurity 22% 38% 57% 71%
46%
Marketing, sales and customer service 41% 16% 34% 50% 73%
57%
Product development / R&D 41% 14% 28% 73%
Strategy and operations 35% 10% 26% 47% 62%
37%
Supply chain / manufacturing 29% 9% 21% 61%
Finance 37% 63%
25% 5% 14%
Figure 6
Human resources 23% 6% 13% 29% 64%
Q: What is your organization’s
current adoption level of generative AI across
the following functions?
28%
(Oct./Dec. 2023) N (Total) = 2,835; Legal, risk and compliance 21% 7% 10% 60%
N (Very high) = 267; N (High) = 1,003;
N (Some) = 1,273; N (Little) = 274
1133
Generative AI: Have we seen this movie before?
The term “unprecedented” is often thrown around Generative AI’s speed factor may give organizations less help the workforce get accustomed to using generative
when talking about business and technology, to the time to ruminate or dabble with small-scale pilots— AI, and will show people how it can help make their
point of being cliché. However, in describing the pace of while reducing the margin for error—and increasing the jobs easier. Also, early wins will likely help produce cost
generative AI’s emergence and advancement—and its consequences of inaction. It also creates opportunities savings and momentum that then can be channeled into
massive potential impact on business (and humanity as a to generate extraordinary business value very quickly. higher value opportunities that are more strategic and
whole)—unprecedented could be an understatement. differentiated in nature, such as enabling new products,
Despite generative AI ’s greatly accelerated pace,
services, business models and ways of working that
Generative AI is already widely available to the public understanding typical adoption patterns based on
simply weren’t possible before generative AI.
and has a running start toward critical mass. Also, similar previous breakthrough technologies can provide
to smartphones, it’s easy for an average person to use valuable lessons that leaders can use to help them
without much training—and can help with activities they understand and fully capitalize on the technology’s rapid
already engage in every day—so the barriers to adoption advancement.
are low. What’s more, generative AI has the strong
As in the past, organizations’ initial efforts will likely
potential to assist with its own future development,
center around efficiency, productivity, cost savings and
which could trigger a cycle of exponential improvement
other incremental improvements. This is expected to
at exponential speed.
14
Now: Key findings
5 Most organizations are primarily relying on off-the-shelf
generative AI solutions.
Where off-the-shelf generative AI
In line with their current emphasis on tactical benefits from generative AI, the vast majority of respondents is used most
were currently relying on off-the-shelf solutions. These included productivity applications with integrated
generative AI (71%); enterprise platforms with integrated generative AI (61%); standard generative AI
71%
applications (68%); and publicly available large language models (LLMs) (56%), such as ChatGPT.
Productivity applications
Relatively few reported using more narrowly focused and differentiated generative AI solutions, such as
industry-specific software applications (23%), private LLMs (32%), and/or open-source LLMs (customized to
68%
Standard applications
their business) (25%).
Reliance on standard, off-the-shelf solutions is consistent with the current early phase of generative AI
61%
adoption, which is primarily focused on improving the efficiency and productivity of existing activities.
Enterprise platforms
However, as use cases for generative AI become more specialized, differentiated and strategic, the
associated development approaches and technology infrastructure will likely follow suit.
56%
Public LLMs
When will we see complex, high-value use cases that are truly differentiated and tailored to the specialized
needs of specific companies, functions and industries? How will organizations combine internal and
external resources to create customized generative AI tools that enable such strategic differentiation? In
particular, will we see off-the-shelf technology offerings be supplemented by private or hybrid public/private
development approaches and technology infrastructures capable of delivering and supporting those
differentiated solutions?
15
Now: Key findings
6 Talent, governance and risk are critical areas where
generative AI preparedness is lacking.
In this initial quarterly survey, 41% of leaders reported their organizations were only slightly or not at
all prepared to address talent concerns related to generative AI adoption, while 22% considered their
organizations highly or very highly prepared. Similarly, 41% of leaders reported their organizations were only
slightly or not at all prepared to address governance and risk concerns related to generative AI adoption,
while 25% considered their organizations highly or very highly prepared (figure 7).
Larger percentages of leaders reported high to very high levels of preparedness in technology infrastructure
(40%) and strategy (34%); however, the survey results show there is still significant room for improvement.
Generative AI barriers related to risk and governance
When it comes to risk and governance, generative AI is definitely not “just another technology.” The
fundamental challenge is how to capitalize on artificial intelligence’s power without losing control of it. After
all, the capability people seem to find most enthralling about generative AI is its ability to so convincingly
simulate human thinking and behavior. Of course, human thinking and behavior aren’t always perfect,
predictable or socially acceptable—and the same is true for the technology, itself.
16
Now: Key findings
Respondents claimed the highest levels of preparation in technology
Preparedness for generative AI
and strategy, while feeling far less prepared in risk and talent.
Technology infrastructure 4% 17% 38% 30% 10%
Strategy 5% 20% 41% 26% 8%
Not prepared
Slightly prepared
Risk & governance 13% 28% 34% 18% 7% Moderately prepared
Highly prepared
Talent 13% 28% 37% 17% 5% Very highly prepared
Figure 7
Q: Consider the following areas. For each, rate your organization’s level of preparedness with respect to broadly adopting generative AI tools / applications?
(Oct./Dec. 2023) N (Total) = 2,835
17
Managing generative AI
implementation risk
Now: Key findings
Monitoring regulatory
Specific generative AI risks and concerns include inaccurate results and information (i.e., “hallucinations”);
47%
requirements and
Establishing a governance
legal risks such as plagiarism, copyright infringement, and liability for errors; privacy and data ownership ensuring compliance
framework for the use
challenges; lack of transparency, explainability and accountability; and systemic bias. The latter of generative AI tools / 46%
applications
exemplifies another category of risk in which AI amplifies and exacerbates a problem that already exists,
such as propagating and systematizing existing social biases, facilitating and accelerating the spread of
Conducting internal
42%
audits and testing
misinformation, helping criminals commit crimes, or fanning the flames of political divisiveness.
on generative AI
tools / applications
Training practitioners
According to the business and technology leaders we surveyed during fourth quarter 2023, the biggest
37% how to recognize and
mitigate potential risks
concerns related to governance were: lack of confidence in results (36%), intellectual property issues
(35%), misuse of client or customer data (34%), ability to comply with regulations (33%), and lack of
36% Ensuring a human
explainability / transparency (31%). validates all generative
AI content
Some of the surveyed organizations were already actively managing generative AI implementation
34%
Using a formal group
risks through actions such as monitoring regulatory requirements and ensuring compliance (47%), or board to advise
on generative
establishing a governance framework for generative AI (46%), and conducting internal audits and testing
AI-related risks
32%
Keeping a formal inventory
on generative AI tools and applications (42%) (figure 8). However, such organizations are in the minority of all generative AI
implementations
and their actions barely scratch the surface of the challenge. This is especially true given that regulatory
26%
requirements typically lag behind the pace of technology innovation—although a US presidential Using outside vendors
to conduct independent
executive order and the European Union’s ambitious Artificial Intelligence Act are clear signs government
audits and testing
21%
Single executive
leaders in many parts of the world are taking the issue of AI risk very seriously.
responsible for managing
generative AI-related risks
Figure 8
Q: What is your organization currently doing to actively manage the risks around
your generative AI implementations?
(Oct./Dec. 2023) N (Total) = 2,835
18
Generative AI is impacting talent
strategies now
2%
Never
10% 17%
No formal Now
time frame
24%
16% Within 1 year
Now: Key findings 2+ years
Generative AI barriers related to talent and workforce
Generative AI has the potential to supplement human workers across a vast array of activities traditionally
thought of as uniquely human. As such, its impact on talent and workforce strategies could be immense.
How will it affect organizations and their workers in the short and long runs? Which types of skills will be
most affected, and when?
31%
The vast majority of leaders we surveyed (72%) said they expect generative AI to drive changes in their
1-2 years
talent strategies sometime within the next two years: now (17%), within 1 year (24%), or in 1-2 years (31%)
(figure 9).
Figure 9
However, less than half (47%) reported that they are sufficiently educating their employees on the
Q: When do you expect to make changes to your talent strategies because of
capabilities, benefits and value of generative AI—survey respondents also cited a lack of technical talent and
generative AI?
skills as the biggest barriers to adoption.
(Oct./Dec. 2023) N (Total) = 2,835
19
Now: Key findings
Against this backdrop, some respondents reported making a high or very high effort to: It should be noted, however, that these reported workforce-related efforts might be limited
recruit and hire technical talent to drive their generative AI initiatives (42%), educate the in scope. Deloitte’s experience suggests that most organizations have yet to substantially
workforce about generative AI (40%), and reskill workers impacted by generative AI (36%). address the talent and workforce challenges likely to arise from large-scale generative AI
Those numbers are much higher for leaders who viewed their organization’s generative AI adoption. A likely reason for this is that many leaders don’t yet know what generative AI’s
expertise as very high (74%, 74% and 73%, respectively) (figure 10). talent impacts will be, particularly with regard to which skills and roles will be needed most.
Preparing workforces for generative AI: Respondents making a high or very high effort in the following areas.
74% 74%
73%
All respondants
55% 55%
42% 50%
40%
Little expertise
36%
30%
27%
24%
Some expertise
16%
14%
10%
High expertise
Recruiting and hiring technical talent to drive Educating our broader workforce to raise Reskilling workers because of the impact
Very high expertise
our generative AI initiatives overall generative AI fluency of generative AI to their roles
Q: What level of effort is your organization taking regarding the following workforce-related areas?
Figure 10
(Oct./Dec. 2023) N (Total) = 2,835
20
“Generating confidence
in workers’ abilities to
collaborate with generative AI,
now, could elevate creativity
and job satisfaction, next.”
21
Now: Key findings
51%
expect generative AI to
7 Leaders see significant societal impacts on the horizon.
increase economic inequality.
Although the leaders we surveyed were generally excited and enthusiastic about generative AI’s potential
business benefits, they were less optimistic about its broader societal impacts. Specifically, 52% of
respondents said they expected widespread use of generative AI to centralize power in the global economy,
while 30% expected it to more evenly distribute global power. Similarly, 51% expected generative AI to
increase economic inequality, while 22% expected it to reduce inequality (figure 11).
What’s more, 49% of respondents believe the rise of generative AI tools / applications will erode the overall
level of trust in national and global institutions.
Is this pessimism or realism? Our survey results appear to reflect the broader moral and ethical debates
about artificial intelligence that are occurring in every corner of society—even in the boardrooms of the
technology companies driving AI development, where AI’s commercial value is being weighed against its
potential value to serve humanity and AI’s potential benefits are being weighed against its potential risks.
The challenges that generative AI poses in corporate governance and risk parallel those in societal
governance and risk. In both domains, the technology’s potential benefits and potential harms are high.
National and supranational organizations and governments will likely need to walk the tightrope of helping
to ensure that generative AI benefits are broadly and fairly distributed, without overly hindering innovation
or providing an unfair advantage to countries with different rules.
22
Now: Key findings
Expected societal impacts of generative AI
Distribution of economic power
5% 25% 18% 42% 10%
30% 52%
distribute centralize
Levels of economic inequality
3% 19% 27% 41% 140%%
22% 51%
decrease inequality increase inequality
Q: How will widespread use of generative AI shift the overall distribution of power in the global economy?
Figure 11
Q: How will widespread use of generative AI tools / applications impact global levels of economic inequality?
(Oct./Dec. 2023) N (Total) = 2,835
23
Support for increased regulation
and global collaboration
Now: Key findings
8 Leaders are looking for more regulation and
78%
more regulation
collaboration globally.
Agree the widespread
proliferation of generative
In a break from traditional business norms, the unique risks associated with generative AI are prompting
AI tools / applications will
many business leaders to call for increased government regulation and increased global collaboration require more regulation
of AI by governments
around AI technologies.
Among the leaders in our survey, 78% said that more governmental regulation of AI is needed, while 72%
said there is currently not enough global collaboration to ensure the responsible development of
AI-powered systems (figure 12).
These results seem to |
185 | deloitte | state-of-gen-ai-report-wave-4.pdf | Now decides next:
Generating a new future
Deloitte’s State of Generative AI in the Enterprise
Quarter four report
January 2025
deloitte.com/us/state-of-generative-ai
Table of contents
Introduction
Key findings
Looking back at 2024
Now: Where we are
Next: Looking ahead
Considerations
Case studies
Authorship & Acknowledgments
About the Deloitte AI Institute
About the Deloitte Center for Integrated Research
About the Deloitte Center for Technology, Media & Telecommunications
Methodology
22
Introduction
Foreword
It was only about 10 years ago when visionary tech leaders started talking about enterprises to be a lot more structurally agile to adapt, embrace and innovate to stay
a future powered by ubiquitous computing and ambient intelligence. Back then it relevant and differentiated.
sounded like science fiction. Today, it’s real. No where is this future more evident than
In the following report, we see that most companies are transforming at the speed
in the rapid advancement and adoption of AI technologies. New models and tools are
of organizational change, not at the speed of technology. This is not surprising but is
gaining greater and greater capabilities and performing more complex reasoning. Even
something that will need to be addressed. That said, many are also already using
what was state of the art a few years ago pales in comparison to what we have today.
GenAI to create business value that exceeds their expectations—with compelling new
In this AI era, many now believe that Moore’s Law is effectively dead. And we have use cases emerging every day.
every reason to believe that the AI flywheel will continue to accelerate with every week
So, what do I say to clients who are in the trenches of this transformation? Don’t lose
and year—often referenced as the greatest secular shift of this quarter century.
focus. Stay curious, and challenge the orthodoxies of your organizations. GenAI and
Despite the technology’s rapid pace, I hear from clients and business leaders who are AI broadly is our reality—it’s not going away. While there are more questions than
wondering when it will meet their transformational expectations—when will business answers, but to stay in the game, leaders must be willing to try, do unconventional
leaders see the value and innovation that has been promised? things, learn and help mature.
Just like the internet, cloud, or even mobile, the transformational opportunities weren’t State of GenAI in the Enterprise is a snapshot in time of this great transformation. An
uncovered overnight. But as they became pervasive, they drove significant disruption opportunity for you to see where and how organizations across industries are finding
to business and technology capabilities, and also triggered many new business their way. I hope it serves to spark new ideas and new approaches that help illuminate
models, new products and services, new partnerships, and new ways of working and the path to your organization’s AI-fueled future.
countless other innovations that led to the next wave across industries. As we have
–Ranjit Bawa, Principal, US Chief Strategy and Technology Officer
experienced the half-life of these waves continues to be shorter. As such, it requires
3
Introduction
Generating a new future
For the past year, Deloitte has been conducting quarterly global survey reports and
executive interviews focused on Generative AI (GenAI) in the enterprise. We titled
our study Now decides next because we believed in GenAI’s potential to dramatically
transform how businesses operate—and that the actions companies take today will
have a decisive impact on their ability to succeed with GenAI in the future. And that’s
exactly what we found.
As with previous transformational technologies, the initial excitement and hype about
GenAI has gradually given way to a mindset of positive pragmatism. Many companies
are already seeing encouraging returns on their early GenAI investments. However,
those companies and others have learned that creating value with GenAI—and
deploying it at scale—is hard work. Although the technology at times seems like magic,
there is no magic wand when it comes to GenAI adoption, deployment, integration
and value creation.
44
Introduction
Key findings
There is a speed limit. Barriers are evolving. Some uses are outpacing others.
GenAI technology continues to advance at incredible Significant barriers to scaling and value creation are still Application of GenAI is further along in some business
speed. However, most organizations are moving at the widespread across key areas. And, over the past year areas than in others in terms of integration, return on
speed of organizations, not at the speed of technology. regulatory uncertainty and risk management have risen in investment (ROI) and expectations. The IT function is
No matter how quickly the technology advances—or organizations’ lists of concerns to address. Also, levels of trust most mature; cybersecurity, operations, marketing and
how hard the companies producing GenAI technology in GenAI are still moderate for the majority of organizations. customer service are also showing strong adoption and
push—organizational change in an enterprise can only Even so, with increased customization and accuracy of results. Organizations reporting higher ROI for their
happen so fast. models—combined with a focus on better governance— most scaled initiatives are broadly further along in their
adoption of GenAI is becoming more established. GenAI journeys.
All statistics noted in this report and its graphics are derived from Deloitte’s fourth quarterly survey, conducted July – September 2024; The
State of Generative AI in the Enterprise: Now decides next, a report series. N (Total leader survey responses) = 2,773. Percentages in this report
and its charts may not add up to 100, due to rounding.
Generative AI is an evolving area of artificial intelligence and refers to AI that in response to a query—a prompt—can create new text,
images, video and other assets. Generative AI systems can interact with humans and are built—or “trained”—on datasets that range in
size and quality from small language models (SLMs) to large language models (LLMs). Generative AI is also referred to as “GenAI.”
Evolving upon GenAI technologies, emerging AI agents are software systems that can complete complex tasks and meet objectives with little
or no human intervention. They are called “agents” because they have the agency to act independently, planning and executing actions to
achieve a specified goal. Related, the vision for agentic AI is that autonomous AI agents will be able to execute assigned tasks consistently and
reliably by acquiring and processing multimodal data, using various tools to complete tasks, and coordinating with other AI agents—all while
remembering what they’ve done in the past and learning from their experience.
5
Introduction
Key findings
The focus is on core business value. The C-suite sees things differently. Agentic AI is here.
A strategic shift is emerging, from technology catch-up Relative to leaders outside of the C-suite, CxOs tend Agentic AI is gaining interest as a breakthrough
to competitive differentiation with GenAI. Beyond the to express a rosier view of their organization’s GenAI innovation that could unlock the full potential of GenAI,
IT function, organizations tend to focus their deepest investments—and how easily and quickly GenAI’s with GenAI-powered systems having the “agency”
GenAI deployments on parts of the business uniquely barriers will be addressed and value achieved. It’s critical to orchestrate complex workflows, coordinate tasks
critical to success in their industries. that CxOs move on from being cheerleaders to being with other agents, and execute tasks without human
champions for achieving organizational efficiency and involvement. However, agentic AI is not a silver bullet and
market competitiveness. all the broad challenges currently facing GenAI still apply.
6
Introduction
Key findings
Our previous quarterly report said the clock was ticking model improvements—including domain and industry
to prove value—and this remains true today. Senior customization—and the promise of AI agents could
decision-makers might not be demanding tangible value help overcome inherent challenges and accelerate
About the State of Generative
and financial results from GenAI yet, but they soon will be. the creation of business value. However, it might be a
AI in the Enterprise:
multiyear journey for some organizations to reach full-scale
More and more organizations are moving from GenAI Wave four survey results
deployment and achieve the ROI they are looking for.
experimentation to deployment and scaling—with
The wave four survey covered in this report was fielded
proven use cases emerging and significant ROI being With GenAI, some level of uncertainty is unavoidable
to 2,773 director- to C-suite-level respondents across six
achieved through the most advanced GenAI initiatives. and the technology will likely continue to advance at
industries and 14 countries between July and September
a rapid pace. Business and technology leaders, for
2024. Industries included: consumer; energy, resources and
What’s more, despite some feelings of disillusionment
their part, should focus on what they can control—
industrials; financial services; life sciences and health care;
and unmet expectations, the vast majority of
namely, organizational readiness, particularly in areas technology, media and telecom; and government and public
organizations we surveyed are taking a realistic
services. The survey data was augmented by additional
such as data, risk management, governance, regulatory
perspective and showing sustained commitment in their insights from 15 interviews with C-suite executives and AI and
compliance and workforce / talent. Addressing issues
quest for value from GenAI, and they seem willing to data science leaders at large organizations across a range of
in these key areas will help position organizations for
industries. For details on methodology, please see p. 45.
do the hard work that needs to be done. Foundation
success with GenAI no matter how the future unfolds.
This quarterly report is part of an ongoing series by the
Deloitte AI InstituteTM to help leaders in business, technology
and the public sector track the rapid pace of Generative AI
change and adoption. The series is based on Deloitte’s
State of AI in the Enterprise reports, which have been
released annually the past five years. Learn more at
deloitte.com/us/state-of-generative-ai.
7
Real-world case studies
The case studies featured in this report are a small subset of the insights from our ongoing
in-depth interviews with business and AI leaders from a wide range of industries. The goal is
to build on the quantitative findings from our quarterly surveys by capturing practical, real-
world insights directly from leaders and organizations on the front lines of GenAI adoption.
Our interviews explore how leading organizations in diverse industries are
using GenAI to create value. Most notably, we are seeing initiatives focused
on applying GenAI to business-specific challenges in areas critical to success
in that organization’s industry. Examples include using GenAI for:
• B rand promotion and integrated business planning in the consumer products industry
• Predictive maintenance for physical assets in the energy industry
• Drug discovery and clinical trial tracking in the pharmaceutical industry
• Cybersecurity and portfolio management in the financial services industry
• S ales enablement, chip development and improved search in the technology industry
• A rchive management and music source separation in the media and entertainment industry
This focus on mission-critical activities suggests a broad strategic shift in the GenAI
landscape, from technology catch-up to competitive differentiation.
Go to case studies
8
Looking back at 2024
999
Now: Looking back at 2024 Level of interest in GenAI
(high + very high)
Looking back at 2024
Q1 Q4
Our first global quarterly survey, conducted in late 2023, revealed great excitement and expectations for GenAI.
However, those feelings were tempered by uncertainty and fear about the technology’s potentially negative impacts
Board 62% 46% -16 pts
on workers and society. Our second and third quarterly surveys focused more deeply on how organizations were
prioritizing tangible results and value creation from their GenAI investments, and on understanding and tackling
the barriers to successful scaling.
C-suite /
-15 pts
executive 74% 59%
A key finding during the year was that promising results from early GenAI pilots were raising expectations and
leaders
driving increased investment in the technology.
Today, interest and excitement about GenAI remain high. However, the initial fervor has gradually given way to a
Technical
86% 86%
leaders
positive yet pragmatic mindset—especially among business leaders at all levels. Meanwhile, technology leaders’
interest and excitement have remained high and steady (figure 1).
LOB /
Although this shift among business leaders might seem like a step backward for GenAI, it is entirely consistent with
functional 64% 56%
the usual life cycle for transformative technologies. It is also a net positive in terms of helping organizations move leaders
past the hype stage so they can directly tackle the serious work of using GenAI to create real business value.
Employees 49% 50%
A key finding during the year was that promising results
from early GenAI pilots were raising expectations and
Figure 1 Q: For the following groups in your organization, rate their
driving increased investment in the technology.
overall level of interest in Generative AI.
State of Generative AI in the Enterprise Survey,
Q1 (Oct./Dec. 2023) N (Total) = 2,774; Q4 (July/Sept. 2024)
N (Total) = 2,773; 14 countries common to both data sets
1100
Now: Looking back at 2024
Over the past year, as organizations gained experience with GenAI, they began to better “Data emerged as the central factor for [our GenAI] success,” said a former software
understand both the rewards and challenges of deploying the technology at scale— engineering manager for one of the world’s leading technology companies. “While
and adjusted their plans and expectations accordingly. Budgets have risen, and the the models and computing power existed, accessing the right data proved to be the
need for C-suites and boards to spur their organizations into action has diminished. biggest bottleneck. To address this, the company implemented a centralized data
At the same time, the need for disciplined action has grown. Technical preparedness strategy, managed by a single data leader, to streamline data acquisition and minimize
has improved, while regulatory uncertainty and risk management have become bigger redundancy—enabling faster model development.”
barriers to progress. Talent and workforce issues remain important; however, access to
specialized technical talent no longer seems to be the dire emergency it once was, at least
in comparison to other priorities. There has been one constant, however: improved data
management continues to be a top priority, even for companies that live and breathe data.
“Data emerged as the central
factor for [our GenAI] success …”
— Former software engineering manager for leading technology company
11
Now: Looking back at 2024
From a technology perspective, the capabilities of The vast majority of respondents (78%) reported they
foundation models and applications have improved expect to increase their overall AI spending in the next
dramatically over the past year. There are smaller, more fiscal year, with GenAI mostly expanding its share of
efficient models; better latency; bigger access windows; the overall AI budget relative to our first-quarter survey
expanded modalities; greater autonomy; and increased results. In particular, the percentage of organizations
model specialization. investing 20%–39% of their overall AI budget on
GenAI climbed by 12 points, while the percentage of
Reliability and trust have improved as well, although both
organizations investing less than 20% of their AI budget
still have a long way to go. Meanwhile, the adoption rate
on GenAI fell by 6 points.
for customized, open-source and/or proprietary large
language models (LLMs) remains limited at 20%–25% of “The way we do business has not changed,” said the VP of
those surveyed. artificial intelligence at a major media and entertainment
company. “For every project, our objective is always to do
Over the past year, respondents reported they
something that has a positive impact on the business. This
believe their organizations have most improved their
has not changed and is not going to change because it’s
GenAI preparedness in the critical areas of technology
what makes sense. However, a large proportion of project
infrastructure (+7 points) and strategy (+5 points). However,
proposals now have a [GenAI] component to them.”
preparedness has seemingly not improved in the other
critical areas of risk and governance and talent.
78%
of respondents expect to increase their overall AI spending
in the next fiscal year.
12
Now: Looking back at 2024
View from the C-suite
Relative to other respondents, the C-suite leaders (CxOs) in our survey generally demonstrated higher levels
of excitement and optimism about their organizations’ GenAI implementations. For example, 21% of C-suite
survey respondents reported they feel GenAI is already transforming their organization, compared to only 8%
of non-C-suite respondents. C-suite executives surveyed are comparatively less worried about barriers such
as trust, risk management, governance and regulatory compliance. They also have a rosier view of how quickly
their organization is moving, and how quickly the barriers to scaling and value creation will be addressed.
Sixty percent of non-C-suite respondents believe it will take 12 months or more to overcome scaling barriers,
compared to only 47% of C-suite respondents.
This doesn’t necessarily mean CxOs are out of touch with the challenges of adopting and deploying GenAI.
It could be they are still playing the primary role of catalyst or cheerleader and are in the process of learning
what it really takes to implement and scale GenAI. What will be important going forward is for CxOs to direct
that enthusiasm to removing barriers and enabling scaling.
Now that GenAI in the enterprise is moving past its infancy, CxOs should take on new roles, including those
of guide, counselor and challenger. Chief executive officers should show top-down support for GenAI, be the
champions for governance and risk initiatives, and foster an environment of trust and transparency. Chief
information officers, chief technology officers and chief data officers should sharpen their focus on identifying and
overcoming the barriers to large-scale GenAI deployment within their domains. Chief financial officers should
ensure responsible spending without stifling innovation. And chief human resource officers should promote
training, reskilling and other human capital investments.
13
Now: Looking back at 2024
The uneven pace of change
With transformational technologies, For businesses, embracing and integrating GenAI back from developing and deploying GenAI tools and
there are always gaps between the pace is much harder—and takes much longer—due to a applications (figure 2). This highlights respondents’
complex mix of factors. This could include dealing with unease about which use cases will be acceptable,
of technological change and the ability of
competing transformational priorities. However, policy, and to what extent their organizations will be held
individuals, businesses and policymakers
legislative and regulatory changes might be more accountable for GenAI-related problems.
to keep up. GenAI is no exception.
challenging overall.
This uneven pace of change creates friction for
Incredible advances in GenAI technology, fueled by
Governments today face the monumental task organizations, which likely contributes to the relatively
massive capital and intellectual investments from
of regulating a technology whose capabilities are moderate pace of transformation we are seeing as
tech companies, are already manifesting in individuals’
still taking shape. One direct consequence is that businesses work through their challenges on the path
everyday lives—through smarter smartphones,
regulatory compliance has emerged from the pack to creating sustained value with GenAI.
improved customer service, AI-enhanced search
to become the top barrier holding organizations
engines, and more.
Barriers to developing and deploying GenAI
Q1 Q4
+10 pts. +6 pts. -10 pts.
38%
36%
32%
28% 26% 26% 27% 26% 27%
25%
24%
22% 21% 20%
19%
18% 17% 17% 17%
15% 15%
14%
Worries about Difficulty Implementation Lack of technical Lack of a Difficulty Lack of an Trouble choosing Cultural Not having the Lack of executive
complying with managing risks challenges talent and skills governance identifying use adoption the right resistance from right comp. commitment
regulations model cases strategy technologies employees infrastructure / and/or funding
data
Figure 2 Q: What, if anything, has most held your organization back in developing and deploying Generative AI tools / applications? (Select up to three challenges)
State of Generative AI in the Enterprise Survey, Q1 (Oct./Dec. 2023) N (Total) = 2,774; Q4 (July/Sept. 2024) N (Total) = 2,773; 14 countries common to both data sets
14
Now: Where we are
1155
Now: Where we are
For our fourth wave report, we wanted to answer
several questions about scaling and value realization.
1
W here do things stand with workforce adoption?
2 H ow many experiments are organizations pursuing, and what are
their success rates?
3
W hich benefits are GenAI initiatives targeting?
4 A re some types of GenAI initiatives / use cases showing more promise
than others?
5
A re they meeting ROI expectations?
16
Now: Where we are
1
Where do things stand with workforce adoption?
Our latest survey results show that access to GenAI is still largely limited to less than enterprise, it generally makes sense to offer broad workforce access to sanctioned
40% of the workforce. Also, for most organizations, fewer than 60% of workers who GenAI tools, supported by clear guidelines for proper use.
have access to GenAI actually use it on a daily basis. This suggests many companies
“Currently, GenAI adoption is driven by internal demand, with early adopters seeking
have yet to integrate GenAI into their standard business workflows. It also raises the
to use the tools to meet their specific needs,” said the head of GenAI in product
chicken-and-egg question of whether limited access to GenAI is inhibiting comfort and
management at a major technology company. “However, we expect a shift towards
uptake with the technology (and stifling innovation), or whether the lack of high-value,
push-driven adoption in the next year, where all business units will be required to
innovative use cases is limiting interest and adoption.
integrate the platform as it becomes an approved and proven tool. This shift will create
For GenAI to become truly transformational, it will likely require greater numbers of pressure for teams to leverage the technology or risk missing out on the benefits it offers.”
workers experimenting and leveraging the technology to identify new, high-impact use
cases within the business. “Within our organization, the demand for GenAI use cases
and innovation primarily comes from middle management and employees, rather than
being driven by the C-suite,” said the director of product management for GenAI, cloud
“ Currently, GenAI adoption is driven
and data centers at a leading semiconductor company. “While the C-suite has been
slower to engage in AI implementation, teams across the company are developing by internal demand, with early
proofs-of-concept and driving AI adoption through internal boards and governance
adopters seeking to use the tools to
structures. This bottom-up approach emphasizes improving workflows and test cases,
with leadership providing support as needed for broader integration.”
meet their specific needs …”
Of course, access alone does not equate success. Providing access to GenAI does
not mean workers will use it. Conversely, workers with a burning desire to use GenAI
— H ead of Generative AI, project management at major technology company
will likely find a way to do so, with or without approval. However, in order to foster
transformation and maintain some level of control over how GenAI is used within the
17
Volume of experiments / POCs
Now: Where we are
2
35%
What is the state of
29%
24%
GenAI experimentation?
7%
We found organizations are still heavily experimenting
3% 3%
with GenAI, and scaling tends to be a longer-term goal.
Over two-thirds of respondents said that 30% or fewer
More than 100 51 to 100 21 to 50 11 to 20 Less than 10 Don’t know
of their current experiments will be fully scaled in the
Volume of experiments / POCs
next three to six months. This suggests companies are
taking time to test GenAI’s capabilities and to figure
out where it can help the most (figure 3).
The lion’s share of organizations are currently
pursuing 20 or fewer GenAI experiments or proofs of
concept (POCs) and expect to fully scale 10%–30% of
those experiments in the next three to six months.
As expected, individual company actions vary, with
larger numbers of experiments being conducted by
organizations that are large, advanced in their use of
AI, and/or operating in key industries of technology,
media and telecommunications; life sciences and
health care; or financial services.
Figure 3 Q: Approximately how many Generative AI experiments or proofs of concept is your organization currently pursuing? What percentage of these AI
experiments or proofs of concept do you anticipate will be fully scaled in the next three to six months?
State of Generative AI in the Enterprise Survey, (July/Sept. 2024) N (Total) = 2,773
snoitazinagro
fo
%
Scaling progress (next 3-6 months)
27%
26%
16%
13%
9%
5%
2% 2%
1%
80%
% of experiments / POCs
snoitazinagro
fo
%
70% 60% 50% 40% 30% 20% 10% 0%
18
Now: Where we are
Which benefits are GenAI
60%
initiatives targeting?
“Improved efficiency and productivity” continue to be
50%
the most commonly sought benefits from GenAI, and
many organizations (40%) reported they are already
achieving their expected benefits in this area to a large
40%
or very large extent. However, our respondents cited
slightly higher levels of success in a small handful of
more strategic benefit areas, particularly “new ideas
and insights” (46%) and “innovation and growth”
30%
(45%) (figure 4).
20%
10%
10% 15% 20% 25% 30% 35% 40% 45% 50% 55% 60%
deveihca
stfieneB
taht
%
eht
,ti
thguos
taht
seinapmoc
gnoma(
)tnetxe
egral
yrev
ro
egral
a
ot
ti
deveihca
Benefits achieved vs. benefits sought
Detect fraud and manage risk
Benefit sought
(% hoping to achieve the benefit)
Figure 4 Q: What are the key benefits you hope to achieve through your Generative AI efforts? (Select up to three benefits) To what extent are you achieving
those benefits to date?
State of Generative AI in the Enterprise Survey, (July/Sept. 2024) N (Total) = 2,773
gniveihca
3
Uncover new ideas and insights
Encourage innovation and growth
Enhance relationships with clients / customers
Improve
Increase speed / ease of developing new systems
efficiency and
productivity
Improve existing products and services
Shift workers from lower- to higher-value tasks
Increase revenue
Reduce costs
46%
seeking
of respondents (seeking the benefit)
reported that they are uncovering
new ideas and insights with GenAI.
19
Now: Where we are GenAI initiatives are most
4
advanced within these functions
Are some use cases showing more promise?
IT 28%
To understand where GenAI is having the deepest impact on organizations, we asked respondents to consider one
of their most advanced GenAI initiatives—an initiative that is most fully scaled—and then to identify which function or
Operations 11%
department it targets.
Marketing 10%
Since GenAI is a highly advanced technology—and one of its best capabilities is generating
Customer service 8%
computer code—it’s no surprise that the IT function came out on top (28%).
Cybersecurity 8%
However, the survey data also shows GenAI being deployed deeply in many other parts of the business as well, including
Product development 7%
operations (11%), marketing (10%), and customer service (8%) (figure 5a).
R&D 6%
5%
Sales
5%
Strategy
Supply chain 4%
Finance 4%
HR 2%
Manufacturing 2%
Legal, risk, compliance 1%
Figure 5a Q: Consider one of your organization’s most advanced (scaled)
GenAI initiatives. In which function or department is this initiative?
State of Generative AI in the Enterprise Survey, (July/Sept. 2024)
N (Total) = 2,773
20
Now: Where we are
Even more revealing, we found that the products company said: “Value creation is measured relatively mundane tasks secondary to the core business.
operationally by the acceleration of development “Our company has an enterprisewide AI leadership
most advanced GenAI applications outside
timelines, with AI providing faster results while staying team, but I think they’re really focused on a co-pilot
of IT overwhelmingly target critical business
within set performance and output quality constraints. strategy and helping all individuals use AI tools to improve
areas that are fundamental to success in a
Our focus is on development speed, rather than their productivity,” said the director of organizational
company’s specific industry (e.g., marketing in the
outperforming human capabilities. And while a tenfold transformation and change at a leading consumer
consumer industry; operations in energy, resources and
acceleration without human involvement remains products company. “We’re a little bit behind the eight
industrial; cybersecurity in financial services).
aspirational, a three- to five-fold increase in speed has ball on internal processes, and AI is sort of on the fringe.
already been realized.” I don’t think business-facing case studies have been
For example, in the life sciences and health care industry,
weaved into an overall enterprise AI strategy.”
where R&D is strategically important, the associate
This is a crucial insight since many business leaders still
director of artificial intelligence at a leading health care
associate GenAI with personal productivity and other
Top three most advanced (scaled) GenAI initiatives by industry
Color of the bubble represents the function
Industry Consumer Energy, resources & industrial Financial services Life sciences & health care Tech, media & telecom Government
Top 3 functions IT 20% Operations 23% IT 21% IT 23% IT 34% IT 96%
using GenAI
Marketing 20% IT 17% Cybersecurity 14% R&D 21% Product dev 17%
applications and
the percentage of Operations 3%
Customer service 12% Strategy 11% Finance 13% Operations 11% Cybersecurity 12%
initiatives in each
Figure 5b Q: Consider one of your organization’s most advanced (scaled) GenAI initiatives. In which function or department is this initiative?
State of Generative AI in the Enterprise Survey, (July/Sept. 2024) N (Total) = 2,773
21
Now: Where we are
5
Are advanced GenAI initiatives meeting ROI expectations?
Return on investment for organizations’ most advanced GenAI initiatives has been
generally positive. Almost all organizations report measurable ROI, and one-fifth of respondents say their most advanced
74%
(20%) report ROI in excess of 30%. Similarly, nearly three-quarters (74%) say their
Generative AI initiative is meeting or
most advanced initiative is meeting or exceeding their ROI expectations (43%
exceeding their ROI expectations.
meeting, 31% exceeding). Also, two-thirds (67%) say their most advanced initiative is
at least moderately integrated into their broader work processes (figure 6).
Most advanced (scaled) GenAI initiatives
ROI to date ROI expectations Level of integration
51% or more 6% Significantly above 7% Completely integrated 4%
31% to 50% 14% Somewhat above 24% Large extent 20%
11% to 30% 41% Meeting 43% Moderate extent 43%
6% to 10% 23% Somewhat below 19% Small extent 25%
Less than 5% 9% Significantly below 5% Not at all, but intend to 7%
Not measuring 5% No intention to integrate 2%
Figure 6 Q: ROI to date: Estimate the ROI to date for this specific initiative. / ROI expectations: How is the ROI from this Generative AI initiative meeting your organization’s expectations? / Level of integration: To what level is the Generative AI initiative integrated
into the broader organization |
186 | deloitte | sea-cons-genai-centre-of-adoption.pdf | GenAI Centre of
Adoption
Scaling GenAI for everybody
Deloitte Workforce Transformation SEA
October 2024
© 2024 Deloitte Consulting Pte Ltd 1
Contents
How GenAI has lowered the barrier to AI 4
Where organisations fall short of leveraging GenAI 5
What organisations can do 6
Measuring GenAI adoption 7
The Centre of Adoption 8
Contact us 9
2
© 2024 Deloitte Consulting Pte Ltd 2
Introduction
Generative AI, or "GenAI," is transforming the landscape of artificial intelligence.
Built on large language models (LLMs) that can create content – text, images,
videos – based on simple prompts, it is rapidly adopted particularly across the
Asia-Pacific (APAC) region, where organisations are moving beyond experiments
and proofs-of-concept to focus on scaling GenAI at an enterprise level.
With investments surging and the Generative AI market projected to hit $200
billion by 20321, its momentum is evidently accelerating.
Despite its rapid growth, many organisations still struggle to fully leverage the
benefits of GenAI.
The urgency for mass adoption, rather than fragmented efforts, is proving critical
to unlocking its full potential.
By doing so, businesses can rapidly prove value through next level innovations,
efficiency, and competitiveness. For this to happen, it is essential that all
employees actively participate in this transformation.
In this paper, we explore what’s truly happening within organisations—
where employees are already using GenAI, regardless of formal management’s
endorsement. We identify four critical gaps in scaling GenAI, particularly from a
people perspective, and discuss how organisations can elevate employees’
proficiency and maturity around GenAI adoption through targeted capability
building.
We introduce the Center of Adoption (CoA) as an approach to drive safe, scalable,
and effective outcomes around GenAI adoption across key functional areas.
Source: 1Generative Artificial Intelligence | Deloitte US
©©© 222000222444 DDDeeellloooiiitttttteee CCCooonnnsssuuullltttiiinnnggg PPPttteee LLLtttddd 333
How GenAI has lowered the barrier to AI
Generative AI has democratised advanced AI by putting it in the
hands of every employee
It has put advanced AI in the hands of everyone
Did you know?
GenAI tools and trainings are readily accessible at little to no cost
(or infrastructure investment), thanks to an abundance of platforms,
tutorials and communities
43% of employees across Asia
Pacific are using GenAI for
work. Southeast Asia ranks
2ndout of 9 locations in APAC
for GenAI use
It takes the form of things we already know
In APAC, GenAI daily users save
6 hours a week through
No formal training is required to start using GenAI tools, which are
increased speed, qualityof
often modelled on familiar user interfaces (e.g., chat bots)
work, and the ability to generate
new ideas
19%of GenAIusers in Southeast
Asia are Daily Active Users.
This is expected to increase by
It makes people faster – and better – at what they do
232% in the next five years
The more you use GenAI, the better you will get, as it learns
how you work and how to provide ever better responses
Source:
Deloitte Generative AI in Asia Pacific Report
©© 22002244 DDeellooiittttee CCoonnssuullttiinngg PPttee LLttdd 44
Where organisations fall short of leveraging GenAI
GenAI makes AI more accessible than ever, but its full potential can only be
realised if four critical gaps are addressed
The LEADERSHIP Gap: The GOVERNANCE Gap: The EFFICIENCY Gap: The SCALE Gap:
Not creating an environment Undefined ethical standards Not fully leveraging the Not scaling use cases beyond
that encourages, celebrates, and rules of use technology and its capabilities individual application
and drives adoption
• Lack of awareness and urgency, as • Unmonitored or unchecked • Inefficient or sub-optimal use of • Involves uneven and isolated use of
decision-makers do not frequently platform usage prompts/tools/platforms tools/platforms
engage with GAI platforms
• Negative consequences (e.g., data • Increased human error, poor • Benefits only select individuals
• Lack of institutional mechanisms leak) machine responses and at negligible scale
to drive adoption
• Significant riskfor the organisation • Modest efficiency gains (e.g., bad • Neglects multiplier gains that can
prompts) only be reaped through function or
organisation wide application
• While 94% of employees are ready to • A study by Layer X found that 6% of • A multidisciplinary study found that • 70%of organisations with scaled GAI
learn new skills to work with GenAI, workers havecopy-pasted sensitive sub-optimal GenAIusage can reduce capabilitiesreport improved products
only 5% reported that their employers informationinto GAI tools3 worker performance by 19%4 and services2
were providing training on a large
scale1 Alarmingly, 4% of employees were • Workers expect 61% of current tasks 63% of similar organisations have
found to do so weekly3
to be impacted by GenAIin the next 5 reported being able to encourage
years5 innovationand growth2
Sources: 1Harvard Business Review | 2Deloitte State of Generative AI in the Enterprise | 3Layer X | 4MIT Sloan | 5Deloitte Asia Pacific Generative AI Report
©© 22002244 DDeellooiittttee CCoonnssuullttiinngg PPttee LLttdd 55
What organisations can do
Focus on your employees to drive adoption and address critical people gaps,
before investing in platforms
Bottom-up Top-down
1 2
(Enabling the workforce) (Starting with use cases for the business)
ALIGN GUIDE UPLIFT TRAIN SECUR E ARCHIT EC T
Create a common Demonstrate how Provide practical Make it easy for Ensure safe and Establish the
understanding of GenAI can be learning pathways employees to reliable GenAI infrastructure to
GenAI concepts, beneficial to all and opportunities embed GenAI use across all support
language, and employees, and to bring up into their day-to- teams and increasingly
guardrails across help them to see employees’ core day ways of functions advanced use
all employee the business GenAI skills working cases and data
groups impact needs
Focus of this whitepaper
©© 22002244 DDeellooiittttee CCoonnssuullttiinngg PPttee LLttdd 66
Measuring GenAI adoption
Ground your approach in five levels of maturity
L5: ENTERPRISE
ENABLER
L4: TEAM ENABLER
Champions department
and enterprise-wide
L3: ACTIVE USER
Drives and checks efficient
GenAI adoption and
and ethical GenAI use within
integration
L2: LEARNER
the team
Proficient daily use of GenAI
• Focuses on scaling GenAI
to perform strings of tasks, as
L1: NOVICE Trained to effectively and • Ensures AI tools are applied capabilities into strategic
part of BAU workflows
initiatives
efficiently perform specific responsibly, with attention to
Minimal engagement and Gen AI use cases ethical considerations and • Ensures that adoption aligns
• Fully proficient in using GenAI
experimentation of Gen AI for data privacy with business goals
• Actively builds GenAI skills for across various tasks
everyday use
specific use cases (e.g., • Mentor teams to elevate their
• Integrates GenAI into daily
• Little engagement with GenAI content generation) GenAI use and foster
workflows to optimise
responsible GenAI innovation
• Limited understanding of • Understands the potential of processes and enhance
how GenAI can help them GenAI in their role productivity
All employees should strive to be at L2, while functional leaders are expected to be at L5. This may vary between organisations, depending on the strategic,
business and organisational objectives
©© 22002244 DDeellooiittttee CCoonnssuullttiinngg PPttee LLttdd 77
The GenAI Centre of Adoption (CoA)
Through a dedicated and centralised team, organisations can accelerate GenAI
adoption while bridging critical gaps
Key Outcomes
Adoption Speed of Adoption Proficiency
% increase in adoption over period Time taken to go from one level to the next # target employees at a discrete level
WHAT IS A COA?
Customer & Corporate
Technology
Marketing Affairs
Dedicated unit designed to drive the mass adoption of GenAI across all
functions and levels within the organisation.
Central hub focused on building a culture of continuous learning and
Sales Finance
GenAI CoA
innovation to maximise GenAI’s value.
CHRO
People & Regulatory KEY ACTIVITIES
Org. / HR Compliance
Develop and drive a shared GenAI taxonomy, practical standards (non-
Head of AI
technical), governance, and ethical guardrails
Operations Legal
Help employees to identify areas where GenAI will benefit them,
experiment safely, and help them to redefine their roles
Supply Enterprise
Data Privacy Identify skill and competency gaps across leadership and employee
Chain Risk
groups; develop learning pathways, experiences and programs
©© 22002244 DDeellooiittttee CCoonnssuullttiinngg PPttee LLttdd 88
Contact Us
Let’s unlock GenAI’s full potential in your organisation
Indranil Roy Christopher Lewin Clarissa Turner Miri Takakura
Executive Director Executive Director Executive Director Director
Workforce Transformation AI & Data Leader Workforce Transformation Workforce Transformation
Southeast Asia
Southeast Asia Malaysia Singapore
[email protected] [email protected] [email protected] [email protected]
+65 9636 8024
+65 6232 7128 +60 3 7610 7233 +65 9181 7044
Authors & Contributors
Christian Teo Nadiah Johari Chanette Teoh Claire Ng
Editor Sub-Editor Author Author
GenAI CoA Workforce Transformation Workforce Transformation Workforce Transformation
David Ng Pui Leung Cyndi Chan
Author Designer
AI & Data CoRe Creative Services (C&M)
©© 22002244 DDeellooiittttee CCoonnssuullttiinngg PPttee LLttdd 99
Deloitte refers to one or more of Deloitte Touche Tohmatsu Limited (“DTTL”), its
global network of member firms, and their related entities (collectively, the
“Deloitte organization”). DTTL (also referred toas “DeloitteGlobal”) and each of its
member firms and related entities are legally separate and independent entities,
whichcannotobligateorbindeachotherinrespectofthirdparties.DTTLandeach
DTTL member firmand related entity isliable onlyforits own acts and omissions,
andnotthoseofeachother.DTTLdoesnotprovideservicestoclients.Pleasesee
www.deloitte.com/abouttolearnmore.
DeloitteAsiaPacificLimitedisacompanylimitedbyguaranteeandamemberfirm
ofDTTL.MembersofDeloitteAsiaPacificLimitedandtheirrelatedentities,eachof
whichisaseparateandindependentlegalentity,provideservicesfrom morethan
100citiesacrosstheregion,includingAuckland,Bangkok,Beijing,Bengaluru,Hanoi,
Hong Kong, Jakarta,KualaLumpur,Manila,Melbourne,Mumbai,NewDelhi,Osaka,
Seoul,Shanghai,Singapore,Sydney,TaipeiandTokyo.
Thiscommunicationcontainsgeneralinformationonly,andnoneofDTTL,itsglobal
network of member firms or their related entities is, by means of this
communication, rendering professional advice or services. Before making any
decision or taking any action that may affect your finances or your business, you
shouldconsultaqualifiedprofessionaladviser.
Norepresentations,warrantiesorundertakings(expressorimplied)aregivenasto
theaccuracyorcompleteness oftheinformationinthiscommunication,andnone
of DTTL,itsmember firms,related entities, employees or agents shall beliable or
responsible for any loss or damage whatsoever arising directly or indirectly in
connectionwithanypersonrelyingonthiscommunication.
©2024DeloitteToucheTohmatsu
DesignedbyCoReCreativeServices.RITM1858095
© 2024 Deloitte Consulting Pte Ltd 10 |
187 | deloitte | Q3 StateOfGenAI_Report_Wave3_v6.pdf | Now decides next:
Moving from potential
to performance
Deloitte’s State of Generative AI in the Enterprise
Quarter three report
August 2024
deloitte.com/us/state-of-generative-ai
Table of contents
Foreword
Introduction
Now: Key findings
1 Building on initial success
2 Striving to scale
3 M odernizing data foundations
4 M itigating risks and preparing for regulation
5 M aintaining momentum by measuring value
Next: Looking ahead
Authorship & Acknowledgments
About the Deloitte AI Institute
About the Deloitte Center for Integrated Research
About the Deloitte Center for Technology, Media & Telecommunications
Methodology
22
Introduction
Foreword
In the rapidly evolving landscape of artificial intelligence (AI), the connection between The complex discussions around creating value and managing risk makes it clear to me
technology and value has become increasingly apparent. What is known about major that we need to keep humans at the center of all this decision-making. It is the human
technology innovations in the past holds true with Generative AI (GenAI). stakeholders who impact how applications are conceived and developed, how they are
adopted and used, and how they are managed for trust and security. In this, employee
Technology application on its own is not enough. Results and business outcomes
upskilling and change management remain indispensable elements of value-driving
matter. The real measure of success for GenAI will be how it enables enterprise
GenAI programs.
strategies and drives tangible value.
With a focus on business outcomes and human-centered change, I feel the future with
As organizations are scaling, and learning from, their GenAI pilots, I have heard the
GenAI grows brighter by the day, even as the journey ahead will continue to surprise
discourse around GenAI shift from unbridled excitement to a more nuanced and
and challenge us.
critical evaluation of its real impact on business outcomes. I am also beginning to see
organizations think more about tailored GenAI tools—evolving from large language Learn more about the series and sign up for updates at
models (LLMs) to small language models (SLMs) for more targeted needs. They are http://deloitte.com/us/state-of-generative-ai.
also exploring how the rise of AI agents can redefine interactions within their digital
–Jim Rowan, Applied AI SGO Leader
environments, offering new avenues for automation and personalization.
Amid this maturation, regulatory considerations are coming to the fore. Our past
survey results indicated a strong market appetite for smart GenAI regulation and
oversight. Businesses and governments alike are navigating a dynamic landscape
and are struggling to keep pace with the rate of technology innovation. The challenge
is to unlock the benefits of GenAI while facing regulatory uncertainty, orchestrating
governance and building trust. No small task.
33
Introduction
Moving from potential to performance
The clock is ticking for organizations to create significant cases with strong return on investment (ROI) and a clear Generative AI-powered applications? Is regulatory
and sustained value through their Generative AI path to scale will be essential. They’ll need to address uncertainty holding them back? Are they developing a
initiatives. Promising pilots have led to more investments, challenges across the board: people, process, data and comprehensive set of financial and nonfinancial measures
escalating expectations and new challenges. During this technology. Change management and organizational to form a complete picture of benefits achieved? These
pivotal phase, C-suites and boards are beginning to look transformation will need to be given as much consideration questions must be explored in-depth as organizations
for returns on investment. There is a chance that their as technology. journey from Generative AI promise to performance.
interest in Generative AI could wane if initiatives don’t
In this quarter’s survey, we focused on two critical areas to
pay off as much, or as soon, as expected.
scaling—data and governance, and risk and compliance—
Will organizations demonstrate the patience and and how organizations are measuring and communicating
perseverance needed to unlock the transformational value. Are data-related issues hindering efforts? How
potential of Generative AI? To get there, value-led use are organizations ensuring the right oversight of
44
Introduction
Moving from potential to performance (cont’d)
Building on initial success Striving to scale
• I mproved efficiency and productivity and cost reduction are still the top benefits • T wo of three surveyed organizations said they are increasing their investments in
sought by organizations. Those are also cited by 42% of respondents as their most Generative AI because they have seen strong early value to date.
important benefits achieved to date.
• H owever, many are still challenged to successfully scale that value—nearly 70% of
• H owever, 58% reported they realized a more diverse range of most important respondents said their organization has moved 30% or fewer of their Generative AI
benefits, such as increased innovation, improved products and services, or experiments into production.
enhanced customer relationships.
• R espondents said that embedding Generative AI deeply into critical business
functions and processes is the top way to drive the most value from their
Generative AI initiatives.
All statistics noted in this report and its graphics are derived from Deloitte’s third quarterly survey, conducted May
– June 2024; The State of Generative AI in the Enterprise: Now decides next, a report series. N (Total leader survey
responses) = 2,770. Percentages in this report and its charts may not add up to 100, due to rounding.
Generative AI is an area of artificial intelligence and refers to AI that in response to a query can create text,
images, video and other assets. Generative AI systems can interact with humans and are often built using large
language models (LLMs). Also referred to as “GenAI.”
5
Introduction
Moving from potential to performance (cont’d)
Modernizing data foundations Mitigating risks and preparing Maintaining momentum
for regulation by measuring
• Three-quarters of respondents said their organizations
have increased investment around data life cycle
• O rganizations feel far less ready for the challenges • M ore than 40% of respondents said their companies
management to enable their Generative AI strategy.
Generative AI brings to risk management and governance— are struggling to define and measure the exact impacts
Top actions include enhancing data security (54%)
only 23% rated their organization as highly prepared. of their Generative AI initiatives.
and improving data quality (48%).
• In fact, three of the top four things holding organizations • Less than half said they are using specific KPIs to
• Data issues are limiting options—55% of organizations
back from developing and deploying Generative AI tools measure Generative AI performance, and many
reported avoiding certain Generative AI use cases
and applications are risk, regulation (such as the European standard measures of success aren’t currently
because of data-related issues. Top data-related
Union’s AI Act, in effect August 1), and governance issues. being applied.
concerns include using sensitive data in models
and managing data privacy and security. • T o deal with regulatory uncertainty, about half of
organizations reported they are preparing regulatory
forecasts or assessments.
About the State of Generative AI in the Enterprise:
Wave three survey results
The wave three survey covered in this report was fielded to 2,770 director- to C-suite-level respondents across six industries and 14 countries
between May and June 2024. Industries included: Consumer; Energy, Resources & Industrials; Financial Services; Life Sciences & Health Care;
Technology, Media & Telecom; and Government & Public Services. The survey data was augmented by additional insights from 25 interviews
with C-suite executives and AI and data science leaders at large organizations across a range of industries. This quarterly report is part of an
ongoing series by the Deloitte AI InstituteTM to help leaders in business, technology and the public sector track the rapid pace of Generative AI
change and adoption. The series is based on Deloitte’s State of AI in the Enterprise reports, which have been released annually the past five
years. Learn more at deloitte.com/us/state-of-generative-ai.
66
Now: Key findings
77
Now: Key findings Top benefit achieved through
Generative AI initiatives
1
Building on initial success
Organizations say they are seeing value from their early Generative AI forays and those successes are driving more
investment. Two-thirds of the organizations we surveyed (67%) said they are increasing investments in Generative AI Improved 34%
efficiency and
because they have seen strong value to date. A head of AI strategy and governance in the banking industry has seen
productivity
this first-hand: “Before GenAI, most senior leaders only had a vague understanding of what AI was or what it can do. Now,
they have AI at their fingertips, and it has opened their eyes to the possibilities. We have applied for additional resources.”
12% Encouraged
innovation
As in our prior quarterly surveys, improved efficiency and productivity and cost reduction continue to be the most
Improved 10%
common benefits sought from Generative AI initiatives. Those benefits were cited by 42% of wave three respondents
existing
products and 9% Reduced
as their single, most important benefit achieved to date (figure 1).
services costs
However, for most wave three respondents (the other 58%), the top benefit achieved through the new technology is Enhanced 9%
relationships
something other than efficiency, productivity or cost reduction. This includes increased innovation (12%), improved 7% Increased speed
with clients /
and/or ease
products and services (10%), and enhanced customer relationships (9%). The diversity of possible sources of value from
customers
of developing
Generative AI initiatives is exciting to many leaders and shows the potential and versatility of this new technology.
new systems /
Increased 6%
software
revenue
This distribution could mean a couple of different things. Organizations may be seeking efficiency, productivity and
6%
Developed
cost reduction, but aren’t seeing it materialize yet; they may be getting unexpected value from less tangible areas; or
new products
they may be prioritizing these other types of value. There is no one-size-fits-all approach to employing Generative AI, Shifted 4% and services
workers from
and there is a wide range of benefits that could be gained. It is important for organizations to be clear about what
4% Better
lower- to higher-
detection of
kind of value they are seeking before embarking on any Generative AI initiatives—start with value first. value tasks
fraud and risk
management
67%
of organizations we surveyed said they are increasing
Figure 1 Q: What is the most important benefit your organization
investments on Generative AI given strong value seen to date. has achieved to date through your Generative AI initiatives?
(May/June 2024 ) N (Total) = 2,770
88
Now: Key findings
Our executive interviews provided examples of Generative AI use cases that are already and greater innovation and market differentiation, most projects further along in the
delivering real-world value across a wide range of industries. Although they are working scaling process are still focused on improving productivity (figure 2).
toward things like automated decision-making, accelerated research and development,
Generative AI use cases delivering real-world value by industry
Banking Transportation Telecom Insurance Consumer Technology Finance Pharmaceuticals
A customer service A system to provide Support tools An internal medical Customer Continuous Project management Internal tool that
tool that handles customer support deployed for retail claims appeal review segmentation tools improvement tools that quickly provides instant
messages, using both and handle simple and technical field tool that provides leveraged to create processes enhanced create summary enterprise information
chat and voice, and support tickets. staff, and systems increased response more precise and by directly leveraging materials for key (such as standard
provides cross-sell The system can for troubleshooting quality and a decreased customized segments customer feedback stakeholders. operating procedures)
opportunities based automatically pull and preventive time to respond. across geographies. to inform product for thousands of staff.
on the interaction. data for human maintenance, all to development
agents to use for reduce costs. road maps.
more complex tasks.
Figure 2
9
Now: Key findings Behaviors driving the most value
for Generative AI initiatives
What do organizations think will most help drive greater “CEOs and executive leadership teams are getting much
value for their Generative AI initiatives? While many more excited and interested in what’s possible and are
Deeply embedding GenAI into
22%
different factors contribute to Generative AI value looking for use cases to demonstrate the value and
functions / processes
creation, the action cited most often by the leaders benefit,” said the global head of AI, machine learning,
we surveyed is embedding the technology deeply into analytics and data at a pharmaceutical company. “There is Effectively managing risks 13%
business functions and processes (figure 3). a lot of willingness to test, experiment and scale. However,
Deploying the latest
the potential danger is that people might get disappointed
11%
“An LLM is like an engine,” said a VP at a bank’s AI center technology
and lose attention if it’s not paying off fast enough.”
of excellence. “No one just wants the engine of a car
Developing creative and
10%
or a plane; they want a car or a plane. So, there are all C-suite and board members are still intrigued, but there
differentiated applications
these things you need to do to make it part of business are some potential signs of enthusiasm beginning to
Tailoring / customizing
processes, so the business can use it.” The value from wane as the “new technology shine” wears off. Survey
10%
models with proprietary data
any Generative AI initiative won’t be fully realized if it sits respondents said that interest in Generative AI remains
apart. As with other technologies, it will only reach its “high” or “very high” among most senior executives
Hiring the best talent 9%
potential when it is embedded in everyday tasks. Many (63%) and boards (53%); however, those numbers
organizations are already employing enterprise tools have declined since the Q1 2024 survey, dropping 11
Completely measuring
8%
enhanced with this emerging technology resource to percentage points and 8 percentage points respectively. performance
try and make this happen. Time is of the essence as organizations look to scale
their early achievements. Providing enough budget 8%
Although many have seen promising results from
early projects and are increasing investment in
Providing access to as much
7%
of the workforce as possible
Generative AI, it is important that organizations
show sustained and significant value as quickly
as possible.
Figure 3 Q: Which behavior / action do you think will drive the most
value for the Generative AI initiatives in your organization?
(May/June 2024 ) N (Total) = 2,770
10
Now: Key findings
2 Striving to scale A large majority of organizations have deployed less than
a third of their GenAI experiments into production
Selecting and quickly scaling the Generative AI projects with the most potential to create
value is the goal. However, many Generative AI efforts are still at the pilot or proof-of-
Organizations
26%
concept stage, with a large majority of respondents (68%) saying their organization has 24%
GenAI experiments
19%
moved 30% or fewer of their Generative AI experiments fully into production (figure 4).
moved into production
14%
This isn’t necessarily surprising—despite rapid and impressive advances in Generative AI’s
7%
capabilities, its applications are still relatively new and organizations are figuring out what it 4%
3%
1% 1%
can (and can’t) do well. Many organizations are learning through experience that large-scale
Generative AI deployment can be a difficult and multifaceted challenge. As with a lot 0% 10% 20% 30% 40% 50% 60% 70% 80%
of digital transformation efforts, projects can fail or struggle for a variety of reasons.
Figure 4 Q: In your estimation, what percentage of your Generative AI experiments have been deployed to date into your
“Most of our applications are still in the minimum-viable-product or proof-of-concept organization (moved into production)?
phase,” said a senior specialist for AI compliance in the automotive industry. (May/June 2024 ) N (Total) = 2,770
“Scaling across an organization where
Successfully scaling may mean different things to different organizations—based
on their goals, what approach they are taking with Generative AI, and to what
you have thousands of employees
extent scaling is actually necessary. They could be expanding from one market
to multiple markets, from a small group within a function to the entire function,
has several basic requirements, and
or from a portion of a process to multiple, integrated processes. It also depends
they’re quite challenging.” on what Generative AI-powered tools and applications are being used: scaling a
code generator across an IT department is going to be different than scaling a
customized LLM for the finance function, or a new enterprise customer relationship
-Senior specialist for AI compliance in the automotive industry
management application with Generative AI features.
11
Now: Key findings
Despite these differences, some fundamentals are consistent. More broadly, organizations should invest in the foundations of Generative AI
and concurrently assess and advance their strategy, processes, people, data and
“Foremost, you need a strategy,” the senior specialist for AI compliance continued.
technology (figure 5).
“Strategy means you can’t start by purchasing separate solutions ... if you really want
to scale, first you need to base your strategy on platforms.” Many of the fundamentals may look similar to prior digital transformation
efforts, but due to the unique nature of Generative AI, things like robust
This platform-centric approach could include establishing centers of excellence, technology
governance, transparency for building trust, transforming talent, and
platforms to enable multiple use cases, and centralized teams of experts. In our Q2 report
mature data life cycle management take on increased importance.
we advocated for centralized resources that can accelerate deployment of similar use
cases and enable organizations to make the most of scarce Generative AI expertise.
Essential elements for scaling Generative AI initiatives from pilot to production
Figure 5
Strategy Process Talent Data & technology
Ambitious Modular
Integrated Transparency Provisioning
strategy & value Robust architecture
risk to build trust the right AI
management governance and common
management in secure AI infrastructure
focus platforms
Clear, Agile Acquiring Effective
Strong Transformed
high-impact operating model (external) and Modern data model
ecosystem roles, activities
use case and delivery developing foundation management
collaboration and culture
portfolio methods (internal) talent and operations
12
Now: Key findings
How do organizations feel like they are doing across these areas—are they prepared the LLMs still needs to be improved … Data readiness; data is going to be problem
to scale? We asked how highly prepared respondents thought their organizations were forever ... Deep Generative AI understanding as well. There’s not enough people who
across some of the essential scaling elements (figure 6). Technology infrastructure understand and can drive transformation.”
(45%) and data management (41%) fared the best, followed by strategy (37%), risk
To help start a conversation on how to overcome some of these barriers, in
and governance (23%), and talent (20%).
this quarter’s survey we focused on two areas critical to scaling—exploring
This indicates that there are still some fundamental challenges holding organizations how organizations are approaching data and governance, and risk
back from successfully scaling their Generative AI initiatives. A senior director and and compliance.
head of a Generative AI accelerator in the pharmaceutical industry identified a
With respect to data, more organizations’ leaders reported they are initially prepared.
number of pressing issues: “The heritage of our processes and approaches, that
For risk and governance, they know they are not. Both need attention.
is what’s really holding us back right now. Number two is that the performance of
Do organizations think they are ready?
Percentage of organizations that are highly prepared
for GenAI across the following areas 45% 41%
37%
23%
20%
Figure 6 Q: For each area, rate your organization’s level of preparedness Technology Data Strategy Risk & Talent
with respect to broadly adopting generative AI tools / applications?
infrastructure management governance
(May/June 2024 ) N (Total) = 2,770
13
Now: Key findings
3 Modernizing data foundations
75%
of organizations have increased their technology investments
around data life cycle management due to Generative AI.
Compared with the other aspects of Generative AI However, even those executives who consider themselves
readiness, survey respondents judged that their highly prepared will likely need to do more as they progress
organizations are fairly mature with respect to data life in their journeys. Some we interviewed said that as they
cycle management (as a reminder, survey respondents moved from proof of concept to scale, unforeseen data
are from more AI-savvy organizations). This could be issues were exposed—highlighting a need to be agile.
because they had a good foundation to start with or
These issues could be because of the Generative AI-
that, according to our survey, 75% of organizations have
specific demands to data architecture and management.
increased their technology investments around data life
More robust governance—quality, privacy, security,
cycle management due to Generative AI.
transparency—is needed overall, especially around using
This increased focus was evident in our executive data that doesn’t already exist inside the organization (e.g.,
interviews. “There’s a whole series of questions GenAI public domain, synthetic and licensed third-party data).
is triggering about data strategy, that in the past Documenting data sources and labeling has an increased
were far less important,” said the chief technology importance. With more people potentially leveraging
officer at a manufacturing company. “I think we’re data, data access frameworks and literacy require more
probably spending as much time on data strategy and attention. It may change approaches toward cloud or on-
management as on pure GenAI questions, because premises data services. For more advanced LLM users,
data is the foundation for GenAI work.” working with synthetic data may eventually come into play.
1144
Now: Key findings
Levels of concern around data management
Figure 7 Q: For the following, how much concern does your organization have with respect to its data management for
Generative AI implementations?
(high + very high)
(May/June 2024 ) N (Total) = 2,770
58% 58% 57% 49% 38%
Using sensitive data Managing data privacy- Managing data security- Complying with data- Using our own proprietary
in models related issues related issues related regulations data in models
One of these challenges was highlighted by a former vice president of data and intelligence That could be because of data-quality issues, intellectual property concerns, not having
for a media and entertainment company: “The biggest scaling challenge was really the the right data, or worries about using certain kinds of data (e.g., public domain, synthetic
amount of data that we had access to and the lack of proper data management maturity. or licensed third-party data). The concerns that organizations were worried about the
There was no formal data catalog. There was no formal metadata and labeling of data most in our survey included using sensitive data in models (58% had at least a high
points across the enterprise. We could go only as fast as we could label the data.” level of concern), data privacy issues (58%), and data security issues (57%) (figure 7).
Organizations were much more worried about using sensitive data (e.g., customer
Data-related issues could be hindering organizations in their quests for getting
or client data) than they were using their own proprietary data (e.g., sales,
the levels of value that they are seeking. Data-related issues have caused 55%
operational, financial).
of the organizations we surveyed to avoid certain Generative AI use cases.
15
Now: Key findings Improving data-related capabilities
Consistent with those concerns, the top actions The value from Generative AI initiatives will increasingly
organizations are taking to improve their data-related come from organizations leveraging their differentiated
capabilities are enhancing data security (54%), improving data in new ways (whether for fine-tuning LLMs, building
Enhanced 54%
data quality practices (48%), and updating data an LLM from scratch or utilizing enterprise solutions).1 data security
governance frameworks and/or developing new For Generative AI to deliver the kind of impact executives
48%
data policies (45%) (figure 8). expect, companies will likely need to increase their Improved
data quality
comfort with using their proprietary data, which may
practices
be subject to existing and emerging regulations.
Updated 45%
governance
frameworks /
Developed new
43% Collaborated
data policies
with cloud
service provider
“Data quality is key. Understanding what data is or IT integrator
Upgraded IT 37%
to improve
infrastructure
capabilities
good data. Where is that data held? How is it
34% Hired new
talent to fill
secured? How is it permissable? All those things
data-related
Integrated 27%
skill gaps
data silos
are key to making [Generative AI] scalable.”
24% Moved to a
more flexible,
-Chief operations officer & chief of strategy for a financial services firm
open data
architecture
Figure 8 Q: What specific actions has your organization taken
to improve its data-related capabilities to support its
Generative AI initiatives?
(May/June 2024 ) N (Total) = 2,770
16
Now: Key findings
4 Mitigating risks and preparing for regulation
According to our survey respondents, Likely driving these concerns are new and emerging as highly prepared. These issues will be increasingly
risks specific to the new tools and capabilities—like important as activities shift from small-scale pilots to
three of the top four barriers to successful
model bias, hallucinations, novel privacy concerns, trust large-scale deployments and Generative AI becomes
development and deployment of
and protecting new attack surfaces. This environment more deeply embedded into the fabric of organizations.
Generative AI tools and applications are:
may be why organizations feel far less ready for the Highlighting the importance, respondents selected
challenges Generative AI brings to risk management and effectively managing risks as the second-most reported
worries about 36%
governance—since only 23% rated their organization way to drive the most value for Generative AI initiatives.
regulatory compliance
30%
difficulty managing risks
29%
lack of a
governance model
Currently, these are considered even more significant
than other critical barriers such as implementation
challenges, a lack of an adoption strategy, and difficulty
identifying use cases.
17
Now: Key findings
The chief operations officer and chief of strategy in a To help build trust and ensure the responsible use for using Generative AI tools and applications (51%),
financial services company summed up the challenge: of Generative AI-powered tools and applications, monitoring regulatory requirements and ensuring
organizations are generally working to establish compliance (49%), and conducting internal audits /
“How do you democratize Generative AI across your
new guardrails, educate their workforces, conduct testing on Generative AI tools and applications (43%)
business while having all of the right controls in place?
assessments, and build oversight capabilities. (figure 9). Despite their importance for effective scaling,
We have an AI board, we have an ethics framework, we
each of these actions is only being taken by less than
have an accountability model. We want to know who’s
Specific actions surveyed organizations are currently
roughly half of the organizations we surveyed.
using it for what, and that it’s being used in the right way.”
taking include establishing a governance framework
Actions to manage risk
51%
49%
43%
37%
35%
33%
30%
23%
19%
Establishing Monitoring regulatory Conducting internal Training practitioners Ensuring a human Keeping a Using a formal Using outside Single executive
a governance requirements and audits and testing how to recognize validates all GenAI- formal inventory group or board to vendors to conduct responsible for
framework for the ensuring compliance of GenAI tools / and mitigate created content of all GenAI advise on GenAI- independent audits managing GenAI-
use of GenAI tools / applications potential risks implementations related risks and testing related risks
applications
Figure 9 Q: What is your organization currently doing to actively manage the risks around your Generative AI implementations?
(May/June 2024 ) N (Total) = 2,770
18
Now: Key findings
78%
of leaders surveyed in Q1 agreed that more
governmental regulation of AI was needed.
Implementing new processes and controls is rarely easy and will likely require active
change management to build support within the organization. “Before launching
anything, we have strict AI governance,” said the chief analytics officer at a professional
services firm. “In the past we had a bit of a siloed approach, but today, at a minimum,
everything has to go through privacy and compliance because we have a methodical
way of managing risk. This is new and challenging to some.”
On top of risk and governance issues, Q3 surveyed organizations were exceedingly
uncertain about the regulatory environment that may exist in the future (depending on
the countries they operate in). In our first quarterly report, 78% of leaders agreed that
more governmental regulation of AI was needed. However, there is a difference between
theory and practice. Organizations are struggling with regulatory uncertainty, and
worries about interpretation and enforcement may be preventing them from pursuing
certain use cases in specific geographies.
The uncertainty around AI regulation may make it feel like there could be many varied
outcomes, but our research suggests most countries are following a similar path
concerning AI policies.2 Governments are working to balance protection, innovation
and economic benefit, so future actions will likely be in line with the regulatory
traditions of each country and region.
1199
Now: Key findings Insights from our executive interviews
How some real-world organizations are dealing with compliance, risk management
Some organizations reported taking action to prepare
and governance issues
for potential regulatory changes. Top areas include
preparing regulatory forecasts or assessments (50%),
An increasing number of organizations are making risk a central factor when selecting Generative AI use
monitoring by the general counsel (48%), and working
cases and investments. However, many are walking a tightrope—trying to minimize risk without being too
with external partners (46%) (figure 10). However, some
risk averse, which could lead to missed opportunities and open the door to competitors.
organizations aren’t doing anything to prepare; 14%
said they aren’t making any specific plans.
Here are some risk-related actions revealed through our in-depth executive interviews:
How organizations are preparing
for regulatory changes Avoiding
Avoid use cases that could require additional regulatory scrutiny
specific tools
and use cases Shut off access to specific Generative AI tools for staff
For organizations that rely heavily on owned intellectual property, be extremely cautious when
Corporate 50% Limitin |
189 | deloitte | us-ima-next-gen-controllership-2025.pdf | Next-gen controllership
Harnessing AI and emerging technologies
to transform finance and accounting
Table of contents
Laying the groundwork: Global research to trace the controllership technology journey �����������������������������1
Section 1: Artificial intelligence: How AI is reshaping accounting and finance ���������������������������������������������������1
Section 2: Beyond AI: Technologies leading change in controllership ��������������������������������������������������������������10
Section 3: From traditional to tech: How emerging technologies are affecting controllership ������������������15
Section 4: How to thrive: A framework for future decision-making �������������������������������������������������������������������19
Conclusion: A way forward ���������������������������������������������������������������������������������������������������������������������������������������������23
End notes ����������������������������������������������������������������������������������������������������������������������������������������������������������������������������24
About the authors ������������������������������������������������������������������������������������������������������������������������������������������������������������25
About the survey ���������������������������������������������������������������������������������������������������������������������������������������������������������������26
Next-gen controllership | Harnessing AI and emerging technologies to transform finance and accounting
Deloitte’s1 Center for Controllership™
Deloitte’s Center for Controllership is a research, resource, and
collaboration center that helps chief accounting officers (CAOs)
corporate controllers, and others in the controllership function�
Deloitte helps organizations effectively navigate business risks
and opportunities—from strategic, reputation, and financial risks
to operational, cyber, and regulatory risks—to gain competitive
advantage. We apply our experience in ongoing business
operations and corporate life cycle events to help clients become
stronger and more resilient. Our market-leading teams help
clients embrace complexity to accelerate performance, disrupt
through innovation, and lead in their industries. For more
information about Deloitte’s Center for Controllership, please visit
www�deloitte�com/us/cfc
IMA® (Institute of Management Accountants)
IMA® is one of the largest and most respected associations
focused exclusively on advancing the management accounting
profession. Globally, IMA supports the profession through
research, the CMA® (Certified Management Accountant) and
CSCA® (Certified in Strategy and Competitive Analysis) programs,
continuing education, networking, and advocacy of the highest
ethical business practices� Twice named Professional Body of the
Year by The Accountant/International Accounting Bulletin, IMA
has a global network of about 140,000 members in 150 countries
and 350 professional and student chapters. Headquartered in
Montvale, N.J., USA, IMA provides localized services through its
four global regions: the Americas, Asia/Pacific, Europe, and Middle
East/India. For more information about IMA, please visit
www�imanet�org�
Next-gen controllership | Harnessing AI and emerging technologies to transform finance and accounting
Laying the groundwork: Global research to trace
the controllership technology journey
Emerging technologies are transforming the finance and This report presents the findings from this survey alongside
accounting industry. With the adoption of artificial intelligence (AI) considerations from industry experts and professionals, offering
and new functionalities available in the convergence of data insights into common emerging technologies used in the
location, process automation, and data analytics technologies, controllership function, the benefits these technologies may have on
financial institutions can now process transactions faster, more core accounting processes, and how technology may transform the
accurately, and with seemingly greater efficiency. However, the function of controllership. We explored how professionals apply new
integration of these new technologies comes with a set of tools, the challenges in adopting technology solutions and insights into
challenges� Legacy systems, which are often outdated and lack the overcoming these challenges, optimizing the positive impacts, and
necessary compatibility with newer technologies, can make the meeting expectations for the future of controllership.
adoption of new technology innovations difficult and expensive.
The insights gleaned from this research form the bedrock of our
Additionally, the implementation of new technology requires
guidance on how controllers and their teams can leverage a
significant investment in training, infrastructure, and cybersecurity
practical framework for navigating the unpredictable landscape of
measures. Despite these challenges, the benefits of emerging
emerging technology. This framework aims to assist finance and
technology in finance and accounting can be promising, and
accounting professionals to optimize the functional value of
companies that integrate these technologies effectively are likely to
technologies that are set to become a staple in the next generation
gain a competitive advantage in the industry.
of controllership�
From the winter of 2023 to the spring of 2024, IMA and Deloitte’s
Center for Controllership conducted a global survey of more than
900 finance and accounting analysts, managers, directors,
controllers, and CFOs. The global survey aimed to read the pulse of
how the finance and accounting functions are navigating the influx of
emerging technology available against expectations for future
implementations, possible applications, and controllership impact�
Section 1: Artificial intelligence: How AI is reshaping
accounting and finance
What is AI? For a term that is consistently in headlines and at the Traditional AI is often rule-based and provides outputs such as
forefront of business discussions, you may find yourself wondering numbers, labels, or classifications. If you have ever used a virtual
what exactly is AI? assistant on a website or leveraged predictive analytics
technologies for your data, you were using traditional AI. This
The standard answer: Artificial intelligence (AI) is the theory and
category of AI is optimized for processing large amounts of data
development of computer systems able to perform tasks normally
following predefined rules that train the AI to respond to a given set
requiring human intelligence. AI can be categorized into two main
of circumstances. It is distinguished by its response within
categories: traditional and generative (figure 1).
prescribed parameters, but it does not adapt to situations outside
its training set�
Generative AI (GenAI) has the ability to generate new content, as
the name suggests. It is AI that can create content across various
modalities, such as text, images, and code, which would have
previously taken human skill and expertise to create.
1
Next-gen controllership | Harnessing AI and emerging technologies to transform finance and accounting
Figure 1: Defining artificial intelligence: Traditional AI vs. Generative AI
AI overview – Artificial intelligence (AI) is the theory and development of
computer systems able to perform tasks normally requiring human intelligence.
Traditional Generative
Traditional AI is Generative AI is artificial
artificial intelligence intelligence applications that
that is often role-based create new content across
and provides outputs various modalities (e.g., text,
such as numbers, images, audio, code, voice,
labels, or classifications video) that would have
without the ability to previously taken human skill
generate new content. and expertise to create.
Traditional Generative
applications applications
Machine learning Text, image,
video generation
Natural language
processing Synthetic data
generation
Virtual assistants
Automated content
Deep learning
moderation or translation
Predictive analytics
Chatbots
Robotics process
Content creation
automation
Automated education
Speech recognition
Image recognition
2
Next-gen controllership | Harnessing AI and emerging technologies to transform finance and accounting
But what does that mean? The current state of AI
Let’s simplify. AI encompasses many technologies that work AI tools and other rule-based innovations are pervasive, but AI is
together to build innovative solutions that transform society and entering a new era. The hype around AI innovation over the past
business alike� year has reached new levels, and for good reason. What changed?
In short, AI is graduating. It is transforming from rule-based
In the finance function, that can include machine learning, natural
traditional models to foundational data and language models that
language processing, deep learning, predictive analytics, robotic
can generate its own rules�
process automation, and speech recognition�
While a rule-based model focuses on predictions and patterns
Why AI matters
using massive amounts of historical data and language models,
The question is not if AI will affect your work, but when. Our global
GenAI can generate content and insights that builds upon
survey showed that the implementation and use of AI in the
foundational data. AI has advanced technological capabilities that
controllership function is expected to nearly double in the next
can empower controllers and transform how business is done�
three to five years. Furthermore, AI was ranked as the second most
With tools from intelligent automation to machine learning, natural
important technology skill for controllers to have training on in the
language processing, and GenAI, organizations are presented with
next three to five years.
both opportunities and risks in finance and accounting.
GenAI captured the public’s imagination when it burst onto the
There are many AI tools available that the accounting and finance
scene in the second half of 2022. Few technologies have ever
function can leverage. When asked which AI products are currently
debuted to such fanfare. Adoption and use of GenAI have been
being used the most in controllership, respondents identified
sudden and rapid among the public. In one example, OpenAI
Microsoft products such as Azure Synapse as the highest used AI
reported reaching 100 million users within 60 days of releasing
tool. Azure Synapse, like most of the tools in our survey, is mostly
ChatGPT to the public�2
used for analytics purposes�
GenAI may be the next great chapter in the history of information.3
This tool was closely followed by OpenAI, with over 30% of
For businesses, the opportunity to augment professionals and
respondents who use AI claiming to use OpenAI in their organizations.
controllers with machine-assisted intelligence is a generational
opportunity. It’s a paradigm shift that may be poised to unlock Rounding out the top three AI tools that respondents mentioned
doors to new business opportunities and fundamentally change using was Snowflake, which has AI capabilities to understand
how the enterprise organizes and operates� unstructured data, answer free-form questions, and provide
intelligent assistance. Other AI tools also used within controllership
include, Domo, Oracle, Sage Intacct, SAP Concur, and ThoughtSpot.
Deloitte’s insights
While the interest in traditional AI and GenAI is reaching
new heights, organizations are adopting AI tools at a lower
rate than many may have expected. Organizations seem to
be waiting for more niche tools to enter the market or
more advanced out-of-the-box technologies to emerge
with practical applications for the finance and
controllership space�
3
Next-gen controllership | Harnessing AI and emerging technologies to transform finance and accounting
GenAI adoption challenges
The top reported challenge for implementing GenAI tools was AI Other challenges noted by respondents included lack of skilled labor,
integration with existing systems, with 19% of survey respondents limited use cases, trust concerns, reliance on bad data, a lack of
citing this has been a challenge with past implementations� This leadership support, and problems with funding (figure 2).
challenge was followed by security concerns, data governance, and
lack of skilled labor for the top implementation challenges� For
future planned AI implementations, integration with existing systems
remained the biggest concern; however, respondents ranked
challenges with data governance and lack of skilled labor higher for
future expected challenges. It is expected to see challenges with
data governance become a significant lift for many organizations as Deloitte’s insights
AI has been receiving much attention in the current climate.
they plan to implement AI (figure 2).
As it introduces a paradigm shift to accelerating
While specific implementation challenges may vary, one common transformation, finance leaders have been more engaged in
barrier is the alignment of system architecture� This relates to the the excitement, likely driving a willingness to fund
noted challenges around data inconsistencies across applications� implementations. However, that excitement may outperform
Inconsistent data governance across the organization leads to the current impact of AI in the finance and accounting space.
challenges in implementing integrated solutions. Therefore, the willingness, or perceived willingness, to fund AI
tools may focus on more long-term or future investments
Lack of funding and lack of leadership support remained the smallest
until the impact aligns with the hype or offers more
challenges for respondents, both for previous and future
assurance for a return on investment.
implementations; however, respondents identified that lack of
funding was becoming more of a concern going forward�
4
Next-gen controllership | Harnessing AI and emerging technologies to transform finance and accounting
Figure 2: Top AI implementation challenges in the past five years
1 2 3
Integration with
Security concerns Data governance
existing systems
4 5 6
Lack of Lack of Lack of trust
skilled labor use case in technology
7 8 9
Lack of
Availability Lack of
leadership
of clean data funding
support
5
Next-gen controllership | Harnessing AI and emerging technologies to transform finance and accounting
AI is an operations force multiplier for human ambitions in finance
Benefits of GenAI in finance
According to respondents, the top reported benefits of AI are Deloitte’s insights
increased automation enablement and the reduction of monotonous Organizations have historically utilized predictive (forecast
responsibilities (figure 3). The global survey showed that 20% more based on historical data) and prescriptive (forecast-driven
recommendations) analytics in more simplistic use cases.
respondents identified predictive and prescriptive analytics
However, professionals have noted GenAI can increase the
as a benefit in the next five years compared to the previous
power of predictive analytics. With GenAI, the model can
three to five years.
offer a prediction with the additional benefit of context and
explanations around that prediction. Natural language
models will make this more accessible� The adoption of
built-in prescriptive analytics into larger offerings will also
likely drive accessibility.
GenAI has the potential to reduce the burden of manually
intensive tasks on humans, freeing them up to focus on higher
value and more ambitious strategies. It is rocket fuel for
operations that can enable a workforce to utilize technologies to
guide decisions and focus more on critical or strategic tasks�4
Figure 3: Top AI benefits in the next five years
1 2 3
Reduce monotonous
Increase automation Ease of data analysis
responsibilities
4 5 6
Reduce Higher Predictive and
human error productivity prescriptive
analytics
6
Next-gen controllership | Harnessing AI and emerging technologies to transform finance and accounting
GenAI use cases in accounting and finance
Deloitte’s insights - Controllership AI use cases5 GenAI’s broad applicability makes it a useful tool across personas
and functions and throughout businesses. For example, using
Increase automation – Automate journal entries; reduce manual
GenAI, the controllership function can systematize recurring
tasks in order-to-cash cycle; process, match, and pay in
entries and reconciliations, perform source-to-target chart of
procure-to-pay process to support touchless invoice processing
account mapping, review and analyze contract terms, and prepare
Reduce monotony – Create, track, and manage close activities; internal and external financial reporting that includes commentary
produce automatic smart accounting reports based on and insights�7
predefined template
Finance leaders can use GenAI to maintain a pulse on the business
Data analysis – Automatically analyze data and provide optional and adapt to changing market conditions, predict and preempt
solutions; improve variance analysis with unstructured data; strategic macroeconomic blockers, enhance organizational
generate insights in video, text, or image format structure, and provide quick answers to evolving questions and
real-time information. Controllers and finance leaders can use
Reduce human error – Reconcile inconsistent journal entries;
GenAI to run intelligent searches of knowledge bases, standard
assess reliability of accounting entries
operating procedures, and regulatory documents; generate control
Productivity – Reduce time spent on manual processes such as compliance reports to provide domain-specific expertise to
risk interviews, in which GenAI can analyze unstructured data business decisions, and monitor compliance, ethics, and control
sources such as discussions to uncover takeaways, themes, and across the business�8
insights; produce diagrams, slides, and other insight material from
With this ability, GenAI could create a more profound relationship
datasets, allowing humans to focus on any identified exceptions
between humans and technology. AI can be a force multiplier or
Predictive and prescriptive analysis – Validate actuals in the assistant for workers—liberating them from more repetitive tasks
close; provide trend analysis and insights for accountants; identify and enabling the workforce to focus on more creative or strategic
biggest drivers of cash flow; analyze historical fulfillment rates for aspects of their jobs�9
inventory management6
7
Next-gen controllership | Harnessing AI and emerging technologies to transform finance and accounting
Implementation considerations
The use of traditional AI and GenAI in accounting and finance may Deloitte’s insights
Many experts identify the record-to-report process for internal or
vary across functions. Respondents identified that advanced
management reporting as one of the strongest use cases for
analytics and intelligent systems, such as data science and AI, are
FP&A. In addition, flux analysis and managing the close process
being implemented the most in financial reporting and financial
could benefit from GenAI implementations.
planning and analysis (FP&A) within controllership. Other areas
leveraging advanced analytics and intelligent systems include
controls and compliance, treasury, general ledger and close
accounting, and operational accounting�
Figure 4: Generative AI adoption in controllership
1–2 years away Currently
from adopting adopting
No plans to
ever adopt
15% 8%
38% 22% 8% 9%
2–5 years away 6–12 months away Currently
from adopting from adopting using
8
Next-gen controllership | Harnessing AI and emerging technologies to transform finance and accounting
While traditional AI tools will likely continue to exponentially Key takeaways
increase in finance and accounting use cases, it is important to
• The ongoing adoption of traditional AI will likely continue to grow
note that GenAI adoption is quickly gaining traction across finance.
as it becomes standard technology in business. Per the survey
Our global survey showed that 16% of respondents are either
results, emerging adoption of GenAI is also likely to increase over
currently using or currently adopting GenAI, and almost half of
the next few years.
respondents (44%) plan to adopt GenAI in the next five years
(figure 4). • When looking at the emerging AI tools and their various
generative applications, the opportunities they present to finance
and accounting are multifaceted. While many tools currently have
analytics applications, GenAI tools are a paradigm shift to the
finance and controllership landscape because of their broad
applicability and convergence with other emerging technologies.
• With the challenges to AI implementations and concerns over
governance and security, stepping into the opportunity to
maximize benefits may be achieved with a successful AI
implementation framework� This is discussed further in Section 4�
9
Next-gen controllership | Harnessing AI and emerging technologies to transform finance and accounting
Section 2: Beyond AI: Technologies leading change
in controllership
The big players: Next-gen tech First, let’s define the key areas in emerging technology outside of AI:
While the novelty of GenAI brings AI to the forefront of many
Data location and management technology
emerging technology discussions, it’s crucial to note that AI is not
Data location and management refers to systems, methodologies,
the only emerging technology taking up real estate in the
and infrastructure used to store, manage, and retrieve data across
next-generation accounting landscape. Other technologies such as
various physical devices and geographical locations. This
process automation, data analytics, and data location continue to
technology encompasses both the hardware and software
evolve and play a big role in accounting and finance. In this section,
components necessary to ensure data is securely saved and
we will identify the most used technologies, implementation trends,
efficiently accessible when needed. In controllership, this
and emerging functionalities for each category according to the
technology can include on-premise, cloud, or data mesh
survey (figure 5).
approaches to storage and data management�
Figure 5: Emerging technology areas in controllership
Common Process Data analytics Artificial
locations of data automation and visualization intelligence (AI)
10
Next-gen controllership | Harnessing AI and emerging technologies to transform finance and accounting
Data analytics and visualization technology Process automation technology
Data analytics tools convert raw data into actionable insights. It Process automation refers to the use of technology to automate
includes a range of tools, technologies, and processes used to find repetitive and manual tasks within a business process. It includes
trends and solve problems by using data. technologies like robotic process automation (RPA), intelligent
document processing (IDP), workflow orchestration, AI, system
Data visualization technologies enable the graphical representation
integrations, and business rules. Its practical applications include
of information and data, often through visual elements like charts,
automating financial processes such as data validation, forecast
graphs, and maps. Its practical application can include the
reports, and reconciliations�
visualization of ad hoc or strategic analysis, compelling
presentations of context underneath typical variance analyses, and
heightened understanding of data to communicate a wide variety
of use cases including daily sales, revenue analytics, variance
decomposition, and growth trends�
Section 2.1: Data location and management survey trends
With most technology implementation initiatives trending upward, Respondents identified the most used data location technology as
it may come as a surprise that results from our survey SAP, with 18% of respondents implementing SAP within the next
demonstrated the implementation of data location technology five years. Other notable data location technologies include SQL,
within accounting and finance is expected to decrease by 32% in and Oracle. While most data location implementations are
the next five years (figure 6). expected to decrease compared to the previous five years, the
survey showed that Amazon Web Services (AWS) will have a 25%
increase in implementations in the next three to five years
compared to current use, the highest increase compared to other
data location technologies (figure 6).
Deloitte’s insights
While this trend deviates from other technologies, there are
some possible explanations for this perception. The
marketplace is moving toward modern ERPs—a wave that
started about five years ago and likely has five years left.
While ERPs typically involve an on-premise data warehouse,
many organizations are moving toward a modern
cloud-based warehouse. Seeking to participate in the cloud
Deloitte’s insights
data warehouse trend, some traditional ERP vendors have
As noted previously, with the emergence of more
created their own offerings as well. Another emerging trend
cloud-based systems and the data mesh trend, the
is the data mesh strategy, in which individual corporate
perception is that finance and accounting professionals may
functions can own their respective data and then publish to a
experience less involvement in the IT side of data
data catalog for consumption in analytics by other functions�
implementation� Some considerations from professionals in
With the move toward cloud-based solutions and emerging the marketplace note that SAP is geared more to larger
data mesh technology, data location implementations may companies and has a stronger footprint in manufacturing
be moving more toward IT ownership. As a result, finance rooted in its strength in ERP integration. Oracle, however,
and accounting leaders may have less visibility or may have a stronger presence in other industries.rooted in
involvement in data location implementations. its strength in ERP integration. Oracle, however, may have a
stronger presence in other industries�
11
Next-gen controllership | Harnessing AI and emerging technologies to transform finance and accounting
Section 2.2: Data analytics and visualization survey trends
The implementation of data analytics and visualization technology
is expected to remain steady, with approximately 24% of Deloitte’s insights
SAC is a native SAP visualization tool, and with an increased
respondents stating they have implemented this type of
interest in SAP S4HANA and Central Finance, SAC will likely see
technology in the past five years and expect to implement this
an uptick in the opportunity for its use. In addition to these
technology in the next three to five years.
traditional reporting and visualization tools identified in the
The survey results found that PowerBI is the most used data survey, we are seeing organizations use desktop analytics
analytics and visualization tool in controllership, with 35% of toolkits to transform, enhance, and improve quality insight and
respondents stating their organization is using PowerBI and 33% data. Other tools offer visualization capabilities as well as tools
planning to implement the tool in the next three to five years to automate business rules, apply criteria, and pull reports�
(figure 6). This is consistent with what Deloitte has seen in the
marketplace . Native and naturally integrated tools have the added AI impact on data analytics and visualization
While natural language generation has been around in some
benefit of ease of use and larger platform integration.
form for many years, the next-gen AI capabilities may offer
Other high-use data analytics tools include Python, SAP Analytics new applications for analytics and data visualization in
Cloud (SAC), and Tableau . We found that the use of SAC is controllership. Organizations will likely see an increase in AI
expected to increase by 28% in the next three to five years when integrations or add-on capabilities with the data analytics
compared to the past five years, the highest increase in change of and visualization tools on the market. There are multiple ways
the analytics technologies� this could present itself with transformative applications.
Notably, GenAI will likely be an innovative tool for producing
prompt-based data and visualization analytics—including
automated or generative language prompts that can produce
new visualizations, stories, and analyses of data.
Figure 6: Most used data analytics and visualization tools in controllership for the next 3–5 years
1 2 3
PowerBI SAP Analytics Cloud Tableau
An interactive data visualization A cloud solution for business intelligence, A data visualization tool
software product developed by enterprise planning, and predictive that allows users to
Microsoft with a primary focus analytics that provides all analytics connect, analyze, and
on business intelligence capabilities for all users in one product visualize any data
Other technologies available in survey question include Alteryx, Python, Qlik, R, and SAS.
12
Next-gen controllership | Harnessing AI and emerging technologies to transform finance and accounting
Section 2.3: Process automation survey trends
The implementation of process automation technology is expected Other common automation tools used in controllership include
to remain steady with over one-fourth of respondents (26%) stating Power Query and Tableau, with 16% and 13% of respondents
that they plan to implement automation technology in the next five currently using these tools in the accounting and finance function.
years (figure 7). This is consistent with automation implementation (figure 7).
trends Deloitte has seen in the past five years. With automation
Automation Anywhere use has the highest expected growth, with
tools becoming increasingly available and user friendly,
an increase by more than 50% over the next three to five years
organizations are reviewing manual processes more frequently to
compared to current use according to the survey.
identify automation opportunities�
Automation Anywhere is utilized to automate transactional
The global survey showed that the most used tool for data
workflows, such as customer service and service management.
preparation and automation is SQL server-enabled automation
tools, with over 22% of respondents stating their accounting and
finance function has used SQL-enabled automation in the past
three to five years. Forward looking, SQL-enabled automation tools
will continue to be the most used tools with 18% of respondents
stating their organization plans to implement SQL in the next five
years (figure 7).
Deloitte’s insights
What we have seen in the marketplace aligns with our view that
RPA and intelligent automation will continue to grow� These
technologies leverage a synthetic keyboard and mouse to
execute business processes.
AI impact on on process automation
AI is having, and will continue to have, a transformative impact
on process automation. The convergence of these two
emerging technology solutions has wide-ranging applications in
Deloitte’s insights
the finance environment.
SQL’s popularity may be due to the broad applicability of
multiple tools that leverage SQL data. SQL acts as a Process automation tools are already starting to leverage AI for
reconciling source to ‘hub’ systems, so businesses that user-generated automations, allowing all users to create
leverage multiple data tools can use SQL to reconcile automation without the need for deep tech knowledge� For
multiple sources of data or match source to target example, a user can leverage AI to automate a process by
data. Other tools also utilize SQL servers, which further specifying various inputs to produce an output.
drives its expansive implementation. For example, SQL can
Another example may be reporting automation—where a user
be used in ERP systems to automate data validation
can generate reports with a prompt or question using GenAI.
processes. It is levered by various applications to perform
Finance and accounting professionals may notice some
automated reconciliations and for financial forecasting and
prominent automation providers already offering AI integrations
data imaging�
with broad use cases. Some of these may look like utilizing AI to
What sets SQL apart from other systems on the market may accelerate the user experience or generating content, context,
also be how well known it is. It is also noted as being very and output to accelerate reviews. Other applications may be
user friendly, explainable, and traceable. SQL offers a creating and handling customer queries |
190 | deloitte | us-ai-institute-generative-ai-agents-multiagent-systems.pdf | Prompting for action
How AI agents are reshaping
the future of work
Expanded capabilities, use cases and
enterprise impact from Generative AI
November 2024
Deloitte AI Institute
Prompting for action | How AI agents are reshaping the future of work
About the Deloitte AI Institute
The Deloitte AI InstituteTM helps organizations connect the different dimensions
of a robust, highly dynamic and rapidly evolving AI ecosystem. The Institute leads
conversations on applied AI innovation across industries, with cutting-edge insights,
to promote human-machine collaboration in the “Age of With.”
The Deloitte AI Institute aims to promote a dialogue and development of artificial
intelligence, stimulate innovation, and examine both challenges to AI implementation
and ways to address them. The Institute collaborates with an ecosystem composed of
academic research groups, startups, entrepreneurs, innovators, mature AI product leaders
and AI visionaries to explore key areas of artificial intelligence including risks, policies,
ethics, future of work and talent, and applied AI use cases. Combined with Deloitte’s deep
knowledge and experience in artificial intelligence applications, the Institute helps make
sense of this complex ecosystem, and as a result delivers impactful perspectives to help
organizations succeed by making informed AI decisions.
No matter what stage of the AI journey you’re in, whether you’re a board member or a
C-suite leader driving strategy for your organization or a hands-on data scientist bringing
an AI strategy to life, the Institute can help you learn more about how organizations across
the world are leveraging AI for a competitive advantage. Visit us at the Deloitte AI Institute
to access the full body of our work, subscribe to our podcasts and newsletter, and join us
at our meetups and live events. Let’s explore the future of AI together.
www.deloitte.com/us/AIInstitute
2
Prompting for action | How AI agents are reshaping the future of work
Content
Key takeaways
• AI agents are reshaping industries by expanding the potential applications
of Generative AI (GenAI) and typical language models.
• Multiagent AI systems can significantly enhance the quality of outputs
and complexity of work performed by single AI agents.
• Forward-thinking businesses and governments are already implementing
AI agents and multiagent AI systems across a range of use cases.
• Executive leaders should make moves now to prepare for and embrace
this next era of intelligent organizational transformation.
Introduction 4
AI agents: 5
What makes them different—and why they matter
Multiagent AI systems: 7
Amplifying the potential of AI agents
Key benefits of AI agents and multiagent AI systems: 7
Advantages that AI agents are unlocking for organizations today
Transforming strategic insights: 8
A real-world example of a multiagent AI system
Achieving impact through targeted use cases: 11
How AI agents are changing industries and enterprise domains
Enabling new ways of working and new horizons of innovation: 13
Implications for strategy, risk, talent, business processes and technology
The road ahead: 15
What we expect as AI agents continue to evolve
Charting a course into the next era of organizational transformation: 16
Recommended actions for leaders to take now
Get in touch & Endnotes 17
3
Prompting for action | How AI agents are reshaping the future of work
Introduction
How can we operate faster and
more efficiently?
This question has always been at the forefront of strategic
agendas—but Generative AI (GenAI) is helping unlock new
answers. With its ability to produce novel outputs from plain-
language prompts, GenAI has enabled enterprises to significantly
enhance speed and productivity across a range of business tasks.
However, use cases for typical language models have only just
begun to show GenAI’s transformative potential. In this time of
rapid AI evolution, it’s time to think bigger and bolder: from
streamlining routine tasks to redesigning entire workflows.
Now the question for business and government leaders
is becoming:
How can we rethink our business
processes with GenAI?
Large language models (LLMs) and GenAI-powered tools used
by most organizations today serve as helpful assistants: A human
worker enters a prompt, GenAI quickly produces an output.
However, this interaction is largely transactional and limited
in scope.
What if GenAI could be more like a skilled collaborator that will
not only respond to requests but also plan the whole process
to help solve a complex need? What if GenAI could also tap into
the necessary data, digital tools and contextual knowledge to
orchestrate the process end to end, autonomously?
This vision is becoming a reality with the emergence of AI agents
and multiagent AI systems—a powerful advancement in what’s
possible through human-AI partnership. Leading companies and
government agencies are already seeing the value of AI agents
and putting them into practice.
Adapt or fall behind
In this paper, we explore what makes AI agents so groundbreaking.
We then reveal how they are reshaping industries, including At the end of 2023, nearly 1 in 6
government and public services, by enabling new use cases,
surveyed business leaders said
enhancing automation and accelerating the future of intelligent
organizational transformation. GenAI had already transformed
their businesses.1
4
Prompting for action | How AI agents are reshaping the future of work
AI agents: What makes them
different—and why they matter
To grasp the potential value of AI agents and their role in As a result, early GenAI use cases have mostly been limited to
expanding the automation horizon, it is important to understand standalone applications such as generating personalized ads
how they differ from the language models and GenAI applications based on a customer’s search history, reviewing contracts and
familiar to business leaders today. legal documents to identify potential regulatory concerns,
or predicting molecular behavior and drug interactions in
AI agents are reasoning engines that can pharmaceutical research.
understand context, plan workflows,
AI agents excel in addressing these limitations while also
connect to external tools and data, and leveraging capabilities of domain- and task-specific digital tools
to complete more complicated tasks effectively. For example,
execute actions to achieve a defined goal.
AI agents equipped with long-term memory can remember
customer and constituent interactions—including emails, chat
While this may sound broadly like what standalone LLMs or sessions and phone calls—across digital channels, continuously
GenAI applications can do, there are key distinctions that learning and adjusting personalized recommendations. This
make AI agents significantly more powerful. (See table, page 6.) contrasts with typical LLMs and SLMs, which are often limited to
Typical LLM-powered chatbots, for example, usually have limited session-specific information. Moreover, AI agents can automate
ability to understand multistep prompts—much less to plan and end-to-end processes, particularly those requiring sophisticated
execute whole workflows from a single prompt. In essence, they reasoning, planning and execution.
conform to the “input-output” paradigm of traditional applications
and can get confused when presented with a request that must AI agents are opening new possibilities to drive enterprise
be deconstructed into multiple smaller tasks. They also struggle to productivity and program delivery through business process
reason over sequences, such as compositional tasks that require automation. Use cases that were once thought too complicated
consideration of temporal and textual contexts. These limitations for GenAI can now be enabled at scale—securely and efficiently.
are even more pronounced when using small language models
(SLMs), which, because they are trained on smaller volumes of In other words: AI agents don’t just interact. They more
data, typically sacrifice depth of knowledge and/or quality of effectively reason and act on behalf of the user.
outputs in favor of improved computational cost and speed.
5
Prompting for action | How AI agents are reshaping the future of work
A new paradigm for
human-machine collaboration
Through their ability to reason, plan, remember and act,
AI agents address key limitations of typical language models.
Typical language models AI agents
Use case Automate tasks Automate entire workflows/processes
scope
Planning Are not capable of planning or Create and execute multistep plans to achieve
orchestrating workflows a user’s goal, adjusting actions based on
real-time feedback
Memory & Do not retain memory and have limited Utilize short-term and long-term memory to
fine-tuning fine-tuning capabilities learn from previous user interactions and
provide personalized responses; Memory may
be shared across multiple agents in a system
Tool Are not inherently designed to integrate with Augment inherent language model capabilities
integration external tools or systems with APIs and tools (e.g., data extractors, image
selectors, search APIs) to perform tasks
Data Rely on static knowledge with fixed training Adjust dynamically to new information and
integration cutoff dates real-time knowledge sources
Accuracy Typically lack self-assessment capabilities and Can leverage task-specific capabilities, knowledge
are limited to probabilistic reasoning based on and memory to validate and improve their own
training data outputs and those of other agents in a system
6
Prompting for action | How AI agents are reshaping the future of work
Multiagent AI systems:
Amplifying the potential of AI agents
While individual AI agents can offer valuable enhancements, the
truly transformative power of AI agents comes when they work
together with other agents. Such multiagent systems leverage Key benefits of AI agents
specialized roles, enabling organizations to automate and optimize and multiagent AI systems
processes that individual agents might struggle to handle alone.
Capability—AI agents can automate interactions with
Multiagent AI systems employ
multiple tools to perform tasks that standalone language
multiple, role-specific AI agents to models were not designed to achieve (e.g., browsing a
website, quantitative calculations).
understand requests, plan workflows,
coordinate role-specific agents, Productivity—Whereas standalone LLMs require constant
human input and interaction to achieve desired outcomes,
streamline actions, collaborate with
AI agents can plan and collaborate to execute complex
humans and validate outputs. workflows based on a single prompt—significantly speeding
the path to delivery.
Multiagent AI systems typically involve standard-task agents
Self-learning—By tapping short- and long-term contextual
(e.g., user interface and data management agents) working with
memory resources that are often unavailable in a pre-trained
specialized-skill and -tool agents (e.g., data extractor or
language model, AI agents can rapidly improve their output
image interpreter agents) to achieve a goal specified by a user.
quality over time.
At the core of every AI agent is a language model that provides
Adaptability—As needs change, AI agents can reason
a semantic understanding of language and context—but
and plan new approaches, rapidly reference new and
depending on the use case, the same or different language models
real-time data sources, and engage with other agents to
may be used by agents in a system. This approach can allow some
coordinate and execute outputs.
agents to share knowledge while others validate outputs across
the system—improving quality and consistency in the process.
Accuracy—A key advantage of multiagent AI systems is
That potential is further enhanced by providing agents with shared
the ability to employ “validator” agents that interact with
short- and long-term memory resources that reduce the
“creator” agents to test and improve quality and reliability
need for human prompting in the planning, validation and iteration
as part of an automated workflow.
stages of a given project or use case.
Intelligence—When agents specializing in specific tasks
This concept extends what’s possible with individual AI agents
work together—each applying its own memory while utilizing
by taking a team or agency approach. By decomposing a detailed
its own tools and reasoning capabilities—new levels of
process into multiple tasks, assigning tasks to agents optimized
machine-powered intelligence are made possible.
to perform the tasks, and orchestrating agent and human
collaboration at each stage of the workflow, this type of system
Transparency—Multiagent AI systems enhance the ability
has proven much more likely to produce higher quality, faster
to explain AI outputs by showcasing how agents communicate
and more trustworthy outcomes.2, 3
and reason together, providing a clearer view of the collective
decision-making and consensus-building process.
In other words: Multiagent AI systems don’t just reason
and act on behalf of the user. They can orchestrate complex
workflows in a matter of minutes.
7
Prompting for action | How AI agents are reshaping the future of work
Transforming strategic insights
No matter the industry, every organization engages in research, analysis and reporting—whether about
economic conditions, customer and constituent preferences, policy and pricing strategies, or other topics.
Traditionally, these projects require skilled human analysts to perform multiple steps, which can be
time-consuming, utilizing research and analysis tools along with in-house subject matter expertise.
Here’s what a traditional research project typically looks like.
Analyst Analyst identifies topic and Analyst Analyst selects sources, Analyst
scope: A report on the top 5 searches and compiles
Analyst synthesizes themes
GenAI trends in financial services, relevant information, and
and perspectives, outlines a
based on publicly available data organizes materials and notes.
plan for the report and sends to
from the prior 3 months.
business stakeholder for review.
Stakeholder Analyst drafts the report Analyst Stakeholder provides Stakeholder
and sends to stakeholder, feedback on outline.
who provides feedback
and iterates with analyst. Analyst or designer researches images,
Analyst sends approved develops graphics and designs report.
report to designer.
Analyst Proofer reviews report and Proofer Risk & compliance
or Designer provides feedback, which analyst
and/or designer incorporate.
Risk & compliance professionals
are engaged as needed.
Final report
is delivered.
While effective and repeatable, this approach is …
Time-consuming Inefficient Difficult to scale
Completing a single report can take Skilled analysts must perform many Companies and government agencies
days or weeks, making it difficult to repetitive activities that take their can struggle to hire and retain enough
seize emerging opportunities. focus away from higher-level analysis. skilled, experienced analysts to grow
their research capacity.
8
Prompting for action | How AI agents are reshaping the future of work
Deloitte has developed a multiagent AI system that can streamline
AI AGENT TYPES
and improve each step of research and reporting. Here’s how it works.
“I need to write a report about “Please tell me about
GenAI trends in my industry.” your request ....” Standard-task Specialized-skill
agent(s) & -tool agents
One or more agents Role-specific agents
that perform tasks that execute specific
common to all tasks within the
workflows workflow
Analyst Analyst and interface agent discuss and define report User
scope, sources and timeframe for data collection, target industry interface All agents can access …
and audience, etc. Through this process, the analyst defines the • Language models (shared or separate)
deliverable: A report on the top 5 GenAI trends in financial services,
based on publicly available data from the prior 3 months. • External tools & data sources as needed
• Shared short- and long-term memory
Planning agent breaks the goal into
subprocesses, develops a workflow and
identifies necessary tools and specialized
agents to execute the workflow.
File Multimodal Planning
management processing
Prompt Data Web Content Topic Report
expanding sourcing browsing summarization modeling writing Analyst reviews the report and
requests changes. The system
iterates and refines the report.
Report Data Data Image Quality
formatting structuring visualizing selection assurance
Analyst
Specialized agents expand prompts, conduct research, compile and analyze results, identify themes
and draft the report outline. As needed, the multimodal processing agent translates and interprets data
collected from visual and audio sources. Once the outline is approved/adjusted by the analyst, additional
specialized agents draft and design the report complete with customized charts and illustrations.
Throughout the process, the quality assurance agent checks for accuracy, quality and regulatory/brand
compliance, while the data management agent ensures source materials and report iterations are Final report
documented for reference/review. is delivered.
In addition to being effective and repeatable, this AI agent-powered approach is …
Fast Efficient Highly scalable
A single, quality report can be Skilled professionals can focus on In essence, this system provides an instantly
produced in less than an hour. validating, iterating and refining the report. available team of skilled digital workers.
9
Prompting for action | How AI agents are reshaping the future of work
Effective and efficient work
depends on creativity and
knowledge augmented by
well-planned processes and
task-appropriate tools.
That’s what AI agents and
multiagent AI systems can
bring together.
10
Prompting for action | How AI agents are reshaping the future of work
Achieving impact through
targeted use cases
Organizations across industries and sectors are already leveraging the potential of AI agents and multiagent
systems to transform processes, improve efficiency, and expand impact. Let’s explore four use cases
that are possible today—two in specific industries, and two that can be applied in any business.
1 USE CASE 2 USE CASE
Individualized financial advisory Dynamic pricing and
and wealth management personalized promotions
INDUSTRY: Financial services INDUSTRY: Consumer
Financial advisory services often have relied on broad Standard pricing strategies often involve static models that do
categorizations of customers based on age, income and risk not account for real-time market conditions, customer behavior
tolerance. This approach can often miss the complexities of or inventory levels. Multiagent AI systems can rapidly integrate
individual financial situations and goals. In today’s rapidly changing analysis based on vast amounts of real-time data—such as
financial landscape, there is an increasing demand for personalized, competitor pricing, customer purchase history and seasonal
adaptive financial advice. Multiagent AI systems can analyze diverse trends—to dynamically adjust prices. Additionally, they can
data sources—including the customer’s financial history, real-time personalize promotions based on individual customer
market data, life events and even behavioral patterns—to help preferences, attributes and shopping habits with the goal of
advisers create financial plans and investment strategies tailored improving conversion rates and elevating customer satisfaction.
for the specific individual. AI agents can then continuously monitor
and adjust recommendations as circumstances change. POTENTIAL ADVANTAGES ACHIEVED WITH AI AGENTS:
POTENTIAL ADVANTAGES ACHIEVED WITH AI AGENTS:
Faster adaptation
Adjust prices instantly in response to
Hyperpersonalization market changes, inventory levels or
Customize financial advice to each customer’s customer demand—optimizing revenue.
specific needs and goals, considering factors
Personalized offers
that other methods might overlook.
Tailor promotions to each customer’s
Continuous fine-tuning preferences and behavior, increasing the
Automatically update financial plans and likelihood of purchase.
strategies in response to changes in market
Greater profitability
conditions or personal circumstances.
Maximize margins and minimize discounting
Improved customer satisfaction by optimizing pricing and promotions on an
Strengthen customer relationships by ongoing basis.
providing more relevant and timely advice,
leading to higher retention and satisfaction.
Enhanced scalability
Serve a larger number of customers with
high-quality, personalized advice without
raising costs to deliver.
11
Prompting for action | How AI agents are reshaping the future of work
3 USE CASE 4 USE CASE
Talent acquisition and recruitment Personalized customer support
DOMAIN: Human resources (HR) DOMAIN: Customer and beneficiary service
Traditional recruitment processes often involve manual resume Traditional customer and beneficiary support systems often rely on
screening, repetitive candidate assessments and significant scripted interactions, which can fail to resolve complex or unique
administrative work—which can lead to inefficiencies. AI agents inquiries—leading to customer frustration and escalation.
can automate the end-to-end recruitment process by using natural In contrast, multiagent AI systems can understand plain-language
language processing to analyze resumes, assess candidates based requests and generate relevant and natural responses that
on skills and experience, and conduct initial screening interviews consider the customer’s history, preferences and real-time context.
via GenAI-powered avatars. These systems can collaborate with These advanced systems can handle many complex inquiries
HR professionals to ensure that qualified candidates are identified, effectively—reducing the need for escalation to live agents while
prioritized and moved through the hiring pipeline efficiently while improving customer/beneficiary satisfaction.
adhering to relevant regulations.
POTENTIAL ADVANTAGES ACHIEVED WITH AI AGENTS:
POTENTIAL ADVANTAGES ACHIEVED WITH AI AGENTS:
Greater consistency and scalability
Increased efficiency AI agents can operate 24/7 without fatigue,
Automate tasks to allow HR teams maintaining a consistent quality of service
to focus on strategic activities, shortening no matter the volume of inquiries.
the time to hire.
Improved customer experiences
Improved candidate matching Each customer interaction can be adjusted
Analyze a broader range of data points to help to individual needs, improving satisfaction
match candidates to roles more accurately, and engagement.
improving the quality of hires.
Compounding efficiencies
Reduced bias The ability to learn from each interaction can
By standardizing candidate assessments and help reduce response times, improve quality,
focusing on skills and experience, AI agents can help and free up human service agents to focus on
address unconscious bias in the recruitment process. more nuanced customer requests.
Dynamic scalability
Handle large volumes of applications, making it
easier to manage hiring campaigns or recruit for
multiple roles simultaneously.
12
Prompting for action | How AI agents are reshaping the future of work
Enabling new ways of working
and new horizons of innovation
As language models continue to evolve, AI agents and systems are Risk implications
likely to become strategic resources and efficiency drivers for core
business and government activities such as product development, AI agents introduce new risks that necessitate robust security
regulatory compliance, customer service, constituent engagement, and governance structures. A significant risk is potential bias in AI
organizational design and others. We see a future in which algorithms and training data, which can lead to inequitable decisions.
agents will transform foundational business models and Additionally, AI agents can be vulnerable to data breaches and
entire industries, enabling new ways of working, operating cyberattacks, compromising sensitive information and data integrity.
and delivering value. The complexity of AI systems also presents the risk of unintended
consequences due to AI agents behaving unpredictably or making
That’s why it’s important for C-suite and public service leaders decisions not aligned with organizational goals.
to begin preparing now for this next chapter in the evolution of
human-machine collaboration and business innovation. To manage these risks, it is important to set clear parameters for
agent interactions, monitor operational metrics, and continually
Let’s explore some of the new ways of thinking and leading that ensure data ethics, privacy, security and integrity. As AI agents
should be considered during this time of rapid change. are integrated into core business processes, an enterprisewide
governance framework with guidelines on data usage, ethics and
security can further help mitigate risks. This framework should
Strategy implications ensure compliance with relevant regulations and include continuous
monitoring of AI agent interactions. Advanced security measures,
Leaders should begin integrating AI agents and multiagent
such as encryption and multifactor authentication, can help protect
AI systems into their overall strategies and future road maps.
against data breaches and cyberattacks. Training and awareness
This involves reimagining business processes, investing in AI
programs for employees can provide an additional defense
capabilities, and fostering cultures of innovation. Organizations
by helping employees understand the ethical and operational
should develop their own clear road map for AI agent adoption,
considerations of working with AI agents.
identifying key areas where they can drive the most value and
impact on broader business goals.
Effective change management will be crucial for successful FOCUS AREAS
integration. Leaders should think carefully through how they will
• Identify brand and operational risks that may arise around
address organizational resistance, provide training, and ensure
data usage, AI agent interactions with each other and with
that employees understand the value and benefits of AI agents.
tools, and ethics.
This includes developing a comprehensive communication strategy
to keep employees and other stakeholders informed and engaged
• Ensure model outputs are effectively tested and validated.
throughout the adoption process.
• Implement an AI agent governance framework that is
regularly reviewed and updated as AI technologies evolve.
FOCUS AREAS • Monitor emerging risks specific to AI agents such as “agent
autonomy”—i.e., the risk of unintended consequences when
• Identify and prioritize business and service areas agents make decisions with minimal human oversight.
where AI agents can have the most immediate and
measurable impact.
• Develop robust training programs to help employees
understand and use AI agents in ways that improve
productivity and efficiency.
13
Prompting for action | How AI agents are reshaping the future of work
Talent implications Business process implications
The implementation of AI agents is likely to change the traditional AI agents and multiagent AI systems demand careful human
workforce structure. As AI agents take over routine and lower-value evaluation of business processes—sometimes from the
tasks, there will likely be a high demand for human skills related to ground up. While agents will redefine many core processes over
designing, implementing and operating these systems. Leaders time, AI agents can be integrated into existing operating models
should think through what new roles, job descriptions and job today, enhancing the efficiency of current processes without the
architectures are involved in building out the capability and then need for complete system overhauls. This approach makes it
how to identify, recruit, train and retain this specialized talent. easier for organizations to adopt lower-risk agent solutions
incrementally—but requires careful planning, management
Beyond the implications for tech talent, enterprise leaders should and alignment to ensure that AI agents are improving what
be ready to help employees across a wide variety of roles learn people and/or other technology solutions already do well.
how to work with AI agents and even identify new use cases
where they could improve processes. Deployed and managed In use cases where AI agents do make sense to implement,
well, AI agents can open up new realms of potential for human- human involvement will remain vitally important for tasks
machine collaboration—but that potential depends on workers requiring judgment, validation and critical decision-making.
understanding, embracing and being able to perform new roles. This collaboration is important to help ensure that AI outputs
are accurate, reliable and effective. In this paradigm, everyone
working with AI agents serves as a manager—giving orders (via
prompts), clarifying requests, monitoring progress, reviewing
FOCUS AREAS
outputs and requesting or making changes as necessary.
• Communicate the benefits of AI agents, and help
employees adapt to new ways of working.
• Foster a culture of innovation and continuous learning. FOCUS AREAS
Leaders should instill a mindset of innovation and
• Ensure that where agents are implemented into existing
adaptability related to AI agents.
business processes, those processes remain effective
• Explore a redesign of job architectures, workflows while driving greater efficiency and value.
and performance metrics to reflect the new reality
• Establish processes for continuously monitoring and
of humans and AI agents working in tandem.
improving the performance of AI agents. This includes
collecting and analyzing data on the performance of
AI agents, identifying opportunities for improvement, and
making changes as needed to optimize their performance.
14
Prompting for action | How AI agents are reshaping the future of work
Technology and data implications
Implementing AI agents can be costly, requiring substantial The road ahead
investment in technology and infrastructure. Organizations
should carefully evaluate the value proposition and return on The era of AI agent collaboration is still in its early stages.
investment; and develop a phased approach to use cases, with Interest is growing among businesses and technology
a focus on “low-hanging fruit” (i.e., simpler use cases) that can providers, but comprehensive solutions are not yet common.
lay the groundwork for more complex activations. There is much technical work to be done—particularly in
terms of the reasoning and planning capabilities that will
Quality data is the foundation for AI agents to work effectively. enable AI agents.
If data is inaccurate, incomplete or inconsistent, the agents’
outputs and actions may be unreliable or incorrect—creating Improvements are likely to come fast. In recent months
both adoption and risk issues. It’s therefore essential to invest GenAI tools have shown significant improvements in
in robust data management and knowledge modeling. reasoning and agent orchestration capabilities. Many venture
capital firms are investing heavily across the spectrum of AI
Adopting trustworthy AI practices is a key to mitigating risks agent-related technologies, as are many of today’s leading
and ensuring ethical deployment. This includes developing GenAI and technology providers. What is available today is
AI agent solutions that are fair, transparent and accountable, only a glimpse of what’s to come. Indeed, we anticipate a
and addressing potential biases in AI models. significant evolution of core language models, AI agents, and
agent orchestration platforms within the next 12 months.
Future-focused leaders aren’t waiting on the sidelines.
FOCUS AREAS
Across industries, companies are already designing, testing
• Put the right technology infrastructure in place to and—in some cases—implementing agents.
support the adoption and implementation of AI agents
(e.g., AI orchestration platforms and scalable data lakes).
• Ensure data is properly organized, up to date and
accessible to AI agents. This includes having well-defined
data governance policies and procedures as well as
continuous access to real-time data feeds to enable
dynamic, accurate decisions.
• Establish processes for monitoring and managing the
performance and ethics of AI agents and multiagent
AI systems. Without transparent and trustworthy AI,
customer trust and regulatory compliance are at risk.
15
Prompting for action | How AI agents are reshaping the future of work
Charting a co |
191 | deloitte | us-ai-institute-teleco.pdf | AI-powered
Communications Service
Providers
Reinvent the future of enterprise operations and
customer care
Deloitte AI InstituteTM
AAII--ppoowweerreedd CCoommmmuunniiccaattiioonnss SSeerrvviiccee PPrroovviiddeerrss
About the Deloitte AI InstituteTM
The Deloitte AI Institute helps organizations Deloitte’s deep knowledge and experience in AI
connect the different dimensions of a robust, highly applications, the Institute helps make sense of this
dynamic and rapidly evolving artificial intelligence complex ecosystem and as a result provides impactful
(AI) ecosystem. The AI Institute leads conversations perspectives to help organizations succeed by making
on applied AI innovation across industries, with informed AI decisions.
cutting-edge insights, to promote human-machine
collaboration in the “Age of WithTM”. No matter what stage of the AI journey you’re in—
whether you're a board member or a C-Suite leader
The Deloitte AI Institute aims to promote a driving strategy for your organization, or a hands
dialogue and development of artificial intelligence, on data scientist, bringing an AI strategy to life—the
stimulate innovation, and examine challenges to AI Deloitte AI institute can help you learn more about
implementation and ways to address them. The AI how enterprises across the world are leveraging AI
Institute collaborates with an ecosystem composed of for a competitive advantage. Visit us at the Deloitte AI
academic research groups, start-ups, entrepreneurs, Institute for a full body of our work, subscribe to our
innovators, mature AI product leaders, and AI podcasts and newsletter, and join us at our meet ups
visionaries, to explore key areas of the technology and live events. Let’s explore the future of AI together.
including risks, policies, ethics, future of work and
talent, and applied AI use cases. Combined with www.deloitte.com/us/AIInstitute
22
AI-powered Communications Service Providers
Communication Service Providers (CSPs)
can leverage AI-enabled solutions to
build a competitive advantage through
operational efficiencies and enhanced
customer experiences.
For CSPs, the need to optimize The business benefits of artificial transformation of business functions
and automate processes is intelligence (AI) are becoming more and moving from a traditional to an
increasing rapidly, especially apparent each day. With AI—and augmented workforce. Way too
given the challenge of managing especially Generative AI—CSPs can often companies hastily pursue
and servicing complex networks automate mundane tasks through individual, disjointed AI and
while still delivering a seamless prediction and decision-making, Generative AI use cases that
and personalized customer derive insights that improve customer eventually waste resources and
experience. Automating processes, experiences, and generate data prove ineffective in achieving the
improving business insights, and that can be used to train machine goals of the business. CSP leaders
creating value require flexible learning models and simulate should first develop a sound AI
strategies, a creative approach and real-world scenarios. strategy for transforming their
data-driven methodologies that can operations in alignment with
prescribe custom solutions at the Capitalizing on this value is no their business objectives.
right place and at the right time. small feat—achieving AI-enabled
operations involves end-to-end
3
AI-powered Communications Service Providers
What is AI-enabled
CSP operations?
AI-enabled CSP operations embed AI solutions to optimize and
automate field service and network operations, and enhance customer
service. By automating complex tasks, improving decision-making,
optimizing network strategy and performance, and enabling
personalized customer experiences, CSPs can:
Improve Reduce Increase Generate
efficiency costs customer loyalty new revenue
44
AAII--ppoowweerreedd CCoommmmuunniiccaattiioonnss SSeerrvviiccee PPrroovviiddeerrss
Why is Generative AI a game-changer?
Here are a few examples where CSPs Network ops and maintenance
Generative AI is a subset of artificial
can utilize AI and Generative AI to Identify network faults through digital
intelligence in which machines create
drive value in the enterprise: twins and provide prompt-based
new content in the form of text, code,
remediation solutions for on-field
voice, images, videos, and processes.
Personalized customer
technicians for faster resolution.
Generative AI has massively expanded
self-service
the scope of what value AI can Advanced decision making
Enable on-the-fly customer support
bring because it can generate net in the field
based on local language and propose
new content based on past data Augment technicians with advanced
new product/service recommendatio
and patterns, as well as assist in problem-solving to provide them with
ns, generating offers to increase
formulating new solutions. Generative additional solutions they might
satisfaction and retention.
AI also allows users to retrieve not have thought of or to test new
Network planning
data across complex and siloed solutions prior to implementation.
and deployment
data sources to gain insights with
Create simulations of service quality
ease. Other AI techniques such as Amidst an ever-changing
and subscriber experience at
supervised and unsupervised learning business and technology
potential deployment sites based on
models can be used to predict landscape, CSPs who effectively
user behavior, environmental and
outcomes and uncover anomalies. embed Generative AI across
historical usage data.
Together with Generative AI, they their operations may be better
represent a significant opportunity Network stress testing able to adapt quickly and remain
for CSPs to reinvent their operations. Generate scenarios to simulate future competitive. Here's more on what
network consumption patterns AI can do for customer service,
to stress test load and provision field services, and network
resources given specific bandwidth operations.
and network constraints.
5
AI-powered Communications Service Providers
Elevating customer care
Technology innovations are pushing CSPs to reimagine how they
deliver customer care experiences. The low costs of switching from one
provider to another and increased competition require companies to
build loyalty through delivering a differentiated customer experience.
Next best action/Next best offer. Proposes
Customer experiences can now be improved
new product/service recommendations and
through next-generation conversational
sales offers based on customer preferences to
AI, which can create unique, personalized
increase satisfaction and retention.
encounters that are brought to life by
avatars and Generative AI models. Imagine Smart routing. Quickly connects customers
empathetic, multilingual chatbots instantly to the best live agents based on specific needs
solving technical issues, providing personalized when issues are beyond what the system
recommendations, and understanding can do.
the nuances of every customer question.
Additionally, predictive models can anticipate For agents
problems before they arise, proactively solving Generative AI can support customer service
the issue to improve their experience. Need associates by generating suggested scripts,
to speak to an agent? They are better armed solutions, and recommendations based on
with chatbots that can summarize complex current customer issues and past interactions
information at a glance. This seamless, AI- and trends. Agent performance can be
driven approach boosts customer satisfaction improved by creating a one-stop-shop hub
and fosters loyalty through personalized and where agents have fast access to information
delightful experiences. and tools, such as:
Agent assist. Enables agents to collaborate
For customers
with Generative AI virtual assistants trained on
Generative AI can improve outcomes through
internal data in real-time via voice and text to
direct customer interactions and enable
help resolve customer inquiries more efficiently.
customer service associates to be more
Issue summarization. Automatically
efficient and customer centric. Generative AI
summarizes, tags, and logs customer history
virtual assistants can enhance the self-service
for agents' future reference.
experience and effectiveness by interpreting
customer intent and available data via: Agent training. Evaluates customer sentiment,
resolution steps, and KPIs to initiate coaching
Knowledge search. Quickly interprets context
and training.
and searches for relevant information from
complex and siloed customer history and
knowledge management systems to deliver
tailored responses in real-time.
6
AAII--ppoowweerreedd CCoommmmuunniiccaattiioonnss SSeerrvviiccee PPrroovviiddeerrss
Demonstrated customer care benefits
AI can help CSPs
improve customer
experiences with
Increased Reduced Labor Increased Better new- purposeful human-to-
personalization operating costs effectiveness compliance customer machine interactions
with more calls with live agents acquisition and conversations can
deflected to handling more and improved be streamlined and
virtual agents complex issues retention personalized.
77
AI-powered Communications Service Providers
Transforming field service
and logistics
A heavier reliance on at-home networks and increased customer
expectations of network reliability have exacerbated the challenges
for field service and logistics teams. The added strain can lead
to inefficient operations, higher costs, and a less-than-ideal
customer experience.
To address these challenges, CSPs can leverage For technicians
AI across the field service value chain. AI tools Once technicians are on site, Generative AI
can help CSPs optimally source, manage, can serve as a copilot. The technology
and pick/pack inventory that field service enables them to more efficiently complete
technicians bring to job sites. Logistics teams jobs, improving technicians’ experience and
can also leverage optimization algorithms to productivity while bettering the customer
efficiently equip and schedule technicians experience. For example:
based on skillsets in addition to dynamically
Issue resolution. Leverage Generative AI
routing them based on unpredictable
to power a search for personalized solutions
constraints such as schedule changes and
to customer network issues and accelerate
weather disruptions.
troubleshooting. Retrieval augmented
generation and summarization from internal
For logistics managers
databases and customer chat history can
Generative AI can support logistics managers
generate the recommended resolution steps
and help them make more informed, proactive
and explanations for network engineers.1
decisions. Look into solutions such as:
Vision AI for image and video analysis can
Field technician dispatch. Optimize your fleet further enable technicians to diagnose issues
productivity with faster dispatch and dynamic through visual inspection.
routing, significantly reducing the organization's
carbon footprint. The advanced scheduling
can route field technicians based on skillsets,
equipment required, and availability.
Predictive maintenance. Avoid untimely
vehicle and mechanical warehouse
breakdowns, as machines alert managers
of potential problems before they happen.
8
AI-powered Communications Service Providers
Demonstrated field service and logistics benefits
These improvements
can not only result in
reduced costs, but also
Increased Reduced Decreased Lower costs Decreased Increased new revenue through
field service time to dispatch for gas and time to train customer an enhanced brand
technician resolution processing maintenance and onboard retention and greater customer
productivity for jobs time technicians loyalty by serving the
customer more quickly
and effectively.
99
AI-powered Communications Service Providers
Optimizing network operations
Network downtime and service degradations cost CSPs tens of
billions of dollars in losses per year.2 The ability to move from reactive
to proactive asset maintenance can deliver significant savings.
Network Operations start to look different when enabled with AI
technologies. There aren’t as many fire drills to answer.
Many CSPs are now enabling their Network Network operations optimization. Move
Operations Centers (NOCs) with AI. These centers from reactive to proactive automation of Tier-1
serve as the nerve center of telco networks and operations that can deliver significant cost savings
can enable downstream AI use cases that can while minimizing network downtime and service
drive value for CSPs. NOCs of the Future will rely degradations. Generative AI can help technicians
less on human wherewithal and manual processes correlate alarms with meaningful insights and
by implementing AI-enabled solutions that bring take action by automating resolution steps.3
greater agility, precision, and proactiveness.
Predictive surveillance. Predict potential
Building AI-enabled NOCs are not only more
future issues before they occur and monitor
efficient, but also more resilient, scalable, and
network equipment in-real time in order to detect
future-ready.
anomalies that might indicate potential issues.
Leveraging AI helps maintain the physical layer
From an operational efficiency perspective, this
of the network, and proactively avoids sudden
includes leveraging historical maintenance and
equipment malfunction and expense truck rolls.
fault data to train predictive and network-specific
Automated data enrichment. Dramatically
large language models that can evaluate real-time
improve ticket resolution. Technicians have to
streaming data to identify potential network
parse through siloed information sources in
issues and raise alarms for network engineers to
order to resolve issues that arise. Generative AI
quickly intervene. AI can help to reduce operations
can enrich issue data by automatically pulling
costs through preventative actions and improve in relevant information from network service,
the customer experience through proactive configuration, and performance metrics. It then
notifications and backup solutions. correlates it with past incidents and resolution
approaches, which can lead to a quicker
For network operators understanding of each problem, more certainty
Deloitte EMEA's Telecom Engineering Centre around the most effective fix, and faster mean-
time-to-diagnose and mean-time-to-resolve.
of Excellence outlines significant return on
investment for automation in its paper,
The Age of Telecom Network Automation.
Both AI and it's counterpart Generative AI are
emerging as game changers, saving critical
time in such areas as:
10
AI-powered Communications Service Providers
Demonstrated network operations benefits
These results
demonstrate that
AI-enabled operations
Reduced Reduced network Increased network Preventative care can improve network
operational downtime outage & incident for network access infrastructure by
complexity predictability decreasing disruptions,
reducing risk of
network failure, and
improving productivity
of network technicians.
1111
AI-powered Communications Service Providers
Create a winning AI strategy
When developing a sound AI strategy for transforming operations,
CSP leaders should align closely with their business objectives.
1 Develop a strategic Generative AI ambition
Your AI strategy should align to your core business objectives and goals and be clear on the
value it aspires to realize as well as the route to get there via considered, achievable action.
2 Produce a compelling case for transformation
Considerations should be made across economic viability, technology viability, privacy,
risk appetite, capacity required and competitive advantage for the case taken to the board.
3 Establish a purposeful approach to prioritizing use cases
Use a risk versus reward metrics to help determine where to start. AI use cases not only
need to aspire to business objectives, but also be evaluated for risks.
4 Identify key players to inspire and drive transformation
A cultural shift will be necessary to upend current business processes, bring people
along on the journey early.
5 Evolve talent to keep pace
For humans and machines to collaborate effectively, both business and technical teams
need to be fluent in and adaptable to new AI technologies.
6 Assess the technical landscape
The technology required to deliver, monitor, evaluate, and improve AI models should
all be evaluated.
7 Develop and efficient data governance approach
The traditional data capabilities built for traditional analytics can support AI with
additional attention on quality, governance, availability, and ownership clarity.
8 Ensure robust controls
Risks and their corresponding mitigations should be built into AI delivery
for every use case, not as an afterthought.
9 Put risk, privacy and ethical considerations at the forefront
Design governance and control mechanisms to ensure ethical and accountable
AI development aligned to policies and customer expectations.
10 Adapt operating models for AI development
Operating models across core and edge businesses should drive safe and consistent
delivery of AI solutions, instilling confidence in the decisions and insights that result.
12
AAII--ppoowweerreedd CCoommmmuunniiccaattiioonnss SSeerrvviiccee PPrroovviiddeerrss
AI is a transformative opportunity for CSPs. Many organizations are already
realizing significant benefits, both cost saving and revenue generating, by
automating processes and augmenting workforces with tools that can accelerate
and improve performance. The CSPs that look at the technology in a broad way,
with a clear path, across operations may better their long-term position.
As we progress through the AI However, it is imperative that the If CSPs do this correctly, smooth
era, CSPs have an opportunity to journey begins with a cohesive operations, delighted customers,
transform their operations with AI strategy in order to avoid and a differentiated competitive
enhanced automation, precision, a siloed approach that fails to advantage could lie ahead.
and personalization. extract the value from this
revolutionary technology.
1133
AI-powered Communications Service Providers
14
4 1
AI-powered Communications Service Providers
Reach out for a conversation.
Howie Stein Mohamad Said
Baris Sarer
Deloitte Consulting LLP Deloitte Consulting LLP
Deloitte Consulting LLP
[email protected] [email protected] [email protected]
Endnotes
1 Beena Ammanath et al, The Generative AI Dossier, Deloitte, 2023, p. 137.
2 Anubhav Mohanti et al, “The hidden costs of downtime: The $400B problem facing the Global 2000,”
Oxford Economics in partnership with Splunk, 23 July 2024.
3 Beena Ammanath et al, p. 137.
14
This publication contains general information only and Deloitte is not, by means of
this publication, rendering accounting, business, financial, investment, legal, tax, or
other professional advice or services. This publication is not a substitute for such
professional advice or services, nor should it be used as a basis for any decision or
action that may affect your business. Before making any decision or taking any action
that may affect your business, you should consult a qualified professional advisor.
Deloitte shall not be responsible for any loss sustained by any person who relies on
this publication.
About Deloitte
Deloitte refers to one or more of Deloitte Touche Tohmatsu Limited, a UK private
company limited by guarantee (“DTTL”), its network of member firms, and their related
entities. DTTL and each of its member firms are legally separate and independent
entities. DTTL (also referred to as “Deloitte Global”) does not provide services to
clients. In the United States, Deloitte refers to one or more of the US member firms of
DTTL, their related entities that operate using the “Deloitte” name in the United States
and their respective affiliates. Certain services may not be available to attest clients
under the rules and regulations of public accounting. Please see www.deloitte.com/
about to learn more about our global network of member firms.
Copyright © 2024 Deloitte Development LLC. All rights reserved. |
192 | deloitte | us-ai-institute-generative-artificial-intelligence.pdf | A new frontier
in artificial intelligence
2
Implications of Generative AI for businesses
Deloitte AI Institute
Implications of Generative AI for businesses
Implications of Generative AI for businesses
Executive summary Contents
The year 2022 was a watershed year for artificial intelligence Even so, Generative AI is in its infancy and not without risk. Section I
(AI), with the release of several consumer-facing applications Some of the most important risks to address relate to privacy and Decoding the Generative AI magic trick 5
like ChatGPT, DALL.E, and Lensa. The common theme the use security, managing bias, transparency and traceability of results,
of Generative AI–a paradigm shift in the world of AI. While current IP ownership, and equal access, especially for those at greater Section II
generations of AI use pattern detection or rule-following to help risk of job displacement. As such, participants should balance Consumer and enterprise use cases
analyze data and make predictions, the advent of transformer commercialization, regulation, ethics, co-creation, and even for Generative AI 9
architectures has unlocked a new field: Generative Artificial philosophy, as well as expand the group of stakeholder thinkers
Intelligence. Generative AI can mimic the human creative and contributors beyond technologists and enthusiasts. Section III
process by creating novel data similar to the kind it was Commerce and competition in Generative AI 17
trained on, elevating AI from enabler to (potentially) Ultimately, Generative AI could
co-passenger. In fact, Gartner estimates that more than 10% Section IV
create a more profound relationship
of all data will be AI-generated by as early as 2025,1 heralding a new Adopting and commercializing Generative AI 27
age, the Age of With™. between humans and technology,
even more than the cloud, the
Although early traction has been through consumer releases, which
About the Deloitte AI Institute
could be era-defining, Generative AI also has the potential to smartphone, and the internet
add contextual awareness and human-like decision-making
to enterprise workflows, and could radically change how we
did before.
Various analysts estimate the market for
The Deloitte AI Institute helps organizations connect the different
dimensions of a robust, highly dynamic and rapidly evolving
do business. We may be only just beginning to see the impact Generative AI at $200B by 2032.7 This represents ~20% of total
AI ecosystem. The AI Institute leads conversations on applied
of solutions like Google’s Contact Center AI (CCAI), which is designed AI spend, up from ~5% today.8 Said another way, the market
AI innovation across industries, with cutting-edge insights, to
to help enable natural language customer service interactions,2 and will likely double every two years for the next decade.
promote human-machine collaboration in the “Age of With”.
industry-specific solutions like BioNeMo from NVIDIA, which can Numbers aside, we believe the economic impact could be
accelerate pharmaceutical drug discovery.3 As such, Generative AI far greater. To help understand the potential, this paper
The Deloitte AI Institute aims to promote a dialogue and
has attracted interest from traditional (e.g., Venture Capital (VC), is equal parts primer and provocateur, adding structure
development of artificial intelligence, stimulate innovation, and
Mergers & Acquisitions (M&A)) and emerging (e.g., ecosystem to a rapidly changing marketplace. We start with a brief
examine challenges to AI implementation and ways to address
partnerships) sources. In 2022 alone, venture capital firms explainer of the foundational elements, delve into enterprise and
them. The AI Institute collaborates with an ecosystem composed
invested more than $2B,4 and technology leaders made significant consumer use cases, shift focus to how players across the market
of academic research groups, start-ups, entrepreneurs,
investments, such as Microsoft’s $10B stake in OpenAI5 and can build sustainable business models, and wrap up with some
innovators, mature AI product leaders, and AI visionaries, to
Google’s $300M stake in Anthropic.6 considerations and bold predictions for the future of Generative AI.
explore key areas of artificial intelligence including risks, policies,
ethics, future of work and talent, and applied AI use cases.
The far-reaching impacts and potential value when deploying
Combined with Deloitte’s deep knowledge and experience in
Generative AI are accelerating experimental, consumer, and soon,
artificial intelligence applications, the Institute helps make sense
enterprise use cases. And even though much media coverage
of this complex ecosystem, and as a result, deliver impactful
has focused on consumer use cases, the opportunities are
perspectives to help organizations succeed by making informed AI
widespread–and some are already here. Still, questions remain
decisions.
about how individuals and enterprises could use Generative AI to
deliver efficiency gains, product improvements, new experiences,
No matter what stage of the AI journey you’re in; whether you’re
or operational change. Similarly, we are only beginning to see how
a board member or a C-Suite leader driving strategy for your
Generative AI could be commercialized and how to build sustainable
organization, or a hands on data scientist, bringing an AI strategy
business models.
to life, the Deloitte AI institute can help you learn more about how
enterprises across the world are leveraging AI for a competitive
advantage. Visit us at the Deloitte AI Institute for a full body of our
work, subscribe to our podcasts and newsletter, and join us at our
meet ups and live events. Let’s explore the future of AI together.
www.deloitte.com/us/AIInstitute 3
2
Implications of Generative AI for businesses 2
IImmpplliiccaattiioonnss ooff GGeenneerraattiivvee AAII ffoorr bbuussiinneesssseess Implications of Generative AI for businesses
Section I: Decoding the Generative AI magic trick
SECTION I
Decoding the Generative AI
magic trick
The lofty expectations for Generative AI depend
on continued progress and innovation across
an interconnected hardware, software, and data
provider ecosystem.
The tech stack underlying Generative AI,
however, is in some ways similar to others
that came before. It consists of three layers:
t
infrastructure, platform, and applications.
Infrastructure is generally accepted s
as the most established, stable, and
commercialized layer. Incumbents
offer compute, networking, and storage,
including access to specialized silicon
(microprocessors) like NVIDIA’s GPUs and
Google’s TPUs optimized for AI workloads.
Meanwhile, the application layer is evolving
rapidly and consists of leveraging and
extending foundation models, which is s
Generative AI’s equivalent of a platform.
44 5
Implications of Generative AI for businesses Implications of Generative AI for businesses
Section I: Decoding the Generative AI magic trick
Generative AI Tech Stack Foundation Models, however, are what While this framework is applicable across AI
differentiate the Generative AI tech architectures, state-of-the-art Foundation
End-users stack from AI that came before. At its Models today (e.g., GPT-3, Stable Diffusion,
core, a Foundation Model, a term coined by Megatron-Turing) are based on a neural
Stanford University’s Center for Research on network architecture called transformers,
Application Development
3 Foundation Models, is a machine learning invented by a team at Google Brain in
User-facing B2B and B2C apps (ML) model pre-trained on a broad dataset 2017.10 Transformers represent a step
developed in partnership with that can be adapted to solve a range of change in ML performance and differ from
or on top of proprietary models problems.9 Just as Microsoft’s Win32 offers prior architectures in their ability to assign
Application APIs for developers to access base-level context, track relationships, and predict
Ecosystem hardware and OS functions, and NVIDIA’s outcomes. The most mature Foundation
Vertically Integrated CUDA allows graphic-intensive applications Models today are in the text domain,
Open/Closed APIs Foundation Models like game engines simplified access to GPU primarily driven by vast quantities of
Fine-tuned Models resources, the model layer is designed to available training data, which accelerated
Niche proprietary models
Refined models for targeted use case connect ambitious application developers to the development of Large Language Models
with pre-built user-facing
2 optimized hardware to help accelerate the (LLMs), a type of Generative AI foundational
B2B or B2C apps.
adoption of and democratize Generative AI. model. LLMs are trained to generate
text by predicting the next word in
Foundational Models
Model Layer These models are often available to a sequence or missing words within
Open or Close-source models
developers via closed and open APIs, where a paragraph.
developers can fine-tune models with
additional training data to improve context,
relevance, and performance to specific use
Hyperscale Compute
1 cases, all while optimizing delivery costs. Compute, networking, storage, and middleware
Foundation models are typically developed
in four stages, which are illustrated below.
Infrastructure/
Hardware Silicon
Specialized microprocessors/accelerators for training & interference Development of Foundation Models
1 2 3 4
Source: Deloitte
Architecture Pre-training Fine-tuning Production
The structure Training on Adjusting Deployment to
and design of the a massively large parameters production where
model and the dataset to create to improve the model is
algorithm used a defined set performance accessible via APIs
for training of parameters on specific tasks
Training Dataset Curated Dataset
Process Elements: Process Input
6 7
sretemaraP
ledoM
Output
ledoM
detsujdA
sretemaraP
Section I: Decoding the Generative AI magic trick
Moreover, Generative AI can create
artifacts across various modes—code,
images, video, audio, and 3D models.
This could both disrupt and drive step
changes in productivity across a range
of capabilities, from copywriting to
research and software engineering. For
example, in advertising, Generative AI could
create original copy, product descriptions,
and images in seconds. It can generate
synthetic X-ray images in healthcare, helping
physician diagnostic training.
Indeed, Generative AI
could transform how
businesses operate and
interact with customers
and may even redefine
an “employee” as we
know it. This transformation
is already underway
in some consumer and
enterprise spaces.
Source: Deloitte
Implications of Generative AI for businesses Implications of Generative AI for businesses
Section II: Consumer and enterprise use cases
for Generative AI
SECTION II
Consumer and enterprise
use cases for Generative AI
In 2022, OpenAI’s DALL·E 2 captured the world’s
attention with its text-to-image capabilities.11
The model creates images from simple prompts,
from something as direct as “a lion in a jungle”
to something more comical like “two lions playing
basketball in the style of Picasso.”
Ever since, Generative AI has occupied the Efficiency | Optimizing tasks like planning,
news cycle, punctuated by other launches research, and product discovery
like ChatGPT and previews like MusicLM. No
wonder we’ve seen broad-market consumer Instruction | Providing personalized
use cases, like Bing’s internet search guidance or learning content
powered by OpenAI’s ChatGPT.12 These
are emblematic of a Cambrian explosion in Creation | Generating or enhancing
consumer apps, touching everything from content, replicating the creative process
search to therapy.
Entertainment | Building games,
To help contextualize this explosion, virtual personas, and other entertainment
we group consumer use cases—those
that individuals invoke in their personal This is just an early view of the market; there
lives—into four broad categories based will likely be overlapping categories as work
on the utility provided: evolves. Moreover, new, category-defining
use cases are expected to emerge as future
generations of AI (e.g., those that enable
multi-model engagement or run entirely
on-device) mature.
8 9
Implications of Generative AI for businesses Implications of Generative AI for businesses
Section II: Consumer and enterprise use cases
for Generative AI
A Sampling of Consumer Use Cases Available Today
Efficiency Instruction Creation Entertainment
Creating a health Conversing with Generating & editing
Creating original games
& wellness plan virtual companions video files
Creating personalized Creating interior Chatting with pop
Discovering new products
financial plans design mockups culture figures
Conducting research Curating outfits
Teaching new languages Rendering 3D environments
with citations & fashion ideas
Modifying & editing Remixing or
Curating content Synthesizing research papers
design files sampling music
Guiding & informing Creating art Generating original fictional
Answering general questions
personal writing & editing images short stories
Sample vendors
Synthesis.ai Grammarly Luminar AI Jasper
Consensus Lingostar.ai Lensa Scenario
10 11
ytixelpmoC
hgiH
ytixelpmoC
woL
Section II: Consumer and enterprise use cases
for Generative AI
The pace of change can make predictions challenging, Consumer use cases can also be indicators These efficiencies may even redefine
of the possibilities in the enterprise. job expectations, making prompt
but as of early 2023, we expect consumer use cases
However, unlike consumers, enterprises engineering (i.e., asking AI the right
with the following aspects as having staying power: require advanced features, proven ROI, questions) a differentiating skill set.
customization, organizational content, Ultimately, horizontal use cases will
security, and technical support. In today’s create a commercial foundation for more
formative era of Generative AI, the most specialized applications. Enterprises must
popular enterprise use cases—invoked start deploying these early to help build
to drive internal or B2B outcomes—will capabilities and a knowledge base, making
be general purpose or applicable across the value case for vertical applications
industries or functions (“horizontal”). over time.
However, like technologies that came before,
Speed to market Occupational utility Seamless integrations
there are often more sustainable value- Today, some enterprises are already driving tangible
Consumer awareness, increasingly through Products that create value in the workplace, Solutions that integrate into platforms could
creation opportunities in industry-specific
social media, could lower acquisition like writing assistants, may be easier be discovered through existing workflows, returns from investments in horizontal use cases.
enterprise use cases (“vertical”).
costs, allowing companies to piggyback on to fit into a sustainable business model, driving more “sticky” adoption. Grammarly
We’ve seen research teams summarize
coverage, work out product kinks, and scale as opposed to products attached to was early to market with this on PCs and,
Potential targets of horizontal use third-party information, product managers
efficiently with an active and contributory a “hype cycle,” like social media filters. more recently, OpenAI with Bing.
cases are well-established automation write requirements documentation, social
user base.
centers, offer a substantial volume of media marketers refine copy, and customer
training data (e.g., knowledge base, service teams create case summaries and
support chat logs), and are the focus suggested resolutions. However, tangible
of cost optimization and productivity ROI could depend on proprietary
improvement efforts. For example, and serviceable data, secure model
creative marketing tasks like writing partitioning, talented product leaders
advertising copy, blogs, or social media and ML engineers, enabling MLOps
captioning can take hours or days for tooling, and new commercial and operating
humans to author. In contrast, Generative models. These are investments that
AI can complete workable drafts in minutes, enterprises should evaluate, whether they
requiring only editing from humans. see themselves as early adopters, fast
followers, or late entrants.
Source: Deloitte
Implications of Generative AI for businesses Implications of Generative AI for businesses
Sampling of Vertical and Horizontal Enterprise Use Cases
Consumer & Life Sciences & Banking & Fin. Technology Media & Industrial & Government &
Retail Health Care Services Telecom Manufacturing Public Sector
Personalized AR/VR Content Fraud Simulation Personalized AR/ Original Games Geological Academic 24/7
Conversational Generation for & Pattern VR Experience Creation Assessment Office Hours
Retail Experience Digital Therapy Detection Generation • • • • • • for Oil Exploration Virtual Assistant
• • • • • • • • • • • • • • • • • • • •
Customized Predictive & Tax and Automated Trailer & Summary Generative Infrastructure
Product Design & Virtual Patient Compliance Audit Product & Generation Simulation Mapping
Recommendation Triage & Scenario Testing Hardware Design • • • • & Safety Testing & Planning
• • • • • • • • • • • • • • •
Product Details 3D Images of Retail Banking Personalized & Script/Score 3D Env. Disaster Recovery
& Photography Anatomy for Transaction Automated UI/UX Design Rendering: Well Simulation
Generation Education Support Design & Subtitle Sites, Pipelines, • •
• • • • • • • • • Generation etc.
• • •
Fashion Outfit Healthy & Personalized Product Testing Personalized Automated Tech. Fraud, Waste &
Curation Wellness Plan Virtual Financial & Feedback News Equipment Abuse Prevention
• • • Creation Advisor Generation & Content Training Reports
• • • • • • • Generation • • • • •
• • •
Personal Art Drug Discovery Financial Software Sales, CX Original Fictional Generative Research
Creation Through Molecule Reporting & Retention Short Stories Automation w/ Citations
& Edits Simulation Analysis & Insight Support Generation for Smart & Explainers
• • Gen. • • • Factories •
• • •
Personalized Self-serve HR & IT End-to-end Customer Feedback Automated Code Dialogue Generation
Conversational Functions Automated Sentiment Debugging for Virtual Assistants
Retail Experience • • Customer Service Classification & Issue Resolution •
• • • • • • •
Enterprise Search 3D Environment Marketing/Sales Accessibility Support Autonomous Code Personalized
& Knowledge Mgmt. Rendering: Content Generation (text-to-speech & Generation Targeted Ads
• • Metaverse • • speech-to-text) & Completions across platforms
• • • • • • • • •
12 13
gnigremE
erutaM
• Text • 3d Model
• Image • Code
• Audio • Others
• Video
LACITREV
LATNOZIROH
Section II: Consumer and enterprise use cases Section II: Consumer and enterprise use cases
for Generative AI for Generative AI
In contrast, vertical use cases target industry-specific workflows
that require domain knowledge, context, and expertise.
For these, foundation models may need to be Ease of use | Integrations into systems and
fine-tuned or may even require new special- workflows via out-of-the-box connections and
purpose models. For instance, Generative AI low/no code tooling, reducing expensive IT
can be used to create a customized portfolio resources and enabling frontline users.
of securities based on risk-reward descriptions
or recommend personalized treatment plans Security and privacy | Compliance with data
based on a patient’s medical history and security standards (e.g., SOC 2, HIPAA, GDPR)
symptoms. However, delivering performant and role/persona-level access control over
vertical use cases requires a nuanced confidential data.
understanding of the field. In software, for
example, Generative AI can design composable Robust ecosystems | Broad set of
blocks of code based on simple prompts, which development and service partners to extend,
requires tacit knowledge of efficient coding, customize, and co-develop specialized data
coding languages, and an understanding sets, use cases, and applications.
of technical jargon.
Transparency and explainability |
Enterprise buyers have unique purchase Understanding how model outputs and
decisions relative to consumers, as model responses are derived and the ability
performance (speed, relevance, breadth to perform root cause analysis
of sources) is not expected to exclusively on inaccurate results.
drive vendor selection. on early opinions
from both advocates and naysayers, frequently Flexibility and customization | Ability to
cited criteria to adopt Generative AI are: create parameters, train on proprietary data,
and customize embeddings while maintaining
privacy and ownership
of data and tuning.
Generative AI Modality
Source: Deloitte
Implications of Generative AI for businesses Implications of Generative AI for businesses
Section II: Consumer and enterprise use cases Section II: Consumer and enterprise use cases
for Generative AI for Generative AI
Despite its promise, myriad challenges should
be overcome before Generative AI can be deployed
Even as new use cases emerge at an accelerating pace,
at scale. We discuss these in more detail, but there
we believe the market will unfold in six ways:
is also the question of commercial viability. In other
words, for all the fascinating possibilities and use cases
for Generative AI, it still needs to be determined how
vendors will build a sustainable business model.
Today, there are ethical concerns with Generative AI, While horizontal use cases will likely be the first to
including its potential for workforce displacement.16 deliver value, vertical-specific use cases could
However, like previous generations of AI, this command a premium due to the dependence on
technology will likely primarily augment human proprietary data. As such, data will be a currency,
performance. Indeed, AI could be commonplace creating new economies for access to proprietary
in worker’s toolkits, like Workspace among analysts, and synthetic data.
GitHub among coders, or Creative Cloud
among marketers.
6
1
Regulatory actions will likely vary
in speed, reach, oversight, and 5
reporting requirements across 2 All industries can benefit from
major markets (e.g., US AI Bill Generative AI. However, data-rich
of Rights,13 EU AI Act,14 China sectors (e.g., banking, retail,
Cyberspace Administration15). As hospitality) or those whose
such, vendors and enterprises will products leverage data (e.g.,
need to proactively establish information services) may move
practices that ensure data 4 —and should move—faster.
quality, transparency, fairness, 3 Conversely, those based on
safety, and robustness, which will judgment (e.g., law, medicine)
be critical to Trustworthy AI. may be more cautious about
adopting but nevertheless see
the benefit by accelerating the
synthesis of prior knowledge.
Text-based use cases will be commercialized
first, but the potential cost and productivity Given the shift away from low-interest rates, costs will
gains may be greater when commercializing increase, pushing enterprises to invest in use cases with
higher-order tasks as these skills can be more clear ROI. As such, use cases that directly impact cost
expensive to recruit, take longer to train, and (e.g., chatbots), productivity (e.g., search), or revenue
are right-brain (creative) versus left-brain (e.g., marketing copy) could have greater adoption than
(logical), making success subjective. those that eliminate humans.
14 15
Implications of Generative AI for businesses Implications of Generative AI for businesses
Section III: Commerce and competition
in Generative AI
SECTION III
Commerce and competition
in Generative AI
The battle for value capture will be fought
on multiple fronts, and each layer of the stack
will have its competitive dynamics driven
by things like scale, data access, brand,
and a captive customer base.
However, we see two primary competitor To begin, the infrastructure layer, which
archetypes: pure-play providers operating is the most mature of the Generative AI
within a single layer–infrastructure, technology stack, is where hyperscalers
model, and application - and integrated dominate the market. The business
providers that play in multiple layers. As model here is proven: provide scalable
with incumbent technology, we expect compute with transparent, consumption-
consumer pricing to be simple (e.g., per based pricing. To help make Generative
user, per month) and enterprise pricing AI workloads “sticky,” hyperscalers
to be more complex (e.g., per call, per have entered commitments with model
hour, revenue share). However, pricing providers to guarantee future workloads,
simplicity, predictability, and value including Azure with OpenAI,17 Google with
will be important to scaling within the Anthropic,18 and AWS with Stability.ai,19
enterprise beyond early adopters or alongside their proprietary models.
edge use cases.
16 17
Implications of Generative AI for businesses Implications of Generative AI for businesses
Section III: Commerce and competition Section III: Commerce and competition
in Generative AI in Generative AI
While the cloud service providers (CSP) deliver abstracted Next is the model layer, where the market Another less-considered path to monetization could
is evolving fast. This area can be resource
services, there is another enabling layer within be developing and licensing model architectures
intensive; model builders must continually
infrastructure that is rapidly evolving: silicon. revisit architectures (e.g., parameters, or development platforms.
embeddings) to maintain performance.
Here, NVIDIA is a leader with their Ampere They have to attract and retain AI talent In other industries, like semiconductors,
and Hopper series GPUs purpose-built (i.e., architects, engineers, and data ARM (CPU) and Qualcomm (wireless
for training and inference workloads, scientists) to design the frameworks, networking) create large, stable business
respectively, coupled with their Selene guardrails, and learning mechanisms to models built on licensing fees.
supercomputing clusters that speed ensure the robustness and reliability of
up training time.20 Similarly, AMD’s models. Finally, Generative AI workloads
CDNA2 Architecture is purpose-built for can run up large bills due to their
exascale computing on machine learning compute-heavy nature and need for
applications, advancing competition in the specialized silicon.22 No wonder we’ve
high-performance computing market.21 seen players start to recoup the investment
by charging fees or integrating into
monetized products (e.g., GPT-3.5 into Edge,
LaMDA into Google Search).
Infrastructure Layer Model Layer
Offering Description Examples Primary Customer Primary Monetization Offering Description Examples Primary Customer Primary Monetization
Enterprise Developer Consumer Model Metric Enterprise Developer Consumer Model Metric
Hosted and managed Co:here
Amazon Closed-source Model Per token
models built on a vast Google Yes Yes No Consumption
Hyperscale and purpose- Baidu Per minute Providers Per API call
data corpus OpenAI
Cloud Service Provider built compute, storage, Yes Yes No Consumption By CPU/GPU
Google
and networking type
Microsoft Foundation models
Open-source Model Meta Monetized via fine-tuned
maintained by No Yes No
Providers Stability.ai models or model hubs
communities
Specialized services to Amazon
Per hour
Generative AI Service accelerate deployment Use case-specific
Co:here Yes Yes No Consumption Per generation Fine-tuned Model Co:here Per token
Providers (e.g., security, monitoring, versions of foundation Yes No No Consumption
Per embedding Providers C3.ai Per API call
testing, model isolation) Google models
Github
Marketplace, community Subscription
Hugging Per month
Model Hubs or hosting services for Yes Yes No Consumption
Purpose-built Face Per hour
AMD One-time models Rev. share
Chip Provider semiconductors, Yes No No Per component Replicate
NVIDIA lease
including GPUs and CPUs
Proprietary architectures, Co:here One-time Per embedding
Model Service
synthetic data, weights, MostlyAI No Yes No Subscription Per month
Providers
and embeddings RealAI License Per user
18 19
Implications of Generative AI for businesses ImplIimcaptliiocantsio onfs G oef nGeernaetriavteiv Ae IA fIo fro rb buussinineesssseess
Section III: Commerce and competition Section III: Commerce and competition
in Generative AI in Generative AI
Finally, the application layer serves as the gateway Competition within the application layer could
between models and end users. unfold within several markets. However, given the
Today's apps are typically monetized wide range of applications and use cases that may
through subscriptions and recurring
emerge, we should look at “micro-markets.” Broadly,
transactions, a model that will likely persist,
albeit with modifications suited to today’s real and predicted enterprise use cases fall into
Generative AI.
five categories where competitive lines could be drawn:
Application Layer
Offering Description Examples Primary Customer Primary Monetization
Enterprise Developer Consumer Model Metric Accelerate
Improve productivity by speeding up outcomes. These do not eliminate human
Google
SDKs, frameworks, intervention but provide high-quality inputs upon which to build.
Hugging License
Platforms and tools to build and Yes Yes No Per user
Face Rev. share
distribute apps
Microsoft
Personalize
Create intimacy and personalization, which previously would have taken significant
Boomy Subscription Per user
Full-feature solutions to effort. Here, models can leverage personal data to tailor content.
Standalone application Canva Yes No Yes Consumption Per month
modify workflows
Lensa One-time Per service
Automate
Extensions and features AI Art
Subscription Per user Deliver business and technical workflows and, in certain instances, replace humans.
Plugins to supplement tasks and Grammarly Yes No Yes
Consumption Per month Vendors often demo these due to the immediate cost-saving potential.
workflows Jasper
Create
Push the boundaries of intellectual property development, leveraging prompts (a new
art form unto itself) to generate novel content like images, video, text, and media.
Simulate
Create environments in which workflows, experiments, and experiences can be simulated
before being pushed into production, saving time, cost, and physical resources.
20 2211
Implications of Generative AI for businesses Implications of Generative AI for businesses
Accelerate Personalize Automate Create Simulate
Calendar mgmt./
Email outreach Social media marketing Image/logo creation 3D modeling
Admin assistant
Gaming environment
Note taking Keynote speaker notes Advertising copy Marketing campaigns
design
Short-form video
Content marketing Physical goods design Support chatbots Medical testing (R&D)
generation
NLP-based email/ Product ideation
Advertising video editing Content summarization Chemical interactions
app. responses & PRD authoring
Basic code generation Disaster response
Code completion Personal assistant Music scoring
& documentation management
Anthropic Facebook OPT BigScience BLOOM OpenAI DALL.E 2 Cradle
Co:here GATO OpenAI Codex Soundify DreamFusion
OpenAI GPT-3 Microsoft X-CLIP Tabnine Stable Diffusion NVIDIA GET3D
22 23
SNOITACILPPA
SLEDOM
Section III: Commerce and competition Section III: Commerce and competition
in Generative AI in Generative AI
Sampling of Enterprise Micro-Markets A second archetype, in contrast to pure-play We see integration happening in two
providers who monetize through first- ways. First, companies like Anthropic and
and third-party channels, are vertically Midjourney have released applications
integrated or multi-layer players. for specific use cases. Lower in the stack,
These players lead with bundled pricing, companies like NVIDIA have released
proprietary data, special-purpose specialized models, including BioNeMo,
clouds, or cross-domain expertise a pharmaceutical pipeline development
to gain a competitive advantage. accelerator that is optimized
to run on NVIDIA GPUs.
Integrated Players
Offering Description Examples Primary Customer Primary Monetization
Enterprise Developer Consumer Model Metric
Anthropic Per month
Applications built on
Model Co:here Subscription Per user
proprietary, first-party Yes No Yes
and application Midjourney Consumption Per service
models
OpenAI Per download
Fully managed Per hour
Model and Google
infrastructure and model- Yes Yes No Consumption Per API call
infrastructure NVIDIA
Source: Deloitte as-a- service Per embedding
Amazon
Purpose-built horizontal Per minute
This may have implications for the model Silicon and Azure
and vertical clouds for ML Yes |
193 | deloitte | us-cbe-nacd-2024-governance-outlook-deloitte.pdf | This document was prepared solely for your internal use only, and it is the sole property of its copyright owner. Further distribution of the content (in whole or in part)
in any form is prohibited without written permission from NACD. Copyright © 2023 National Association of Corporate Directors. All rights reserved.
Artificial Intelligence: An Emerging
Oversight Responsibility for
Audit Committees?
By Brian Cassidy, Ryan Hittner, and Krista Parsons, Deloitte & Touche LLP
The audit committee has many discrete duties, including overseeing financial reporting and
related internal controls, the independent and internal auditors, and ethics and compliance,
to name just a few. However, these and other duties are part of a broader audit committee
responsibility: risk oversight. While the audit committee does not manage all risks, it is responsible for
overseeing the procedures and processes by which the company anticipates, evaluates, monitors,
and manages risks of all types. Recent developments in artificial intelligence (AI), including the
emergence of generative AI, are leading businesses to evaluate AI’s potential impact to their business
technology strategy. As businesses expand their use of AI, especially into core business processes,
the audit committee will need to understand the challenges and opportunities presented by AI to
address risks related to governance and stakeholder trust.
WHO’S MINDING THE AI STORE NOW?
According to a 2023 survey conducted by Deloitte and the Society for Corporate Governance,
corporate secretaries see AI strategy and oversight as still evolving. The findings show that few
respondents (13%) had a formalized AI oversight framework, although many (36%) were considering
the development and implementation of AI oversight policies and procedures.
22
2024 Governance Outlook
This document was prepared solely for your internal use only, and it is the sole property of its copyright owner. Further distribution of the content (in whole or in part)
in any form is prohibited without written permission from NACD. Copyright © 2023 National Association of Corporate Directors. All rights reserved.
These results are particularly interesting when compared to a 2022 Deloitte survey, in which 94
percent of respondents said AI was critical to their company’s short-term success.1 This may suggest
some level of information asymmetry between management and the board, congruent with the
notion that AI is in a state of flux. Thus, at least for now, the AI landscape might best be characterized
as an abstract governance puzzle.2
THE AI GOVERNANCE PUZZLE
Oversight Structure [Not] on the Agenda Risky Lack of Opinion
29% reported that AI over- 44% indicated that AI has not 68% didn’t know (or didn’t
sight was not assigned to been on any agenda (full respond) when asked how
any committee or the full board or committee); 37% the company mitigates
board; 16% placed it with the have discussed on an ad hoc AI-related risk.
audit committee. or as needed basis.
RISKS AND OPPORTUNITIES
FAMILIAR AND DIFFERENT SET OF RISKS
With new technology comes the possibility of new risks. Some AI risks present well-trodden chal-
lenges that arise in other technology areas and can be overseen and understood in the context of an
ongoing enterprise risk management (ERM) process,3 such as the COSO ERM framework. However,
other risks may be unfamiliar and/or amplified. A few illustrative examples are highlighted below.
X Shadow IT Environments: Use of IT assets by personnel without the knowledge or
oversight of IT security professionals can occur with any type of software or hardware.
However, unauthorized use of generative AI by personnel may compound data-related
risks. This risk may be increased given the lack of AI policy in many organizations.
Further, employees leveraging generative AI to write code may inadvertently introduce
vulnerabilities through code generated by AI.
1 Business leaders were defined as company representatives who met one or more of the following qualifiers: (1)
responsible for AI technology spending or approval of AI investments, (2) responsible for the development of AI
strategy, (3) responsible for implementation of AI technology, (4) acting as AI technology subject-matter specialist,
or (5) otherwise stated they were influencing decisions around AI technology. See Nitin Mittal, Irfan Saif, and Beena
Ammanath, Fueling the AI transformation: Four key actions powering widespread value from AI, right now, State of AI
in the Enterprise, 5th Edition report, Deloitte, October 2022.
2 Natalie Cooper, Bob Lamm, and Randi Val Morrison, “Future of tech: Artificial intelligence (AI),” Board Practices Quar-
terly, Deloitte, August 2023.
3 Alexander J. Wulf and Ognyan Seizov, “‘Please understand we cannot provide further information’: Evaluating content
and transparency of GDPR-mandated AI disclosures,” AI & Society (2022).
23
2024 Governance Outlook
This document was prepared solely for your internal use only, and it is the sole property of its copyright owner. Further distribution of the content (in whole or in part)
in any form is prohibited without written permission from NACD. Copyright © 2023 National Association of Corporate Directors. All rights reserved.
X IP Ownership and Infringement: Generative AI users can input confidential or
protected data, which may result in an array of adverse outcomes, including
disclosure of such confidential or protected data to third parties. Outputs using this
type of data may also constitute infringement of intellectual property.4 Furthermore,
as generative AI applications are used to craft increasingly sophisticated media
across multiple formats, it may not be clear who owns the rights to any resulting
intellectual property.
X Cybersecurity Bad Actors: A frequent concern across many types of technology stems
from malicious actors who circumvent security protocols. Generative AI use cases may
amplify some types of cybersecurity risks. For example, hackers may use generative AI
to write code for purposes of infiltrating data environments or create phishing messages
that more accurately mimic human language and tone.
Finding the appropriate balance between AI’s benefits and risks depends on a constellation of
factors. Outputs produced by generative AI change over time as the technology learns from data.
But just like with humans, it is possible for this subcategory of AI technology to learn things that are
incorrect. For that reason, traditional risk management strategies may not be well-equipped for the
challenges that arise from generative AI use.
GENERATIVE AI RISK EXAMPLES
Low Transparency Hallucination Bias Potential Value Alignment
How generative AI Generative AI When trained on Even with safe-
derives its output products and services nonrepresentative guards, generative
can be a “black box,” may generate data, generative AI AI output may
making it difficult to output that seems output could exhibit contradict its
explain and/or audit. accurate but is systematic errors. intended purpose.5
actually false or
cannot be justified.
4 Christian Heinze, “Patent infringement by development and use of artificial intelligence systems, specifically artificial
neural networks,” in A Critical Mind: Hanns Ullrich’s Footprint in Internal Market Law, Antitrust and Intellectual Property,
eds. Christine Godt and Matthias Lamping, MPI Studies on Intellectual Property and Competition Law, vol. 30 (Hei-
delberg, Germany: Springer, 2023), pp. 489–515.
5 Vic Katyal, Cory Liepold, and Satish Iyengar, “Artificial intelligence and ethics: An emerging area of board oversight
responsibility,” On the Board’s Agenda, Deloitte, 2020.
24
2024 Governance Outlook
This document was prepared solely for your internal use only, and it is the sole property of its copyright owner. Further distribution of the content (in whole or in part)
in any form is prohibited without written permission from NACD. Copyright © 2023 National Association of Corporate Directors. All rights reserved.
Regardless of whether the risk is familiar, completely new, and/or amplified, the resultant conse-
quences may be notable. Failure to mitigate any subcategory of AI-related risks may lead to many
adverse outcomes such as reputational damage, financial losses, legal action, and regulatory
infractions. A starting point for addressing such concerns might include using mitigation strategies
that are already known to work in other contexts, such as the COSO ERM framework referred to
earlier. For AI-centric guidance related to implementation and scaling, it may be worth considering
the benefit of systems such as the NIST AI Risk Management Framework.
WITH RISKS COME BENEFITS, TOO
If AI presented nothing but risk, it seems unlikely that it would have emerged as “the” technology of
the future. Clearly, AI has benefits, some of which may not be known for some time. One particular
set of benefits is squarely in the audit committee’s wheelhouse—namely, the potential to streamline
and enhance a company’s internal audit, financial reporting, and internal control functions. There
are also aspects of generative AI technology that, while still evolving, may one day fundamentally
change an organization’s financial systems. While there is much uncertainty, the future transforma-
tive potential of generative AI may add much to the current array of use cases. In the shorter term,
various subcategories of AI are already capable of improving the quality of financial reporting via
reviewing transactions, identifying errors, addressing internal control gaps, and detecting fraud.
If AI isn’t being used within these areas, the audit committee might ask if the company is exploring
potential use cases—and if the company is not, the committee might ask to hear the reasons behind
that decision.
USE OF AI TECHNOLOGY MAY HAVE MANY BENEFITS
Cost Savings Boosted Revenues Development Time New Insights
Process automations AI-infused products AI may shorten time Appropriate
and improvements and services may to market by increas- generative AI
may improve task provide new growth ing the speed of use may bolster
efficiency. opportunities. early-stage testing. employee creativity.
25
2024 Governance Outlook
This document was prepared solely for your internal use only, and it is the sole property of its copyright owner. Further distribution of the content (in whole or in part)
in any form is prohibited without written permission from NACD. Copyright © 2023 National Association of Corporate Directors. All rights reserved.
COMMON AI USE CASE EXAMPLES
USE CASE DESCRIPTION OPPORTUNITIES RISKS
Use of intelligent The technology may Poorly designed or
automation to match reduce costs by maintained systems
invoices to payments, processing a large may generate errors
including classification volume of transactions that are time consum-
Invoices
of expenses with a high degree of ing to undo.
and Payments
accuracy.
Leverage of natural By producing the Natural language and
language and gener- initial drafts or identi- generative AI trained
ative AI processing fying common errors, on biased data may
to create legal docu- generative AI may misapply the law or
Contract Review ments or review them create efficiencies and make up precedent.
or Generation for errors lower legal liability in a
cost-effective manner.
Incorporating Modeling and analytics Lack of robust testing
predictive analytics to AI technology may be and regular updates
improve the accuracy capable of identifying can cause modeling
of functions like inven- patterns at a speed that and analytics AI to
Forecasting
tory management and outpaces human-led become more inaccu-
and Modeling
revenue forecasting data analysis efforts. rate over time.
Use of generative AI Employees may use The technology may
to develop models generative AI to drive expose confidential
or applications efficiencies in day-to- data with generative
that create effi- day tasks and help AI inputs or may
Code ciencies for routine identify possible gener- create outputs that
Development personnel activities ative AI use cases. involve intellectual
property infringement.
AI AND THE AUDIT COMMITTEE
The tendency to assign oversight of emerging risks to the audit committee means it is sometimes
described as the “kitchen sink” of the board. However, as noted earlier, this is consistent with the
audit committee’s overarching role in risk oversight. It’s also worth considering that it is common for
topics taken on by the audit committee at the outset to eventually be overseen by other committees.
Some aspects of AI oversight seem more aligned with the audit committee’s work than others. And
when it comes to considering such congruence questions, it may be helpful to think about the audit
committee’s current levels of technology fluency and comfort. For instance, given the audit commit-
tee’s traditional governance areas, it may be prudent for it to oversee AI use in financial reporting.6
6 The audit committee may want to also think about indirect impacts. Depending on the use case, AI technology may
have an array of indirect effects on financial measures (GAAP or otherwise).
26
2024 Governance Outlook
This document was prepared solely for your internal use only, and it is the sole property of its copyright owner. Further distribution of the content (in whole or in part)
in any form is prohibited without written permission from NACD. Copyright © 2023 National Association of Corporate Directors. All rights reserved.
In other parts of AI oversight, it may be less clear whether the audit committee is a “good fit.” For
example, the impact of generative or natural language AI on the workforce may be more aligned
with the oversight of the compensation/talent committee or the full board.
The “temporary assignment” of AI to the audit committee may make sense for other reasons, as well.
First, AI remains an emerging technology and is likely to continue to change rapidly. Second, there is
extensive governmental interest in AI, which may result in legislation that will require adjustments in
its oversight. Thus, determining now that AI, or aspects of AI, should be overseen by another commit-
tee or committees may turn out to be premature.
An audit committee might choose to assess its AI risk tolerance across oversight areas such as auditing,
financial reporting, and internal control functions. It may be helpful to contextualize that analysis
by comparing it to other areas of the company. For example, company divisions that routinely use
technology enhancements in client-facing operations may have a higher appetite for risk. But a
higher risk tolerance in operational settings does not necessarily correlate with how risks are viewed
when it comes to financial reporting impacts.
An important part of the AI governance puzzle for the audit committee is assessing risk. But, at least
for now, this task is currently made more difficult by a shifting regulatory landscape. Governments
and regulators around the world are considering whether regulation and policy can address AI risks.
Their progress toward developing and enacting policies and regulations over AI is uneven across the
globe and in different stages of development and enactment. And to make things more complex,
stakeholder groups—shareholders, customers/clients, employees, suppliers, and community—all have
varying and sometimes conflicting expectations around use and governance of AI. For these reasons,
there may be a benefit to continuously assessing AI risks and benefits over waiting for emerging
and future legislative proposals or regulatory guidance. But to accurately make such continual
assessments, it’s important that the audit committee and the board have sufficient knowledge to ask
questions around the organization’s adoption and use of AI.
POTENTIAL AUDIT COMMITTEE OVERSIGHT QUESTIONS
TO CONSIDER
X What are the company’s current and potential future use cases for AI, and do any of
them have an impact on financial reporting or other audit committee oversight areas?
X Has management considered opportunities to use AI that may enhance or improve
financial reporting processes?
X What processes are, or will be, used to evaluate dependencies that may arise in other
areas where the audit committee may have primary oversight, like cybersecurity or
data management?
X Are processes for use of AI congruent with the company’s risk appetite in terms of level
of proactiveness and mitigation strategy?
X Given the speed of AI technology development, are existing processes being assessed
and updated with appropriate frequency?
27
2024 Governance Outlook
This document was prepared solely for your internal use only, and it is the sole property of its copyright owner. Further distribution of the content (in whole or in part)
in any form is prohibited without written permission from NACD. Copyright © 2023 National Association of Corporate Directors. All rights reserved.
Brian Cassidy Ryan Hittner Krista Parsons
Brian Cassidy is an Audit & Assurance partner with Deloitte & Touche LLP and the US Audit & Assurance Trustworthy
AI leader.
Ryan Hittner is an Audit & Assurance principal with Deloitte & Touche LLP and the US Artificial Intelligence & Algorithmic
Assurance coleader.
Krista Parsons is an Audit & Assurance managing director with Deloitte & Touche LLP. She is also the Governance
Services coleader and the Audit Committee Program leader for Deloitte’s Center for Board Effectiveness.
This publication contains general information only and Deloitte is not, by means of this publication, rendering account-
ing, business, financial, investment, legal, tax, or other professional advice or services. This publication is not a substitute
for such professional advice or services, nor should it be used as a basis for any decision or action that may affect your
business. Before making any decision or taking any action that may affect your business, you should consult a qualified
professional adviser. Deloitte shall not be responsible for any loss sustained by any person who relies on this publication.
Deloitte refers to one or more of Deloitte Touche Tohmatsu Limited, a UK private company limited by guarantee (“DTTL”),
its network of member firms, and their related entities. DTTL and each of its member firms are legally separate and
independent entities. DTTL (also referred to as “Deloitte Global”) does not provide services to clients. In the United
States, Deloitte refers to one or more of the US member firms of DTTL, their related entities that operate using the
“Deloitte” name in the United States and their respective affiliates. Certain services may not be available to attest clients
under the rules and regulations of public accounting. Please see http://www.deloitte.com/about to learn more.
28
2024 Governance Outlook |
194 | deloitte | us-ai-institute-responsible-use-of-generative-ai.pdf | Proactive risk management
in Generative AI
Deloitte AI Institute
A call for proactive risk management in Generative AI
About the AI Institute
The Deloitte AI Institute helps organizations connect
the different dimensions of a robust, highly dynamic
and rapidly evolving AI ecosystem.
The AI Institute leads conversations on helps make sense of this complex ecosystem,
applied AI innovation across industries, with and as a result, deliver impactful perspectives
cutting-edge insights, to promote human- to help organizations succeed by making
machine collaboration in the “Age of With”. informed AI decisions.
The Deloitte AI Institute aims to promote No matter what stage of the AI journey
a dialogue and development of artificial you’re in; whether you’re a board member
intelligence, stimulate innovation, and or a C-Suite leader driving strategy for your
examine challenges to AI implementation organization, or a hands on data scientist,
and ways to address them. The AI Institute bringing an AI strategy to life, the Deloitte AI
collaborates with an ecosystem composed institute can help you learn more about how
of academic research groups, start-ups, enterprises across the world are leveraging
entrepreneurs, innovators, mature AI product AI for a competitive advantage. Visit us at
leaders, and AI visionaries, to explore key the Deloitte AI Institute for a full body of
areas of artificial intelligence including risks, our work, subscribe to our podcasts and
policies, ethics, future of work and talent, newsletter, and join us at our meet ups and
and applied AI use cases. Combined with live events. Let’s explore the future of
Deloitte’s deep knowledge and experience in AI together.
artificial intelligence applications, the Institute
www.deloitte.com/us/AIInstitute
2
A call for proactive risk management in Generative AI 3
Generative AI is dominating public
interest in artificial intelligence
By some estimations, Generative AI is the end of the
Internet search and the tool that will revolutionize
many aspects of how we work and live. We’ve heard
that before in AI. The newest applications often
conjure public excitement.
Yet, Generative AI is different than most In the business realm, there is growing
other kinds of AI in use today. Large intrigue around how Generative AI can be
language models, for example, can respond used in the enterprise. As with all cognitive
to user prompts with natural language tools, the outcomes depend on how they are
outputs that convincingly mimic coherent used, and that includes managing the risks,
human language. What is more, there is which for Generative AI have not been as
effectively no barrier to using some of these deeply explored as the capabilities.
models because they do not require any Some primary questions are, can business
knowledge of AI, much less an understanding users trust the outputs of this kind of
of the underlying math and technologies. AI application, and if not, how can that
be achieved?
3
A call for proactive risk management in Generative AI
New bots on the block
To this point, AI has broadly been used CIOs and technologists may already know
to automate tasks, uncover patterns that Generative AI is not “thinking” or being
and correlations, and make accurate creative in a human way, and they also likely
predictions about the future based on know that the outputs are not necessarily
current and historical data. Generative AI is as accurate as they might appear. Non-
designed to create data that looks like real data. technical business users, however, may not
Put another way, Generative AI produces digital know how Generative AI functions or how
artifacts that appear to have the same fidelity much confidence to place in its outputs.
as human-created artifacts. Natural language The business challenge is magnified by the
prompts, for example, can lead the neural fact that this area of AI is evolving at a rapid
network to generate images that are in some pace. If organizations and end users are
cases indistinguishable from authentic images. challenged just to keep up with Generative
For large language models that create text, AI’s evolving capabilities, how much more
the AI sometimes supplies source information, difficult might it be to anticipate the risks
underscoring to the user that its outputs are and enjoy real trust in these tools?
factually true, as well as persuasively phrased.
“Trust me,” it seems to say.
4
A call for proactive risk management in Generative AI
Trust is not an inherent quality of AI but instead the product
of AI governance, risk mitigation, and the intentional alignment
of people, processes, and technologies across the enterprise.
The trustworthiness of Generative AI
depends on how an organization uses it,
and as enterprises wade into this fast-moving
field of AI, there are factors of trust and ethics
that should be addressed.
5
A call for proactive risk management in Generative AI
1| Managing hallucinations
and misinformation
A Generative model references its dataset to There is also the risk of inherent bias within
concoct coherent language or images, which is the models, owing to the data on which they are
part of what has startled and enticed early users. trained. No single company can create and curate
With natural language programs, while the all of the training data needed for a Generative
phrasing and grammar may be convincing, AI model because the necessary data is so
the substance may well be partially to expansive and voluminous, measured in tens of
entirely inaccurate, or sometime, when terabytes. Another approach then is to train the
representing a statement of validity, model using publicly available data, which injects
false. One of the risks with this kind of natural the risk of latent bias and therefore the potential
language application is that it can “hallucinate” for bias in the AI outputs.
an inaccurate output in complete confidence.
It can even invent references and sources that A fundamental risk is that users may place
are non-existent. The model would be forgiven complete confidence in erroneous or
as its function is to generate digital artifacts that biased outputs and make decisions and
look like human artifacts. Yet, coherent data take actions based on a falsehood. One
and valid data are not necessarily the same, way to help mitigate this risk is through AI
leaving end users of large language models to governance, and many of the leading practices
contend with whether an eloquent output is associated with other kinds of AI also apply
factually valuable at all. to generative models: workforce upskilling,
waypoints for decision making across the
AI lifecycle, structured oversight, ubiquitous
documentation, and the many other activities
that promote Trustworthy AI™.
36
A call for proactive risk management in Generative AI
2|The matter of attribution
Generative AI outputs align with the original How do we contend with attribution when a
training data, and that information came from tool is designed to mimic human creativity by
the real world, where things like attribution and parroting back something drawn from the data
copyright are important and legally upheld. it computes? If a large language model outputs
Data sets can include information from online plagiarized content and the enterprise uses that
encyclopedias, digitized books, and customer in their operations, a human is accountable
reviews, as well as curated data sets. Even when the plagiarism is discovered, not
if a model does cite accurate source the Generative AI model. Recognizing
information, it may still present outputs the potential for harm, organizations may
that obscure attribution or even tread implement checks and assessments to help
across lines of plagiarism and copyright ensure attribution is appropriately given. Yet,
and trademark violations. if human fact-checking of AI attribution
becomes a laborious process, how much
productivity can the enterprise actually
gain by using Generative AI?
Finding the balance between trust in attribution
and human oversight will be an ongoing
challenge, with significant legal and brand
implications for the enterprise.
37
A call for proactive risk management in Generative AI
3|Real transparency
and broad user explainability
End users can include people who have Enterprise-wide AI literacy and risk awareness is
limited understanding of AI generally, much becoming an essential aspect of any company’s
less the complicated workings of large day-to-day operations. This is perhaps even
language models. The lack of a technical more important with Generative AI. Business
understanding of Generative AI does not users should have a real understanding of
absolve the organization from focusing on Generative AI because it is the end user
transparency and explainability. If anything, (and not necessarily the AI engineers and data
it makes it that much more important. scientists) who contends with the risks
and the consequences of trusting a tool,
Today’s Generative AI models often come regardless of whether they should. To promote
with a disclaimer that the outputs may be the necessary AI understanding, CIOs and
inaccurate. That may seem like transparency, business leaders may look to existing workforce
but the reality is many end users do not training and learning sessions, explanatory
read the terms and conditions, they do not presentations to business users, and fostering
understand how the technology works, and an enterprise culture of continuous learning.
because of those factors, the large language
model’s explainability suffers. To participate in
risk management and ethical decision making,
users should have accessible, non-technical
explanations of Generative AI, its limits and
capabilities, and the risks it creates.
38
A call for proactive risk management in Generative AI
Accountability
on the road ahead
Even as Generative AI becomes better able No matter how powerful it becomes, we
to mimic human creativity, we should still need the analysis, scrutiny, context
remember and carefully consider the awareness, and the humanity of people at
human side of this equation. Everyone the center of our AI endeavors.
will be affected by Generative AI in one
way or another, from outsourced labor to This AI era is the Age of
layoffs, changing professional roles, and even
potentially legal issues. Generative AI will have With™, where humans
real impact, and because an AI model has
work with machines to
no autonomy or intent, it cannot be held
accountable in any meaningful sense.
achieve something neither
At scale, the possibility of transparency
could do independently.
with Generative AI becomes elusive and
“keeping the human in the loop” becomes Now is the time to
a growing problem. It is also unclear at this
derive viable methods
point the degree of consequences that may
result from mass adoption of Generative AI,
of accountability, trust,
such as the proliferation of fake facts to the
detriment of objective and complete truth. and ethics, linking the
These challenges are unlikely to hinder
Generative AI product
Generative AI’s adoption.
and its outcomes with its
creator, the enterprise.
9
A call for proactive risk management in Generative AI
Reach out for a conversation.
Beena Ammanath Wessel Oosthuizen Dr. Kellie Nuttall Jefferson Denti Audrey Ancion Jan Hejtmanek Roman Fan Anne Sultan
Global Deloitte AI Institute Deloitte AI Institute Africa, Deloitte AI Institute Deloitte AI Institute Brazil, Deloitte AI Institute Deloitte AI Institute Deloitte AI Institute China, Deloitte AI Institute France,
Leader Lead Australia, Lead Lead Canada, Lead Central Europe, Lead Lead Lead
Deloitte AI Institute Deloitte Africa Deloitte Australia Deloitte Brazil Deloitte Canada Deloitte Central Europe Deloitte China Deloitte France
United States, Lead [email protected] [email protected] [email protected] [email protected] [email protected] [email protected] [email protected]
Deloitte Consulting, LLP
[email protected]
Dr. Bjoern Bringmann Prashanth Kaddi Masaya Mori Nicholas Griedlich Naser Bakhshi Tiago Durao Sulabh Soral
Deloitte AI Institute Deloitte AI Institute India, Deloitte AI Institute Japan, Deloitte AI Institute Deloitte AI Institute Deloitte AI Institute Deloitte AI Institute
Germany, Lead Lead Lead Luxembourg, Lead Netherlands, Lead Portugal, Lead United Kingdom, Lead
Deloitte Germany Deloitte India Deloitte Japan Deloitte Luxembourg Deloitte Netherlands Deloitte Portugal Deloitte United Kingdom
[email protected] [email protected] [email protected] [email protected] [email protected] [email protected] [email protected]
10
This publication contains general information only and Deloitte is not, by means of this publication, rendering
accounting, business, financial, investment, legal, tax, or other professional advice or services. This publication is not
a substitute for such professional advice or services, nor should it be used as a basis for any decision or action that
may affect your business. Before making any decision or taking any action that may affect your business, you should
consult a qualified professional advisor.
Deloitte shall not be responsible for any loss sustained by any person who relies on this publication.
About Deloitte
Deloitte refers to one or more of Deloitte Touche Tohmatsu Limited, a UK private company limited by guarantee
(“DTTL”), its network of member firms, and their related entities. DTTL and each of its member firms are legally
separate and independent entities. DTTL (also referred to as “Deloitte Global”) does not provide services to clients.
In the United States, Deloitte refers to one or more of the US member firms of DTTL, their related entities that
operate using the “Deloitte” name in the United States and their respective affiliates. Certain services may not be
available to attest clients under the rules and regulations of public accounting. Please see www.deloitte.com/about
to learn more about our global network of member firms.
Copyright © 2023 Deloitte Development LLC. All rights reserved. |
195 | deloitte | us-ai-institute-scaling-GenAI-final (1).pdf | Scaling Generative AI
13 elements for sustainable growth and value
Scaling Generative AI | 13 elements for sustainable growth and value
About the Deloitte AI InstituteTM
The Deloitte AI Institute helps organizations connect
the different dimensions of a robust, highly dynamic
and rapidly evolving AI ecosystem. The AI Institute leads
conversations on applied AI innovation across industries,
with cutting-edge insights, to promote human-machine
collaboration in the “Age of With”.
The Deloitte AI Institute aims to promote a dialogue
and development of artificial intelligence, stimulate
innovation, and examine challenges to AI implementation
and ways to address them. The AI Institute collaborates
with an ecosystem composed of academic research
groups, start-ups, entrepreneurs, innovators, mature AI
product leaders, and AI visionaries, to explore key areas
of artificial intelligence including risks, policies, ethics,
future of work and talent, and applied AI use cases.
Combined with Deloitte’s deep knowledge and experience
in artificial intelligence applications, the Institute helps
make sense of this complex ecosystem, and as a result,
delivers impactful perspectives to help organizations
succeed by making informed AI decisions.
No matter what stage of the AI journey you’re in; whether
you’re a board member or a C-Suite leader driving
strategy for your organization, or a hands on data
scientist, bringing an AI strategy to life, the Deloitte AI
institute can help you learn more about how enterprises
across the world are leveraging AI for a competitive
advantage. Visit us at the Deloitte AI Institute for a
full body of our work, subscribe to our podcasts and
newsletter, and join us at our meet ups and live events.
Let’s explore the future of AI together.
www.deloitte.com/us/AIInstitute
2
Scaling Generative AI | 13 elements for sustainable growth and value
Near the top of every enterprise agenda is a question of how to leverage Generative AI (GenAI).
With use cases proliferating horizontally across functions and vertically within business units,
the next step is...sustainably scaling GenAI for strategic business value.
Generative AI, like origami, transforms a resource
(data and paper, respectively) into a compelling
output. Just as origami artists fold paper to
resemble interesting shapes, Generative AI
computes data to approximate human cognition
and creativity.
3
Scaling Generative AI | 13 elements for sustainable growth and value
Getting more GenAI into production
Deloitte’s State of GenAI in the Enterprise Q3
report revealed that many businesses are
encountering challenges when making the
transition from GenAI proof-of-concept to
scaled deployment.1
Seventy percent of surveyed organizations indicate that less than one third of their
GenAI experiments have made it to production. This suggests that while enterprises are
investing in GenAI, they are not yet seeing the full potential ROI. A common challenge is
defining what is required to achieve GenAI scale at a practical level.
We define scale broadly as the ability of a system to handle a growing amount of
work or its potential to be enlarged to accommodate growth with steadily
decreasing unit costs. For GenAI specifically, scaling also means moving from
experimentation to implementation in a way that is sustainable, secure, and
aligned with business goals.
GenAI at scale generates more diverse and representative outputs, it can handle more
complex tasks, and its speed, output quality, and accuracy are enhanced. As a result,
operational costs become more efficient and business impact is governed, measured,
and communicated.
4
Scaling Generative AI | 13 elements for sustainable growth and value
At the highest level, GenAI scaling factors can be grouped into the familiar areas of strategy, process, talent, and
data and technology. Each area presents challenges to be navigated and contains leading practices that help point the way
to GenAI value realization.
Strategy Process Talent Data and Technology
Ambitious Modular
Integrated Transparency Provisioning
strategy and value Robust architecture
risk to build trust the right AI
management governance and common
management in secure AI infrastructure
focus platforms
Clear, Agile Acquiring Effective
Strong Transformed
high-impact operating model (external) and Modern data model
ecosystem roles, work,
use case and delivery developing foundation management
collaboration and culture
portfolio methods (internal) talent and operations
Essential elements for scaling Generative AI initiatives from pilot to production
5
Scaling Generative AI | 13 elements for sustainable growth and value
STRATEGY
Ambitious strategy and
value management focus
An organization’s GenAI strategy and vision need to be comprehensive, integrated with
broader business objectives, and aligned with other existing AI programs. Executive
buy-in and a top-down mandate are essential for aligning functions and decision-making.
Leadership sets priorities and strategy, and without an executive mandate, it is difficult
to coordinate change across multiple teams. A cohesive GenAI strategy defines business
objectives, sets measurable goals, identifies valuable areas for application, and measures
realized value. As a part of strategy development, it’s important to show progress against
short-term goals and inform any iterative improvements needed to the strategy.
Establish a
comprehensive vision
with a top-down mandate
6
Scaling Generative AI | 13 elements for sustainable growth and value
STRATEGY
Clear, high-impact use
case portfolio
There are six common macro archetypes for GenAI: Q&A-based search, summarization,
content generation, content transformation, virtual agent, and code generation. In seeking
value-driving applications, organizations should look across the archetypes for low-
barrier, high-impact use cases for core business domains. These drive efficiencies and
savings that can be reinvested in innovation. Other high-impact use cases may be more
transformational and differentiating with enterprise-wide applicability. Whether deploying
a proven application or striving for something novel, all applications require technical
feasibility and a viable business case. What is more, existing processes will likely need to be
reimagined to incorporate and leverage the capabilities of GenAI use cases in workflows.2
At its core, the use case portfolio needs to be focused on answering business questions
and meeting quantified goals. We see leading organizations create business cases that
weave together the value GenAI can provide to multiple teams, rather than evaluating
the value of individual applications. This is done most effectively by running a number
of use cases in parallel. It makes efficient use of resources and allows for rapid portfolio
management should a specific use case prove less compelling without sacrificing
momentum of the overall Gen AI portfolio.
Explore low-barrier, high-
impact use cases to drive
efficiencies and savings
7
Scaling Generative AI | 13 elements for sustainable growth and value
STRATEGY
Strong ecosystem
collaboration
GenAI is maturing rapidly, with existing providers and new market entrants alike driving
capabilities and lateral applications. The array of GenAI solutions and the speed with which
they are evolving can make it challenging to select the appropriate tools and platforms
that enable enterprise strategy. To reach target outcomes, enterprise leaders should build
strategic relationships with technology and data ecosystem stakeholders and keep pace
with GenAI development. By monitoring elements like product roadmaps, total cost of
ownership, and labor delivery models, business leaders can gain an understanding as
to how their GenAI programs should evolve and how ecosystem players can accelerate
progress and results as strategic partners, rather than as transactional vendors. Use a
framework to support a structured approach to evaluating solutions based on factors such
as data readiness, AI maturity, risk appetite, and total cost of ownership.
Evolve with existing
providers and new market
entrants alike
8
Scaling Generative AI | 13 elements for sustainable growth and value
PROCESS
Robust governance
Inconsistent processes can lead to risks and inefficiencies, while consistent governance
processes help standardize workflows for data collection, solution engineering, output
validation, and performance monitoring. Common delivery frameworks (e.g., LLMOps)
bring together GenAI development and deployment into a unified, governed lifecycle that
is secure and compliant. A common misconception is that strong processes can hinder
speed and creativity. Our experience suggests the opposite. By understanding how work
needs to be done and the accompanying guardrails, teams are empowered to explore
ways to generate value without fear that they may be making a mistake. Clear boundaries
allow freedom for bold action and innovation, while a lack of clarity may lead to more
conservative approaches. Governance includes documented roles and responsibilities
that drive stakeholder accountability in decision-making across the AI lifecycle, and
inform the controls for risk identification and mitigation. Governance also standardizes
how stakeholders identify, prioritize, and approve GenAI applications. As processes are
amended, organizations need to be mindful about disrupting existing automated or
manual controls and take steps to establish assurance in those amended processes.
Even as the regulatory landscape is in flux, organizations should proactively establish
governance processes that meet existing or likely regulatory requirements.
Create repeatable
governance processes to
help standardize work
9
Scaling Generative AI | 13 elements for sustainable growth and value
PROCESS
Integrated risk management
For GenAI to reach its full potential business value and adoption, it must be trusted
and secure.3 Attempting to scale without accounting for trust in data and the machine
that consumes it can have implications for regulatory compliance, finance and strategy,
cybersecurity and privacy, adoption and change management, and brand reputation—
the consequences of which can limit or even erase GenAI’s intended value. Risk and
trust need to be considered and addressed across the GenAI lifecycle, from design and
development through deployment and scaled implementation. This includes validation
processes and feedback loops for human oversight to manage solution performance
and accuracy. It also includes guardrails to ensure privacy, drive ongoing compliance, and
promote agility in proactively responding to emerging risks. Data security is particularly
essential. Differentiated GenAI applications are fueled by sensitive, proprietary enterprise
data. Thus, training and usage can potentially expose or leak business-critical data and
create risks to the organization. This is not a one-time event—organizations must make
this part of regular work, rather than a separate consideration.
Address risk and data
security across the
GenAI lifecycle
10
Scaling Generative AI | 13 elements for sustainable growth and value
PROCESS
Agile operating model
and delivery methods
The operating model impacts how the enterprise aligns technology, processes, and roles
and responsibilities to create strategic business value. An integrated model connects the
blueprint for value with AI business cases to inform how work is delivered and helps drive
alignment across the enterprise. As the marketplace matures and new capabilities and
risks impact AI lifecycles and governance, the organization needs to be agile in matching
internal opportunities with the right technologies. To help, organizations may turn to
technical experts or an AI Center of Excellence (COE) that equips decision makers with
the insight to align the vision for success with the organization’s AI maturity and ambition.
This supports a cohesive approach to orchestrating the elements of GenAI development
and application. It helps avoid AI and data silos and instead drive toward reusable
building blocks, coordinated sourcing strategy, informed build-versus-buy decisions, and
security and risk management.
Support a cohesive
approach to orchestrating
the components
11
Scaling Generative AI | 13 elements for sustainable growth and value
TALENT
Transparency to build trust
in secure AI
Trust in GenAI is essential to increasing workforce adoption and realizing benefits. With
GenAI, employees may have existing biases, inhibitions, skills gaps, or even a fear that they
could be replaced by a machine. Trust in GenAI grows out of transparency, where every
stakeholder understands how the enterprise is pursuing GenAI applications, how they are
intended to create value, and how the workforce can leverage these tools as efficiency and
productivity enhancers. Transparency around the benefits targeted by GenAI solutions
helps correct misinformation and creates an opportunity to improve the workforce
experience. Trust is also important for external stakeholders, third parties, and customers,
and a transparent approach to GenAI use includes consent for data collection, notification
of how GenAI outputs may impact users, and documentation across the AI lifecycle to
inform audits and compliance.
Help stakeholders understand
the GenAI vision and how it
creates value for them
12
Scaling Generative AI | 13 elements for sustainable growth and value
TALENT
Transformed roles, work,
and culture
Deployments at scale can disrupt the status quo, transforming employee responsibilities and
how work is accomplished. As an enterprise reimagines strategy, processes, and technology
to drive GenAI value, the workforce needs to be brought on the journey as value is created
through individuals doing work differently. Organizations should nurture adoption by
documenting and communicating responsibilities and process amendments to workflows.
Poor communications may cause misunderstanding about GenAI’s potential and limitations,
leading to unrealistic expectations or resistance. Conversely, effective communications
align stakeholders around the same vision for scale and value, including as they relate to
governance, policy, IT security, risk, and funding. Topics to communicate include outcomes
and lessons learned, the organization’s AI roadmap, the impact on end users (e.g., customers
or employees), and guidance to the workforce on how to balance day-to-day tasks with
AI skills development. Ongoing adoption should be measured to identify optimization
opportunities and internal leading practices. This should inform the overall use case roadmap
and activation strategy. Simply put, upstream conversations should take place before
continuing to build technical solutions that are underdelivering against expectations.
Nurture adoption by
documenting responsibilities
and process amendments
13
Scaling Generative AI | 13 elements for sustainable growth and value
TALENT
Acquiring (external) and
developing (internal) talent
Organizations deploying GenAI need to consider the skilled human talent required
across the GenAI lifecycle. Skills mapping can reveal where the enterprise needs to
expand or improve the workforce. Recruiting new talent is one avenue, such as by
attracting new employees from educational facilities (e.g., universities). In reimagining
work with GenAI, the organization may attract new leaders who are eager to use
technology to deliver business value, as well as top talent seeking opportunities to
learn and develop. Yet, most of a company’s GenAI capabilities will grow out of training
and upskilling existing employees, and as GenAI touches every part of the enterprise,
the entire workforce requires training to adopt and use it. To this end, businesses may
create overall AI literacy programs, training plans tailored to employee personas (e.g.,
technical, functional, sales, marketing, etc.), and opportunities (e.g., hackathons and
digital playgrounds) for employees to apply new knowledge and build competence in
GenAI application, management, and monitoring. A GenAI COE can help orchestrate
human-centered continuous learning to promote adoption.
Balance talent
acquisition with
workforce upskilling
14
Scaling Generative AI | 13 elements for sustainable growth and value
DATA & TECHNOLOGY
Modular architecture and
common platforms
IT architecture needs to evolve as technologies mature and as the organization’s needs
change. Flexibility in modular systems includes leveraging microservices and APIs
(Application Programming Interfaces) for tech stack integration, as well as techniques for
improving output reliability (e.g., retrieval augmented generation, fine-tuning). This enables
platform and model “lift and shift” and supports partnerships with hyperscalers that can
provision on-prem or cloud-based environments via contracts that reward increased
volume with lower unit costs. In prioritizing a modular architecture, organizations can
facilitate user growth with a cost-per-user model, automate guardrails for managing GenAI
risk, leverage GenAI capabilities in enterprise software platforms, and establish an internal
marketplace where users can select models, access prompt catalogs, and leverage
existing solutions. Modular architecture and delivery also accommodate low-code
platforms for business users and provide a clear pathway to industrializing capabilities.
Prioritize a flexible
IT architecture to
facilitate enhancements
15
Scaling Generative AI | 13 elements for sustainable growth and value
DATA & TECHNOLOGY
Provisioning the right
AI infrastructure
GenAI infrastructure includes reusable assets, data pipelines, solution development
environments, and a range of post-deployment management and feedback capabilities.
Bringing the right secure infrastructure to the right place in the GenAI value chain is
necessary for sustainable, cost-effective scale. Taking an AI Factory approach enables
reusable components and data products while also integrating sourcing strategy,
cybersecurity considerations, demand generation, prioritization, governance, and business
outcomes. While focusing on speed to value and taking an agile, incremental approach
to infrastructure development, organizations can look to iterative design and continual
evaluation of cost mechanisms against a per-user or per-use model. One important
consideration is that executives are likely to be more comfortable funding enhancements
to existing capabilities, as opposed to building net-new systems. Using existing investments
and approaching scale as building incremental capabilities can help encourage investments
by overcoming a misperception that a GenAI endeavor is starting from scratch.
Take an agile
approach to enable
continuous improvement
16
Scaling Generative AI | 13 elements for sustainable growth and value
DATA & TECHNOLOGY
Modern data foundation
As organizations increasingly shift to hybrid-cloud environments, data integration challenges
may increase, with proprietary and third-party data sources existing on disparate platforms.
In addition to master data, GenAI applications consume other forms of data (e.g., reference,
unstructured first-party) that traditionally sit in the realm of knowledge management. Value
creation opportunities from GenAI are blending knowledge and data management capabilities.
Data quality and accessibility issues can limit value and potentially create a perception that
scaled solutions are not viable nor valuable. A GenAI-ready data foundation includes the
processes, philosophies, approaches, and approvals for data sharing and use. As a part of this,
evaluate the organization’s data findability, accessibility, interoperability, reusability, and storage.
Rather than starting from scratch, the organization’s existing data governance efforts can likely
be extended and adjusted to accommodate unstructured data. Data should also be curated
and integrated across departmental lines. Consider a parallel workstream for data readiness
evaluation and progression focused on clean and organized data, efficient data pipelines,
and robust data governance practices. By ensuring systems are secure and foundational
data capabilities are aligned with the GenAI strategy and governance, enterprises can evolve
data availability, engineering, and management to enable adoption and scale. At the same
time, it is worth noting that interim value can be harvested, albeit at a lower potential, while
comprehensive and foundational data modernization activities are underway.
Align data capabilities and
processes with GenAI strategy to
support quality and accessibility
17
Scaling Generative AI | 13 elements for sustainable growth and value
DATA & TECHNOLOGY
Effective model management
and operations
Trustworthy, compliant GenAI applications require coordinated solution management,
including continuous monitoring for impartial output accuracy, waypoints for decision-
making, and data feedback loops for continuous improvement. Cost management is also a
factor. GenAI deployment raises questions around variable and fixed costs, and business
leaders need visibility into managing and forecasting end-to-end costs for infrastructure,
tools, personnel, maintenance, and models. Insourcing key functions may permit
differentiation or better economics over time, and insourcing decisions need to be balanced
against the cost to build a capability, the ramifications of moving to a fixed versus variable
cost, and the expenses associated with capability management (e.g., hiring and training,
oversight, technology acquisition, facilities).
Monitor for impartial
output accuracy and focus
on cost management
18
Scaling Generative AI | 13 elements for sustainable growth and value
Measuring success with GenAI at scale
The value of scaled GenAI deployments is found
in how they advance an integrated enterprise
strategy and drive toward business goals.
Establishing realistic goals for quantitative KPIs (beyond productivity and efficiency metrics,
such as hours saved) allows the enterprise to assess whether the scaled deployment is
achieving its intended business impact. With a use case portfolio that balances cost- and
revenue-oriented value levers, there are key indicators that reveal whether the
enterprise is on the right track:
• Increased speed to market, from ideation to deployment
• A decline in proof-of-concept demand, as demand shifts to low-code environments
available to business users
• A decrease in unit cost for new capabilities/solutions, with technical solutions and code
being reusable, thus reducing development efforts
• An increase in the number of foundational capabilities that help the organization access
GenAI advancements as they emerge
• An increase in domain-specific models allowing for more use cases and broader
application across the organization
• Increased use of capabilities and solutions, owing to a growing number of users in the
enterprise
• An increase in stated value realization on a cumulative basis due to GenAI
• An increase in internal certification/badging of existing employees in GenAI capabilities,
both functional and technical
• Use of GenAI to redefine a business process, rather than embedding GenAI in existing
business processes
19
Scaling Generative AI | 13 elements for sustainable growth and value
GenAI capabilities are improving and multiplying,
and at this point, few organizations are likely to
have achieved each element of scale to their
greatest capacity. The leading practices, governed
processes, and ecosystem of complementary
technologies are still being developed and defined.
While change is inevitable, pursuing the
elements of scale today positions the
organization to go live with GenAI for
business value as this transformative
technology evolves.
20
Scaling Generative AI | 13 elements for sustainable growth and value
Let’s connect
Reach out for a conversation on scaling Generative AI
Lou DiLorenzo Jr. Edward Van Buren Rohit Tandon
US AI & Data Strategy Government & Public Services US AI & Insights
Practice Leader Leader – Applied AI Practice Leader
US CIO & CDAO Programs Deloitte Consulting LLP Deloitte Consulting LLP
Executive Sponsor [email protected] [email protected]
Deloitte Consulting LLP
[email protected]
Acknowlegements
The authors would like to thank the following leaders and colleagues for their contributions to this effort.
Kevin Abraham, Beena Ammanath, Aniket Bandekar, Kevin Byrne, Ricky Franks, Justin Hienz, Kevin Hutchinson, David Jarvis,
Carissa Kilgour, Lena La, Geoff Lougheed, Parth Patwari, Brittany Rauch, Jim Rowan, Kristin Ruffe, Baris Sarer, Dean Sauer, Laura
Sangha Pati Aditya Kudumala Jenn Malatesta
Shact, Brenna Sniderman, Ian Thompson, and Saurabh Vijayvergia.
USI AI & Insights Life Sciences Global AI Leader Commercial Officer
Practice Leader Deloitte Consulting LLP Deloitte & Touche LLP Endnotes
Deloitte Consulting LLP [email protected] [email protected]
1 Jim Rowan, Beena Ammanath, Brenna Sniderman et al, “Now decides next: Moving from potential to performance, Deloitte’s
[email protected]
State of Generative AI in the Enterprise,” Quarter three report Deloitte, August 2024.
2 Rowan, Ammanath, Sniderman et al, “Now decides next.”
3 Deloitte, “TrustworthyAITM, Bridging the ethics gap surrounding AI,” accessed 3 October 2024.
21
Deloitte refers to one or more of Deloitte Touche Tohmatsu Limited, a UK private company limited by guarantee, and its
network of member firms, each of which is a legally separate and independent entity. Please see www.deloitte.com/about
for a detailed description of the legal structure of Deloitte Touche Tohmatsu Limited and its member firms. Please see
www.deloitte.com/us/about for a detailed description of the legal structure of Deloitte LLP and its subsidiaries. Certain
services may not be available to attest clients under the rules and regulations of public accounting.
This publication contains general information only and Deloitte is not, by means of this publication, rendering accounting,
business, financial, investment, legal, tax, or other professional advice or services. This publication is not a substitute
for such professional advice or services, nor should it be used as a basis for any decision or action that may affect your
business. Before making any decision or taking any action that may affect your business, you should consult a qualified
professional advisor. Deloitte shall not be responsible for any loss sustained by any person who relies on this publication.
Copyright © 2024 Deloitte Development LLC. All rights reserved. |
196 | deloitte | us-artificial-intelligence-and-mergers-and-acquisitions.pdf | Artificial intelligence and
mergers and acquisitions
Observations from the frontlines and how
to prepare for the coming shift
Artificial intelligence and mergers and acquisitions: Observations from the frontlines and how to prepare for the coming shift
Artificial intelligence (AI) has a powerful new variant,
Generative AI (GenAI). GenAI shows promise as game-
changing technology given its combination of novel features,
accessibility by nontechnical users, and scalability across an
enterprise. This combination of traits has the potential to un-
lock new sources of value across the enterprise.
Unsurprisingly, organizations want to know what this all 4. This is only the beginning, and when it comes to
means for dealmaking and, most importantly, how to GenAI, winners will be defined by their ability to
realize its value to mergers and acquisitions (M&A). navigate early challenges and identify the right
choices needed to win.
In an M&A context, it is easy to see how GenAI could
create a competitive edge for early adopters. Its ability to What about the second question: How can enterprises
ingest, interpret, and summarize significant quantities of capitalize on this opportunity? Given the indications
data; automate manual and labor-intensive processes; above, we recommend serious consideration of the
and uncover new insights and questions are all potential following actions:
avenues for enhancing returns during M&A. The
1. Stand up or strengthen sensing capabilities using
opportunities are numerous, but there is a clear risk:
internal and external resources, to keep a pulse on
Leaders who choose to defer action may lose ground to
AI and GenAI activity, considering direct and indirect
those who seize first-mover advantage.
competitors and partners.
Before diving in headfirst, M&A leaders and executives
2. Recast the M&A strategy by taking into consideration
should ask: How will GenAI affect M&A, and how can
how AI and GenAI might affect existing value chains
we capitalize on this opportunity?
and opportunities to capitalize on disruption and drive
To answer the first question, four key predictions are greater growth and value creation throughout the
well-founded: portfolio.
1. Deals will increasingly focus on (i) the acquisition of 3. Identify and invest in experts that can help
AI- and GenAI – capabilities, assets, and data, and (ii) validate and amplify AI and GenAI opportunities
the divestiture of business models vulnerable to and that bring a blend of commercial, operational, and
AI disruption. technical perspective.
2. Meaningful application of GenAI will enhance the 4. Prioritize and test AI use cases to develop a deeper
M&A process across the entire life cycle, improving understanding of capabilities and limitations and to aid
speed, quality of insights, and financial outcomes with identifying the most promising opportunities to
during execution. implement across the enterprise.
3. GenAI will continue to gain momentum in M&A as
early adopters employ it as a key lever to create value With the stage set, let us dive into the details.
from the top line to “heart of the business” functions.
2
Artificial intelligence and mergers and acquisitions: Observations from the frontlines and how to prepare for the coming shift
Four key predictions
1. Deals will increasingly focus on (i) the acquisition and look for ways to further exploit opportunities.
of AI- and GenAI – capabilities, assets, and data, and Similarly, for enterprises that find the basis of their
(ii) the divestiture of business models vulnerable to competition has fundamentally changed, some may
AI disruption. elect to divest noncore businesses to reinvest in new
and differentiated capabilities.
Private equity and strategics are increasingly buying AI
and GenAI capabilities. According to Crunchbase, GenAI 2. Meaningful application of GenAI will enhance the
and AI startups raised almost $50 billion in 2023, a 9% M&A process across the entire life cycle, improving
increase over 2022 levels.1 Furthermore, we see AI and speed, quality of insights, and financial outcomes
GenAI deal activity across nearly all sectors. See below during execution.
for recent industry examples:
With its ability to digest large quantities of data,
• Technology: Acquisition of AI based capabilities to synthesize and summarize findings quickly, develop
enhance AI Business offerings to improve customer quantitative and qualitative analyses, provide
experience and productivity. recommendations and predictions based upon pattern
recognition, and refine outputs through deep learning,
• Life sciences: Acquiring products from a clinical-stage
AI and GenAI can make an impact across the full M&A
drug discovery firm, which uses AI for a proprietary
life cycle.
drug discovery engine.
To date, much of the focus has been earlier in the
• Insurance tech: Acquisition of AI-driven cyber risk
life cycle. This is likely driven by companies starting to
analytics and GenAI-enhanced underwriting and
apply it where they are most comfortable and feel the
quoting.
least risk. Key examples include using AI and GenAI to
evaluate markets, products, and technologies to inform
• Oil and gas: M&A and investment in AI-enabled digital
strategies, identify gaps or vulnerabilities in product
models that can increase operational efficiency by
portfolios, and prioritize targets that fill those gaps. In
enhancing reservoir characterization.
fact, several private equity funds are already engaged
As these acquisitions show, AI and GenAI are a rising in exercises to understand how GenAI could have an
trend in multiple sectors. Goldman Sachs Research impact on their portfolio of investments.
predicts that AI investment could approach $200 billion
globally by 2025,2 which is likely to set the stage for AI is also helping to identify targets in a novel way. In
place of typical screening tools and criteria, AI and
future business strategies in an increasingly AI-driven
machine learning (ML)-enabled screening tools help
global market.
uncover previously “hidden” options by presenting new
Additionally, there has been a significant increase in targets that resemble their short list of top targets.
private equity deal activity focused on AI and GenAI, In fact, one client is training AI on what “successful”
considering rapid growth potential. To avoid being left portfolio companies look like—without defining inputs
behind, industry leaders will need to create or defend or outputs. Instead, the AI has established its own
competitive advantage with new investments in assets, criteria to uncover targets with a higher probability of
data, and AI capabilities. We are already seeing the value capture. Lastly, GenAI is helping clients review and
first signs of disruption across multiple sectors, which summarize critical supplier or customer contracts to
creates opportunities for the disrupters and potentially inform execution, integration, or separation strategies.
significant challenges for the disrupted.
Figure 1 shows the five key stages of the M&A life cycle
As the promise of GenAI becomes more real, new and associated key use cases identified by Deloitte.
entrants will pose competitive threats for incumbents
3
Artificial intelligence and mergers and acquisitions: Observations from the frontlines and how to prepare for the coming shift
Figure 1: AI and GenAI use cases across the M&A life cycle
Examples of AI and GenAI M&A use cases
• Product portfolio analysis: Assess product mix and recommend growth strategies
• Nested wargaming: Game out competitive moves and counter-moves by a key competitor
Target
• (Un)structured data analysis: Automated data extraction and transference of deal financial data into analytics and tools
screening/M&A
• Market sensing and analytics in PE: Identify and summarize PE investment thesis, leveraging market research and
strategy & planning
valuation • Deal sourcing/target screening: Identify and prioritize high performing assets for future deals to correlate with investment
strategy
• Valuation global standards chatbot: GenAI Engine that answers valuation questions based on global standards
• Diligence observations/focus: Evaluate financial and operational data to identify key risks
• Management interview preparation: Prepare management interview guides and additional data requests
• CDD prep and voice of customer execution: Examine target customer segments, preliminary trends
and summarize themes
Due
• Functional due diligence (HR, IT, Ops, etc.): Analyze and compare HR practices and policies (e.g., leaves, severance)
diligence
• Culture diligence: Use public data sources to gather information on the target company’s culture
• Management EBITDA drafting: Create management adjusted EBITDA build and draft description of adjustments
• Report tie-out: Compare draft report with finance workbook schedules, and note where values do not reconcile
• Working capital optimization: Generate insights about payment terms for customer/vendors and working capital
Negotiations & • Term sheet analysis: Analyze and summarize key agreement and financing terms
deal structure • Deal closing conditions: Draft Day 1 criteria and closing conditions based on sell-and buy-side objective
• Blueprinting, Day 1 checklists, TSAs: Automate operating model designs and draft of Day 1 planning deliverables
• Benchmarking analysis: Prepare benchmarking of financial and operational KPIs
• IT landscape analytics: Generate summary of comparison between seller and buyer applications, infrastructure and IT
Post-deal services
planning & • Contract analysis: Encapsulate key terms in contracts for contract harmonization
execution • Day 1 communications: Generate Day 1 communications (e.g., deal announcements, stakeholder FAQs, and social media
posts)
• Culture analysis: Synthesize survey and focus group data and recommend actionable steps to address culture differences
• Chatbot for Day 1 support: Leverage GenAI chatbot to answer questions related to deal and Day 1 readiness
• Synergy assessment: Quantify operational and financial synergies through transformation initiatives
Restructuring &
• Value creation and synergies: Analyze value creation levers, and prioritize cost savings initiatives
transformation
• Critical path management: Develop and track critical path for transformation and value realization
4
laed-erP
laed-tsoP
Artificial intelligence and mergers and acquisitions: Observations from the frontlines and how to prepare for the coming shift
While much of the focus has been on the earlier stages As summarized in figure 2, the degree of readiness,
of the life cycle, companies will build on early learnings value potential, stakeholder engagement, and security
and apply those learnings to reduce the risk and reaches its greatest strength at various parts of the
increase the value during downstream M&A activities. M&A life cycle.
Figure 2: Focus areas for GenAI in M&A based on Deloitte’s sponsored survey of M&A executives
Target Initial Detailed Integration/
identification assessment due diligence separation planning
Lower readiness
Market
readiness
Higher readiness
Lower value
creation potential
M&A value
potential
Higher value
creation potential
Limited stakeholder
engagement
Stakeholder
engagement
Extensive stakeholder
engagement
Lower security
concern
Security
Higher security
concern
Early developments in GenAI have created a higher readiness in early M&A life cycle use
cases. While there may be higher value creation potential in later life cycle use cases,
there are also additional considerations with stakeholder engagement and security.
Note: Figure 2 includes responses from a brief study conducted with M&A executives across industries.
5
Artificial intelligence and mergers and acquisitions: Observations from the frontlines and how to prepare for the coming shift
The costs, risks, and rewards of building and executing The question is not whether GenAI will affect M&A, but
GenAI use cases are still taking shape. But the instances rather at what pace? The technology’s potential to recast
that produce meaningful return on investment in the look and feel of dealmaking is significant, but several
the form of better insights, increased productivity, challenging headwinds must be navigated to bring that
and accelerated execution will emerge as real potential to fruition.
differentiators—and likely pave the way for applications
GenAI suffers from hallucinations: making incorrect
later in the M&A life cycle.
inferences from its source data that may seem
3.GenAI will continue to gain momentum in M&A correct. As with any tool, results and quality must
as early adopters employ it as a key lever to create be validated. GenAI is likely to open a gap and lag in
value from the top line to “heart of the business” understanding and development for the average or early
functions. career employee. Additionally, regulatory and ethical
complexities continue to evolve and at a seemingly
While we anticipate acquisitions of AI- and GenAI-
slower pace than AI. We also see access to or ownership
augmented business will continue to be a focus, we also
of large, high-quality, proprietary data increasing in
see that early experimentation with AI is uncovering
importance as a source of advantage.
opportunities to improve top-line growth, reduce costs,
and minimize execution risk. In fact, a recent Deloitte Perhaps GenAI will come to differentiate M&A winners
survey3 found that 79% of CEOs believe AI will increase from laggards. On the other hand, AI technologies
efficiencies, and 52% believe AI will drive revenue growth may simply become mission-critical capabilities that all
for their enterprises. companies adopt equally—tomorrow’s analog to the
internet or electricity.
As the survey signals, top-line growth is not the only
consideration coming into focus. The associated cost Having sketched the likely developments and the
opportunities and operational benefits are becoming remaining areas of uncertainty, what should an M&A-
clearer as well. Some buyers are already incorporating oriented organization do today to prepare for an AI-
modest cost savings associated with more well-founded fueled future?
use cases such as deploying advanced chatbots to
reduce customer service costs, automating coding and
documentation tasks to lower software development
costs, or even personalizing marketing content while
trimming associated spend.
We anticipate that as buyers gain more experience,
they will likely gain confidence in their ability to estimate
and deliver the impact of these use cases. With that
confidence, they will naturally expand their AI repertoires
to more “heart of the business” functions, such as cost of
goods sold, along with back-office functions such as IT,
finance, HR, and legal.
4.This is only the beginning, and when it comes
to GenAI, winners will be defined by their ability
to navigate early challenges and identify the right
choices needed to win.
6
Artificial intelligence and mergers and acquisitions: Observations from the frontlines and how to prepare for the coming shift
From strategy to action:
Four steps to take now
Strategy and deal teams that have not yet acted should an enterprise needs, what businesses or capabilities it
consider four moves to inform their strategy. We see no longer needs, and what to acquire or divest based on
each move as a “no regrets” decision that can help that intelligence.
position a company for effective and sustainable growth
Step 3. Identify and invest in experts that can help
in parallel with the AI revolution.
validate and amplify opportunities.
Step 1. Stand up or strengthen sensing capabilities.
As companies pursue AI acquisitions, they will have to
In a world of increasing “unknown unknowns” and identify internal talent or access external expertise, or
accelerating rate of advancement, companies can benefit both, to assess AI and GenAI targets.
by formalizing their approach to sensing.
That expertise will be critical to evaluating the quality of
For example, companies should examine sources the underlying technology and impact to existing assets.
of new potential threats that may arise from GenAI Increasing the level of understanding across business
disruption. Such sources may expand the definition of leaders, including commercial and R&D will be critical
competitors to include smaller, nontraditional entrants to identifying new internal diligence leads and sourcing
and markets beyond existing products and offerings. deals.
New technological developments should be tracked
In tandem, diligence teams will need to build and employ
and “scored” to indicate level and type of impact on
a framework that evaluates the suitability of a target’s
existing businesses or relative attractiveness of new
AI capabilities and its potential as a disrupter or enabler
opportunities. Additionally, insights from secondary
of enterprise strategy. Teams will also need to evaluate
sources should be validated through firsthand knowledge
and quantify future investment needed to enable their
of developments, either via third parties or direct
strategy. This can include detailed software due diligence,
conversations with those who possess knowledge of
including evaluation of source code and data, and
emerging capabilities. Increased awareness will not only
product testing for accuracy of underlying technology.
provide valuable perspective on the current state of play
but should enhance abilities to see where things may
Step 4. Prioritize and test M&A AI use cases.
go in the future and better position companies to make
smarter bets earlier. Planning can only take you so far, and we believe that
“learning by testing” is critical in this early stage of new
Step 2. Recast M&A strategy through an AI lens.
technology adoption.
Companies should reexamine their industry structures
Companies should leverage a cross-functional team’s
and reimagine their business models through an AI lens.
perspective to assist with prioritization of AI and GenAI
use cases. Regarding prioritization, teams should
The first consideration is to understand and challenge
consider evaluation of use cases based on their customer
existing assumptions that the industry will operate in
value, business impact, feasibility, and investment needs.
the future as it does today. The second consideration
Teams will also need to establish measures of success
involves disaggregating the value chain and pushing on it:
and identify learnings that will help make use cases
Where could AI both disrupt the chain and create shifts in
effective at scale. Lastly and most importantly, teams
power? Lastly, given these dynamics, companies should
should begin to test these use cases and avoid analysis
make deliberate choices in a “where to play and how
paralysis. The key is to get started on the journey and not
to win” strategic choice framework. This approach can
to overthink which foot to make the first step with.
assist with uncovering the specific AI capabilities or data
7
Artificial intelligence and mergers and acquisitions: Observations from the frontlines and how to prepare for the coming shift
New tools, new rules
AI presents unparalleled capabilities that can super- Deloitte has seen these shifts at work as we advise and
charge productivity, identify value in novel ways, gen- serve our clients during this technological revolution. As
erate rapid insights, and assist with identifying and we have observed these trends and forces firsthand, it
mitigating risks. These capabilities are likely to rapidly seems certain that enterprises across all industries will
change the work we do today and reshape how we think have lessons ahead. They will come from experience,
about M&A. GenAI not only has the potential to change not theory—and those who learn them earliest stand to
M&A from a process standpoint, but to also influence reap the greatest benefits.
the deals we seek, the way we compete, and the sourc-
es of value we identify across the enterprise. Lastly, the
pace of adoption is increasing, and those who wait face
disruption from those who act.
8
Artificial intelligence and mergers and acquisitions: Observations from the frontlines and how to prepare for the coming shift
Endnotes
1. Gené Teare, “Global startup funding in 2023 clocks in at lowest level in 5 years,” Crunchbase, January 4, 2024.
2. Abhinandan Jain, “2023: The year AI took over investments – What to expect in 2024?,” Alltech Magazine, January 6, 2024.
3. Deloitte, “The majority of CEOs surveyed believe Generative AI will increase their organizations’ efficiencies: ‘Summer 2023 Fortune/Deloitte CEO
Survey’,” press release, July 24, 2023.
9
Artificial intelligence and mergers and acquisitions: Observations from the frontlines and how to prepare for the coming shift
Authors
Will Engelbrecht Erik Dilger
Principal Managing Director
Deloitte Consulting LLP Deloitte & Touche LLP
[email protected] [email protected]
Jeffrey Canon Sean McKenzie
Managing Director Manager
Deloitte Consulting LLP Deloitte Consulting LLP
[email protected] [email protected]
Sandeep Dasharath
Senior Manager
Deloitte Consulting LLP
[email protected]
About Deloitte
Deloitte refers to one or more of Deloitte Touche Tohmatsu Limited, a UK private company limited by guarantee, and its network
of member firms, each of which is a legally separate and independent entity. Please see www.deloitte.com/about for a detailed
description of the legal structure of Deloitte Touche Tohmatsu Limited and its member firms. Please see www.deloitte.com/us/
about for a detailed description of the legal structure of Deloitte LLP and its subsidiaries. Certain services may not be available to
attest clients under the rules and regulations of public accounting.
This publication contains general information only and Deloitte is not, by means of this publication, rendering accounting,
business, financial, investment, legal, tax, or other professional advice or services. This publication is not a substitute for such
professional advice or services, nor should it be used as a basis for any decision or action that may affect your business. Before
making any decision or taking any action that may affect your business, you should consult a qualified professional adviser.
Deloitte shall not be responsible for any loss sustained by any person who relies on this publication.
Copyright © 2024 Deloitte Development LLC. All rights reserved.
10 |
197 | deloitte | us-ai-institute-scaling-GenAI-final.pdf | Scaling Generative AI
13 elements for sustainable growth and value
Scaling Generative AI | 13 elements for sustainable growth and value
About the Deloitte AI InstituteTM
The Deloitte AI Institute helps organizations connect
the different dimensions of a robust, highly dynamic
and rapidly evolving AI ecosystem. The AI Institute leads
conversations on applied AI innovation across industries,
with cutting-edge insights, to promote human-machine
collaboration in the “Age of With”.
The Deloitte AI Institute aims to promote a dialogue
and development of artificial intelligence, stimulate
innovation, and examine challenges to AI implementation
and ways to address them. The AI Institute collaborates
with an ecosystem composed of academic research
groups, start-ups, entrepreneurs, innovators, mature AI
product leaders, and AI visionaries, to explore key areas
of artificial intelligence including risks, policies, ethics,
future of work and talent, and applied AI use cases.
Combined with Deloitte’s deep knowledge and experience
in artificial intelligence applications, the Institute helps
make sense of this complex ecosystem, and as a result,
delivers impactful perspectives to help organizations
succeed by making informed AI decisions.
No matter what stage of the AI journey you’re in; whether
you’re a board member or a C-Suite leader driving
strategy for your organization, or a hands on data
scientist, bringing an AI strategy to life, the Deloitte AI
institute can help you learn more about how enterprises
across the world are leveraging AI for a competitive
advantage. Visit us at the Deloitte AI Institute for a
full body of our work, subscribe to our podcasts and
newsletter, and join us at our meet ups and live events.
Let’s explore the future of AI together.
www.deloitte.com/us/AIInstitute
2
Scaling Generative AI | 13 elements for sustainable growth and value
Near the top of every enterprise agenda is a question of how to leverage Generative AI (GenAI).
With use cases proliferating horizontally across functions and vertically within business units,
the next step is...sustainably scaling GenAI for strategic business value.
Generative AI, like origami, transforms a resource
(data and paper, respectively) into a compelling
output. Just as origami artists fold paper to
resemble interesting shapes, Generative AI
computes data to approximate human cognition
and creativity.
3
Scaling Generative AI | 13 elements for sustainable growth and value
Getting more GenAI into production
Deloitte’s State of GenAI in the Enterprise Q3
report revealed that many businesses are
encountering challenges when making the
transition from GenAI proof-of-concept to
scaled deployment.1
Seventy percent of surveyed organizations indicate that less than one third of their
GenAI experiments have made it to production. This suggests that while enterprises are
investing in GenAI, they are not yet seeing the full potential ROI. A common challenge is
defining what is required to achieve GenAI scale at a practical level.
We define scale broadly as the ability of a system to handle a growing amount of
work or its potential to be enlarged to accommodate growth with steadily
decreasing unit costs. For GenAI specifically, scaling also means moving from
experimentation to implementation in a way that is sustainable, secure, and
aligned with business goals.
GenAI at scale generates more diverse and representative outputs, it can handle more
complex tasks, and its speed, output quality, and accuracy are enhanced. As a result,
operational costs become more efficient and business impact is governed, measured,
and communicated.
4
Scaling Generative AI | 13 elements for sustainable growth and value
At the highest level, GenAI scaling factors can be grouped into the familiar areas of strategy, process, talent, and
data and technology. Each area presents challenges to be navigated and contains leading practices that help point the way
to GenAI value realization.
Strategy Process Talent Data and Technology
Ambitious Modular
Integrated Transparency Provisioning
strategy and value Robust architecture
risk to build trust the right AI
management governance and common
management in secure AI infrastructure
focus platforms
Clear, Agile Acquiring Effective
Strong Transformed
high-impact operating model (external) and Modern data model
ecosystem roles, work,
use case and delivery developing foundation management
collaboration and culture
portfolio methods (internal) talent and operations
Essential elements for scaling Generative AI initiatives from pilot to production
5
Scaling Generative AI | 13 elements for sustainable growth and value
STRATEGY
Ambitious strategy and
value management focus
An organization’s GenAI strategy and vision need to be comprehensive, integrated with
broader business objectives, and aligned with other existing AI programs. Executive
buy-in and a top-down mandate are essential for aligning functions and decision-making.
Leadership sets priorities and strategy, and without an executive mandate, it is difficult
to coordinate change across multiple teams. A cohesive GenAI strategy defines business
objectives, sets measurable goals, identifies valuable areas for application, and measures
realized value. As a part of strategy development, it’s important to show progress against
short-term goals and inform any iterative improvements needed to the strategy.
Establish a
comprehensive vision
with a top-down mandate
6
Scaling Generative AI | 13 elements for sustainable growth and value
STRATEGY
Clear, high-impact use
case portfolio
There are six common macro archetypes for GenAI: Q&A-based search, summarization,
content generation, content transformation, virtual agent, and code generation. In seeking
value-driving applications, organizations should look across the archetypes for low-
barrier, high-impact use cases for core business domains. These drive efficiencies and
savings that can be reinvested in innovation. Other high-impact use cases may be more
transformational and differentiating with enterprise-wide applicability. Whether deploying
a proven application or striving for something novel, all applications require technical
feasibility and a viable business case. What is more, existing processes will likely need to be
reimagined to incorporate and leverage the capabilities of GenAI use cases in workflows.2
At its core, the use case portfolio needs to be focused on answering business questions
and meeting quantified goals. We see leading organizations create business cases that
weave together the value GenAI can provide to multiple teams, rather than evaluating
the value of individual applications. This is done most effectively by running a number
of use cases in parallel. It makes efficient use of resources and allows for rapid portfolio
management should a specific use case prove less compelling without sacrificing
momentum of the overall Gen AI portfolio.
Explore low-barrier, high-
impact use cases to drive
efficiencies and savings
7
Scaling Generative AI | 13 elements for sustainable growth and value
STRATEGY
Strong ecosystem
collaboration
GenAI is maturing rapidly, with existing providers and new market entrants alike driving
capabilities and lateral applications. The array of GenAI solutions and the speed with which
they are evolving can make it challenging to select the appropriate tools and platforms
that enable enterprise strategy. To reach target outcomes, enterprise leaders should build
strategic relationships with technology and data ecosystem stakeholders and keep pace
with GenAI development. By monitoring elements like product roadmaps, total cost of
ownership, and labor delivery models, business leaders can gain an understanding as
to how their GenAI programs should evolve and how ecosystem players can accelerate
progress and results as strategic partners, rather than as transactional vendors. Use a
framework to support a structured approach to evaluating solutions based on factors such
as data readiness, AI maturity, risk appetite, and total cost of ownership.
Evolve with existing
providers and new market
entrants alike
8
Scaling Generative AI | 13 elements for sustainable growth and value
PROCESS
Robust governance
Inconsistent processes can lead to risks and inefficiencies, while consistent governance
processes help standardize workflows for data collection, solution engineering, output
validation, and performance monitoring. Common delivery frameworks (e.g., LLMOps)
bring together GenAI development and deployment into a unified, governed lifecycle that
is secure and compliant. A common misconception is that strong processes can hinder
speed and creativity. Our experience suggests the opposite. By understanding how work
needs to be done and the accompanying guardrails, teams are empowered to explore
ways to generate value without fear that they may be making a mistake. Clear boundaries
allow freedom for bold action and innovation, while a lack of clarity may lead to more
conservative approaches. Governance includes documented roles and responsibilities
that drive stakeholder accountability in decision-making across the AI lifecycle, and
inform the controls for risk identification and mitigation. Governance also standardizes
how stakeholders identify, prioritize, and approve GenAI applications. As processes are
amended, organizations need to be mindful about disrupting existing automated or
manual controls and take steps to establish assurance in those amended processes.
Even as the regulatory landscape is in flux, organizations should proactively establish
governance processes that meet existing or likely regulatory requirements.
Create repeatable
governance processes to
help standardize work
9
Scaling Generative AI | 13 elements for sustainable growth and value
PROCESS
Integrated risk management
For GenAI to reach its full potential business value and adoption, it must be trusted
and secure.3 Attempting to scale without accounting for trust in data and the machine
that consumes it can have implications for regulatory compliance, finance and strategy,
cybersecurity and privacy, adoption and change management, and brand reputation—
the consequences of which can limit or even erase GenAI’s intended value. Risk and
trust need to be considered and addressed across the GenAI lifecycle, from design and
development through deployment and scaled implementation. This includes validation
processes and feedback loops for human oversight to manage solution performance
and accuracy. It also includes guardrails to ensure privacy, drive ongoing compliance, and
promote agility in proactively responding to emerging risks. Data security is particularly
essential. Differentiated GenAI applications are fueled by sensitive, proprietary enterprise
data. Thus, training and usage can potentially expose or leak business-critical data and
create risks to the organization. This is not a one-time event—organizations must make
this part of regular work, rather than a separate consideration.
Address risk and data
security across the
GenAI lifecycle
10
Scaling Generative AI | 13 elements for sustainable growth and value
PROCESS
Agile operating model
and delivery methods
The operating model impacts how the enterprise aligns technology, processes, and roles
and responsibilities to create strategic business value. An integrated model connects the
blueprint for value with AI business cases to inform how work is delivered and helps drive
alignment across the enterprise. As the marketplace matures and new capabilities and
risks impact AI lifecycles and governance, the organization needs to be agile in matching
internal opportunities with the right technologies. To help, organizations may turn to
technical experts or an AI Center of Excellence (COE) that equips decision makers with
the insight to align the vision for success with the organization’s AI maturity and ambition.
This supports a cohesive approach to orchestrating the elements of GenAI development
and application. It helps avoid AI and data silos and instead drive toward reusable
building blocks, coordinated sourcing strategy, informed build-versus-buy decisions, and
security and risk management.
Support a cohesive
approach to orchestrating
the components
11
Scaling Generative AI | 13 elements for sustainable growth and value
TALENT
Transparency to build trust
in secure AI
Trust in GenAI is essential to increasing workforce adoption and realizing benefits. With
GenAI, employees may have existing biases, inhibitions, skills gaps, or even a fear that they
could be replaced by a machine. Trust in GenAI grows out of transparency, where every
stakeholder understands how the enterprise is pursuing GenAI applications, how they are
intended to create value, and how the workforce can leverage these tools as efficiency and
productivity enhancers. Transparency around the benefits targeted by GenAI solutions
helps correct misinformation and creates an opportunity to improve the workforce
experience. Trust is also important for external stakeholders, third parties, and customers,
and a transparent approach to GenAI use includes consent for data collection, notification
of how GenAI outputs may impact users, and documentation across the AI lifecycle to
inform audits and compliance.
Help stakeholders understand
the GenAI vision and how it
creates value for them
12
Scaling Generative AI | 13 elements for sustainable growth and value
TALENT
Transformed roles, work,
and culture
Deployments at scale can disrupt the status quo, transforming employee responsibilities and
how work is accomplished. As an enterprise reimagines strategy, processes, and technology
to drive GenAI value, the workforce needs to be brought on the journey as value is created
through individuals doing work differently. Organizations should nurture adoption by
documenting and communicating responsibilities and process amendments to workflows.
Poor communications may cause misunderstanding about GenAI’s potential and limitations,
leading to unrealistic expectations or resistance. Conversely, effective communications
align stakeholders around the same vision for scale and value, including as they relate to
governance, policy, IT security, risk, and funding. Topics to communicate include outcomes
and lessons learned, the organization’s AI roadmap, the impact on end users (e.g., customers
or employees), and guidance to the workforce on how to balance day-to-day tasks with
AI skills development. Ongoing adoption should be measured to identify optimization
opportunities and internal leading practices. This should inform the overall use case roadmap
and activation strategy. Simply put, upstream conversations should take place before
continuing to build technical solutions that are underdelivering against expectations.
Nurture adoption by
documenting responsibilities
and process amendments
13
Scaling Generative AI | 13 elements for sustainable growth and value
TALENT
Acquiring (external) and
developing (internal) talent
Organizations deploying GenAI need to consider the skilled human talent required
across the GenAI lifecycle. Skills mapping can reveal where the enterprise needs to
expand or improve the workforce. Recruiting new talent is one avenue, such as by
attracting new employees from educational facilities (e.g., universities). In reimagining
work with GenAI, the organization may attract new leaders who are eager to use
technology to deliver business value, as well as top talent seeking opportunities to
learn and develop. Yet, most of a company’s GenAI capabilities will grow out of training
and upskilling existing employees, and as GenAI touches every part of the enterprise,
the entire workforce requires training to adopt and use it. To this end, businesses may
create overall AI literacy programs, training plans tailored to employee personas (e.g.,
technical, functional, sales, marketing, etc.), and opportunities (e.g., hackathons and
digital playgrounds) for employees to apply new knowledge and build competence in
GenAI application, management, and monitoring. A GenAI COE can help orchestrate
human-centered continuous learning to promote adoption.
Balance talent
acquisition with
workforce upskilling
14
Scaling Generative AI | 13 elements for sustainable growth and value
DATA & TECHNOLOGY
Modular architecture and
common platforms
IT architecture needs to evolve as technologies mature and as the organization’s needs
change. Flexibility in modular systems includes leveraging microservices and APIs
(Application Programming Interfaces) for tech stack integration, as well as techniques for
improving output reliability (e.g., retrieval augmented generation, fine-tuning). This enables
platform and model “lift and shift” and supports partnerships with hyperscalers that can
provision on-prem or cloud-based environments via contracts that reward increased
volume with lower unit costs. In prioritizing a modular architecture, organizations can
facilitate user growth with a cost-per-user model, automate guardrails for managing GenAI
risk, leverage GenAI capabilities in enterprise software platforms, and establish an internal
marketplace where users can select models, access prompt catalogs, and leverage
existing solutions. Modular architecture and delivery also accommodate low-code
platforms for business users and provide a clear pathway to industrializing capabilities.
Prioritize a flexible
IT architecture to
facilitate enhancements
15
Scaling Generative AI | 13 elements for sustainable growth and value
DATA & TECHNOLOGY
Provisioning the right
AI infrastructure
GenAI infrastructure includes reusable assets, data pipelines, solution development
environments, and a range of post-deployment management and feedback capabilities.
Bringing the right secure infrastructure to the right place in the GenAI value chain is
necessary for sustainable, cost-effective scale. Taking an AI Factory approach enables
reusable components and data products while also integrating sourcing strategy,
cybersecurity considerations, demand generation, prioritization, governance, and business
outcomes. While focusing on speed to value and taking an agile, incremental approach
to infrastructure development, organizations can look to iterative design and continual
evaluation of cost mechanisms against a per-user or per-use model. One important
consideration is that executives are likely to be more comfortable funding enhancements
to existing capabilities, as opposed to building net-new systems. Using existing investments
and approaching scale as building incremental capabilities can help encourage investments
by overcoming a misperception that a GenAI endeavor is starting from scratch.
Take an agile
approach to enable
continuous improvement
16
Scaling Generative AI | 13 elements for sustainable growth and value
DATA & TECHNOLOGY
Modern data foundation
As organizations increasingly shift to hybrid-cloud environments, data integration challenges
may increase, with proprietary and third-party data sources existing on disparate platforms.
In addition to master data, GenAI applications consume other forms of data (e.g., reference,
unstructured first-party) that traditionally sit in the realm of knowledge management. Value
creation opportunities from GenAI are blending knowledge and data management capabilities.
Data quality and accessibility issues can limit value and potentially create a perception that
scaled solutions are not viable nor valuable. A GenAI-ready data foundation includes the
processes, philosophies, approaches, and approvals for data sharing and use. As a part of this,
evaluate the organization’s data findability, accessibility, interoperability, reusability, and storage.
Rather than starting from scratch, the organization’s existing data governance efforts can likely
be extended and adjusted to accommodate unstructured data. Data should also be curated
and integrated across departmental lines. Consider a parallel workstream for data readiness
evaluation and progression focused on clean and organized data, efficient data pipelines,
and robust data governance practices. By ensuring systems are secure and foundational
data capabilities are aligned with the GenAI strategy and governance, enterprises can evolve
data availability, engineering, and management to enable adoption and scale. At the same
time, it is worth noting that interim value can be harvested, albeit at a lower potential, while
comprehensive and foundational data modernization activities are underway.
Align data capabilities and
processes with GenAI strategy to
support quality and accessibility
17
Scaling Generative AI | 13 elements for sustainable growth and value
DATA & TECHNOLOGY
Effective model management
and operations
Trustworthy, compliant GenAI applications require coordinated solution management,
including continuous monitoring for impartial output accuracy, waypoints for decision-
making, and data feedback loops for continuous improvement. Cost management is also a
factor. GenAI deployment raises questions around variable and fixed costs, and business
leaders need visibility into managing and forecasting end-to-end costs for infrastructure,
tools, personnel, maintenance, and models. Insourcing key functions may permit
differentiation or better economics over time, and insourcing decisions need to be balanced
against the cost to build a capability, the ramifications of moving to a fixed versus variable
cost, and the expenses associated with capability management (e.g., hiring and training,
oversight, technology acquisition, facilities).
Monitor for impartial
output accuracy and focus
on cost management
18
Scaling Generative AI | 13 elements for sustainable growth and value
Measuring success with GenAI at scale
The value of scaled GenAI deployments is found
in how they advance an integrated enterprise
strategy and drive toward business goals.
Establishing realistic goals for quantitative KPIs (beyond productivity and efficiency metrics,
such as hours saved) allows the enterprise to assess whether the scaled deployment is
achieving its intended business impact. With a use case portfolio that balances cost- and
revenue-oriented value levers, there are key indicators that reveal whether the
enterprise is on the right track:
• Increased speed to market, from ideation to deployment
• A decline in proof-of-concept demand, as demand shifts to low-code environments
available to business users
• A decrease in unit cost for new capabilities/solutions, with technical solutions and code
being reusable, thus reducing development efforts
• An increase in the number of foundational capabilities that help the organization access
GenAI advancements as they emerge
• An increase in domain-specific models allowing for more use cases and broader
application across the organization
• Increased use of capabilities and solutions, owing to a growing number of users in the
enterprise
• An increase in stated value realization on a cumulative basis due to GenAI
• An increase in internal certification/badging of existing employees in GenAI capabilities,
both functional and technical
• Use of GenAI to redefine a business process, rather than embedding GenAI in existing
business processes
19
Scaling Generative AI | 13 elements for sustainable growth and value
GenAI capabilities are improving and multiplying,
and at this point, few organizations are likely to
have achieved each element of scale to their
greatest capacity. The leading practices, governed
processes, and ecosystem of complementary
technologies are still being developed and defined.
While change is inevitable, pursuing the
elements of scale today positions the
organization to go live with GenAI for
business value as this transformative
technology evolves.
20
Scaling Generative AI | 13 elements for sustainable growth and value
Let’s connect
Reach out for a conversation on scaling Generative AI
Lou DiLorenzo Jr. Edward Van Buren Rohit Tandon
US AI & Data Strategy Government & Public Services US AI & Insights
Practice Leader Leader – Applied AI Practice Leader
US CIO & CDAO Programs Deloitte Consulting LLP Deloitte Consulting LLP
Executive Sponsor [email protected] [email protected]
Deloitte Consulting LLP
[email protected]
Acknowlegements
The authors would like to thank the following leaders and colleagues for their contributions to this effort.
Kevin Abraham, Beena Ammanath, Aniket Bandekar, Kevin Byrne, Ricky Franks, Justin Hienz, Kevin Hutchinson, David Jarvis,
Carissa Kilgour, Lena La, Geoff Lougheed, Parth Patwari, Brittany Rauch, Jim Rowan, Kristin Ruffe, Baris Sarer, Dean Sauer, Laura
Sangha Pati Aditya Kudumala Jenn Malatesta
Shact, Brenna Sniderman, Ian Thompson, and Saurabh Vijayvergia.
USI AI & Insights Life Sciences Global AI Leader Commercial Officer
Practice Leader Deloitte Consulting LLP Deloitte & Touche LLP Endnotes
Deloitte Consulting LLP [email protected] [email protected]
1 Jim Rowan, Beena Ammanath, Brenna Sniderman et al, “Now decides next: Moving from potential to performance, Deloitte’s
[email protected]
State of Generative AI in the Enterprise,” Quarter three report Deloitte, August 2024.
2 Rowan, Ammanath, Sniderman et al, “Now decides next.”
3 Deloitte, “TrustworthyAITM, Bridging the ethics gap surrounding AI,” accessed 3 October 2024.
21
Deloitte refers to one or more of Deloitte Touche Tohmatsu Limited, a UK private company limited by guarantee, and its
network of member firms, each of which is a legally separate and independent entity. Please see www.deloitte.com/about
for a detailed description of the legal structure of Deloitte Touche Tohmatsu Limited and its member firms. Please see
www.deloitte.com/us/about for a detailed description of the legal structure of Deloitte LLP and its subsidiaries. Certain
services may not be available to attest clients under the rules and regulations of public accounting.
This publication contains general information only and Deloitte is not, by means of this publication, rendering accounting,
business, financial, investment, legal, tax, or other professional advice or services. This publication is not a substitute
for such professional advice or services, nor should it be used as a basis for any decision or action that may affect your
business. Before making any decision or taking any action that may affect your business, you should consult a qualified
professional advisor. Deloitte shall not be responsible for any loss sustained by any person who relies on this publication.
Copyright © 2024 Deloitte Development LLC. All rights reserved. |
198 | deloitte | us-advisory-ai-data-readiness.pdf | AI data readiness (AIDR)
Getting your data ready for AI adoption at scale
July 2024
AI data readiness (AIDR) | July 2024
Introduction
Data is becoming increasingly important for the success of a
business as organizations adopt to changes in the business
environment; become more digital, data-driven, and use data
to influence decision-making; and become more responsive to
customer needs. Data has historically been used to drive various
aspects of business and has been an enabler for emerging
technologies, including artificial intelligence (AI), which has been
a game-changer in recent years.
AI, at its core, is a sophisticated and multifaceted concept,
intricately woven from three fundamental components:
Each of these elements plays a crucial role in the functioning and
impact of AI and has specific risks and challenges that need to be
mitigated through an effective set of implemented AI
Business context Technique/ algorithm Data governance requirements.
What makes up AI? Common risks/challenges for governing AI
BBuussiinneessss ccoonntteexxtt •• PPuurrppoossee aanndd vvaalluuee o off A AI I ••OOppeeraratitoionnala cl ocnotnrtorlosls
•• AAccccoouunnttaabbiilliittyy ffoorr AAI Iu ussee ••HHuummaann-i-nin-t-hthe-elo-loopop
TThhee bbuussiinneessss ccoonntteexxtt oorr pprroobblelemm i nintetennddeedd t oto b e
baed dardedsrseesds wedit hw tithhe tuhsee u osfe AoIf m AoI dmelosd/aellgso/ariltghomristh ms •• IImmppaacctt oonn ppeeoopplele and ••RReespspoonnses eto t ou nuinnitnetnednedde do uotucotcmomeses
aencdo seycsotesymstem
TTeecchhnniiqquuee//aallggoorriitthhmm •• AApppplliiccaabbiilliittyy ttoo uussee c caassee ••LLifiefe c yccylcele c oconntrtorlosls
SSppeecciifificc tteecchhnniiqquuee,, tteecchhnnoollooggyy, ,o orr c coommbbininaatitoionn o of ft htheeses et hthatat •• OObbffuussccaattiioonn//eexxpplalaininaabbiliiltityy ••PPeerfroformrmanancec ein idnidciactaotrosrs
aarree uusseedd ttoo aaddddrreessss aa ssppeecciiffiicc u ussee c caassee o or rb buussinineesss sp proroblbelmem • Vendor/ platform dependency • Data and model drift
• Vendor/ platform dependency • Data and model drift
((ee..gg..,, nnaattuurraall llaanngguuaaggee pprroocceessssiningg ( N(NLLPP),) ,n neeuuraral ln neetwtwoorkr)k)
DDaattaa •• DDaattaa ggoovveerrnnaannccee and ••DDaatata r eresisliielinencycy
DDaattaasseettss ((iinntteerrnnaall oorr eexxtteerrnnaall)) u usseedd t too b buuilidld a anndd t rtarainin A AI I astnadn sdtaarnddsards ••DDaatata m moovevmemenetnt
mmooddeellss//aallggoorriitthhmmss,, aanndd tthheeirir l elevveel lo of fc cuuraratitoionn a anndd f ifit-tf o-fro-ur-sues e •• DDaattaa eetthhiiccss aanndd pprrivivaaccyy ••DDaatata u usese/f/itfi tfo fro rp uprupropsoese
((ee..gg..,, aavvaaiillaabbiilliittyy ooff vveeccttoorrss,, wweeigighhttss, ,r reessuultlsts)) •• D Da at ta a q qu ua al li it ty y ••TThhiridrd-p-paartryt yd daatata
1
AI data readiness (AIDR) | July 2024
Defining the business problem is a linchpin in maintaining Identifying the appropriate algorithm or technique is another
a sharp focus on requirements throughout the creation of an critical step in implementing an AI solution, once the business
AI model. This initial step serves as a compass, guiding the requirements have been identified. This involves considering
development process by helping to clearly articulate the business factors such as scalability, interpretability, and computational
efficiency. Consequently, this step assists in laying the
requirements (i.e., specific challenges and/or opportunities) that the
groundwork for subsequent phases of model development,
AI model aims to address.
including data pre-processing, feature engineering, and model
evaluation, to reasonably ensure that the AI model is effective in
addressing the targeted business challenge.
Data availability
Typical data-related challenges for organizations
Data quality and fit for purpose
Typical data-related challenges for organizations
• Is the required data associated with the business
problem available within the organization?
• In case of availability of data, what are the insights into
• Is the data on which the AI model is constructed,
nuances of data availability that assist practitioners in
capable of providing meaningful insights
making informed decisions regarding data collection,
or predictions?
pre-processing, and augmentation?
• Are there potential challenges related to data quality
or quantity requiring measures to address these
issues throughout the model development process?
The effective implementation of AI hinges on adeptly managing
various data challenges, especially in the context of heightened
complexity in data life cycle management for AI applications.
Typical challenges include:
• The quality and availability of data, with poor data quality
potentially impeding AI system development.
• Ethical considerations, including privacy and security,
which highlight the importance of regulatory compliance.
• Data governance, standards, regulatory compliance, and
data resilience are emphasized to help minimize risks and
reasonably ensure accountability in AI decision-making.
To mitigate the errors and inefficiencies, it is crucial to implement
effective data quality processes, including data cleansing,
validation, and monitoring. Data quality standards and practices
are essential to reasonably ensure that the data used for training AI
models is accurate, representative, and unbiased.
2
AI data readiness (AIDR) | July 2024
Use cases related to AI over the years and associated data concerns1
AI usage Data concerns examples
1. Personalized customer service • Availability of historical transactional data
(e.g., AI powered chatbots and virtual assistants) • Accuracy of data and use of data that is fit for purpose
• False positives in data used to train the fraud detection model
2. Real-time fraud detection and security
• Manual reviews vs. level of automation to validate data quality
• Compliance to data privacy rules
3. AI powered robo-advisor
• Source/method for acquisition of data used for advanced analytics models
• Processes to manage sourcing, evaluation, procurement, integration, and
4. Credit risk assessment maintenance of third-party datasets
• Bias in data uses for assessments and decisions
1. Todd Bigham et al., AI and risk management: Innovating with confidence, Deloitte, 2018.
3
AI data readiness (AIDR) | July 2024
What is AI data readiness (AIDR)?
An organization’s preparedness in implementing strategies to help guide effective AI deployment by reasonably ensuring that its data is
available, high quality, properly structured, and aligned with its AI use cases.
That depends. What
How can I benefit data do we have to Well, what are your
from AI? use for AI? data requirements?
Considerations for AIDR:
• Understanding of the business
context and objective for AI
• Structuring a process to identify
and evaluate available data
Business Users Data scientists Data & tech. teams
• Articulating data readiness gaps
and improvement opportunities
Use AI to achieve business Identify data requirements for Provide data based on data
needs and drive decision making business needs, and ensure data requirements from the • Facilitating a common
is ready for AI use data scientists taxonomy across stakeholders
for AI data readiness
4
AI data readiness (AIDR) | July 2024
What are the steps to implement and reasonably
ensure the readiness of AI data?
Identifying the data scope, evaluating data readiness, and implementing improvements for data readiness are pivotal in creating an effective
AI model.
Define Evaluate data Improve data
data scope readiness readiness
Define the scope of data whether it Utilize aspecialized Data Develop and execute a plan to
focuses on a single, specific use case Readiness Assessment Tool to improve data readiness through
or aims for an enterprise-wide evaluate the readiness of in-scope combination of near and
transformationto determine targeted data for the intended AI application longer-term actions to accelerate
and effective planning. and use. AI build and deployment.
5
AI data readiness (AIDR) | July 2024
Define data scope
Defining the data scope is a crucial initial step for financial
institutions embarking on the journey into artificial intelligence.
Specifically, the scope for AI data readiness involves evaluating
risk tolerance, harnessing the insights of use-case owners through
strategic collaboration, and ultimately identifying key characteristics
to help articulate the problem or objective addressed by the
AI model.
This scope can range from a focused application like fraud detection
to a broader, enterprisewide embrace of artificial intelligence.
Several activities are involved in creating a well-defined data scope:
Activities Considerations
• Required data inputs designed to optimize AI model performance
1. Identify required data inputs • Data availability
• Specific data types, both structured and unstructured
• Data sources available and understanding how to apply them to the AI model
2. Define data sources • Identification and documentation of the data sources to be leveraged (e.g.,
internal databases, external APIs, third-party datasets, or acquired data)
3. Establish data collection and • Data cleaning, normalization, feature engineering, and augmentation
pre-processing requirements as required
• Adherence to data privacy regulations and safeguarding sensitive information
4. Consider data privacy and security • Access control to reasonably ensure that authorized personnel with specific
roles can view or modify sensitive data
• Definition of time frame and scale of datasets
5. Define data scope boundaries
• Limitations or exclusions to be imposed on the datasets
In the context of a fraud detection use case for a bank, establishing • Geographical information: Considering the geographic
a precise data scope is paramount. The data scope for this scenario location of transactions can be crucial for identifying anomalies.
could include:
Unusual transactions in locations not typically associated with
the customer’s behavior could be red flags.
• Financial transactions: The primary focus may likely be on
data related to financial transactions encompassing details such
as transaction amounts, time stamps, and transaction types.
• Customer behavior patterns: Analyzing historical customer
behavior is essential. This includes studying spending patterns,
transaction frequency, and typical transaction sizes associated
with each customer.
6
AI data readiness (AIDR) | July 2024
Evaluate data readiness: Five dimensions
In the realm of financial services, where data is as valuable as Having an AI data readiness approach allows for a structured
currency, the readiness of this data for AI implementation is not process to evaluate the preparedness of a client’s data landscape
just a technical requirement but a strategic imperative. across five critical dimensions:
Dimensions Capabilities for evaluation
1.Data Availability
Data that’s well-organized, structured, and easily accessible in a • Data Management
timely manner to boost efficiency in storage, retrieval, and • Data Integration and Utilization
processing,while promoting reusability along with abstraction. • Advanced Analytics
• Data Storage
2.Data volume & diversity
Sufficient and diverse datasets (e.g., representing real-world • Historical Data
scenarios) allow for AI solutions to identify complex patterns • Data Sourcing
and deliver more accurate predictions. • Data Diversity (Features)
3.Data quality & integrity
By adopting and upholding leading data quality standards • Data Accuracy and Fitness
and processes, AI models can work with accurate, • Standardization & Protocols
consistent & fit-for-purpose data, leading to reliable and • Metadata
• Documentation & Reporting
accurate outcomes.
4.Data governance
Implementing a robust governance framework for data and • Data Strategies
AI can help manage data throughout its lifecycle, establish • AI Governance and Documentation
policies, standards,data ownership,and set guidelines • Data Collaboration
• Data Assessment
around use of data for AI.
5.Data ethics & responsibility
Data ethical considerations incorporated in data policies for • Regulatory Compliance
use of data for AI to drive improvement in accountability of • Data Protection & Access Monitoring
AI-based decision-making, emphasize safety, and foster • Use Case Specific Data Rules
transparency around data usage for AI processes.
#
Our AIDR questionnaire can help measure an AIDR score to facilitate a
AIDR GO / NO-GO DECISION to move forward with the AI Build.
score
7
AI data readiness (AIDR) | July 2024
Each dimension is a pillar that upholds the integrity and efficacy of many essential building blocks and that the necessary factors are
of AI applications. The capabilities for evaluation listed above can considered for constructing the initial AI model.
reasonably ensure that these pillars are strong both individually and
This meticulous approach not only facilitates the ease of
cohesively to support the overarching goal of
implementing subsequent models, but also reasonably ensures
AI-driven transformation.
the ongoing health and performance of the initial AI model.
A highly effective strategy for achieving data readiness is to dedicate
ample time to thoroughly analyze the existing landscape across each
dimension outlined above. This process helps ensure the availability
Improve data readiness
Improving AI data readiness is important because high-quality accuracy and reliability of data is crucial for training AI models.
and well-structured data is one of the foundations of successful AI
The following are five tactical steps to improve AI data readiness,
models and algorithms. By improving data readiness, organizations
based on the evaluation of capabilities across each of the
can unlock the full potential of AI and derive meaningful insights, as
five domains .
5. Communicate
and monitor
4. Develop • Meet with
key stakeholders
improvement plans
to evaluate progress
and resolve blockers
3. Go/ No-Go • Develop initiatives to • Create data readiness
workshop help address the top scorecard to measure
findings while AI project KPIs to
2. Determine • Hold go/no-go considering the risk demonstrate impact
workshops with tolerance levels of improved
risk tolerance stakeholders established in the data readiness
previous step
to gauge
1. Analyze • Collaborate which AIDR • Establish clear
results with client to dimensions from steps in improving
determine the risk the Assessment defining a data
tolerance levels of Tool need to be governance framework
• Review the implementing a
addressed based
outcomes of the solution to the
on the risk Target
Assessment use case for
tolerance levels
Tool and identify each areas of State vision
the top findings improvement
and areas
• Consider client’s
of improvement
risk appetite
within the
business context
Expected outcomes:
• Improved data environment maturity
• Accelerated AI model development
8
AI data readiness (AIDR) | July 2024
Conclusion
AI has emerged as a transformative force in today’s data-driven
world. While readiness of data is critical for harnessing the
potential of AI applications, several key takeaways have emerged,
including challenges, prospects, and considerations for adoption.
As organizations are investing in data infrastructure and
formulating AI strategies, the continuous advancements, and a
commitment to addressing data related concerns will assist in
driving the future of AI applications.
AI data readiness will thus be the foundation for unlocking AI’s
full potential in a wide range of applications. Your data is not just
information; it can be the key to your potential opportunities in
leveraging AI solutions.
Are you ready to unleash the power of AI?
How can we help?
AI data readiness approach
Deloitte’s AIDR approach is a tool for assessing data
readiness in preparation for AI implementation.
AI Data Readiness Assessment Tool
The Data Readiness Assessment Tool is leveraged to
evaluate the current state of a company’s data
environment in the five key dimensions of
datareadiness.
AI Data Readiness Score/Results
The AIDR Assessment Toolaggregates the responses
from each question to show a score for each dimension,
rolled up into an aggregate score.
9
AI data readiness (AIDR) | July 2024
Reach out to get started:
Our team is standing by to help and is excited for the opportunity to assist with your AI data readiness journey.
Vic Katyal Cory Liepold Satish Iyengar
Risk & Financial Advisory Risk & Financial Advisory Risk & Financial Advisory
Principal and Chief Operating Officer Principal Managing Director
Deloitte & Touche LLP Deloitte & Touche LLP Deloitte & Touche LLP
Ajay Ravikumar Akiva Ehrlich Shlomi Cohen
Risk & Financial Advisory Digital Controls Artificial Intelligence and Data
Senior Manager Advisory Partner Managing Director
Deloitte & Touche LLP Deloitte Israel & Co. Deloitte Israel & Co.
Amit Golan
Financial Industry Risk and Regulatory
Senior Manager
Deloitte Israel & Co.
Contributors:
Soumya Gollamudi, Risk & Financial Advisory, Manager, Deloitte & Touche LLP
Brian Nam, Risk & Financial Advisory, Senior Consultant, Deloitte & Touche LLP
Amanda Schwartz, Risk & Financial Advisory, Analyst, Deloitte & Touche LLP
Sahana Gurumurthy, Risk & Financial Advisory, Senior Solution Advisor, Deloitte & Touche Assurance & Enterprise Risk Services India Pvt. Ltd.
10
Data health reporting | Insights and actions
About Deloitte.
This document contains general information only and Deloitte is not, by means of this document, rendering accounting, business,
financial, investment, legal, tax, or other professional advice or services. This document is not a substitute for such professional
advice or services, nor should it be used as a basis for any decision or action that may affect your business. Before making any
decision or taking any action that may affect your business, you should consult a qualified professional adviser.
Deloitte shall not be responsible for any loss sustained by any person who relies on this document.
As used in this document, “Deloitte” means Deloitte & Touche LLP, a subsidiary of Deloitte LLP. Please see www.deloitte.com/us/
about for a detailed description of our legal structure. Certain services may not be available to attest clients under the rules and
regulations of public accounting.
Copyright © 2024 Deloitte Development LLC. All rights reserved.
11 |
199 | deloitte | us-ai-institute-gen-ai-for-enterprises.pdf | Generative AI
is all the rage
Deloitte AI Institute
GGeenneerraattiivvee AAII iiss aallll tthhee rraaggee
Implications of Generative AI for businesses
About the Deloitte AI Institute
The Deloitte AI Institute helps organizations connect Combined with Deloitte’s deep knowledge and
the different dimensions of a robust, highly dynamic experience in artificial intelligence applications,
and rapidly evolving AI ecosystem. The AI Institute the Institute helps make sense of this complex
leads conversations on applied AI innovation across ecosystem, and as a result, deliver impactful
industries, with cutting-edge insights, to promote perspectives to help organizations succeed by
human-machine collaboration in the “Age of With”. making informed AI decisions.
The Deloitte AI Institute aims to promote a dialogue No matter what stage of the AI journey you’re in;
for and development of artificial intelligence, whether you’re a board member or a C-Suite leader
stimulate innovation, and examine challenges to AI driving strategy for your organization, or a hands
implementation and ways to address them. The AI on data scientist, bringing an AI strategy to life, the
Institute collaborates with an ecosystem composed of Deloitte AI institute can help you learn more about
academic research groups, start-ups, entrepreneurs, how enterprises across the world are leveraging AI
innovators, mature AI product leaders, and AI for a competitive advantage. Visit us at the Deloitte AI
visionaries to explore key areas of artificial intelligence Institute for a full body of our work, subscribe to our
including risks, policies, ethics, future of work and podcasts and newsletter, and join us at our meet ups
talent, and applied AI use cases. and live events. Let’s explore the future of AI together.
www.deloitte.com/us/AIInstitute
2
22
GGeenneerraattiivvee AAII iiss aallll tthhee rraaggee
Let’s take a moment to cut
through the hype.
The AI field took a turn with the release
of powerful Generative Artificial Intelligence
(AI) models, and as a result, the world is
seeing the automation of some skills around
creativity and imagination sooner than many
expected. For some organizations, Generative
AI holds valuable potential for higher order
opportunities, like new services and business
models. Deloitte offers a method for
selecting Generative AI use cases, as well
as some next steps for business leaders in
the Age of With™.
33
Implications of Generative AI for businesses 2
GGeenneerraattiivvee AAII iiss aallll tthhee rraaggee
The rise of Generative AI
Generative AI has captured attention
in global media and the public square,
prompting questions and discussions
around this transformative technology.
Businesses, research organizations, and Generative AI in general and LLM-
even lay users are experimenting with powered chatbots in particular
Generative AI, and given the excitement are not without risks, and this
and interest, it is important to look is prompting discussions around things
more closely at the potential capabilities like the potential for job losses and
and implications for business. legal questions around intellectual
property and ownership. What is more,
Generative AI is a subset of artificial because the chatbot mimics coherent
intelligence in which machines create human phrasing, it may give some the In this article, we look closely at the
new content in the form of text, code, impression that the AI understands the potential benefits and limitations
voice, images, videos, processes, and prompts to which it responds, which of Generative AI, introduce a method
even the 3D structure of proteins. Some can lead users to anthropomorphize to qualify if, where and how these
forms of Generative AI have been well the chatbot (i.e., the ELIZA effect, cognitive tools could be used, and
established in this decade, but it was as seen in the work of computer offer important factors for business
a large language model (LLM) powering scientist Joseph Weizenbaum). leaders to weigh when adopting
an easily accessible chat interface Generative AI.
that enabled Generative AI to have its Deloitte is working on a variety of
breakthrough moment and surprise projects exploring the opportunities In a prior article, “Implications
even specialists in the field. and business value Generative AI of Generative AI for businesses,”
can create for our clients. From Deloitte offered a deep dive on
As with other types of AI before it, experiences and conversations thus the qualities and capabilities
this new poster child of AI is stimulating far, the clear path ahead, as with of Generative AI, the state of the
the imagination as organizations and all AI, is to attempt to discover and market, and what that means for
individuals consider how to use this tool capitalize on capabilities while also organizations wading into this fast-
to benefit both business and society. responsibly managing the risks that evolving technology field. And in
Generative AI can help in incremental forthcoming articles, we will discuss
are already emerging.
digitization and basic productivity use questions from legal, ethics, risk, and
cases (e.g., more effective text-based talent and technology perspectives
channels). But its grander potential and provide insights into industry
is in the higher order opportunities, use cases.
such as new services or business models
that were previously uneconomical.
444
Generative AI is all the rage
The rise of Generative AI While this is still the
beginning, it’s moving fast.
Among organizations across industries, In late 2022, with the release of an
there is interest in differentiating AI use easy-to-use Generative AI chatbot, more
cases, from public service applications people began to discover and imagine
to addressing climate change and how this new technology can be used
transforming business functions (see in the creative space. The chatbot use
Deloitte’s AI Dossier). AI is viewed as a case opened the door for thinking more
tool that can automate skills and tasks broadly about how Generative AI can
performed by humans, and AI can be so be used for tasks, ranging from writing
successful in this regard that humans can copy to generating 3D structures and to
forget skills that have been automated. outputting organizational processes. As
Examples include writing assistants, such, we are now seeing the automation
home automation, and automotive of some skills around creativity and
navigation systems. Would most people imagination sooner than many expected.
have the ability to navigate a new city
without a mobile phone? There is a lot left to discover. In this
Age of With™, the era of humans
We have seen these kinds of automations working with intelligent machines
emerge across a variety of areas and to achieve things greater than what
skillsets. The assumed roadmap for either could do alone, Generative AI
AI was that, in the shorter term, AI is will impact the future of work and
most valuable as a way to automate become a common tool throughout
operational skills, and creative skills will various aspects of our daily lives.
remain the exclusive province of human In some cases, the applications may
thinking for the foreseeable future. With be clearly visible, but more often
Generative AI, this roadmap has taken an than not, they may be working in
unexpected turn. the background.
5
GGeenneerraattiivvee AAII iiss aallll tthhee rraaggee
The evolution of Generative AI
The ability of Generative AI to create a convincing (albeit low-
quality and greyscale) image of a human face emerged in 2014.
Since then, the image quality has increased, and today, almost
anything that can be described in words can also be generated
as an image, using a textual description called a “prompt.”
Throughout 2022, social media users tinkered with Generative
AI platforms and shared the results. We have seen avocado
armchairs and photorealistic images of astronauts riding horses
on the Moon. Cosmopolitan magazine was the first to publish
a cover page created by an AI-based image generation tool,
and there has even been a case of a user who submitted an
AI-generated image to a fine art competition—and won first
place. Today, we are seeing similar improvements in other kinds
of Generative AI. You may even have found this article via a chat
with an AI system that integrates with a search engine.
Images created with Generative AI.
66
Generative AI is all the rage
How Generative AI works:
Understanding the basics
To understand how Generative AI will impact
business and life, we need to understand what
it is, what it can do, and what it cannot do, yet.
Machine learning has dominated the field In a process referred to as training,
of AI for decades. Generally, this approach the algorithm is supplied with a large
to AI development is rooted in the dataset of input/output examples to
concept of learning from examples, rather extract patterns from the input, which
than following explicitly programmed allow conclusions about the expected
rules. This is important as there are many output. Spam filters, for example, use
tasks that humans perform based on these patterns to identify similarities in
tacit knowledge (and thus can provide data points and relate those to different
examples) but cannot describe the classes (i.e., sorting email to a spam
underlying rules to do so. For example, folder). While the input data has become
humans know how to recognize a face, more complex over time, from simple
but the rules that would instruct an AI arrays of numbers to high-resolution
system to do the same are much less photos, the output side of a model has
clear. The approach of learning from to this point been mostly limited to
examples has led to the development of categories like “spam” or “not spam,”
powerful tools that can identify intricate “cat” or “dog,” or numerical values such
patterns in complex data. as 7°C or $29. This approach powers
nearly all AI that has been deployed so far,
the result is “single purpose” applications
that can only perform one task.
7
Generative AI is all the rage
INPUT OUTPUT
Figure 1:
Used car data An example of
a single purpose
Type Engine Milage Year machine learning
model, using a
Limo Electric 70k 1996
Predicted price regression model
Regression to predict the
S. wagon Petrol 100k 2005 $17,000
model resale value for
a given car.
Truck Diesel 150k 2016
Figure 2:
INPUT OUTPUT
An example of how
Email Predicted label a single purpose
model can be used
Hi Bjorn,
No spam to sort e-mails
Have a look at this: A generative adversarial network
Classification by “spam” or
(GAN) is a class of machine learning frameworks
model Spam “not spam.”
designed by Ian Goodfellow
Enter Generative AI
The main difference between “traditional quality data processed over weeks
AI” and Generative AI is that in the latter, on large-scale, GPU-enabled, high
the output is of a higher complexity. performance computing clusters.
Rather than just a number or a label, the Only a few institutions have the necessary
output can be an entire high-resolution resources and talent to build such
image, a full page of newly written text models. Running a model also requires
(which is generated word by word), or any a lot of compute, which is why access to
other digital artifact. This introduces an these kinds of models is often provided
interesting new element: There is usually via an application programming interface
more than one possible correct answer. (API). This allows developers to use the
This results in a large degree of freedom models with their existing software
and variability, which can be interpreted products without need for additional
as creativity. infrastructure. These models are versatile
and can be fine-tuned for specific tasks,
Generative AI models are typically hence they are called Foundation Models.
large and resource hungry. Creating Unlike single-purpose AI, they are suited
them requires terabytes of high- for multi-purpose tasks.
Figure 3:
INPUT OUTPUT
With Generative AI,
Generated text user prompts lead
Prompt
to artifacts that
Generative AI is a broad field of computer sicence focused on can contain a large
What is AI?
model creating intelligent machines that can perform tasks degree of freedom
that typically require human-like congnitive abilities,
and variability.
such as perception, reasoning, learning, problem-
solving, and decision-making.
8
Generative AI is all the rage
Regarding risks and limitations
Current Generative AI models have Similar to other AI models, Foundation
limitations. Perhaps the most well-known Models can reproduce latent bias in the
is termed “hallucination,” which refers to training data, and of course, they lack
a high-confidence response that is not comprehension and the ability to reason
grounded in the training data. In other as humans do. This has implications for
words, the response is fictional. For the broader concept of Trustworthy
some applications, like art generation, AI™. After all, they are language models,
this is a non-issue and perhaps even image models, or voice models but not
a desired “creative” feature of Generative knowledge models.
AI. For other applications, however,
such as copywriting or computer code Despite limitations, Foundation
generation, hallucinations can result in Models can function at such a high
artifacts that are not entirely valid or true, quality that many new use cases
which undercuts the potential value of become possible.
Generative AI.
Another limiting factor is that today’s
Generative AI models generate artifacts
based on the model itself and the
Some known limitations
input prompt. Other additional sources
of current Generative AI
and datasets cannot currently be
integrated directly into the model’s
internal information processing without
Hallucination | Generative AI systems create
costly re-training or fine-tuning, which
confident responses that cannot be grounded in any
means Generative AI models can only
of its training data.
access information extracted from the
data on which they were trained. For
similar reasons, they cannot provide Bias | Similar to other learned models, Foundation
references and sources for the generated Models inherit the bias contained in the training data.
content, which complicates validation.
Furthermore, current models have
Lack of human reasoning | Generative AI systems
a context window of a few thousand
are based on statistical features, which is not a solid
words, which is the limit for the size
of the combined input and output. foundation for logical reasoning.
However, Generative AI models can
be combined with other systems (e.g.,
Limited context window | Current models have
search, conversational AI) to leverage the
a context window of a few thousand words, which is the
benefits of both parts. For example, with
limit for the combined input and output of the model.
a chatbot, a conversational AI system can
serve as an orchestration layer between
the Generative AI model, a search engine,
and the user, which helps to amplify the
user experience.
9
Generative AI is all the rage
Generating revenue
using Generative AI
Using this technology for business benefit can
be conceived along two distinct approaches.
First, the models can be used as they a job advertisement or a floor plan,
are available today, a simple interface all the way to the 3D model of an engine
that allows near-direct access to the part, a molecule with certain properties,
underlying model in the form of a chat or a workflow, to name a few. Use cases
for text or an image generation tool. with high usage frequency are preferred,
The second approach is to integrate as there will be more example data to
Generative AI with other technologies fine-tune and improve a model, and
to automate processes. For example, subsequently a more substantial impact.
Generative AI can allow for human- Other factors to consider in selecting
level expressive interactions, while high-value use cases are existing skill and
a conversational AI system (i.e., a chat- cost bottlenecks with human generated
or voicebot) controls the flow and artifacts. The quality of the artifact may
ensures factual accuracy. An example in some cases require human effort, but
is an automated, Generative AI-powered if it can be created with Generative AI to
call center. We expect the second a commensurate quality, then the human
approach will provide the most value. can be liberated to work on higher quality
tasks. By turning lower-level creative tasks
A good start to identifying use cases over to Generative AI, we could see things
is to find processes or tasks where like databases providing stock content
a digital artifact of some kind is created (e.g., images, sounds, or texts) turned
or processed. This could range from upside down as these digital artifacts can
be created instantly with a prompt.
If a task requires effort to execute but is easy
to validate, it might be a good use case.
Deloitte has designed a Digital Artifact a task without Generative AI; and the
Generation/Validation method to help necessary effort to validate or fact check
innovation leaders determine whether the output from the Generative AI. This
an idea can be turned into a beneficial leads to a two-dimensional classification,
use case leveraging Generative AI. At categorizing use cases based on the
the core of this method are two of the required human effort and the ability
most critical elements to consider: of the user to validate the results.
the human effort required to complete
10
GGeenneerraattiivvee AAII iiss aallll tthhee rraaggee
There is a sweet spot for Generative AI use cases
Generative AI is useful where the artifact generation effort is high and validation is easy
Digital Artifact Generation/Validation method
Identifying desirable use cases
HIGH
Generative AI assists best in use cases
where human effort is high, while
validation is easy.
CONSIDER Generation effort
3 How much human effort is required to
2
achieve the desired result
1
Validation effort
4 How much human effort is required to
check the plausibility or correctness of
Generation the result
effort
ASSESS
NEGLECT
LOW
DIFFICULT Validation effort EASY
Examples plotted above
1 Create a joke 2 Draw an image of 3 Draft a contract 4 Draft a contract
an elephant under (without legal expertise) (with legal expertise)
While creating a good
a palm tree
joke requires some effort If you do not have legal If you do have legal
into designing the punch- Drawing any sophisticated expertise, drafting expertise, drafting
line and best storytelling- image requires reasonable a contract is very hard and a contract still requires
style, it is easy to validate effort for most people validating it difficult. effort, but validating it is
simply by reading it. regardless of the tools Generative AI is useful signifacantly easier.
available. On the other where the artifact
hand, validation is easy generation effort is high
since you can just look at and validation is easy.
the picture.
1111
Generative AI is all the rage
For example, re-writing text can be There is an additional complication
a daunting and time-consuming task for that should be considered. If the model
a human. Generative AI tools can take outputs are consistently correct, users
original text and quickly produce may, over time, become less rigorous in
a re-written result, a shorter text, fact checking. Inevitably, however, the
a summary, or even a different writing model will make an error, and part of the
style. A user who is familiar with the challenge is that the errors may not be
original content can validate the accuracy obvious, particularly when Generative AI
or correctness of the output. Thus, this is used to create more complex things,
could be a promising application of like programming code. Thus, when
Generative AI. assessing the ease of validation, weigh
the importance of ongoing attention
Yet, if the user is reviewing outputs that to fact checking.
are outside of their area of expertise,
validation becomes more complicated.
The Generative AI output may
read as coherent and convincingly
accurate, but the potential for a
“hallucination” remains. If users lack
the knowledge to validate the output
and spot hallucinations, the use case is
revealed to require high levels of effort
for validation and mitigating the risks
from hallucinations.
12
Generative AI is all the rage
Insights from Deloitte projects on Generative AI:
Reaping benefits from Generative AI requires
more than identifying a good use case
Identifying use cases is only part of the avoid the temptation to go forward alone
challenge. Whenever a transformative and instead find support and knowledge
technology emerges, some organizations from partners, colleagues, and third-party
are spurred to experiment for the sake organizations operating in this space.
of its novelty, which can lead to “random
acts of digital” that do not deliver the The inherent complexity in current
anticipated return. Driving business projects leveraging Generative AI
results with Generative AI requires a requires a cross-disciplinary team
strategy and collaboration from a cross- to guide and govern the AI lifecycle.
disciplinary team. In addition, with Professionals from a variety of domains
a technology that is advancing and can help the business answer critical
maturing as quickly as Generative AI, questions, including:
DOMAIN
Ideation Business Customer & Enterprise Human Risk Regulations
& Product Operations Marketing Technology Capital Management & Laws
Development
STAKEHOLDERS
Creatives, CEO, COO, CMO CIO, CTO, IT CHRO Risk officers Legal &
designers Line of Business Compliance
leaders
KEY QUESTIONS
What can How does the How can the use Can the existing Does the What risks emerge What current and
Generative AI Generative AI fit case be leveraged MLOps-tech stack workforce possess when deploying expected laws
permit that into and enhance to build customer and platform the skills to use Generative AI (e.g., and regulations
reduces human existing processes engagement, licenses fuel Generative AI, jailbreaks, prompt- concern the use
effort and can be and enterprise and how much Generative AI, or and what are the spoofing), and of Generative AI,
rapidly validated? strategy? transparency is are third-party implications for how do these risks and are existing
appropriate? services required? talent acquisition impact Generative governance and
and upskilling? AI value? MLOps processes
sufficient to meet
those laws and
regulations?
13
Generative AI is all the rage
Based on our observations and experience, we
recommend business leaders avoid jumping head-first
into the hype. Instead, we encourage decision makers to:
1 2 3
Develop a strategy for Become familiar with the Bring together a cross-
Generative AI and integrate and underlying technologies that make disciplinary team of people with
harmonize it with the enterprise’s Generative AI possible, as well as the the domain knowledge to think
existing AI strategy. The same current capabilities and limitations. creatively about potential use cases.
principles that guide an AI-fueled Educate your workforce in the When business leaders, technology
organization apply to the use of usage, risks, and capabilities of AI leaders, and creatives work together
Generative AI (e.g., access to curated to establish a baseline of knowledge with external experts, they are able
enterprise data; AI governance; through training. Also monitor over to identify valuable applications
process transformation to leverage time how the technology advances and also design Generative AI
cognitive workers, etc.). With and the impact on business risks deployments, to mitigate business
a technology evolving this quickly, and opportunities, as they emerge. and cyber risks and meet applicable
avoid the temptation to go forward This article series may support laws and regulations.
alone. Find support and knowledge your efforts.
from partners and third-party
organizations operating in this space.
4 5 6
Leverage Deloitte’s Ensure the collection and Assess use cases against
Digital Artifact Generation/ curation of proprietary data, Trustworthy AI™ principles,
Validation method as this is key for tailored use cases including challenges around bias
to identify where Generative AI that provide a differentiator or and misinformation, attribution,
might impact your value chain, competitive advantage. transparency, and enterprise
with incremental digitization from accountability for the impact from
basic productivity use cases to Generative AI.
higher order opportunities, such
as new, differentiating services
or business models.
14
Generative AI is all the rage
Deloitte is excited to move into the future with
our clients and colleagues, as well as with our
connections in academia and the broader
AI ecosystem around the world.
The discussions so far show that there There is a lot
is a need for a deeper understanding
of Generative AI, from the underlying to cover and the
technology to its impact on the future
conversations are far
of work. As such, it is important to look
closely at the implications for risk, trust,
from over. Deloitte
and governance, which is investigated
in a forthcoming article, “Proactive risk is a trusted advisor
management in Generative AI.” There are
also legal considerations for Generative as we push beyond
AI, which we plan to cover in “Legal
the initial buzz
implications of using Generative AI
(What the AI System won’t tell you).”
around this new
technology and into
how Generative AI
can be used for good
in the Age of WithTM.
15
GGeenneerraattiivvee AAII iiss aallll tthhee rraaggee
Reach out for a conversation.
Beena Ammanath Wessel Oosthuizen Dr. Kellie Nuttall Jefferson Denti Audrey Ancion
Global Deloitte AI Institute Deloitte AI Institute Africa, Deloitte AI Institute Deloitte AI Institute Brazil, Deloitte AI Institute
Leader Lead Australia, Lead Lead Canada, Lead
Deloitte AI Institute Deloitte Africa Deloitte Australia Deloitte Brazil Deloitte Canada
United States, Lead [email protected] [email protected] [email protected] [email protected]
Deloitte Consulting, LLP
[email protected]
Jan Hejtmanek Roman Fan Anne Sultan Dr. Bjoern Bringmann Prashanth Kaddi
Deloitte AI Institute Deloitte AI Institute China, Deloitte AI Institute France, Deloitte AI Institute Deloitte AI Institute India,
Central Europe, Lead Lead Lead Germany, Lead Lead
Deloitte Central Europe Deloitte China Deloitte France Deloitte Germany Deloitte India
[email protected] [email protected] [email protected] [email protected] [email protected]
Masaya Mori Nicholas Griedlich Naser Bakhshi Tiago Durao Sulabh Soral
Deloitte AI Institute Japan, Deloitte AI Institute Deloitte AI Institute Deloitte AI Institute Deloitte AI Institute
Lead Luxembourg, Lead Netherlands, Lead Portugal, Lead United Kingdom, Lead
Deloitte Japan Deloitte Luxembourg Deloitte Netherlands Deloitte Portugal Deloitte United Kingdom
[email protected] [email protected] [email protected] [email protected] [email protected]
Special thanks to contributors:
Jakob Nikolaus Moecke, Senior Consultant, Deloitte AI Institute Germany Elon Allen, Partner, Monitor Deloitte Australia
Philipp Joshua Wendland, Senior Consultant, Deloitte AI Institute Germany Bram den Hartog, Partner, Monitor Deloitte Australia
Alexander Mogg, Lead Partner, Monitor Deloitte Germany Aisha Greene, Senior Manager, Deloitte AI Institute Canada
Kate Fusillo Schmidt, Senior Manager, Deloitte AI Institute US Anke Joubert, Senior Manager, Deloitte AI Institute Luxemburg
Erica Dodd, Senior Manager, Deloitte AI Institute Australia Dr. Gordon Euchler, Director, Deloitte Germany
Jessica Carius, Senior Consultant, Deloitte AI Institute Australia
111666
This publication contains general information only and Deloitte is not, by means of
this publication, rendering accounting, business, financial, investment, legal, tax, or
other professional advice or services. This publication is not a substitute for such
professional advice or services, nor should it be used as a basis for any decision or
action that may affect your business. Before making any decision or taking any action
that may affect your business, you should consult a qualified professional advisor.
Deloitte shall not be responsible for any loss sustained by any person who relies on
this publication.
About Deloitte
Deloitte refers to one or more of Deloitte Touche Tohmatsu Limited, a UK private
company limited by guarantee (“DTTL”), its network of member firms, and their related
entities. DTTL and each of its member firms are legally separate and independent
entities. DTTL (also referred to as “Deloitte Global”) does not provide services to
clients. In the United States, Deloitte refers to one or more of the US member firms of
DTTL, their related entities that operate using the “Deloitte” name in the United States
and their respective affiliates. Certain services may not be available to attest clients
under the rules and regulations of public accounting. Please see www.deloitte.com/
about to learn more about our global network of member firms.
Copyright © 2023 Deloitte Development LLC. All rights reserved. |
200 | deloitte | us-ai-institute-ceo-guide-to-generative-ai-enterprises.pdf | A CEO's guide
to envisioning the
Generative AI enterprise
Leading a Generative AI-fueled enterprise:
A CEO series
Deloitte Global CEO Program
Deloitte AI InstituteTM
AA CCEEOO''ss gguuiiddee ttoo eennvviissiioonniinngg tthhee GGeenneerraattiivvee AAII eenntteerrpprriissee
About the About the
Deloitte Global CEO Program Deloitte AI Institute
The Deloitte Global CEO Program is The Deloitte AI Institute helps sense of this complex ecosystem, and as
dedicated to advising chief executive organizations connect the different a result, deliver impactful perspectives
officers throughout their careers—from dimensions of a robust, highly dynamic to help organizations succeed by making
navigating critical points of inflection, to and rapidly evolving AI ecosystem. The AI informed AI decisions.
designing a strategic agenda, to leading Institute leads conversations on applied AI
through personal and organizational innovation across industries, with cutting- No matter what stage of the AI journey
change. The program offers innovative edge insights, to promote human-machine you’re in; whether you’re a board member
insight and immersive experiences to help: collaboration in the “Age of With”. or a C-Suite leader driving strategy for your
organization, or a hands on data scientist,
• Facilitate the personal success of
The Deloitte AI Institute aims to promote bringing an AI strategy to life, the Deloitte
individual executives, new or tenured,
a dialogue and development of artificial AI institute can help you learn more about
throughout their life cycle.
intelligence, stimulate innovation, and how enterprises across the world are
• Elevate the relationships between them, examine challenges to AI implementation leveraging AI for a competitive advantage.
their leadership teams, and their boards and ways to address them. The AI Institute Visit us at the Deloitte AI Institute for a
collaborates with an ecosystem composed full body of our work, subscribe to our
• Support the strategic agenda for their
of academic research groups, start-ups, podcasts and newsletter, and join us at
organizations in times of disruption
entrepreneurs, innovators, mature AI our meet ups and live events. Let’s explore
and transformation.
product leaders, and AI visionaries, to the future of AI together.
explore key areas of artificial intelligence
www.deloitte.com/us/ceo
including risks, policies, ethics, future of www.deloitte.com/us/AIInstitute
work and talent, and applied AI use cases.
Combined with Deloitte’s deep knowledge
and experience in artificial intelligence
applications, the Institute helps make
22
A CEO's guide to envisioning the Generative AI enterprise
Few technologies have debuted to as much
consumer and media fanfare as Generative AI,
especially upon the November 2022 launch of the
first conversational Generative AI chatbot.
Over the past year, user adoption,
experimentation, and awareness of
Generative AI’s seemingly boundless
possibilities have continued to expand.
This exponential growth has instilled
a growing belief among businesses
and CEOs that Generative AI has the
potential to significantly augment,
if not substitute, even the most
intricate and unstructured avenues
of value creation.
For example, computer code, once
considered the work of specialized
masters, can now be easily created by
Generative AI. Even elements of the
nuanced art of strategy development,
In a Generative AI-driven Moreover, the hard investment
a most critical executive
autonomous enterprise, such tradeoffs that CEOs have had
discipline, can be increasingly
capabilities will likely become more to face have limited their ability
augmented by AI. Consider the
commonplace over time. to develop critical capabilities,
application and use of scenarios or
including foundational technology
strategic foresight in the formulation
At the risk of adding fuel to an already and workforce investments. The
and refinement of enterprise strategy.
frenetic hype cycle, Generative AI Generative AI shift requires
Ever more powerful and generative
opens up possibilities for entirely new business leaders—most acutely,
AI could:
business models and market capture. CEOs—to alter how they lead
• Dramatically broaden and structure But the reality of the past year is the enterprise.
the basis of evidence by which to stark: There has been a lot of activity
anticipate the future of markets and interest, and plenty of proofs of
concept and demos, but a disjointed
• Create rich and divergent stories
approach has prevented most
about different plausible futures
companies from fully harnessing
• Continuously monitor the the potential of Generative AI.
environment for signals regarding
the critical uncertainties that
underpin the plausible futures
and their likelihood
• Assess the suitability of strategic
positions and options in the
context of the different futures and
suggest adaptations
3
AAA CCCEEEOOO'''sss ggguuuiiidddeee tttooo eeennnvvviiisssiiiooonnniiinnnggg ttthhheee GGGeeennneeerrraaatttiiivvveee AAAIII eeennnttteeerrrppprrriiissseee
AAnn ooppppoorrttuunniittyy ttoo rreebbuuiilldd
FFoorr cceerrttaaiinn jjoobbss aanndd iinndduussttrriieess,, eessppeecciiaallllyy tthhoossee
tthhaatt iinnvvoollvvee kknnoowwlleeddggee wwoorrkk,, GGeenneerraattiivvee AAII iiss
ppooiisseedd ttoo hhaavvee aa wwiiddeesspprreeaadd iimmppaacctt,, aanndd
eeaarrllyy mmoovveerrss ccaann ttaakkee aaddvvaannttaaggee..
44
A CEO's guide to envisioning the Generative AI enterprise
Our analysis has shown that Generative AI is much more than the
successful digital transformation can evolution of a chatbot—it can be the
result in up to $1.25 trillion (USD) in compressed digital representation
additional market cap, and Generative of the entire enterprise, capturing
AI is proving to be a powerful knowledge and communicating it
accelerant for transformation.1 Over through natural language (as opposed
the next decade, productivity gains to programming languages). To truly
and capabilities enabled by AI are capture its actual value, rather than
expected to increase global GDP by focusing on accomplishing discrete
$7 trillion, while the Generative AI tasks or “shallow” use cases that
market doubles every other year.2,3 are disconnected from the core
business, CEOs have an opportunity
CEOs can capture this value by to envision how to align Generative
setting the right vision, drawing AI to their overall business strategy.
their perspective from both a Generative AI embodies the potential
strategic understanding of the to encapsulate and disseminate the
technology and its potential to drive entirety of an enterprise, not merely in
value and marketplace advantage. completing tasks but in reshaping the
Indeed, Generative AI represents fundamental business framework.
the unequivocal catalyst reshaping
industries and redefining business We call this vision of the future the
strategies, empowering forward- autonomous enterprise: a future-
thinking CEOs as architects of an state organization that capitalizes
AI-infused future. on the unique advantages of pairing
humans with AI to help people
become far more effective and the
work more fulfilling as digital agents
complement and support them.
55
AA CCEEOO''ss gguuiiddee ttoo eennvviissiioonniinngg tthhee GGeenneerraattiivvee AAII eenntteerrpprriissee
Realizing a future-state
autonomous enterprise
PPrreevviioouussllyy,, vvaalluuee ccrreeaattiioonn tthhrroouugghh aauuttoommaattiioonn
wwaass lliimmiitteedd bbyy tthhee iinnaabbiilliittyy ttoo pprroocceessss llaarrggee
aammoouunnttss ooff uunnssttrruuccttuurreedd ddaattaa,, wwhhiicchh rreessttrriicctteedd
iitt ttoo ttaasskkss rreeqquuiirriinngg llooww ccrreeaattiivvee ddiiffifficcuullttyy,,
llooww ccoonntteexxtt vvaarriiaabbiilliittyy,, aanndd hhiigghh aaccccuurraaccyy..
TThhaatt’’ss nnoo lloonnggeerr tthhee ccaassee wwiitthh GGeenneerraattiivvee AAII..
66
A CEO's guide to envisioning the Generative AI enterprise
As we know from studying the Technical teams in the background In the autonomous enterprise of
progression of information technology are constantly monitoring the data the future, the blueprints of the
over time, cognitive automation and models the digital agents are organization, its complex ways of
systems are only going to become using to compile such assessments. working, and years of institutional
more intelligent.4 Generative AI Doing so maintains confidence in knowledge are at our fingertips,
accessible through sophisticated AI
capabilities could enable the use of the integrity and accuracy of the
models. The transformative potential
digital bots or agents that operate forecasts. After minor modifications,
of Generative AI transcends prior
throughout an enterprise in a the CEO and executive team agrees
limitations in value creation through
supportive role. Such bots could on next steps. Humans then execute
automation, heralding a future where
be given goals instead of specific the agreed vision by activating
fully autonomous bots operate
commands and could develop plans, additional agents that independently
within enterprises, helping humans
execute tasks, and even assign other assist them with making project plans,
formulate strategies, execute tasks,
digital agents tasks. designing products and digital twins,
and adapt to market dynamics,
and generating marketing content.
Imagine a planning meeting in an The digital agents can even alert fundamentally altering the landscape
autonomous enterprise. Digital agents humans when they should adjust of business operations.
are tasked with synthesizing the strategies to changes in market
company’s prior fiscal year sales and conditions to ensure resiliency.
creating a forecast based on current
and expected market conditions.
The CEO and the executive team
interrogate the enterprise AI model
about its forecasting methods
and assumptions, which are
communicated with clear rationales.
7
AAA CCCEEEOOO'''sss ggguuuiiidddeee tttooo eeennnvvviiisssiiiooonnniiinnnggg ttthhheee GGGeeennneeerrraaatttiiivvveee AAAIII eeennnttteeerrrppprrriiissseee
Humans with AI
CCrreeaattiinngg tthhiiss lleevveell ooff vvaalluuee tthhrroouugghh GGeenneerraattiivvee AAII rreeqquuiirreess
CCEEOOss ttoo rreeiimmaaggiinnee wwaayyss ooff wwoorrkkiinngg aanndd tthhee rroollee ooff hhuummaann
ccoonnttrriibbuuttiioonnss ttoo tthhee wwoorrkkppllaaccee.. AArrttiiccuullaattiinngg aa ccoommppeelllliinngg
vviissiioonn ooff hhuummaannss wwiitthh AAII ((tthhee hhuummaann ++ AAII aaddvvaannttaaggee))
ccaann hheellpp aa CCEEOO oouuttppaaccee tthhee ccoommppeettiittiioonn..
88
A CEO's guide to envisioning the Generative AI enterprise
First and foremost, the CEO should to create differentiation; and it will
be specific about how Generative AI be the role of its leaders to find that
can increase human employees’ skills, “differentiation” by designing unique
efficiency, and productivity, thanks ways for humans and AI to interact.5
to new interfaces that ease human Humans will continue to excel at
interaction and allow for engagement ensuring that a strategy balances
through natural language. Current the conflicting goals of multiple
Generative AI capabilities enhance stakeholders, meeting new customers
individual productivity by partnering face-to-face, or leveraging outside-
with humans to serve three primary the-box creativity to overcome
roles: synthesizing disparate data seemingly unsurpassable obstacles.
sources, copiloting as virtual assistants Because Generative AI is more
for complex tasks, and creating adept at churning out iterations than
content much more efficiently than generating breakthrough ideas,6 the
humans. In this way, Generative benefits of Generative AI may only
AI democratizes capabilities and be limited by our imaginations: The
acts as a great equalizer to level World Economic Forum, surveying
up talent: New or inexperienced 800 global business leaders, predicts
workers can immediately increase that “leadership and imagination skills
their contributions and value while will be largely unaffected by AI.”7 If this
experienced workers reach new prediction holds, the CEO’s role will
productivity levels. be to imagine how the creativity of
their people can be combined with the
Second, CEOs should recognize capabilities of AI to build competitive
that an autonomous enterprise advantage. In fact, CEOs can strive to
frees humans to focus on problems create an environment in which the
requiring a human touch. In fact, human workforce is incentivized to
organizations competing in a create and share content. Otherwise,
marketplace where every company AI models will over time begin to be
has access to the same Generative trained on their own content, creating
AI tools will likely need to rely on an “AI echo chamber” that could
enduring human capabilities, such markedly degrade the quality and
as curiosity, empathy, and creativity diversity of output.8
9 9
AA CCEEOO''ss gguuiiddee ttoo eennvviissiioonniinngg tthhee GGeenneerraattiivvee AAII eenntteerrpprriissee
The CEO’s role: Set the vision,
tell the story, and invest well
To better capture value and realize the full potential of the autonomous
enterprise, CEOs play a vital role in three significant areas:
setting the vision, communicating it, and making the right investments
to accelerate the journey toward that future.
1 | Set the vision. discrete use cases, pilots, and
projects to scale AI in order to
The CEO is not only critical to driving
realize its full value. To do this, they
digital transformation, but also
can lean in on three drivers of AI
sets the tone for how ambitious
scale: community, commonality,
a transformation will be.9 CEOs’
and coordination.11
most unique role is to develop
and articulate a clear vision—an
opportunity for a radically enhanced,
augmented, and eventually
automated business model that can
bring value to employees, customers,
and other stakeholders. But the Nurture a community Discover and exploit Coordinate through
autonomous enterprise will look of workers who commonality in order a central group in
different for each organization, and are enthusiastic to to build capabilities order to gain singular
CEOs must determine the salience, explore the potential across the enterprise visibility to all initiatives
as the application, speed, pace of of Generative AI tools. on integrated platforms, and to better prioritize
change, and potential for advantage In doing so, these rather than delivering high-impact and
will vary by business.10 That said, communities will be a set of disparate transformational
the aggregate pace of change is able to identify areas initiatives or capabilities. investments.
only accelerating. where there may be
duplicate efforts or
To shape their vision, CEOs may be
similar structures.
inclined to take a technology and
By building communities and better
apply it directly to their business
coordinating, organizations can find
model, systematically examining the
commonalities that lead to more integrated
opportunities for AI to be infused at
every step of the existing enterprise platforms and well orchestrated use cases.12
value chain. But, like the shift from This will likely lead to opportunities to
candles to light bulbs, Generative AI simplify the value chain and create a more
could provide CEOs the opportunity integrated enterprise.
to fully reshape and redefine their
business models, thinking beyond
10
AA CCEEOO''ss gguuiiddee ttoo eennvviissiioonniinngg tthhee GGeenneerraattiivvee AAII eenntteerrpprriissee
2 | Communicate the vision. experience and the values of the as needed to adapt to new ways
people it serves—from how humans of working. That’s not to mention
There are multiple barriers to
create, connect, and make decisions tackling concerns around privacy,
Generative AI that only a clearly
to how they consume, learn, and security, trust, explainability,14
articulated vision can unlock. In
grow. Leaders must emphasize how and regulation.
addition, our research shows that
their employees and customers can
AI starts out in a trust deficit. When flourish with machines, rather than In the coming articles of Leading a
customers know a brand is using AI, against them. Intentional, humanity- Generative AI-fueled enterprise: A CEO's
their trust in the brand declines by a powered augmentation of AI will series, we’ll help CEOs navigate these
factor of 12, and they are significantly create autonomous enterprises, challenges by guiding them through
more likely to rate a brand as low productive environments, and organizational readiness, ecosystem
in reliability. From a workforce flourishing futures that reflect what strategy, and leadership imperatives.
perspective, workers perceive it really means to be human. This series is intended to support
employers as less empathetic when CEOs on their AI journeys as their
AI tools are offered. Furthermore, 3 | Invest to accelerate organizations evolve from digital
it's not uncommon that leaders transformation. enterprises to intelligent enterprises,
have their own concerns with AI, and finally, to the autonomous
The CEO’s path to enterprise
ranging from privacy, security, and enterprise that is right for them.
adoption should give teams
transparency of results, to the loss of
confidence as well as resources
human connection—a critical factor Paul Graham, co-founder of
and freedom to experiment,
in bridging the trust gap with AI and technology startup accelerator Y
with commitments to hard
employee experiences.13 Failing to Combinator, has noted “When you’re
investments. The journey to an
address these trust risks can lead to dealing with exponential growth,
autonomous enterprise means
significant potential for value erosion. the time to act is when it feels too
building a foundation in digital and
early.”15 A year after the earliest
For CEOs at AI-fueled organizations, AI capabilities, such as technology versions of Generative AI have made
trust is imperative to building a infrastructure, for the flexibility and their big debut, few are likely to
narrative that inspires confidence computing power needed to properly claim that it’s too early to act. It’s
in employees and customers empower AI; data management, for not too late for CEOs to act—yet—
alike. Powerful narratives rooted feeding the organization’s digital on a bold vision to drive value and
in trust start with envisioning a blueprint into AI models; and change competitive advantage through the
positive future where AI enables management through upskilling, autonomous enterprise.
and complements the human cultural changes, and restructuring
11
AA CCEEOO''ss gguuiiddee ttoo eennvviissiioonniinngg tthhee GGeenneerraattiivvee AAII eenntteerrpprriissee
Reach out for a conversation
Benjamin Finzi Nitin Mittal Bill Briggs
Global CEO Program Leader Global Generative AI Leader Chief Technology Officer
Deloitte Consulting LLP Deloitte Consulting LLP Deloitte Consulting LLP
[email protected] [email protected] [email protected]
Anh Nguyen Phillips Louis DiLorenzo Jr.
Global CEO Program AI & Data Strategy Practice Leader
Research Director National US CIO Program Leader
Deloitte Touche LLP Deloitte Consulting LLP
[email protected] [email protected]
Contributors
A special thanks goes to the following colleagues for their work in this effort.
Deborshi Dutt | AI Strategic Growth Offering Leader, Florian Klein | Strategic Foresight Leader, The Center for the
Deloitte Consulting LLP | [email protected] Long View, Deloitte Consulting Germany | [email protected]
Jas Jaaj | Managing Partner, Generative AI | Abhijith Ravinutala | Writer, Novel and Eponential
[email protected] Technologies (Office of the CTO) | [email protected]
Baris Sarer | Principal, Telecom, Media & Technology, Caroline Brown | Eminence Team Lead, Office of the CTO,
Strategy & Analytics | [email protected] Deloitte Consulting LLP | [email protected]
Jonathan Goodman | Global Chair, Monitor Deloitte,
Deloitte Canada | [email protected]
1122
AA CCEEOO''ss gguuiiddee ttoo eennvviissiioonniinngg tthhee GGeenneerraattiivvee AAII eenntteerrpprriissee
About the CEO series:
Leading a Generative AI-fueled enterprise
A veritable ocean of content exists in regard to Generative AI adoption for
enterprises. Through Leading a Generative AI-fueled enterprise: A CEO series,
we aim to provide a ship for CEOs and leaders to navigate that ocean.
Not all companies may need to board this ship, but for industries that
involve knowledge work, Generative AI is poised to have widespread
impact, and CEOs can take advantage.
Endnotes
1 Diana Kearns-Manolatos, “Unleashing Value from Digital Transformation: Paths and Pitfalls,” Monitor Deloitte.
February 14, 2023.
2 Goldman Sachs, “Generative AI Could Raise Global GDP by 7%.” April 5, 2023.
3 Gopal Srinivasan et al, “A New Frontier in Artificial Intelligence,” Deloitte AI Institute. 2023.
4 Mike Bechtel, “Prepare for the Future of Information Technology,” Deloitte Consulting LLP.
5 Michael Griffiths et al, “Human Inside: How Capabilities Can Unleash Business Performance,” Deloitte Insights. 2020.
6 David Meyer, “’Generative AI Is Not Yet an Automation Technology’: A Decade later, the Authors of a Seminal Paper on Job
Risks Are Back with a Reevalution,” Fortune. September 18, 2023.
7 Mark Rayner, “AI: 3 Ways Artificial Intelligence Is Changing the Future of Work,” World Economic Forum. August 14, 2023.
8 Oguz A. Acar. "Has Generative AI Peaked?" Harvard Business Review. November 8, 2023.
9 Ibid.
10 Dr. Kellie Nuttall and Stuart Scotis, "Generation AI: Ready or not, here we come!" Deloitte Access Economics & AI Institute.
11 Mike Walsh and Nitin Mittal, “To Scale GenAI, Companies Need to Focus on 3 Factors,” Harvard Business Review.
December 1, 2023.
12 Ibid.
13 Deloitte analysis. TrustID and AI: Measuring the Impact of AI on Customer and Employee Trust. August 2023.
14 In reference to AI, explainability refers to a layer of transparency in AI systems, allowing users to understand how data is
used and how AI systems and algorithms make decisions. See: Lukas Kruger, “Explaining explainable AI,” Deloitte UK.
15 Paul Graham, “When you’re dealing with exponential growth, the time to act is when it feels too early,” Post on X. March 11,
2020. Accessed November 13, 2023.
1133
This publication contains general information only and Deloitte is not, by means of
this publication, rendering accounting, business, financial, investment, legal, tax, or
other professional advice or services. This publication is not a substitute for such
professional advice or services, nor should it be used as a basis for any decision or
action that may affect your business. Before making any decision or taking any action
that may affect your business, you should consult a qualified professional advisor.
Deloitte shall not be responsible for any loss sustained by any person who relies on
this publication.
About Deloitte
Deloitte refers to one or more of Deloitte Touche Tohmatsu Limited, a UK private
company limited by guarantee (“DTTL”), its network of member firms, and their related
entities. DTTL and each of its member firms are legally separate and independent
entities. DTTL (also referred to as “Deloitte Global”) does not provide services to
clients. In the United States, Deloitte refers to one or more of the US member firms of
DTTL, their related entities that operate using the “Deloitte” name in the United States
and their respective affiliates. Certain services may not be available to attest clients
under the rules and regulations of public accounting. Please see www.deloitte.com/
about to learn more about our global network of member firms.
Copyright © 2023 Deloitte Development LLC. All rights reserved. |
201 | deloitte | us-ai-institute-ceo-guide-to-scaling-generative-ai.pdf | Three roles CEOs need to play
to scale Generative AI
Leading a Generative AI-fueled enterprise:
A CEO series
Deloitte Global CEO Program
Deloitte AI InstituteTM
TThhrreeee rroolleess CCEEOOss nneeeedd ttoo ppllaayy ttoo ssccaallee GGeenneerraattiivvee AAII
About the About the
Deloitte Global CEO Program Deloitte AI Institute
The Deloitte Global CEO Program is The Deloitte AI Institute helps sense of this complex ecosystem, and as
dedicated to advising chief executive organizations connect the different a result, delivers impactful perspectives
officers throughout their careers—from dimensions of a robust, highly dynamic to help organizations succeed by making
navigating critical points of inflection, to and rapidly evolving AI ecosystem. The AI informed AI decisions.
designing a strategic agenda, to leading Institute leads conversations on applied AI
through personal and organizational innovation across industries, with cutting- No matter what stage of the AI journey
change. The program offers innovative edge insights, to promote human-machine you’re in; whether you’re a board member
insight and immersive experiences to help: collaboration in the “Age of With”. or a C-Suite leader driving strategy for your
organization, or a hands on data scientist,
• Facilitate the personal success of
The Deloitte AI Institute aims to promote bringing an AI strategy to life, the Deloitte
individual executives, new or tenured,
a dialogue and development of artificial AI institute can help you learn more about
throughout their life cycle.
intelligence, stimulate innovation, and how enterprises across the world are
• Elevate the relationships between them, examine challenges to AI implementation leveraging AI for a competitive advantage.
their leadership teams, and their boards and ways to address them. The AI Institute Visit us at the Deloitte AI Institute for a
collaborates with an ecosystem composed full body of our work, subscribe to our
• Support the strategic agenda for their
of academic research groups, start-ups, podcasts and newsletter, and join us at
organizations in times of disruption
entrepreneurs, innovators, mature AI our meet ups and live events. Let’s explore
and transformation.
product leaders, and AI visionaries, to the future of AI together.
explore key areas of artificial intelligence
www.deloitte.com/us/ceo
including risks, policies, ethics, future of www.deloitte.com/us/AIInstitute
work and talent, and applied AI use cases.
Combined with Deloitte’s deep knowledge
and experience in artificial intelligence
applications, the Institute helps make
22
Three roles CEOs need to play to scale Generative AI
The strategic opportunities presented
by Generative AI require CEOs to dive deep
into their organizations’ technology agenda.
For many CEOs, that means becoming
tech-savvy enough to detect how
Generative AI could redefine
their business models, including
understanding disruptions to their
industries, identifying the competitive
advantage in their enterprise’s AI
adoption, and understanding where
this advantage would likely erode the
fastest. We last wrote about the CEO’s
need to set a vision for adopting
Generative AI, communicate that
vision, and invest in transformation.1
However, the path from vision to
action is not always clear. While many
executives recognize the importance
CEOs have always had multiple roles,
of AI, up to 87% don’t feel equipped
at times serving as skilled dealmakers
to transform their business with
with the acumen for favorable
it, according to recent surveys.2
negotiations; as venture capitalists
Given the types of Generative AI
who place bets on winning strategies
choices that need to be made, their
and manage portfolios; or as
outsized impact, and the significant
champions who evangelize important
organizational change demanded by
business priorities. In the context
such transformation, CEOs should
of Generative AI, CEOs must apply
dive into key decisions they would
their experiences from these
normally delegate. That’s because
roles to three distinct areas:
they are actively shaping their
securing computing power,
organization’s vision and defining its
selecting an ecosystem for their
ambitions: whether to be a first mover
large language models (LLMs),
or fast follower, whether AI is needed and standing up centers
for innovation or productivity, and of excellence.
whether they should be building or
buying AI capabilities.
3
TThhrreeee rroolleess CCEEOOss nneeeedd ttoo ppllaayy ttoo ssccaallee GGeenneerraattiivvee AAII
The buck stops with the CEO
Yesterday’s white-hot innovations are prone to
becoming today’s modernization or efficiency headaches.
44
Three roles CEOs need to play to scale Generative AI
Take multicloud for example. Therefore, CEOs need to take time upfront to make key decisions,
Many enterprises rushed into such as:
piecemeal cloud agreements without
establishing a central decision-making
hub, and years later found themselves
with technical sprawl that desperately
needed streamlining.3 Without
caution, Generative AI adoption
could take the exact same path. With Should we How can our How will we How do we
more and more players rolling out focus our organization measure direct embed trust
AI options and the hype burgeoning, investments build flexibility and indirect and guardrails
CEOs should learn from the lessons of on a few key in our execution cost and in the AI model
the past. Given how new Generative choices, or approach? performance development?
AI is, not even your technical leaders should we implications?
may have the expertise necessary maximize
to navigate this field alone. optionality while
the competitive
market for
Generative AI
plays out?
Some enterprises we’ve interviewed have
already run into issues of making AI investments
without reaping strong benefits. The time is
now for CEOs to make effective decisions.
55
TThhrreeee rroolleess CCEEOOss nneeeedd ttoo ppllaayy ttoo ssccaallee GGeenneerraattiivvee AAII
Securing access
to computing power
With the widespread adoption of Generative AI, the
necessity for swift model training and execution
has emerged as a critical business requirement.
66
Three roles CEOs need to play to scale Generative AI
Conventional compute infrastructure Business models are bound chip models, they are negotiating
relies on central processing units to evolve as AI increasingly with the CEOs of chipmakers to
(CPUs) that handle data sequentially. becomes a part of knowledge understand and secure the right level
However, for highly parallel workloads, work, which further emphasizes of resources for their company.
especially when dealing with LLMs, the importance of the CEO
the use of graphics processing in shepherding that change. As dealmakers, CEOs should be sure
units (GPUs) and other specialized Regardless of the level of a CEO’s to consider investor sentiments
AI chips enable massively parallel AI ambitions, they will likely need to and engage with their executive
processing, a crucial element for think creatively about partnerships leadership team, specifically their
efficiently processing terabytes of to secure computing power for chief information officer (CIO) and
data through algorithms based on Generative AI. As the dealmaker- chief technology officer (CTO),
Generative AI. As we discussed in in-chief, they play a pivotal role in to facilitate alignment between
Tech Trends 2024, companies are ensuring access to critical resources hardware procurement strategies
actively tackling this challenge by that can redefine their enterprise. and overarching business objectives.
embracing GPUs as the primary Additionally, emerging technologies
For some organizations, their needs
resource for training AI models.4 such as edge computing present
may be met by niche cloud providers
The integration of such dedicated AI new opportunities for decentralized
who specialize in GPUs, a number
chips is poised to become standard AI processing. By staying abreast
of which are cropping up in global
practice in enterprises, offering of technological advancements
markets.7 However, those with higher
early adopters a competitive edge, and market trends, CEOs can
AI ambitions of pursuing innovation
particularly in a fragile supply chain.5 make informed decisions to
and competitive advantage may want
Research predicts that the market future-proof their organization’s
to secure more robust computing
for specialized chips will be well over computing infrastructure.
power. Many such CEOs are engaged
US$60 billion in 2024 and climb up to
in conversations with Generative AI
US$120 billion by 2027.6
hardware companies. Without delving
into the technical details of the actual
7
TThhrreeee rroolleess CCEEOOss nneeeedd ttoo ppllaayy ttoo ssccaallee GGeenneerraattiivvee AAII
Selecting an AI ecosystem
for LLMs
The value of Generative AI hinges upon the data
it consumes. Or, as the adage says, “Garbage In, Garbage
Out.” This presents a problem for CEOs, as most LLMs
available today are not built with out-of-the-box domain,
industry, or organization-level specificity.
88
Three roles CEOs need to play to scale Generative AI
Though CEOs may face challenges determines whether it will be a first adoption. For example, the utility
with data regulations and standards, mover in building custom LLMs company Enbridge built separate
private LLMs can deliver clear or buy them later.9 Adopting a copilot tools for developers to
advantages in choice, cost, and venture-capitalist-like mindset, the code and for office staff to navigate
control, while enabling enterprises CEO can leverage understanding productivity applications, thereby
to retain their strategic intellectual of the marketplace and established offering diverse benefits to each
property.8 To capture this value and relationships with major players team.10 As we’ve written previously,
scale, enterprises need to select the to determine which bets can be finding commonality in AI needs can
LLMs and broader AI ecosystems that made safely, while considering ensure that the enterprise builds a
suit their specific needs. their broader portfolio and being cohesive platform for scaling AI, as
thoughtful about every round of opposed to buying one-off products
CEOs can also look at their investment. for disparate use cases.11
organization’s data as a bargaining
chip. Many enterprises have an In this pursuit, CEOs need to ensure Furthermore, CEOs must recognize
untapped trove of data, which could the LLM selection process isn’t the evolving nature of AI models and
be highly valuable to companies technologies, necessitating
purely a technical endeavor.
building AI products. CEOs can They must go beyond traditional continuous evaluation and
consider valuing current and future- procurement methodologies, relying optimization of LLM solutions. This
state data assets as potential inputs instead on strategic relationships might entail engaging with industry
to new business models, as long and market insights to identify thought leaders, cultivating strategic
as they concurrently secure or ideal partners. Given the pace of partnerships with research
anonymize data to avoid trust and innovation, today’s exciting AI model institutions, and more. CEOs can tap
regulatory concerns. They will likely can become obsolete tomorrow, their CTOs and CIOs to maintain and
need to keep all options in mind as and organizations can’t rely on check in on key AI relationships that
they make investment decisions. lengthy vendor assessments. But can drive business growth
by fostering the right collaborations and innovation.
As with securing hardware, the CEO
across the Generative AI ecosystem,
need not delve too deep into the
organizations can start to build
technical specifics of different AI
custom models for specific functions
models. Instead, the CEO calibrates
(e.g., Finance) and then scale up their
the company’s AI ambition and
9 9
TThhrreeee rroolleess CCEEOOss nneeeedd ttoo ppllaayy ttoo ssccaallee GGeenneerraattiivvee AAII
Standing up AI centers
of excellence
A recent Deloitte and Fortune CEO survey
found that 80% of organizations are already
implementing or likely to implement
Generative AI to accelerate innovation,
while a whopping 96% are doing so to
increase efficiencies.12
1100
Three roles CEOs need to play to scale Generative AI
To reap these expected benefits, unwavering support and Case in point: Generative AI adoption
organizations are standing up AI resources to facilitate inception. can be seen as a new frontier in
centers of excellence (COEs). These They can ensure that critical cognitive efficiency, enabling us
organizational hubs serve as catalysts elements of standing up the COE— to tap into the power of a human
for innovation, enabling organizations such as hardware, data needs, intelligence unburdened by repetitive
to conceptualize, develop, and deploy and governance—are sufficiently activities and able to focus on
AI solutions at scale. funded, while delegating the exploring, connecting, and elevating
remaining aspects. the human experience.15 In the era
A COE can bring together a cross- of Generative AI, valuable work is
The CEO also has to engage in winning
functional group of AI experts and not just repeating known tasks, but
hearts and minds of all stakeholders,
stakeholders to focus organizational asking the right questions, developing
including customers, employees,
efforts and create a consistent and innovating new solutions,
the board, and society at large.
approach to governance and assessing the generated outputs,
Working with their chief legal or risk
guardrails (e.g., by hiring ethicists), and fine-tuning your model for better
officer, CEOs can help ensure that
which are increasingly salient topics performance—uniquely human tasks.
all members of the enterprise, from
to consumers and employees.13
the executive leadership team to
Moreover, this organizational Finally, the CEO’s role in articulating
middle managers and beyond, feel
structure may help organizations the vision, mission, and purpose of AI
prepared for what’s on the horizon.
differentiate their AI transformation extends beyond internal stakeholders
Many employees are fearful of AI
from that of their competitors— to external partners, investors, and
transformation in their organizations
partnering with a consulting provider industry peers. By leveraging AI
and are eager to understand what
may provide the outside perspective centers of excellence as platforms
future jobs and skills may look like.14
and network needed to succeed. for knowledge exchange and
CEOs should foster a culture of AI
collaboration, CEOs can position their
fluency and innovation and champion
CEOs need not lead the organizations as frontrunners in the
a clear purpose for AI adoption, as a
implementation of these centers. race to Generative AI advantage.
matter of supercharging humans (not
Rather, their role is to champion
replacing them).
establishment and provide
11
TThhrreeee rroolleess CCEEOOss nneeeedd ttoo ppllaayy ttoo ssccaallee GGeenneerraattiivvee AAII
The AI revolution is bound to alter CEO roles
for the years to come. Already, CEOs have to
be more tech-savvy than ever, given how
important technology is to competitive advantage
and ways of working.
1122
Three roles CEOs need to play to scale Generative AI
As AI becomes even more embedded This is especially true when AI
into knowledge work, the details of adoption is still nascent. CEOs can
AI adoption are expanding out of the bring their experience to bear on
tech leader’s domain to become a the many macro and micro decisions
CEO priority. that will need to be made when a
technology is both very new and
very impactful.
As we continue our series on leading an
AI-fueled organization, we’ll delve into
more aspects of the CEO’s role in preparing
their organizations to pivot to the
Generative AI future.
13
TThhrreeee rroolleess CCEEOOss nneeeedd ttoo ppllaayy ttoo ssccaallee GGeenneerraattiivvee AAII
Reach out for a conversation
Benjamin Finzi Nitin Mittal Bill Briggs
Global CEO Program Leader Global Generative AI Leader Chief Technology Officer
Deloitte Consulting LLP Deloitte Consulting LLP Deloitte Consulting LLP
[email protected] [email protected] [email protected]
Anh Nguyen Phillips Deborshi Dutt
Global CEO Program AI Strategic Growth Offering Leader,
Research Director Deloitte Consulting LLP
Deloitte Touche LLP [email protected]
[email protected]
Contributors
A special thanks goes to the following colleagues for their work in this effort.
Kate Schmidt | COO, AI Strategic Growth Offering,
Deloitte Consulting LLP | [email protected]
Abhijith Ravinutala | Writer, Office of the CTO,
Deloitte Consulting LLP | [email protected]
Caroline Brown | Eminence Team Lead, Office of the CTO,
Deloitte Consulting LLP | [email protected]
1144
TThhrreeee rroolleess CCEEOOss nneeeedd ttoo ppllaayy ttoo ssccaallee GGeenneerraattiivvee AAII
About the CEO series: Leading a Generative AI-fueled enterprise
A veritable ocean of content exists in regard to Generative AI adoption for enterprises. Through
Leading a Generative AI-fueled enterprise: A CEO series, we aim to provide a ship for CEOs and
leaders to navigate that ocean.
Not all companies may need to board this ship, but for industries that involve knowledge work,
Generative AI is poised to have widespread impact, and CEOs can take advantage.
Endnotes
1 Benjamin Finzi et al, “A CEO’s guide to envisioning the Generative AI enterprise,” Deloitte. November 30, 2023.
2 Weber Shandwick, “Un/Predictions 2024.” February 22, 2024.
3 Mike Bechtel and Bill Briggs, “Above the clouds: Taming multicloud chaos,” Deloitte Insights. December 06, 2022.
4 Bechtel and Briggs, “Tech Trends 2024,” Deloitte Insights. December 06, 2023.
5 Lucas Mearian, “Chip industry strains to meet AI-fueled demands—will smaller LLMs help?,” Computer World.
September 28, 2023.
6 Gartner, “Gartner Forecasts Worldwide AI Chips Revenue to Reach $53 Billion in 2023.” August 22, 2023.
7 Krystal Hu, “CoreWeave raises $2.3 billion in debt collateralized by Nvidia chips,” Reuters. August 03, 2023.
8 Chris Arkenberg et al, “Taking control: Generative AI trains on private, enterprise data,” Deloitte Insights. November 29, 2023.
9 Tanmay Chopra, “When it comes to large language models, should you build or buy?,” TechCrunch. January 25, 2023.
10 Bechtel and Briggs, “Genie out of the bottle: Generative AI as growth catalyst,” Deloitte Insights. December 06, 2023.
11 Mike Walsh and Nitin Mittal, “To Scale GenAI, Companies Need to Focus on 3 Factors,” Harvard Business Review.
December 01, 2023.
12 Benjamin Finzi and Brett Weinberg, “Fall 2023 Fortune/Deloitte CEO Survey Insights,” Deloitte. November 13, 2023.
13 Brad Hise and Jenny Dao, “Ethical considerations in the use of AI,” Reuters. October 02, 2023.
14 Heather Stockton, “Inspiring Employees To Trust The Coming GenAI-Fueled Workplace,” Forbes. January 18, 2024.
15 Walsh and Mittal, “Industry 5.0 will be fueled by minds, not just machines,” Fortune. January 15, 2024.
1155
This publication contains general information only and Deloitte is not, by means of
this publication, rendering accounting, business, financial, investment, legal, tax, or
other professional advice or services. This publication is not a substitute for such
professional advice or services, nor should it be used as a basis for any decision or
action that may affect your business. Before making any decision or taking any action
that may affect your business, you should consult a qualified professional advisor.
Deloitte shall not be responsible for any loss sustained by any person who relies on
this publication.
About Deloitte
Deloitte refers to one or more of Deloitte Touche Tohmatsu Limited, a UK private
company limited by guarantee, and its network of member firms, each of which is a
legally separate and independent entity. Please see www.deloitte.com/about for a
detailed description of the legal structure of Deloitte Touche Tohmatsu Limited and
its member firms. Please see www.deloitte.com/us/about for a detailed description
of the legal structure of Deloitte LLP and its subsidiaries. Certain services may not be
available to attest clients under the rules and regulations of public accounting.
This publication contains general information only and Deloitte is not, by means of
this publication, rendering accounting, business, financial, investment, legal, tax, or
other professional advice or services. This publication is not a substitute for such
professional advice or services, nor should it be used as a basis for any decision or
action that may affect your business. Before making any decision or taking any action
that may affect your business, you should consult a qualified professional advisor.
Deloitte shall not be responsible for any loss sustained by any person who relies on
this publication.
Copyright © 2024 Deloitte Development LLC. All rights reserved. |
202 | deloitte | us-scaling-artificial-intelligence.pdf | Scaling artificial intelligence (AI)
Navigating the horizons of innovation
Scaling artificial intelligence (AI) | Navigating the horizons of innovation
AI technology is creating new opportunities and transforming To help answer these questions and guide this transformative
industries in this fast-changing world. For insurers, using AI journey, we have formulated a three-phase framework that provides
effectively is not only a way to keep up with the competition; a road map for success. It is based on our extensive knowledge of
it’s also a way to excel in a digital age where innovation is essential the industry and the opportunity and potential value to be achieved.
for success. With so many possible options and challenges, scaling This framework provides a systematic approach to shape your
AI across your enterprise can seem overwhelming. ambition and strategy, enabling insurers to contextualize their
current actions and plan a way forward in scaling AI efficiently.
Today, many executives are wondering: How do you deal with By dividing the journey into three separate horizons—optimize,
the difficulties of this journey? How do you move from testing differentiate, and disrupt—insurers can navigate the complexities
to deploying to achieving scale? How do you generate and of scaling AI with confidence and clarity.
demonstrate value and keep the momentum?
It’s one thing to explore the potential of AI through small-scale
experiments and prototypes and another to incorporate AI solutions
smoothly into your operations, delivering real benefits and driving
lasting growth. Making this transition to scale successfully requires
a structured approach and a clear understanding of where
to prioritize your efforts.
Aspirational
Stage of journey Opportunity
Disrupt
Next wave: Reimagining • Transform business models
Horizon 3
the business
Disrupt • Transform ways of working
Ex: Autonomous agents to help manage
products/policies/portfolios, multi-
agent conversations, automated fraud
investigation, insurance-specific LLMs
Horizon 2 Differentiate
Differentiate Holistic business impact • Increase revenue
Ex: Call center virtual agents, automated
• Improve experiences
end-to-end support, anomaly-based
cybersecurity and identity verification, • Reduce costs
automated policy coverage review
Horizon 1
Optimize
Optimize
Internal experimentation • Improve productivity
and optimization
• Lower risk profile
Ex: Intelligent document processing,
document contextualization and
Today summarization, improved chatbots,
synthetic data and code generation
2
Scaling artificial intelligence (AI) | Navigating the horizons of innovation
Horizon 1: Optimize Horizon 2: Differentiate
Internal experimentation Holistic enterprise
and prototyping transformation
Horizon 1 is the first step in scaling AI, where the main goal is to As your organization moves toward Horizon 2, the aim is to
experiment, test, and verify AI applications internally. It focuses on achieve comprehensive enterprise transformation and scale.
evaluating how to adopt, scale, and manage new technologies that Horizon 2 prepares the ground to combine traditional AI and
have low risk. Organizations in Horizon 1 are still emerging in terms generative AI solutions and makes the case for a wider enterprise
of AI maturity, are developing foundational capabilities (e.g., soft AI strategy. To get tangible value from AI, organizations need
launch production environment), and are establishing AI policy to change their focus from creating point solutions to building
and governance. integrated solutions using a variety of AI technologies with a full view
to enhance how they deliver experience and value for their internal
Additionally, as part of this step, the goal is to verify productivity and customers (i.e., employees) and external customers (i.e., advisers,
efficiency improvements through successful internal point solutions agents, policyholders).
across the insurance value chain, enhancing operational efficiency
and preparing for future use-case scalability. Examples include For instance, a comprehensive solution to assess the effectiveness
AI-driven intelligent document processing for simplified of marketing messages can use several AI technologies, such
information extraction and validation, and generative AI coding as descriptive analytics to measure audience reach and quality,
tools for non-developers. machine learning to measure the effectiveness of content,
and generative AI to measure and summarize insights and
recommendations for improvement. At Deloitte, we have developed
a systematic approach, called a “string of pearls,” on how to scale
successfully across an organization using AI solutions.
This horizon involves major changes in the organization, with
targeted investments and solution development across the business
functions, and it takes a holistic view of how data, business assets,
and technologies work together at the enterprise level—thus
creating value across multiple dimensions to increase revenue,
enhance experiences, and reduce costs.
3
Scaling artificial intelligence (AI) | Navigating the horizons of innovation
Horizon 3: Disrupt
Next wave: Reimagining
the business
The ultimate goal of effective AI scaling and adoption occurs in
Horizon 3, where AI can help transform traditional business models
by creating new ways of using technology. This horizon is where the
vision of AI innovators to build AI-powered enterprises becomes
reality. The goal is to achieve a “humans with machines” way of
working, where humans and machines collaborate to create the
opportunity to rethink the business model and/or operating model.
For example, with AI’s cognitive abilities that resemble human
intelligence, autonomous AI decision-making can mimic expert
judgment without human involvement for some applications
like reinforcement learning and deep learning techniques; and
improvements in causal AI enabled by evolving models and pattern
recognition techniques. Many of the most potential use cases
emerge from the combination of leading AI techniques (autonomous
agents, reinforcement learning) with advancements in technology
(transformer/probabilistic models) and hardware (Internet of Things
devices). These state-of-the-art solutions will not only change ways
of working and business models but entire industries, opening new
possibilities for innovation and growth.
4
Scaling artificial intelligence (AI) | Navigating the horizons of innovation
Embracing the
AI journey
Each horizon builds on the previous one, creating a base for
sustainable and scalable AI development. Horizon 1 allows
processes to be fine tuned, preparing the ground for more extensive
transformations in Horizon 2. Horizon 3, then, becomes the domain
of innovation, where organizations can explore the limits of what
AI can do, reinventing their business models and operations.
Scaling AI is a dynamic and complex journey that requires a careful
and strategic approach. Our scaling AI horizon framework offers a
road map for organizations to go through the different stages of AI
development, from optimizing internal processes to differentiating
their enterprise and ultimately disrupting traditional business
models. As technology keeps evolving, the ability to scale AI
effectively becomes not just a competitive edge but a necessity
for staying relevant in an increasingly digital and intelligent world.
5
Scaling artificial intelligence (AI) | Navigating the horizons of innovation
Contacts
Sandee Suhrada
Principal
Deloitte Consulting LLP
[email protected]
Udit Narula
Manager
Deloitte Consulting LLP
[email protected]
Vishvam Raval
Senior consultant
Deloitte Consulting LLP
[email protected]
6
Scaling artificial intelligence (AI) | Navigating the horizons of innovation
7
About Deloitte
Deloitte refers to one or more of Deloitte Touche Tohmatsu Limited, a UK private
company limited by guarantee, and its network of member firms, each of which is a
legally separate and independent entity. Please see www.deloitte.com/about for a
detailed description of the legal structure of Deloitte Touche Tohmatsu Limited and
its member firms. Please see www.deloitte.com/us/about for a detailed description
of the legal structure of Deloitte LLP and its subsidiaries. Certain services may not
be available to attest clients under the rules and regulations of public accounting.
This publication contains general information only and Deloitte is not, by means of
this publication, rendering accounting, business, financial, investment, legal, tax, or
other professional advice or services. This publication is not a substitute for such
professional advice or services, nor should it be used as a basis for any decision
or action that may affect your business. Before making any decision or taking any
action that may affect your business, you should consult a qualified professional
adviser. Deloitte shall not be responsible for any loss sustained by any person who
relies on this publication.
Copyright © 2024 Deloitte Development LLC. All rights reserved.
8861248 |
203 | deloitte | us-realizing-transformative-value-from-ai-and-genAI-in-life-sciences-040924.pdf | Realizing Transformative Value
from AI & Generative AI
in Life Sciences
Life Sciences companies have an opportunity to unlock
$5-7 billion (Bn) dollars in value from the use of artificial
intelligence (AI). We estimate that nearly 90% of value will be
derived from three functional areas: research and development
(R&D), manufacturing and supply chain, and commercial.
Over the past 18 months, Generative AI (GenAI) has broadened the breadth of value that AI could deliver.
Deloitte conducted a study of 20, end-to-end AI use cases, which when linked together like ‘pearls on a string’
can transform value streams (e.g., product launch, clinical development) across functional areas. Supported by
specialist interviews, benchmarking surveys, and pro forma forecasts, the objective was to assess the total
enterprise value opportunity of AI from cost reduction, cost avoidance, and revenue generation.
We estimate a top 10 biopharma company with average revenue of $65-75 Bn could capture
between $5-7 Bn of peak value by scaling the use of AI over 5 years. This varies based on organization
size (e.g., $35 Bn in revenue could achieve $2.5-3.5 Bn in peak value).
Estimated Enterprise Opportunity: $5–$7Bn
FIGURE 1. VALUE CREATION BREAKDOWN BY FUNCTION
Example Levers Impacted by AI Value Creation Breakdown
Research & Revenue Uplift Levers Cost Reduction Levers: 40%
Development • Time to market • Time to market
30 - 45% of Value • Revenue from drug repurposing • Cost to pre-clinical trial 60%
10%
Manufacturing Revenue Uplift Levers Cost Reduction Levers:
& Supply Chain • Revenue from surplus • # of deviations
90%
15 - 25% of Value manufacturing yield • Production cycle time
Commercial Revenue Uplift Levers Cost Reduction Levers: 45%
25 - 35% of Value • Patient conversion rate • Marketing content creation
• Time to/on therapy • Payer contract administration 55%
5%
Enabling Areas Revenue Uplift Levers Cost Reduction Levers:
5 - 15% of Value • Reduction in contract revenue • Overall SDLC cycle time
95%
leakage • Media / PR spend
Revenue Uplift
Cost Reduction
2 Realizing Transformative Value From AI And GenAI In Life Sciences
R&D represents the top value opportunity at 30-45%. AI applied to novel drug identification and accelerating
drug development could provide both cost savings and revenue uplift. This is followed by commercial at
25-35%, where marketing costs could be optimized and activities such as script utilization could be enhanced
by AI. Manufacturing, supply chain, and enabling areas (including IT, HR, and Finance) are primarily candidates
for cost transformation through efficiency realization and vendor cost reductions using AI.
We estimate a 5-year timeline for an enterprise to capture peak value from use of AI. The value accretion
schedule also differs by functional area due to each area’s inherent characteristics. Typically, enabling areas
(including IT, HR, and Finance) have the most opportunities for cost savings and avoidance, which results in
faster time to value realization. In contrast, R&D and supply chain have long-term capital requirements that
are difficult to amend, thus extending the timeline to reaching peak value.
FIGURE 2. AVERAGE 5-YEAR VALUE ACCRETION SCHEDULE OF AI IMPACT (PERCENTAGE OF
PEAK VALUE REALIZED)
$0.9B - $1.2B $1.8B - $2.5B $3.1B - $4.3B $4.1B - $5.7B $5.0B - $7.0B
Research & Development 22%
30% 32% 34% 35%
Supply Chain 13%
14%
17%
Commercial 19% 20%
38%
Enabling Areas 34%
33%
31% 30%
28%
21% 18% 16% 15%
Year 1 Year 2 Year 3 Year 4 Year 5
ASSUMPTIONS
1Foundational data and infrastructure are in place to enable transformational use case development
2Each function implements the full portfolio of transformational AI use cases (e.g., AI clinical trials, AI manufacturing, AI marketing)
METHODOLOGY
• Each bar represents the value a top 10 biopharma company could capture from AI over a 5-year time frame
• A top-down (% of revenue) and bottom-up (% of operating margin) approach was applied to evaluate value potential based on an
individual organization’s growth potential and operational efficiency
• The peak value range is a blended average using the two evaluation approaches
• For the top 10 biopharma companies average total revenue were $65Bn to $75Bn and average operating margin of$20Bn to $25Bn in
2022
3 Realizing Transformative Value From AI And GenAI In Life Sciences
Can GenAI live up to its predicted
value potential?
Anecdotal success stories and endless media attention have generated hype around the utility of GenAI.
Deloitte has spent the last year not only thinking critically about the value of GenAI, but also implementing the
technology, driving its adoption, and monitoring its value. This includes scaling dozens of AI and GenAI use
cases for our life sciences clients and launching our own GenAI platform for our internal colleagues.
Our experience suggests that there is realizable value from GenAI. But, it’s critical for companies to separate
hype from reality to best understand the true impact GenAI could deliver.
FIGURE 3. DEBUNKING MYTHS AROUND GEN AI
Value Realization
GenAI should deliver immediate bottom- Typically, cost reductions are likely to occur within 1-2 quarters
line value. of deployment primarily from efficiency gains and cost avoidances.
Revenue gains could take 3-4 quarters to materialize.
Talent Disruption
GenAI could be a lever to quickly right- In the short-term, GenAI could drive individual FTE productivity
size organizations. gains.
Adoption
If you build it, they will come. GenAI Adoption of GenAI is more likely to be successful when the technology
tools, once launched could be embraced is embedded in existing ways of working / tools in concert with
and utilized to the maximum. purposeful upskilling of users with prompt engineering skills.
Market Speed
Innovation is moving so quickly that LLMs We have reached critical milestones with LLMs. Within the next 1-2
and AI strategy should be updated every years, incremental shifts (e.g., multimodal processing) rather
6 months. than paradigm shifts are likely.
Enterprise Structures
Setting up a Center of Excellence is the A strong enterprise mandate, governance, and value capture
only path to adoption and success. methodology are the pathways to success regardless if the model is
centralized or federated.
4 Realizing Transformative Value From AI And GenAI In Life Sciences
What are the GenAI ‘no regret bets?’
To de-risk investments and accelerate progress, organizations should kickstart their GenAI programs with ‘no
regrets bets’ that can deliver value in a relatively short timeframe. This not only serves as a proof point to
catalyze enterprise adoption, but it also creates opportunities to fund additional investments with realized
gains.
There isn’t a one-size-fits-all approach when it comes to ‘no regrets bets.” However, based on our recent work
in implementing GenAI programs, we have learned that the following ‘bets’ are likely to reflect a low
complexity, high value profile for most organizations.
FIGURE 4. POTENTIAL VALUE FOR KEY 'NO REGRETS BETS'
VALUE TO VALUE TO WHY THIS IS
DESCRIPTION THE BUSINESS UNIT THE ENTERPRISE NO REGRETS
Research & Scientific Literature
Development Summarization
Generate easy-to- Greater productivity + Cost reduction GenAI can cut through
consume summaries of from faster hypotheses research noise and go
+ Revenue Uplift
scientific literature testing straight to insights with
minimal resource
investment
Intelligent Study
Deliverable Authoring
Automate the drafting Greater speed + Cost reduction Companies have a
of clinical study report from less rework and massive treasure
+ Cost avoidance
(CSR) deliverables automated drafting trove of past
documents that can
be tapped into to
automate creation
Supply Chain & SOP Management AI
Manufacturing Assistant
Automate updating Greater productivity + Cost reduction GenAI can help avoid
all relevant SOPs with from automated and costly quality control
+ Cost avoidance
simple prompts cascaded updates to issues by improving
SOPs how employees
perform their job
Amplified Quality
Events Mgmt.
Identify, investigate, Enhanced compliance + Cost avoidance Misclassifications
and remediate quality from faster identification can result in huge
events using AI and remediation of quality penalties that can be
events mitigated using GenAI
5 Realizing Transformative Value From AI And Gen AI In Life Sciences
FIGURE 4B. POTENTIAL VALUE FOR KEY NO REGRETS BETS
VALUE TO VALUE TO WHY THIS IS
DESCRIPTION THE BUSINESS UNIT THE ENTERPRISE NO REGRETS
Commercial AI-Generated Content
Generate ideas and Improved outcomes + Cost reduction Small productivity gain
design artifacts using AI from more personalized in this large cost bucket
+ Revenue uplift
content and faster can result in outsized
adaptation to customer bottom-line impact
needs
Contract
Performance Advisor
Intelligent investigation Minimized leakage + Cost reduction Pharma companies
of payer contracts to of revenues through spend billions in
+ Revenue uplift
identify discrepancies better monitoring rebates, GenAI can
identify deviations
which could result in
large payoffs
Enabling Areas MLR Optimization
Identify high-risk Streamlined process + Cost reduction Marketing pieces can
claims in marketing of incident management be quickly reviewed
+ Cost avoidance
materials for review and from preemptive risk resulting in faster time
automated adjustment remediation to market
Competitive
Intelligence
Generate competitive Improved insights + Cost reduction GenAI can tap into
insights through market/ from faster and more market data that is
+ Revenue uplift
industry data accurate data synthesis currently underused
to facilitate more
informed decisions
In future articles, we will dive deeper into how organizations
should approach 'no regrets bets' identification and
execution.
6 Realizing Transformative Value From AI And Gen AIIn Life Sciences
The time to act is now
The life sciences industry is at an inflection point – and harnessing AI and GenAI as a catalyst for
transformation is vital. Winning tomorrow requires organizations to take the right steps toward embracing this
technology today.
Here are top 5 actions that you could take in order to initiate momentum on your AI
and GenAI value journey:
Establish a Empower a leader(s) with a mandate to own and drive an
1
Leadership Mandate enterprise AI + GenAI agenda
Align on a Prioritize 2 – 3 strategic opportunity areas to serve as
2
Strategic Blueprint enterprise north stars
Identify Activate business units and IT/Digital to identify initial
3
No Regrets Bets “no regrets bets” that align to priority areas
Create Minimum Establish a governance function that can manage AI + GenAI
4
Viable Governance risks, investments, ethical use, and progress while
encouraging innovation
Launch Deliver solutions that can demonstrate tangible value
5
Pilot Solutions and prove out adoption
However, successfully driving large-scale AI transformation programs requires organizational
evolution. We list 4 major changes required as companies move forward on their AI value journey...
Mindset Leadership Investment Cultural Execution
Evolution Evolution Evolution Evolution Evolution
Move beyond the Goal leaders Treat AI investments AI should not be Evolve beyond front
endless cycle of against measurable as core enablers looked at like a and back office
near term proof AI targets and value of enterprise tool, but as a skill methods and adopt
of concepts, and goals in order to business strategies that all employees a “two in the box”
place long term drive AI evangelism and not as will need to approach where
bets on AI in key and accountability experimental possess to business & IT are
areas investments maximize efficiency goaled together
7
Realizing Transformative Value From AI And Gen AI In Life Sciences
Authors
Aditya Kudumala
Deloitte Consulting LLP
[email protected]
Adam Israel
Deloitte Consulting LLP
[email protected]
Sai Lella
Deloitte Consulting LLP
[email protected]
Jonathan Fan
Deloitte Consulting LLP
[email protected]
Wendell Miranda
Deloitte Services LLP
[email protected]
About Deloitte
Deloitte refers to one or more of Deloitte Touche Tohmatsu Limited, a UK private company limited by guarantee (“DTTL”), its network of member firms,
and their related entities. DTTL and each of its member firms are legally separate and independent entities. DTTL (also referred to as “Deloitte Global”)
does not provide services to clients. In the United States, Deloitte refers to one or more of the US member firms of DTTL, their related entities that
operate using the “Deloitte” name in the United States and their respective affiliates. Certain services may not be available to attest clients under the
rules and regulations of public accounting. Please see www.deloitte.com/about to learn more about our global network of member firms.
This publication contains general information only and Deloitte is not, by means of this publication, rendering accounting, business, financial, investment,
legal, tax, or other professional advice or services. This publication is not a substitute for such professional advice or services, nor should it be used as a
basis for any decision or action that may affect your business. Before making any decision or taking any action that may affect your business, you should
consult a qualified professional advisor. Deloitte shall not be responsible for any loss sustained by any person who relies on this publication.
Copyright © 2024 Deloitte Development LLC. All rights reserved.
8 |
205 | deloitte | us-navigating-the-artificial-intelligence-frontier.pdf | Navigating the
artificial intelligence
frontier
An introduction for internal audit
Navigating the artificial intelligence frontier | An introduction for internal audit
Insights to ground your AI strategy
During 2023, artificial intelligence (AI) captured the imagination of the world, fueling discussion among businesses and policymakers by
demonstrating the transformative power of how these technologies could redefine work. Whether interacting with vast sources of knowledge
and business data through humanlike interactions, accelerating how people work, or revealing new opportunities that were not previously
possible through manual efforts, the benefits of AI are far reaching.
AI is a broad domain. However, significant attention has been given to a specific field of AI known as Generative Artificial Intelligence (GenAI)
following the mass global interest in applications like OpenAI’s ChatGPT. Adoption and use of (GenAI) has been sudden and rapid among the
public. OpenAI reported reaching 100 million users within 60 days of releasing ChatGPT to the public.1 Given the opportunity GenAI presents,
and the fact employees are using GenAI “side of desk” for work tasks, it is no wonder that organizations are investing heavily in enterprise use
cases. With the rapid acceleration and integration of GenAI into business functions, AI and GenAI risk management will continue to be a hot
topic for internal audit teams throughout 2024 and beyond.
For internal audit, this presents two key considerations: how to provide assurance and assess the risks associated with AI (including
GenAI), and how to leverage its potential to evolve and innovate internal audit’s own ways of working. In this publication, we explore
these two faces of AI.
Internal audit’s role in assurance over AI
AI and GenAI offer significant opportunities for organizations. At the same time, they present a frontier of new risks for boards and audit
committees to navigate. To mitigate and minimize these risks, organizations are actively investing in the development of risk
management frameworks and controls to enable them to innovate with confidence.
These new AI controls will be needed to help manage data privacy and security risks, as well as ethical considerations and concerns
about the reliability of outputs created by GenAI. Internal audit functions are also looking at the developing regulatory landscape to
ensure that their organization is preparing for the arrival of regulations across all geographies they operate in.
In conjunction with the publication of the regulations and guidance, the pace of AI development and deployment for the United States is
expected to intensify as the US government pushes to be a global leader in AI development and innovation.
Harnessing the power of AI to reimagine internal audit’s ways of working
Alongside their organizations’ efforts to leverage AI, internal audit leaders are also trying to understand the potential impact or
opportunity and art of the possible for their own functions.
As a firm, we believe the integration and use of enabling technologies in internal audit, such as AI, is critical to helping functions
maximize their impact and value. The digital landscape is broad, covering many other domains including automation, audit
management systems, cloud-based solutions, visualization, data analytics, and process mining. While they can be deployed in
isolation, the power of digital is in their combination. As such, internal audit functions need a strategic and coordinated approach
across both the function and the internal audit life cycle.
GenAI will play an important role within internal audit functions’ digital strategies, not only in providing new capabilities but also in
helping to engage leadership and staff in continuous improvement and innovation by reimagining traditional approaches. For those who
are successful in digitizing their functions, the rewards are clear—enhanced quality, increased assurance and better insights, new levels
of productivity, increased staff satisfaction levels, and greater impact on the organization and for their broader stakeholders. Yet digital
capabilities remain a significant gap and the number-one opportunity for many functions.
1 Krystal Hu, “ChatGPT sets record for fastest-growing user base – analyst note,” Reuters, February 2, 2023.
1
Navigating the artificial intelligence frontier | An introduction for internal audit
Decoding the jargon: Useful AI terminology to know
Before internal audit functions can hope to assess and assure the risks of AI or look to explore the art of the possible from its use, internal
auditors must acquire a baseline of AI fluency. To help you with your AI 101, we outline some of the key terminologies and basics below.
Artificial intelligence
Artificial intelligence is a broad “umbrella” term
given to the field of computer science that
focuses on creating systems that can perform
tasks requiring human intelligence.
Machine learning
Machine learning refers to algorithms that make informed decisions and
learn over time without being explicitly programmed to do so. Machine
learning helps to train AI models to identify and predict patterns based on
human-processed data, rather than relying on hard-coded rules.
Deep learning
Deep learning is a powerful and advanced machine learning paradigm
that leverages neural networks to improve model performance.
The models simulate human reasoning to make intelligent decisions
and learn over time based on observed results.
Generative artificial intelligence
GenAI is a highly sophisticated subset of AI using foundation models to
create content across a variety of modalities. The models which support
the generation of content (often referred to as foundation models) are
underpinned by advanced machine and deep learning capabilities.
2
Navigating the artificial intelligence frontier | An introduction for internal audit
Types of GenAI (modalities)
The primary GenAI foundation models are focused around generating new content using our primary forms of communication,
such as text and imagery. However, there are many variations, and models continue to develop at pace. For simplicity, it can be
helpful to think of the models as being able to produce outputs across an increasing combination of the following data
modalities (e.g., multimodal):
Explain to my In Python, code a A bowl of soup A teddy bear Play "we have
colleagues the program that that is a portal painting to reduce
business impact of predicts the to another a portrait the number
Generative AI likelihood of dimension as of plastic
in 50 words customer digital art bags" in a
conversion sleepy tone
Text Code Image Video Audio
Generative AI, by creating new
content or predicting future
trends, can drive innovation,
optimize operations, enhance
customer interactions, and
enable personalized offerings.
This results in improved
business efficiency, customer
satisfaction, and potentially
opens up new revenue streams.
Large language models
Of all the modalities, large language models (LLMs) have gained most of the attention from organizations for their ability to generate
text. Popularized through tools like OpenAI’s ChatGPT, LLMs are a specific type of text-based model that have been trained on
petabytes worth of global data. The parameters within these models represent the model’s level of understanding about each word and
their context within the training datasets. In the case of LLMs, more parameters allow them to capture more complex patterns in the
data they were trained on, typically leading to improved accuracy on language-related tasks.
At a simplistic level, LLMs predict an output based on inferences built on their training and the inputs they receive. Given the level of
data they have been trained on, their ability to provide highly convincing and compelling responses in a humanlike interaction is why
they have gained such attention.
3
Navigating the artificial intelligence frontier | An introduction for internal audit
The mechanics of GenAI
GenAI is a highly sophisticated subset of AI. While the vast majority of internal auditors will not need a deep technical understanding akin to
data scientists, it can be helpful to appreciate the general mechanics of GenAI to consider where risks can arise and to determine the level of
technical skills that an internal auditor may need to provide assurance over the organizations’ use of these tools.
How GenAI works
Applications … what we see
GenAI applications generate content from user prompts across various
modalities (e.g., text, image, video, audio) based on how the underlying model
GenAI application Outputs was trained.
Why do these applications seem so human?
Like traditional AI, foundation models are models (1) that predict outputs
based on inferences built on the inputs they receive. However, through
fine-tuning (2), prompt engineering (3), and adversarial training (4), these
models can generate outputs geared toward meeting human intent.
Prediction What are foundation models?
GenAI models OpenAI’s GPT-4 and NVIDIA’s Megatron are two examples of foundation
models, specifically large language models, which use deep learning to
Models process massive amounts of data to form “memories” on the input
datasets through tokenization (5), thereby shaping the models’ parameters
Cloud and data (6). There are common foundation model architectures—for example,
platforms
Data Transformer (7), Diffusion (8)—which drive the modalities for each model.
Training on the world’s knowledge
AI
Foundation models are trained on petabytes worth of global data to shape
infrastructure
Compute power understanding, tone, and behavior while considering human communication styles.
Powering our journey to tomorrow
The scale of the compute capacity required to train and process
foundation models necessitates the usage of leading GPUs (graphics
processing units) (e.g., A100 NVIDIA) and TPUs (tensor processing units)
(e.g., Google TPU v4) on scalable infrastructure.
Key terms
1. GenAI model 4. Adversarial training 7. Transformer model
A neural network that has undergone The technique of pitting different deep A model that can “transform” words into
training to generate outputs based on a learning models against each other in a context-aware representations that Google and
given input prompt. training game or competition. University of Toronto invented in 2017.
2. Fine-tuning 5. Tokenization 8. Diffusion model
The process of refining foundation models to The process of splitting text into smaller units. Construction of high-resolution images from
make them suitable for specific applications. noise. Mostly used in speech-to-image and text-
6. Parameters to-image models.
3. Prompt engineering Trainable values within the model that are
The act of creating or modifying the prompt adjusted based on the training data to optimize
given to a model to obtain an optimal answer the output.
or output.
4
Navigating the artificial intelligence frontier | An introduction for internal audit
What can AI do today?
The capabilities that AI can provide today are allowing organizations to challenge their ways of working and reveal new possibilities. Not all of
these will be relevant to internal audit, but some could be applied across the internal audit life cycle to evolve and innovate approaches.
Example AI capabilities include:
SENSE PERCEIVE
Sense physical Sense visual
See objects See faces See actions Hear voices
data data
Sense screen Convert speech
Sense light Detect objects Detect faces Detect motion
pixels to text
Sense Identify
Sense sound Classify objects Recognize faces Identify actions
keystrokes speaker
Determine
Sense Sense mouse
Perform OCR Determine age Hear sounds gender from
temperature clicks
voice
Identify
Determine Recognize
emotion in
gender sounds
voice
Recognize
emotion
LEARN KNOW
Represent
Learn by Learn facts Retrieve
and store
technique and skills information
knowledge
Populate global Retrieve
Learn from
Learn skills knowledge relevant
examples
base documents
Populate
Retrieve
Learn by trial contextual
Learn facts relevant
and error knowledge
answer units
base
Learn by
analyzing Retrieve
Maintain truth
structure specific facts
Capabilities with most relevance and potential application to internal audit.
5
Navigating the artificial intelligence frontier | An introduction for internal audit
COMMUNICATE PLAN ACT
Act in
Understand and generate Plan
Understand language physical
language production
environment
Detect Translate Generate Plan robot Convert text
Classify text
language languages narrative motion to speech
Analyze Generate
Answer Move robot
Extract entities sentiment in image and Plan routes
questions limbs
text video captions
Analyze
Recognize Act in virtual
emotion in Dialogue
relationships environment
text
Generate
mouse clicks &
keystrokes
Generate
animated
avatar
CREATE REASON AND SOLVE PROBLEMS
Create text Create videos Infer
Create
Create custom Make logical
marketing Cluster
videos inferences
content
Make
Create sales
probabilistic Recommend
content
inferences
Create support Predict numeric
Classify
content value
Create images Create speech Solve problems
Search for
Create general Create custom
optimal Optimize
images voices
solution
Create
Satisfy
advertising
constraints
images
Create
Create models
chemicals
Capabilities with most relevance and potential application to internal audit.
6
Navigating the artificial intelligence frontier | An introduction for internal audit
AI is not as new as you think … but with GenAI we are
heading into unchartered waters
Before becoming too caught up in the GenAI hype, it is worth recognizing that most people are already using AI in their daily lives without
realizing it. For example, tools like auto-complete, spellcheck, smart calendar scheduling, and suggestions on the most effective ways to
visualize data in applications, such as Power BI, are all using forms of AI. Natural language processing (e.g., chatbots, sentiment analysis),
speech recognition (audio to text), robotics, and perception sensing (e.g., object detection) have been in existence for some time. Chances are
your organization, and potentially your own internal audit function, are already engaged in forms of machine learning. If you have not yet
explored existing AI capabilities, there are significant opportunities and benefits that can be gained before heading into the world of GenAI.
The clear potential from democratizing access and the acceleration in development of GenAI tools is creating very significant opportunities
and brings with it new areas of risk that many organizations have yet to understand.
Internal audit’s role in assurance over AI
GenAI presents a broad spectrum of risks, many of which are still emerging. Among the main concerns raised by GenAI are:
Risk Description
Privacy Personal information shared with third-party Software-as-a-Service AI may not comply with
privacy laws and puts customer/employee data at risk of exposure.
Information gathered (e.g., by web scraping) may contain IP protected content, and prompts
Intellectual property must be carefully written not to leak any secret know-how. There are also challenges with
protecting IP of content generated by AI.
GenAI tools may be targeted by adversaries to reveal sensitive information and/or take malicious
Malicious behavior
actions on networks and data.
GenAI tools may be used in an unintended manner and to circumvent organizational policies,
Ethical use
laws, and regulations (e.g., submitting content in competitive events).
Hallucination Models might output facts that are false. Sources and citations are unavailable for most models.
Bias in training data (e.g., over/under-representation of a population cohort, sexism, racism) can
Bias
generate biased outputs.
Lack of considerations for model performance limitations (dependent on training data used)
Model performance
could lead to sub-optimal business outcomes (e.g., poor quality reports).
7
Navigating the artificial intelligence frontier | An introduction for internal audit
The regulatory landscape
There have been several developments in the AI regulatory landscape, which continues to move at pace. Guidance has been published to aid
organizations as they navigate the use not only of GenAI, but of all forms of AI. Some of the key voices in the regulatory landscape include:
• EU AI Act (latest development from December 2023) – The European Union AI Act, which was provisionally agreed to among member
states and is expected to come into action in the first quarter of 2024, is a regulatory risk-based approach to classify AI systems and manage
the development, distribution, and use of AI systems.
• US White House Executive Order on AI (issued October 2023) – The Biden-Harris administration has released an executive order (EO)
aimed at enhancing safe, secure, and trustworthy development and use of AI throughout the federal government. This EO, while not a law
or regulation, encourages federal agencies to explore AI uses responsibly and manage associated risks, and could lead to new policies
impacting AI developers. To advance security and safety in AI’s development and use, the EO invokes emergency authority to require
disclosure of powerful AI systems and large-scale computing operations. It also addresses concerns around GenAI, encouraging the
identification of synthetic content and the use of labels to distinguish between authentic and AI-generated content.
• International Organization for Standardization (ISO) and International Electrotechnical Commission (IEC) published two key
standards for managing AI risks and systems (published in February 2023 and December 2023) – ISO/IEC 23894:2023 (published in
February 2023) provides risk management guidance for organizations developing, deploying, or using AI systems. It outlines principles and
processes for integrating risk management practices throughout the AI life cycle. ISO/IEC 42001:2023 (published in December 2023) focuses
on establishing, implementing, maintaining, and continually improving an AI management system. It specifies requirements to facilitate the
responsible development, deployment, and use of AI.
• NIST (National Institute of Standards and Technology) framework (published in January 2023)2 – NIST has collaborated with
organizations from both public and private sectors to develop the NIST AI risk management framework. The guidance is voluntary and aims to
help organizations understand the considerations that should be made during the design, development, use, and evaluation of AI systems.
2 Marilena Do Rosario, "What you need to know about NIST's AI Risk Management Framework published in January 2023," Deloitte Financial Services Blog, February 2, 2023.
8
Navigating the artificial intelligence frontier | An introduction for internal audit
What should internal audit be doing for their boards and
audit committees?
While GenAI technology is still developing, it is already being adopted by organizations at pace. Internal audit functions are seeking to
understand to what extent their organization is using this technology and to what extent they are planning to invest in it. As internal audit
functions grapple with this new risk domain, we recommend the following activities:
AI strategy and governance
Internal audit should consider its organization’s approach to the governance of AI. This should include
01
a review of the organization’s AI strategy, business case(s), and to what extent AI risks have been
considered. Consideration should be given to what extent senior executives have been involved in
defining the AI strategy and associated guardrails, given they can have organizational consequences.
Policy, standards, and guidelines
02
Internal audit should consider reviewing any AI policy the organization has developed, including
acceptable usage guidance and/or policy that defines the parameters of AI system development.
AI inventory
Internal audit should consider whether an AI inventory has been developed by the business
03
including both active and developing AI projects with details on their status and risk management
considerations. Organizations are taking differing approaches to this, but ultimately AI risks cannot
be managed unless there is clarity over AI use.
Regulatory readiness
Internal audit should understand how the organization is staying up to date with the fast-moving
04
regulatory environment. Organizations need to consider regulations in all the geographies they
operate in. If this assessment is not thorough, they run the risk of having to “roll back” deployed AI
use cases, which could cause significant business disruption.
AI risk management and culture
Current risk management processes may need to be amended to ensure that risks associated with AI are
proficiently covered.3 AI risk management frameworks and risk assessments are being developed and
05
should be integrated into the current risk management processes and procedures to ensure systems
utilizing AI are effectively managed, governed, and monitored. Risk appetite statements may also need to
be updated, and many organizations are adapting existing governance arrangements to be fit for AI, such
as AI ethics councils and the creation of AI centers of excellence.
3 Lukas Kruger and Michelle Seng Ah Lee, "Embedding controls and risk mitigations throughout the GenAI development lifecycle," Deloitte, accessed March 28, 2024.
9
Navigating the artificial intelligence frontier | An introduction for internal audit
Harnessing the power of AI to reimagine
internal audit’s ways of working
What about internal audit’s own use of AI?
The use of more established AI capabilities (e.g., natural language processing and machine learning) have been present within more advanced
internal audit functions for some time, often found within analytics teams. As access to GenAI and data security issues are overcome, we
expect to see internal audit functions of all shapes and sizes to significantly scale their use of GenAI. For now, the reality is that most internal
audit functions have not engaged in GenAI beyond exploration of ChatGPT or conceptual applications. Only very few are actively developing
proofs of concept, but this is just a matter of time. As AI continued to grow in popularity, it is hard to imagine a world where it is not at the
forefront of our businesses. The good news is that AI is not as scary as it seems.
Enabling technologies are becoming increasingly accessible and this is only being accelerated through the wider efforts of organizational
Information Technology (IT) functions looking at the same challenging questions. Organizations do not need to become digital experts
overnight or start replacing auditors with a team of data scientists (although increasing digital fluency and being able to access some of these
skill sets will be important).
10
Navigating the artificial intelligence frontier | An introduction for internal audit
A glimpse into the GenAI-driven internal audit life cycle
The application of GenAI on internal audit’s life cycle is only limited by the imagination and creativity of teams. From our discussions with
internal audit functions, the following applications and use cases are where many in the industry see potential:
Risk assessment Plan development Engagement planning Execution Reporting
Supporting auditor Supporting auditor Supporting auditor Analysis of data through Initial draft report.
research and research and research and natural language Initial draft report
understanding of risk understanding of risk, understanding of risk and questioning. review and QA.
for a specific industry. business processes, and business processes in
expected controls in advance of planning. Suggested interview Editorial QA, e.g.,
Supporting audit advance of engagement questions for different simplifying language,
universe creation, e.g., planning. Suggested control stakeholders’ personas. sentiment analysis.
guidance on universe objectives and test
design and process Suggested audits procedures based on Critical assessment of risk Summation of reports
universe. against the risk- assessed in-scope risk areas. and control descriptions, for audit committee
audit universe. e.g., if it covers who, what, summaries.
Suggested data where, and when.
Suggested scheduling sources and potential Generation of video/
and resource allocation analytics tests. Initial draft of workpaper. audio reporting.
based on known
constraints, e.g., Generated scripts for Drawing themes from Customized stakeholder
number of staff, their data extraction and interview notes/audio. communications.
skills and seniority. analytics execution.
First draft of scope/terms Summation/ Report language
of reference. interrogation of audit translation.
evidence documents.
Drafting emails to
Initial workpaper review communicate the audit
and quality assurance. report.
Initial draft of issue/
observations.
AI is only one element of internal audit’s digital landscape. Significant benefits can be achieved through automation, audit management
systems configuration and design, cloud-based solutions, visualization, data analytics, and process mining. While they can be deployed in
isolation, the power of digital is in their combination. As such, internal audit functions need a clear digital strategy and coordinated approach
across the function. For further information on how functions should approach a purpose-driven and digitally powered future, we
recommend reading our Internal Audit 4.0 framework.
11
Navigating the artificial intelligence frontier | An introduction for internal audit
What should internal audit be doing to accelerate
its adoption of AI?
Increase your digital fluency
01 Start engaging with learning and development now. You do not need everyone to become data scientists,
engineers, or digital experts. However, being familiar with the terminology, types of capability, and
potential for these tools will help accelerate adoption.
Determine your digital strategy and potential
Determine how GenAI can help you achieve your broader functional strategy and outcomes.
Systematically review your ways of working to identify potential use cases. But do not limit your digital
02 strategy to just GenAI; there are many applications and use cases relating to other areas of machine
learning, such as natural language processing, sentiment analysis, topic modeling, linear regression, and
neural networks that can already be harnessed and provide opportunities for experimentation. Equally,
do not overlook the opportunities that exist from maximizing functionality from audit management
systems and embracing analytics, visualization, and other tools such as process mining.
Engage with your technology teams
03
Understand your organization’s stance toward AI, both from a data privacy and security perspective and
its appetite for shaping existing solutions within the safety of your organization’s environment.
Clean up your data
The quality of AI both in terms of its training and its output will be a product of the quality of data it is
given. Many organizations (including internal audit) have poor data quality, version control, or outdated
04 versions of documents that have not been removed from intranets for years. While you are waiting for
some of the tools to become more accessible, getting your house in order will pay dividends to the value
AI can deliver. For example, analyzing your risk and control frameworks, scope documents, findings, and
recommendations to create a tokenized database of internal audit content could help turn currently
untapped information into a goldmine of knowledge and insight.
Work through and manage the risks
The risks outlined in this publication are as relevant to internal audit’s use of GenAI as they are to the
05
business. Good governance is critical, and functions should be challenging themselves to put in robust
governance processes and controls around the use, development, testing, access, and ongoing
monitoring of AI within internal audit.
Develop a culture of innovation
Organizational culture can make or break the success of innovative technology and new ways of working.
06 The limits of what GenAI could be used for are only contained by the imagination of individuals. Functions
that have a culture of innovation, curiosity, and the willingness to experiment have usually fared better
than those that were less willing to embrace change. Functions should consider innovation programs,
encourage experimentation, and reward the right behaviors.
12
Navigating the artificial intelligence frontier | An introduction for internal audit
Where do we go from here?
Whether it is the assurance over risks posed by AI or exploring how you might use these technologies in your own internal audit
function, learning and improving your digital fluency is key. GenAI will also require a mindset shift. Its prevalence and the speed at which
it is evolving will drive a need to reimagine the human-technology relationship. Interacting with GenAI will become part of the daily
routine, enabling new possibilities but bringing the potential for overreliance on AI outputs.
Organizations, including internal audit functions, will need to assess the risks and opportunities associated with GenAI, balancing the
benefits from efficiencies gained through reduced manual effort with the need to check and verify accuracy. What is clear is that tools
this powerful offer so much potential that they will be here for the long term. The attention given to GenAI and the investments being
made means internal audit will need to engage and do so quickly.
GenAI is a fast-moving and developing field of AI. As an organization, we are taking the same journey as many of our clients. We believe
GenAI has the potential to transform the internal audit profession and have already made significant investments in both our approach
to assuring GenAI and supporting organizations in their use of these technologies. If you would like to talk to our dedicated team of
specialists, please get in touch.
Internal audit contacts
Sarah Fedele Neil White Michael Schor Geoffrey Kovesdy
Principal Principal Principal Principal
US Internal Audit Leader Deloitte & Touche LLP Deloitte & Touche LLP Deloitte & Touche LLP
Deloitte & Touche LLP [email protected] [email protected] [email protected]
[email protected]
GenAI internal audit contacts
Explore Internal Audit 4.0
framework for further
insights on the power of
Alex Vorpahl Madeline Mitchell
digital for internal audit
Senior Manager Senior Manager
Deloitte & Touche LLP Deloitte & Touche LLP
[email protected] [email protected]
13
This publication contains general information only and Deloitte is not, by means of this publication,
rendering accounting, business, financial, investment, legal, tax, or other professional advice or
services. This publication is not a substitute for such professional advice or services, nor should it be
used as a basis for any decision or action that may affect your business. Before making any decision or
taking any action that may affect your business, you should consult a qualified professional adviser.
Deloitte shall not be responsible for any loss sustained by any person who relies on this publication.
As used in this document, “Deloitte” means Deloitte & Touche LLP, a subsidiary of Deloitte LLP. Please
see www.deloitte.com/us/about for a detailed description of our legal structure. Certain services may
not be available to attest clients under the rules and regulations of public accounting.
Copyright © 2024 Deloitte Development LLC. All rights reserved. |
207 | deloitte | us-lshc-ai-medtech-2024.pdf | Is Generative AI changing
the game for medtech?
Deloitte research suggests while GenAI has
started delivering value, its potential is far greater
Is Generative AI changing the game for medtech?
Contents
Executive summary 3
AI and GenAI have started to deliver value to
medtech companies across functions 4
GenAI could help medtech firms achieve cost-efficiencies
of up to 12% of their revenue 6
Building blocks to maximize value from AI and GenAI 8
Conclusion 10
Authors 12
2
Is Generative AI changing the game for medtech?
Executive summary
Across industries, artificial intelligence (AI) and Generative However, using AI and GenAI in isolation may not drive this
AI (GenAI) discussions are moving from potential to value transformative value. Adopting a string-of-pearls approach—
realization.1 For medtech companies, this value can be in integrating multiple GenAI use cases with other AI technologies,
the form of cost reductions, cost avoidance, and new data, and digital tools—is important for achieving cost
revenue generation. efficiencies and other benefits. This requires appropriate
“building blocks” to scale AI and GenAI, including creating an
To assess where medtech companies have realized value2
enterprise ambition for AI use and operating structures such as
from AI and GenAI and what could be next, the Deloitte
a Center of Excellence. These should be accompanied by new
Center for Health Solutions surveyed 85 leaders from medtech
ways of working, governance for responsible and ethical AI use,
organizations during the summer of 2024 and conducted
and proactive communication of an AI value narrative.
follow-up interviews. We found that:
• AI and GenAI have begun to deliver value across functions,
with 42% of surveyed executives reporting benefits in product
development and 35% in IT and cybersecurity functions.
• GenAI could enable medtech companies to achieve cost
efficiencies of 6% to 12% of their total revenue in the next two
to three years. This could be in the form of cost reductions,
cost avoidances, and other benefits. For a medtech company
with $20 billion to $26 billion in revenue, this translates to an
estimated $1.2 billion to $3.2 billion.
3
Is Generative AI changing the game for medtech?
AI and GenAI have started
to deliver value to medtech
companies across functions
Using AI to transform processes is not new to medtech organizations “ Our strategy is to utilize AI and GenAI in
(see sidebar, “AI in action at medtech companies”). But beyond this
our products and across all functions as
anecdotal evidence at the organizational level, the industrywide
impact of AI and GenAI is not well known. Our survey results aim much as possible.”
to bridge this gap by illustrating where and how these technologies
have affected medtech companies. —VP, large medtech company
Our research suggests that, overall, medtech companies have
moved quickly to leverage GenAI’s capabilities. Fifty-seven percent of
surveyed leaders reported that their organization is implementing or
scaling GenAI use cases or “quick wins” that have provided benefits
across functions. The pace of GenAI adoption also appears to have
accelerated the use of other AI, as our survey data indicates.
AI in action at medtech companies
Medtech companies have used AI to save time,
reduce costs, avoid additional expenses, and
create revenue. Here are some examples:
Digital asset procurement optimization: Siemens Personalized technology: Meticuly uses AI and 3D
Healthineers developed an AI-powered digital asset printing to deliver custom medical implants within
management repository, enabling its 60,000 employees two to seven days. Meticuly’s ML algorithms analyze
to search for, use, and reuse digital assets, such as stock the patient’s natural bone structure, including any
imagery, for marketing. This initiative saved the company an defects or irregularities and the impacted area,
estimated EUR 3.5 million by reducing the need to purchase to design a bespoke implant. Such implants could
new digital assets.3 reduce the likelihood of interoperative challenges.5
People analytics and workforce development: Johnson
and Johnson leveraged AI to transform and modernize
its HR operations. AI-driven models predict employee
attrition based on industry trends, performance, and career
progression. Machine learning (ML) is applied to human
capital data to assess the state of skills in the workforce and
help create personalized development plans and learning
curricula for employees.4
4
From code to cure, how Generative AI can reshape the health frontier | The shifting health care market landscape
Respondents reported that their organizations have generated the
greatest value from AI and GenAI in product development (42%)
and IT and cybersecurity (35%) functions (see figure 1). For product
development, medtech companies have used AI for concept design
and prototyping.6 Beyond device design, GenAI has benefited some
organizations in summarizing scientific literature and drafting trial
documents, such as vendor contracts and site checklists.
For IT and cybersecurity, organizations have used GenAI to generate
high-quality code and improve data management processes. This
includes creating meta labels, cleaning data, and anonymizing data
to enhance security and usability.7
While close to one-third of surveyed leaders said they have already
realized value from AI and GenAI across functions, another third
expects to realize value across most functions over the next two
to three years. In commercial operations, AI could help analyze
sales and customer data to provide the next-best engagement
recommendations to sales teams and improve conversion rates.
Leveraging AI-driven platforms could accelerate the creation, review,
and deployment of marketing and sales content to cut marketing
spending.8 One interviewee reported that his organization is using AI
5
ROF
ESU
IANEG
Is Generative AI changing the game for medtech?
Figure 1: Value realized and expected from AI and GenAI use across functions
Product IT and Commercial HR and other Supply chain and Finance
Functions development cybersecurity internal services manufacturing
Have already realized
value from AI and GenAI 42% 35% 30% 23% 22% 20%
Expect to realize value 35% 28% 37% 31% 37% 36%
from AI and GenAI in the
next 2 to 3 years
GenAI use cases being scaled
Study document Report generation Optimizing marketing Training content Quality event Financial insight
generation from unstructured content creation creation management generation enterprise data sources
Clinical study
summarization and Data management Medical, legal, and HR portal content Inventory Collection
insight generation regulatory review creation optimization automation
Scientific writing A du et bo um ga git ne gd K sO uL m p me ars rip ze ac tit oiv ne Ma ck oi mng m c uo nn it ce an tit o a nn d Pr ao nd du c dt e t lr ia veck rying M&A due diligence
more inclusive
Q. In which of the following areas has your organization realized value from the use of AI including GenAI? In which of these areas does it expect to realize value in
the next 2 to 3 years? Respondent to select only one option. [N=85]
Q. Where has GenAI helped your organization to realize benefits? Is your organization scaling the use of GenAI for this purpose? Please select all that apply.
Respondents answered this question based on their visibility into GenAI use across functional areas as assessed by our screener questions. Figure shows the top 3
GenAI use cases being scaled across functional areas.
Commercial [N=58], Supply chain and manufacturing [N=63], R&D [N=69], Finance [N=54], HR and other internal services [N=55] and IT and cybersecurity [N=68]
chatbots to automate order intake and solve customer
queries. Given its contextualization abilities, GenAI could further
improve self-service capabilities for medtech customers across
digital channels.9
As part of digitalized supply chains, AI could automate activities
such as supplier checks and management of quality events
while optimizing inventory management, distribution, and
warehousing. For a faster path to net-zero, AI could provide insights
to optimize capital investment and energy usage and cost-effectively
reduce emissions.10
AI can also assist as a self-service partner for business leaders
to support agile financial decision-making. For instance, AI
could provide quick access to complex financial insights such as
scenario analysis for budget planning. Continuous financial data
monitoring using AI and advanced analytics could also help uncover
valuable investment and savings opportunities, which could
improve margins.11
Is Generative AI changing the game for medtech?
GenAI could help medtech firms
achieve cost-efficiencies of up to
12% of their revenue
By effectively deploying GenAI and other AI technologies, medtech Overall, our survey respondents anticipate AI and GenAI to
companies could achieve cost efficiencies, including cost reductions, reduce SG&A costs by 7% to 19% over the next two to three years,
cost avoidances, and other benefits, ranging from 6% to 12% of which could benefit the commercial and shared services functions,
revenue. For instance, a large medtech company with $20 billion to including HR, finance, and IT. For a large medtech company, this
$26 billion in revenue could realize cost efficiencies of $1.2 billion to could mean an estimated savings of up to $1.5 billion through
$3.2 billion through AI implementations across functions in the next efficiency gains and vendor cost optimizations. Similarly, a large
two to three years (figure 2). Actual efficiencies gained could vary medtech company could save up to $1.4 billion (5% to 12%) in
depending on the scale and integration of AI within an organization. COGS by applying AI to activities such as predictive maintenance,
contract creation, and vendor management in supply chain
To arrive at these estimates, we analyzed anticipated cost savings
and manufacturing.
percentages from AI and GenAI use across the various categories—
selling, general, and administrative costs (SG&A); R&D expense; As discussed previously, more surveyed leaders reported
and cost of goods sold (COGS)—as reported by surveyed medtech realizing value from AI and GenAI in product development than
leaders. We applied these percentages to publicly reported financial other functions. According to survey respondents, applying AI and
data from the top 10 medtech companies by revenue in 2023.12 GenAI could save up to 20% of R&D costs, translating to $0.3 billion
in savings for a large medtech company over the next two to
three years.
Figure 2: Estimated potential efficiency gains via AI use for a large medtech company in the next 2–3 years
Potential savings %
Top 10 medtech firm Functional areas likely
Type of costs from use of AI in the Potential savings estimate
average cost* to benefit
next 2–3 years*
Selling, general and
Commercial, Shared
administrative costs $7.2B 7%-19% $0.5B - $1.5B
services
(SG&A)
Supply chain and
Cost of goods sold (COGS) $11.7B 5%-12% $0.6B - $1.4B
manufacturing
R&D costs $1.5B 7%-20% $0.1B - $0.3B Product development
Estimated potential savings for a large medtech company
$1.2B - $3.2B
in the next 2–3 years
Methodology:
* Potential dollar savings are based on applying savings estimates collected through our survey to the average of different costs—R&D expenses, cost of
goods sold (COGS), and selling, general, and administrative (SG&A)—reported in the financial statements of the top 10 medtech companies in 2023. Cost
savings percentages applied are based on interquartile ranges from the survey data.
Q. To your best estimate, what share of cost savings could AI including GenAI enable your organization to achieve in the next 2 to 3 years? Please pick the %
of impact AI has had across the following cost categories. (N=85)
6
Is Generative AI changing the game for medtech?
However, using AI and GenAI in isolation may not drive one another to drive efficiencies and compound value (see sidebar,
transformative benefits. Medtech companies may need to undertake “Enhancing health care practitioner engagement with a string-of-
a string-of-pearls approach. By stringing together multiple workflows pearls approach”). Such an approach can be applied to processes
through GenAI and other AI, data, and digital technologies, across functions—from R&D to commercial and shared services.
companies could transform entire processes. This approach
creates a series of business process enhancements that build on
Enhancing health care practitioner engagement with a string-of-pearls approach
A string-of-pearls approach with GenAI at the core could enable companies to deliver appropriate content to
health care practitioners (HCPs) through appropriate channels at an appropriate time.
As part of this approach, GenAI and ML could analyze prescriptions, sales data, and digital interaction data to
micro segment and target HCPs for engagement. Based on this segmentation and AI-driven insights on content
consumption patterns, GenAI could categorize, organize, and create new content. ML algorithms could then
determine the sequence and frequency of content delivery across channels and measure the effectiveness of
these tactics to optimize omnichannel engagement (figure A).
Figure A: Orchestrating omnichannel engagement with GenAI
“Next-best action” objectives Use cases GenAI ML Descriptive analytics Data
Understand the
1 e wn hv icir ho n thm ee bn rt a i nn d C ino tm elp lige eti nti cv ee mar bS k rc e ar t na , dp co e im
n
a tn p ed e llit gd it ei is v nt e cil , el and Geo-lev ree pl i on rs ti ig nh gt , trend pubW lie cb d- is sc cr la op se ud re d s,a Ata P; LD
operates
Find and engage with
HCPs based on hidden HCP micro- Identify “latent” drivers Segment HCPs based on APLD data, digital
2 potential and segmentation of customer behavior prescribing patterns HCP 360 profiles interactions, sales
disposition
actionable insight
3 C o pr ea g rte a sog no nizr aei lz ic ze o a a n tin t oed nn t for cat (teC ago gon grt iie nzn a gt )t ion Tag con tt ae xn ot n a os mse yts using Content i nc so in gs hu tmption Standard ci oz ned te t na txonomy,
4 D b coue nnv te d el lo nep t n a en wd geC no en rt ae tn iot n tailG oe ren de r ta ot e p rc eo fn et re en nt c es Conte pn rt e e fff ere ec nti cv eeness/ maM pe ps is na gg te o/ c so en gt men et n ts Firs pt- up ba lr isty h ea rn ad l lt ih ai nr cd e-p sarty
5 D e cuxe p sli tev ore mir e enta rc sr eg se tt oe d O enm cu gn asi t gc o eh m man e en r n e tl Behavio dr ea tl e-b cta is oe nd signal fC rh eqa on u pn e tne iml c, y s i, z e c aq a tu id oe e nn n c ce e, Call plan ps l, a p na sid media loC ig nR s s,M i gc h ud tsa st ,t o a mm, c eea drl il s a c e pe gn lm at ne e sr n t
Measure Closed loop Automated cross- Measure effectiveness of Measure audience reach APLD data, digital
6 effectiveness of measurement channel, cross-audience tactics and quality marketing data, HCP
tactics interpreter MTA interaction data
Source: Deloitte analysis
7
Is Generative AI changing the game for medtech?
Building blocks to maximize value
from AI and GenAI
Interviewees highlighted that though many medtech organizations leveraging less than 10% of AI’s potential.” We outline six key building
have kick-started their AI and GenAI journey, they may just be blocks for medtech companies to leverage AI and GenAI to a greater
scratching the surface. As one interviewee commented, “I feel we are extent (figure 3).
FigurFei 3g:u Brueil d4i:n Bg ubiloldckins gfo br lloevcekrsa gfionrg lAeI vaenrda GgeinnAgI AI and GenAI
1 Craft a strategic blueprint: Anchor AI investments to meet overarching business
objectives and focus on 3 to 5 high-value opportunity areas to scale AI.
2 Build operating structures: Establish measurable AI targets and possibly a Center of
Excellence (COE) to align investments and resources, supporting prioritized AI and
GenAIuse cases.
3 Focus on value realization: Take a disciplined approach to evaluate, approve, and
adopt AI and GenAIto help ensure value.
4 Create tech and scaling capabilities: Select an appropriate technology platform to
leverage LLMs while forging ecosystem partnerships to scale AI and GenAI.
5 Build new ways of working: Emphasize AI's role in enhancing human capabilities, and
develop training programs to improve AI literacy and optimize workflows.
6 Promote responsible AI use: Establish governance to address data privacy, security,
and bias, and foster the accountable and ethical deployment of AI.
Source: Deloitte analysis
Source: Deloitte analysis
1. Craft a strategic blueprint 2. Build operating structures
Scaling AI across the enterprise may no longer be a question of Building distinct operating structures, such as establishing an
“when” but “to what extent.” Medtech companies should view AI AI Center of Excellence (COE), could help medtech companies
and GenAI investments as core enablers of their enterprise strategy cross-leverage talent and resources to weave AI and GenAI into
rather than as experimental projects. While executing a multitude of their business. Such COEs, while not necessarily a requirement
use cases can deliver value, a focus on depth versus breadth could for success, should enable prioritizing and managing AI and GenAI
be more beneficial. investments and should include an explicit capability for value
realization.
Action steps:
Action steps:
• Frame a strategic blueprint: Define the business ambition
for AI use and anchor investments to meet short- and long-term • Source and prioritize use cases: Identify and prioritize
business goals. a portfolio of AI and GenAI use cases relevant to identified
opportunity areas.
• Identify bold plays: Determine three to five high-value
opportunity areas where AI and GenAI use could provide the most • Create a roadmap: Develop a plan to align investments and
value and focus on scaling for impact. resources to support use-case development and scaling.
• Set measurable AI targets: Build explicit metrics of value to be
captured through use-case enablement. Assigning measurable
targets to AI leaders can further promote AI advocacy and ensure
8
accountability.
Is Generative AI changing the game for medtech?
3. Focus on value realization 5. Build new ways of working
Sustaining the enthusiasm and momentum to scale AI requires While initial experiments may have progressed, large-scale AI
a disciplined approach to assessing and making appropriate deployment involves a stronger focus on bridging the trust gap
investments and communicating their value. AI appears to be just with AI, which GenAI could likely exacerbate. As one interviewee
another step on the digital transformation path for companies. By pointed out, “We want to make AI a part of our culture and part
focusing on building and communicating an ongoing value narrative and parcel of the way we do things going forward.” This involves
from AI use, companies could prevent disillusionment, especially as fostering willingness among business users by clearly demonstrating
newer digital technologies become the next “shiny thing.” value, providing adequate training, and creating guardrails for the
responsible use of AI and GenAI tools.
Action steps:
Action steps:
• Track and realize value: Create mechanisms to measure
outcomes and optimize investments while assessing progress • Focus on human augmentation to encourage adoption:
toward AI ambitions. During rollouts, demonstrate how AI enhances human capabilities
rather than just automates tasks.
• Articulate an enterprise value story: Create and promote a
narrative of AI impact on identified opportunity areas to relevant • Build AI fluency: Develop training programs to improve AI
stakeholders—from the board and CEO to business leaders. literacy, covering AI concepts, applications, and limitations.
Optimize workflows to ensure seamless AI integration and user
proficiency.
4. Create technology and scaling capabilities
Build enterprise architecture that collates key components required
6. Promote responsible AI use
to quickly create and deploy AI and GenAI applications, tools, and
capabilities. Such architecture should include access to open-source Medtech companies should establish clear guidelines and promote
and proprietary platform capabilities to utilize large language ethical practices around AI to help ensure responsible deployment.
models, sandbox environments, and common solution archetypes. The string-of-pearls approach can only be effectively leveraged if
companies ensure AI use is harmonized with the evolving regulatory
Action steps:
environment.
• Determine an appropriate platform(s): Select a cost-effective
Action steps:
platform, or platforms, to support capability development (e.g.,
semantic search, information extraction, classification, simulation • Establish ground rules: Govern the ethical use of AI by
exercises) tailored to execute multiple GenAI use cases. addressing issues such as data privacy, security, and bias.
• Build ecosystem partnerships: Identify and work with industry • Shape internal governance: Enable business users to
consortia and working groups as they emerge. Continuously understand and contend with risks from AI use (e.g., data security
evaluate platforms and partnerships to enhance AI and GenAI and privacy) and model outputs (e.g., hallucinations and biases,
capabilities and infrastructure. lack of transparency).
• Promote transparency and accountability: Clearly document
processes and communicate openly about AI model/tool
capabilities and limitations.
99
Is Generative AI changing the game for medtech?
Conclusion
The medtech industry appears to be at an inflection
point—scaling AI and GenAI for transformation is
important. Organizations should consider making
“bold plays” that integrate these technologies into
their operations.
In upcoming publications, we will focus on how
medtech companies can execute bold plays through
a string-of-pearls approach and the impact these
could have on functions such as R&D, supply chain,
and commercial.
10
Is Generative AI changing the game for medtech?
Endnotes
1. Jim Rowan et al., Now decides next: Moving from potential to performance, August 2024.
2. F or our survey respondents, we defined value as cost savings, cost eliminations, new sources of revenue, improvements or additions to
products and services, improved customer relations, and other benefits organizations may have achieved through AI and GenAI use.
3. Bynder, “Siemens Healthineers saves millions in costs with Bynder’s AI-powered DAM,” accessed August 8, 2024.
4. Finn Bartram, “AI and automation: Suresh Raman of Johnson & Johnson on how to effectively harness AI technology in people operations,”
Authority Magazine, November 20, 2023.
5. Y Consulting, “Meticuly: Revolutionizing the future of medical implants with AI and 3D printing,” August 25, 2024.
6. Mark Crawford, “Artificially intelligent design for orthopedic devices,” Orthopedic Design & Technology, March 10, 2023.
7. Vikram Agarwal, “2024 J.P. Morgan Healthcare Conference: As GenAI Expands Medtech Capabilities, the Industry Gains Momentum,”
LinkedIn, January 25, 2024.
8. Anthill, “AI in pharma marketing: Meaning, strategy and best practices,” accessed September 20, 2024.
9. DigitalOcean, “How to use AI for sales: Techniques and tools,” accessed September 19, 2024.
10. Tanya Gupta, “AI in supply chain: Use cases, examples and how AI [sic] used in supply chain management?,” Clear, updated August 8, 2024.
11. Michael Abramov, “The human-AI collaboration: How humans and AI can work together in finance,” Keymakr Blog, August 2, 2024.
12. Medical Device and Diagnostic Industry (MD+DI), “Top 40 medical device companies,” February 29, 2024.
13. Nitin Mittal, “Getting real about GenAI,” Deloitte Insights, April 1, 2024.
11
FIsr oGmen ceordaeti tvoe cAuIr ceh, ahnogwin Gge tnheer gaatimvee AfoI rc amne rdetsehcahp?e the health frontier | Unlocking new levels of efficiency, effectiveness, and innovation
Authors
Sheryl Jacobson Dr. Jay Bhatt
US Consulting Medtech Practice Leader Managing Director
Deloitte Consulting LLP Deloitte Center for Health Solutions and
[email protected] Deloitte Health Equity Institute
Deloitte Services LP
[email protected]
Mukund Lal Wendell Miranda
Senior Manager Deputy Manager
Deloitte Consulting LLP Deloitte Services LP
[email protected] [email protected]
Aditya Kudumala Apoorva Singh
Principal Senior Research Analyst
Deloitte Consulting LLP Deloitte Services LP
[email protected] [email protected]
Srivathson Chennakesavan Leena Gupta
US Life Sciences MedTech Analytics & Life Sciences Research Leader
Cognitive Leader Deloitte Center for Health Solutions
Deloitte Consulting LLP Deloitte Services LP
[email protected] [email protected]
Dr. Asif Dhar
Global Life Sciences and Health Care
Consulting Leader
Deloitte Consulting LLP
[email protected]
Acknowledgments:
The authors would like to thank Maulesh Shukla, Natasha Elsner, Dana Schmucker, Tomislav Medan, Spencer Hanson,
Adam Israel, and Rob Jacoby for their insights, expertise, and critical feedback on the research. Additionally, the authors
would like to thank Rebecca Knutsen, Laura DeSimio, Chris Giambrone, Jesse Daniels, Deb Asay, and many others who
contributed to the success of this project.
12 12
This publication contains general information only and Deloitte is not, by means of
this publication, rendering accounting, business, financial, investment, legal, tax, or
other professional advice or services. This publication is not a substitute for such
professional advice or services, nor should it be used as a basis for any decision or
action that may affect your business. Before making any decision or taking any action
that may affect your business, you should consult a qualified professional advisor.
Deloitte shall not be responsible for any loss sustained by any person who relies on
this publication.
About Deloitte
Deloitte refers to one or more of Deloitte Touche Tohmatsu Limited, a UK private
company limited by guarantee (“DTTL”), its network of member firms, and their related
entities. DTTL and each of its member firms are legally separate and independent
entities. DTTL (also referred to as “Deloitte Global”) does not provide services to
clients. In the United States, Deloitte refers to one or more of the US member firms of
DTTL, their related entities that operate using the “Deloitte” name in the United States
and their respective affiliates. Certain services may not be available to attest clients
under the rules and regulations of public accounting. Please see www.deloitte.com/
about to learn more about our global network of member firms.
Copyright © 2024 Deloitte Development LLC. All rights reserved.
9709314 |
208 | deloitte | us-state-of-gen-ai-report-q2.pdf | Now decides next:
Getting real about
Generative AI
Deloitte’s State of Generative AI in the Enterprise
Quarter two report
April 2024
deloitte.com/us/state-of-generative-ai
Table of contents
Foreword
Introduction
Now: Key findings
1 Value creation
2 Scaling up
3 B uilding trust
4 E volving the workforce
Next: Looking ahead
Authorship & Acknowledgments
About the Deloitte AI Institute
About the Deloitte Center for Integrated Research
About the Deloitte Center for Technology, Media & Telecommunications
Methodology
22
Introduction
Foreword
We have traveled a long way since the Generative AI space race kicked off in November 2022—and
yet, we know we are still at the beginning of this long and exciting transformation. Every day, we
talk with clients about how much there is to focus on in the moment, how explosive the pace of
change is, and how challenging it can be amid the excitement to take a longer-term view.
We see organizations starting to achieve benefits and move toward a near future where this
“We are in the first inning of a
early stage of Generative AI tools is widely dispersed and driving new value. But there are also
some hard realities to deal with as business leaders look to scale and realize the potential of this thousand-inning game and there’s
powerful technology.
so much to be figured out.”
The State of Generative AI in the Enterprise: Getting real about Generative AI captures a new snapshot
of this transformative time from the perspectives of nearly 2,000 business and technology leaders,
all from organizations that are actively deploying and scaling Generative AI today. Echoing our -Chief analytics officer in financial services
many clients, from these executives we hear that while excitement persists it may be at its peak as
leaders come up against cultural challenges, questions about how to manage their workforces, and
issues with trust that—at least for now—stand in the way of unlocking Generative AI’s full value.
All told, it is exciting that Generative AI’s potential is beginning to weave its way deeper into the
foundations of how organizations operate and we continue to learn more about emerging leading
practices. Amid those developments, we also continue to see that achieving value with Generative
AI connects hand in hand with keeping humans at the center.
Learn more about the series and sign up for updates at http://deloitte.com/us/state-of-generative-ai.
Nitin Mittal, Costi Perricos, Kate Schmidt, Brenna Sniderman and David Jarvis
3
Introduction
Getting real about Generative AI
Is the infatuation phase over? Quarter two of Deloitte’s Our research shows that organizations are increasingly Two of the most critical challenges for scaling are
global quarterly survey found many organizations prioritizing value creation and demanding tangible building trust (in terms of making Generative AI
beginning to get down to the serious work of making results from their Generative AI initiatives. This requires both more trusted and trustworthy) and evolving
Generative AI’s vast potential a reality. them to scale up their Generative AI deployments— the workforce (addressing Generative AI’s potentially
advancing beyond experimentation, pilots and proofs massive impact on worker skills, roles and head count).
This report presents findings from the second in
of concept. Transitioning to large-scale deployments
Deloitte’s ongoing series of quarterly global surveys Here we’ll take an in-depth look at all four of these
will increase Generative AI’s impact on the business
on Generative AI in the enterprise. To gain additional areas—value, scaling, trust and workforce—to help
and expand its reach to a much larger segment of the
context for our wave two research, we also organizations move forward more effectively on their
workforce. Successful scaling, in turn, presents a wide
conducted a series of in-depth interviews with senior Generative AI journeys. Future survey reports will
range of challenges, encompassing everything from
executives from a broad range of industries. focus selectively on other key challenges to successful
strategy, processes and people to data and technology.
Generative AI scaling and value creation.
4
Introduction
Getting real about Generative AI (cont’d)
Value creation Scaling up
• T he percentage of organizations reporting they were already achieving their expected • L eaders see scaling as essential for creating value, increasing Generative AI’s impact
benefits to a “large” or “very large” extent is 18%–36%, depending on the type of on the business and expanding the technology’s user base. The scaling phase is when
benefit being pursued. Generative AI’s potential benefits are converted into real-world value. It’s also, however,
when an organization’s potential concerns can become real-world barriers to success.
• O rganizations that reported “high” or “very high” levels of Generative AI expertise
are scaling Generative AI much more aggressively—and are achieving their desired • Common areas of concern include data security and quality, explainability of Generative AI
benefits to a much greater degree than others. outputs, and worker mistrust or lack of familiarity with Generative AI tools.
• O rganizations primarily plan to reinvest the savings from Generative AI into innovation • W orkforce access to approved Generative AI tools and applications remains quite low,
(45%) and improving operations (43%)—addressing the value equation from both sides. with nearly half of surveyed organizations (46%) reporting they provided approved
Generative AI access to just a small portion of their workforces (20% or less). However,
most workers with internet access will have access to public Generative AI tools and
could be using them without consent.
All statistics noted in this report and its graphics are derived from Deloitte’s second quarterly survey, conducted
January – February 2024; The State of Generative AI in the Enterprise: Now decides next, a report series. N (Total
leader survey responses) = 1,982.
Generative AI is an area of artificial intelligence and refers to AI that in response to a query can create text,
images, video and other assets. Generative AI systems can interact with humans and are often built using large
language models (LLMs). Also referred to as “GenAI.”
5
Introduction
Getting real about Generative AI (cont’d)
Building trust Evolving the workforce
About the State of
• Lack of trust remains a major barrier to large-scale • M ost organizations (75%) expect the technology to Generative AI in the
affect their talent strategies within two years; 32%
Generative AI adoption and deployment. Two key
Enterprise: Wave two
of organizations that reported “very high” levels of
aspects of trust we observed are: (1) trust in the quality
survey results
Generative AI expertise are already making changes.
and reliability of Generative AI’s output and (2) trust
from workers that the technology will make their jobs
The wave two survey covered in this report was fielded
• T he most expected talent strategy impacts are process
easier without replacing them. to 1,982 director- to C-suite-level respondents across
redesign (48%) and upskilling or reskilling (47%).
six industries and six countries between January
• T rust issues have not prevented organizations from and February 2024. Industries included: Consumer;
• Generative AI is expected to increase the value of
Energy, Resources & Industrials; Financial Services;
rapidly adopting Generative AI for experiments
some technology-centered skills (such as data analysis) Life Sciences & Health Care; Technology, Media &
and proofs of concept, with 60% reporting they
Telecom; and Government & Public Services. Our
as well as human-centered skills (such as critical thinking,
are effectively balancing rapid implementation with Q2 survey findings are augmented with over 20 executive
creativity and flexibility), while decreasing the value of
interviews. This second report is part of a yearlong
risk management. Trust is likely to become a bigger
other skills. series by the Deloitte AI Institute to help leaders
issue, however, as organizations transition to large-
in business, technology and the public sector track
scale deployment. Many reported they are currently • I n the short term, more organizations said they expect the rapid pace of Generative AI change and adoption.
The series is based on Deloitte’s State of AI in the Enterprise
investing significant time and effort into building the technology to increase head count (39%) than to
reports, which have been released annually the past five years.
guardrails around Generative AI. decrease head count (22%)—perhaps due to increased
Learn more at deloitte.com/us/state-of-generative-ai.
needs for Generative AI and data expertise.
• Organizations that reported “high” or “very high” levels of
expertise recognize the importance of building trust in
Generative AI across numerous dimensions (e.g., input /
output quality, transparency, worker empathy) and are
implementing processes to improve it to a much greater
extent than are other organizations.
6
Now: Key findings
77
Now: Key findings
1
Value creation
Proving, measuring and communicating value is crucial Value objectives and priorities for Generative AI can— Therefore, many forward-thinking organizations are
to an organization’s Generative AI journey. In our survey and should—vary by organization, industry and use case. implementing Generative AI without specific ROI targets
and interviews, many organizations reported they Where the technology’s potential impact is strategic as they realize they can’t afford to get left behind in this
were increasingly emphasizing the need for Generative and truly game-changing, the need and latitude for critical and fast-moving market.
AI initiatives and investments to have clear value experimentation, learning and innovation are much
objectives and deliver tangible results, rather than simply greater (with less emphasis on immediate payback) than
being viewed as experiments or learning experiences. in situations where productivity and cost savings are the
primary expected benefits.
As one executive at a Fortune 500 manufacturing
company noted: “We have a very strict internal rule Moreover, Generative AI is so new—and advancing so
that if we don’t see value from our Generative AI quickly—that accurately estimating benefits is much
solutions, we won’t do it or we won’t scale it.” harder than for an established technology with a proven
track record.
That said, there are many ways to define and
measure value—especially for a technology with the “Any technology that’s a little over a year old, nobody’s
transformational potential of Generative AI. Although going to have a year’s worth of data to do a backward-
financial return on investment (ROI) is important, value looking ROI,” said one tech company executive we
drivers such as innovation, strategic positioning and interviewed. “And with the fundamental and foundational
competitive differentiation can be even more important. changes Generative AI offers, it’s very hard to even offer
a forward-looking [total cost of operating] or ROI because
there’s so many possibilities of impact and varied ways
to integrate it into your business.”
88
Now: Key findings
Organizations are starting to demand tangible “large” or “very large” extent is 18%–36%, depending on Generative AI “experts” are achieving their desired
business value from Generative AI, and some the type of benefit being pursued. benefits to a much greater degree.
are beginning to achieve real-world benefits.
As one public sector executive told us, “The big selling In every category, organizations that rated themselves
point is if I make an investment and do something like
The organizations we surveyed expect Generative AI to as having “high” or “very high” levels of Generative AI
this, what’s the tangible return and what are some easy
deliver a broad range of benefits, with the most common expertise reported much greater success at achieving their
returns? And then what are more complicated longer-term
objective—at least in the short term—being improved desired benefits. Their advantage was greatest in strategic
returns that take more investment money? If I can do some
efficiency and productivity (56%), which is consistent with and growth-related areas such as improving products and
of the easier ones and build on them, it can translate into
the results from last quarter’s survey. The percentage of services and encouraging innovation and growth.
‘I think this would be worth it to invest a lot more money.’ I
respondents who said their organizations’ Generative AI
believe a lot of entities in our sector are at that point.”
initiatives were already achieving expected benefits to a
Achieving benefits
Of those seeking the benefit, the percentage of respondents achieving the benefit to a large extent or more
Very high expertise 70% 63% 54% 55% 48% 48% 48% 40% 36% 42%
Overall 28% 35% 27% 18% 36% 30% 25% 29% 30% 22%
Improve existing Encourage Improve Reduce Increase speed / Uncover new Increase Enhance Detect fraud / Shift workers
products innovation efficiency and costs ease of dev new ideas and revenue relationships with manage risk from lower- to
and services and growth productivity systems / software insights clients / customers higher-level tasks
Figure 1 Q: What are your anticipated benefits and to what extent are you achieving those benefits to date?
(Jan./Feb. 2024); N (Total) = 1,982; N (very high) = 96
9
Now: Key findings
“Expert” organizations are scaling Generative AI According to our survey, organizations reporting “very high” expertise reported, on average, implementing
much more aggressively. high” levels of Generative AI expertise are deploying AI at scale in 1.4 functions, out of eight total functions,
much more rapidly and extensively than others. In fact, while those with “some” expertise are doing so in only
Generative AI expert organizations are likely having
73% said they are adopting the technology at a “fast” 0.3 functions. Further, 38% of those with “very high”
more success at capturing benefits because they are
or “very fast” pace (versus only 40% of organizations expertise reported implementing Generative AI at
scaling up much more aggressively, compared to the
with “some” level of expertise). They are also scaling scale in marketing, sales and customer service—
other categories, which provides a larger base for
Generative AI at higher rates across functions and using versus only 10% of organizations with “some” level of
generating benefits.
it more within functions. For example, those with “very expertise.
Companies that report expertise are moving quickly.
80%
73%
66% 64%
61% 62%
47%
48%
40%
39%
34%
33%
23%
19%
Adopting at a Providing more of their Adopting at higher levels Investing more in Investing more in Using code Using open-source
faster pace workforce access to across functions hardware cloud consumption generators more LLMs more
GenAI
Adopting Generative AI Implementing Generative Increasing hardware Increasing cloud Currently using Generative Currently using
“fast” or “very fast” >40% of workforce has AI for marketing, sales investment because of investment because of AI code generator open source large
access to Generative AI and customer service Generative AI strategy Generative AI strategy language models
tools / applications
Figure 2 (Jan./Feb. 2024 ) N (Total) = 1,982; N (Very high) = 96; N (Some) = 1,021 Very high expertise Some expertise
10
Now: Key findings
Insights from our executive interviews align closely with and has saved us significant amounts of money … and that are industry-specific and narrowly focused but more
survey findings, showing that leading organizations are we have scaled very broadly across many of our sites and strategically impactful (e.g., Generative AI tools for
aggressively scaling up their Generative AI efforts both continue to scale further across more equipment across semiconductor design that are used only by a small
horizontally (across multiple functions or domains) more sites.” subset of workers but have a very large impact on
and vertically (within a single function or domain). This the business).
Similarly, from a broad market perspective we are seeing
combination of horizontal and vertical scaling may help
an increasingly sharp distinction between horizontal use
achieve value creation more effectively.
cases that cut across industries (e.g., office productivity
As one chief transformation officer in manufacturing noted, suites and enterprise resource planning systems with
“[We have] an application that is being incredibly successful integrated Generative AI) and vertical use cases that
11
Now: Key findings Areas to reinvest time and cost savings
Organizations primarily plan to reinvest the strategic objectives such as innovation and growth,
savings from Generative AI into innovation and
and are likely already working more aggressively to
additional operations improvements.
develop strong Generative AI capabilities.
Among the overall respondent pool, organizations Driving 45%
By contrast, organizations in industries that are
innovation
said they primarily planned to reinvest cost
currently not being disrupted by Generative AI are opportunities
and timesavings from Generative AI into driving
more likely to focus on benefits such as individual
innovation (45%) and improving operations (43%), 43% Improving
worker productivity and operations improvement,
operations
addressing the value equation from both sides. It’s
areas with less of a sense of urgency and less Developing 29% across the
interesting to note that a significant percentage new products organization
tolerance for risk. Such organizations can still benefit
and services
of organizations (27%) also planned to reinvest in
greatly from Generative AI—just in a different way. 28% Expanding
our market
scaling Generative AI adoption, creating a cycle of
They also have a valuable opportunity to watch Scaling GenAI 27%
Generative AI reinvestment and growth. adoption across
and learn from the experiences of other industries
the organization
28% Improving
that are currently being disrupted—lessons that
Organizations with “very high” Generative AI cybersecurity
could serve them well if and when Generative AI infrastructure
expertise are even more focused than others on
Training and 23%
disruption reaches their own industry. upskilling
driving innovation (51%). They are also less inclined
employees 20% Enhancing risk
than others to reinvest savings from Generative
management
“To enable GenAI value in our business, we need to
systems
AI into improving operations and more inclined to Enhancing IT 19%
change our mindset and develop R&D capabilities
infrastructure
prioritize developing new products and services. 16% Exploring new
to realize a long-term vision enabled by GenAI,” said
business models
Creating a 13%
the CEO of a digital media company. “Right now, [our
The right reinvestment approach depends on an return for 9% Creating
mindset] is short-term and just about tangible cash shareholders new jobs
organization’s specific needs. Organizations currently
value for one-off use cases.”
facing strategic disruption or transformation from
Generative AI have a greater imperative to focus on Figure 3 Q: Where does your company plan to reinvest cost or timesavings
generated through implementation of GenAI capabilities (select top 3)?
(Jan./Feb. 2024 ) N (Total) = 1,982
12
Now: Key findings
2 Scaling up
A key to value creation, scaling increases Generative potential issues become real-world barriers. And with organization, incorporating more datasets, expanding the
AI’s impact on the business and expands its user Generative AI, many of those barriers are still being user base (internal and external) to improve upon existing
base—both of which have a strong multiplier effect on identified and understood. results, and refining the current solution for more value.
Generative AI’s benefits. Yet, many organizations find it This phased approach gives us a sense of assurance the
“There are always issues that emerge through the
challenging to make the leap from pilots and proofs of investment is worthwhile before we commit significantly
adoption and scaling transition that aren’t expected—
concept to large-scale deployment. more resources.”
the question we have to consider is how hard are they to
Scaling is complex and requires effort across a variety of overcome,” said a chief technology officer we interviewed. Off-the-shelf Generative AI solutions for common use
interrelated elements spanning strategy, process, people, “For example, [one of our] use cases had some technical, cases such as office productivity are arguably the easiest
data and technology. Although the challenges associated policy and cybersecurity issues, but they were relatively to deploy at scale, but they still require substantial
with scaling Generative AI are common to many digital easy to overcome, so we scaled. Conversely, for [two investment, effort and training. For unique and/or more
transformation initiatives, issues such as risk management other] use cases more issues emerged linked to the skill strategic Generative AI solutions and use cases, the
and governance, workforce transformation, trust and data level to work with the outputs of the AI solution. These complexity and challenges increase by leaps and bounds,
management take on even greater importance. What have been harder to address, so scaling has been slower.” along with the potential for greater returns.
worked well in the past might not work the same way with
A public sector chief information officer outlined another
this new technology.
approach: “[For us, successful scaling is] building on
The scaling phase is when potential benefits are previous successes and then taking those initiatives
converted into real-world value. It is also, however, when to another level. Expanding to other areas of the
13
Now: Key findings
Workforce access to approved GenAI tools and applications remains low. Our executive interviews pointed to a number of reasons for this overall low penetration
rate, mostly revolving around risk versus reward—especially data-related risks. Do the
Nearly half of our respondents (46%) reported they provided approved Generative AI
potential rewards of Generative AI justify the risks, and can the risks be mitigated? In
access to just a small portion of their workforces (20% or less). Organizations reporting
particular, we found widespread concern that allowing workers to use public large language
“very high” levels of Generative AI expertise are further along, with nearly half (48%)
models (LLMs) and Generative AI tools might lead to problems with protection of intellectual
providing approved Generative AI access to at least 40% of their workforces. Even for these
property and customer privacy.
“expert” organizations, worker access to approved tools remains the exception, not the rule.
Percentage of workforce with access to Generative AI
76%
Overall
49%
46%
36%
Little expertise
31% 29% 28%
27%
25%
23% 24%
16% 16%
Some expertise 14% 16%
8%
5% 3% 3% 4% 6% 6% 7%
1% 2%
High expertise
Up to 20% 20%–40% 40%–60% 60%–80% More than 80%
Very high expertise
Percentage of the workforce
Figure 4 Q: How much of your overall workforce, do you estimate, have access to your organization’s sanctioned (approved) Generative AI tools/applications?
(Jan./Feb. 2024) N (Total) = 1,982, N (Very high) = 96, N (High) = 606, N (Some) = 1,021, N (Little) = 257
14
Now: Key findings
Other concerns that came up in our executive sensitive data and intellectual property into public LLMs
interviews include: in an entirely uncontrolled way. This status is likely to
continue in the absence of practical policies for allowing
• Generative AI outputs that can be unpredictable and
and managing widespread Generative AI access.
subject to inaccuracies (i.e., “hallucinations”)—which
undermine trust, particularly when combined with Organizations should be actively developing sustainable
lack of transparency and explainability processes and policies for enabling ubiquitous but
responsible Generative AI use and managing the
• Potential loss of control over what Generative AI
associated risks at scale. Widespread but controlled
apps are being used within the organization and
access to Generative AI will help people get more
who is using them
comfortable with the technology and enable them to
understand what it can and cannot do—giving them a
• Worker resistance to using Generative AI due to lack
more realistic and informed perspective while opening
of familiarity or concerns about being replaced
the door to new opportunities for Generative AI value
Given the potential challenges and risks, a cautious
creation across the enterprise.
approach to allowing workers to use Generative AI tools
arguably makes sense. However, tight restrictions on
Generative AI are best viewed as a temporary stopgap
measure—not a viable long-term solution. Logically,
any worker with internet access will have access to
public Generative AI tools and could be using them
without their employer’s consent—potentially leaking
1155
“It has been surprising to see how
low the bar is to do something
quick and dirty—this is both
exciting and scary, but the big
challenge is to scale—this is a
whole new ball game … but scaling
is hard without centralization.”
-Director of data science and AI in the technology industry
1166
Now: Key findings Areas of strength
3 Building trust Growing trust
said their organization’s trust in
Lack of trust continues to be one of the biggest barriers According to a chief technology officer we interviewed,
72%
all forms of AI has increased since
to large-scale adoption and deployment of Generative AI. “The explainability piece is really holding us back right now
Generative AI emerged in late 2022
In this context, two key aspects of trust are: (1) trust in the … once we get a better handle on that, I think we will really
quality and reliability of Generative AI’s output (supported be able to accelerate our adoption.”
Balancing speed and risk
by improved transparency and explainability), and (2) trust
Ultimately, most organizations will likely each end up using
reported their organization is effectively
from workers that Generative AI will make their jobs easier
LLMs customized and fine-tuned for their specific balancing integrating Generative AI
and won’t replace them. 60%
rapidly while implementing processes
domain, industry and use case, rather than just scaling
that mitigate potential risks
Regarding worker trust, one executive we interviewed up a general-purpose LLM. This specificity will help
noted that “once people start seeing efficiencies and Generative AI produce outputs that are more precise,
the benefits the tools have to their work, that will drive transparent and explainable.
Opportunities for improvement
adoption and sustained success.” In other words, greater
Lack of trust and related risks have thus far not
exposure to Generative AI tools will help people become
Lacking confidence
prevented organizations from rapidly adopting
more comfortable with the technology and understand
Generative AI for experiments and proofs of
how it can help them do their jobs.
selected “lack of confidence in results” as
concept; however, this will likely change as
33% one of their top risks related to Generative
As for trusting Generative AI’s outputs, the technology’s organizations transition to large-scale deployment.
AI tools / applications (#3 of 10 overall)
fallibility in the form of “hallucinations” is well known and
According to our wave two survey, 60% of respondents
is actively being addressed through improved training
Measuring trust
believed their organization is effectively balancing
and guardrails. For many organizations, transparency and
rapid integration of Generative AI while implementing of organizations said they are measuring
explainability are even bigger issues. In its current form,
processes that mitigate potential risks. Also, 72% said their worker trust and engagement as part
36%
Generative AI still operates largely as a black box—taking
of altering their talent strategies because
organization’s trust in Generative AI has increased since the
an input and producing an output with no real way for
of the adoption of Generative AI
technology emerged in late 2022.
humans to understand how that output was reached.
Figure 5 (Jan./Feb. 2024 ) N (Total) = 1,982
17
Now: Key findings
Our executive interviews suggest, however, that addressing trust issues is likely to “Expert” organizations recognize the importance of building trust in
become critically important as organizations transition from experimentation to Generative AI and are putting effort into it.
large-scale deployment—especially for organizations where the imperative to deploy
Despite the importance of trust for successful Generative AI deployment and scaling, 40%–
Generative AI is more tactical than strategic, and thus less time sensitive.
45% of our overall respondents said they are, to a “large” or “very large” extent, implementing
Generative AI when deployed at scale becomes far more important to the business processes to improve trust in their Generative AI initiatives through various aspects (such as
and affects a much larger pool of human users, making trust a much bigger issue. Trust data quality, output reliability and organizational empathy). However, among organizations that
related to data quality, LLM training and output reliability becomes particularly important. reported “very high” Generative AI expertise, the focus on trust is much higher across every
aspect (59%–73%). This likely reflects both their greater appreciation for the importance of trust
“If you don’t have the right dataset or data quality, it is very hard for the application to be
and their greater reliance on Generative AI as an integral and crucial part of the business.
helpful,” said a chief technology officer we interviewed. “GenAI solutions are very sensitive
to good quality and well-structured data. If the data is not correct, it is very hard to know
that the output is wrong.”
Companies implementing processes to generate trust in GenAI
In our survey, 33% of respondents cited lack of confidence in results as one of Generative
To a “large” or “very large” extent
AI’s top risks (third in the list of top 10 risks). Only 36% of the organizations surveyed were
measuring worker trust and engagement as part of adapting their talent strategies to Overall Very high expertise
Generative AI.
45% 60%
Transparency with employees
Demonstration of consideration,
40% 59%
empathy and kindness in use of GenAI
43% 73%
Quality Generative AI input data
41% 67%
Reliable Generative AI output
Figure 6 (Jan./Feb. 2024 ) N (Total) = 1,982; N (Very high) = 96
18
Now: Key findings
4 Evolving the workforce
Workforce challenges affect Generative AI scaling on both the front and back ends. On the Most organizations expect Generative AI to affect their talent strategies.
front end, organizations need valuable and scarce talent with expertise in Generative AI
Three-quarters of survey respondents (75%) expect to change their talent strategies
(and data management) to develop and refine their solutions. They also need the overall
within two years in response to Generative AI. Organizations reporting “very high”
workforce to be comfortable enough with the technology to be willing to use it for improving
Generative AI expertise expect to change their talent strategies even faster, with
efficiency and effectiveness. On the back end, organizations need to understand how the
32% already making changes. This is consistent with our broader finding that such
workforce could be affected by large-scale Generative AI deployment and then develop
organizations are scaling up their initiatives much more aggressively than are others,
talent strategies, programs and policies that make sense for the business and workers alike.
leading to greater and more immediate talent impacts.
Addressing these critical and complex workforce issues is an urgent enabler for Generative
AI adoption and scaling, even as organizations work to figure out the technology side
of the problem.
Timeline for change in talent strategies
18% 26% 31% 16% 10%
Now Within 1 year 1-2 years 2+ years Don’t
know /
no formal time
frame
Figure 7 Q: When do you expect to make changes in talent strategies because of generative AI?
(Jan./Feb. 2024 ) N (Total) = 1,982
19
Now: Key findings
The most common talent strategy responses are with “very high” expertise were much more focused on resistance |
209 | deloitte | us-state-of-gen-ai-report.pdf | Now decides next:
Insights from the leading
edge of generative AI adoption
Deloitte’s State of Generative AI in the Enterprise
Quarter one report
January 2024
Table of contents
Foreword
Introduction
Now: Key findings
1 Excitement about generative AI remains 4 Current generative AI efforts remain more
high, and transformative impacts are focused on efficiency, productivity and cost
expected in the next three years. reduction than on innovation and growth.
2 M any leaders are confident about their 5 Most organizations are still primarily relying
organization’s generative AI expertise. on off-the-shelf generative AI solutions.
3 Organizations that report very high 6 Talent, governance and risk are critical areas
expertise in generative AI tend to feel more where generative AI preparedness is lacking.
positive about it—but also more pressured
7 Leaders see significant societal impacts on
and threatened.
the horizon.
8 Leaders are looking for more regulation and
collaboration globally.
Next: Looking ahead
Authorship & Acknowledgments
About the Deloitte AI Institute
About the Deloitte Center for Integrated Research
About the Deloitte Center for Technology, Media & Telecommunications
Methodology
22
Foreword
Now decides next
The arrival of generative AI heralds disruption and From these wave one insights, we can gain a clearer
opportunity across industries. Organizations are picture of how leaders are using generative AI,
exploring how generative AI can be used to unlock challenges, and lessons learned thus far. This helps
business value, supercharge efficiency and productivity, reveal some of the essential questions leaders should
and open the door to entirely new products, services be asking now and actions they should be taking to
and business models. As business leaders contend prepare their enterprise for what comes next.
with this new technology and make decisions about the
There is still much to discover with generative AI.
future of the enterprise with generative AI, it is helpful
As it matures and is deployed at scale for a litany of
to keep one’s finger on the pulse of adoption.
applications, new questions and challenges will become
To that end, The State of Generative AI in the Enterprise: clearer. Our quarterly reports will be available to help
Now decides next, captures the sentiments of 2,835 you make sense of this fast-moving space, consider
business and technology leaders involved in piloting or practical guidance based on what we have learned,
implementing generative AI in their organizations. In this and take a forward-looking view in your business
inaugural release of the quarterly report series, leaders future with generative AI.
indicated persistent excitement for using generative
Learn more about the series and sign up for updates at
AI and many expect substantial transformative
deloitte.com/us/state-of-generative-ai.
impacts in the short term. Yet, they also acknowledged
uncertainty about generative AI’s potential implications
Deborshi Dutt, Beena Ammanath, Costi Perricos and
on workforces and society as the technology is
Brenna Sniderman
widely scaled, calling for greater investment in talent,
governance and global collaboration.
3
Introduction
Now decides next: Insights from the leading edge of generative AI adoption
Will generative AI (gen AI) be the greatest, most impactful technology innovation in Generative AI seems to be following the same pattern, only much, much faster. ChatGPT
history? Will it completely transform how humans live and work? Or will it turn out to was publicly released on November 30, 2022, largely as a technology demonstration.
be just another technology du jour that promised revolutionary change but ultimately Two months later, it had already attracted an estimated 100 million active users—
delivered only incremental improvement? Right now, we can’t be certain. making it the fastest-growing consumer application in history.1
What we do know is that many breakthrough technologies of the past have followed Since then, generative AI has continued to advance by leaps and bounds and many new
a common adoption pattern: initial awareness; excitement that led to hype; mild tools and use cases have emerged—providing a powerful glimpse at the technology’s
disappointment as hype met reality; and then explosive growth once the technology vast potential to transform how people live and work.
reached critical mass and proved its worth.
4
Introduction
Insights from the leading edge (cont.)
About The State of
Generative AI in
During this frenzied period of generative AI advancement To help make smart decisions, leaders need objective, timely
and adoption, leaders in business, technology and information about current generative AI developments— the Enterprise
the public sector are under tremendous pressure to and where things are headed. Which is why Deloitte is
To help leaders in business, technology and the
understand generative AI—and to figure out how to harness conducting this ongoing quarterly survey. Our goal is to take public sector track the rapid pace of generative
AI change and adoption, Deloitte is conducting a
its capabilities most effectively (or at least avoid being the pulse of generative AI adoption, offer a view of what’s
series of quarterly surveys. The series is based
disrupted). They also sense that now decides next; that their happening, track evolving attitudes and activities, and deliver on Deloitte’s State of AI in the Enterprise reports,
which have been released annually five years
decisions and actions today will significantly affect how practical, actionable insights that can help leaders like you
running. The wave one survey was fielded to more
generative AI unfolds in the future, for better or worse. make informed and confident decisions about AI, strategy,
than 2,800 director- to C-suite-level respondents
across six industries and 16 countries between
investment and deployment.
It’s been said that people tend to overestimate the effect of October and December 2023. Industries included:
Consumer; Energy, Resources & Industrials;
a technology in the short run and underestimate its effect in In this report, we examine our first quarterly survey findings
Financial Services; Life Sciences & Health Care;
the long run. This phenomenon has occurred many times in in detail, supported by insights from Deloitte’s AI-related Technology, Media & Telecom; and Government
& Public Services. Learn more at deloitte.com/us/
the past and could very well happen again with generative AI. work with organizations across every major industry and
state-of-generative-ai.
Note here that given generative AI’s dizzying pace of change, many geographic regions. We also offer a forward-looking
the gap between the short run and long run might be view to help you decide what generative AI actions may make
measured in days, weeks or months—not years or decades. sense for your own organization and situation.
All statistics noted in this report and its graphics are derived from Deloitte’s
first quarterly survey, conducted October – December 2023; The State of
Generative AI in the Enterprise: Now decides next, a report series. N (Total
leader survey responses) = 2,835
Generative AI is an area of artificial intelligence and refers to AI that in
response to a query can create text, images, video and other assets.
Generative AI systems can interact with humans and are often built using
large language models (LLMs). Also referred to as “gen AI.”
5
Now: Key findings
This first pulse of our generative AI quarterly surveys was
completed in December 2023, and included more than
2,800 AI-savvy business and technology leaders directly
involved in piloting or implementing gen AI at major
organizations around the world. Here’s what they had to
say about sentiment, use cases, challenges and more.
66
Generative AI elicits a range
Now: Key findings
of strong emotions
1 Excitement about generative AI remains high, and
transformative impacts are expected in the next three years.
62% Excitement
Nearly two-thirds (62%) of the business and technology leaders surveyed reported excitement as a top
Fascination 46%
sentiment with regard to generative AI; however, that excitement was tinged with uncertainty (30%) (figure 1).
The vast majority of respondents (79%) said they expect generative AI to drive substantial transformation
30% Uncertainty
within their organization and industry over the next three years—with nearly a third expecting substantial
transformation to occur now (14%) or in less than one year (17%) (figure 2). Trust 17%
The survey results suggest that many AI-fueled organizations are on the verge of scaling up their efforts
16% Surprise
and embracing generative AI in a more substantial way. This aligns with what we’re seeing in the marketplace,
where organizations around the world are racing to move from experimentation and proofs-of-concept
Anxiety 10%
to larger-scale deployments across a variety of use cases and data types—pursuing both speed and
value capture while managing potential downside risks and societal impacts. 8% Confusion
In future surveys, we will be closely monitoring progress in this area—particularly with regard to
Fear 6%
organizations’ expertise, capabilities, tangible outcomes, and responses to rapidly emerging advances in
generative AI technology. 4% Exhaustion
Anger
1%
31%
of the leaders we surveyed expect substantial transformation
Figure 1
in less than one year; 48% expect it in one to three years.
Q: Thinking about generative AI, what emotions do you feel most about the technology?
(Oct./Dec. 2023) N (Total) = 2,835
77
Now: Key findings
When is generative AI likely
to transform your organization?
1%
14%
Never
Now
20%
17%
Beyond
three years
Less than
one year
48%
In one to
three years
Figure 2
Q: When is generative AI likely to substantially transform your organization and your industry, if at all?
(Oct./Dec. 2023) N (Total) = 2,835
8
44% rate their organization’s generative AI expertise as
Now: Key findings high or very high, but is such expertise even possible
given the pace of the technology’s advancement?
2 Many leaders are confident about their organization’s
generative AI expertise.
Self-assessed expertise with
A large percentage of our survey respondents (44%) said they believe their organizations currently have generative AI runs high
high (35%) or very high (9%) levels of expertise with generative AI. This result is somewhat surprising given
how rapidly generative AI is evolving (figure 3).
1%
But within the specific context of our survey, high levels of confidence seem entirely reasonable since
No expertise
9%
we deliberately chose experienced leaders with direct involvement in AI initiatives at large organizations 10%
Very high
already piloting or implementing generative AI solutions. However, given how rapidly the field is unfolding, it Little
expertise
expertise
may be worth questioning the extent to which any leader should feel highly confident in their organization’s
expertise and preparedness. In fact, even today’s foremost AI experts who are personally developing
generative AI technologies at times seem genuinely surprised by their own creations’ capabilities.2
35%
High
Do some leaders consider their organizations to have high expertise based largely on the knowledge
expertise
and experience gained from small-scale pilots with a small number of generative AI tools? If so, leaders
45%
and organizations might actually become less confident over time as they gain experience with the larger
Some
challenges of deploying generative AI at scale. In other words, the more they know, the more they might
expertise
realize how much they don’t know. This is a trend we’ve seen time and again with other technological
advancements, and one we’ll be watching closely in our future surveys.
Figure 3
Q: How would you assess your organization’s current level of overall expertise
regarding generative AI?
(Oct./Dec. 2023) N (Total) = 2,835
9
Expertise with generative AI drives
attitudes toward adoption
Now: Key findings
3 Organizations that report very high expertise in generative
Very high Some
expertise expertise
AI tend to feel more positive about it—but also more
Trust prevails Rank trust
39% 9%
among top
over uncertainty
pressured and threatened.
emotions felt
11% 38% Rank uncertainty
among top
Relative to other respondents, leaders who rated their organization’s overall generative AI expertise as “very
emotions felt
high” tended to feel much more positive about the technology; however, they also feel more pressure to
adopt it—and see it as more of a threat to their business and operating models (figure 4).
Analysis showed this group using more modalities, deploying generative AI across more enterprise functions, Broad interest 78% 38% Say employees
show high interest
sparks
and pursuing more use cases. As you can see in the figure 4, leaders who reported very high levels of
in gen AI
transformation
expertise were also more likely to report higher levels of trust and lower levels of uncertainty. They also
31% 9% Say gen AI
tended to show broader interest in generative AI and expected faster transformation for their organizations.
is already
transformative
At the same time, these respondents’ greater understanding of generative AI appears to be shaping their
perspective on potential impacts—positive and negative. Many reported they viewed widespread adoption of
the technology as a threat to how their organizations operate and conduct business, amplifying the pressure
Widespread 33% 16% Feel widespread
and urgency they felt to adopt generative AI and scale it.
adoption is a
adoption
threat to business
generates
pressure
44% 25% Feel greater
Leaders of organizations with very high expertise are more likely to pressure to
adopt gen AI
view generative AI as a threat to their business and operating models.
Figure 4
(Oct./Dec. 2023) N (Total) = 2,835,
N (Very high) = 267; N (Some) = 1,273
10
Key benefits organizations hope
to achieve with generative AI
Now: Key findings
4 Current generative AI efforts remain more focused
Improve 56%
on efficiency, productivity and cost reduction than on
efficiency and
productivity
innovation and growth. 35% Reduce costs
Improve existing 29%
The majority of organizations surveyed are currently targeting tactical benefits such as improving
products and
services
efficiency / productivity (56%) and/or reducing costs (35%). Also, 91% said they expect generative AI to
29% Encourage
innovation
improve their organization’s productivity, and 27% expect productivity to increase significantly. A smaller
and growth
percentage of organizations reported targeting strategic benefits such as innovation and growth (29%)
Shift workers from 26%
(figure 5). lower to higher
value tasks
26% Increase speed
This is consistent with past technology adoption patterns. Initially, most organizations logically focus on
and/or ease of
developing new
incrementally improving their existing processes and capabilities—capturing value from low-hanging fruit
systems / software
Increase 25%
while building knowledge, experience and confidence with the new technology. Later, they expand or shift
revenue
their focus to improvements that are more innovative, strategic and transformational—using the new
technology to drive growth and competitive differentiation and advantage through capabilities that simply
23% Enhance
relationships
weren’t possible before.
with clients /
customers
Surveyed leaders that cited higher levels of AI expertise show earlier signs of moving up this curve. They Uncover new 19%
ideas and
are more focused on uncovering new ideas and insights (23% vs. 19% for the overall respondent pool),
insights
with less emphasis on efficiency and productivity (44% vs. 61% for the overall respondent pool) and cost
18% Detect fraud and
manage risk
reduction (26% vs. 38% for the overall respondent pool)—although those tactical benefits continue to be Figure 5
Q: What are the key benefits you hope to achieve through your generative AI efforts?
(Oct./Dec. 2023) N (Total) = 2,835
11
Now: Key findings
their bigger focus. In addition, nearly three-quarters of organizations that cited very high generative
AI expertise had already begun integrating the technology into their product development and R&D
activities, which are key drivers of innovation and growth.
As more organizations gain expertise and experience with generative AI, will they reinvest their dividends
from improving efficiency and productivity toward pursuing more strategic benefits such as innovation
and growth? Or will they use those dividends in other ways? This is another area we’ll be monitoring
closely in future pulse surveys.
Certainly, productivity and efficiency can be transformational, especially given the massive scale
generative AI has the potential to enable. However, the greatest value and strategic differentiation will
likely come from using the technology to innovate. First, by helping to generate new products, services
and capabilities that wouldn’t be possible otherwise. And, second, by enabling new business models and
ways of working across an enterprise.
In addition, organizations that cited very high generative AI expertise were already taking a much more
comprehensive approach than average, with significantly higher adoption levels across a broad range
of functional areas. In specific areas such as HR, and legal, risk and compliance, those organizations’
generative AI adoption rates were nearly three times higher than for the total respondent pool (figure 6).
91%
of all organizations expect their productivity to increase due
to generative AI.
12
Now: Key findings % of those who are using generative AI
Total Little expertise Some expertise High expertise Very high expertise
in a limited or at-scale implementation
Level of generative AI
adoption
IT / cybersecurity 22% 38% 57% 71%
46%
Marketing, sales and customer service 41% 16% 34% 50% 73%
57%
Product development / R&D 41% 14% 28% 73%
Strategy and operations 35% 10% 26% 47% 62%
37%
Supply chain / manufacturing 29% 9% 21% 61%
Finance 37% 63%
25% 5% 14%
Figure 6
Human resources 23% 6% 13% 29% 64%
Q: What is your organization’s
current adoption level of generative AI across
the following functions?
28%
(Oct./Dec. 2023) N (Total) = 2,835; Legal, risk and compliance 21% 7% 10% 60%
N (Very high) = 267; N (High) = 1,003;
N (Some) = 1,273; N (Little) = 274
1133
Generative AI: Have we seen this movie before?
The term “unprecedented” is often thrown around Generative AI’s speed factor may give organizations less help the workforce get accustomed to using generative
when talking about business and technology, to the time to ruminate or dabble with small-scale pilots— AI, and will show people how it can help make their
point of being cliché. However, in describing the pace of while reducing the margin for error—and increasing the jobs easier. Also, early wins will likely help produce cost
generative AI’s emergence and advancement—and its consequences of inaction. It also creates opportunities savings and momentum that then can be channeled into
massive potential impact on business (and humanity as a to generate extraordinary business value very quickly. higher value opportunities that are more strategic and
whole)—unprecedented could be an understatement. differentiated in nature, such as enabling new products,
Despite generative AI ’s greatly accelerated pace,
services, business models and ways of working that
Generative AI is already widely available to the public understanding typical adoption patterns based on
simply weren’t possible before generative AI.
and has a running start toward critical mass. Also, similar previous breakthrough technologies can provide
to smartphones, it’s easy for an average person to use valuable lessons that leaders can use to help them
without much training—and can help with activities they understand and fully capitalize on the technology’s rapid
already engage in every day—so the barriers to adoption advancement.
are low. What’s more, generative AI has the strong
As in the past, organizations’ initial efforts will likely
potential to assist with its own future development,
center around efficiency, productivity, cost savings and
which could trigger a cycle of exponential improvement
other incremental improvements. This is expected to
at exponential speed.
14
Now: Key findings
5 Most organizations are primarily relying on off-the-shelf
generative AI solutions.
Where off-the-shelf generative AI
In line with their current emphasis on tactical benefits from generative AI, the vast majority of respondents is used most
were currently relying on off-the-shelf solutions. These included productivity applications with integrated
generative AI (71%); enterprise platforms with integrated generative AI (61%); standard generative AI
71%
applications (68%); and publicly available large language models (LLMs) (56%), such as ChatGPT.
Productivity applications
Relatively few reported using more narrowly focused and differentiated generative AI solutions, such as
industry-specific software applications (23%), private LLMs (32%), and/or open-source LLMs (customized to
68%
Standard applications
their business) (25%).
Reliance on standard, off-the-shelf solutions is consistent with the current early phase of generative AI
61%
adoption, which is primarily focused on improving the efficiency and productivity of existing activities.
Enterprise platforms
However, as use cases for generative AI become more specialized, differentiated and strategic, the
associated development approaches and technology infrastructure will likely follow suit.
56%
Public LLMs
When will we see complex, high-value use cases that are truly differentiated and tailored to the specialized
needs of specific companies, functions and industries? How will organizations combine internal and
external resources to create customized generative AI tools that enable such strategic differentiation? In
particular, will we see off-the-shelf technology offerings be supplemented by private or hybrid public/private
development approaches and technology infrastructures capable of delivering and supporting those
differentiated solutions?
15
Now: Key findings
6 Talent, governance and risk are critical areas where
generative AI preparedness is lacking.
In this initial quarterly survey, 41% of leaders reported their organizations were only slightly or not at
all prepared to address talent concerns related to generative AI adoption, while 22% considered their
organizations highly or very highly prepared. Similarly, 41% of leaders reported their organizations were only
slightly or not at all prepared to address governance and risk concerns related to generative AI adoption,
while 25% considered their organizations highly or very highly prepared (figure 7).
Larger percentages of leaders reported high to very high levels of preparedness in technology infrastructure
(40%) and strategy (34%); however, the survey results show there is still significant room for improvement.
Generative AI barriers related to risk and governance
When it comes to risk and governance, generative AI is definitely not “just another technology.” The
fundamental challenge is how to capitalize on artificial intelligence’s power without losing control of it. After
all, the capability people seem to find most enthralling about generative AI is its ability to so convincingly
simulate human thinking and behavior. Of course, human thinking and behavior aren’t always perfect,
predictable or socially acceptable—and the same is true for the technology, itself.
16
Now: Key findings
Respondents claimed the highest levels of preparation in technology
Preparedness for generative AI
and strategy, while feeling far less prepared in risk and talent.
Technology infrastructure 4% 17% 38% 30% 10%
Strategy 5% 20% 41% 26% 8%
Not prepared
Slightly prepared
Risk & governance 13% 28% 34% 18% 7% Moderately prepared
Highly prepared
Talent 13% 28% 37% 17% 5% Very highly prepared
Figure 7
Q: Consider the following areas. For each, rate your organization’s level of preparedness with respect to broadly adopting generative AI tools / applications?
(Oct./Dec. 2023) N (Total) = 2,835
17
Managing generative AI
implementation risk
Now: Key findings
Monitoring regulatory
Specific generative AI risks and concerns include inaccurate results and information (i.e., “hallucinations”);
47%
requirements and
Establishing a governance
legal risks such as plagiarism, copyright infringement, and liability for errors; privacy and data ownership ensuring compliance
framework for the use
challenges; lack of transparency, explainability and accountability; and systemic bias. The latter of generative AI tools / 46%
applications
exemplifies another category of risk in which AI amplifies and exacerbates a problem that already exists,
such as propagating and systematizing existing social biases, facilitating and accelerating the spread of
Conducting internal
42%
audits and testing
misinformation, helping criminals commit crimes, or fanning the flames of political divisiveness.
on generative AI
tools / applications
Training practitioners
According to the business and technology leaders we surveyed during fourth quarter 2023, the biggest
37% how to recognize and
mitigate potential risks
concerns related to governance were: lack of confidence in results (36%), intellectual property issues
(35%), misuse of client or customer data (34%), ability to comply with regulations (33%), and lack of
36% Ensuring a human
explainability / transparency (31%). validates all generative
AI content
Some of the surveyed organizations were already actively managing generative AI implementation
34%
Using a formal group
risks through actions such as monitoring regulatory requirements and ensuring compliance (47%), or board to advise
on generative
establishing a governance framework for generative AI (46%), and conducting internal audits and testing
AI-related risks
32%
Keeping a formal inventory
on generative AI tools and applications (42%) (figure 8). However, such organizations are in the minority of all generative AI
implementations
and their actions barely scratch the surface of the challenge. This is especially true given that regulatory
26%
requirements typically lag behind the pace of technology innovation—although a US presidential Using outside vendors
to conduct independent
executive order and the European Union’s ambitious Artificial Intelligence Act are clear signs government
audits and testing
21%
Single executive
leaders in many parts of the world are taking the issue of AI risk very seriously.
responsible for managing
generative AI-related risks
Figure 8
Q: What is your organization currently doing to actively manage the risks around
your generative AI implementations?
(Oct./Dec. 2023) N (Total) = 2,835
18
Generative AI is impacting talent
strategies now
2%
Never
10% 17%
No formal Now
time frame
24%
16% Within 1 year
Now: Key findings 2+ years
Generative AI barriers related to talent and workforce
Generative AI has the potential to supplement human workers across a vast array of activities traditionally
thought of as uniquely human. As such, its impact on talent and workforce strategies could be immense.
How will it affect organizations and their workers in the short and long runs? Which types of skills will be
most affected, and when?
31%
The vast majority of leaders we surveyed (72%) said they expect generative AI to drive changes in their
1-2 years
talent strategies sometime within the next two years: now (17%), within 1 year (24%), or in 1-2 years (31%)
(figure 9).
Figure 9
However, less than half (47%) reported that they are sufficiently educating their employees on the
Q: When do you expect to make changes to your talent strategies because of
capabilities, benefits and value of generative AI—survey respondents also cited a lack of technical talent and
generative AI?
skills as the biggest barriers to adoption.
(Oct./Dec. 2023) N (Total) = 2,835
19
Now: Key findings
Against this backdrop, some respondents reported making a high or very high effort to: It should be noted, however, that these reported workforce-related efforts might be limited
recruit and hire technical talent to drive their generative AI initiatives (42%), educate the in scope. Deloitte’s experience suggests that most organizations have yet to substantially
workforce about generative AI (40%), and reskill workers impacted by generative AI (36%). address the talent and workforce challenges likely to arise from large-scale generative AI
Those numbers are much higher for leaders who viewed their organization’s generative AI adoption. A likely reason for this is that many leaders don’t yet know what generative AI’s
expertise as very high (74%, 74% and 73%, respectively) (figure 10). talent impacts will be, particularly with regard to which skills and roles will be needed most.
Preparing workforces for generative AI: Respondents making a high or very high effort in the following areas.
74% 74%
73%
All respondents
55% 55%
42% 50%
40%
Little expertise
36%
30%
27%
24%
Some expertise
16%
14%
10%
High expertise
Recruiting and hiring technical talent to drive Educating our broader workforce to raise Reskilling workers because of the impact
Very high expertise
our generative AI initiatives overall generative AI fluency of generative AI to their roles
Q: What level of effort is your organization taking regarding the following workforce-related areas?
Figure 10
(Oct./Dec. 2023) N (Total) = 2,835
20
“Generating confidence
in workers’ abilities to
collaborate with generative AI,
now, could elevate creativity
and job satisfaction, next.”
21
Now: Key findings
51%
expect generative AI to
7 Leaders see significant societal impacts on the horizon.
increase economic inequality.
Although the leaders we surveyed were generally excited and enthusiastic about generative AI’s potential
business benefits, they were less optimistic about its broader societal impacts. Specifically, 52% of
respondents said they expected widespread use of generative AI to centralize power in the global economy,
while 30% expected it to more evenly distribute global power. Similarly, 51% expected generative AI to
increase economic inequality, while 22% expected it to reduce inequality (figure 11).
What’s more, 49% of respondents believe the rise of generative AI tools / applications will erode the overall
level of trust in national and global institutions.
Is this pessimism or realism? Our survey results appear to reflect the broader moral and ethical debates
about artificial intelligence that are occurring in every corner of society—even in the boardrooms of the
technology companies driving AI development, where AI’s commercial value is being weighed against its
potential value to serve humanity and AI’s potential benefits are being weighed against its potential risks.
The challenges that generative AI poses in corporate governance and risk parallel those in societal
governance and risk. In both domains, the technology’s potential benefits and potential harms are high.
National and supranational organizations and governments will likely need to walk the tightrope of helping
to ensure that generative AI benefits are broadly and fairly distributed, without overly hindering innovation
or providing an unfair advantage to countries with different rules.
22
Now: Key findings
Expected societal impacts of generative AI
Distribution of economic power
5% 25% 18% 42% 10%
30% 52%
distribute centralize
Levels of economic inequality
3% 19% 27% 41% 140%%
22% 51%
decrease inequality increase inequality
Q: How will widespread use of generative AI shift the overall distribution of power in the global economy?
Figure 11
Q: How will widespread use of generative AI tools / applications impact global levels of economic inequality?
(Oct./Dec. 2023) N (Total) = 2,835
23
Support for increased regulation
and global collaboration
Now: Key findings
8 Leaders are looking for more regulation and
78%
more regulation
collaboration globally.
Agree the widespread
proliferation of generative
In a break from traditional business norms, the unique risks associated with generative AI are prompting
AI tools / applications will
many business leaders to call for increased government regulation and increased global collaboration require more regulation
of AI by governments
around AI technologies.
Among the leaders in our survey, 78% said that more governmental regulation of AI is needed, while 72%
said there is currently not enough global collaboration to ensure the responsible development of
AI-powered systems (figure 12).
These results seem to r |
210 | deloitte | us-the-new-data-duo-for-ai-powered-growth.pdf | Marketing and IT:
The new data duo for
AI-powered growth
How marketers can bridge the data
Subtitle, date or author second line
divide to leverage the full power of AI
Marketing and IT: The new data duo for AI-powered growth
WHAT’S INSIDE
Increasing consumer complexity 1
Data + AI + marketing 3
Benefits of a unified data ecosystem combined with AI 7
How to create a unified data ecosystem and enable 9
AI-powered marketing
Technologies to create a unified data ecosystem 12
Resources to guide your next steps 16
Marketing and IT: TheC naenw S VdOatDa sduurov ifvoer t AhIe- pfuotwuerere odf gmroedwitah?
INCREASING CONSUMER COMPLEXITY
Consumer behaviors are constantly evolving, and their So perhaps it’s not surprising that creating and
high expectations for speed, convenience, and tailored delivering personalized messages to customers is
experiences make it complex for brands to effectively something that marketers have worked on for many
understand and meet their needs. To make a substantial years. Over time, advancements in technology have
connection with their target audience and satisfy those made the collection, processing, and activation of data,
expectations, organizations must not only be flexible especially first-party data, more influential in supporting
and agile, but stay up to date with changing consumer this strategy. As marketers explored more advanced
values and preferences. solutions, such as machine learning (ML) and artificial
intelligence (AI), to power hyper-personalization, they
From a marketing perspective though, you could enabled more automated processes, which helped their
argue that this thinking is nothing new. For as long as organizations increase efficiency and cut costs. Although
marketers have honed their craft, they’ve understood marketers have relied on AI for some time now (maybe
that the organizations that know their customers and without even realizing it), the generative AI revolution is
most effectively provide a personalized experience creating lots of excitement, numerous questions, and
are the ones that can drive engagement, acquisition, some trepidation about what this technology means
and lifelong loyalty. In fact, research shows that a well- for marketing. For many organizations, it can still be
executed, hyper-personalized marketing strategy can very difficult to power a customer-centric and hyper-
deliver eight times the return on investment (ROI) and lift personalized marketing strategy that effectively links
sales by 10% or more (figure 1).1 back to and connects with its customers.
Figure 1. Research shows ROI and lift sales
Source: Deloitte, Connecting with meaning, accessed 2023
41
Marketing and IT: The new data duo for AI-powered growth
But why is this the case? For years, organizations have Traditionally siloed across most organizations, there
collected massive amounts of customer data to analyze has been a rise of marketing and IT leaders teaming
and inform their decision-making (figure 2). Additionally, up and using innovative data approaches to provide
the rapid advancement of AI solutions should be a new and improved experiences to their customers.
catalyst for change, greatly enhancing an organization’s Throughout this paper, we explore the benefits of
ability to create marketing content and engage with the marketing and IT duo and how this partnership
customers more efficiently and effectively. So why are leverages data and AI to improve customer experiences
marketers continuing to struggle? and unlock business outcomes.
To answer this question, we surveyed a diverse group of
marketing and IT leaders throughout the world to:
• Identify the use cases that marketers prioritize
highly but struggle to execute.
• Understand the common challenges that
marketers experience with use case execution.
• Define what marketers truly mean by
“personalization” and “bringing their customer
experiences to life.”
Figure 2. Survey results involving artifical intelligence models with predictive AI and generative AI
2
Marketing and IT: The new data duo for AI-powered growth
DATA + AI + MARKETING
Data lies at the core of every modern organization, This may be easier said than done though, as we see
and it’s being created, stored, and analyzed at an organizations continuing to grapple with the challenge
unprecedented rate. Especially for marketers, this of effectively implementing critical and highly prioritized
explosion of data presents enormous opportunities use cases. For example, over half of the marketing
for those who are prepared to take advantage of it. leader respondents to our survey indicated the following
With the rapid growth of AI, it’s become increasingly marketing capabilities as high priorities for their
critical for organizations to ensure that the necessary organizations, but more than 40% of those respondents
data to inform their AI solutions is sound and readily also said that their organizations lacked maturity in
available. Organizations can combine their data actually executing each of these capabilities.
foundation and AI capabilities with human expertise
to understand people’s needs and the external factors When exploring the major challenges that organizations
that influence them. Establishing this holistic picture run into when implementing or using AI/ML to support
can help marketers become more efficient in both their their marketing use cases, our research uncovered an
investment strategy and time to execute, ultimately impactful collection of barriers (figure 4).
enabling them to deliver the hyper-personalized
experiences that their customers expect (figure 3).
Figure 4. What are the major challenges or
Figure 3. Marketers focus to deliver hyper- barriers that your organization experiences
personalized experiences by establishing a when implementing and using AI/ML to support
holistic picture marketing and advertising use cases?
63
Marketing and IT: The new data duo for AI-powered growth
Although not an exhaustive list, there are several There are many opportunities for organizations that are
factors that can contribute to these struggles and an properly prepared to collect, process, and unify their
organization’s inability to overcome them. data, particularly as the volume of gathered data grows.
The ability to organize, access, and act on data is critical.
LIMITED DATA SHARING AND INTEROPERABILITY Organizations that fail to address data management and
BETWEEN SYSTEMS unification requirements may find future data influxes to
be more of a challenge than a chance for innovation.
While many organizations collect and store first-party
data (and may combine it with second- or third- FRAGMENTED NETWORK OF CONTENT CHANNELS
party data) to use in marketing campaigns, a lack of
connectivity and interoperability between the data can be Consumers have seemingly endless ways to research
problematic and limit the impact of marketing analytics their interests, be inspired, or make a purchase.
or activation efforts. Although not the only culprit, Conversely, marketers are dealing with a larger and
siloed or independent data systems often contribute more fragmented network of content channels to
to an organization’s lack of data unification and sharing, reach and engage people. This only amplifies the
especially as the volume of collected data grows. challenges that organizations and marketers must
navigate—especially since, as research shows, people
Based on our research, the top challenge organizations expect to be consistently treated as individuals across
face when sharing data internally mirrors a top challenge all of these channels. Organizations need to adapt to
associated with organizations using AI/ML for their consumers’ changing interests in real time and serve up
marketing use cases: ensuring sufficient integrations or personalized content no matter where they are.
interoperability across data platforms.
Figure 5. Customers expect to be consistently treated as indivuals across all content channels.
Source: Deloitte, Connecting with meaning, accessed 2023
4
Marketing and IT: The new data duo for AI-powered growth
Figure 6. how teams within an organization believe
SLOW AND COMPLEX TECH ADOPTION
first-party data is used to support an organization’s
marketing use cases.
The slow and oftentimes complex orchestration of
technologies within organizations can hinder the
development of a project before it even starts.
In recent years, marketing teams have advanced
their tech fluency. Since 2015, the fastest-rising skills
mentioned in job postings for marketing leaders are “key
performance indicators (KPIs)” and “cloud solutions,”
two areas that fit comfortably in the world of data and
technology.2 Additionally, our research revealed that
72% of marketers possess either primary decision-
maker or executive approver authority within their
organizations when it comes to choosing marketing or
data technology solutions.
This shift in skill set and responsibility has moved the
focus of IT more toward system integration support and
away from its traditional role as the primary enterprise
technology decision-maker. For a while, this worked out,
but over time and with the growth of first-party data and
AI, conflicting priorities and perspectives between IT and
Ultimately, this can lead to misalignment on the core
marketing teams surfaced.
objective(s) for an AI project, which can encourage
both teams to continue to work independently. This
On the IT side, there’s a strong will to lead enterprise
can create an environment for AI adoption projects
data and AI projects. It makes sense: This team oversees
to sputter along or fail altogether as both teams
the organization’s data governance and security policies,
struggle to agree on important project components,
and often builds and trains AI/ML models. From the
such as the business case, technical and financial
marketing team’s side, the desire to spearhead these
requirements, and evaluation criteria.
data and AI projects stems from the fact that they
are the ultimate consumers of these data assets (for
Through our research, we continue to see that today
instance, AI/ML model scores). Marketers need them
in how teams within an organization believe first-party
to understand customers and build better marketing
data is used to support an organization’s marketing
activation tactics, such as personalized content
use cases (figure 6).
generation or an optimized media buying strategy.
85
Marketing and IT: The new data duo for AI-powered growth
IMPACT OF DATA PRIVACY POLICIES These changes can impact everything from campaign
planning to post-campaign reporting, and they can affect
Ever-changing data privacy regulations require privacy the accuracy of third-party ad platforms and brand
leaders to constantly update and conform their measurement systems. A marketer’s ability to target,
organization’s data compliance and governance rules. measure, and understand its consumers is directly
Additionally, the expansive but often fragmented affected, which makes it vital that organizations align on
collection of data privacy laws propels many and establish a privacy-centric approach to marketing.
organizations to adopt internal policies that apply
regulations, such as the California Consumer Privacy This can be especially challenging for global brands
Act of 2018 (CCPA) or the General Data Protection that must untangle a web of regional, national, and
Regulation (GDPR), across any consumer that interacts international data privacy laws. Regardless of the
with the organization, regardless of that individual’s complexity, owning and building a foundation of
location. Additionally, heightened controls among consented first-party data is crucial to AI-powered
internet browsers and hardware companies are marketing. Our research indicates that high-growth
impacting traditional data collection, and with these companies are focused on using AI to support their
disruptions in gathering valuable data, marketers are marketing processes because AI is the business
facing significant disruption in how they drive business multiplier that can help organizations keep up with
impact and measure ROI. shifting consumer demands and gain necessary insights
in an efficient and privacy-centric way (figure 7).
Figure 7. High-growth companies are focused on using AI to support their marketing processes.
6
Marketing and IT: The new data duo for AI-powered growth
BENEFITS OF A UNIFIED DATA ECOSYSTEM
COMBINED WITH AI
It’s important to remember that consumers expect create actionable insights through conversation. In turn,
organizations to understand their wants and needs as this can make insights commonplace and reduce the
well as adapt to changes in them almost instantaneously. time needed for organizations to transform their vast
Providing a superior customer experience is increasingly amounts of data into insights and decisions.
important for organizations, especially in today’s
crowded marketplace. In fact, 97% of organizational For years, marketers have utilized analytics applications
leaders agree that customer experience management that have AI/ML built in as a core component to help
is an integral business strategy for creating loyal and unlock insights. The introduction of generative AI has
long-lasting customer relationships.3 However, to meet created the opportunity to take this output to the next
or even anticipate and exceed their customers’ needs, level by democratizing access to generated insights and
organizations should adopt a unified data ecosystem data-driven decision-making.
that integrates with their existing data systems, as
well as harnesses the power of AI to create authentic Imagine asking a generative AI application to “Tell me
customer experiences. This unified data ecosystem which online products underperform and how that
serves as the infrastructure that helps organizations impacts revenue.” After a few moments, the application
take advantage of AI. In today’s dynamic world, speed returns a list of the 10 products with the highest declines
and predictability can be game-changing tools for in forecasted revenue for the next quarter. Finding ways
organizations as they design and deliver the hyper- to make it faster to get to insights regardless of technical
personalized experiences that customers expect. or analysis skills will be incredibly impactful.
By building a foundation with organized and trusted INCREASE ADOPTION OF AI IN MARKETING
data, you build confidence across your organization that
everyone is operating from the same source of truth. As part of our research, we discovered that 78% of
From there, you can enable AI to make data-driven markets are planning to increase the use of AI to
decisions quicker and unlock forecasting capabilities enhance their marketing capabilities and processes
that let you better predict and adapt to the needs of the over the next 12–18 months. Conversely, only 31% of
business. In addition to positioning your organization marketers indicated that their organizations have a
to be successful with AI, you may realize the following well-defined strategy in place that balances generative AI
benefits from a unified data ecosystem: capabilities with robust data privacy measures to enable
AI-powered marketing. This signifies that while many
MAKE EVERYONE AN ANALYST organizations are aware of—and some even actively
utilize it—generative AI still remains a new territory
The unification of data, AI, and business intelligence that may not be fully understood, but marketers are
(BI) enables marketers to develop a dynamic collection intrigued by its potential.
of self-service BI dashboards that a broad group of
end users—not just analysts—can easily use and For example, content guidelines and policies can be
manipulate. The ability to access, analyze, and act on up- effective tools for marketers but sometimes difficult
to-date data can empower marketers, business leaders, to design and deploy. Content strategists can utilize
and other important users with real-time insights that generative AI to brainstorm and develop the content
drive value. pillars that incorporate their organization’s mission,
vision, and brand values. Additionally, generative AI can
This concept of “everyone is an analyst” introduces be an effective tool to enforce the brand guidelines
a new way to work by making data and marketing that ensure an organization’s voice is consistent in
analytics capabilities more accessible to all interested content and community engagements across numerous
stakeholders. Essentially, it can empower an platforms and channels.
understanding of data at a superficial level by enabling
marketers and other business users with the ability to
7
Marketing and IT: The new data duo for AI-powered growth
Figure 8. Surveyed marketers are interested in implementing
generative AI capabilities.
Marketers can also use generative AI to tailor a conversational experiences that retrieve information
message’s tone for different audiences, or create custom from a wide variety of relevant data sources, such as
imagery based on an individual’s specific characteristics localized weather, product and media catalogs, or
or behavior—it can act as a helping hand in the effort current events. This AI-powered capability can help
to create hyper-personalized marketing. In fact, 73% decrease the time-to-service and enhance the speed
of marketers surveyed indicated an interest in using and accuracy with which customer questions are
generative AI in this way (figure 8). addressed, resulting in reliable support that helps you
stand out to your customers.
Working in parallel with ad platform technology that’s
powered by AI, marketers can create sustainable EXPERIMENT AND ITERATE FASTER WITH AI
frameworks to facilitate consistency and resonance
and make ongoing decisions about their content and A structured, test-and-learn approach to experiment
its impact. For example, marketers can integrate AI with AI can supercharge your innovation for quicker and
and data into their digital content creation process to more efficient results.
create quick-turn digital assets that activate across
multiple channels. Organizations can use ad platform technology powered
by AI to help optimize their in-platform metrics to
COMBINE AI AND HUMAN EXPERTISE TO realize a high ROI. Marketers can use multivariate
OPTIMIZE RESOURCES testing strategies to identify cohorts, maximize
engagement, and iterate with increasing improvements,
AI is naturally adept at tackling questions where the all while relying on the technology to do the heavy
answers are precise and the logic is clear. To that end, lifting when it comes to creative experimentation. For
it can be especially beneficial to utilize AI to automate instance, AI can generate content based on different
tasks, especially the time-consuming or less strategic versions of metadata descriptors to create new
ones. This can help drive better campaign performance iterations to be tested.
or improve team efficiency by reallocating people to
focus on tasks less suited for AI, such as critical problem- Organizations can build and deploy AI models, often
solving or strategic decision-making on how to market a provided by a cloud-based platform, with their first-
new product or service. party data, to enable predictive analytics capabilities,
such as customer lifetime value modeling or propensity
For example, some advertising platforms provide AI- to convert, which can be activated across marketing
powered features, and marketers can utilize these to channels to help marketers optimize their strategy for
augment their targeting strategies and capture growth hyper-personalized customer engagement and serve
opportunities or drive incremental conversions across the right message to the right customer at the right
their marketing channels. time. By enabling these audiences, marketers can test
personalized messaging and experiment with more
Generative AI can also be used to improve loyalty and parts of the customer journey.
engagement on your website by automating advanced
181
Marketing and IT: The new data duo for AI-powered growth
HOW TO CREATE A UNIFIED DATA ECOSYSTEM
AND ENABLE AI-POWERED MARKETING
By now, you may be convinced of the benefits of a Figure 9. Who in your company is championing the
unified data ecosystem and AI-powered marketing. The adoption and use of AI to support marketing and
next step: realizing this dream in your organization. advertising use cases?
The transformation required for this process represents
an opportunity to start fresh with the way you work with
data, and not just optimize a broken process. It will be a
journey, but over time, changes can be made that better
align IT and marketing teams. You can move toward a
centralized vision, reduce data silos to encourage more
data collaboration, and democratize access to analytical
insights for all interested stakeholders.
FIND AN IMPLEMENTATION APPROACH THAT
WORKS FOR EVERYONE
Perhaps the best place to start this journey is with
the teams and individuals who will be responsible for
the design, implementation, use, and support of the
new data and AI capabilities. The objective here is to
bring the marketing, business, and IT teams together
to align on a data and AI implementation that fits with
the organization’s culture, values, and growth strategy.
Below are some suggestions that can help this go
more smoothly.
ACKNOWLEDGE MISALIGNMENT BETWEEN
ORGANIZATIONAL LEADERS
Our research shows that executive leaders and IT
Delivering transformative projects, such as building managers are active champions for AI adoption. We
a unified data ecosystem with AI capabilities, can found that when digital transformation is driven from
be a sizable investment that creates tension and the top down, it tends to be more successful (figure 9).
disagreements among all invested stakeholders.
Accepting this conflict is a crucial first step. In fact, research from Deloitte’s State of AI in the
Unsurprisingly, our research indicated a healthy spread Enterprise, 5th Edition reports tthat a vision or plan
of teams that are championing the adoption of AI to from an organization’s executive leadership for how
support marketing use cases. When those situations AI will be used is the most important factor in the
occur, embrace the conflict and use it to identify the root development of an AI-ready culture. Additionally, the
causes that prevent harmony among the key decision- research indicates these “high-outcome organizations,”
makers. This practice may require teams and individuals which adopt leading practices associated with the
to tweak their mindsets about data and AI, but doing so strongest AI outcomes, are significantly more likely to
can help each team, especially marketers, to develop report revenue-generating results—such as entering
new disciplines that bring a strategic, full-funnel, and new markets, expanding services to new constituents,
cross-channel view to how an organization can use data creating new products or services, or enabling new
and AI. business and service models. The rewards can be
lucrative for organizations that are aligned and execute
on a common vision for the use of AI.
192
MMaarrkkeettiinngg aanndd IITT:: TThhee nneeww ddaattaa dduuoo ffoorr AAII--ppoowweerreedd ggrroowwtthh
IDENTIFY TEAMS TO INCLUDE IN PROCESS
With an emphasis on creating
new growth, we’ve leveraged out
Although many teams will benefit from a data and AI
foundation, some teams who would benefit more, such e-commerce and customer profile
as digital advertising teams, don’t have the IT or technical data for ML-driven audience
resources to support them. Many organizations work
segmentation and driving rapid
with external partners to build their marketing use
experimentation of personalized
cases, and they use something like predictive audience
building and activation to quickly demonstrate ROI and customer experience.
prioritize more strategic use cases. The North Face is
a great example of this. The company saw a need to
Frank Tingley
better understand its customers, and it collaborated
Senior Director of Analytics
with Deloitte to develop a cloud-based solution with AI/
ML capabilities that used its e-commerce and customer The North Face
profile data to increase purchase frequency and drive
member acquisition.
Figure 10. Revenue-generating outcomes—High- vs. low-outcome organizations
(Selecting “Achieved to a high degree”)
Source: Deloitte, State of AI in the Enterprise, 5th Edition, October 2022
1130
Marketing and IT: The new data duo for AI-powered growth
When thinking about the teams to include and building a Figure 11. Types of partners organizations used to
road map for teams to benefit quickly, there are a handful develop their customer data and AI strategies.
of important, people-related questions to consider.
1. Between our IT and marketing leadership teams, do
we have a clear and aligned understanding of what a
unified data ecosystem combined with AI can solve?
2. In order for this investment to drive value, which
teams will ultimately access and utilize these
capabilities?
3. Which teams can take advantage of these data and
AI capabilities to drive experimentation and growth?
4. How can the organization ensure people readiness?
What AI skill levels are present among current
employees?
5. How do the marketing, business, and IT leaders align
on this journey? As the growth and financial security
of organizations continues to be scrutinized, how
can the infusion of data and AI help leaders better
understand, plan for, and exceed their target metrics?
6. With respect to multi-partner orchestration, what
external organizations need to be involved in the
process to maximize value?
While many marketing leaders may be more familiar with
LEVERAGE MULTIPLE PARTNERS
the process of collaborating with a media or creative
agency of record, the importance of data in marketing is
External support can be quite valuable for a
so crucial that some forward-thinking marketing leaders
transformational project with data and AI. After
are hiring a “data partner/agency of record” to help
identifying the internal teams that should be involved,
bridge the data divide with IT.
consider which partners could meaningfully contribute
to your project.
Developing and implementing an impactful strategy
around unified data and AI can be challenging. A plan
When we asked about the types of partners
that calls on the expertise of a variety of partner types is
organizations used to develop their customer data and
critical for success. Especially with the recent explosion
AI strategies, each type listed below was selected by
of generative AI, the AI landscape and ecosystem is
more than 45% of marketing and IT leaders (figure 11).
rapidly evolving with new technology and partnerships.
Organizations need reliable guidance on strategies
When this question was changed to inquire about
that can help them build and deploy applications that
the implementation of their customer data and AI
successfully balance safety, responsibility, and ROI.
strategies, creative and media agencies were the only
partner types to not be selected by at least 45% of
marketing and IT leaders.
1114
Marketing and IT: The new data duo for AI-powered growth
TECHNOLOGIES TO CREATE A
UNIFIED DATA ECOSYSTYEM
Another key piece of the puzzle is identifying the right that enable data analysts to build and operationalize
technology platform to build the data foundation ML models on structured, semi-structured, and even
that powers your AI engines. Data and AI are highly unstructured data using simple SQL—in a fraction of
interdependent, so a sound data foundation and the time it would take to build a model from scratch.
strategy heavily influences the ability to develop and These ML models can be shared with a managed AI/ML
deploy initial AI use cases and gain traction toward platform that allows for more advanced AI use cases
reaching a mature state of AI adoption. Organizations designed to help you build, deploy, and scale machine
need to build AI into their data foundation and strategy learning models faster, for any use case, including
and embed AI into the data life cycle. This can be building generative AI apps.
accomplished through a customer data platform (CDP)
or data cloud solution, which can be built internally using Specific features like this are designed and well-
cloud platform technology, engaging with a consulting or positioned to make AI/ML development faster, easier,
delivery partner, or making a purchase off the shelf from and more accessible than ever before.
a technology partner.
Ultimately, organizations can work with a data cloud
BUILD A DATA CLOUD provider to help design and drive use case testing that
can quickly prove value without needing a new, full-
With a collection of cloud-based, data, and AI-powered scale enterprise system implemented. From that point,
systems, data cloud providers can help organizations organizations can help build an appropriate business
manage every stage of the data life cycle and transform case, assess the output, and evolve scale machine
their marketing efforts through the benefit of a learning models faster, for any use case, including
connected, open, and intelligent data ecosystem. With building generative AI apps.
this collection of cloud-based services, organizations
can implement a data platform that unifies the data, Specific features like this are designed and well
business intelligence, and AI capabilities needed to positioned to make AI/ML development faster, easier,
provide transformative experiences for customers, and more accessible than ever before.
unlock timely insights across various data sources, and
enable organizations to act on data-derived decisions Ultimately, organizations can work with a data cloud
that drive impact. provider to help design and drive use case testing that
can quickly prove value without needing a new, full-
Many data cloud providers also offer preconfigured AI/ scale enterprise system implemented. From that point,
ML frameworks and toolsets so developers don’t have organizations can help build an appropriate business
to start from scratch when they begin their AI projects. case, assess the output, and evolve.
More specifically, some data warehouses offered by
data cloud providers have built-in AI/ML capabilities
1152
Marketing and IT: The new data duo for AI-powered growth
LICENSE AN ENTERPRISE CDP Additionally, some organizations are turning toward a
“dual-zone” CDP approach (as introduced in Deloitte
CDPs have become a fundamental tool for organizations Digital’s Bridge the customer data divide with a dual-
to gain real insight into their customers’ preferences zone CDP) to create a structure for IT and marketing
and intent. A CDP can provide a solution built on top of leaders to collaborate and align on the organization’s
existing technology and infrastructure, which can enable technology—from data storage and processing to
organizations to quickly integrate their existing data marketing analysis and activation. Dual-zone unbundles
sources and leverage those insights to drive efficiency. CDP capabilities from one or more sources and
reorganizes them into two distinct (but still connected)
Of course, a CDP is not a one-size-fits-all piece of zones, each with clear ownership and responsibilities.
technology, and understanding how to proceed requires
some upfront discussions among IT and business As a result, organizations can deepen their
leaders to research and understand the different CDP understanding of customers, reduce risk through
options emerging in the marketplace today and define a greater privacy compliance, elevate the experiences of
clear path forward (figure 12). customers, and drive new revenues.
Figure 12. Common CDP archetypes
Four basic CDP archetypes are emerging in the marketplace today—each with its own strengths and gaps.
1136
Marketing and IT: The new data duo for AI-powered growth
Figure 13. Solving needs across the customer divide
Enterprise IT and marketing typically have different priorities when it comes to customer
data-related capabilities. Identifying the right solution for your whole organization begins
with understanding where those needs diverge—and where they overlap.
Source: Deloitte, Bridge the customer data divide with a dual-zone CDP, November 2022
EXPERIMENT WITH AI-POWERED SOLUTIONS
Generative AI solutions are
With high-quality data in place, it’s possible to transform not just about keeping up with
key marketing capabilities, such as content marketing, content demands; they are about
in real time and craft inspiring creative content using AI
|
211 | deloitte | us-state-of-gen-ai-q3.pdf | Now decides next:
Moving from potential
to performance
Deloitte’s State of Generative AI in the Enterprise
Quarter three report
August 2024
deloitte.com/us/state-of-generative-ai
Table of contents
Foreword
Introduction
Now: Key findings
1 Building on initial success
2 Striving to scale
3 M odernizing data foundations
4 M itigating risks and preparing for regulation
5 M aintaining momentum by measuring value
Next: Looking ahead
Authorship & Acknowledgments
About the Deloitte AI Institute
About the Deloitte Center for Integrated Research
About the Deloitte Center for Technology, Media & Telecommunications
Methodology
22
Introduction
Foreword
In the rapidly evolving landscape of artificial intelligence (AI), the connection between The complex discussions around creating value and managing risk makes it clear to me
technology and value has become increasingly apparent. What is known about major that we need to keep humans at the center of all this decision-making. It is the human
technology innovations in the past holds true with Generative AI (GenAI). stakeholders who impact how applications are conceived and developed, how they are
adopted and used, and how they are managed for trust and security. In this, employee
Technology application on its own is not enough. Results and business outcomes
upskilling and change management remain indispensable elements of value-driving
matter. The real measure of success for GenAI will be how it enables enterprise
GenAI programs.
strategies and drives tangible value.
With a focus on business outcomes and human-centered change, I feel the future with
As organizations are scaling, and learning from, their GenAI pilots, I have heard the
GenAI grows brighter by the day, even as the journey ahead will continue to surprise
discourse around GenAI shift from unbridled excitement to a more nuanced and
and challenge us.
critical evaluation of its real impact on business outcomes. I am also beginning to see
organizations think more about tailored GenAI tools—evolving from large language Learn more about the series and sign up for updates at
models (LLMs) to small language models (SLMs) for more targeted needs. They are http://deloitte.com/us/state-of-generative-ai.
also exploring how the rise of AI agents can redefine interactions within their digital
–Jim Rowan, Applied AI SGO Leader
environments, offering new avenues for automation and personalization.
Amid this maturation, regulatory considerations are coming to the fore. Our past
survey results indicated a strong market appetite for smart GenAI regulation and
oversight. Businesses and governments alike are navigating a dynamic landscape
and are struggling to keep pace with the rate of technology innovation. The challenge
is to unlock the benefits of GenAI while facing regulatory uncertainty, orchestrating
governance and building trust. No small task.
33
Introduction
Moving from potential to performance
The clock is ticking for organizations to create significant cases with strong return on investment (ROI) and a clear Generative AI-powered applications? Is regulatory
and sustained value through their Generative AI path to scale will be essential. They’ll need to address uncertainty holding them back? Are they developing a
initiatives. Promising pilots have led to more investments, challenges across the board: people, process, data and comprehensive set of financial and nonfinancial measures
escalating expectations and new challenges. During this technology. Change management and organizational to form a complete picture of benefits achieved? These
pivotal phase, C-suites and boards are beginning to look transformation will need to be given as much consideration questions must be explored in-depth as organizations
for returns on investment. There is a chance that their as technology. journey from Generative AI promise to performance.
interest in Generative AI could wane if initiatives don’t
In this quarter’s survey, we focused on two critical areas to
pay off as much, or as soon, as expected.
scaling—data and governance, and risk and compliance—
Will organizations demonstrate the patience and and how organizations are measuring and communicating
perseverance needed to unlock the transformational value. Are data-related issues hindering efforts? How
potential of Generative AI? To get there, value-led use are organizations ensuring the right oversight of
44
Introduction
Moving from potential to performance (cont’d)
Building on initial success Striving to scale
• Improved efficiency and productivity and cost reduction are still the top benefits • Two of three surveyed organizations said they are increasing their investments in
sought by organizations. Those are also cited by 42% of respondents as their most Generative AI because they have seen strong early value to date.
important benefits achieved to date.
• However, many are still challenged to successfully scale that value—nearly 70% of
• However, 58% reported they realized a more diverse range of most important respondents said their organization has moved 30% or fewer of their Generative AI
benefits, such as increased innovation, improved products and services, or experiments into production.
enhanced customer relationships.
• Respondents said that embedding Generative AI deeply into critical business
functions and processes is the top way to drive the most value from their
Generative AI initiatives.
All statistics noted in this report and its graphics are derived from Deloitte’s third quarterly survey, conducted May
– June 2024; The State of Generative AI in the Enterprise: Now decides next, a report series. N (Total leader survey
responses) = 2,770. Percentages in this report and its charts may not add up to 100, due to rounding.
Generative AI is an area of artificial intelligence and refers to AI that in response to a query can create text,
images, video and other assets. Generative AI systems can interact with humans and are often built using large
language models (LLMs). Also referred to as “GenAI.”
5
Introduction
Moving from potential to performance (cont’d)
Modernizing data foundations Mitigating risks and preparing Maintaining momentum
for regulation by measuring
• Three-quarters of respondents said their organizations
have increased investment around data life cycle
• Organizations feel far less ready for the challenges • More than 40% of respondents said their companies
management to enable their Generative AI strategy.
Generative AI brings to risk management and governance— are struggling to define and measure the exact impacts
Top actions include enhancing data security (54%)
only 23% rated their organization as highly prepared. of their Generative AI initiatives.
and improving data quality (48%).
• In fact, three of the top four things holding organizations • Less than half said they are using specific KPIs to
• Data issues are limiting options—55% of organizations
back from developing and deploying Generative AI tools measure Generative AI performance, and many
reported avoiding certain Generative AI use cases
and applications are risk, regulation (such as the European standard measures of success aren’t currently
because of data-related issues. Top data-related
Union’s AI Act, in effect August 1), and governance issues. being applied.
concerns include using sensitive data in models
and managing data privacy and security. • To deal with regulatory uncertainty, about half of
organizations reported they are preparing regulatory
forecasts or assessments.
About the State of Generative AI in the Enterprise:
Wave three survey results
The wave three survey covered in this report was fielded to 2,770 director- to C-suite-level respondents across six industries and 14 countries
between May and June 2024. Industries included: Consumer; Energy, Resources & Industrials; Financial Services; Life Sciences & Health Care;
Technology, Media & Telecom; and Government & Public Services. The survey data was augmented by additional insights from 25 interviews
with C-suite executives and AI and data science leaders at large organizations across a range of industries. This quarterly report is part of an
ongoing series by the Deloitte AI InstituteTM to help leaders in business, technology and the public sector track the rapid pace of Generative AI
change and adoption. The series is based on Deloitte’s State of AI in the Enterprise reports, which have been released annually the past five
years. Learn more at deloitte.com/us/state-of-generative-ai.
66
Now: Key findings
77
Now: Key findings Top benefit achieved through
Generative AI initiatives
1
Building on initial success
Organizations say they are seeing value from their early Generative AI forays and those successes are driving more
investment. Two-thirds of the organizations we surveyed (67%) said they are increasing investments in Generative AI Improved 34%
efficiency and
because they have seen strong value to date. A head of AI strategy and governance in the banking industry has seen
productivity
this first-hand: “Before GenAI, most senior leaders only had a vague understanding of what AI was or what it can do. Now,
they have AI at their fingertips, and it has opened their eyes to the possibilities. We have applied for additional resources.”
12% Encouraged
innovation
As in our prior quarterly surveys, improved efficiency and productivity and cost reduction continue to be the most
Improved 10%
common benefits sought from Generative AI initiatives. Those benefits were cited by 42% of wave three respondents
existing
products and 9% Reduced
as their single, most important benefit achieved to date (figure 1).
services costs
However, for most wave three respondents (the other 58%), the top benefit achieved through the new technology is Enhanced 9%
relationships
something other than efficiency, productivity or cost reduction. This includes increased innovation (12%), improved 7% Increased speed
with clients /
and/or ease
products and services (10%), and enhanced customer relationships (9%). The diversity of possible sources of value from
customers
of developing
Generative AI initiatives is exciting to many leaders and shows the potential and versatility of this new technology.
new systems /
Increased 6%
software
revenue
This distribution could mean a couple of different things. Organizations may be seeking efficiency, productivity and
6%
Developed
cost reduction, but aren’t seeing it materialize yet; they may be getting unexpected value from less tangible areas; or
new products
they may be prioritizing these other types of value. There is no one-size-fits-all approach to employing Generative AI, Shifted 4% and services
workers from
and there is a wide range of benefits that could be gained. It is important for organizations to be clear about what
4% Better
lower- to higher-
detection of
kind of value they are seeking before embarking on any Generative AI initiatives—start with value first. value tasks
fraud and risk
management
67%
of organizations we surveyed said they are increasing
Figure 1 Q: What is the most important benefit your organization
investments on Generative AI given strong value seen to date. has achieved to date through your Generative AI initiatives?
(May/June 2024 ) N (Total) = 2,770
88
Now: Key findings
Our executive interviews provided examples of Generative AI use cases that are already and greater innovation and market differentiation, most projects further along in the
delivering real-world value across a wide range of industries. Although they are working scaling process are still focused on improving productivity (figure 2).
toward things like automated decision-making, accelerated research and development,
Generative AI use cases delivering real-world value by industry
Banking Transportation Telecom Insurance Consumer Technology Finance Pharmaceuticals
A customer service A system to provide Support tools An internal medical Customer Continuous Project management Internal tool that
tool that handles customer support deployed for retail claims appeal review segmentation tools improvement tools that quickly provides instant
messages, using both and handle simple and technical field tool that provides leveraged to create processes enhanced create summary enterprise information
chat and voice, and support tickets. staff, and systems increased response more precise and by directly leveraging materials for key (such as standard
provides cross-sell The system can for troubleshooting quality and a decreased customized segments customer feedback stakeholders. operating procedures)
opportunities based automatically pull and preventive time to respond. across geographies. to inform product for thousands of staff.
on the interaction. data for human maintenance, all to development
agents to use for reduce costs. road maps.
more complex tasks.
Figure 2
9
Now: Key findings Behaviors driving the most value
for Generative AI initiatives
What do organizations think will most help drive greater “CEOs and executive leadership teams are getting much
value for their Generative AI initiatives? While many more excited and interested in what’s possible and are
Deeply embedding GenAI into
22%
different factors contribute to Generative AI value looking for use cases to demonstrate the value and
functions / processes
creation, the action cited most often by the leaders benefit,” said the global head of AI, machine learning,
we surveyed is embedding the technology deeply into analytics and data at a pharmaceutical company. “There is Effectively managing risks 13%
business functions and processes (figure 3). a lot of willingness to test, experiment and scale. However,
Deploying the latest
the potential danger is that people might get disappointed
11%
“An LLM is like an engine,” said a VP at a bank’s AI center technology
and lose attention if it’s not paying off fast enough.”
of excellence. “No one just wants the engine of a car
Developing creative and
10%
or a plane; they want a car or a plane. So, there are all C-suite and board members are still intrigued, but there
differentiated applications
these things you need to do to make it part of business are some potential signs of enthusiasm beginning to
Tailoring / customizing
processes, so the business can use it.” The value from wane as the “new technology shine” wears off. Survey
10%
models with proprietary data
any Generative AI initiative won’t be fully realized if it sits respondents said that interest in Generative AI remains
apart. As with other technologies, it will only reach its “high” or “very high” among most senior executives
Hiring the best talent 9%
potential when it is embedded in everyday tasks. Many (63%) and boards (53%); however, those numbers
organizations are already employing enterprise tools have declined since the Q1 2024 survey, dropping 11
Completely measuring
8%
enhanced with this emerging technology resource to percentage points and 8 percentage points respectively. performance
try and make this happen. Time is of the essence as organizations look to scale
their early achievements. Providing enough budget 8%
Although many have seen promising results from
early projects and are increasing investment in
Providing access to as much
7%
of the workforce as possible
Generative AI, it is important that organizations
show sustained and significant value as quickly
as possible.
Figure 3 Q: Which behavior / action do you think will drive the most
value for the Generative AI initiatives in your organization?
(May/June 2024 ) N (Total) = 2,770
10
Now: Key findings
2 Striving to scale A large majority of organizations have deployed less than
a third of their GenAI experiments into production
Selecting and quickly scaling the Generative AI projects with the most potential to create
value is the goal. However, many Generative AI efforts are still at the pilot or proof-of-
Organizations
26%
concept stage, with a large majority of respondents (68%) saying their organization has 24%
GenAI experiments
19%
moved 30% or fewer of their Generative AI experiments fully into production (figure 4).
moved into production
14%
This isn’t necessarily surprising—despite rapid and impressive advances in Generative AI’s
7%
capabilities, its applications are still relatively new and organizations are figuring out what it 4%
3%
1% 1%
can (and can’t) do well. Many organizations are learning through experience that large-scale
Generative AI deployment can be a difficult and multifaceted challenge. As with a lot 0% 10% 20% 30% 40% 50% 60% 70% 80%
of digital transformation efforts, projects can fail or struggle for a variety of reasons.
Figure 4 Q: In your estimation, what percentage of your Generative AI experiments have been deployed to date into your
“Most of our applications are still in the minimum-viable-product or proof-of-concept organization (moved into production)?
phase,” said a senior specialist for AI compliance in the automotive industry. (May/June 2024 ) N (Total) = 2,770
“Scaling across an organization where
Successfully scaling may mean different things to different organizations—based
on their goals, what approach they are taking with Generative AI, and to what
you have thousands of employees
extent scaling is actually necessary. They could be expanding from one market
to multiple markets, from a small group within a function to the entire function,
has several basic requirements, and
or from a portion of a process to multiple, integrated processes. It also depends
they’re quite challenging.” on what Generative AI-powered tools and applications are being used: scaling a
code generator across an IT department is going to be different than scaling a
customized LLM for the finance function, or a new enterprise customer relationship
-Senior specialist for AI compliance in the automotive industry
management application with Generative AI features.
11
Now: Key findings
Despite these differences, some fundamentals are consistent. More broadly, organizations should invest in the foundations of Generative AI
and concurrently assess and advance their strategy, processes, people, data and
“Foremost, you need a strategy,” the senior specialist for AI compliance continued.
technology (figure 5).
“Strategy means you can’t start by purchasing separate solutions ... if you really want
to scale, first you need to base your strategy on platforms.” Many of the fundamentals may look similar to prior digital transformation
efforts, but due to the unique nature of Generative AI, things like robust
This platform-centric approach could include establishing centers of excellence, technology
governance, transparency for building trust, transforming talent, and
platforms to enable multiple use cases, and centralized teams of experts. In our Q2 report
mature data life cycle management take on increased importance.
we advocated for centralized resources that can accelerate deployment of similar use
cases and enable organizations to make the most of scarce Generative AI expertise.
Essential elements for scaling Generative AI initiatives from pilot to production
Figure 5
Strategy Process Talent Data & technology
Ambitious Modular
Integrated Transparency Provisioning
strategy & value Robust architecture
risk to build trust the right AI
management governance and common
management in secure AI infrastructure
focus platforms
Clear, Agile Acquiring Effective
Strong Transformed
high-impact operating model (external) and Modern data model
ecosystem roles, activities
use case and delivery developing foundation management
collaboration and culture
portfolio methods (internal) talent and operations
12
Now: Key findings
How do organizations feel like they are doing across these areas—are they prepared the LLMs still needs to be improved … Data readiness; data is going to be problem
to scale? We asked how highly prepared respondents thought their organizations were forever ... Deep Generative AI understanding as well. There’s not enough people who
across some of the essential scaling elements (figure 6). Technology infrastructure understand and can drive transformation.”
(45%) and data management (41%) fared the best, followed by strategy (37%), risk
To help start a conversation on how to overcome some of these barriers, in
and governance (23%), and talent (20%).
this quarter’s survey we focused on two areas critical to scaling—exploring
This indicates that there are still some fundamental challenges holding organizations how organizations are approaching data and governance, and risk
back from successfully scaling their Generative AI initiatives. A senior director and and compliance.
head of a Generative AI accelerator in the pharmaceutical industry identified a
With respect to data, more organizations’ leaders reported they are initially prepared.
number of pressing issues: “The heritage of our processes and approaches, that
For risk and governance, they know they are not. Both need attention.
is what’s really holding us back right now. Number two is that the performance of
Do organizations think they are ready?
Percentage of organizations that are highly prepared
for GenAI across the following areas 45% 41%
37%
23%
20%
Figure 6 Q: For each area, rate your organization’s level of preparedness Technology Data Strategy Risk & Talent
with respect to broadly adopting generative AI tools / applications?
infrastructure management governance
(May/June 2024 ) N (Total) = 2,770
13
Now: Key findings
3 Modernizing data foundations
75%
of organizations have increased their technology investments
around data life cycle management due to Generative AI.
Compared with the other aspects of Generative AI However, even those executives who consider themselves
readiness, survey respondents judged that their highly prepared will likely need to do more as they progress
organizations are fairly mature with respect to data life in their journeys. Some we interviewed said that as they
cycle management (as a reminder, survey respondents moved from proof of concept to scale, unforeseen data
are from more AI-savvy organizations). This could be issues were exposed—highlighting a need to be agile.
because they had a good foundation to start with or
These issues could be because of the Generative AI-
that, according to our survey, 75% of organizations have
specific demands to data architecture and management.
increased their technology investments around data life
More robust governance—quality, privacy, security,
cycle management due to Generative AI.
transparency—is needed overall, especially around using
This increased focus was evident in our executive data that doesn’t already exist inside the organization (e.g.,
interviews. “There’s a whole series of questions GenAI public domain, synthetic and licensed third-party data).
is triggering about data strategy, that in the past Documenting data sources and labeling has an increased
were far less important,” said the chief technology importance. With more people potentially leveraging
officer at a manufacturing company. “I think we’re data, data access frameworks and literacy require more
probably spending as much time on data strategy and attention. It may change approaches toward cloud or on-
management as on pure GenAI questions, because premises data services. For more advanced LLM users,
data is the foundation for GenAI work.” working with synthetic data may eventually come into play.
1144
Now: Key findings
Levels of concern around data management
Figure 7 Q: For the following, how much concern does your organization have with respect to its data management for
Generative AI implementations?
(high + very high)
(May/June 2024 ) N (Total) = 2,770
58% 58% 57% 49% 38%
Using sensitive data Managing data privacy- Managing data security- Complying with data- Using our own proprietary
in models related issues related issues related regulations data in models
One of these challenges was highlighted by a former vice president of data and intelligence That could be because of data-quality issues, intellectual property concerns, not having
for a media and entertainment company: “The biggest scaling challenge was really the the right data, or worries about using certain kinds of data (e.g., public domain, synthetic
amount of data that we had access to and the lack of proper data management maturity. or licensed third-party data). The concerns that organizations were worried about the
There was no formal data catalog. There was no formal metadata and labeling of data most in our survey included using sensitive data in models (58% had at least a high
points across the enterprise. We could go only as fast as we could label the data.” level of concern), data privacy issues (58%), and data security issues (57%) (figure 7).
Organizations were much more worried about using sensitive data (e.g., customer
Data-related issues could be hindering organizations in their quests for getting
or client data) than they were using their own proprietary data (e.g., sales,
the levels of value that they are seeking. Data-related issues have caused 55%
operational, financial).
of the organizations we surveyed to avoid certain Generative AI use cases.
15
Now: Key findings Improving data-related capabilities
Consistent with those concerns, the top actions The value from Generative AI initiatives will increasingly
organizations are taking to improve their data-related come from organizations leveraging their differentiated
capabilities are enhancing data security (54%), improving data in new ways (whether for fine-tuning LLMs, building
Enhanced 54%
data quality practices (48%), and updating data an LLM from scratch or utilizing enterprise solutions).1 data security
governance frameworks and/or developing new For Generative AI to deliver the kind of impact executives
48%
data policies (45%) (figure 8). expect, companies will likely need to increase their Improved
data quality
comfort with using their proprietary data, which may
practices
be subject to existing and emerging regulations.
Updated 45%
governance
frameworks /
Developed new
43% Collaborated
data policies
with cloud
service provider
“Data quality is key. Understanding what data is or IT integrator
Upgraded IT 37%
to improve
infrastructure
capabilities
good data. Where is that data held? How is it
34% Hired new
talent to fill
secured? How is it permissable? All those things
data-related
Integrated 27%
skill gaps
data silos
are key to making [Generative AI] scalable.”
24% Moved to a
more flexible,
-Chief operations officer & chief of strategy for a financial services firm
open data
architecture
Figure 8 Q: What specific actions has your organization taken
to improve its data-related capabilities to support its
Generative AI initiatives?
(May/June 2024 ) N (Total) = 2,770
16
Now: Key findings
4 Mitigating risks and preparing for regulation
According to our survey respondents, Likely driving these concerns are new and emerging as highly prepared. These issues will be increasingly
risks specific to the new tools and capabilities—like important as activities shift from small-scale pilots to
three of the top four barriers to successful
model bias, hallucinations, novel privacy concerns, trust large-scale deployments and Generative AI becomes
development and deployment of
and protecting new attack surfaces. This environment more deeply embedded into the fabric of organizations.
Generative AI tools and applications are:
may be why organizations feel far less ready for the Highlighting the importance, respondents selected
challenges Generative AI brings to risk management and effectively managing risks as the second-most reported
worries about 36%
governance—since only 23% rated their organization way to drive the most value for Generative AI initiatives.
regulatory compliance
30%
difficulty managing risks
29%
lack of a
governance model
Currently, these are considered even more significant
than other critical barriers such as implementation
challenges, a lack of an adoption strategy, and difficulty
identifying use cases.
17
Now: Key findings
The chief operations officer and chief of strategy in a To help build trust and ensure the responsible use for using Generative AI tools and applications (51%),
financial services company summed up the challenge: of Generative AI-powered tools and applications, monitoring regulatory requirements and ensuring
organizations are generally working to establish compliance (49%), and conducting internal audits /
“How do you democratize Generative AI across your
new guardrails, educate their workforces, conduct testing on Generative AI tools and applications (43%)
business while having all of the right controls in place?
assessments, and build oversight capabilities. (figure 9). Despite their importance for effective scaling,
We have an AI board, we have an ethics framework, we
each of these actions is only being taken by less than
have an accountability model. We want to know who’s
Specific actions surveyed organizations are currently
roughly half of the organizations we surveyed.
using it for what, and that it’s being used in the right way.”
taking include establishing a governance framework
Actions to manage risk
51%
49%
43%
37%
35%
33%
30%
23%
19%
Establishing Monitoring regulatory Conducting internal Training practitioners Ensuring a human Keeping a Using a formal Using outside Single executive
a governance requirements and audits and testing how to recognize validates all GenAI- formal inventory group or board to vendors to conduct responsible for
framework for the ensuring compliance of GenAI tools / and mitigate created content of all GenAI advise on GenAI- independent audits managing GenAI-
use of GenAI tools / applications potential risks implementations related risks and testing related risks
applications
Figure 9 Q: What is your organization currently doing to actively manage the risks around your Generative AI implementations?
(May/June 2024 ) N (Total) = 2,770
18
Now: Key findings
78%
of leaders surveyed in Q1 agreed that more
governmental regulation of AI was needed.
Implementing new processes and controls is rarely easy and will likely require active
change management to build support within the organization. “Before launching
anything, we have strict AI governance,” said the chief analytics officer at a professional
services firm. “In the past we had a bit of a siloed approach, but today, at a minimum,
everything has to go through privacy and compliance because we have a methodical
way of managing risk. This is new and challenging to some.”
On top of risk and governance issues, Q3 surveyed organizations were exceedingly
uncertain about the regulatory environment that may exist in the future (depending on
the countries they operate in). In our first quarterly report, 78% of leaders agreed that
more governmental regulation of AI was needed. However, there is a difference between
theory and practice. Organizations are struggling with regulatory uncertainty, and
worries about interpretation and enforcement may be preventing them from pursuing
certain use cases in specific geographies.
The uncertainty around AI regulation may make it feel like there could be many varied
outcomes, but our research suggests most countries are following a similar path
concerning AI policies.2 Governments are working to balance protection, innovation
and economic benefit, so future actions will likely be in line with the regulatory
traditions of each country and region.
1199
Now: Key findings Insights from our executive interviews
How some real-world organizations are dealing with compliance, risk management
Some organizations reported taking action to prepare
and governance issues
for potential regulatory changes. Top areas include
preparing regulatory forecasts or assessments (50%),
An increasing number of organizations are making risk a central factor when selecting Generative AI use
monitoring by the general counsel (48%), and working
cases and investments. However, many are walking a tightrope—trying to minimize risk without being too
with external partners (46%) (figure 10). However, some
risk averse, which could lead to missed opportunities and open the door to competitors.
organizations aren’t doing anything to prepare; 14%
said they aren’t making any specific plans.
Here are some risk-related actions revealed through our in-depth executive interviews:
How organizations are preparing
for regulatory changes Avoiding
Avoid use cases that could require additional regulatory scrutiny
specific tools
and use cases Shut off access to specific Generative AI tools for staff
For organizations that rely heavily on owned intellectual property, be extremely cautious when
Corporate 50% Limiting
exposi |
212 | ey | ey-gen-ai-for-lending-brochure.pdf | How pursuing
GenAI can transform
mortgage lending
By applying GenAI innovations across the lifecycle,
mortgage lenders can gain a strategic advantage.
IN BRIEF In this era of unprecedented technology, mortgage
• Mortgage lenders previously lenders have a transformative opportunity to drive
have embraced artificial
operational efficiencies while enhancing customer
intelligence (AI) and machine
learning (ML). Yet they experiences by leveraging generative AI (GenAI). With
have been slow to adopt
artificial intelligence (AI) and machine learning (ML)
generative AI (GenAI).
already underpinning technological advancements in
• For mortgage lenders
embarking on their GenAI this space, we would expect to see lenders flocking
journey, multiple use cases
to GenAI to gain a tactical advantage. GenAI makes
impacting key operational
components offer a path extracting insights and automating processes connected
forward.
with unstructured data easier than ever before, and
the mortgage industry is rich with data across the loan
lifecycle, including credit, marketing, servicing and
back office.
Yet according to the Fannie Mae Mortgage Lender
Sentiment Survey released in October 2023, only 7%
of mortgage lenders are currently using GenAI; 71%
are either just beginning to explore this technology
or are not considering it at all. The complexity of the
technology, evolving regulations, concerns over data
privacy and intellectual property issues have made
adoption a challenge. But those who find a way to
navigate the complexities and drive successful adoption
will have the opportunity to outperform their industry
peers in revenue, profitability and customer experiences.
01 | How pursuing GenAI can transform mortgage lending
“
Lenders can explore and invest in GenAI capabilities starting
with use cases that have already shown a significant positive
impact in other industries. Starting on a small scale allows
lenders to identify immediate gains, thereby providing a
valuable learning experience. Moreover, this measured
approach boosts the confidence to implement broader and
more ambitious GenAI applications while maintaining a
sustainable pace of progression.
ADITYA SWAMINATHAN
EY Americas Consumer Lending and Mortgage Leader
Mortgage lender adoption of AI/ML
Source: Mortgage Lender Sentiment Survey, Fannie Mae, October 2023
7%
29%
Current users
22%
Trial users
Investigating
Not used or explored
42%
02 | How pursuing GenAI can transform mortgage lending
1
CHAPTER 1
How to remediate concerns around potential GenAI risks
Mortgage lenders are taking a measured approach to exploring GenAI.
The reluctance around GenAI is rooted more in perceived On the plus side of the ledger, GenAI could counteract
obstacles and concerns than a lack of awareness of the some of the market pressures that mortgage lenders
technology’s potential. In our interactions with clients, are facing. In the EY 2023 Annual Mortgage Executive
lenders have expressed concerns that the implementation Research Report, 60% of leaders across 20 top global
of GenAI will be a complex, expensive process that banks, midsize regional banks, and nonbank and FinTech
disrupts their existing infrastructure. Mortgage lenders lenders reported the need to increase origination volume.
also worry about data security, privacy issues, and Historically high interest rates dampened new loan and
regulatory compliance. refinancing activity, which in turn has created greater
urgency to gain market share. Among the lenders
While the risks and the concerns are real, they can be
surveyed, 70% also cited reducing operating costs as
addressed through a deliberate, measured approach
a top challenge. The mortgage lifecycle – origination,
that selects the right mix of technology and implements
servicing and default – involves time-consuming tasks
governance processes. Integration often does not require
that require reviewing cumbersome unstructured content
a drastic technological overhaul, as existing systems can
including loan applications, title report, and appraisal
be enhanced with AI capabilities, making the transition
reports, as well as extensive human interaction. GenAI
more achievable and less intimidating.
has the potential to streamline these tasks, automate
routine processes, facilitate swift and accurate decision-
making, and ultimately generate substantial cost savings.
Perceived value in GenAI
Source: Mortgage Lender Sentiment Survey, Fannie Mae, October 2023
Improve operational efficiency 73%
Reduce human error 9%
Enhance consumer/borrower experience 7%
Better control risks 5%
Other 5%
03 | How pursuing GenAI can transform mortgage lending
2
CHAPTER 2
Applying GenAI across the mortgage lending value chain
Mortgage lenders can gain significant advantages in critical areas.
Opportunities to boost efficiency across key operational
components span the entire mortgage lifecycle.
These uses are extensive and scalable, evolving with
technological advancements and incorporating more
sophisticated applications over time.
Origination: Servicing: Default: Cross-functional
applications:
By analyzing existing GenAI can serve as a GenAI can analyze
customer metadata, virtual loan assistant, borrower data and GenAI can reduce the
GenAI has the potential handling a large previous interactions need for manual data
to increase origination volume of requests to determine the most entry by automating,
volume by generating simultaneously and effective communication organizing, and
personalized loan identifying complex or strategies with the categorizing underwriting
offerings and creating sensitive issues that customer. It can optimize and servicing
tailored products for require the expertise payment allocation and documentation, including
specific customer groups. of a human agent. predict the likelihood credit reports, income
The prospect pool can be Through extensive use of payments on statements, tax returns,
expanded by tapping into of chatbots, GenAI can delinquent accounts. and insurance policies.
a wider range of sourcing improve call deflection AI and ML models can Back-office operations
channels, including social and containment. also proactively identify also can be automated
media. Establishing By automating key borrowers who are at to complete regulatory
an enriched prospect tasks, including call increased risk of default. compliance checks, such
profile can allow for dispositioning and notes as loan file completeness
more effective outreach capture, productivity of reviews, disclosures and
through customized the servicing workforce certification. Customer
loan offers, a higher can be increased. complaints can be
propensity to close the Agents also benefit from identified and logged
loan, and an increased enhanced knowledge with greater accuracy,
likelihood that the tools that help them which accelerates
borrower will secure the query the complex the resolution. GenAI
mortgage. state-level regulations also can channel
and internal rules and unstructured data from
procedures. various resources into a
searchable knowledge
management hub.
04 | How pursuing GenAI can transform mortgage lending
In addition to these specific use cases, lenders can benefit
from GenAI adoption by their vendors who are moving
to embed the technology in support systems, platforms
and applications.
GenAI use cases and opportunities across the mortgage lifecycle
Origination Servicing Default Cross-functional
• Personalized loan • Early intervention • Personalized • Document
offerings and default collection automation
prevention communications
• Social media lead • Searchable,
generation • Loss mitigation and • Payment allocation synthesized
catastrophic events optimization knowledge center
• Fraud detection and
credit assessment • Virtual loan assistant • Default prediction • Regulatory
and account models compliance
management
• Customer complaints
Source: EY Consumer Lending Team
05 | How pursuing GenAI can transform mortgage lending
3
CHAPTER 3
GenAI use cases forge the path to lending modernization
Where mortgage lenders should start making initial investments in GenAI.
For lenders starting their GenAI journeys, the following • Knowledge center: When receiving a customer query,
three use cases offer an ideal entry point. agents in traditional organizations often have to
access multiple databases to find the answer. Agents
• Personalized loan offerings: Traditional lending
are further challenged by the possibility that the
institutions primarily offer standardized loan products
information in the database may be outdated. Difficult
with minimal customization, which may not cater
interfaces also can create bottlenecks by requiring
to the specific needs of all customers. The lack of a
the use of different tools to find information. Once
personalized approach that aligns the loan offerings
the answer is found, organizations run the risk of
with the customer’s financial condition also increases
agents interpreting or conveying the same information
the chance of default. Standardized loan products
differently.
come with high operational costs due to the manual
process of scrutinizing individual loan applications GenAI can help connect and combine knowledge across
and underwriting them. different databases, creating a knowledge center for
agents to access. It can also parse and summarize
GenAI can customize loans, leveraging customer data
new laws, rules, and regulations applicable to the
to design products tailored to individual needs, which
organization. Having dashboards and interfaces that
in turn enhances customer satisfaction and retention.
are easier to navigate simplifies the information-
It can increase loan origination rates by analyzing
gathering process and saves time. With all agents
existing customer metadata. Based on a customer’s
working from the same information, they can provide
personal and financial situation, GenAI can offer data-
a consistent customer experience.
based insights that would allow the lender to adjust
the loan terms, potentially decreasing the chances of
Answering customer queries for information
default. Through greater automation of the entire loan
sanctioning process, operational costs can be reduced
Future state with GenAI —
and efficiency improved.
Difference in customer loan offerings • Connecting data from disparate sources
• Accurate and relevant content
Future state with GenAI —
• Simplified dashboards and interfaces
• Consistent voice
• Customized loan products
• Accurate targeting of specific customer groups
Source: EY Consumer Lending Team
• Personalized approach
• Operational efficiency
Source: EY Consumer Lending Team
06 | How pursuing GenAI can transform mortgage lending
• Customer complaints: When a customer calls with a
complaint, the traditional organization has agents who Authors:
manually draft complaint summaries, a time-consuming
task that can lead to inaccuracies and errors. Without
clear parameters, the manual categorization of
Aditya Swaminathan
complaints can be difficult. While logging the complaint,
EY Americas Consumer Lending
the agent receiving the call is also expected to treat the and Mortgage Leader
caller with empathy, providing a human touch to the Ernst & Young LLP
customer experience. [email protected]
An application that leverages a Large Language Model
Sameer Gupta
(LLM) can help transcribe and summarize the complaint
EY North America Financial
into call notes. By using predefined categories, GenAI
Services Organization Advanced
can classify complaints, increasing organization and
Analytics Leader
making it easier to identify trends. Freed up from Ernst & Young LLP
manual tasks, agents can focus on extending empathy [email protected]
and apologies, which improves customer satisfaction.
William Coe
Handling customer complaints
Senior Manager,
Business Transformation Consulting
Future state with GenAI — Ernst & Young LLP
[email protected]
• AI-transcribed complaints increase efficiency
• AI complaint categorization saves time and allows
for easier identification of complaint trends
Contributors:
• Agent attention is on the customer, improving
customer satisfaction
Dan Thain
Source: EY Consumer Lending Team Principal, Financial Services
1 Mortgage Lender Sentiment Survey, Fannie Mae, October 2023 Business Consulting
Ernst & Young LLP
Conclusion [email protected]
With advancing technology and evolving consumer
expectations, a transformative opportunity awaits Joe Owen
forward-thinking mortgage lenders. By exploring and Senior Manager, Consulting
Ernst & Young LLP
investing in GenAI technologies, lenders stand to gain a
[email protected]
first-mover advantage and play a pivotal role in shaping
the future of the consumer lending space.
Luke Caussade
Manager, Consulting
Ernst & Young LLP
[email protected]
07 | How pursuing GenAI can transform mortgage lending
EY | Building a better working world
EY exists to build a better working world, helping
to create long-term value for clients, people and
society and build trust in the capital markets.
Enabled by data and technology, diverse EY
teams in over 150 countries provide trust through
assurance and help clients grow, transform and
operate.
Working across assurance, consulting, law,
strategy, tax and transactions, EY teams ask better
questions to find new answers for the complex
issues facing our world today.
EY refers to the global organization, and may refer to one or more,
of the member firms of Ernst & Young Global Limited, each of which is a
separate legal entity. Ernst & Young Global Limited, a UK company limited
by guarantee, does not provide services to clients. Information about
how EY collects and uses personal data and a description of the rights
individuals have under data protection legislation are available
via ey.com/privacy. EY member firms do not practice law where
prohibited by local laws. For more information about our organization,
please visit ey.com.
Ernst & Young LLP is a client-serving member firm of
Ernst & Young Global Limited operating in the US.
What makes EY distinctive in financial services
Over 84,000 EY professionals are dedicated to financial services,
serving the banking and capital markets, insurance, and wealth and asset
management sectors. We share a single focus — to build a better financial
services industry, one that is stronger, fairer and more sustainable.
© 2024 Ernst & Young LLP.
All Rights Reserved.
US SCORE no. 23260-241US_2
2402-4424343 BDFSO
ED None
This material has been prepared for general informational purposes only and
is not intended to be relied upon as accounting, tax, legal or other professional
advice. Please refer to your advisors for specific advice.
ey.com |
213 | pwc | wef-leveraging-generative-ai-for-job-augmentation-and-workforce-productivity-2024.pdf | In collaboration with
PwC
Leveraging Generative AI
for Job Augmentation and
Workforce Productivity:
Scenarios, Case Studies
and a Framework for Action
I N S I G H T R E P O R T
N O V E M B E R 2 0 2 4
Images: Unsplash.com
Contents
Foreword 3
Executive summary 4
Introduction 6
1 GenAI’s potential for promoting job augmentation and 7
workforce productivity
2 The unwritten future of GenAI in the workforce 11
3 Insigths from early adopters 16
4 Framework for action 21
Conclusion 27
Appendix: Scenario methodology 28
Acknowledgements 29
Contributors 30
Endnotes 31
References 32
Disclaimer
This document is published by the
World Economic Forum as a contribution
to a project, insight area or interaction.
The findings, interpretations and
conclusions expressed herein are a result
of a collaborative process facilitated and
endorsed by the World Economic Forum
but whose results do not necessarily
represent the views of the World Economic
Forum, nor the entirety of its Members,
Partners or other stakeholders.
© 2024 World Economic Forum. All rights
reserved. No part of this publication may
be reproduced or transmitted in any form
or by any means, including photocopying
and recording, or by any information
storage and retrieval system.
2
November 2024 Leveraging Generative AI for
Job Augmentation and Workforce Productivity:
Scenarios, Case Studies and a Framework for Action
Foreword
Till Leopold
Peter Brown
Head, Work, Wages and Job
Partner, PwC UK; Global
Creation, World Economic
Workforce Leader, PwC
Forum
Generative artificial intelligence (GenAI) is geographies, we found that successful deployment
transforming the world of work. According to the of GenAI depends as much, or more, on people
World Economic Forum’s latest Future of Jobs than the technology itself. Workers need to
survey, within the next five years, employers expect understand, trust, and adopt GenAI. This requires
GenAI advancements to reshape a substantial not only training and support but also a cultural shift
number of jobs, potentially affecting up to 40% of within the organization to embrace new ways of
total global working hours. working.
As the capabilities of this transformative technology This report covers insights from the early adopters
continue to evolve, organizations are wondering we interviewed, as well as four different scenarios
how they can use GenAI to drive job augmentation for how the deployment of GenAI in organizations
and workforce productivity, and the actions they could play out. It also offers an actionable
can take to harness its full potential. framework which organizations can use to shape
their GenAI workforce approach.
To find out, the World Economic Forum and PwC
embarked on a new piece of research focusing We would like to thank all the organizations and
on how early adopters of GenAI are leveraging it experts who generously shared their time and
across the workforce, the impact it is having and experience with us. We hope this report will be a
the lessons they have learned along the way. useful resource for organizations across the world
as they navigate the opportunities and challenges
Based on interviews with more than 20 that GenAI brings for job augmentation and
organizations across a wide range of industries and workforce productivity.
3
Executive summary
Generative artificial intelligence (GenAI) has conversely, because of concern over its fast
the potential to drive significant improvements progress and potential for job disruption—is
in workforce productivity at the level of tasks, one which misses out on the opportunities of
organizations and economies. Delivering those productivity gains and job augmentation. A world
gains depends, among other things, on the of high trust but limited improvements in GenAI
deployment of GenAI to augment jobs, i.e. contains significant risks; while one where both
to partially perform tasks in such a way that trust and quality and applicability improve in tandem
technology effectively supports or enhances human is likely to see the biggest gains in workforce
capabilities through human-machine collaboration. productivity and job augmentation.
Drawing on a review of existing research, scenario
analysis and case studies of early adopters, this Insights from early adopters
report proposes a framework for action that fosters
job augmentation. The four near future scenarios outlined provide
a useful background to insights derived from
Global context interviews with more than 20 early adopters
from a wide range of industries and regions across
What sets GenAI apart from previous developments the world. These organizations are pursuing GenAI
in artificial intelligence is its ability to widen access partly out of confidence in productivity gains. They
to the use of AI and eliminate the barrier of also believe that GenAI will improve the quality
specialized knowledge. GenAI has the potential to of work, and the experience of their employees.
contribute to economic and productivity growth A different motivation is a desire to pre-empt the
by creating efficiencies through freeing up working potential disruption of their business.
time spent on lower-value tasks to engage in higher
value-added activities. Moreover, GenAI has the The organizations quickest to adopt GenAI in their
potential to augment human workers by enhancing workforce are those that could be described as
their skills and capabilities, thereby increasing their ‘data-driven’. They emphasize the need to develop
productivity and enabling new and diverse forms of and test GenAI solutions in small groups before
value creation. rolling them out to the rest of the organization,
allowing for issues to be identified and addressed
However, GenAI’s potential to enhance productivity before wider implementation. They also put
may vary across countries, industries, and significant emphasis on risk management, including
organizations. To effectively deploy GenAI in the designing processes that have ‘humans in the
workforce, organizations must also address a loop’, forming internal committees or councils that
range of factors including trust, skills, culture and establish internal rules, standards, and frameworks
the demonstration of business value from GenAI and assess use cases and consider sustainability
investments. implications of using GenAI at scale.
Scenario analysis To identify the potential for workforce productivity
gains and job augmentation, early adopters
With such a fast-moving technology, it is hard to combine both bottom-up and top-down
predict how even the relatively near-term future will approaches, with strong support from leadership
play out. To help think through the possibilities, it and reliance on the innovative capabilities of their
is useful to think in terms of scenarios based on workforce. It is in day-to-day practice where most
two key uncertainties that will shape the near future use cases are identified and developed. According
of GenAI-enabled job augmentation, productivity to this perspective, the most promising use
and innovation. The first core uncertainty relates cases are those embraced and championed by
to the level of trust in GenAI, which refers to the employees themselves.
confidence that employees and organizations have
in GenAI-driven tools and their outputs as well Framework for action
as employee trust in their employers, technology
providers, and governments. The second core Combining insights from the scenarios and
uncertainty relates to whether the applicability and lessons learned from early adopters, the report
quality of GenAI will continue to improve in the proposes an actionable framework for promoting
short-term or remain the same. job augmentation and workforce productivity
growth with GenAI. Focusing on factors within an
Any combination of these two dimensions is organization’s control, it is designed to be useful
possible, leading to very different outcomes. A both to organizations just starting out on their
world where trust is low—either because GenAI GenAI workforce deployment journey as well as
does not progress significantly from today; or, those seeking to scale existing efforts.
4
The framework highlights a number of key elements Engage elements focus on facilitating that GenAI
around two core themes: Enable and Engage. The workforce applications are effectively adopted and
Enable elements focus on establishing foundations integrated into workflows to generate the desired
and guiding principles and include: GenAI vision benefits. These elements include: Culture and
and strategy; Data and technology infrastructure; change management; Skills development and
and Regulatory compliance and governance. The redeployment; and Use case management.
5
Introduction
This report aims to unearth the experience of early Section 4 builds on the previous sections to offer
adopters of generative artificial intelligence (GenAI) an actionable framework that organizations may
deployment in the workforce to derive lessons adapt for their own use to augment jobs and
learned and provide an actionable framework enhance productivity through GenAI adoption.
for promoting job augmentation and enhancing The framework aims to enable organizations to
workforce productivity. It examines the key harness the potential of GenAI while adhering to
elements that organizations must have in place ethical standards, emerging legal requirements,
to facilitate these outcomes. Research interviews and considering the development and well-being of
conducted for this report are global in scope, employees using the technology.
encompassing a wide range of geographies,
industries and organizations including commercial, The findings suggest that, with the right enabling
public-sector and social entities. conditions, GenAI has the potential to augment
jobs and enhance productivity. However, this
The emergence of GenAI in the workplace has requires organizations to go through a phase of
created significant interest, from the boardroom understanding the technology’s value for their
to the breakroom. Section 1 examines these specific needs, identifying appropriate use cases,
expectations, hopes and concerns, and outlines and thoroughly testing the solutions. Moreover,
current barriers for individuals and organizations to safeguarding that workers understand, trust and
effectively leverage the technology to achieve better adopt GenAI is essential before use cases can be
people and business outcomes. scaled; thus, in addition to training and support, a
cultural shift within the organization is also critical
At present, the future of GenAI in the workforce to embrace new ways of working. Based on
remains uncertain and undefined. Despite rapid the insights from the interviews, the successful
developments, the technology is still in its infancy, deployment of GenAI depends as much or more on
making it impossible to extrapolate the extent to people than on the technology itself.
which productivity gains and job augmentation
may be achieved in the near future. Acknowledging This report has been developed as part of The
the unpredictable nature of the future, Section Jobs Initiative, coordinated by the World Economic
2 considers four scenarios to enable various Forum, which aims to build the jobs of tomorrow
stakeholders to think through the multiple ways in and ensure good jobs for all in the context of
which GenAI in the workforce could evolve. ongoing labour market disruptions. One key
focus area for the initiative is promoting strategies
Section 3 presents findings from interviews with for leveraging GenAI for job augmentation and
more than 20 early adopting organizations that workforce productivity growth. It is one of a series of
have generously shared their experiences, lessons current World Economic Forum reports that explore
learned and expectations regarding the emerging the transformative role of artificial intelligence across
impact of GenAI on productivity gains and job industries and a variety of key themes.
augmentation to provide valuable insights into the
practical implications and potential returns of GenAI
workforce deployment.
6
GenAI’s potential
1
for promoting job
augmentation and
workforce productivity
This section provides an overview of the debate several other prominent LLMs shortly thereafter,
about generative artificial intelligence (GenAI) in public interest in GenAI has surged, raising
the workforce and its potential for promoting job expectations about its potential to transform the
augmentation and productivity growth. It also global labour market. According to the World
highlights current expectations and assessments Economic Forum’s latest Future of Jobs survey,
surrounding GenAI as well as barriers to its more within the next five years, employers expect a
widespread workforce adoption, two years after the substantial number of jobs to be reshaped due to
public launch of one of the most prominent large GenAI advancements1, potentially affecting up to
language models (LLMs). 40% of total global working hours.2
By leveraging natural language processing This first section of the report will review the current
technology, GenAI enables users to interact with state of the debate on GenAI’s potential, with a
it as though they were conversing with a human, particular focus on job augmentation, workforce
reducing barriers to usage and the need for productivity growth and barriers to the technology’s
specialized technical knowledge. Since the public more widespread workforce adoption.
launch of ChatGPT 3.5 in November 2022, and
1.1 GenAI and job augmentation
Like other recent advances in automation and AI More frequently, GenAI may partially automate
technologies, the rise of GenAI has led to concerns some tasks of a job role but simultaneously improve
about possible job displacement. This apprehension human workers’ ability to perform other tasks. In
is partly rooted in the technical potential of the line with recent research, this paper refers to this
technology itself and partly in skepticism about process as job augmentation (see Box 1 and Fig.
employers’ and governments’ ability to support 1).6
individuals through AI-induced job disruptions.3 One
recent survey indicated that 47% of employees who As GenAI technologies and labour markets
had used GenAI expressed concerns that it may continue to evolve, it is likely that some job roles
affect the nature of their work in a negative way.4 may become more fully automated while others
may be further augmented in the future. In similar
Research examining the potential impact of GenAI ways to earlier industrial transformations, both
on jobs commonly operates on the premise that job automation and job augmentation may be
job roles and occupations are composed of expected to lead to additional job creation – both
various tasks, some of which may be susceptible directly, creating wholly new jobs in various fields,7
to varying degrees of automation by GenAI. For and indirectly through macroeconomic spillover
instance, tasks that are repetitive or routine are effects from increased productivity and additional
more exposed to automation than those requiring economic value creation. The focus of this report
significant human interaction. While a wide range of is on the immediate term and in putting into place
tasks may be fully automated by GenAI, research enabling conditions for job augmentation now and
to date has found very few examples of jobs that in the next years.
could be displaced in this way in their entirety.5
7
BOX 1 Automation and augmentation
This report distinguishes between the following collaboration. Job augmentation may go beyond
definitions: technical productivity increase to also enhance job
quality and worker well-being.2
Job automation refers to the use of GenAI to fully
1. Raisch, S. and S. Krakowski, “Artificial
perform tasks that were previously performed by
Intelligence and Management: The Automation–
humans in a given occupation.1
Augmentation Paradox”, Academy of
Management Review, vol. 46, no. 1, 2021.
Job augmentation refers to the use of GenAI
to partially perform tasks in such a way that 2. World Economic Forum, Augmented
technology effectively supports or enhances Workforce: Empowering People, Transforming
Manufacturing, 2022.
human capabilities through human-machine
FIGURE 01 GenAI: Example of a more exposed and less exposed job
Software Developers (more exposed) Human Resource Managers (less exposed)
28.7% 43.2% 28% 16.1% 22.2% 61.7%
Higher potential for automation: Higher potential for automation:
— Analyse data to improve operations — Determine resource needs of projects or operations
— Analyse performance of systems or equipment — Manage budgets or finances
Higher potential for augmentation: Higher potential for augmentation:
— Prepare informational or instructional materials — Explain regulations, policies or procedures
— Evaluate the characteristics, usefulness or performance of — Train others or operational or work procedures
products or technologies
Lower potential for automation or augmentation: Lower potential for automation or augmentation:
— Coordinate with others to resolve problems — Interview people to obtain information
— Communicate with others about business strategies — Coordinate group, community or public activities
Automation Augmentation Lower potential Non-language tasks
Source
World Economic Forum, Jobs of Tomorrow: Large Language Models and Jobs, 2023.
1.2 GenAI and workforce productivity growth
GenAI’s potential impact on productivity is one of knowledge. GenAI has the potential to contribute
its most anticipated benefits, particularly because to economic and productivity growth by creating
of the slowdown in productivity growth in many efficiencies through freeing up working time
economies.8 Although current forecasts vary spent on lower-value tasks to engage in higher
widely, it has been suggested that GenAI’s impact value-added activities. For instance, automating
on productivity could add trillions to the global help desk queries may allow customer service
economy over the next decade.9 workers to focus on more complex issues that
increase customer satisfaction. One recent study,
What sets GenAI apart from previous developments surveying more than 100,000 workers from 11
in AI is its ability to widen access to the use GenAI-exposed occupations, found that workers
of AI and eliminate the barrier of specialized estimated ChatGPT could reduce working times
8
by 50% for one-third of their job tasks.10 As may disproportionally boost productivity for workers
respondents interviewed as part of the research with less experience or skill, thereby reducing entry
for this report highlighted, to realize these potential barriers to the digital economy.16
productivity gains it is important to capture time
saved as value at an organizational level. At an industry level, exposure to GenAI-driven
task automation and augmentation varies widely
Moreover, GenAI has the potential to augment across sectors, with not all industries being equally
human workers by enhancing their skills and impacted or standing to benefit from GenAI. As
capabilities, thereby increasing their productivity described above, previous research has identified
and enabling new and diverse forms of value which tasks are most exposed to LLMs, highlighting
creation.11 For instance, GenAI may augment their higher or lower potential for automation or
human capabilities in creative tasks, though it augmentation. For example, one recent study
does not currently surpass human creativity on its found that software developers from three large
own.12 Research also suggests that GenAI may technology firms increased the number of tasks
help narrow productivity gaps between lower- and completed by over 26% using GenAI.17 When these
higher-skilled workers.13 exposure levels are aggregated at the industry level,
it becomes evident that the impact of GenAI may
GenAI’s potential to enhance productivity may vary vary significantly across industries. For instance,
across countries, industries and organizations. At the technology and financial sectors could face
a country level, more developed economies may substantial task automation, while the healthcare
face higher disruption risk due to prevalence of and education sectors may benefit more from task
knowledge work, but they are also better equipped augmentation.18
to adopt GenAI more quickly and at scale.14 Many
of these countries also face a decrease in labour Importantly, as discussed in Section 3 of this
supply, which may boost the demand for new report, productivity growth is not the only driver
technologies such as GenAI to seek efficiency for organizations to deploy GenAI. Many also
improvements.15 Emerging economies may similarly expect improved quality of work and better work
benefit from productivity growth by addressing experiences for their employees, increasing
infrastructure constraints and shortages in basic employee engagement and talent retention.
digital skills. Early research indicates that GenAI
1.3 Current barriers to scaling GenAI adoption in
the workforce
Historically, slow and inconsistent adoption of To build trust and facilitate the ethical use of GenAI,
AI technologies has restricted their impact and there is strong demand for transparency and
effectiveness.19 As of mid-2024, only 12% of responsible deployment. Increasingly, organizations
workers report that they use GenAI at work on are implementing responsible GenAI principles
a daily basis.20 Current barriers to GenAI uptake to build trust in decision-making processes by
encompass concerns related to trust, skills improving explainability and mitigating risks. At both
acquisition, change in culture and unclear business national and supranational levels, some territories
value. are tightening regulations on AI to promote trust
and ethical use by setting clear boundaries and
enforcing accountability. For example, the European
Trust AI Act includes the “human-in-the-loop” principle,
emphasizing human accountability in decision
making. This principle may help to increase trust in
Trust is a crucial factor that must be considered GenAI by establishing accountability and respecting
when embracing new technologies. GenAI models human values.24
are sometimes referred to as “black box” systems
due to the complexity of their algorithms, raising At an industry level, concerns have been raised
concerns about the outcomes they generate about a comparatively small number of industry
and transparency.21 In line with this, CEOs see players holding significant influence over the
cybersecurity, spread of misinformation, legal or development of GenAI as well as its regulatory
reputational damage, and increased levels of bias environment.25 Government regulation is partly
as primary concerns related to the adoption of aimed at creating a level playing field where all
GenAI.22 In addition to bias and discrimination, parties follow the same regulations and criteria.
workers are specifically worried about the lack Nevertheless, inconsistent AI laws worldwide
of oversight, transparency, explainability and could also have the reverse effect, disadvantaging
accountability.23 organizations that are operating in the most strictly
regulated jurisdictions.
9
Skills 60% over the next three years, reaching 7.6%
of IT budgets by 2027.30 Others exhibit greater
reluctance, questioning the assuredness of the
At a workforce level, two out of five employers returns these investments may bring. Companies
report that a lack of adequate AI-related skills is often cite costs as a significant barrier to GenAI
an obstacle to the integration of GenAI at work.26 adoption, with a number of them unsure of the
Increased demand for GenAI skills outside of tech technology’s potential benefits.31 The uncertainty is
roles is evident when examining the share of job amplified by the limited evidence available on the
postings in non-tech roles that now request these impact of GenAI on firm performance.32 Preliminary
skills.27 Yet, there is a prevailing concern among outcomes show promise, but the conclusive
78% of senior executives that their companies benefits remain unclear. For instance, the edge
may fail to train their employees rapidly enough to from adopting GenAI could diminish with increasing
keep pace with technological advancements in the competitive pressures in the market.33
coming years.28 This concern is reinforced as 37%
of more than 56,000 workers from 50 countries Preliminary findings indicate that the most significant
and regions surveyed by PwC have not used promise of GenAI in revolutionizing business models
GenAI applications for work in the past year, and an could reside at the intersection of specialized
additional 25% have only done so once or twice. expertise and innovative problem-solving, propelling
Even in the highly exposed financial sector almost rapid advancements in proposition innovation that
one-quarter of workers reported not to have used encompass new offerings, customers, markets,
GenAI for work. In the telecommunication sector, channels and customer relationships. Interviews
which showed the highest overall use of GenAI, conducted for this report with various early
19% of workers have never used GenAI in their adopters of GenAI show that organizations remain
work.29 cautious about the use of GenAI in externally facing
products and services. While ultimately something
they may strive for, so far, most organizations are
Culture experimenting and scaling up GenAI within the
comparatively safer walls of their own internal
organization.
The culture of an organization is a crucial factor in
the adoption of new technologies such as GenAI. To shift the focus from smaller-scale incremental
Organizations interviewed for this report stress the improvements to new business models, the
importance of change management: successful advancement of GenAI, in combination with
introduction of GenAI depends on experiments and other emerging technologies, will be one of the
finding use cases. This requires a stimulating and most important determinants. For instance, the
supportive culture. For example, it is important to emergence of multimodal LLMs enables the
cultivate mindsets such as a future-positive attitude, concurrent processing and generation of various
growth mindset and agility, which are crucial for data forms (including text, imagery and audio),
employees to embrace GenAI in the workplace. integrating these elements to create a thorough
Young (technology) companies and data-driven understanding. It can bring new strategic business
organizations tend to adopt new technology easier benefits such as improved decision-making,
and faster because they are less hampered by enhanced user experience and operational
existing, established ways of working and embody efficiency.34 With more clarity on the speed and
a data culture and digital literacy. magnitude of such developments, the potential of
GenAI to lead to business model reinvention will
become more apparent.
Business value
Certain enterprises are readily committing
substantial resources to GenAI technologies,
with investments in GenAI projected to grow by
10
The unwritten future of
2
GenAI in the workforce
There remains a high degree of uncertainty about productivity and innovation. These scenarios are
the future trajectory of GenAI in the workforce not forecasts or idealized visions but illustrate what
and the extent to which its potential for job realistically could happen.
augmentation and productivity growth may be
realized. This section presents four different near This section presents four scenarios for the near-
future scenarios for how the deployment of GenAI term future of GenAI based on future studies
in organizations could play out. Organizations, methods (Figure 2; see the Appendix for a more
leaders and workers alike will need to consider detailed description of the report’s scenario
these alternative futures when forming their hopes, methodology). Scenarios were developed through
expectations and strategies regarding GenAI workshops and trend analysis, focusing on two
adoption. key uncertainties that will shape the near future of
GenAI-induced job augmentation, productivity and
“Amara’s law” states that observers tend to innovation: 1) trust in GenAI, and 2) improvements
overestimate the short run impacts of new in applicability and quality of the technology. These
technology and underestimate the longer- scenarios are applicable to organizations that are
term ones. The long-term impact of GenAI on exposed to GenAI, with leadership aiming to deploy
productivity, augmentation and innovation remains GenAI, regardless of current workforce adoption or
uncertain.35 While some current task- and firm-level external influences. The analysis excludes situations
use cases for tools like ChatGPT have shown a where governmental entities may force or restrict
40% decrease in task time and an 18% increase GenAI use.
in quality,36 experience has shown that it can take
a long time for technology to become sufficiently
widespread to affect productivity at an economy- GenAI-induced job augmentation
wide scale.37 Applying GenAI in tasks, processes
and productivity growth: Two
and structures requires experimentation and finding
core uncertainties
and applying use cases – and this simply takes
time.38 This is also confirmed by the case study
interviews in Section 3. The first core uncertainty relates to the level of trust
in GenAI, which refers to the confidence that
Moreover, many tasks that humans currently employees and organizations have in GenAI-driven
perform, for example in the areas of transportation tools and their outputs. It also refers to the trust
and manufacturing, are multifaceted and require of employees in the organization, the technology
real-world interaction, which GenAI is not currently provider and the government to prevent issues such
able to improve upon.39 The question is whether as privacy breaches, exploitation and information
organizations will reach a point where massive leaks. As outlined in Section 1, trust is crucial
scaling-up may take place, leading to productivity for GenAI adoption and is influenced by different
gains and job augmentation on a macroeconomic factors.
level in the longer term.
The second core uncertainty relates to whether the
applicability and quality of GenAI will continue to
Scenario thinking: navigating an improve in the short term or remain the same. High
applicability means GenAI tools are practical and
uncertain future
useful across various use cases and industries.
High quality means the outputs are accurate,
Studies of the future recognize its unpredictability reliable and have a low percentage of errors. When
and aim to anticipate and prepare for the impact combined, these qualities make GenAI valuable
of potential developments, in this case: the and dependable. Improved applicability and quality
impact GenAI could have on job augmentation, would lead to new use cases, user models and
productivity and innovation in the near future. The functionalities, allowing GenAI to (further) augment
scenarios presented in this section are tools to and automate tasks, enhance jobs, create new
navigate uncertainty and inform strategic decisions. industries and serve as a foundation for future
They explore uncertainties and present possible technologies.
outcomes of GenAI-induced job augmentation,
11
High
Scenario 1 Scenario 4
High Hopes Shifting Gears
High trust High trust
Current applicability/quality Expanding applicability/quality
Scenario 2 Scenario 3
Broken Promises Lost Opportunities
Low trust Low trust
Current applicability/quality Expanding applicability/quality
Low
Remains at current level Improves
Applicability and quality of GenAI
2.1 Scenario 1: High Hopes
not able to effectively interpret or validate the results
High trust, current applicability & it produces. This leads to inaccurate decision-
making or reliance on flawed insights. So, high trust
quality
does not result in increasing productivity; on the
contrary, it leads to work having to be redone (for
In this scenario, enthusiasm for GenAI workforce example, one recent study showed that participants
adoption is high. Leadership hopes GenAI will who used an LLM to solve a particular business
contribute to the solving of labour shortages and problem exhibited a 23% lower correctness of the
anticipates it will improve the quality of work. There response compared to those who completed the
is a fear of missing out on opportunities as well. task without GenAI, due to ineffective use of the
Organizations, fearful of disruption, are afraid of t |
214 | pwc | 28-ceo-survey-regional-report.pdf | Capturing opportunities today,
reinventing for tomorrow
28th Annual CEO Survey:
Middle East findings
Snapshot of the Middle East findings
Resilient GCC economies spark Reinvention, urgent - accelerated Industries converge and compete
01 02 03
optimism amid uncertainty by AI, climate and regulation: over new domains of growth:
90% 60 72%
%
of GCC CEOs are confident about of regional CEOs think their businesses will of regional CEOs expect to do a
growth in their company revenue not be viable within 10 years or less deal outside of their industry or
over the next 12 months without adaptation, with the majority citing sector in the next three years
regulatory change as a major factor
77% 70% 53%
of KSA CEOs and 80% of UAE CEOs of GCC CEOs believe GenAI will increase of regional companies have targeted a new
confident in economic growth in their profitability within 12 months customer base within the last five years
territory
43%
61% 79%
of regional CEOs are already competing in
of regional CEOs expect to increase of regional CEOs have initiated climate
new sectors or industries
headcount within 12 months, although friendly investments in the last five years
34% of GCC CEOs cited skills shortages
as a major concern
40%
of GCC CEOs cited that cyber risk is the
top threat for the next 12 months,
followed by geopolitical conflict
Foreword
Our annual survey of CEOs globally and across the Middle East reflects the collective voice of business leaders -
offering valuable insights into the opportunities they see, the challenges they face and the path forward. This year
we captured more responses than ever before, with almost 300 CEOs sharing their views. In our region, hearing
these voices has never been more important, as we grapple with the profound megatrends of climate change and,
technological disruption, increasing regulation and an evolving geopolitical landscape.
What we’ve heard is clear: CEOs in our region are among the most confident globally about economic growth in
their territories and their own revenue growth in the coming year, with many planning to expand their workforces.
Businesses here are already investing in new technologies and strategies, particularly around AI and sustainability;
and with the region amplifying its voice in the global climate conversation, there is a clear commitment to driving
sustainable growth.
However, for these leaders, the urgency to reinvent is clear. A striking 60% of regional CEOs now believe their
businesses will not survive ‘within the next 10 years or less’ without significant adaptation - a notable increase
from last year, when less than half expressed similar concerns.
CEOs in the Middle East recognise that traditional models of business are increasingly unsustainable in the face
of transformative catalytic shifts. Alongside the climate crisis and AI-driven disruption, they also recognise there is
a battle to capture value in new domains as industry lines blur and companies face fierce new competitors,
reshaping market dynamics. These are forcing CEOs to rethink how they can innovate to secure that ever-critical
advantage. The imperative is evident: CEOs must balance the opportunities of today while also reinventing their
businesses for tomorrow.
I want to sincerely thank the chief executives who provided their valuable time in participating in this survey. The
rich insights provided have enabled us to chart a clear picture of the opportunities and challenges shaping the
future of business in our region.
Hani Ashkar
Middle East Senior Partner
PwC Middle East
01
Resilient GCC
economies spark
optimism amid
uncertainty
Resilient GCC economies spark optimism amid uncertainty
CEOs in the Middle East remain among the most confident globally about their company’s revenue growth and the Unsurprisingly, in the wider Middle East,
region's economic outlook for the year ahead, despite the geopolitical unrest unfolding in 2024. confidence is lower than in GCC countries as the
economic impact of regional conflicts have
According to the regional findings of our 28th PwC Annual Global CEO Survey, based on responses from chief extended to neighbouring countries, such as
executives representing 11 countries across the Middle East, this sentiment is the strongest amongst CEOs in the Jordan, Egypt and Lebanon. Jordan recorded a
Gulf Cooperation Council (GCC) countries, with 90% optimistic about short-term revenue growth over the next 12 6.6% year-on-year decrease in tourist arrivals
months. 71% of CEOs in these countries also indicate optimism in their own territory’s economic growth - ahead of through August due to its proximity to conflict
their peers in the wider region and globally - with 80% of CEOs in the UAE and 77% in Saudi Arabia forecasting zones in the region, while there was a 62% drop in
economic growth in the next 12 months, followed by CEOs in Oman (69%) and Qatar (63%). Expectations of Suez Canal revenue in Egypt as a result of the
growth from CEOs globally when talking about their own territories sat at a more modest 57%. reduction in Red Sea traffic in the first half of this
year.1
Despite its proximity to regional conflicts and
ongoing challenges from inflation and the currency
Q. How do you believe
Global Middle East GCC crisis, Egypt, however, has experienced a
economic growth (i.e. gross
remarkable economic turnaround this year, with a
domestic product) will change, US$35bn investment from the UAE providing a
57% 64% 71%
if at all, over the next 12 major boost. This has supported the
implementation of critical reforms, including the
months in your territory?
liberalisation of the currency regime, which has
played a key role in reducing inflation and boosting
economic growth.2 As a result, 63% of CEOs in
Improve Egypt remain optimistic about economic growth in
their market. In contrast, only 45% of CEOs in
Stay the same Jordan share this optimism, significantly trailing
their regional and global peers.
Decline 21%
16%
Looking at confidence through an industry lens,
14%
the consumer markets sector is the most confident
22%
19% about revenue growth in the short term (within the
14%
next 12 months), while the technology, media, and
telecommunications sector leads in confidence
2025 2025 2025
over the more medium term.
Note: Percentages may not total 100% due to rounding.
Q. How confident are you about your company’s prospects for revenue growth?*
Richard Boxshall
Consumer
91%
Chief Economist and
Markets
Leader of the PwC
91%
Global Economics
Network
Health
90%
93%
Variations in CEO sentiment across
the region reflect the differing
Technology, Media 90% growth prospects between the
Telecommunications
GCC and other territories. While
98% the outlook for the GCC remains
stronger, supported by substantial
investments in the non-oil sector
Financial 90% driving economic diversification
Services and resilience, geopolitical
90% uncertainties and their impact on
inflation and supply chains remain
a concern for CEOs.
Transport
87%
and Logistics
93%
Energy, Utilities
79%
& Resources
88%
Over the next 12 months Over the next three years *Net confidence for Middle East companies
Note: Percentages may not total 100% due to rounding.
Non-oil investment fuels growth
Given the Middle East’s investment in the non-oil sectors, its green economy initiatives and technological
accelerations, and an emphasis on localisation, it’s not surprising that non-oil investment is fuelling growth. Non-oil
GDP increases across the GCC averaged 3.7%, easily surpassing the overall economic growth rate of 1.8%.3 This
has helped offset the impact of OPEC+ oil production cuts. The Middle East also continues to emerge as a thriving
hub for dealmaking, where we see the region’s sovereign investment funds playing a pivotal role in driving private
equity deal volumes, particularly in emerging sectors as highlighted in our 2024 TransAct Middle East mid-year
update.4
In fact, according to our regional findings, the UAE is the seventh most likely country where global CEOs are
planning to invest, outside of their home territory - while within the region, Saudi Arabia, the UAE and Egypt are the
top three countries that regional CEOs are planning to invest into outside of their existing territories.
Anticipating revenue increases, CEOs in the region are looking to scale their operations to capture growth
opportunities at even higher rates than last year. Our survey data indicates that they are more likely to increase
headcount than their global peers: 61% of regional CEOs expect to add headcount within the next 12 months,
compared to just 42% globally, and up from 55% in 2024. In the GCC, 64% of CEOs plan to increase headcount,
despite a third (34%) citing skills shortages as a major concern. This reiterates the need for organisations to
prioritise workplace upskilling to adapt to technological advancements and tackle challenges such as supply chain
disruptions, driven by geopolitical tensions and the climate crisis.
From an employee perspective, regional workforce responses to our 2024 Hopes and Fears survey reflect the same
sentiment - 63% said technological change, especially the rise of AI, GenAI and robotics, would impact their jobs in
the next three years, compared to only 46% globally. Additionally, more than half (54%) of respondents stated that
climate change would impact their jobs, compared to 37% globally.5
Global Middle East
17%
39%
42%
Looking at headcount trends by sector, 70% of CEOs in the healthcare industry plan to increase employee numbers next year, alongside over 60% in consumer markets,
transportation and logistics and technology, media and telecommunications. Additionally, more than half of CEOs in the energy, utilities and resources and financial services
sectors anticipate workforce growth, reflecting a strong commitment to expansion across key industries.
Sector-specific concerns include inflation for consumer
markets (47%) and health industries (40%), while cyber
risks and geopolitical tensions are prominent in
transportation and logistics (53%) and technology,
media and telecommunications (50%). In the energy,
utilities and resources sector, cyber risks (47%) and
macroeconomic volatility (39%) dominate concerns,
while financial services leaders are worried about
technological disruptions (42%) and economic
instability (19%).
5202
Q. To what extent will your company increase or decrease headcount in the next 12 months?
GCC
Decrease 11% 12%
Little to no
28% 24%
change
Increase 61% 64%
Note: Percentages may not total 100% due to rounding.
Confidence in the face of ongoing threats: Geopolitical conflicts, cyber risks and skills shortages
Alongside this picture of varying levels of confidence, Middle East CEOs identified a range of issues set to be top of
mind as they approach decisions in the year ahead. Geopolitical conflict (41%), cyber risks (36%) and inflation
(30%) were cited as their top threats, while a lower availability of skilled workers emerged as a main concern,
particularly for business leaders in the GCC (34%). Notably, inflation, which was the top concern last year at 38%,
has now been overtaken by geopolitical conflict as the top risk.
The perceived magnitude of these threats varies somewhat across the region. Geopolitical conflict remains a
significant concern, particularly for CEOs in Jordan, where 55% say that their organisation will be ‘highly’ or
‘extremely’ exposed to this threat in the coming year, compared to 41% of the regional average.
In the UAE, cyber risks are identified as the biggest threat by CEOs, with 38% of business leaders anticipating high
exposure to such risks in 2025. Meanwhile skills shortages are a primary concern for 41% of CEOs in Oman and
rank among the top three threats for CEOs in the UAE and Egypt.
“The key challenge will be the lack of talent across the board, whether it is in IT, digital, relationship management or
compliance. Our only competitive advantage lies with our people, and we must ensure to recruit and retain the best
talent.” CEO, Financial Services
02
Reinvention,
urgent -
accelerated by AI,
climate and
regulation
Despite a robust short-term confidence in business
growth among regional business leaders, the impetus Q. If your company continues running on its current path, for how long do you think your business
to reinvent is stronger than ever this year - and more will be economically viable?
immediate. CEOs in the Middle East anticipate greater
pressure to evolve in the next decade or even less,
driven mostly by continuing emerging technologies, Global
42%
climate change, anticipated increased regulation and an
intensifying competition over new domains of growth as
55%
industry lines blur. They understand that the next
decade will bring profound change, and they must be
ready to adapt. This sentiment resonates strongly
Middle East
through our survey findings. 60%
Last year, almost half of Middle East CEOs expressed 37%
concerns about their organisation’s economic viability
over the next 10 years if they failed to evolve. This year,
our survey reveals an increased urgency to reinvent - GCC
64%
64% of GCC CEOs and 60% of overall Middle East
CEOs believe they will need to adapt their businesses
3%
in 10 years or less to remain viable. This heightened
concern not only surpasses last year’s levels but also
far exceeds the current global average of 41%.
10 years or less More than 10 years
Note: Percentages may not total 100% due to rounding.
Recognising the significant impact of transformative shifts on the region, CEOs in the Middle East are driven by an
urgent need to rethink strategies, looking to embrace innovation and build resilience for sustainable growth. The
message is clear: evolve or face irrelevance.
A closer look at key industries in the region reveals that the imperative for reinvention is widely felt. Our survey data
indicates that over 70% of CEOs in healthcare industries and the energy, utilities and resources sectors believe their
organisations will not be viable within the next decade without adapting. Similarly, more than 60% of leaders in
transport and logistics and technology, media and telecommunications share this sentiment.
Additionally, over half of CEOs in consumer markets and nearly half in the financial services sector recognise the
same critical need for transformation, reiterating that the need to evolve is no longer a choice – it's a necessity.
So how have business leaders in the region been driving change?
For CEOs in the Middle East, this has been an
opportunity to radically transform their Q. To what extent has your company taken the following actions in the last five years?
fundamental approach to creating, delivering
and capturing value. According to our
findings, more than half of our regional CEOs
have driven change by innovating products 54%
53% 53%
52%
and services over the past five years, while
50%
53% have focused on targeting a new
customer base to expand their market reach.
Examples of such innovation include 43%
UAE-based Detectiome's Revonco,6 an 40%
39%
38%
AI-powered multi-cancer early detection test
that is redefining healthcare, and Tabby, 34% 35%
MENA's first fintech unicorn,7 which is 32%
revolutionising financial services in the region.
26%
25%
24%
Additionally, 43% have collaborated with
other organisations, 39% have targeted new
routes to market and 34% have implemented
new pricing models - higher than the global
averages. This underscores the fact that
regional CEOs have more proactively
embraced innovation, diversification and
strategic partnerships to ensure their
organisations are better equipped for future Targeted a new Developed innovative Collaborated with Targeted new Implemented new
success. customer base products or services other organisations routes to market pricing models
53% Global Middle East GCC
Note: Percentages may not total 100% due to rounding.
Reinvention is equally critical from an investor’s perspective, as they seek to understand how the companies they invest in
have targeted a new
are navigating crises, strengthening resilience and ensuring long-term value creation. According to PwC’s 2024 Global
Investor Survey8 investors are closely examining the reinvention imperative - especially the adoption of emerging
customer base
technologies - to assess whether businesses they are investing in are positioned to capitalise on innovative opportunities.
in the last five years Four in five investors who invest in companies in the Middle East indicated technological change as the most fundamental
driver compelling companies to rethink how they create, deliver and capture value.
CEOs focus on regulation, strategy and innovation for future viability
For business leaders in the region, anticipated Additionally, almost a third (29%) of CEOs highlighted rising product and service costs as a key external driver for
regulatory changes were identified as the most critical evaluating economic viability, slightly below the global average of 32%. Embracing disruptive technologies was
external factor influencing economic viability over the cited by 27% of respondents as another main external driver of economic viability within the next decade, followed
next decade or less. Over a third of Middle Eastern closely by strong incumbent competition at 26%. On the internal front, 22% of regional CEOs pointed to a lack of
CEOs cited these expected changes, compared to 42% skills as a major factor impacting their company’s potential viability within the next 10 years or less.
globally. Over the next decade, GCC countries are
expected to implement significant regulatory changes in
the areas of AI, technology and climate, among others.
Q. What are the top five external factors influencing a Middle East company's
This will be key to shaping future enterprise and will
economic viability within the next 10 years?
offer transformative opportunities for businesses to
drive innovation, enhance competitiveness and achieve
sustainable growth. Robust frameworks for AI Changes in the
38%
governance, data protection and cybersecurity will regulatory environment
foster a secure environment for technological
advancements, while climate-focused policies will
Increasing
enable businesses to leverage sustainability 29%
products/services costs
opportunities. The region also anticipates economic
diversification initiatives, labour market reforms and
Disruptive
trade regulations, including digital trade platforms and 27%
technology
free trade agreements that will further reduce barriers,
attract foreign investments, and open new markets.
Strong incumbent
26%
“Compliance with advanced regulations remains competition
essential. Achieving a balance between innovation and
regulatory requirements will require a proactive Decreasing demand
23%
approach to ensure that our initiatives comply with both for products/services
local and international standards.” - CEO, Technology,
Media and Telecommunications
Among those Middle East CEOs looking beyond a 10-year horizon, more than half identified a growing demand for
products and services as a critical external factor, followed by regulatory changes (52%) and disruptive technologies
at (34%).
Internally, more than half (59%) emphasised making bold strategic choices as the most crucial factor for long-term
viability, slightly ahead of their global peers at 55%. These strategic decisions will enable businesses to address
future disruptions and seize emerging opportunities - enabling them to build resilience in a dynamic regional
landscape.
Other key internal factors impacting potential long-term economic viability included organisational efficiency (39%)
and having the right skills for a competitive environment (35%), underscoring the need for adaptability and
readiness to thrive.
Megatrends redefine industries: AI and climate creating new domains of growth
Middle East countries are adopting Artificial Intelligence (AI) at an unprecedented pace, fuelled by ambitions to
diversify economies and build future-ready industries. Business leaders see AI as a transformative catalyst for
innovation, with GenAI tools optimising processes and accelerating outcomes. And trust in having AI embedded into
key processes is particularly high, with half of GCC CEOs trusting it to a ‘large’ or ‘very large’ extent, compared to
Ali Hosseini only one third of their global peers.
Chief AI and
Technology Officer This growing confidence is backed by investments by regional governments and private enterprises in AI research,
PwC Middle East development and innovation hubs, while fostering responsible AI adoption. Saudi Arabia has made substantial
progress in the Global AI Index, climbing 17 positions to rank 14th globally, while national strategies have facilitated
a deep trust in AI - including the Saudi Vision 2030 and the UAE’s National AI Strategy 2031.
Our regional findings also indicate that in the GCC, a notable 88% of CEOs have adopted GenAI in the last 12
months, exceeding global averages and reflecting greater confidence in the technology’s potential.
The Middle East has benefited from
a high rate of AI adoption and at a
greater pace than competitors Q. Did your company adopt generative AI to any degree in the last 12 months?
globally, which has led to increases
in time efficiencies, profitability,
revenue and a tech-savvy
workforce. To maintain their edge,
88%
companies need to accelerate
86%
AI-led innovation and integration, 83%
with a particular focus on GenAI to
unlock new value opportunities and
be future ready
We adopted generative AI to any degree in the last 12 months
Global Middle East GCC
Note: Percentages may not total 100% due to rounding.
In fact, over the next three years, AI, including GenAI, is set to become a core component of technology platforms,
business processes, and the development of new products and services in the region. For example, Falcon 3,
developed by the Technology Innovation Institute (TII),9 delivers high-quality results with low compute requirements,
while Jais, a collaboration between G42’s Inception and Mohammad Bin Zayed University of Artificial Intelligence
(MBZUAI),10 preserves Arabic heritage and democratises AI access. In the GCC, 93% of CEOs predict AI will be
systematically integrated into tech platforms, compared to 78% globally. Additionally, 90% expect AI to enhance
business processes and workflows (vs. 76% globally), 85% to embed it in workforce and skills (vs. 68% globally),
and 81% anticipate its use in new product and service development (vs. 63% globally).
This reiterates the agility and proactivity of regional business leaders in adopting AI to drive digital transformation,
maintain competitiveness and foster growth. This sentiment is only set to grow stronger, with the region expected to
prioritise investments in AI infrastructure, forge global partnerships with leading tech giants, and establish robust
data security frameworks to drive sustainable AI growth in 2025.11
GenAI adoption is also rapidly accelerating across industries in the Middle East, with adoption rates exceeding 85%
in sectors such as consumer markets, transport and logistics, health industries, energy, utilities and resources,
technology, media and telecommunications and financial services within the past 12 months. Trust in embedding AI
and GenAI was particularly strong amongst CEOs from the consumer markets, transport and logistics and
technology, media and telecommunications industries.
As CEOs in the region embrace GenAI at scale, a striking 70% of business leaders in the GCC have indicated that it
will increase profitability in the next 12 months, up from last year - and higher than the global average of just 49%.
Q.To what extent will generative AI increase or decrease the profitability of your company in the
next 12 months? (Net increase)
49%
Global
Middle East 67%
GCC 70%
This confidence is reinforced by the tangible benefits observed over the past year, with GCC CEOs reporting that
GenAI has driven greater efficiencies, increased revenue and profitability, and facilitated job creation.
The most notable findings on GenAI this year were as follows:
01 02 03
68% of GCC CEOs More than half of GCC CEOs 36% of GCC business leaders
acknowledged improved reported revenue growth (vs. highlighted job creation through
efficiencies in their own time at 32% globally) while 53% saw GenAI, more than double the
work (vs. 53% globally) and an increase in profitability (vs. global average of 17%.
63% reported efficiencies in 34% globally).
employees’ time (vs. 56%
globally). In fact, 72% of CEOs
in technology, media and
telecommunications, 69% in
healthcare industries and 65%
in financial services have
expressed strong confidence in
GenAI’s potential to enhance
employee efficiency.
Q.To what extent did generative AI increase or decrease the following in your company in the last 12 months? (Net increase)
68% 68% 65%
63%
56% 52% 53% 51% 52%
53%
36%
32%
34%
32%
17%
Efficiencies in my own Efficiencies in my Profitability Revenue Headcount
time at work employees' time at work
Global Middle East GCC
Note: Percentages may not total 100% due to rounding.
Investor confidence also echoes these findings, with regional data from the PwC Global Investor Survey 2024
revealing that investors remain optimistic about the promise of GenAI - 74% of respondents to this survey believe
that GenAI will increase productivity in the companies they invest in or cover in the Middle East, compared to 66%
of global respondents believing the same about the territories they are investing in. And 67% of respondents believe
that GenAI will increase profitability in the companies they invest in or cover in the region, compared to 62% of
global respondents. This optimism aligns with the broader trend of key regional economies positioning themselves
as global frontrunners in AI adoption and innovation.
“In the near future, disruptive technologies will drive the economy. The challenge is that we don't know
what technologies will come and what disruption they will make. We must be ready to adapt to these new
67%
technologies and think out of the box.” - CEO, Consumer Markets
of respondents believe that
GenAI will increase
profitability in the companies
they invest in or cover in the
region, compared to 62% of
global respondents.
Climate change, the other critical megatrend - a powerful catalyst for reinvention
Climate change, the other critical megatrend, has also been a powerful catalyst for reinvention. Its combination of
opportunities and challenges have prompted CEOs to rethink strategies, adopt sustainable practices and position
their organisations for long-term resilience and growth. This is evident in our survey findings which indicate that
Yahya Anouti nearly 80% of CEOs in the GCC have initiated climate-friendly investments in the past five years, signaling a positive
Partner, Strategy& PwC regional momentum towards sustainability. Notably, we see particularly strong commitments from CEOs in transport
Middle East Sustainability and logistics (90%), consumer markets (84%) and financial services (84%) sectors, reflecting a growing focus on
Platform Leader sustainable practices in some of the region’s fastest growing sectors. The Saudi Investment Bank’s debut Tier 1
Sustainable Sukuk Issuance of US$ 750 million,12 for example, has been a milestone in the bank’s commitment to
sustainable finance and reinforces its position as a leader in responsible banking practices in the Kingdom of Saudi
Arabia.
Our survey demonstrates a clear In our latest Sustainability in the Middle East report13 published in 2024, four in five executives indicated that their
picture for Middle East CEOs: companies now have a formal sustainability strategy in place – with more than half saying that this strategy is fully
sustainability can drive economic embedded across their organisations.
opportunity and deliver measurable
Our regional CEO Survey findings have revealed that for more than half of business leaders in the GCC and nearly
benefits. With almost 80% initiating
half in the Middle East - climate investments are yielding returns that are higher than global averages, despite the
climate-friendly investments in the last
high upfront costs. This indicates that there is now a growing acknowledgment among regional CEOs that
five years, the region is proving that
sustainability can align with profitability and presents an opportunity to explore the factors enabling these higher
bold action on climate can align with
returns, such as government incentives, technological adoption or strategic investments in renewable energy.
profitability. The challenge now is for
leaders to accelerate innovation, push However, the balance between managing costs and maximising revenue continues to pose a substantial challenge
boundaries, and turn sustainability into for business leaders. When it comes to CEO buy-in on sustainable investments, only 14% of CEOs in the region
a cornerstone of their competitive say they have accepted returns below the minimum acceptable rate for other investments in the past 12 months,
advantage. compared to 25% globally - the case for the second consecutive year.
And while almost 80% of regional CEOs have made climate friendly investments in the last five years (particularly in
the transport and logistics, consumer markets and financial services sectors at 90%, 84% and 84% respectively),
they are less likely than their global peers to accept significantly lower rates of return on climate-friendly
investments. In a world increasingly shaped by environmental challenges, business leaders must recognise the need
to further integrate sustainability into core strategies - including investment decisions - to align profitability
with purpose.
Q. To what extent have climate-friendly investments initiated by your company in the last five years
caused increases?
54%
49%
48%
44%
36%
33%
Revenue from products/services sales Costs
Global Middle East GCC
Note: Percentages may not total 100% due to rounding.
Findings from our survey have also indicated that among Middle East CEOs who have not made any climate-friendly investments in the last 12 months, several key barriers
hinder their ability to decarbonise. Regulatory complexity emerges as a key challenge due to the lack of mandatory regional sustainability regulations, impacting companies
engaged in cross-border trade with jurisdictions like the European Union and the United States, where such regulations are enforced.14 Similarly, the perception of lower returns
on climate-friendly investments remains a greater challenge regionally, with the percentage of CEOs concerned about this nearly doubling the global average. The lack of
available financing is also a challenge, with regional CEOs reporting this issue at more than twice the rate of global leaders.
I aim to strengthen our focus on sustainability by deeply integrating climate-friendly investments into our business model. This aligns with global trends and
stakeholder expectations, ultimately contributing to a positive environmental impact and long-term profitability.” CEO, Financial Services
Addressing these obstacles requires a reinvention of traditional business models, enabling business leaders to advocate for regulations that can support climate action agendas,
embrace innovative financing strategies and relook at climate-friendly investments as opportunities for long-term value creation.
Q. To what extent, if at all, are the following factors inhibiting your company’s ability to decarbonise its business model?
[NET: To a large extent & to a very large extent]
Global
Middle East
36% 36%
35%
GCC 34% 33% 34%
31%
31%
24%
23%
20%
20%
18%
14%
6%
Lower returns for Regulatory complexity Lack of demand from Lack of available finance Lack of buy-in from
climate-friendly (e.g. policy changes, external stakeholders my management
investments inconsistent local (e.g. customers, team or the board
requirements) investors)
Note: Percentages may not total 100% due to rounding.
5202
03
Industries
converge and
compete over new
domains of growth
As CEOs in the Middle East evaluate the impact of the transformative forces of AI and climate change on their
existing industries and businesses, they are focusing on unlocking new value streams. This is driving sector
convergence, breaking down traditional boundaries and fostering collaboration. For example, AI-powered solutions
are linking healthcare with technology to create precision medicine and advanced diagnostics as in the case of the
Ahmad Abu Hantash partnership between G42 Healthcare and Mubadala Health in the UAE.15 Meanwhile, the climate crisis is driving
Partner, Technology energy companies such as ADNOC |
215 | pwc | ai_jobs_barometer_2024.pdf | PwC’s 2024
AI Jobs Barometer
How will AI affect jobs, skills, wages,
and productivity?
pwc.com/aijobsbarometer
PwC’s 2024 AI Jobs Barometer goes beyond
predictions about AI’s impact to find evidence by
analysing over half a billion job ads. The Barometer
reveals how AI is transforming the world of work,
making people and businesses more productive
while changing what it takes for workers to succeed.
Headline Findings
4.8x 25%
Sectors with highest AI penetration are Jobs that require AI specialist skills carry up
seeing almost fivefold (4.8x) greater labour to a 25% wage premium in some markets.
productivity growth. Rising labour productivity
can generate economic growth, higher
wages, and enhanced living standards.
3.5x 25%
Growth in jobs that require AI specialist skills Skills sought by employers are changing at a
has outpaced all jobs since 2016 (well before 25% higher rate in occupations most able to
ChatGPT brought fresh attention to AI), with use AI. To stay relevant, workers in these jobs
numbers of AI specialist jobs growing 3.5 will need to build or demonstrate new skills.
times faster than all jobs.
PwC’s 2024 AI Jobs Barometer 2
Half a billion job ads reveal
AI’s impact
AI is the Industrial Revolution of knowledge work, transforming how all workers can apply
information, create content, and deliver results at speed and scale. How is this affecting
jobs? With the AI Jobs Barometer, PwC set out to find empirical evidence to help sort fact
from fiction.
PwC analysed over half a billion job ads from 15 countries to find evidence of AI’s impact at
worldwide scale through jobs and productivity data.
PwC tracked the growth of jobs that demand specialist AI skills (such as machine learning
or neural networks) across countries and sectors as an indication of AI penetration.1 We
find that AI penetration is accelerating, especially in professional services, information &
communication, and financial services. Workers with specialist AI skills command significant
wage premiums, suggesting that their abilities to deploy AI are valuable to companies.
1 AI’s true penetration into the economy may be even greater than reflected in this analysis. By focusing on job ads, this analysis captures AI’s impact on
job changers, but does not capture AI usage or upskilling for existing employees.
PwC’s 2024 AI Jobs Barometer 3
The AI Jobs Barometer uses half a billion job ads from 15 countries to
examine AI’s impact on jobs, skills, wages, and productivity
But AI’s impact is not limited to only those workers who have specialist AI skills. Many,
if not most, workers who use AI tools in their work do not have or need these specialist
skills. For example, a limited number of workers with specialist AI skills may design
an AI system or tool for a company that is then used by hundreds or thousands of the
company’s customer service agents, analysts, or lawyers - none of whom have specialist
AI skills. In fact, one thing that makes a well-known form of AI - generative AI - such a
powerful technology is that typically it can be operated using simple everyday language
with no technical skills required.
To capture AI’s impact on all jobs, PwC analysed all jobs (and sectors) by their level of ‘AI
exposure.’ A higher level of AI exposure means that AI can more readily be used for some
tasks. Examples of occupations with higher AI exposure are financial analysts, customer
service agents, software coders, and administration managers. The analysis revealed that
sectors with higher AI exposure are experiencing much higher labour productivity growth.
At the same time, the skills demanded by employers in AI-exposed occupations are
changing fast. Read on to learn more.
PwC’s 2024 AI Jobs Barometer 4
Key Terms
‘AI specialist skills’: Specialist, technical AI skills like deep learning or
cognitive automation. See Appendix One for AI skills list.
‘AI specialist jobs’: Jobs that require specialist, technical AI skills.
‘All jobs’: All jobs in all occupations.
‘AI-exposed’: Describes all jobs or sectors in which AI can readily be used
for some tasks (based on definition of AI Occupational Exposure
developed by Felten et al.)
PwC’s 2024 AI Jobs Barometer 5
AI penetration is
accelerating
Attention to AI’s impact on the jobs market exploded in November 2022 with the launch of
ChatGPT 3.5. However, the data shows that AI had quietly exerted a growing impact on the
jobs market years before. Growth in AI specialist jobs has outpaced growth in all jobs since
2016, well before ChatGPT brought fresh focus to AI.
Today, there are seven times as many postings for specialist AI jobs as there were in 2012.
In contrast, postings for all jobs have grown more slowly, doubling since 2012. Put another
way, openings for jobs that require specialist AI skills have grown 3.5 times faster than
openings for all jobs since 2012.
Growth in Al jobs has outpaced all jobs since at least 2016
Number of Job Postings, relative to 2012
12
10
8
6
4
2
0
2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023
AI Jobs All Jobs
Source: PwC analysis of Lightcast data. The analysis represents six of the fifteen countries: US, UK, Singapore, Australia, Canada, and New Zealand.
Nine countries have been excluded due to data prior to 2018 being unavailable: France, Germany, Belgium, Denmark, Spain, Italy, Netherlands, Norway,
and Sweden.
The 2022 peak in job postings above represents exceptionally high demand for workers
which gradually eased in 2023 as job market conditions returned toward normal.
PwC’s 2024 AI Jobs Barometer 6
Knowledge work sectors have higher
AI penetration
Knowledge work sectors in particular are seeing growing demand for jobs that require
specialist AI skills. The share of job ads requiring these skills is higher in professional
services, information & communication, and financial services - precisely those sectors
predicted to be most exposed to AI.2 Financial services has a 2.8x higher share of jobs
requiring AI skills vs other sectors, professional services is 3x higher, and information &
communication is 5x higher.
Share of job postings by sector requiring Al skills
5%
4%
3%
2%
1%
0%
2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023
Information and Communication Professional Services Financial Services Other sectors
Sources: PwC analysis of Lightcast data, UK Government Impact of AI on Jobs 2023. “Other sectors” includes Agriculture, Mining, Power, Water, Retail
trade, Transportation, Accommodation, Real Estate, Administrative activities, Arts and Entertainment, Household activities, Construction, Manufacturing,
Education, Health and Social Activities and Extraterritorial Activities sectors. Chart includes all 15 countries in this study.
2 AI Occupational Exposure (AIOE), constructed by Felten et al (2021), scores and measures the degree to which occupations rely on abilities in
which AI has made the most progress in recent years, meaning AI can more readily be used for some tasks in those occupations.
PwC’s 2024 AI Jobs Barometer 7
AI specialist jobs command up to a 25%
wage premium on average
AI’s value to companies is made clear by what is happening with the wages of workers
with AI specialist skills - the very people who are making the AI revolution possible.
25%
Up to 25% wage premium for
workers with specialist AI skills
PwC’s 2024 AI Jobs Barometer 8
As we have seen, growth in jobs demanding AI specialist skills has outpaced growth in all
jobs since 2016. What’s more, these jobs carry up to a 25% wage premium on average,
underlining the value of these skills to companies.
Below are average AI wage premiums for five countries for which there is sufficient
data to perform the analysis. To show how this wage premium can affect individual
occupations, wage premiums for selected occupations are given.
Wage premium for job vacancies which require Al skills by country
Country Al Wage Premium
Occupation USA UK Canada Australia Singapore
Database Designers
+53% +58% +8% +14% +35%
and Administrators
Lawyers +49% +27% - - -5%
Sales and Marketing
+43% +14% +3% +7% +3%
Managers
Financial Analysts +33% +32% - - +11%
Applications Programmers +32% +24% - +7% +34%
Systems Analysts +30% +34% +15% +7% +28%
Accountants +18% +5% +17% - +4%
Average wage premium
+25% +14% +11% +6% +7%
across all jobs
Sources: PwC analysis of Lightcast data, ISCO-08 Occupation Codes (4-digit level). 2023 data. These findings do not demonstrate a causal relationship.
These estimates are calculated by comparing the average salaries of AI job postings to those of non-AI postings for the same occupations. Two filters
are applied to ensure (1) the count of AI job postings and (2) the ratio of AI jobs:non-AI jobs being compared is above a certain threshold. The analysis
provided represents five of the 15 countries: UK, USA, Singapore, Canada and Australia. The remainder of the countries have been omitted from this
analysis as the data was less extensive: New Zealand, Italy, France, Germany, Spain, Belgium, Netherlands, Denmark, Norway and Sweden.
For example, job ads for US sales managers that require AI specialist skills offer wages that
are on average 43% higher than job ads for sales managers that do not require AI skills.
Canada’s accountants can enjoy a 17% wage premium if they have AI specialist skills, and
UK employers are willing to pay a 27% premium for lawyers equipped with AI skills.
PwC’s 2024 AI Jobs Barometer 9
AI appears to be driving a
productivity revolution
So far this report has discussed jobs which require specialist AI skills like deep learning
or natural language processing. But many, if not most, workers who use AI tools in their
work do not have these skills. To understand how AI is affecting all jobs, PwC examined
jobs and sectors by their levels of ‘AI exposure’ which means the degree to which AI can
readily be used for some tasks. PwC’s analysis revealed how higher levels of AI exposure
appear to be affecting workers’ productivity, numbers of job openings, and the skills that
jobs require. First, let’s see how AI may be affecting productivity.
Labour productivity growth has been sluggish in many nations for years. OECD countries
have experienced a lost decade of labour productivity growth with weak average annual
rises of 1.1% from 2011 to 2020, followed by declines in both 2021 and 2022.3
4.8x
higher growth in labour
productivity in Al-exposed
sectors
3 OECD, Labour Productivity and Utilisation. The pandemic had a negative impact on productivity in 2020-2022.
PwC’s 2024 AI Jobs Barometer 10
Al exposure and labour productivity growth rate by sector. Each dot
represents a country.
Sources: PwC analysis of OECD data, Felten et al. (2021). The AI Occupation Exposure (AIOE) constructed by Felten et al’s (2021) AI Occupational
Exposure (AIOE) scores and measures the degree to which occupations rely on abilities in which AI has made the most progress in recent years, meaning
AI can more readily be used for some tasks. The AIOE score is a relative measure, where higher numbers indicate greater exposure to AI, meaning that
even negative values still imply a certain degree of exposure to AI. To measure the growth rate in labour productivity, PwC used the OECD’s GVA per
person employed metric, indexed on 2018. Due to the availability of the OECD data, PwC focused on just six sectors. The 2023 OECD labour productivity
data has not been released.Therefore the labour productivity growth rate between 2018 and 2022 is considered. If the view that AI is increasing productivity
is correct, it would be expected that the pattern of stronger productivity growth for AI-exposed sectors would continue or accelerate in 2023. The ‘4.8x
higher growth’ is a comparison of averaged labour productivity growth rates; absolute growth rates are 0.9% and 4.3%.
PwC’s 2024 AI Jobs Barometer 11
2202
-
8102
etar
htworg
ytivitcudorp
ruobaL
30%
20%
10%
0%
-10%
-20%
-1.5 -1 -0.5 0 0.5 1 1.5 2 2.5
Sectors less exposed to Al Al Sectoral Exposure Sectors more exposed to Al
Average .9% Average 4.3%
noitcurtsnoC
gnirutcafunaM
dooF
,liateR
&
secivreS tropsnarT
lanoisseforP
secivres
noitamrofnI ygolonhceT
laicnaniF secivreS
This stagnant labour productivity is a serious problem because it is a drag on economic
growth, reducing potential tax revenues, chipping away at investment in public services
and flatlining living standards.
Recently there has been much speculation that AI can supercharge workers’ productivity.
The good news is there is now evidence to suggest that this is not just wishful thinking,
and is already fast becoming reality.
We have seen that three sectors - financial services, IT, and professional services - have
higher AI exposure and higher AI penetration. How is this affecting productivity?
The data shows that these three sectors are seeing nearly 5x faster productivity growth
than sectors with lower AI exposure (such as transport, manufacturing and construction).
While it is not possible to prove causation, this is an intriguing pattern. Unlike the
computer revolution which took significant time to enhance productivity (economist
Robert Solow once observed that the impact of the computer age was evident
everywhere but in the productivity statistics), the data suggests AI is already doing so,
right now. AI may be compressing the ‘productivity J-curve’4 in which new technologies
can take significant time to cause a sharp uptick in productivity.
PwC’s 2024 Global CEO Survey confirms that 84% of CEOs whose companies have
begun to adopt AI believe it will increase efficiency in their employees’ time at work.5
Increasing productivity means more than just doing the old things faster. It also means
finding new, AI-powered ways to create value. In fact, 70% of CEOs say that AI will
significantly change the way their company creates, delivers and captures value over the
next three years.
Al does more than help workers do the old things
faster. Al opens the door to new business models
and ways of creating value.
The implications for business are huge. Global CEOs anticipate that one form of AI -
generative AI - will deliver significant top and bottom line benefits, with 46% saying it will
increase profitability, and 41% saying it will increase revenue.
Investors agree. PwC’s 2023 Global Investor Survey shows that investors believe
accelerated adoption of AI is critical to the value equation, with 61% of investors saying
faster adoption is very or extremely important. When responses indicating ‘moderately
important’ are included, the proportion jumps to 85%.
All of this adds up to a positive story for the global economy: a revolution in productivity
and value creation.
4 Productivity J-curve,’ Brynjolfsson et al., National Bureau of Economic Research.
5 Around a third of the respondents in our 2024 Global CEO Survey have begun to adopt AI. Of these, 84% believe it will increase employees’ efficiency.
These findings suggest that companies leading the way on AI deployment are seeing the benefits.
PwC’s 2024 AI Jobs Barometer 12
AI is helping to ease
labour shortages
In AI-exposed occupations such as customer services and IT - a number of which have
acute labour shortages - jobs are still growing, but 27% more slowly on average.
This could be good news for many nations facing shrinking working age populations and
vast unmet needs for labour in many sectors. AI can help to ensure that the labour supply
is available for the economy to reach its full potential.
27%
Lower job growth in
Al-exposed occupations
(though jobs still growing overall)
PwC’s 2024 AI Jobs Barometer 13
Job openings are still growing in Al-exposed occupations, but more slowly
Sources: PwC analysis of Lightcast data, ISCO-08 Occupation Codes (2-digit level) and Felten et.al AI Occupation Exposure. The cross-country
comparison on the right hand side considers the difference in the growth in job postings between the occupations most exposed to AI and those least
exposed to AI.
It is important to emphasise that job numbers in AI-exposed occupations are still growing.
The data suggests that AI does not herald an era of job losses but rather more gradual
jobs growth, helping to enable companies to find the workers they need.
PwC’s 2024 AI Jobs Barometer 14
3202-9102
sgnitsop
ni
etar
htworG
250%
Cleaners and Helpers
200%
Construction and
150% Manufacturing Labourers
Sales and Service Workers
Plant and Machinery
Operators
Clerical Support Workers
100%
Administrative &
Commercial Managers
50% Agriculture Labourers
Business Professionals
IT Professionals
0%
-2.00 -1.50 -1.00 -0.50 0.00 0.50 1.00 1.50 2.00
-50%
Occupation less Occupation is more
-100% exposed to Al exposed to Al
Average growth 80% Al Occupation Exposure Average growth 58%
What this means for workers:
Build skills for an AI age
The skills required by employers in AI-exposed occupations are changing fast. Old skills
are disappearing from job ads - and new skills are appearing - 25% faster than in roles
less exposed to AI.
25%
higher skills change in Al-exposed
occupations
PwC’s 2024 AI Jobs Barometer 15
Change in skills demanded by employers for occupations more (and less)
exposed to Al
Sources: PwC analysis of Lightcast data, ISCO-08 Occupation Codes (2-digit level), Felten et al. (2021). The net skill change is based on Deming and
Noray (2020) and is calculated by using the difference between 2019-2023 in the total number of skills required by job occupations using the ISCO-08 4-
digit occupational codes. The AI Occupation Exposure is from Felten et al’s (2021) and measures the degree to which occupations rely on abilities in
which AI has made the most progress in recent years, meaning AI can more readily be used for some tasks. The correlation coefficient is .31 and is the
statistical measure that quantifies the strength and direction of a linear relationship between net skill change and AI Exposure. To calculate the average net
skill changes for the most and least exposed occupations to AI,an average of the net skill change of the top and bottom quartile of occupations is taken
based on their exposure to AI. See Appendix Two for formula.
Workers in AI-exposed roles may need to demonstrate or acquire new skills to stay relevant
in a jobs market that is fast-evolving. PwC’s 2024 Global CEO Survey makes it clear that
69% of CEOs anticipate that generative AI will require most of their workforce to develop
new skills, rising to 87% of CEOs who have already deployed generative AI. Workers need
to take ownership of their learning, rapidly developing the skills to remain relevant and to
embrace the opportunity AI brings.
PwC’s 2024 AI Jobs Barometer 16
3202-9102
neewteb
egnahc
lliks
teN
14
Greater
change
in skills 12
demanded Web and multimedia developers
by
employer
10
Software
Web
developers
8 technicians
6
Athletes and
sports players Mathematicians,
actuaries and
4 statisticians
Smaller
Hand
change
in skills launderers Judges
demanded 2
by
employer Psychologists
Roofers
0
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2
Al Occupation Exposure
Occupation less Occupation is more
exposed to Al exposed to Al
Average: 2.7 Average: 3.4
(Least exposed quartile) (Most exposed quartile)
There are clues to which skills workers may want to build to prosper in an AI age. Some
of the skills rising fastest in demand are those which cannot easily be performed by AI.
Below are four of the skills categories rising fastest in demand, and for each category a few
examples are provided of specific skills with growing demand. From dam construction to
sports instruction, some skills with booming demand are relatively hard for AI to perform.
FASTEST growing skill categories
Skill sub-category Growth in skills
sub-category
Yoga +426%
Performing
Sport Instructors +178%
Arts, Sports, +155%
Swimming +20%
and Recreation
Creative Arts +18%
Child Safeguarding +156%
Personal Care +82% Laser Hair Removal +84%
and Services Skin Treatments +41%
Funeral Arrangements +11%
Solar Development +87%
Energy and +58% Water Metering +58%
Utilities Energy Trading +44%
Dam Construction +33%
Sediment Sampling +84%
Ecological Restoration +57%
Environment +48%
Waste Collection +32%
Flood Controls +17%
Sources: PwC analysis of Lightcast data. Data based on 2019-2023. The overall growth in skill categories is calculated as the change in the average
share of the skill category for all countries between 2019 and 2023.
On the other hand, what skills are declining in demand? Below are four skills categories
with the steepest declines in employer demand, with a few illustrative examples of
specific skills with falling (or rising) demand within each category.
The AI transformation is clear to see in categories like Information Technology where
demand for AI-related skills like ‘AI/Machine Learning Inference’6 is flourishing, while
6 AI/Machine Learning Inference means applying a machine learning model to a dataset to generate an output, insight, or prediction.
PwC’s 2024 AI Jobs Barometer 17
demand for some skills that may be more readily replaced by AI (such as coding in
Javascript) is falling. The Analysis category shows a similar pattern with soaring demand
for Natural Language Programming (an AI skill) and declining demand for Regression
Analysis, a type of analysis AI can help to perform.
SLOWEST growing skill categories
Skill sub-category Growth in skills
sub-category
AI/ML Inference +113%
Information -26% Smart Devices +81%
Technology Cloud Operations -7%
Javascript -37%
Game Design +12%
Visual Effects -11%
Design -23%
Computer Graphics -30%
Interface Design -46%
Pipeline Management +6%
Consumer Sales -11%
Sales -20%
Online Auctions -23%
Cold Calling -37%
Natural Language +64%
Programming
Analysis -14% Asset Analytics +3%
Data Synthesis -8%
Regression Analysis -21%
Within the slowest growing skills categories, some sub-categories buck the trend and are growing fast. Some of these
(like Al/ML Inference) are Al skills.
Sources: PwC analysis of Lightcast data. Data based on 2019-2023. The overall growth in skill categories is calculated as the change in the average
share of the skill category for all countries between 2019 and 2023.
No going back to yesterday’s jobs market -
but vast opportunities await those who
adapt to an AI age
AI is redefining what it means to be a financial analyst, a software coder, a customer
service agent (and many more roles), opening up whole new possibilities for workers to
deliver impact. Workers who learn to harness AI are likely to have bright futures in which
they can generate greater value and could consequently have greater bargaining power
for wages - all within a context of rising societal prosperity.
Workers agree. PwC’s 2023 Global Workforce Hopes and Fears Survey shows workers
expect mostly positive benefits from AI with 31% expecting AI to increase their
productivity/efficiency and 21% expecting AI to create new job opportunities.
Many who predict AI will cause a sharp decline in job numbers are asking the wrong
question. Those who predict AI will have a negative impact on total job numbers often
look backward, asking whether AI can perform some tasks in the same way as they
have been done in the past. The answer is yes. But the right question to ask is this:
How will AI give us the power to do entirely new things, generating new roles and even
new industries?
PwC’s 2024 AI Jobs Barometer 19
AI makes human labour more relevant and valuable, opening
up new opportunities for people to develop new skills and enter new
roles. AI will create new jobs for people that we haven’t yet begun
to imagine. Many of the fastest-growing jobs of today - from cloud
engineer to digital interface designer - didn’t exist 10 or 20 years ago
and have been generated by technology. Like a spreadsheet or a saw,
AI is a tool that makes people more powerful and capable. Workers
who build the skills to harness AI will be more valuable than ever.
Pete Brown, Global Workforce Leader, PwC UK
AI often performs best in partnership with people. Without oversight, AI can miss context
and nuance or give poorer quality output. Only humans can fully appreciate and navigate
the people, processes, and context of individual organisations and situations.
As technology gets better at being technology, humans can get
better at being humans. There is clear evidence that AI often delivers
the best outcomes when used in partnership with people. The AI era
requires a new style of leadership, an openness to bold transformation
and inventive thinking about how AI and people together can create
new forms of value.
Carol Stubbings, Global Markets and TLS Leader, PwC UK
Our analysis (particularly the finding about AI’s potential impact on productivity) suggests
that AI’s effect on jobs may be similar to that of the internal combustion engine in the 20th
century which reduced numbers of some jobs (such as horse trader) while at the same time
creating far more jobs than it displaced (from truck driver to road engineer to traffic police).
PwC’s 2024 AI Jobs Barometer 20
AI provides much more than efficiency gains. AI offers
fundamentally new ways of creating value. In our work with clients,
we see companies are using AI to amplify the value their people
can deliver. We don’t have enough software developers, doctors,
or scientists to deliver all the code, healthcare, and scientific
breakthroughs the world needs. There is a nearly limitless demand for
many things if we can improve our ability to deliver them.
Scott Likens, Global AI and Innovation Technology Leader, PwC US
Far from heralding the end of jobs, AI signals the start of a new era in which workers can
be more productive and valuable than ever.
Instead of focusing only on how AI can take on some tasks formerly done by people,
we should think inventively about how to make the most of AI to create new industries
and new roles for people. Embracing AI in this way is one way to bring about continued
positive outcomes for workers. Economist Eric Brynjolfsson observed, ‘If AI is used
mainly to mimic humans, to replace humans with machines, it is likely to lead to lower
wages and more concentration of wealth. If we use AI mainly to augment our skills, to do
new things, then it is likely to lead to widely shared prosperity and higher wages.’7
7 The Second Machine Age, Eric Brynjolfsson
PwC’s 2024 AI Jobs Barometer 21
Next steps for companies, workers,
policymakers
There is no going back to yesterday’s jobs market, but - if carefully managed - the AI
revolution could bring a bright future for workers and companies. Below are steps that
companies, workers, and policymakers can take to help realise AI’s promise to grow
productivity and fuel rising shared prosperity.
Here is what companies can do. Business leaders can embrace, experiment, and create
new uses of AI. They can think beyond using AI to do things the way they have been done
in the past and instead use AI to generate new ways to create value. While AI can help to
make existing processes more efficient, companies can realise even more benefit from
AI by using it to reinvent business models or pioneer new product lines. Thinking
inventively about how to use AI helps the company to be the disruptor rather than the
disrupted, and it helps to create new opportunities for people.
PwC’s 2024 AI Jobs Barometer 22
Business leaders should view AI as a complement to people that is best used with
human oversight. Leaders should track the ever-shifting ‘jagged frontier’8 which marks
where AI performs brilliantly versus where AI lacks capabilities or works best with human
assistance. Companies can support employees to make the most of AI by offering
training and helping them see how AI empowers them (and can even make their jobs
more enjoyable by freeing them to work more autonomously and be more confident in
their roles)9.
Firms can consider hiring on the basis of candidates’ skills rather than focusing solely
on their degrees, job history, or previous job titles. This helps firms find the workers they
need, and it helps workers more readily adapt to a fast-changing jobs market. A study
by PwC and the World Economic Forum conducted across 18 economies shows that
a skills first approach has the potential to expand the talent pool by 100 million people.
Companies can take a skills first approach for existing employees too, treating workers as
people with sets of skills and talents that can be fluidly applied across the organisation.10
These ‘skills based organisations’ can more flexibly deploy workers, helping both
companies and workers adapt to the AI transformation while opening up broader talent
pools, developing more resilient talent pipelines for the jobs of tomorrow, and achieving
enhanced levels of employee motivation, satisfaction, performance, and retention.11, 12
Workers, for their part, should embrace AI, experimenting with it and seeking ways it can
complement and enable them in their work.13 Workers should build the skills to be sought
after in an AI age (for example, skills that either complement AI or are hard for AI to do).
Some workers may need to adapt more than others to succeed in an AI era; for example,
some workers may need only a little training to adopt AI tools while other workers may
need to move to new occupations which require more extensive retraining or upskilling.
Workers, companies, and policymakers share responsibility for helping all workers adapt
to an AI era.
8 ‘Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of AI on Knowledge Worker Productivity and Quality,’
Fabrizio Dell’Acqua et al, Harvard Business School working paper
9 MIT Sloan Management Review: ‘Achieving Individual - and Organisational - value with AI,’ 2022.
10 Skills based organisations are integrating skills throughout the talent management lifecycle by implementing skills-based training programs for
upskilling and reskilling, as well as establishing skills-based career pathways for redeployment.
11 Skills-based sourcing & hiring playbook, Rework America Alliance, 2022
12 AI can help with skills based hiring by, for example, automatically generating and updating skills profiles and working out adjacent skills people are
likely to have or could readily learn.
13 Workers whose companies do not offer AI tools can experiment with public AI tools like ChatGPT. Workers should not use proprietary company data
on public tools, but public tools still provide a wealth of opportunities to get to know AI’s power.
PwC’s 2024 AI Jobs Barometer 23
Policymakers can encourage the use of AI to grow productivity and prosperity, for
example by building the supportive policy environment, digital infrastructure, and skilled
workforce to help realise AI’s potential. Countries with the strongest growth in jobs that
demand AI skills (an indicator of AI usage and penetration) offer lessons for policymakers
in how to create an environment conducive to making the most of AI. The three countries
in this study with the highest proportion of jobs that require AI skills are Singapore,
Denmark, and the US. These are the same three countries that top the IMF’s AI
Preparedness Index ranking which measures areas such as digital infrastructure, human-
capital and labor-market policies, innovation and economic integration, and regulation
and ethics. Policymakers who would like their people to benefit from the AI revolution
should take note.
Proportion of total job vacancies requiring Al related skills by country,
2012-2023
Singapore has the highest proportion of Al related
2.5%
job vacancies increasing to 4.8% in 2023
2.0%
1.5%
1.0%
0.5%
0.0%
2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023
Singapore Denmark United States New Zealand Germany United Kingdom France All countries
Sources: The IMF’s AI Preparedness Index ranks countries’ preparedness to adopt AI based on four pillars: Digital Infrastructure, Innovation & Integration,
Human Capital & Policies and Regulation & Ethics.
Policymakers can support workers with training/retraining and safety nets, and shape
the education system to help prepare workers for an AI age in which critical thinking,
creativity, and adaptability are likely to be key skills. Finally, policymakers can strive to
make sure that growing prosperity from AI adoption is widely shared.
PwC’s 2024 AI Jobs Barometer 24
Key areas for action
Policymakers
E ncourage the use of Al to grow productivity and prosperity
Ensure growing prosperity from Al adoption is widely shared
Support the use of Al to augment rather than replace workers
Support workers with training/retraining, worker protections, and safety nets
Shape the education system to help prepare workers for an Al age
Ensure the responsible use of Al with PwC’s Responsible Al framework
Businesses
E mbrace, exper |
217 | pwc | revolutionising-Customs-with-AI.pdf | Revolutionising Customs with AI
Dream big, start small
Foreword
Rajat Chowdhary
Partner, Technology Consulting
PwC Middle East
customs authorities are at a critical crossroad. Global trade is
becoming more complex and security demands are intensifying.
AI offers a transformative opportunity to revolutionise customs
operations – yet its adoption remains fragmented.
This thought leadership on Revolutionising customs with AI: Dream
big, start small, explores how the technology can move beyond
siloed applications to drive end-to-end transformation across the
customs value chain.
By harnessing AI’s capabilities holistically, customs administrations
can achieve greater operational efficiency, enhance security and
gain deeper insights into trade flows, ultimately building a
next-generation customs system that is agile and resilient in the face
of modern trade demands.
Bastian Vomhof
Director, Customs Consulting
PwC Luxembourg
AI is on everyone's mind. For customs administrations, the path to fully
realising the transformative potential of AI begins with elevating their digital
maturity. This journey towards digital transformation is essential for
achieving simplification, enhanced security, and improved trade facilitation
through AI.
At PwC, we bring unparalleled experience and expertise in guiding and
upskilling customs administration through this transformation. We offer
access to global best practices, skills, and support you in experimentation
and innovation to enable AI.
Join us as we envision a future where AI redefines the next generation of
customs excellence.
PwC | Revolutionising customs with AI 2
Introduction
In an interconnected and complex global economy,
customs administrations find themselves at a critical
juncture. As trade volumes surge due to e-commerce,
supply chains diversify, and illicit activities become more
sophisticated. Traditional customs operations struggle to
keep pace, highlighting the urgent need for
transformation. AI is at the forefront of this evolution,
poised to revolutionise customs by driving operational
efficiency, enhancing security, and simplifying
processes.
Currently, many customs administrations explore AI
through fragmented, siloed efforts, such as
experimenting with machine learning for risk
management or using automation for document
processing. While these isolated use cases may offer
localised benefits, they do not provide a cohesive
strategy for unlocking AI’s full potential when applied
across the entire customs value chain.
This paper highlights that AI adoption starts with
experimentation and prototyping, which ultimately lead
to a broader strategic vision. It examines the challenges
in this journey and demonstrates how AI can streamline
and automate customs processes, enhance security,
and facilitate legitimate trade.
customs authorities can embrace a new era of
human-centric AI to simplify and automate their work,
leveraging this momentum to transition to a higher level
of digital transformation for greater efficiency and
effectiveness. In its final sections, the paper outlines
strategies for accelerating AI adoption, offering practical
steps to drive full-scale transformation.
In this journey, AI is not just a solution; it is the
foundation of a next-generation customs system that is
faster, more secure, and capable of adapting to the
evolving demands of modern trade.
PwC | Revolutionising customs with AI 3
Evolving challenges in customs
Customs administrations face challenges in adapting to the digital age. Many have outdated
IT systems and regulations and struggle to analyse ever increasingly volumes of data.
These issues are compounded by the fierce competition for the skills needed to experiment
with emerging technologies. Nevertheless, there are now tangible examples of Customs
Administrations effectively leveraging the benefits of AI implementation.
Challenge Description Case Example
DG TAXUD built a
customs administrations must perform risk collaborative AI platform that
assessments based on hundreds of millions of allows EU Member States to
records gathered through the EU’s Import collaborate and enhance
Control System before loading in a third country their national risk
Safety &
or arrival in the EU. assessment capabilities.
security
In a complex regulatory landscape, customs
In Brazil, customs declaration
authorities must manage diverse international
and verification processes
trade agreements and compliance
were manual and error-prone
requirements. Evolving standards require
Compliance until automated with an
real-time monitoring and reporting, with
& regulation AI-enabled verification tool.
non-compliance risking revenue and security.
Recent global events like pandemics and
To solve supply chain
geopolitical tensions have exposed the fragility disruptions, the UAE, in
of supply chains. customs authorities need to partnership with WEF,
address disruptions in the flow of goods and launched the TradeTech
Supply chain adapt to the unique challenges posed by initiative to harness AI to
improve customs procedures.
& logistics smaller, high-volume e-commerce shipments."
As customs processes become more advanced, Indian customs developed an
a substantial amount of data is generated. AI-enabled system to store
customs authorities require solutions to and analyse data effectively,
integrate this data from various sources and providing valuable insights
Data
analyse it for improved operational efficiency. for decision-making.
management
PwC | Revolutionising customs with AI 4
Transitioning to cognitive customs:
Enhancing efficiency, security, and trade
To manage the evolving challenges in international trade and customs, authorities should aim
to transition towards a cognitive customs state, characterised by advanced technological and
operational capabilities. This transformation requires substantial investment, eventually
embedding AI in the customs operations DNA.
While many customs authorities are currently at the experimental or opportunistic stages of AI
adoption, moving towards a cognitive stage can help custom authorities to achieve many
intermediate benefits, such as develop simpler, fully automated systems that leverage
predictive analytics, optimise processes autonomously, and learn from historical data to
improve decision-making.
This cognitive shift ensures greater efficiency, security, and adaptability in managing the
increasingly complex global trade environment. Additionally, it allows customs administrations
to be proactive in addressing emerging risks, improve trade facilitation, and create a more
transparent and streamlined supply chain, ultimately fostering economic growth and
international cooperation.
Customs AI Maturity Framework
Adopts emerging Cognitive shift: AI as
technologies with part of the Customs
flexible architecture Administration DNA
Process-driven
Basic capabilities operations
PwC | Revolutionising customs with AI 5
ygolonhceT
ytirutam
Non exhaustive
Where do you AI use cases
stand today?
High
Increase container scanning using
AI
Detect abnormal packages with
unsupervised learning
Innovator Leader
Enabled Established Virtual chatbot assistants for trade
and compliance queries
Auto-classify goods from free text
descriptions
Low
Low Operational Maturity High
Real-time translation with any
language of traders and travelers
AI will help customs organisations
unravel untapped potential
Customs authorities should invest in and experiment with AI-enabled technologies to advance
towards a cognitive customs state. This can be done by strategically deploying AI for different
customs use cases, helping unlock new levels of operational efficiency, enhance security, and
streamline processes. While the benefits may take some time to materialise, AI will eventually
drive exponential growth in capabilities, leading to transformative changes in customs
operations over the long term.
After overcoming initial setbacks and adapting to AI
technologies, customs operations begin to see
transformative changes, leading to high transformative
and operational maturity
What actually happens
Cognitive
capabilities
Time
Valley of disappointment What you think should happen
There might be initial challenges in AI adoption that
can be attributed to the complexities involved in Indicates initial expectations, where stakeholders
integrating AI into existing systems. This might might expect a direct and steady improvement in
result in the delay in realisation of AI benefits, cognitive capabilities as AI technologies
slowing growth in the development of cognitive are implemented
capabilities
PwC | Revolutionising customs with AI 6
Integrating AI across the value chain
Customs Administrations should evaluate their entire value chain and pinpoint specific use
cases where AI can Revolutionise and transform their operations.
Non exhaustive
Value chain What it entails Countries Technology Outcomes
Improved security
AI identifies high-risk by enhancing the
Assessing risk along
cargo shipments and detection of
with aiding customs in
Pre-arrival forecasts trade trends, dangerous
proactive planning EU, Canada, enhancing security shipments
Hong Kong,
and efficiency
South Korea
Increased detection
Detecting cargo AI detects anomalies
accuracy and
anomalies in X-ray and fraud patterns,
expedited the
scans and combating improving contraband
Arrival clearance process
smuggling and tax Canada, Hong identification and
evasion by spotting Kong, South combating smuggling
fraud patterns Korea and tax evasion
Automates the
Customs duties and process of
AI-enabled
taxes are calculated determining the HS
Custom Harmonised System
based on the goods' code of goods and
clearance type and value World Customs (HS) Codes calculates their duty
Organisation classification tool
rate in real time
Faster audit
AI ensures trade
Analysing invoices compliance by processes, reduced
Post and certificates analysing invoices and human error, and
automatically to certificates, detecting increased
clearance combat smuggling and UAE fraud to combat compliance
tax evasion smuggling and tax
evasion
Improved
AI-enabled analysis tool compliance and
Analysing customs
is utilised to examine simplified customs
Policy- data for insights,
customs data, providing clearance process
making aiding data-driven Brazil
insights for data-driven
policy decisions
policy decisions
PwC | Revolutionising customs with AI 7
Impact of enablers on AI adoption
A comprehensive business impact analysis, based on various criteria such as business
model, ease of adoption, competitive advantage, and technological disruptions, can help to
focus on specific needs and requirements.
Non exhaustive
Impact on
Enablers Key component Description Investment
adoption
Manage innovation in an
adaptive manner,
Innovation
ensuring the required
management
Agile agility to cope with Low Medium High Low Medium High
changes.
Place customs officers
User centricity and trade at the centre of
UX design the AI integration process. Low Medium High Low Medium High
Understand and govern
own data, make future
Data centricity reforms more data-driven,
Data governance and address privacy Low Medium High Low Medium High
and management concerns.
Invest in capacity building
of customs staff to foster
Skills
a cultural shift. Forming
development
Training interdisciplinary teams to Low Medium High Low Medium High
avoid siloed initiatives.
Break silos within
customs administrations
and between technology
Collaboration
providers to share best
Low Medium High Low Medium High
Strategic
practice, tools and
partnerships
resources.
Leveraging internationally
Technology supported standards to
integration simplify integration and
Interoperability Low Medium High Low Medium High
reduce costs.
PwC | Revolutionising customs with AI 8
How can PwC help you?
• Capture the vision and ambition for AI.
Current state
• Gap assessment of the existing customs capabilities in
and vision three aspects: people, process, data and technology.
• Benchmarking study for alignment and best practice.
• Identifying and prioritising use cases.
• Design user journeys and organisational structure.
Feasibility and • Carry out feasibility studies.
prioritisation • Define governance structure, including key performance
indicators, decision rights, and Responsible,
Accountable, Consulted, and Informed (RACI) matrices.
• Rapid prototyping.
• Develop strategic choices.
Strategy and
• Action planning and roadmapping – for example, on how
action planning
the customs administrations can move into the desired
future state, outlining key milestones and objectives.
• Shape technical descriptions, including requirements,
desired functionalities, and visual prototypes.
• Leverage key enablers that can support your journey,
Prototyping
such as technology, resources, and partnerships.
• Design customs ecosystem architecture, including
integration requirements.
• RFP preparation including pre-qualification, technical
evaluation, scope of work, and technical/functional
Tender and specifications.
• Define Service-Level Agreement (SLA) parameters for
supplier evaluation
different components.
• Support on pre-bid meeting and clarification response.
• Vendor response analysis (technical and commercial).
• Project management activities, including risk mitigation
and daily project coordination.
Programme • Supply installation testing (use case testing) and go-live
monitoring.
management
• SLA monitoring for edge devices, applications, and IT
infrastructure.
• Evaluation of change requests across project life cycle.
PwC | Revolutionising customs with AI 9
eniltuO
metsyS
smotsuC
neG
txeN
tnemssessA
poleved
dna
ngiseD
tnemelpmI
evil
oG
Conclusion
Customs authorities are increasingly adopting AI to boost efficiency, security, and compliance, yet
much of AI’s potential remains untapped due to foundational gaps. To unlock AI’s power, customs
authorities must establish essential enablers to translate technology into tangible gains.
Experimental, proof-of-concept pilot projects, focused on real impacts, are essential for
determining how AI can best fit customs operations.
Alongside this experimental approach, better innovation management is crucial, with an agile
workflow that includes end-user input. While rapid results should remain a priority, AI’s broader
role in driving digital transformation, including improvements in data governance and technology
infrastructure, must not be overlooked.
Establishing mixed AI teams that combine data scientists, risk managers, and other departments
will improve engagement and buy-in across the organisation, reducing the risks of siloed
initiatives. Additionally, advancing customs legislation to allow data standardisation will make AI a
more effective, versatile tool for customs.
To fully harness AI in customs, these initial steps toward a larger vision are essential. Building a
foundation for better coordination, stakeholder involvement, and a results-driven mindset,
customs administrations can leverage AI to deliver safer, more efficient, and reliable processes for
themselves and end-users.
References
1. https://www.wto.org/english/res_e/booksp_e/wcotech22_e.pdf
2. https://www.elibrary.imf.org/display/book/9798400200120/9798400200120.xml?code=imf.org
4. https://www.wto.org/english/res_e/booksp_e/wco_wto_annex_the_case_studies.pdf
5. https://www.apec.org/docs/default-source/groups/sccp/compendiumofsmartcustomspracticesforapec
economies_0424.pdf?sfvrsn=acc161e1_2
6. https://mag.wcoomd.org/magazine/wco-news-104-issue-2-2024/automating-image-analysis-
china-customs-implements-new-model-for-the-development-and-deployment-of-algorithms
7. https://taxation-customs.ec.europa.eu/document/download/bf00c70a-df7f-475f-a6ab-6194b3b89efb_en
Contact Us
PwC Middle East PwC Luxembourg
Rajat Chowdhary Philippe Pierre
Partner, Technology Partner, EU Global Leader
Mobile: +971504293733 Mobile: +352 621 334 313
Email: [email protected] Email: [email protected]
Sharang Gupta Bastian Vomhof
Director, Technology Director, Customs Consulting
Mobile: +971 504326559 Mobile: +352 621 334 109
Email: [email protected] Email: [email protected]
Dipesh Guwalani
Senior Manager, Technology
Mobile: +971 565205132
Email: [email protected]
Contributors
Xavier Lisoir
Managing Director, Customs & AI Ravi Jhawar
Director, Customs Architecture
Mobile: +352 621 334 114
Email: [email protected] Mobile: +352 621 334 430
Email: [email protected]
Soham Rane
Manager, Technology
Mobile: +966 543697178
Email: [email protected]
About PwC
PwC is a global network operating in 151 countries, with over 364,000 professionals dedicated to delivering excellence in Assurance, Advisory, and Tax Services. Beyond our traditional Customs Compliance services for Economic
Operators, we are recognized experts in transformation and modernisation of Customs Administrations worldwide. Our international perspective allows us to serve a diverse clientele, including the EU Commission, GCC public safety,
border security, and Customs Authority, as well as numerous national Customs administrations. Our consulting projects span from strategy to implementation, supporting Customs Administrations in their digital transformation journeys. We
deliver large-scale digital programmes that make Customs more data-driven, efficient and save. Beyond strategic advice, we design and implement innovative Customs solutions in-house, leveraging cutting-edge technologies such as data
analyPticsw, cClou d| c oRmepuvtinog,l auntdi oartnificisiali nintgell igcenucse. tOoumr cosm pwrehiethns ivAe Iapproach ensures practical, hands-on solutions that drive tangible results for Customs Administrations.
12
© 2024 PwC. All rights reserved
Sensitivity: Public |
218 | pwc | AI-insurance_21.08.2023_END.pdf | Artificial Intelligence
for insurance
companies
Presentation by
Petr Novák, Radek Hendrych,
Magdalena Kardela-Wojtaszek
Agenda
1.
Who we are? 3
2.
AI in the current world 6
3.
Selected use cases 10
4.
Our approach and capabilities 14
5.
Summary 22
© PwC Risk Management & Modeling 2
1
Who we are?
© PwC Risk Management & Modeling
Introducing our Team
Petr Novák Radek Hendrych
Director Senior Manager
#datastrategy #actuary #modeling
We are a newly established team within PwC CZ, with a #dataanalysis Data Actuarial #IFRS17 #SolvencyII
strong focus on digital, data-driven services and products. #datadelivery
Insight Excellence
Together, we bring insurance business-subject-matter #datagovernance
experts that can deliver complex transformation projects
from business design and data management up to
regulatory and actuarial consequences.
Our competence is bolstered by an expansive teams of
senior specialists, who provide crucial support to each
Business
project.
Development
Insurance
risk team
100+
Magdalena
Kardela-Wojtaszek
Manager
experts from PwC with
#insurance #productdesign
extensive experience in #businessdevelopment
data, risk and business
© PwC Risk Management & Modeling 4
Meet the Team supporting Us
40+
Our strong skills are supported by large teams of experienced specialists including insurance business experts, risk managers,
data managers, data modelers, actuaries, and more. In total, there are over 40 specialists based in the Czech Republic, with the
potential to expand to over 100 specialists across the Central and Eastern European region.
experts in PwC CZ
100+
Risk managers
Operations Efficiency specialists Data analysts
experts in PwC CEE
Data engineers
Risk modelers
Product Development Managers
Data scientists
Business analysts
Actuaries
Data architects
Project managers
Insurance experts
Business Intelligence specialists
© PwC Risk Management & Modeling 5
2
AI in the
current world
© PwC Risk Management & Modeling
AI in the current world
Artificial Intelligence has been making great strides. It has shifted from an incomprehensible subject of a chosen few “Einsteins” to a daily used assistant.
Companies invest enormous amounts of money in AI to revolutionize various aspects of their operations and gain a competitive edge in the market.
Why is it important to think about AI ?
Cost Reduction
Innovation and Enhanced
Improved
and Process
Competitive customer Economic growth
efficiency
Optimization
Advantage experience
© PwC Risk Management & Modeling 7
PwC's support in achieving
AI-powered business goals
How to effectively use
AI to build up your
company? How to
transform your business
to stay ahead?
PwC can help you to answer those questions
as well as define the right AI vision and
strategy of your company.
© PwC Risk Management & Modeling 8
Clear path to the successful AI implementation
▪1 Management awareness - We will show ▪5 Way of work - We will define the
Our approach the managers what the current and changes in your way of working to
expected AI capabilities are and how include or enhance AI use
they are used/can be used in insurance
Our proven approach ▪6 Platform and tool - We will define the
sector. We will moderate unreasonably
contains a set of deliverables main functionalities needed to use AI
high expectations and challenge low
that help you to define a expectations ▪ 7 Regulation and risks - We will articulate
reasonable path of AI the main risks related to AI
▪2 Use case ideation - We will prepare a
implementation for your implementation in your company and
workshop to think up relevant AI use
explain the existing and emerging
business.
cases
regulations
▪3 Vision - We will help you to formulate the
▪8 Roadmap - We will place all the main
company AI vision
activities onto a roadmap
▪4 Organization and team - We will
propose to you variants of organizational
setup, services, competencies and roles
to implement AI and benefit from AI
© PwC Risk Management & Modeling 9
3
Selected
use cases
© PwC Risk Management & Modeling
AI models | Use cases
AI deployment can benefit insurers across various domains, encompassing pricing, underwriting, claims handling, customer service, and fraud prevention.
Presented below are a exemplary use cases that illustrate how AI can impact internal processes and customer service quality.
Pricing and underwriting Distribution Claims handling
• Analyze volumes of data and make accurate • Understand your customers better and tailor their products, • Optimize internal processes, reduce manual errors,
predictions about risk factors and prices services, and marketing efforts accordingly improve operational efficiency, and lower costs
• Analyze various factors, such as individual • Offer personalized product recommendations and increase • Analyze historical data and patterns and identify
behavior, driving patterns, or health data, to cross-selling opportunities suspicious claims
create personalized pricing models (in terms
• Analyze customer behaviors, market trends, and other
of risk premium or profit margin)
relevant data to predict future outcomes
Analysis & monitoring
• Automate the underwriting process to make
better underwriting decisions
• Identify patterns, trends, and anomalies
Customer service
• Analyze large datasets and perform complex
• Analyze vast amounts of market data, including
actuarial model / calculations (e.g., in terms of
reserving, pricing, CLV) • Use AI-powered chatbots and virtual assistants to handle competitor pricing, customer preferences to stay
customer inquiries competitive in a rapidly evolving market
• Analyze customer feedback from various channels to • Automate compliance monitoring and ensure
understand sentiment, identify issues, and improve adherence to regulations and reduce human error
customer satisfaction
• Automate the generation of reports, dashboards,
and insight
© PwC Risk Management & Modeling 11
Use cases | Claims handling
AI use cases AI use cases
Explore the example of how • Document and photo analysis • Letter preparation for client
• Verification of policy conditions for claims • Client communication
artificial intelligence (AI) • Claim anti-fraud check
integration revolutionizes • Providing claim decisions
• Adjusting case reserves
claim processing, reviews,
• Settlement settings
and settlements.
Automated
Claim Review Claim Rejected
& Investigation
Benefits
Claim Revision
Claim Event Claim Reported
(Ex-post)
• Claim process optimization
Manual • Reduction of manual mistakes
Claim Review Claim Settled • Operational efficiency improvement
& Investigation • Cost reduction
• Enhanced customer experience
• Accurate reserve estimations and
calculations
AI use cases AI use cases AI use cases AI use cases
• Real-time support for claim • Document and photo analysis • Letter preparation for client • Fraud detection
registration • Verification of policy conditions • Client communication • Process monitoring
• Analysis of claim form and for claims • Payment processing (automated/manual)
attached documents for • Claim anti-fraud check • Claim portfolio analysis:
completeness and correctness • Providing claim validity patterns, trends, anomalies
• Claim / client scoring recommendations. identification
(automated / manual)
• Adjusting case reserves • Automated report, dashboard,
• Case reserving and insight generation
AI use cases
• Real-time claim status support
• Actuarial loss reserving
© PwC Risk Management & Modeling 12
Use cases | Pricing
Premium
composition
AI use cases
•Advanced statistical / machine
Profit margin learning techniques to optimize
Benefits
profit margin (e.g., dynamic
discounts) -competitiveness
•Accurate risk assessment
•Customized premiums based on
customer/product risk profile
Expenses
•Competitive pricing
(costs & commissions)
AI use cases
Risk premium •Advanced statistical / machine
learning techniques to model
(expected claim outgo) expected risk more precisely
© PwC Risk Management & Modeling 13
4
Our approach
and capabilities
© PwC Risk Management & Modeling
Data science team
Usually, at the center of AI implementation, there is a data science team. The team should not be
composed only from data scientists, but also data engineers, analysts and DevOps engineers should
be part of it.
These required roles are highly sought after in the job market, making recruitment and retention
challenging. We will provide you with job descriptions and the proper mix of employees and
contractors. To effectively attract and retain these experts, several critical factors must be considered:
the specific use cases, the tools employed, the methodology applied, and the team's composition.
Not all AI development must be realized from scratch by the data science team.
A lot of ready-made AI solutions and knowledgeable suppliers are on the market. But it is important,
data science team provides AI solutions to the company, and it should be their right to decide about
developing the solution themselves or with the help of a supplier. We will address this situation in
your case and bring the elements to decide on whether to Make or Buy.
What are the sources of dissatisfaction among data scientists, and what factors could erode their
loyalty and enthusiasm? One critical aspect is investing time and effort into developing an AI solution
that ends up unused by the company. Surprisingly, this situation is quite common. Another
demotivating factor is navigating through bureaucratic processes involving multiple levels of approval
beyond their control, often leading to extended delays and inefficiencies in project progression.
Additionally, the extended duration needed for data preparation can be discouraging; although they're
keen to create AI models, the necessary data isn't readily available. Lastly, the presence of low-
quality data makes it difficult for data scientists to perform their tasks effectively.
© PwC Risk Management & Modeling 15
Functions of modern data science framework
Modeling data Modeling
availability environment
• Data easily accessible from modeling environment • Modern programming language (e.g., Python)
• Data science team in control of data extraction • Collaboration tools for smooth cooperation inside team
– Shared virtual storage
• Long-term data storage and versioning
– Code version control (e.g., GIT)
– Libraries version control (e.g., conda, poetry)
• Data storage separate from modeling environment – Experiments and models version control (e.g., What does this
• Data extraction controlled by a different team mean?)
• Data provided in flat files without any version control • Virtual machines for computation / memory-intensive tasks
• Outdated programming languages (e.g., SAS)
• Complicated and messy collaboration inside the team
– File sharing through emails
– No code version control
– Different version of libraries on local machines
– No control over experiments and models
• Only local machines with limited computation power
© PwC Risk Management & Modeling 16
Functions of modern data science framework
Documentation Deployment
and outputs to production
• Versioning of outputs (model registry) • Production code in the same framework as development
• Documentation integrated in the data science environment
framework • Project structure created taking deployment into consideration
• Integrated testing
• No systematic control over model versions • Containerization and CI/CD pipeline
• Documentation in separate files without versioning • Scheduling and automation of tasks (e.g., monthly scoring)
• Data science team in charge of production settings
• Production system separated from development environment (or in
different programming language)
• No consideration of deployment during modeling
• Complicated testing procedure
• Difficult integration (and updates) in production systems
• Manual running of production scripts
• Separate DevOps / MLOps team in charge of production
© PwC Risk Management & Modeling 17
Functions of modern data science framework
Monitoring Planning and
and validation project management
• Real-time monitoring / validation of model performance • Modern project management tools (e.g., Jira, Azure DevOps)
• Monitoring / validation integrated in data science platform • Time estimation and real-time tracking for individual tasks
• Alerts and early warnings in case of unexpected behavior • Integration with documentation, code, and outputs
• Fully automated reporting interface
• Project management outside the data science framework
• Ad hoc one-time monitoring / validation reports • Complicated tracking and governance
• Reports has to be created outside data science platform
• No integration with documentation, codes, or outputs
• No real-time information about potential problems
• Requiring manual effort, consuming time and human resources
© PwC Risk Management & Modeling 18
Functions of modern data science framework
Modeling
environment
• Faster project delivery
• Increased quality of delivery
• Higher effectiveness and costs savings
• Reduction of risk of error / miscommunication
• More extensive control over the product
• Improved project management
• Flexibility and ability to quickly react to new situations
• Support experimentation and innovation
© PwC Risk Management & Modeling 19
Building data science framework/platform
On top of traditional Example technical solution implemented in MS Azure: External raw data
data solutions, we
implement a platform • IDE: Jupyter notebooks*
dedicated to machine • Computation framework: Azure Machine Learning Studio Data lake DWH
learning and AI use • AutoML: Azure Machine Learning**
cases. The platform • Model registry: Azure Machine Learning** (MLFlow)
consists of all the tools
• Experiments: Azure Machine Learning** (MLflow) Consumers Data science platform
data scientists would
• Feature store: Azure Machine Learning** (MLFlow)
need to accomplish
• Monitoring & logging: Azure Machine Learning** (MLflow) + Azure Monitor
their tasks. Artifact Project Model
• Orchestration: Azure Data Factory (Airflow), Azure Databricks* Workflows
CI/CD
store management registry
• Artifact store: Azure DevOps Artifacts
The platform could be
• Project management: Azure DevOps Boards + Wiki
implemented on- Code Computation
• Code versioning: Azure DevOps Repos Orchestration
premise or in-cloud - versioning framework
• CI/CD: Azure DevOps Pipelines
e.g., Microsoft Azure.
• Storage: Azure Data Lake Storage + Delta Lake + Azure SQL Database Monitoring &
AutoML Storage
* or other open source IDE (e.g., VS Code) logging
** or Azure Databricks (Spark)
IDE Experiments Feature store
© PwC Risk Management & Modeling 20
Building data science framework/Way of working
The ultimate goal of data scientists is to prepare solutions that
automatically provide smart advice from data or that automatically
answer questions or can just conversate.
The important role to benefit from data science is the business or
product owner. They are atheperson who can connect business
opportunities with data science capabilities and has the power to
decide where to use data science.
There are 3 related processes to achieve the goal:
1. Business delivery - from idea, through cost/benefit analysis,
objective specification, changes in the organization, product and
client service up to business monitoring
2. Data science delivery - from data preparation, through model
training and deployment up to model monitoring
3. DevOps delivery - automation of data processing for the model,
creation of CI/CD pipelines, encapsulation of the model to an
application, operation and technical support
© PwC Risk Management & Modeling 21
5
Summary
© PwC Risk Management & Modeling
Data culture strategy
Having the best of breed data platforms and excellent data specialists does not mean the company
will succeed in gaining maximum value from data. Something that is required to win in the
competitive fight in the digital world.
No endeavors by data specialists can cover all the company’s and the employees’ needs to get
responses from data.
The employees must learn how to understand and use data to find answers to their basic questions:
so-called BI self service. There are effective tools in the market powered by AI, which can help them
in this effort. It is also important the employee experience the difference between a decision based on
data and a decision based on their intuition. In addition, the employees should recognize and
understand the problem when using low-quality data.
Another part of data culture lies in the employees’ understanding how they can benefit from and
work with data specialists, mainly data scientists and what AI can bring to them.
PwC has the experience and know-how to focus and structure data educationand how to use the
tools to achieve a steep learning curve and the best results.
© PwC Risk Management & Modeling 23
Interested?
Contact us.
Petr Radek Magdalena
Novák Hendrych Kardela-Wojtaszek
Data Delivery, Actuarial and Risk Modeling, Insurance business development,
Risk Management and Modeling Risk Management and Modeling Risk Management and Modeling
PwC Czech Republic PwC Czech Republic PwC Czech Republic
T: +420 602 282 972 T: +420 734 542 531 T: +420 732 999 650
M: [email protected] M: [email protected] M: [email protected]
Thank you
PwC Czech Republic
Risk Management & Modeling
pwc.cz/rmm
© 2023 PricewaterhouseCoopers Česká republika, s.r.o. All rights reserved. “PwC” is the brand under which member firms of PricewaterhouseCoopers International Limited (PwCIL) operate
and provide services. Together, these firms form the PwC network. Each firm in the network is a separate legal entity and does not act as agent of PwCIL or any other member firm. PwCIL
does not provide any services to clients. PwCIL is not responsible or liable for the acts or omissions of any of its member firms nor can it control the exercise of their professional judgment or
bind them in any way.
© PwC Risk Management & Modeling |
219 | pwc | SEE-NextGen-Survey-Report-2024.pdf | PwC’s Global NextGen
Survey 2024 – SEE
Report
Transformation, succession and the next generation of family business leaders
The mid-2020s are particularly significant for the Central and Eastern European (CEE) region. May
2024 heralded the 20th anniversary of the accession of Czech Republic, Estonia, Hungary, Latvia,
Lithuania, Poland, Slovakia and Slovenia to the European Union, joined three years later by Bulgaria
and Romania. This growing maturity is reflected in the current challenges faced by family businesses,
which are of crucial importance to the regional economy, with up to half of private sector
employeesbeing employed by family-owned and run companies.
The majority of Southeastern European (SEE) family businesses are still run by the first generation of
post-communist era entrepreneurs who formed them in the early 1990s. According to the 2023 CEE
Family Business Survey, 64% of CEE family businesses (double the 32% global figure) are in the
hands of the first generation and 36% have passed to second or further generations (vs. 68%
globally). However, with many of the first wave now in their 60s or 70s, this decade is seeing the
approach of a changing of the guard in leadership in family businesses. NextGen, or the next
generation of family business leaders who are typically aged between young adulthood and their early
40s, are getting closer to attaining senior executive roles.
The global PwC Global NextGen Survey is particularly pertinent to SEE, illustrated by the record
number of participants this year of respondents from SEE and the wider CEE—and the high
percentage of family businesses still run by their founders. There are clear signs that these pioneering
businesses are at a crossroads—and are looking inward and outward to plan a strategy to remain
viable and prosper in the coming decades.
The contribution of family businesses to SEE economies cannot be ignored and
underestimated. In fact, in many countries in our region, these companies are the backbone
of the national economies. Succession planning is key to future-proofing all businesses but
has particular significance for family businesses. NextGen in many cases are already in
managerial positions—preparing to take over as CEOs or Board members. The SEE edition
of the report explores the key challenges and priorities on the horizon NextGen. It also shows
that local traditions and business mentality have their impact on how these companies are
increasingly seeing the imperative to transform.”
Bojidar Neytchev
Partner, Entrepreneurial and Private Business Leader
PwC SEE
PwC | Global NextGen Survey 2024 2
Key findings
Only have a minute? Consider these notable findings from the SEE edition of the
NextGen survey.
NextGen, or the next generation of family business leaders, are typically aged between young
adulthood and their early 40s, and are getting closer to assuming control of family-run
businesses in the SEE region.
Achieving growth is a key priority for NextGen in the next two years. Over half (52%) of
respondents see achieving growth as a top priority, followed by expanding into new sectors or
markets and talent management. There is a reasonably good match between business
priorities and where NextGen are engaged.
NextGen globally and in SEE feel they can add most value in terms of professionalising and
modernising management practices (22% in SEE vs. 21% globally), diversifying the services
and products that the business offers (16% in SEE vs 9% on a global level) and international
expansion (14% in SEE vs. 8 % globally).
NextGen in SEE are reasonably positive on issues related to clarity of roles and
responsibilities (63%). Also, 55% of NextGen in SEE believe that their company has a clear
governance structure, compared to 68% of the current generation. At the same time, they are
less positive about their company’s digital capabilities and confess that in more than one-third
(33% vs. 36 % globally) of their companies there is a resistance to embracing changes.
Despite a realisationthat there is a need to harness generative AI, SEE family businesses
have been slow to implement the technology. Although 47% (compared to 30% globally) are
in the early stages of exploration, 33% of family businesses in SEE have yet to begin their
generative AI journey. Only 6% have implemented AI in some or many parts of their
organisations. Furthermore, only 12% of family businesses in SEE have a person or team in
the company directly responsible for generative AI.
Responses show a clear difference in how much NextGen in SEE are engaged in generative
AI now - and how much they expect to be in the future. Only 12% of NextGen in SEE are
currently engaged in generative AI - but a further 53% expect to be engaged in it in the future.
PwC | Global NextGen Survey 2024 3
Key findings
Only have a minute? Consider these notable findings from the SEE edition of the
NextGen survey.
NextGen is overall positive about the technology’s potential, despite being aware of the risks
of AI adoption. However, only 10% of NextGen in SEE believe that generative AI will increase
their company profitability in the next 12 months (half of the global figure) and only 29% (44%
globally) believe that in the next three years, generative AI will significantly change the way
their companies create, deliver and capture value.
A slightly greater majority of NextGen in SEE (86% vs 82%) are personally interested in
generative AI than global averages. Perceived knowledge levels in the region are high. 59%
of Nextgen in SEE feel they are personally knowledgeable about generative AI vs 53%
globally.
Only 4% of NextGen in SEE report that their family businesses have already defined
governance around using AI responsibly, although a further 59% believe they need to do this.
NextGen in SEE generally feel more positive than global averages about family trust levels.
However, only 27% believe there are high levels of trust between family members and non-
family members within the business. Less than half (47%) believe there are high levels of
trust between NextGen family members and the current generation.
NextGen is concerned about succession planning. Just over half (51%) of NextGen in SEE
are aware of a succession plan in their family business, but many of those were not involved
in its development and further 6% do not know if there is a succession plan.
53% of NextGen in SEE believe the ability or readiness of the current generation to retire is a
difficult aspect of succession and 45% believe proving themselves as a new leader is will be
also be difficult.
PwC | Global NextGen Survey 2024 4
Achieving growth is the key
priority
Achieving business growth is comfortably the key priority for NextGen in SEE over the next two
years, with well over half of respondents citing growth as most important. Given that the second
biggest priority is expanding into new sectors, this chimes with the results of last year’s Family
Business Survey, where 91% of CEE respondents reported that growth is important because it
enables investment in their company’s future. This suggests that both current and NextGen leaders
are doing all they can to adapt to uncertain times in order to pursue growth.
Question
From your own personal point of view, what would be your top three priorities for the company over the next two
years?
In which areas, if any, are you personally actively engaged at present or likely to be engaged in the future?
SEE Global
59
Achieving business growth
67
41
Expanding into new sectors or markets
43
Talent management -attracting/retaining the best 41
young talent 43
Improving the working conditions/practices of our 24
employees 27
Ensuring we offer the right products and services 22
for today’s customers 51
22
Investing in innovation and R&D
27
20
Adopting new technologies
37
Increasing our focus on investments for 20
sustainability and impact 27
8
Upskilling the digital capabilities of our workforce
22
Supporting our local community via increased 8
investment or business activity 16
Reconsidering our asset allocation and 8
investments 14
Reducing the organisation's environmental impact 4
18
Increasing our focus on privacy and cybersecurity 2
2
PwC | Global NextGen Survey 2024 5
In today's dynamic market, family businesses must strategically focus on
both organic growth and new business growth to ensure long-term
success. For organic growth, staying vigilant to structural changes in
established markets is essential. Embracing digital transformation can
optimize operations and enhance customer experiences. Investing in
talent and fostering a culture of innovation can drive continuous
improvement and adaptability. For new business growth, identifying
opportunities in emerging markets experiencing structural changes can Luis Ndreka
provide a significant edge. Exploring new sectors and forming strategic
CEO of Lufra
partnerships can open additional growth avenues. By focusing on these
Foods, Albania
dual strategies, family businesses can navigate challenges and secure a
prosperous future.”
Despite the current generation of family business leaders citing changing market conditions,
innovation and development as fundamentally important, family businesses have a reputation for
having somewhat traditional and conservative mindsets. Both globally and in the SEE region,
NextGen reveals a willingness to explore new ideas and business practices. The principal areas
where they believe they can add the most value are: professionalisation and modernisationof
management practices and international expansion.
This more than hints that NextGen has one eye on leading business transformation once they assume
the roles of key decision-makers. This is especially relevant considering diversifying the services and
products that family businesses offers and separating family ownership from management can be two
of the key actions that NextGen may bring to the top tables.
Orbico started as a logistical and distribution startup in the late eighties and early nineties. At
that time, we could develop and grow by adding new partners, services, and territories to our
portfolio. Last twenty years we have been considered as a reliable partner of global leading
producers and are encouraged to acquire existing players in our region of Central and
Southeast Europe.”
Stjepan Roglic
Deputy Chairman of the Supervisory Board
Orbico d.o.o, Croatia.
PwC | Global NextGen Survey 2024 6
Question
Where do you think that you can personally add the most value to your family business?
SEE Global
Professionalising and modernising management practices 22
21
Diversifying the services and products that the business 16
offers 9
14
International expansion
8
10
Separating family ownership from management
8
Having a clearly defined purpose, i.e.ensuring the business 8
is not just about making profits 11
8
Investing in new business ideas
10
8
Reinvesting more profit into developing the business
8
Having a business strategy fit for the digital age 6
10
Starting my own venture (supported/financed by the family or 2
operating under the family holding)) 4
2
Attracting and retaining talent
4
Starting my own venture (not supported/financed by the 2
family) 2
Upskilling staff 3
Partnering with start-ups 2
NextGen in SEE is notably less positive than current leadership on issues related to clarity of roles
and responsibilities, governance structures and digital structures. There is a clear generation gap in
terms of role clarity and governance, with NextGen both globally and in SEE over 10% more
pessimistic. Furthermore, one-third (33%) of NextGen in SEE and slightly more of their global
counterparts (36%) believe there is institutional resistance in their family business to embracing
change. Additionally, less than one-quarter (24% vs. 34% globally) see appropriate protocols or a
constitution in place. The survey suggests, therefore, that NextGen clearly feels there is work to be
done in transforming the governance and the strategic direction of SEE family businesses.
Question
How strongly do you agree that…?
SEE Global
Global
There are clear roles and responsibilities for those involved in running the business 63 63
We have a clear governance structure 55 51
We have strong digital capabilities 41 31
There is a resistance within the company to embrace change 33 36
We have family protocols/a constitution in place 24 34
The current generation does not fully see opportunities related to technology
24 29
transformation within the business
PPwwCC || GGlloobbaall NNeexxttGGeenn SSuurrvveeyy 22002244 7
A gap between NextGen personal
interest and company action on
generative AI
In any business or sector, implementing GenAI is a marathon, not a
sprint, requiring a careful balance between urgency and prudence: move
too slowly, and you lose out to competitors; move too quickly, and the
risks increase significantly. To bridge the gap between awareness and
value creation, family businesses should treat GenAI implementation as
a strategic initiative rather than an operational one centered solely on
functions, tools, or technology. They should focus on building a
trustworthy, traceable data foundation and appropriate governance which Snezhana Ilieva
are essential for responsible implementation, with humans playing a key
Director AI and
role in the process. An emphasis should be placed on building
Data Science,
awareness amongst employees - effective use of generative AI depends
PwC SEE
on staff proficiency. Finally, businesses should build a balanced
ecosystem of partners to address GenAI needs across the tech stack
while avoiding vendor lock-in.”
High interest and knowledge surround AI but limited
action so far
The recent CEE edition of the Global CEO Survey alluded to CEOs being somewhat more neutral than
their global counterparts on generative AI, with 10% less believing the technology will significantly
change the way their company creates, delivers and captures value over the next three years.
Nonetheless, CEOs still see a clear direction of travel in terms of AI, with well over half (59%) predicting
that generative AI will be vital to transformation.
The NextGen Survey identifies comparable patterns. Most NextGen in SEE (86% vs 82% globally) are
personally interested in generative AI—and a majority perceive their knowledge levels to be high. 59%
of NextGen in SEE consider themselves knowledgeable about generative AI, which is higher than the
53% global figure.
Despite an awareness that there is a need to harness generative AI, SEE family businesses have been
slow to implement the technology. Although 47% (compared to 30% globally) are in the early stages of
exploration, 33% of family businesses in SEE have yet to begin their generative AI journey. Only 6%
have implemented AI in some or many parts of their organisations. Furthermore, only 12% of family
businesses in SEE have a person or team in the company directly responsible for generative AI.
PwC | Global NextGen Survey 2024 8
These results point to two things. Firstly, there may be more evidence of a generation gap between
current leaders and NextGen. NextGen clearly recognise the transformational potential of generative
AI—but also report that their family businesses have been slow off the mark in terms of adoption.
Secondly, connected to the above—it is clear that the pace of generative AI technological
advancement is relentless, but the reaction of many family businesses hasn’t gotten even close to
matching this speed.
Despite their broad realisation that there is a need to harness AI technology, SEE family businesses
have moved towards generative AI adoption at a fairly pedestrianpace in many cases. They have to
pick up speed in order to benefit from the technology—and quickly.
Question
How would you describe your family business’s current level of adoption of generative AI?
SEE Global
8
No activity: Our company has currently prohibited its use
9
33
No activity: We have not yet started to explore 40
47
Early stages of exploration
30
2
Currently testing and piloting 7
2
Tested and paused
1
6
Currently implemented in a few areas
6
Currently implemented in many areas 1
2
Other
1
5
Don`t know
Digitialisation and automation of processes is an inevitable part of business. We are actively
investing in this direction to be more efficient, to create more convenience for customers, and
to make the work of our teams even more precise. No matter how fast generative AI
develops, it will not replace humans, their intelligence, and their creativity. I believe in people
and the immense potential of each succeeding generation.”
Vladimir Nikolov
Operational President at FANTASTICO, Bulgaria
PwC | Global NextGen Survey 2024 9
Forecasting growth in the deployment of AI in the medium,
rather than short term
Given the slow pace of AI implementation in SEE family businesses, it is perhaps unsurprising that
just over one in ten (12%) of NextGen report their organisation has a person or team directly
responsible for generative AI. Also, responses show a clear difference in how much NextGen in SEE
are engaged in generative AI now—and how much they expect to be later. Only 12% of NextGen in
SEE are currently engaged in generative AI—but a further 53% expect to be engaged in the future.
Question
To what extent do you agree with the following statements about generative AI?
SEE Global
In the next three years, generative AI will require most of our 41
workforce to develop new skills 48
In the next three years, generative AI will significantly change 29
the way our company creates, delivers and captures value 44
Generative AI will increase our company’s profitability in the 10
next 12 months 21
Generative AI will mean a reduction in our company’s 8
headcount in the next 12 months 18
Generative AI has already changed our company’s technology 4
strategy
15
There is an important question that
these results pose—especially
when considering that over one-
third of SEE NextGen believe
generative AI will require the
majority of their workforce to learn
new skills. Will there be a significant
reconfiguration of strategic priorities
towards generative AI once
NextGen assumes executive roles
in family businesses?
PwC | Global NextGen Survey 2024 10
The 41% difference between NextGen who are currently engaged in AI and those who believe they
will be in the future certainly points to an acceleration towards AI once NextGen takes the reins.
Additionally, only 10% of NextGen in SEE believe that generative AI will increase the company’s
profitability in the next 12 months—which is half of the 21% reported globally.
This gives the impression that the handover from the first wave of SEE entrepreneurs to a more tech-
savvy NextGen is seen by the latter as a catalyst for more transformation and profitability in the
medium to long term through AI adoption.
The transition from interest to implementation in generative AI in family businesses involves strategic
planning, education and integration efforts. Family businesses should start educating their leaders
and employees about what generative AI is, its potential and how it can benefit their specific
businesses.
A clear majority of NextGen in SEE are generally positive about what generative AI can potentially
bring to their family businesses. There are also clear signals that there is some trepidation, primarily
about the risks surrounding AI—in particular cybersecurity—and also concerning the sheer pace of
the technology’s evolution. Approaching half (41%) of NextGen see phishing attacks, data breaches
and other cyber risks as likely to increase due to generative AI. Significant numbers of between one-
quarter and one-fifth see risks in the spread of misinformation, bias towards specific groups of
customers or employees, and legal liabilities and reputational risks.
Question
To what extent do you agree that generative AI is likely to increase the following in your company in the next 12
months?
SEE Global
48
Cybersecurity risk (e.g.phishing attacks, data breaches)
41
33
Spread of misinformation
27
29
Legal liabilities and reputational risks
22
25
Bias towards specific groups of customers or employees
20
PPwwCC || GGlloobbaall NNeexxttGGeenn SSuurrvveeyy 22002244 11
Despite being aware of the inherent risks in AI adoption, there is overall positivity about the
technology’s potential among NextGen. This is demonstrated by a clear majority of 76% (compared to
73% globally and 68% in other parts of CEE) seeing AI as a powerful force for business
transformation. The perceived key benefits of AI by NextGen in SEE are increased operational
efficiencies, cost savings and improved decision making. AI is also seen as being capable of helping
SEE businesses adopt new technologies, upskill the digital capabilities of the workforce and achieve
growth.
Question
To what extent do you agree with the following statements about AI generally? And to what extent do you agree
or disagree with the following statements about AI and your family business?
SEE Global
76
AI is a powerful force for business transformation
73
65
AI seems to evolve so quickly that it’s hard to keep up
66
61
It’s difficult to know how to capitalise on AI
51
In the next three years, AI will lead to more competitor pressure 61
(e.g., new products, entrants, prices changes) 64
I feel I can personally help the business to navigate emerging 49
technologies / AI 42
Being an AI champion will help me move into a leadership 39
position 40
There is an opportunity for family businesses to take a leading 29
role in the responsible use of technology and AI 50
This optimism is tempered, however, by two-thirds (66%) of NextGen reporting they believe that AI
seems to evolve so quickly that it's hard to keep up, and that well over half (58%) feel that AI will bring
more pressure from competitors.
The survey statistic that perhaps speaks loudest about the extent of work to be done by SEE family
businesses is that only 4% of NextGen report their family business has already defined governance
around using AI responsibly, although a further 59% believe they need to do this. SEE family
businesses need to quickly get moving in the AI direction that NextGen clearly indicates as the one
they must travel on.
PwC | Global NextGen Survey 2024 12
Current leaders and NextGen
have work to do in building and
maintaining trust
It is not a surprise that as young and dynamic persons NextGen
understand the trust issue that the consumers of their family business
may have in emerging technologies used. On the other hand, given this
response we may expect that NextGen will prioritise building consumers'
trust in technologies when they get more involved and in control of the
family operations. So, I would rather see this as a positive sign indicating
a potential future development of family businesses. We may expect that
more attention will be devoted in those businesses on technology issues Miroslav
as well as on transparency and integrity matters. Depending on the Marchev
family businesses' profile they can approach big technology providers to
Country
help them introduce new solutions, team up with startups to explore new
Managing
opportunities or eventually try to develop new and more trustful smart
Partner, PwC
solutions internally. Anyway, interesting times in this respect are ahead
North Macedonia
of us and NextGen will clearly try to lead the way on technology related
topics.”
NextGen in SEE generally feel more positive than the global
average about family trust levels. However, only 27% believe
there are high levels of trust between family members and
non-family members within the business and less than half
(47%) believe there are high levels of trust between NextGen
family members and the current generation. This apparent
lack of trust among different groups within the business is not
the best basis for some companies to undertake succession
planning, which is discussed in the section below.
PwC | Global NextGen Survey 2024 13
Question
How much trust would say there is between…?
5 –High levels of trust 4 1-3 –Lower levels of trust
Global
5 –High Levels
NextGen family members and the current
47 31 22 32
generation
Family members outside the business and family
45 27 29 28
members working in the business
Family owners and non-family management 31 29 41 22
Family members and non-family members within
27 37 37 23
the business
*Based on CEE editions of the 2023 Family Business Survey and the 2024 NextGen Survey.
These issues around trust also have consequences for the business as a whole—as family
businesses in the SEE region tend to build their reputation upon trust. There is an acceptance that
consumers may also have trust issues. Although approaching half of NextGen in SEE (45%) believe
their consumers have medium levels of trust in businesses to responsibly use emerging
technologies, only 12% of NextGen believe consumers have high levels of trust in this context.
As explained in the 2024 Edelman Trust Barometer Global Report, while family businesses remain
the most trusted type of business, implementation of innovation is just as important as invention.
Mismanaged innovations are more likely to create a backlash than build consumer trust.
With the apparently slow pace of AI experimentation and even slower implementation generally in
SEE family businesses, NextGen (and the current generation of leadership) has a major challenge on
their hands. This challenge isn’t only around increasing their companies' understanding of the
business benefits and risks of AI—but implementing it in a way that retains and increases the
confidence of consumers.
PPwwCC || GGlloobbaall NNeexxttGGeenn SSuurrvveeyy 22002244 14
In order to earn consumers’ trust over time, businesses need to focus on open communication
and transparency when starting to implement emerging technologies. They need to provide
detailed information about data collection and processing, as well as about all data protection
measures undertaken. Education and awareness will foster the process and ensure
consumers are aware of the benefits of emerging technologies and how the business is using
them to provide a higher quality product or service. Finally, CEE family businesses have built
a strong reputation and trust throughout time that can be positively leveraged in the
implementation process.”
Mihaela Kozanova
Cluster Business Development Manager of Sofia Hotels Management
Bulgaria
PPwwCC || GGlloobbaall NNeexxttGGeenn SSuurrvveeyy 22002244 15
More effective succession planning
is needed to bridge the generation
gap
PwC can help to smooth up the transfer of the business to the next
generation. This includes helping to establish a legal structure that
supports the family business's unique needs and the unwritten rules that
govern family interactions within the business context. We can also
assist in defining managerial roles for family members, complementing
these with external experts where necessary to bring in fresh
perspectives and specialised knowledge. PwC ensures that the
succession process does not negatively impact continuance of the family Milivoje
business and or business executives, making all parties involved in the Nesovic
process and with mutual respect and recognition.”
Partner, EPB
Leader, PwC
Serbia, Bosnia
and Montenegro
While emotional, social and personal issues play important roles in all businesses, intergenerational
and interfamilial ties greatly heighten these dynamics in family companies. Given the advancing years
of the first generation of family business entrepreneurs, it follows that handing over to the next
generation of leaders is at the forefront of their thoughts. Last year’s CEE Family Business Survey
showed that succession planning is a growing challenge for many family-owned businesses in the
region.
In the CEE Family Business Survey, almost three-quarters of leaders (74%) reported that ensuring
their business stays in the family is a key personal long-term goal. However, the survey also
suggested that many first-generation leaders had not taken appropriate steps to achieve this goal.
Only 69% of them have some form of governance policy in place within the business compared to
81% globally—and 10% fewer than global averages had a document including a last will and
testament (25% in CEE, 35% globally).
This concern over succession planning is shared by NextGen. Just over half (51%) of NextGen in
SEE are aware of a succession plan in their family business. However, a significant minority of 18%
were not involved in its development. Concerningly for the family businesses’ importance to SEE
economies, as many as 43% reported that their business had no succession plan in place, and a
further 6% said they were unaware of a plan.
PwC | Global NextGen Survey 2024 16
Question
Are you aware of your family having a succession plan in place?
I don`t know if No, there is Yes, bit I was not involved Yes, and we have developed
there is a plan no plan in its development the plan together
SEE Global
6
12 17
43 28 22
20 22
18
33 39 39
2024 2024 2022
*Text
A number of factors may be at play in regard to this apparent lack of planning in numerous family
businesses. A majority of NextGen identify the ability or readiness of the current generation to retire
as a key issue in succession planning. 53% see this as a fairly or very difficult aspect of succession
and 45% believe proving themselves as a new leader/board member will be challenging.
Our whole family has always been involved in the activities of FANTASTICO GROUP. There
is currently a third generation of the family working in the company, and I would say that
continuity is more of a natural process as we all, including myself, have been working as a
team since our youth. We have gone through various positions to know the processes well,
including in a supermarket. At the same time, we rely on experience, but we are also brave
enough to accept the new ideas of younger colleagues. This applies both to our family and to
the whole company's team. About 3,400 people are part of our team, and 84 of them have
over 20 years of experience in the company.”
Vladimir Nikolov
Operational President at FANTASTICO, Bulgaria
PwC | Global NextGen Survey 2024 17
The question perhaps is—does the current generation share NextGen’s confidence that they are
ready to lead family businesses into the coming decades? There may also be intergenerational
communication issues, as well as the ones of trust. SEE family businesses would do well to
remember that passing on ownership and responsibility is not a straightforward matter of passing on
the torch upon retirement. Rather, the process hinges upon the establishment of an appropriate legal
framework. As well as ownership and executive decision-making, this structure should also pay
attention to the values and mission of the company, how knowledge is passed on, and importantly—
how all of this is communicated based on established rules.
Succession planning isn’t, however, just about intergenerational succession, it is also about
opportunities. NextGen in SEE generally feel positive about their career opportunities and ambitions
in their family business. A vast majority (92%) moderately or strongly agree that they have the
opportunity to learn and grow within their family business. This perhaps points to an engaged, hungry
NextGen who are slightly unsure, but also excited and positive about their and their family
businesses’ futures.
Question
In your view, how easy or difficult are the following aspects of succession?
Difficult Easy
SEE 24 76
Learning about business essentials (i.e.
accounting, finance, operations) Global 29 68
SEE 29 65
Discovering your own strengths and
passions
Global 31 66
SEE 33 63
Understanding and getting insights on your
own Family Business / Office
Global 32 64
SEE 45 49
Proving yourself as a new leader or board
member
Global 48 46
SEE 53 41
Ability or readiness of the current generation
to retire
Global 55 37
PPwwCC || GGlloobbaall NNeexxttGGeenn SSuurrvveeyy 22002244 18
NextGen should be drivers
of sustainability goals
With the introduction of CSRD and sustainability reporting family
business will be impacted on different scales. Large companies under
local laws and regulations will be directly impacted with an obligation to
prepare sustainability reports, however impact will be felt by small and
medium private enterprises as it impacts their customers and suppliers
which will require them to adopt throughout their value chain. Family
Slaven Kartelo
businesses which recognise this as an opportunity and investment
Partner,
(instead of cost) and set smart sustainable strategy and operations will
Sustainability
have multiple benefits, including conti |
220 | pwc | healthcare-reboot-for-the-gcc.pdf | A Healthcare Reboot
for the GCC
How AI can supercharge your
healthcare ecosystem
Introduction
Artificial Intelligence (AI) in healthcare is no longer a distant dream. Natural language processing (NLP) can, among
other things, now record doctors’ notes, interpret medical histories, and analyse scientific literature more thoroughly
than humans. Deep-learning algorithms applied to wearable sensors, genomic data, blood work, scans, and other
medical information can develop personalised treatment plans. And virtual medical assistants powered by AI can
offer health coaching, shape diets, and even help prevent and predict illness.
AI in healthcare has evolved from basic data patterns to complex predictive models. Leading institutions use AI to
analyse vast amounts of clinical data, improving diagnosis and workflows. Mayo Clinic’s virtual primary care system
in the United States, for example, uses AI to optimise patient interviews and diagnostic processes, with physicians
selecting AI-recommended diagnoses in 84% of cases.1
In 2008, the Singapore Ministry of Health established Synapxe (originally the Integrated Health Information System)
to connect people and systems for a healthier Singapore. Partnering with major technology providers, Synapxe
developed the AI-powered Health Discovery Plus (HD+) system, enhancing healthcare services and workflows. HD+
supports remote monitoring and management of conditions like hypertension and chronic kidney disease. Developed
during the pandemic, it significantly improved patient care by processing over 660,000 vital sign reports and issuing
over 45,000 clinical alerts. Innovations like HD+ demonstrate AI’s potential in remote patient monitoring, diagnostics,
and patient-managed treatment planning, driving global healthcare toward AI-augmented care.
Figure 1: Moving from Advanced Analytics to Large Language Models
Range of techniques that aim to understand and predict /
Advanced Analytics forecast data patterns through statistical analysis and other
complex methodologies
Computer systems designed to replicate human
Artificial Intelligence intelligence, enhancing human cognitive functions and
decision-making processes
Subset of AI that enables machines to analyse and learn
Machine Learning from existing datasets, to make decisions and predictions
more accurately
Algorithms that use prompts or existing data and learned
GenAI patterns to generate new and original content, such as text,
code, images, videos, audio
Large Subset of GenAI, trained on massive datasets to process
Language and interpret human language and translate it to
Models computational embeddings
Although interest in healthcare AI opportunities has been growing exponentially over the past few years, its roots
began in some of healthcare’s greatest challenges. For the past five decades, healthcare technology has been
plagued by a lack of interoperability - getting data and people to communicate intelligently, quickly, and efficiently.
The more recent emergence of Generative AI (GenAI) addresses a major challenge in healthcare: the overwhelming
amount of unstructured data from diverse sources. It does this by extracting insights from complex datasets, such
as millions of whole genome sequences or human brain intricacies. As leading physician Eric Topol explains in his
published work, Deep Medicine, these advancements will ultimately enhance the health outcomes for individuals and
improve the efficiency of healthcare providers.2
01
The need for AI is clear: by integrating patients, providers, payers, and pharmaceutical companies through AI,
a unified front is created to overcome some of the sector’s greatest challenges. Providers leverage AI-powered
diagnostics and treatment planning for personalised patient care. Payers benefit from proactive risk management
and cost optimisation. Pharmaceutical companies streamline drug discovery and development, accelerating the
journey from research to market. This holistic integration ensures a collaborative approach, where information flows
seamlessly across the sector, fostering innovation, efficiency, and a collective commitment to advancing the quality
and accessibility of healthcare services.
Effectively deploying advanced healthcare analytics also fits in with the growing global focus on population health
and value-based care, and the widely accepted Quintuple Aim of healthcare improvement,3 which encompasses the
following goals:
Figure 2: The Quintuple Aim of healthcare improvement
Patient
Clinical
experience
outcomes
The Quintuple
Aim of
healthcare
improvement
Health
equity
Financial
sustainability
Clinician
wellbeing
02
The GCC: a region ripe for AI-enabled
technological innovation
The Gulf Cooperation Council (GCC) governments are aligning with the global digital revolution, shifting their focus to
e-commerce, smart cities, e-services, and digital health. These indicators and growing enthusiasm have paved the
way for AI to be integrated within businesses from all sectors, including a leapfrog adoption cycle for AI across the
healthcare sector.
To realise the full potential of AI, societies and organisations must commit to advanced healthcare analytics. This
commitment is already underway in the GCC region. In fact, our research highlights the GenAI market opportunity
across the GCC, estimating potential overall economic impact of $23.5 billion per year by 2030.4 The research also
indicated that GenAI-fueled improvements in efficiency and effectiveness would have the greatest impact in Saudi
Arabia and the UAE, with significant benefits also seen in Qatar, Kuwait, Oman, and Bahrain and healthcare was one
of the key industries to be affected.
GCC AI readiness indicators
High internet usage Global cybersecurity World digital Government AI-readiness
index competitiveness ranking index
GCC has one of the highest KSA and UAE ranked among UAE ranks 12th among 64 The UAE ranks 18th globally
internet usage rates globally, the top five countries in the economies in 2023 ranking in 2023, reflecting significant
with 99% of the population world by the International by the International Institute strides in AI implementation
online, compared to the Telecommunication Union for Management in public services. Saudi
world average of 63%.5 (ITU).6 Development (IMD).7 Arabia and Qatar rank 29th
and 34th respectively, out of
193 countries.8
Recent examples of AI healthcare initiatives in the region:
The UAE has been making great strides in leveraging AI in healthcare - from Abu Dhabi’s Department of Health
becoming the first entity to develop the Policy on Use of Artificial Intelligence in the Healthcare Sector,9 through to
the Ministry of Health and Prevention (MoHAP) establishing the health sector’s inaugural Centre of Excellence for AI
in 2023.10 Earlier this year, Thumbay Institute of AI in Healthcare hosted the region’s first international conference on
AI training and upskilling for healthcare professionals.11
The Dubai Health Authority’s EJADA AI system has also achieved remarkable results through a preemptive
disease prevention system, while the UAE healthcare firm M42, a joint venture between Mubadala Health and G42
Healthcare, has recently launched Med42, a clinical Large Language Model that provides high-quality answers to
medical questions. This follows the earlier release of their Arabic-enabled ‘Jais’ model, showcasing the country’s
commitment to advancing healthcare through AI.
Recently in Saudi, the Saudi Data and AI Authority established the AI Ethics Principles to integrate AI technologies,
regulate AI data, safeguard data privacy, promote responsible AI development, and minimise risk.12 And in 2022 the
Saudi Food and Drug Authority published guidance on the regulation of AI and ML-based medical devices, to ensure
the safety and reliability of these technologies.13
With governments laying the groundwork for AI governance, healthcare entities now have an urgent responsibility to
integrate these technologies within their businesses, to make sure they are on track with the global mobilisation as
they strive to meet enterprise challenges.
03
Stabilisation, value-based care, and regulation
drive AI direction across the GCC
In the current environment, healthcare demand is largely well-served by high-quality, specialised providers, and
market expansion is tapering as it approaches a steady state. As business value growth potential levels off for
healthcare providers across the GCC, the focus for players is reorienting towards differentiation and improving
efficiencies.
Stabilisation is also inevitable with the various government reforms in place aimed at expanding private health
insurance coverage. Abu Dhabi and Dubai have witnessed this with the rollout of mandatory insurance, resulting in
steady hospitalisation rates of around 10% following full implementation.14 Other GCC countries are expected to
follow a similar trend.
With the increased focus on value-based care — succinctly defined as the overall cost of improving outcomes that
are valued by patients and populations — the overarching message is evident: providers are incentivised for the
value of integrated care they provide. In Abu Dhabi, downward pressure on facility multipliers has been witnessed
from a reimbursement standpoint, thus impacting overall business top line growth. Similar to the case in Dubai,
the potential introduction of service-based packages in the region is also expected to exert negative pressure on
expected reimbursement rates. Providers are thus forced to deliver care more effectively by leveraging operational
efficiencies — an approach that can be facilitated through the use of AI. By applying AI algorithms, processes such
as resource allocation, including staff scheduling, room utilisation, and equipment maintenance, can be streamlined,
thus improving overall efficiency.
Growing regulatory requirements and maturing insurance markets have also increased the imperative for physicians
and other clinicians to accurately document patient information, thereby increasing the administrative burden and
limiting the time they have available to fulfil their true purpose: treating patients face to face. This is amplified by
the shortage of clinical staff in the GCC and the heavy reliance on external staff to fill the resource gap, with recent
findings showing close to 60% of physicians and nurses in KSA and Oman being expatriates (this figure rises to more
than 90% in Dubai alone, highlighting how expat-driven the workforce is).15 To mitigate this issue, GenAI offers the
solution for auto-generating necessary clinical documentation and relieving the administrative burden.
Given the challenge presented by governments moving forward with AI governance and the market pressures in the
region, how does the healthcare sector move forward with exploiting the opportunities inherent in AI technology? As
with many disruptive technology adoptions, it begins by making the business case.
04
Getting started in healthcare AI implementation
Implementing an AI programme can be an exciting yet daunting endeavour for leadership teams looking to harness
its transformative power. In healthcare today, AI offers unprecedented opportunities to drive innovation, enhance
decision-making and unlock new levels of efficiency and competitive advantage.
By structuring AI work into three phases, leadership can establish a clear plan from strategy through execution and
measurement. This structured approach aligns teams and maximises AI’s impact on business operations, customer
experiences and technology innovation. To drive this process, an AI Centre of Excellence (CoE) can be established,
which will serve as a stepping stone to capture, ideate and develop use cases, and build internal capacity in a
responsible and sustainable way.
Figure 3: AI implementation programme overview
Monitoring &
AI transformation blueprint
evaluation
Business
strategy
Business use
cases
Roadmap and
delivery
Voice of the
customer
Technology
capabilities
Technology strategy
Monitoring of data
Data governance
management practices &
policies
alignment with regulations
Data structure for Effectiveness of data
data management governance policies
Data quality, security Assessment of data
& compliance with quality & integrity to ensure
industry regulations accuracy & reliability
Data management & governance
Enterprise program management Enterprise architecture Agile delivery
Delivery excellence (CI/CD
Change management and training
pipeline, ML ops)
05
Phase 1: Ambition
Leadership starts by identifying appropriate business and technology team members. The Ambition phase sets the
overarching business, technology, and data strategy for the AI programme, taking an outside-in, customer/product-
centric approach. Key activities include defining AI vision and goals, aligning customer needs and market trends,
and establishing clear success criteria for measuring AI’s impact on customer experiences. This phase assesses
the readiness and investment requirements for AI adoption, and establishes governance structures and stakeholder
engagement mechanisms to ensure alignment and accountability throughout the AI journey.
Leadership teams should define quantifiable business goals using key performance indicators (KPIs) to demonstrate
AI’s impact, ranging from improving patient outcomes and operational efficiency to driving revenue growth and
ensuring regulatory compliance. By setting clear, measurable objectives, leadership can track progress, evaluate
efficacy, and demonstrate tangible value to stakeholders.
Figure 4: Example of AI goal setting and KPI’s
Enhanced patient
Operational efficiency Clinical decision support
outcomes
•Goal: Reduce readmission rates by x% •Goal: Reduce average patient wait times •Goal: Improve diagnostic accuracy rates
with a specified timeframe by X minutes through AI-powered by X% using AI-driven imaging analysis
•Goal: Improve patient satisfaction scores scheduling and resource allocation and decision support tools
by X points over next year •Goal: Increase utilisation rate of hospital •Goal: Decrease medication errors by X%
•Goal: Increase early detection rates for facilities by X% through predictive through AI-powered medication
chronic conditions by X% through maintenance and optimisation reconciliation/prescription assistance
AI-driven predictive analytics •Goal: Reduce admin overhead costs by •Goal: Increase adherence to
X% by automating repetitive tasks (e.g., evidence-based guidelines by X% with
billing, coding) using AI AI-driven clinical decision support
systems
Population health
Revenue growth Regulatory compliance
management
•Goal: Increase revenue per patient •Goal: Reduce incidence of preventable •Goal: Achieve compliance with regulatory
encounter by X% through AI enabled diseases by X% through AI-driven requirements (e.g., HIPAA, GDPR)
personalised treatment plans targeted population health analytics and targeted through AI-driven data
interventions interventions security/governance measures
•Goal: Expand service offerings and •Goal: Increase patient engagement and •Goal: Reduce audit risk/penalties related
market reach by X% through AI-enabled adherence to treatment plans by X% to compliance violations by X% through
telemedicine/remote patient monitoring using AI-powered patient education AI-enabled risk management and
solutions programs monitoring systems
•Goal: Achieve higher reimbursement •Goal: Improve community health
rates by X% through improved outcomes by X% through proactive
documentation accuracy & compliance health monitoring and intervention
facilitated by AI strategies enabled by AI
It is also important to thoroughly assess what data is available, including the different sources, formats and quality of
the data that will need to be leveraged. Identify any gaps or concerns that might exist, and develop a plan to source,
prepare and integrate the data required in a robust and responsible way.
06
Phase 2: AI transformation blueprint
The AI transformation blueprint phase builds a roadmap that balances value and risk, and flows from the business
strategy and KPIs of the Ambition phase into developing AI business use cases and priorities for engaging AI
technology capabilities. Leadership assesses where the greatest opportunities lie for optimising existing processes
within their current workflows. Leaders may also choose to prioritise certain initiatives based on unique needs and
priorities, such as customer expectations or specific products and services in the enterprise’s portfolio.
Potential business use cases
The figure below illustrates a selection of potential business use cases for AI in healthcare, highlighting a spectrum of
applications.
Figure 5: Risk-impact assessment for potential use cases
•Aid patients with tasks such as appointment scheduling, physician / service
Care journey automation availability, registration, benefits and eligibility checks, etc.
Remote patient monitoring •Monitoring (e.g., vitals monitoring) and engaging with the patients (e.g., medication
& engagement adherence, reminders) remotely and outside of traditional care delivery settings
•Auto-generated document that summarises some, or all of the patient's existing
Shared health summaries
conditions, clinical notes, discharge summaries, medications, etc.
Patient
engagement Patient education and •Provision of educational content and tools to patients to help manage their health
consumerism and allow them to be part of the healthcare decision-making process
•Engage with a primary care virtual assistant, providing an initial assessment and
Virtual clinician
recommendations without oversight from a clinician
•Automating clinical documentation (e.g., discharge summaries, clinical notes, referral
Clinical documentation letters, prescription, etc.) based on live interaction between patient and physician
during an episode of care
•AI algorithms that leverage vast amounts of medical database and patient data (e.g.,
Clinical diagnosis and
electronic health records, transcripts, natural human speech, written material,
personalised medicine
Clinical decision images, videos) for clinical diagnosis and generation of personalised treatment plans
support and •Assign diagnostic and procedural codes to patient diagnoses, treatments,
Clinical coding and
documentation procedures or equipment used, based on analysed clinical information from health
charge entry
records
Clinician performance •Feedback on clinician performance (e.g., adherence to clinical guidelines, billing
evaluation and training procedures) based on analysis of clinical encounter transcripts
Payment verification & •Verify payments automatically by matching with patient accounts and update patient
payment posting accounts automatically
•Analyse refunds, adjustments, write-offs and overpayments, identify any
Account reconciliation
discrepancies, and process financial corrections
Mid-office and
Insurance pre-approval & •Extract patient information, verify coverage and generate and submit
back-office
benefits authorisation pre-authorisation requests to insurance companies
administrative
Claim submission & denial •Review claims and denials by identifying any errors or missing information
functions
management (minimising risk of rejected claims) and streamline claim submission process
High Risky bets Priority wins
R me om nio tote ri np ga t aie nn dt C aa ur te o mjo au tr in oe ny veP ria fiy cm ate ion nt &
engagement payment posting
Clinical diagnosis Patient Shared Account reconciliation
& personalised education & health Insurance pre-approval &
medicine consumerism summaries benefits authorisation
Claim submission &
Virtual Clinical denial management
clinician documentation
Clinical coding
and charge entry
Clinician
performance
evaluation and
training
Deprioritised risks Incremental gains
Low
High Risk Low
of implementing AI to the use case
07
In developing and prioritising business use cases, leaders must define them in detail, including their objectives, target
user groups, data requirements, and expected outcomes. Enabling technology needs are concurrently assessed,
and a technology stack is selected or developed to effectively support AI use cases. The rapid advancement of AI
technologies can lead to significant changes in risk and feasibility within a few months, necessitating continuous
reassessment to ensure alignment with the latest developments and opportunities.
Partnership strategies
Partnership strategies also play a key to this analysis, covering both technological capability and the regulatory sides.
Technology partners can provide access to cutting-edge, off-the-shelf AI solutions and domain-specific accelerators.
Partnering with regulatory bodies and standards organisations can help mitigate legal and ethical risks associated
with AI implementation.
Workforce and skills
A thorough review of the existing workforce skill sets must take place to identify any gaps where upskilling is
required, or new roles to be created calling for fresh skills and experience entirely. Our latest CEO Survey found that
67% of Healthcare CEO’s believe that GenAI will require most of their workforce to develop new skills.16 Workforce
preparation is essential to successful execution.
Risk assessment
Leadership should also assess the potential for unintended consequences or disruptions to existing workflows and
revenue cycle operations. By conducting a thorough risk assessment and weighing these factors, business leaders
can make informed decisions about which use cases to prioritise and mitigate risks effectively. Risk also has to
be measured against value to determine priority. High-risk factors, such as external versus internal customers (i.e.
patient facing vs. employee facing), regulatory compliance, data security, and stakeholder buy-in, must be carefully
evaluated to mitigate potential challenges and ensure successful implementation.
Once potential use case areas have been identified, leaders can evaluate which elements can be served using
in-house technology capabilities versus which will require strategic partnerships. At this point teams can also
collaborate to define clear objectives, success criteria, and KPIs for each use case.
By following a structured approach to use case development, leadership teams can ensure that AI initiatives are
aligned with strategic objectives and positioned for success in driving tangible business value and innovation. This
phase also involves building or enhancing data infrastructure, sourcing and preparing training data, and developing
or acquiring AI models and algorithms. Finally, the AI Transformation phase outlines a comprehensive implementation
plan, including timelines, resource allocation, risk management strategies, and milestones for measuring progress
and achieving key deliverables.
For payers and pharma players, a similar approach can be adopted — prioritising individual use cases based on
their value addition as well as their risk of implementation. Rather than having customer centres answering benefits-
related questions, AI can automatically generate Explanation of Benefits tables that outline services and cost-
sharing amounts, streamlining the workflow and reducing admin work. Payers can also facilitate prior authorisation
processes through automated intake, validation and triaging of PA request documentation. This will help reduce
waiting times for approvals, allowing patients quicker access to necessary treatments. Also, given the significant
number of grievances and complaints insurers receive, automating the response process can expedite the workflow,
as well as help in highlighting potential risks to minimise litigation and maintain brand reputation.
Pharma players can also leverage AI across different processes. By predicting and visualising molecular
conformations, new molecule generation is enabled with data-informed models. AI can also enhance search and
analytics capabilities through pretrained LLMs on available drug safety data and real-world cases.
08
Phase 3: Monitoring and evaluation
This phase focuses on evaluating the outcomes of AI initiatives and measuring their impact against the overarching
strategy. This involves defining KPIs and metrics to assess the effectiveness, efficiency, and ROI of AI programmes.
These metrics may include customer satisfaction scores, revenue growth, cost savings, operational efficiency
gains, and other relevant indicators. It also includes conducting regular assessments to track progress, identify
improvement areas, and adjust strategies and tactics as needed. Additionally, this phase aims to communicate
results and insights to key stakeholders, including senior leadership, business units, and external partners to
recognise successes, address challenges, and reinforce the value of AI in transforming customer experiences and
driving business outcomes.
09
With great power comes great responsibility:
building an enterprise governance framework for
responsible AI
For business leaders, there are plenty of reasons to be excited about AI, starting with its power and ease of use. But,
as with any emerging technology, there are also potential new risks. Some of these hazards may come from your
company’s use, others from malicious actors.
To manage both kinds of risks and harness AI’s power to drive sustained outcomes and build trust, you will need
responsible AI - a methodology designed to enable AI’s trusted, ethical use. It has always been important, but it has
become crucial in the dawning era of generative AI.
Healthcare leaders will start with a responsible AI end-to-end enterprise governance framework which focuses on the
risks and controls that will guide the AI journey. The elements of such a framework are shown below:
Figure 6: Responsible AI framework
The responsible use of AI necessitates an integrated approach that encompasses technical, ethical, social and legal considerations, all
of which are essential to maximise the advantages of AI use while substantially reducing any associated risks.
Ethical principles Human oversight Data integrity
AI technologies must be developed and AI technologies must be overseen by Datasets employed in training AI models
adopted in accordance with core ethical humans to guarantee their adherence to must be of high quality and diverse to
standards, including transparency, ethical and legal standards and to prevent bias and promote fairness.
fairness, privacy, and accountability. mitigate unforeseen outcomes. Developers must address potential biases
that could arise from both the algorithms
and their application.
Regulation &
Privacy & security Education & awareness
governance
AI systems must be designed, prioritising Individuals and organisations must be Governments and industries must
privacy and security, safeguarding well-informed about the capabilities & collaborate to establish robust regulations
against unauthorised access and misuse limitations of AI, as well as its potential and standards that promote the ethical
of personal data. effects on society and the environment. development and application of AI
systems.
10
What’s next?
AI is no longer a technology of the future: it is very much here and in use today. It holds the promise of empowering
healthcare leaders to overcome some of the sector’s most historically daunting challenges—detecting diseases at
their earliest stages, streamlining patient care, reducing medical errors, decreasing clinician burnout, lowering system
costs, and improving outcomes that matter to the populations it serves.
It is not a question of whether healthcare will implement AI, but of how the implementation will proceed and continue
to evolve. This raises a host of other questions for healthcare players in the GCC:
01 Where should healthcare players place their bets?
Should players take immediate action and invest in high-cost technologies, following global counterparts and
02
best practices, even if they may pose some form of risk?
Or should players progress in small steps focused on low risk and low cost solutions (quick wins) to check
03
the box for AI integration?
04 Do organisations have the minimum level of data quality, completeness, and maturity to enable the use of AI?
05 How will healthcare providers gain the trust of patients in clinical diagnosis and data privacy?
How smoothly can they transform existing healthcare systems and workflows into digitised, AI-enabled
06
processes?
The answers lie in a structured approach like the one we have detailed, driven by establishing a robust AI CoE from
strategy to execution that can prioritise the right use cases while empowering the workforce with necessary skills and
tools to augment their knowledge and capabilities.
The faster leadership teams can embrace AI and start their journey, the faster they can reap the many advancements
that are driving costs down while improving quality of care. Achieving the delicate balance between innovation and
risk mitigation is key to unlocking the full potential of this powerful and transformative technology
AI stands poised to reimagine healthcare. Where will your imagination lead your enterprise?
11
References
1. https://nuscriptmed.com/ai-achieves-high-diagnostic-accuracy-in-virtual-primary-care-setting/
2. https://drerictopol.com/portfolio/deep-medicine/#:~:text=In%20Deep%20Medicine%2C%20leading%20physician,medicine%20and%20
reducing%20human%20mortality.
3. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8608191/
4. PwC. Reshaping the Middle East: A CEO’s playbook to win the $23.5 billion Generative AI opportunity. https://www.strategyand.pwc.com/m1/en/
strategic-foresight/sector-strategies/technology/reshaping.html
5. World Bank/ (2021). Internet users (% of population). World Bank Data. https://data.worldbank.org/indicator/IT.NET.USER.ZS
6. International Telecommunication Union (ITU) Global Cybersecurity Index. https://www.itu.int/en/ITU-D/Cybersecurity/Pages/global-cybersecurity-
index.aspx
7. IMD World Digital Competitiveness Ranking 2023 Report. https://worldcompetitiveness.imd.org/countryprofile/AE/digital
8. Oxford Insights. (2023). Government AI Readiness Index 2023. https://oxfordinsights.com/ai-readiness/ai-readiness-index/
9. https://www.doh.gov.ae/en/news/the-department-of-health-becomes-region
10. Ministry of Health and Prevention. (2023). MoHAP Launches Health Sector’s First National Centre of Excellence for AI. https://mohap.gov.ae/en/
media-center/news/19/10/2023/mohap-launches-health-sectors-first-national-centre-of-excellence-for-ai
11. https://gulfnews.com/uae/health/uae-thumbay-institute-of-ai-in-healthcare-to-host-regions-first-international-conference-driving-ai-training-for-
health-professionals-1.102298294
12. https://sdaia.gov.sa/en/SDAIA/about/Documents/ai-principles.pdf
13. https://beta.sfda.gov.sa/en/regulations/87661
14. Dubai Health Authority. Annual Health Statistic Book. https://www.dha.gov.ae/en/open-data Department of Health Abu Dhabi. https://www.doh.gov.
ae/en/resources/opendata
15. Frost Sullivan/Mashreq. 2020 Annual Overview of Healthcare in the GCC Growth opportunities for 2021 and beyond. https://www.mashreq.com/-/
jssmedia/pdfs/corporate/healthcare/2020-Annual-Overview-of-Healthcare-in-the-GCC.ashx
16. https://www.pwc.com/m1/en/ceo-survey/27th-ceo-survey-middle-east-findings-2024.html
Contact us
Amar Patel Timur Korshlow
Partner, Healthcare Sector Lead Partner, Advanced Analytics,
for Deals, PwC Middle East
PwC Middle East Email: [email protected]
Email: [email protected]
At PwC, our purpose is to build trust in society and solve important problems. We’re a network of firms in 151
countries with nearly 364,000 people who are committed to delivering quality in assurance, advisory and tax
services. Find out more and tell us what matters to you by visiting us at www.pwc.com.
Established in the Middle East for over 40 years, PwC Middle East has 30 offices across 12 countries in the region
with around 11,000 people. (www.pwc.com/me).
PwC refers to the PwC network and/or one or more of its member firms, each of which is a separate legal entity.
Please see www.pwc.com/structure for further details.
© 2024 PwC. All rights reserved |
221 | capgemini | Capgemini-Custom-Private-Gen-AI-Assistants-POV-1.pdf | Enterprise-specific
AI agents keep the
Gen AI promise
Custom enterprise-specific Gen AI makes it possible to reach business outcomes faster with
more insightful and original output. It can be the most effective way for a company to
maximize the value of its own structured and unstructured data.
The awarding of the 2024 Nobel Science Prizes in public data and adding processing power is not
physics and chemistry for artificial intelligence- enough to deliver domain-specific improvements
related discoveries confirmed its significance for and the expansion of capabilities needed to justify
investment. They lack the specialized, high-quality
modern life. According to Nobel laureate Geoffrey
data needed to operate with expertise in a company’s
Hinton, one aspect of AI related to generative AI -
specialism. This data is locked behind corporate
machine learning - will “have a huge influence. It will
firewalls, or otherwise unavailable for generalist
be comparable with the Industrial Revolution, but
LLM training. As a result, even top-tier chatbots like
instead of exceeding people in physical strength, it’s
ChatGPT or Gemini, which are based on LLMs, can
going to exceed people in intellectual ability.”1 produce flawed answers, or “hallucinations”.
Nevertheless, generative AI has not been in popular Accordingly, there are sound reasons to view the
use for long, with some well-publicized instances of generative AI phenomenon with some detachment,
unintended generative AI mistakes. Consequently, but its fundamental benefits to enterprise have
organizations are cautiously optimistic about not yet been widely or fully tapped in the economy.
Enterprise-specific, i.e., custom private Gen AI, also
the potential benefits of integrating it into their
known as agentic AI, can be improved to achieve
strategies, processes, and business models.2
higher accuracy using transparent, accessible
While large learning models (LLMs) hold potential sources and verifiable results. This form has an
to re-shape business, they run into a performance enhanced ability to focus on specific areas and
ceiling when dealing with specialist areas that they produce context-sensitive results through training on
have not been trained on. Users are cautioned not refined data from specific business domains. Agentic
to trust their responses outright, especially for implementation allows expanded use cases in more
questions involving unique, organization-specific complex and sophisticated scenarios.
information. Simply training them with more
1Babbage from The Economist, The 2024 Nobel prizes: a triumph for AI, Oct. 9, 2024
2Capgemini Research Institute, Harnessing the value of generative AI, p. 27
2 Enterprise-specific AI agents keep the Gen AI promise © Capgemini 2024. All rights reserved
Specialize for better outcomes
General purpose LLMs can be effective tools to dormant, siloed and forgotten. Meanwhile, the
increase productivity, but their lack of specificity company reinvents the wheel, wastes time and
reduces the relevance and quality of their output. budget, and misses opportunities.
They tend to produce generic results that do
To achieve generative AI’s maximum benefit to
not deliver the full potential achievable from
enterprise productivity and creativity, a custom-
an organization’s own intellectual property and
designed language model can be trained on
data, which should lead to actionable insights and
an organization’s own dataset in its entirety
productivity gains.
using structured and unstructured data. This
Innovation that sets a company apart is most likely form of enterprise AI agent has a higher level of
to come from an organization’s own proprietary reasoning, producing more refined responses on
insights. According to business intelligence provider, specialist subjects.
Gartner, 68% of enterprises struggle to integrate AI
The ideal generative AI model boosts productivity
into workflows that rely heavily on internal data.3
across the entire workforce. For instance, it can
Custom private Gen AI makes it possible to respond streamline the process a business development
faster to business opportunities, with more team follows to respond to RFPs, speed up selection
insightful and original content. It can be the most of qualified talent for HR, or enhance the quality of
effective way for a company to maximize the value customer interactions in a contact center.
of its IP in various forms. Without the mass data
interpretive power of Gen AI, much of this will lie
3 Gartner, RAG in enterprise data strategy
3 Enterprise-specific AI agents keep the Gen AI promise © Capgemini 2024. All rights reserved
Safeguarding Gen AI data
Along with making Gen AI a more effective business resource, there are
significant ethical and compliance reasons to prefer a custom private Gen
AI system. Its closed loop structure allows tighter governance, security
protocols, and continuous monitoring to keep it in line with a company’s
own ethical and security guidelines, and in compliance with data privacy and
sovereignty, where applicable.
Knowing the methods used for data training are key to managing legal risks.
These risks are real, with credible copyright and trademark infringement
cases already underway.4 Proprietary LLMs are safer due to their closely
controlled, transparently-sourced training data.
By striking a balance between robust security controls and proactive
safeguards for ethical AI performance, organizations can protect their critical
assets while cultivating trust and strengthening operational resilience.
Data security has climbed higher up the board agenda due to tighter
operational resilience regulation. A custom private Gen AI assistant allows
companies to impose strict data security measures, retain full ownership
of their intellectual property, and gain clearer insights into potential
vulnerabilities.
4Reuters, AI companies lose bid to dismiss parts of visual artists’ copyright case, August 13, 2024
4 Enterprise-specific AI agents keep the Gen AI promise © Capgemini 2024. All rights reserved
The data ingestion
framework - the key
to delivering full
Gen AI value
The decisive component of a custom private LLM for a query, making knowledge management and on a large, known dataset, RAG is an information
is its Data Ingestion Framework (DIF). The DIF is domain-specific assistance more productive. Queries retrieval system that picks out specific, relevant data
crucial for analysis of and attributing meaning to are thus answered with data from the correct from available databases and documentation. An LLM
documents and other materials in a dataset and documents and the most applicable sections, a critical is static, while the RAG model is dynamic. The LLM’s
thus its productiveness. It extracts, organizes, and capability for domain-specific use cases and effective strength is in specific, well-defined scenarios while
prepares data for future retrieval. It applies metadata knowledge management. RAG is capable of broad matching with up-to-date
for ontological purposes, ensuring that the model information, but with less precision, making guardrails
Supporting a trained LLM with customized retrieval-
can access the right information at the right time. The critically important. 5
augmented generation (RAG) improves the accuracy
aim is for better targeting of the required information
of responses. Where the LLM is an AI system trained
5Towards Data Science, The Practical Limitations and Advantages of RAG, April 15, 2024
5 Enterprise-specific AI agents keep the Gen AI promise © Capgemini 2024. All rights reserved
Multi-stage guardrails
RAG systems rely on guardrails at multiple stages Intermediate guardrails then act as checkpoints
to ensure quality and mitigate risks. Guardrails during the retrieval process, validating selected
also keep Gen AI systems in alignment with data for alignment with policies and the context of
organizational values by eliminating harmful and the query.
biased outputs.
A final output filter evaluates the response once the
The first layer of control begins with input information is retrieved, but before it is delivered
filtering, which examines the specific content of to the user or downstream business processes.
user prompts for compliance and risk factors. For This filter ensures that the response adheres
instance, a query requesting sensitive strategies to key requirements, such as confidentiality,
or methodologies, e.g., how to redesign internal appropriateness, e.g., absence of toxic or harmful
processes, could inadvertently expose confidential content, and compliance with company policies
information. This is particularly important in and regulatory standards. It also confirms that the
regulated sectors like banking, healthcare, and output meets user expectations of accuracy and
defense, where strict compliance standards prohibit relevance, serving as a final quality control step.
sharing sensitive data with external AI systems
or vendors.
6 Enterprise-specific AI agents keep the Gen AI promise © Capgemini 2024. All rights reserved
Cost controls
The cost of running LLMs can spiral if left unchecked summaries or context-specific outputs rather than
due to their reliance on intensive computing processing entire datasets. This hybrid approach can
resources and token-based pricing models. This is cut compute expenses by 25–35%.7
directly tied to the volume of prompts processed
Cost control measures in custom LLMs integrated
and the length of generated responses. This
with RAG systems with token usage monitoring
makes extensive or complex queries exponentially
enable enterprises to harness the power of
expensive. 65% of enterprises in 2024 reported
generative AI without financial surprises. By
difficulty in forecasting LLM usage costs, with some
focusing on task-specific applications and
companies experiencing over 30% unanticipated
streamlined workflows, organizations can better
budget overruns due to insufficient monitoring and
forecast expenses and align their AI investment with
control.6
business priorities.
A major benefit of using RAG with custom LLMs
is the ability to shift computational burden to
retrieval mechanisms, which are cheaper to run.
RAG frameworks retrieve enterprise data in real-
time, allowing LLMs to focus solely on generating
6Gartner, Enterprise AI cost control report
7Forrester, Build Efficient And Robust GenAI Apps With Prompt Engineering And
Advanced LLM App Architectures
7 Enterprise-specific AI agents keep the Gen AI promise © Capgemini 2024. All rights reserved
Examples of Gen AI solutions using custom private generative AI agents tailored to
domain and business function
Banking and Healthcare Research and
Finance Sales
insurance providers development
Automating data Credit memo Drafting longitudinal Identifying cross- Synthetic data
analysis across generation patient summaries selling opportunities generation
financial reporting, by using a Gen AI
marketing, Covenant monitoring Clinical trial package agent as a portfolio Advanced drug
operations, supply generation sales executive discovery and
chains Suspicious Activity therapeutics
Report and other Personalized medicine Generating first-shot
Proactive data-driven financial crime report RFP/RFQ responses Novel protein design
recommendations filings Research assistant
to reduce manual agent Customer intent/ Clinical trial and
analysis time Pitchbook generation insights agent research facilitation
Claims processing, e.g,
legal case package
creation
Underwriting
assistant
8 Enterprise-specific AI agents keep the Gen AI promise © Capgemini 2024. All rights reserved
Use case -
Streamlining an
insurance underwriting
workflow
A market-leading US insurance underwriter saw the opportunity to use
generative AI to improve underwriting workflows. Their implementation
of a custom private generative AI model analyzed historical claims, policy
data, and external risk factors to draft underwriting recommendations. It
also generated detailed explanations for its recommendations, enabling
underwriters to then make informed decisions faster.
This reduced underwriting case turnaround times by 40%, improved risk
assessment accuracy, and supported the creation of personalized policies.
This ultimately enhanced customer satisfaction and profitability.
9 Enterprise-specific AI agents keep the Gen AI promise © Capgemini 2024. All rights reserved
Our experience
Capgemini has proven expertise in activating data and AI to its full potential
for data-driven businesses. Building on our partnerships with hyperscalers
and the AI innovation ecosystem, we help our clients deliver value and
generate competitive advantage with a portfolio of tailored, scalable Gen
AI solutions.
We help maximize the value of your enterprise data by creating accurate,
context-aware Gen AI assistants. They empower your employees and
customers using your own data for specific business needs, while
safeguarding data. These agents are typically used to streamline customer
service, marketing, contract management, content generation, financial
analysis, and more, at controlled cost of use.
10 Enterprise-specific AI agents keep the Gen AI promise © Capgemini 2024. All rights reserved
Experts to contact
Pinaki Bhagat
AI & Generative AI Solution Leader,
Financial Services
Ashvin Parmar
Vice President, Generative AI CoE Leader,
Financial Services
11 Enterprise-specific AI agents keep the Gen AI promise © Capgemini 2024. All rights reserved
About Capgemini
Capgemini is a global business and technology transformation partner, helping organizations to
accelerate their dual transition to a digital and sustainable world, while creating tangible impact for
enterprises and society. It is a responsible and diverse group of 340,000 team members in more than
50 countries. With its strong over 55-year heritage, Capgemini is trusted by its clients to unlock the
value of technology to address the entire breadth of their business needs. It delivers end-to-end
services and solutions leveraging strengths from strategy and design to engineering, all fueled by
its market leading capabilities in AI, cloud and data, combined with its deep industry expertise and
partner ecosystem. The Group reported 2023 global revenues of €22.5 billion
Get the future you want | www.capgemini.com
© 2024 Capgemini. All rights reserved. |
224 | capgemini | 2024_07_10_Capgemini-News-Alert_Generative-AI-for-Software-Engineering-1.pdf | Press contact:
Florence Lievre
Tel.: +33 1 47 54 50 71
Email: [email protected]
Generative AI is set to be adopted by 85% of the software workforce over the
next two years
Three in five organizations see innovative work as the biggest benefit of generative AI use in software
engineering; software professionals say generative AI will boost their comms with business teams
Paris, July 10, 2024 – Generative AI (Gen AI) is expected to play a key role in augmenting the
software workforce, assisting in more than 25% of software design, development, and testing
work in the next two years. According to the Capgemini Research Institute’s latest report
“Turbocharging software with generative AI: How organizations can realize the full potential of
generative AI for software engineering”, a large majority (80%) of software professionals
believe that, by automating simpler repetitive tasks, Gen AI tools and solutions will significantly
transform their function, freeing up time for them to focus on higher-value-adding tasks. More
than three quarters of software professionals are confident that generative AI has the potential
to boost collaboration with non-technical business teams.
While the generative AI adoption for software engineering is still in its early stages, with 9 in 10 organizations
yet to scale, the report found that organizations with active Gen AI initiatives are already reaping multiple
benefits from its adoption – fostering innovation coming first place (61% of organizations surveyed) followed
by improving software quality (49%). They saw also an improvement of between 7 to 18% (on average) in
the productivity1 of their software engineering functions. For certain specialized tasks, time saving was as
high as 35%.
Organizations surveyed highlighted that they plan to leverage the additional time freed up by generative AI
for innovative work such as developing new software features (50%) and upskilling (47%); while reducing
headcount being the least-adopted route (just 4% of responding organizations). New roles, such as
generative AI developer, prompt writers or generative AI architect are also emerging.
Improved collaboration between tech and business teams
From better communication to explaining what the code is doing in natural language, Gen AI makes the
connection between software engineers and other business teams more effective. 78% of software
professionals are optimistic about Gen AI’s potential to enhance collaboration.
Augmented software workforce and employee satisfaction
According to the survey, generative AI tools are used today by 46% of software engineers for assisting them
on tasks. Almost three quarters agree that generative AI's potential extends beyond writing code. While
coding assistance is the leading use case, generative AI also has applications in other software development
lifecycle activities, such as code modernization or user experience (UX) design.
1 Overall improvement in the productivity of the individual from all types of tasks accelerated by generative AI. Productivity advantage
increasing with organization size.
Capgemini News Alert
Both senior and junior software professionals also report higher levels of satisfaction from using Gen AI
(respectively 69% and 55%). They see generative AI as a strong enabler and motivator.
However, according to the report 63% of software professionals declare using unauthorized Gen AI tools to
assist them in tasks. This rapid take-up, without proper governance and oversight in place, exposes
organizations to functional, security, and legal risks like hallucinated code, code leakage, and IP issues.
Pierre-Yves Glever, Head of Global Cloud & Custom Applications at Capgemini, said: “Generative AI has
emerged as a powerful technology to assist software engineers, rapidly gaining adoption. Its impact on
coding efficiency and quality is measurable and proven, yet it holds promise for other software activities.
However, we must remember that the true value will emerge from a holistic software engineering approach,
beyond deploying a single ‘new’ tool. This involves addressing business needs with robust and relevant
design, establishing comprehensive developer workspaces and assistants, implementing quality and security
gates, and setting up effective software teams. The focus should be on what genuinely generates value.
Exciting times lie ahead!”
To access the full report: Link
Methodology:
The Capgemini Research Institute surveyed 1,098 senior executives (director and above) and 1,092 software
professionals (architects, developers, testers, and project managers, among others). 20 in-depth interviews
were conducted with leaders from the industry, partners, and startups, along with several software
professionals.
About Capgemini
Capgemini is a global business and technology transformation partner, helping organizations to accelerate
their dual transition to a digital and sustainable world, while creating tangible impact for enterprises and
society. It is a responsible and diverse group of 340,000 team members in more than 50 countries. With its
strong over 55-year heritage, Capgemini is trusted by its clients to unlock the value of technology to address
the entire breadth of their business needs. It delivers end-to-end services and solutions leveraging strengths
from strategy and design to engineering, all fueled by its market leading capabilities in AI, cloud and data,
combined with its deep industry expertise and partner ecosystem. The Group reported 2023 global revenues
of €22.5 billion.
Get The Future You Want | www.capgemini.com
About the Capgemini Research Institute
The Capgemini Research Institute is Capgemini’s in-house think-tank on all things digital. The Institute
publishes research on the impact of digital technologies on large traditional businesses. The team draws on
the worldwide network of Capgemini experts and works closely with academic and technology partners. The
Institute has dedicated research centers in India, Singapore, the United Kingdom and the United States. It
was recently ranked #1 in the world for the quality of its research by independent analysts.
Visit us at https://www.capgemini.com/researchinstitute/
Capgemini News Alert |
226 | capgemini | 22Oct2024-Conversations-for-tomorrow_Edition_9_Report-V1.pdf | Gener(AI)ting
the future
Quarterly review
N°9 — 2024
Gener(AI)ting the future
Generative AI today is at the threshold of an outburst of creative exuberance. The cover
for this edition of Conversations for Tomorrow represents the myriad possibilities that
arise at the intersection of light and shadow – not too dissimilar to the possibilities that
generative AI creates. At the same time, it also highlights fleeting patterns and dark
spots which one must bear in mind as they move forward. The content and design of
this issue reflect the opportunities, the challenges, and the risks that generative AI is
now throwing up in front of organizations.
2
Capgemini Research Institute Gener(AI)ting the future
Foreword At Capgemini, we But adopters have also had to acknowledge
help organizations AI’s significant carbon footprint. Over
prepare for tomorrow one-third of organizations in our research
by distilling the are already tracking their Gen AI carbon
unique insights and perspectives of leaders emissions.
from global business, academia, the startup
community, and wider society. Organizations globally are rapidly
embedding Gen AI across functions, with
Gener(AI)ting the Future a ripple effect for wider society. In this
edition of Conversations for Tomorrow we
In Conversations for Tomorrow, the focus on this AI-generated future.
Capgemini Research Institute identifies
the strategic imperatives for the future of We would like to thank all the leaders and
business and the society it serves. In this experts who have enriched this edition of
ninth edition of the journal, among other the journal with their insights. By sharing
areas, we explore: the perspectives of such a diverse range of
accomplished individuals, we aim to present
• The rapid rise of generative AI (Gen AI) a comprehensive overview of Gen AI and its
contribution to generating a new future.
• The rate at which organizations across
industries are adopting the technology
• The CEO of one of the hottest AI startups
• The use cases that it enables
• One of the most-respected AI scientists
globally, who is also a board member at
• Its impact on sustainability journeys
Amazon
• How it is likely to change work and the
• A former member of the European
workforce
Parliament who played a key role in the
• How and why it needs to be regulated EU AI Act
Our annual research into the state of Gen • A leading professor at Stanford
AI shows that organizations are embracing
• Senior executives from Adobe,
generative AI, this is reflected by an uptick in
Salesforce, OECD, Telefónica, and Itaú
investment levels. The vast majority (80%) of
Unibanco
organizations in our survey increased their
investment in 2023; 20% maintained their
• Capgemini’s own subject-matter experts
investment levels; and no organization has
decreased its investment in Gen AI from last Pulling together such a wide range of views
year. was an extremely instructive exercise for
us. We hope you enjoy reading this edition
In the past year, every organizational as much as we enjoyed putting it together
domain, from sales and marketing to IT, for you.
operations, R&D, finance, and logistics, has
seen an increase in the rate of adoption.
Early adopters are seeing benefits from
improved operational efficiency to
enhanced customer experience. Moreover,
generative AI adoption among employees
is robust in most organizations, with the
majority allowing its use.
33
Gener(AI)ting the future Capgemini Research Institute
Contents
4
Capgemini Research Institute Gener(AI)ting the future
PP..1144
THE CEO CORNER
P.16 P.16
Arthur Mensch Aiman Ezzat
CEO CEO
Mistral AI Capgemini
5
Gener(AI)ting the future Capgemini Research Institute
PP..2266
EXECUTIVE CONVERSATIONS WITH…
P.28 P.64
Erik Brynjolfsson Clara Shih
Professor, Stanford CEO, Salesforce AI
P.36 P.76
Audrey Plonk Andrew Ng
Deputy Director, Directorate CEO, LandingAI
for Science, Technology and
Innovation (STI), OECD
P.44 P.86
Dragoş Tudorache Chema Alonso
Former MEP, EU AI Act Chief Digital Officer,
co-rapporteur, European Telefónica
Parliament
P.54 P.94
Scott Belsky Ricardo Guerra
Chief Strategy Officer, Chief Information Officer,
Adobe Itaú Unibanco
Contents
66
Capgemini Research Institute Gener(AI)ting the future
PP..110044
PERSPECTIVES FROM CAPGEMINI
GENERATIVE AI FOR
MANAGEMENT
P.106
PP.. 113300
Elisa Farri
and
Gabriele Rosani INSIGHTS
from The Management Lab
by Capgemini Invent FROM THE CAPGEMINI
RESEARCH INSTITUTE
P.131
Gen AI in enterprise
GENERATIVE AI: THE The rise of generative AI investments in
ART OF THE POSSIBLE organizations
P.116 P.138
Robert Engels Gen AI in software engineering
Head of Generative AI Lab, How software engineering is being
Capgemini shaped by generative AI
OPERATIONAL AI IS
CHANGING HOW WE LOOK
AT DATA
P.124
Anne Laure Thibaud
Executive Vice President, Data,
AI & Analytics Group Offer
Leader, Capgemini
and
Steve Jones
Executive Vice President, Data
Driven Business & Gen AI,
Capgemini
77
Gener(AI)ting the future Capgemini Research Institute
Executive Summary
Executive
Summary
Generative AI (Gen AI) is making rapid Andrew Ng, CEO, LandingAI, emphasizes the
inroads into organizational structures, boost to productivity: “For many jobs, AI will
transforming them rapidly from the inside only automate or augment 20-30% of tasks.
out. As businesses across industries begin So, there's a huge productivity boost, but
to implement Gen AI, several key themes people are still required for the remaining
emerge that highlight its potential to impact 70% of the role.”
organizations, workforce, and society.
Gen AI has the potential
to reimagine the future
We are most creatively
of the workforce
confident when we
Gen AI has potential to unlock human are five years old.
creativity, allowing employees to focus on
We lose our creative
complex strategic tasks. Clara Shih, CEO,
Salesforce AI, comments: “AI will allow confidence as we
workers to move away from repetitive tasks
get older because
to focus on doing what humans do best,
which is building relationships, unlocking of the skills gap,
creativity, making connections, and
exposure to criticism,
addressing higher-order problems.”
and just the lack of
Scott Belsky, Chief Strategy Officer and
access to creative
Executive Vice President, Design and
Emerging Products, Adobe, adds: “We are tools. Generative
most creatively confident when we are five
AI is fundamentally
years old. We lose our creative confidence
as we get older because of the skills gap, changing this.”
exposure to criticism, and just the lack of
access to creative tools. Generative AI is
Scott Belsky
fundamentally changing this.” Chief Strategy Officer, Adobe
8
Capgemini Research Institute Gener(AI)ting the future
Executive Summary
Whenever a
traditional activity
gets replaced or
augmented with
one based on bits,
it usually brings
significant energy
and environmental
benefits.”
Erik Brynjolfsson
Stanford
This shift is fostering a culture of continuous In the article, Generative AI for
learning and adaptability. Erik Brynjolfsson, Management, Elisa Farri, Vice-President
Professor at the Stanford Institute for at Capgemini Invent, and Gabriele Rosani,
Human-Centered AI and Director of the Director at Capgemini Invent, comment:
Stanford Digital Economy Lab, discusses the “AI's capability to collaborate on a cognitive
importance of workforce skill enhancement: and emotional level, offering insights and
“AI requires significant changes in the contributing to complex decision-making
economy to create full impact, particularly processes, is an area that many managers
in terms of organization and skills of the have yet to fully realize or integrate into
workforce. Identifying which skills are their strategic thinking.”
important, followed by self-learning and
training programs, are required to prepare They add: “Executives need to cultivate the
the workforce. Secondly, businesses will ability to adopt the co-thinking mindset.
need to restructure and adapt to capitalize Whether in individual tasks or team
on new technologies.” endeavors, mastering this shift will become
a vital competitive advantage.”
Gen AI also has a significant impact
on managers and leadership.
9
Gener(AI)ting the future Capgemini Research Institute
Executive Summary
In Gen AI deployment, Sharing Capgemini perspectives, Anne Laure
Thibaud, Executive Vice President, Data
ethical considerations
AI and Analytics Group Offer Leader, and
are paramount Steve Jones, Executive Vice President, Data-
driven Business and Gen AI, remark: “The
assumption is that Gen AI cannot be trusted
Gen AI has emerged as a transformative
in the same way as a human employee and
innovation. However, its potential
given the opportunity, will act outside its
for misuse emphasizes the need for
boundaries. Organizations are compelled to
organizations to uphold strong ethical
build all the information about their culture,
standards.
mission, and guardrails into the AI they use
to retain control of it.”
Aiman Ezzat, CEO, Capgemini, stresses
the importance of safe use of Gen
AI: “Organizations should establish
employee guidelines for safe use of Gen
AI and validating outputs to eliminate
bias.”
Arthur Mensch, CEO, Mistral AI,
underscores the importance of managing
AI-driven products within ethical
boundaries: “When an organization is
making an AI-driven product, it must
consider the decisions and outputs the
system will make. These decisions and
these outputs should be constrained to
respect the company's role.”
Audrey Plonk, Deputy Director of
the OECD Directorate for Science,
Technology, and Innovation, is
responsible for the OECD’s digital
policy portfolio. She elaborates on data
concerns: “Data privacy considerations
are a key aspect that organizations and
individuals are exploring extensively.
There is a lot of work to be done to
improve transparency and determine
which sources of data should be used to
train AI models. It is essential to put the
appropriate safeguards in place, including
data protection considerations.”
10
Capgemini Research Institute Gener(AI)ting the future
Executive Summary
The path to responsible Ricardo Guerra, CIO, Itaú Unibanco,
highlights the responsibility of
AI
organizations to use AI ethically:
“Organizations have to take a lot of the
To secure the future of AI, comprehensive responsibility for use and governance of AI
safeguards and collaborative efforts are and other technologies. But governments
essential. Unified actions from policymakers must still stay informed and try to
and organizations are crucial to ensure the implement supportive regulation.”
responsible use of Gen AI.
Steering towards a
Dragos Tudorache, former member of the
sustainable future with
European Parliament and Rapporteur on the
EU AI act, says: “Most companies working Gen AI
with AI already had general principles or
codes of conduct or self-regulation in place.
Gen AI will play a pivotal role in addressing
There were guidelines outlined by UNESCO,
climate change. Additionally, businesses
OECD, and even by the European Parliament.
must adopt sustainable practices that align
But we realized these measures were
with environmental goals.
insufficient to mitigate the very real risks,
such as discrimination bias.”
On climate engineering, Andrew Ng
elaborates: “Given the world's collective
inability to move CO2 emissions in the way
we know it needs to, I think it is past time
to take climate engineering more seriously.
Organizations should
I think AI, specifically large AI foundation
establish employee models of climate, have a large role to play
in that.”
guidelines for safe
use of Gen AI and Ricardo Guerra talks about sustainable
data centers: “We're learning when to
validating outputs to
use different solutions and emphasize
eliminate bias.” investing in sustainable data centers and
green technologies. We're also closely
Aiman Ezzat monitoring the market, prioritizing
CEO, Capgemini providers that offer green solutions.”
Erik Brynjolfsson says: “Whenever a
On the need to draft the EU AI Act, he adds:
traditional activity gets replaced or
“We needed to put stronger safeguards in
augmented with one based on bits, it
place that command respect and, ultimately,
usually brings significant energy and
help society to trust in the interaction with
environmental benefits.”
this technology, hence the decision to
formulate the policy.”
11
Gener(AI)ting the future Capgemini Research Institute
Executive Summary
The rise of open-source allow everyone to scrutinize the model for
potential sources of bias, demystifying the
and small language
‘black box’ nature of AI models.”
models
Arthur Mensch comments: “Smaller models
also mean the applications are less costly
Small language models (SLMs) are highly
to run and, more importantly, if you have
cost-effective, resource-efficient, and
a model that is 100 times smaller, you can
have minimal environmental impact.
call it 100 times more for the same cost,
Open-source models promote innovation,
bringing a little more intelligence to your
enhance collaboration, and ensure greater
application with each call.”
transparency in development and usage.
Clara Shih adds: “The future of AI will be a
Aiman Ezzat says: “There are clear
combination of both large and small models
advantages to both open and closed
because of climate impact, as well as for cost
approaches. Openness boosts innovation
and performance reasons.”
and drives collaboration. Open models also
When an organization
is making an AI-driven
product, it must
consider the decisions
and outputs the system
will make. These
decisions and these
outputs should be
constrained to respect
the company's role.”
Arthur Mensch
CEO, Mistral AI
12
Capgemini Research Institute Gener(AI)ting the future
Executive Summary
Ricardo Guerra says: “Adopting Gen AI
requires a culture of innovation. With Gen
AI, we need to engage the business and
The future of AI will design teams actively, as they must identify
opportunities beyond mere tech adoption.”
be a combination of
both large and small
Innovative startups
models because of
point to the future of
climate impact, as
Gen AI
well as for cost and
From content creation to multi-agent
performance reasons."
systems, alternative computing, and hybrid
AI, emerging tech startups are pushing the
Clara Shih
boundaries of the next generation of AI
CEO, Salesforce AI
applications.
Getting Gen AI right
Synthesia uses AI to create customizable
requires sound video content featuring realistic avatars,
allowing businesses to generate engaging
technical strategy and a
video presentations. Soundraw offers
culture of innovation an AI-powered platform for generating
original music without the risk of copyright
infringement.
Effective Gen AI adoption requires a well-
coordinated strategic approach with a
In the realm of alternative computing,
strong technical foundation, support from
Mythic develops analog chips for faster,
leadership, and an organizational culture of
more efficient AI tasks such as matrix
innovation.
multiplications, while Groq creates AI-
optimized language processing units (LPUs)
Chema Alonso, Chief Digital Officer,
designed for running LLMs.
Telefónica, suggests: “You need to have a
robust technical strategy based on cloud
Liquid AI is pioneering the development of
and sound data, and the rest will fall into
highly efficient, task-specific models using
place. Secondly, you need to have strong
liquid neural networks, with applications
support from top management. Finally,
such as drone navigation, showcasing Gen
you need sufficient budget. Once you have
AI’s wide range of possibilities.
that, you need to make sure that your whole
organization is very well trained on Gen AI –
what can and cannot be done.”
13
Gener(AI)ting the future Capgemini Research Institute
The CEO Corner
Arthur Mensch
CEO
Mistral AI
The CEO
in discussion with
Corner
Aiman Ezzat
CEO
Capgemini
14
Capgemini Research Institute Gener(AI)ting the future
15
Gener(AI)ting the future Capgemini Research Institute
The CEO Corner
Arthur Mensch Aiman Ezzat
CEO, Mistral AI CEO, Capgemini
Arthur Mensch is a French entrepreneur and With more than 20 years’ experience
scientist. at Capgemini, Aiman Ezzat has a deep
knowledge of the Group’s main businesses.
In 2023, Arthur Mensch, along with Guillaume
He has worked in many countries, notably
Lample and Timothée Lacroix, founded Mistral
the UK and the US, where he lived for more
AI with the mission of making generative AI
than 15 years.
ubiquitous and pioneering a new approach to AI
- one that is more open, portable, independent, Aiman was appointed CEO in May 2020.
and accessible to all. Prior to that, from 2018 to 2020, he served
as the Group’s COO and, from 2012 to
After more than 10 years of academic work
2018, as CFO. Aiman is also on the Board
focused on the possibilities of machine learning
of Directors of Air Liquide and is a member
in the field of brain imaging and on optimization
of the Business Council and the European
of machine learning, he joined DeepMind Paris in
Round Table (ERT) for Industry.
2020 as a researcher, where he spent three years
and played a key role in the development and
deployment of flagship projects in generative AI.
1166
Capgemini Research Institute Gener(AI)ting the future
The CEO Corner
What inspired you to form a new player [in
Mistral] in the generative AI (Gen AI) space –
and why in Europe?
— Arthur: My co-founders and I have been working in the
Gen AI space for over 10 years, previously in large US-based
organizations. When development accelerated at end-2022,
it gave us an opportunity to create some very strong models
in a short period of time. We secured funding, assembled
a dedicated team and the GPUs [graphic processing units]
required to train the LLMs [large language models], and
were ready to go.
Why Europe? Europe is a great place to start a company. The
education systems in France, Poland, or the UK, for example,
are great for training AI scientists. We brought in recent
PhDs from Paris; we were able to get the most important
thing to get started – the team. As the only player in Europe
in the field of conventional language models, we had some
strong geographical business opportunities.
We use both an open-source
model and a portable platform
for model deployment."
Arthur Mensch
1177
Gener(AI)ting the future Capgemini Research Institute
The CEO Corner
What do you see as the advantages of the open-source gen
AI model?
— Arthur: We use both an open-source model and a portable platform for
model deployment. Even our commercial models are licensed. This allows
users to customize the models to their needs.
It offers portability and comfort. With a model that you can deploy on any
platform, on a private cloud or on-premise or on dedicated instances on
the cloud, you can use the technology where your data is. So, this adapts to
the data-governance constraints of the enterprises, and our customers very
much appreciate this flexibility.
— Aiman: There are clear advantages to both open and closed approaches.
Openness boosts innovation and drives collaboration. Open models also
allow everyone to scrutinize the model for potential sources of bias,
demystifying the “black box” nature of AI models. There are also challenges.
Customizing open models for a particular industry or organization is tricky,
but using open models out of the box can lead to suboptimal performance.
Fine-tuning any foundation model, open-source or proprietary, is a time-
consuming, resource-intensive process that requires significant financial
investment. Hence, it is important for enterprises to assess ROI carefully
before pushing out the Gen AI boat.
"Open models also allow
everyone to scrutinize
the model for potential
sources of bias."
Aiman Ezzat
1188
Capgemini Research Institute Gener(AI)ting the future
The CEO Corner
Do you see organizations using a generalized Gen AI model going
forward or many different specialized models?
— Arthur: We see the field moving in these two directions simultaneously. A strong
generalized model gives a good platform for testing solutions. But this can be a
slow and costly process, offering poor ROI for specific tasks. You want your LLM
to offer an intelligent, dynamic solution for a specific issue, whether that’s parsing
the logs of an IT system or parsing the conversation between a customer and a
customer agent.
From a scientific point of view, smaller models
can solve specific issues, but they must be
finely tuned. We want to bring solutions to
market that develop the smallest possible
If you have a model
model to solve a specific defined task, which
will allow for low-latency applications. that is 100 times
smaller, you can call it
Smaller models also mean the applications
are less costly to run and, more importantly, 100 times more for the
if you have a model that is 100 times smaller,
same cost, bringing a
you can call it 100 times more for the same
cost, bringing a little more intelligence to little more intelligence
your application with each call. We call this
to your application
“compressed knowledge”. We specialize
models in order to make differentiated with each call."
applications that go fast, that call LLM often
and that are cost-controlled. Arthur Mensch
— Aiman: There’s a very clear market for both generalized and specialized models.
A generalized model can serve those use cases that don’t require extensive
customization. These are “low-hanging fruit” that rapidly demonstrate the power of
Gen AI.
Developing and training specialized models for some basic use cases might even
be counterproductive in terms of cost and sustainability. That said, there are use
cases that benefit from specialized models, for instance, in terms of performance
characteristics or in detecting and responding to specific nuances of the industry
or use case. Any use case that requires high performance or deep domain
expertise will likely continue to go down the path of specialized models. At the
same time, specialized models potentially require significant resources in terms
of maintenance and regular updates, so organizations might prefer a generalized
model for use cases with less stringent requirements. I see a future where both
types of models coexist harmoniously.
1199
Gener(AI)ting the future Capgemini Research Institute
The CEO Corner
Any use case that
requires high
performance or deep
domain expertise
will likely continue to
go down the path of
specialized models."
Aiman Ezzat
What are the most innovative use cases that you are
seeing in Gen AI?
— Arthur: In financial services, for instance, Mistral has built models that
extract information from financial reports and summarize it for bankers to
analyze. This harnesses the power of generative AI to process a large amount
of text and detect weak signals, which is very much the core business of
banks. The other successful deployment is in customer services.
— Aiman: We have been working on several innovative cases using Gen
AI across industries. In life sciences for instance, we have developed with
generative AI a solution to design new drug molecules. This method
significantly boosts the process of generating new structures, offering
researchers a potent tool for designing molecules aimed at specific
biological targets. It illustrates AI's transformative potential in accelerating
and refining drug discovery, particularly in the preliminary phases.
2200
Capgemini Research Institute Gener(AI)ting the future
The CEO Corner
Given the energy required to create and train the large
models, what are the sustainability implications for Gen
AI?
— Arthur: Most of the compute and energy resources required to run these
systems are used at inference time rather than at training time. So you train
for a couple of months, and when the models are deployed on many, many
GPUs, then the large energy consumption is more linked to the usage than
to the training itself. There are trade-offs between the amount you spend
on training and the compression that you can achieve. If you invest more in
training, you can make smaller models, achieving the same performance as a
larger model with less compute. These smaller models consume less energy
to deploy at inference time.
At Mistral, we focus on
compressing knowledge and
making models that are smaller
Most of the compute and
than those the competition
produces. Limiting carbon energy resources required to
emissions is a cause that is
run these systems are used
very dear to our heart and the
reason why we deployed our at inference time, rather than
solutions in Europe. In Sweden,
at training time."
in particular, renewables
compose a high proportion of
energy consumption. Arthur Mensch
— Aiman: Our research shows that more than three-quarters of organizations are
conscious of environmental concerns around Gen AI. As a leader in the eco-digital
revolution, we at Capgemini recognize the need to weigh the immense potential of Gen
AI against its cost to the planet and society. We are committed to taking a “sustainable
by design” approach to developing Gen AI solutions that harness cutting-edge data, AI,
and climate tech to maximize business outcomes in a sustainable manner.
Mitigation strategies include optimizing the amount of data required to train the
models, working on smaller, task-specific energy-efficient models that employ more
efficient training and operating algorithms, and powering the AI infrastructure with
renewable energy as well as using more energy-efficient datacenters. We also promote
transparency in AI development and operation by monitoring and disclosing the energy
consumption and carbon footprints of Gen AI models. Our Gen AI lifecycle analysis tools
help organizations to mitigate environmental impact.
2211
Gener(AI)ting the future Capgemini Research Institute
The CEO Corner
How should large organizations address ethical considerations and
potential bias in deployment of AI models?
— Arthur: When an organization is
making an AI-driven product, it has to
consider the decisions and outputs the
system will make. So, these decisions "The rare
and these outputs should be constrained
talent that we
to respect the company's role. What it
means is that before deployment of a recommend
new AI product, the first thing to think
every
about is how do you evaluate success.
How do you ensure that the model is organization
behaving as it should and not producing
look for is
unwanted outputs? And is it able to
provide a nuanced but unbiased answer the software
to complex questions?
engineer who
Owing to our open approach, the can also do data
customer can make their own editorial
science."
choices from these evaluations.
— Aiman: Large organizations should be Arthur Mensch
conscious of a variety of risks: Inherited
risk, intellectual property, correctness,
data leakage, and user privacy.
Organizations should establish employee
guidelines for the safe use of Gen AI
and validating outputs to eliminate
bias. At Capgemini, we have applied a
governance model to ensure this.
22
Capgemini Research Institute Gener(AI)ting the future
The CEO Corner
How are you bridging the AI talent demand-supply gap?
— Arthur: It has been a challenge to get the best AI scientists. We
recommend hiring very strong data scientists who can undertake software
development. Since we are making the tools and the foundation for the
model itself, training the model is not a necessity within the enterprise
setting. To make the most interesting products, clients must understand
how to use the platform.
So, the capacity for doing this is really adjacent to what we used to call
data science a decade ago. It's the ability to run experiments, to evaluate
certain systems, to see what is failing, and to see how to try and improve it.
This scientific mindset, running experiments on a computer and measuring
success, which is really the data scientist's job. The changes with the data
scientist's job today is that the software requirement is stronger because,
if you want to make an interesting application, you also need to dive deep
into the way you assemble the software, connect it to the LLM, the LLM to
the database, and an LLM to tools. Having a system mindset is necessary to
create successful applications. The rare talent that we recommend every
organization look for is the software engineer who can also do data science.
— Aiman: We are investing over €2 billion over three years in Gen AI and
have already trained over 120,000 team members on generative AI tools
thanks to our Gen AI Campus. We have also launched a dedicated platform
to industrialize our custom generative AI projects. We will also focus on
obtaining certifications and building centers of excellence, as well as specific
go-to-market skills. Ultimately, Gen AI training will be a key requirement in
all of our development and training curricula.
2233
Gener(AI)ting the future Capgemini Research Institute
The CEO Corner
How do you see gen AI driving transformation in large
organizations?
— Arthur: The first step is to take a model – a Mistral model, for instance
– and connect it to the enterprise context. The enterprise context is
located across different databases or SaaS [Software-as-a-Service]
systems. You can then generate assistants with access to the enterprise
context, to help every employee navigate the enterprise processes and
organization. That's typically what our customers do first. They create a
knowledge management tool or general assistant for employees.
— Aiman: Driving transformation with generative AI goes beyond the
technology. Success depends on a broad strategic vision that covers
everything from applying it to the right use cases, potentially adapting
internal processes, to optimizing customer-facing operations. In addition,
the value of generative AI depends on two key foundations: the data and
the human elements. Leaders need to have the right data foundations
in place to ensure they are realising the full potential of Gen AI. Equally
important is training employees to not only use AI effectively but also to
trust it, which is key to adoption.
2244
Capgemini Research Institute Gener(AI)ting the future
The CEO Corner
Arthur Mensch Aiman Ezzat
CEO, Mistral AI CEO, Capgemini
"The first step "There’s a very
is to take a clear market
model – a for both
Mistral model, generalized
for instance and specialized
– and connect models."
it to the
enterprise
context."
222555
Gener(AI)ting the future Capgemini Research Institute
Executive Conversations
Executive
conversations
with…
26
Capgemini Research Institute Gener(AI)ting the future
Executive Conversations
STANFORD SALESFORCE AI
Erik Brynjolfsson Clara Shih
Professor CEO
h p.28 h p.64
OECD LANDINGAI
Audrey Plonk Andrew Ng
Deputy Director, Directorate for CEO
Science, Technology and Innovation (STI)
h p.76
h p.36
EUROPEAN PARLIAMENT TELEFÓNICA
Dragoş Tudorache Chema Alonso
Former MEP, EU AI Act co-rapporteur Chief Digital Officer
h p.44 h p.86
ADOBE ITAÚ UNIBANCO
Scott Belsky Ricardo Guerra
Chief Strategy Officer Chief Information Officer
h p.54 h p.94
27
Gener(AI)ting the future Capgemini Research Institute
Executive Conversations
ERIK
BRYNJOLFSSON
Professor at the Stanford Institute
for Human-Centered AI, and
Director of the Stanford Digital
Economy Lab
28
Capgemini Research Institute Gener(AI)ting the future
Executive Conversations
GENERATING
GROWTH
THROUGH AI
Erik Brynjolfsson is the Jerry Yang the Co-founder of Workhelix. One of the most
and Akiko Yamazaki Professor cited authors on the economics of information,
and Senior Fellow at the Stanford he was among the first researchers to measure
Institute for Human-Centered AI the productivity contributions of IT and the
(HAI), and Director of the Stanford complementary role of organizational capital and
Digital Economy Lab. He is also the other intangibles. He is the author of nine books,
Ralph Landau Senior Fellow at the including the bestseller The Second Machine
Stanford Institute for Economic Age: Work, Progress, and Prosperity in a Time
Policy Research (SIEPR), Research of Brilliant Technologies (2014) with co-author
Associate at the National Bureau Andrew McAfee, and Machine, Platform, Crowd:
of Economic Research (NBER), and Harnessing Our Digital Future (2017).
29
Gener(AI)ting the future Capgemini Research Institute
Executive Conversations
"AI – IN PARTICULAR GENERATIVE
AI – IS THE ELECTRICITY OF OUR
ERA, INCREASINGLY UBIQUITOUS
AND SPAWNING COUNTLESS
COMPLEMENTARY INNOVATIONS."
How is generative AI transformational?
The biggest driver of productivity growth for
businesses and the economy as a whole, is what
economists call “general-purpose technologies”
or GPT, the same initialism AI researchers now use
for “generative pre-trained transformers.” AI – in
particular generat |
227 | capgemini | CRI_Turbocharging-Software-with-Gen-AI-1.pdf | Turbocharging software with Gen AI: How organizations can realize the full potential of generative AI for software engineering
Turbocharging
software with Gen AI
How organizations can realize the full potential
of generative AI for software engineering
#GetTheFutureYouWant
#GetTheFutureYouWant
fo
elbaT
tnetnoC
2
Turbocharging software with Gen AI: How organizations can realize the full potential of generative AI for software engineering
Capgemini Research Institute 2024
3
Turbocharging software with Gen AI: How organizations can realize the full potential of generative AI for software engineering
Capgemini Research Institute 2024
evitucexE
yrammuS
4
Turbocharging software with Gen AI: How organizations can realize the full potential of generative AI for software engineering
Organizations are reaping multiple benefits • Organizations are utilizing these productivity gains on
innovative work such as developing new software features
from leveraging generative AI for software
(50%) and upskilling (47%). Very few aim to reduce
engineering.
headcount (4%).
• Generative AI is having a positive impact on software
• The leading benefits for organizations are enabling
professionals’ job satisfaction.
more innovative work, such as developing new software
features/services (observed by 61% of surveyed • 69% of senior software professionals and 55% of junior
organizations), improving software quality (49%), and software professionals report high levels of satisfaction
increasing productivity (40%). from using generative AI for software.
• 78% of software professionals are optimistic about
• Organizations using generative AI have seen a 7–18% generative AI’s potential to enhance collaboration
productivity improvement1 in the software engineering between business and technology teams.
function as per early estimates. This is highest for
specialized tasks such as coding assistance2 (34% as the
maximum potential for time savings with 9% on average)
and creating documentation (35% as the maximum
potential for time savings with 10% on average). This
research analyzed time savings in various software
engineering tasks using generative AI tools and not cost
savings which can be significantly different.
Capgemini Research Institute 2024
%
78
evitucexE
yrammuS
5
Turbocharging software with Gen AI: How organizations can realize the full potential of generative AI for software engineering
Generative AI adoption is at an early stage • Generative AI is expected to play a key role in augmenting
the software workforce with better experience, tools and
but will accelerate sharply.
platforms, and governance (assisting in more than 25% of
software design, development, and testing work by 2026).
• Adoption of generative AI for software engineering is still
in its early stages, with 9 in 10 organizations yet to scale. • Coding assistance is the leading use case, but generative
AI also finds applications in other software development
• 27% of organizations are running generative AI pilots,
lifecycle (SDLC) activities (test case generation,
and 11% have started leveraging generative AI in their
documentation, code modernization, UX design assistance,
software functions.
etc.)
• Three in four (75%) large organizations (annual revenue
greater than $20 billion) have adopted (piloted/
• Most use cases have yet to be adopted by a majority of
scaled) generative AI compared to 23% of their smaller
organizations (39% are focusing on coding assistance and
counterparts (annual revenue between $1–5 billion).
37% on UX design assistance as top adopted use cases).
• Adoption (including pilots) is expected to increase
significantly in the next two years from 46% of software
workforce using generative AI tools today (for any kind of
training, experimenting, piloting, and implementing, with
authorized or unauthorized access) to an estimate of 85%
in 2026.
Capgemini Research Institute 2024
evitucexE
yrammuS
6
Turbocharging software with Gen AI: How organizations can realize the full potential of generative AI for software engineering
Lack of foundational prerequisites and • Using unauthorized tools without proper governance and
oversight exposes organizations to functional, security,
unofficial usage of generative AI pose
and legal risks like hallucinated code, code leakage, and
significant functional, security, and
IP issues.
legal risks.
• 27% of organizations have the platforms & tools, and 32%
have talent prerequisites in place, to implement generative
AI for software engineering.
• Over 60% lack governance and upskilling programs for
generative AI for software engineering.
• Of those software professionals who use generative AI,
63% use unauthorized tools.
• Nearly a third of the workforce is self-training on
generative AI for software as less than 40% of employees
are receiving training from their organizations.
Capgemini Research Institute 2024
evitucexE
yrammuS
7
Turbocharging software with Gen AI: How organizations can realize the full potential of generative AI for software engineering
How can organizations harness the full retention, solving complex issues, and collaborating
with business.
potential of generative AI for software
engineering? • Identify requirements for new capabilities and
source them.
• Prepare for generative AI use by delivering technology
• Select and prioritize high benefit use cases. prerequisites:
• Build a repository of platforms and tools for a seamless
• Mitigate risks around security, IP/copyright issues, and and augmented software engineering experience.
code leakage using a thorough risk management approach.
• Privately and safely contextualize generative AI
assistants with organization’s own content.
• Transform your software organization to ensure optimal
usage of generative AI:
• Adopt a measurement protocol for generative AI impact
• Augment your software teams with a generative AI monitoring and use case prioritization.
assistant. A majority of junior (53%) as well as senior
professionals (58%) believe that generative AI tools
will augment their day-to-day work within the next • Put people at the heart of this transformation by creating
two years. For instance, generative AI tools can help a learning culture at your organization.
junior professionals learn faster and come up to speed
• Provide upskilling and cross-skilling opportunities.
quickly, while they allow senior professionals to focus
on grooming juniors by ensuring their learning and • Address employees’ work displacement concerns.
Capgemini Research Institute 2024
8
Turbocharging software with Gen AI: How organizations can realize the full potential of generative AI for software engineering
Who should This report provides insights into the use of the full potential of generative AI for software
generative AI for software engineering and engineering.
offers recommendations that will be useful to
read this Business leaders in technology, IT, product,
organizations across industries in harnessing
strategy, R&D/engineering, general management,
and innovation who have responsibility for –
report and
and oversight of – their organization’s software
engineering function will find it particularly useful.
why? 1000+ This report draws on insights from a
comprehensive multi-sectoral survey of 1,098
senior executives (director level and above)
and 1,092 software professionals (including
architects, developers, testers, and project
organizations with annual revenue
managers) from organizations with over $1 billion
greater than $1 billion, represented by a
in annual revenue. The report covers the major
minimum of one software professional
considerations for implementing generative AI
and one software leader, are part of this
in software engineering and includes in-depth
research.
qualitative insights from 20 industry leaders,
professionals, and entrepreneurs.
Capgemini Research Institute 2024
9
Turbocharging software with Gen AI: How organizations can realize the full potential of generative AI for software engineering
This report is a part of Capgemini Research
Institute’s series on Generative AI
Gen AI in organizations - annual research
Gen AI for management*
Gen AI in supply chain* Gen AI for marketing Gen AI for software Gen AI in R&D
engineering and engineering
Gen AI and
consumers
Gen AI and Gen AI and Gen AI and
Gen AI in business operations* Gen AI in manufacturing* Gen AI in customer service* sustainability* ethics/ cybersecurity*
trust*
Data mastery*
Special edition of our premium journal Conversations for tomorrow on Gen AI*
To find out more, please go to https://www.capgemini.com/insights/research-institute/ *Upcoming reports
Capgemini Research Institute 2024
noitcudortnI
10
Turbocharging software with Gen AI: How organizations can realize the full potential of generative AI for software engineering
Since the dawn of the modern computer age, there Today, by leveraging the power of large language models
has been a disconnect between natural language and (LLMs), generative AI can enhance developers’ productivity,
machine language. With hardware and software advances, improve software quality, and accelerate time to market.
programming has evolved in waves over time and this gap has Marco Argenti, Chief Information Officer at Goldman Sachs:
begun to close (see Figure 1). “Goldman Sachs is using artificial intelligence to turn software
developers and others into superhumans.” 4
This evolution now appears near complete, as natural
language becomes the lingua franca. With recent rapid In generative AI, the software workforce has a tool to
advances in AI and high-performance computing, we can accelerate key tasks (such as design, coding, migrating,
now simply “chat” with computers and – through human testing, deploying, support and maintenance) with minimal
supervision and accountability – let the AI assistant augment effort and a minimal learning curve.
tasks ranging from programming, generating test cases
and user stories, to documenting, among others. As Andrej
Karpathy, one of the founders of OpenAI and former director
of AI at Tesla, famously quipped following the introduction of
ChatGPT: "The hottest new programming language is English”.3
Capgemini Research Institute 2024
noitcudortnI
Figure 1.
Increasing levels of value creation from evolution of software development languages and platforms
Evolvement of software development languages & platforms
ENIAC
1940
Source: Capgemini analysis
noitaerc
eulaV
11
Turbocharging software with Gen AI: How organizations can realize the full potential of generative AI for software engineering
“The hottest new
programming
language is English”
Low code/No code
Cloud native
DevOps Automation
Python
Java
C++
C-Programming
Cobol
IBM 704
Assembler
Generative AI boost
1950 1960 1970 1980 1990 2000 2010 2020 2030
Machine & assembly language High-level programming language Object-oriented language Development platforms
Capgemini Research Institute 2024
However, generative AI brings risks and challenges. implementation approach to harness the potential of generative
Uncontrolled use can lead to hallucinated code, IP issues, AI while managing its risks. With this research we attempt to assess
private data leakages, and security vulnerabilities. Software the impact of generative AI on the software engineering function,
engineering organizations need a new strategy and covering such questions as:
• How will generative AI impact the various stages of software
development lifecycle (SDLC)?
• How can organizations quickly adopt and scale generative AI to
drive productivity and innovation?
• How will generative AI impact software engineers’ ways of
working?
• What are the challenges for software engineering and how best
can we manage the risks associated with generative AI?
• How can organizations continuously measure and optimize
impact of generative AI on their software engineering function?
noitcudortnI
12
Turbocharging software with Gen AI: How organizations can realize the full potential of generative AI for software engineering
Capgemini Research Institute 2024
13
Turbocharging software with Gen AI: How organizations can realize the full potential of generative AI for software engineering
What do we Defining the term “software” • Custom software: Specific, advanced programs
developed for a specific purpose for an individual
Software is a strategic capability, transforming the way or company, which can be modified or changed.
mean by businesses design their products and services, function Custom software is not commercially available
overall, compete, and provide value to customers. Software but is built and operated for internal purposes.
is vital to modern business, whether as a product itself or
“Generative AI
integrated into enterprise apps or products. • Consumer software: Sold directly to end users,
consumer software includes apps, web portals,
There are three main categories of software: and information tools such as maps, financial data,
for software
news, games, and music players.
• Business software: Used by organizations to run,
scale, and optimize day-to-day business functions and • Embedded software: A piece of software to
engineering”? processes and/or interact with their customers and program hardware or non-PC devices to facilitate
partners. functioning. These are specialized environments
and applications for a specific hardware stack with
There are two broad types of business software: performance, power, and functionality requirement
• Packaged software: Third-party standard programs and constraints.
grouped to provide different tools from the same
family in a package, commercially available under the
licensor’s standard terms, payable with either a one-
off or annual fee.
Capgemini Research Institute 2024
14
Turbocharging software with Gen AI: How organizations can realize the full potential of generative AI for software engineering
Generative AI has Generative AI’s potential for software Generative AI’s impact on the SDLC
engineering
potential for all With the increasing proliferation of software in products,
services, operations; software teams are under pressure
Software engineering has shifted strongly towards greater
categories of software,
to deliver more, better, faster. Generative AI has the
automation and simplification, particularly with the advent
potential to yield benefits across the SDLC. Figure 2
but this research focuses of generative artificial intelligence (generative AI). The rise
shows some of the tasks and activities in SDLC that can
of large language models (LLMs) has been key. LLMs are
benefit from the use of generative AI tools. It is worth
largely on software deep-learning AI algorithms that can recognize, summarize,
noting that it is a subset of all activities encompassing
translate, predict, and generate content by building on
engineering for custom, very large datasets. They have facilitated the increasing SDLC. It can be integrated at any stage – from business
needs analysis and writing agile user stories to software
adoption by consumers and organizations of software
embedded, or consumer design, coding, documentation, packaging, deployment,
engineering.
testing, and operations – augmenting the work of
software which goes Generative AI has the potential to transform the software software engineers and helping increase efficiency,
engineering process, as it can be integrated into tech stacks improve quality, and enhance job satisfaction.
through the entire
to unlock new features and updates for software currently
Generative AI also touches the roles of many data
in use. Many leaders are striving to integrate AI-enabled
software development analysts, business analysts, platform/software designers,
plug-ins or incorporate AI-powered technology into their
and software engineers, developers, and tester.
own enterprise and software engineering platforms. Our
lifecycle.
previous research shows that generative AI will assist in
writing one out of every five lines of code in the coming
year.5
Capgemini Research Institute 2024
15
Turbocharging software with Gen AI: How organizations can realize the full potential of generative AI for software engineering
Figure 2.
Potential application areas of generative AI in the SDLC
(DevOps) Software Lifecycle
Business Use case modeling Coding assistance (code generation, Legacy code Code explanation Platform provisioning Software observability with
demand/requirement User stories generation completion) modernization Code documentation & configuration analysis and recommendations
analysis and writing Reverse engineering Unit tests generation (migration, conversion, Code vulnerabilities
etc.)
Business
...............................................D....e...s...i.g...n...........................................................C...o...d....i.n...g............................................................B....u...i..l.d................................................................T...e...s...t...........................................................R....e...l.e...a...s..e..........................................................D....e...p...l.o....y.......................................................O.....p...e...r..a...t..e.......................................................M.....o...n....i.t..or
demand
UX/UI design Software architecture Software refactoring Software packages Test Case generation Software packages assembly Incidents resolution
configuration Test Data sets Release notes Tickets assistance
(Agile) Product Teams / (Waterfall) Development Teams
Backlog and roadmap planning Product value stream performance Team effectiveness analysis and improvement Team communication and collaboration
Effort estimations recommendations Process facilitation (plannings, retrospective, burndown, etc.)
Industrialized Software Engineering Platform
Agile Process Management/ALM Developer workplace (IDE) DevOps automation Tests automation Generative AI foundations
Source: Capgemini Research Institute analysis.
Capgemini Research Institute 2024
16
Turbocharging software with Gen AI: How organizations can realize the full potential of generative AI for software engineering
01
Organizations are reaping
significant benefits from
leveraging generative AI for
software engineering.
Capgemini Research Institute 2024
17
Turbocharging software with Gen AI: How organizations can realize the full potential of generative AI for software engineering
Figure 3. Augmenting innovation and
One in two organizations adopting generative AI sees improvements in enabling innovative work and quality of software.
improving software quality
are the leading benefits.
Percentage of organizations seeing benefits through the adoption of generative AI,
as mentioned by software leaders
Three in five organizations see innovative work – for
Enabling innovative work example, developing new features and services using
61%
(e.g., developing new features, services etc.) software – as the biggest benefit of generative AI use
in software engineering (see Figure 3). Of software
Quality of software 49%
professionals surveyed, 80% believe that, by automating
simpler repetitive tasks, generative AI will free up time
Productivity 41%
for them to focus on innovation and value-adding tasks,
fostering greater creativity.
Collaboration 36%
Akram Sheriff, Senior Software Engineering Leader (Gen
Security 34% AI, AI, ML) at Cisco Systems elaborates: “One of the biggest
drivers of generative AI adoption is innovation. Not just on
Time to market/reduction in lead-time 33% the product side but also on the process side. While senior
professionals are leveraging generative AI combined with their
Cost of software development 25% domain expertise for product innovation, junior professionals
see value in AI process and tool innovation, and in automation
Technical debt 12%
and productivity optimization.”
Compliance and risk management 9%
Source: Capgemini Research Institute, Generative AI in Software Engineering, Senior Executive Survey, April 2024, n =
412 software leaders that have scaled up or are running pilots with generative AI in software engineering.
Capgemini Research Institute 2024
18
Turbocharging software with Gen AI: How organizations can realize the full potential of generative AI for software engineering
Generative AI also enables improvements in software
quality. It can help deliver higher-quality code with fewer
errors and improvements in test coverage and quality. Both
factors give organizations a productivity boost at team and
organizational levels. For example, Emirates NBD, a large
banking group in the Middle East, not only accelerated
developer productivity by up to 20% in complex tasks, but
also improved the company’s code quality by 20% by using
GitHub Copilot’s code suggestions.6
Head of AI at a leading Australian telco, explains: “With use of
generative AI for software engineering, the number of test cases
could be increased by 30%, greatly enhancing test coverage and
quality.”
Capgemini Research Institute 2024
19
Turbocharging software with Gen AI: How organizations can realize the full potential of generative AI for software engineering
Figure 4. For telecom businesses, generative AI can play a significant
Telecom and retail sectors see enablement of innovative work as a top benefit from generative AI. role in the development of such data-powered, innovative
applications as network management and maintenance
as well as customer service/sales apps offering hyper-
Percentage of organizations by sector, who have active initiatives and see enable- personalization. BT Group’s Digital unit has an AI-powered
ment of innovative work as a top benefit, as per software leaders product lifecycle management strategy. Within four months
of deploying Amazon’s CodeWhisperer, it had automated
Telecommunications 86% nearly 12% of repetitive work, allowing the pilot workforce to
focus on more strategic goals.7
Retail 76%
Similarly, the retail industry is leveraging generative AI
Life sciences and healthcare 74% to gather and analyze customer preferences, competitor
insights, past sales history, etc., and create robust and precise
Consumer products 71%
requirements documentation as the basis of engaging
Global 61% customer-facing apps. Wayfair, a home goods company, is
considering using generative AI to reduce the technical debt
High tech 58%
accumulated in their software stack over years.8
Energy transition & utilities 56%
Banking 55%
Aerospace & defence 53%
Automotive 52%
Public services 52%
Insurance 45%
Source: Capgemini Research Institute, Generative AI in Software Engineering, Senior Executive Survey, April 2024, n = 412 senior
executives that have scaled up or running pilots with generative AI in software engineering.
Capgemini Research Institute 2024
20
Turbocharging software with Gen AI: How organizations can realize the full potential of generative AI for software engineering
Organizations with active Figure 5.
Larger organizations have seen greater productivity improvement with generative AI.
generative AI initiatives
have seen an average
Productivity improvement range of a software professional, grouped by organization revenue size
7–18% improvement in
productivity across the 19% 19% 19%
18%
18%
SDLC.
15%
11%
Those organizations actively using generative AI in software
9% 9%
engineering have seen an average total productivity
7% 6%
improvement of 7–18% across the SDLC today, compared to
4%
non-usage of generative AI. The increasing maturity of tools
and processes along with growing professional experience,
means productivity is likely to continue to improve.
We also found that productivity advantage increases with Global Average USD 1 billion to USD 5 billion to USD 10 billion to USD 20 billion to More than USD 50
< USD 5 billion < USD 10 billion < USD 20 billion < USD 50 billion billion
organization size (see Figure 5).
Top range of productivity improvement Bottom range of productivity improvement
Source: Capgemini Research Institute, Generative AI in Software Engineering, Senior Executive Survey, April 2024, n =
412 software leaders that have scaled up or running pilots with generative AI in software engineering. Top and bottom
productivity ranges are found by the 80th and 20th percentile respectively of individual productivity improvement data.
Capgemini Research Institute 2024
21
Turbocharging software with Gen AI: How organizations can realize the full potential of generative AI for software engineering
Ancileo, Singapore-based insurance software-as-a-service
(SaaS) company, used generative AI to increase developer
productivity. Sylvain Dutzer, Chief Technical Officer, Ancileo:
“Ancileo is using Amazon Q to supercharge our developers by
helping them understand existing codebase and troubleshoot
directly in their integrated development environment (IDE). This
allows our team to reduce time resolving coding-related issues by
30%. Even our architects use it to help find the best solutions to
specific problems based on context.” 9
Nitin Tandon, Chief Information Officer of financial services
firm Vanguard: “We are enabling productivity gains for
developers by experimenting ‘rapidly and safely’ with generative
AI tools — with human oversight and expertise.” 10
Improvements in coding speed (78%) and testing speed (54%)
are the top reasons cited for this improvement. Generative
AI can produce test cases directly from requirements, with
significant time savings. Where testing an app requires
certain application programming interfaces (APIs), AI test
code generators can create these snippets quickly.
Capgemini Research Institute 2024
22
Turbocharging software with Gen AI: How organizations can realize the full potential of generative AI for software engineering
Generative AI yields Figure 6.
Generative AI shows significant productivity improvement in terms of time savings for documentation and coding assistance tasks.
productivity improvement
for a set of specialized Maximum and average time savings through generative AI usage for a set of activties (current estimates)
software tasks
35%
34%
We analyzed specific tasks from a software professional’s
daily work to understand the impact of generative AI.
Some tasks are better suited for generative AI, given the
maturity of the tools available and the experience of the 20% 20%
workforce. As shown in Figure 6, creating literature and
documentation, and writing code and scripts show the
greatest timesaving. This tapers off for the remaining 10%
9%
major task categories in the SDLC. However, as toolchains
5%
and platforms improve, this benefit is likely to spread. It is
1%
important to note that saving time using generative AI tools
is significantly different from saving cost. Assessing cost
savings was not a part of the scope of this research. Creating literature Writing code and Debugging and Project management
and documentation scripts testing
Maximum improvement Average improvement
Source: Capgemini Research Institute, Generative AI in Software Engineering, Software Professionals Survey, April 2024, n = 368
software professionals that are actively using generative AI. Maximum improvement is represented by the 95th percentile’s results,
while an average user is represented by the statistical average.
Capgemini Research Institute 2024
23
Turbocharging software with Gen AI: How organizations can realize the full potential of generative AI for software engineering
Organizations are utilizing Figure 7.
Innovative work and upskilling are the top areas where organizations are channelizing productivity gains.
productivity gains on
innovative work and How is your organization planning to leverage the additional time freed up by generative AI?
upskilling, not headcount
Focus software professionals' efforts on innovation for e.g.,
50%
reduction. developing new features, services etc.
Upskill software professionals on business skills and understanding 47%
According to our survey, 79% believe generative AI will
significantly reduce the workload and free up additional time Focus software professionals' efforts on complex, high-value tasks 46%
for software professionals. This freed up time is being used
for higher-value-adding tasks including enhanced innovation
Upskill software professionals on advanced technical capabilities 45%
and upskilling, as shown in Figure 7.
Mousumi Bhattacharya, Director of IT at Centene, a US-based Invest in cross-skilling of software professionals 37%
managed care company: “Generative AI has tremendous
potential to improve productivity by shifting professional Train software professionals to ensure quality, security,
31%
efforts and time from mundane and repetitive things to more IP, ethical issues standards are being met
meaningful, creative and challenging tasks.”
Reducing technical debt 26%
Stephane Dupont, EVP and Head of Operations at Airbus,
the leading European aerospace company: “I see it as a
Reduce the size of workforce 4%
coding assistant, giving developers more time to think about
the architecture, the new features, next steps, quality, etc., and
spending less time on pure code development.”
Source: Capgemini Research Institute, Generative AI in Software Engineering, Senior Executives Survey, April 2024,
N = 870 senior executives who believe that generative AI will free up additional time for software professionals
Capgemini Research Institute 2024
24
Turbocharging software with Gen AI: How organizations can realize the full potential of generative AI for software engineering
Brian Lanehart, president, and CTO of financial technology
provider Momnt: “Being able to completely communicate an
entire application request to generative AI will reduce a task
timeline significantly. That means an engineer or team is freed up
to think creatively or strategically.” 11
Reducing headcount is the least-adopted route (taken by
only 4% of responding organizations); and new roles, such as
generative AI developer, generative AI Architect, AI platform
architect, prompt engineer, etc. have evolved. The head of
AI at a leading Australian telco: “Even as autonomous vehicles
“I see it as a coding assistant, giving developers more time to think
are a reality, human supervision and ability to take control is
still required. Similarly, software engineers won’t be replaced by about the architecture, the new features, next steps, quality, etc.,
generative AI – they will start thinking about the actual design
and spending less time on pure code development.”
process, long-term strategy, next phase of software, etc. rather
than spending a year writing code.“
Stephane Dupont
EVP and Head of Operations at Airbus,
the leading European aerospace company
Capgemini Research Institute 2024
25
Turbocharging software with Gen AI: How organizations can realize the full potential of generative AI for software engineering
Generative AI benefits in Figure 8.
Senior and junior professionals see positive impacts of generative AI on their job satisfaction
software engineering
extend to job satisfaction
Extent to which generative AI impacts these areas within your organization’s
software engineering function
and happiness.
69%
Our research shows that generative AI has a positive impact
on software professionals’ job satisfaction and reduces 55%
51%
attrition rates (see Figure 8). Fabio Veronese, Head of ICT 47%
Industrial Delivery at ENEL Group: “We are more ambitious. For
us, improving development productivity with generative AI is not
just about lines of code. It is also about developer experience.”
%
69
Senior software professionals Junior software professionals
(having experience > 3 years) (having experience <= 3 years)
Senior software professionals believe that Improve job satisfaction Reduce attrition
generative AI will have a positive impact on job
satisfaction
Source: Capgemini Research Institute, Generative AI in Software Engineering, Software
Professionals Survey, April 2024, N = 215 software professionals
Capgemini Research Institute 2024
26
Turbocharging software with Gen AI: How organizations can realize the full potential of generative AI for software engineering
Most of the current workforce sees generative AI as a strong Figure 9.
enabler and motivator – 35% associate it with being “assisted Most of the workforce feels positive about generative AI tools for software engineering.
and augmented,” and 24% feel “excited and happy” about its
adoption (see Figure 9).
How does the workforce feel as regards to the adoption of generative AI
While there is currently an emphasis on generative AI’s
utility in code completion and writing, three in four senior
executives believe it will significantly transform their
software engineering organization.
Tommy MacWilliam, Engineering Manager for Infrast |
228 | capgemini | Semiconductors-report.pdf | The semiconductor
industry in the AI era
Innovating for tomorrow’s demands
#GetTheFutureYouWant
fo
elbaT
stnetnoc
2
The semiconductor industry in the AI era
Snapshot of research Downstream industries
methodology express concerns
over semiconductor
supply
Who should read this Organizations The semiconductor
report and why? anticipate surging industry is innovating,
semiconductor demand but softwarization
remains a challenge
Capgemini Research Institute 2025
3
The semiconductor industry in the AI era
Resilience and Partnerships are taking Conclusion
sustainability gather the industry forward
momentum
Exploring big tech's How the semiconductor Research
shift to in-house
industry can capitalize methodology
chip design
on emerging
opportunities
Capgemini Research Institute 2025
evitucexE
yrammus
4
The semiconductor industry in the AI era
AI adoption is powering a surge in demand for customization, introduce more comprehensive application
semiconductors: programming interfaces (APIs) and software development kits
(SDKs), as well as reinforcing security features. Consequently,
While semiconductor industry organizations forecast a 15%
one in three downstream organizations is exploring or is
rise in two years, the downstream organizations (those reliant
actively engaged in in-house chip design, enabling greater
on semiconductor supply for their products or services and
customization while also gaining more control over
operations) anticipate their demand for chips to increase at
their supply chains. Further, sustainability, supply chain
a higher growth rate of 29%. Increased adoption of artificial
resilience, and security are critical concerns for downstream
intelligence (AI) and generative AI (Gen AI) is driving the
organizations going forward.
need for specialized neural processing units (NPUs) and
high-performance graphics processing units (GPUs) that can Amid “softwarization” challenges, innovation shines
handle massive computations and large datasets efficiently. through:
Additionally, downstream organizations expect their demand
The semiconductor industry continues to excel in innovation
for AI chips, custom silicon chips, and memory-intensive chips
across a number of areas. Although various players stand out
to increase over the next 12 months.
in specific areas, our analysis reveals three types of innovation
Amid buoyant demand lies concern: that are consistently prioritized across the industry:
Over half of downstream organizations doubt the • Design innovation: Advances in chip architectures, such as
semiconductor industry's ability to meet their needs. 3D integrated circuit (IC) design and multi-die integration
Technological advancements, including GPU computing for AI are pushing the boundaries of performance and energy
and machine learning and inference acceleration are important efficiency, while half of design organizations are investing in
to these industries, which are continually seeking to enhance Gen AI to shorten design cycles.
Capgemini Research Institute 2025
evitucexE
yrammus
5
The semiconductor industry in the AI era
• Manufacturing innovation: Advances in extreme ultraviolet extend the chip lifecycle, and enhance customization in an
(EUV) lithography and the shift towards smaller process evolving market, the industry finds monetizing its software
nodes (i.e., 3 nanometer and 2 nm) enable the production a challenge.
of more powerful and efficient chips. Nearly half of
Focus on supply chain resilience and sustainability:
manufacturers also rely on AI and ML to optimize processes.
Only two in five semiconductor organizations are confident
• Packaging innovation: 3D packaging and the use of chiplets
in the resilience of their supply chains. Organizations focus
(tiny integrated circuits that can be combined to create
on onshoring and “friendshoring” (basing supply chains in
complex components) are enhancing functionality and
countries that are geopolitical allies) to enhance stability and
performance without increasing physical footprint.
reduce single-region dependency. Consequently, the industry
Hardware security remains paramount, with significant anticipates domestic sourcing will improve by 17% over the
investment in secure chip design, hardware-based encryption, next two years. Three-quarters (74%) of organizations expect
and root of trust (RoT, a trusted source within a cryptographic to increase their US investments, and 59% will increase
system) technologies. But while there is steady progress in investment in the EU.
integrating software and hardware to create more adaptable
Besides its continued focus on improving power efficiency,
and programmable semiconductor solutions, monetization
the industry is becoming more eco-friendly by cutting
remains a hurdle.
energy consumption, implementing water recycling and
However, the “softwarization of semiconductors” is falling reuse systems, using less toxic alternative chemicals, and
short of the industry's expectations. While this innovation is minimizing waste.
crucial for semiconductor companies to expand use cases,
Capgemini Research Institute 2025
evitucexE
yrammus
6
The semiconductor industry in the AI era
Recommendations for the semiconductor industry: • Coordinate strategies with governmental initiatives
To capitalize on emerging opportunities, semiconductor such as grants for R&D and collaborate within the industry
organizations should consider the following: ecosystem to drive shared innovation and standardization.
• Utilize AI and Gen AI to automate design processes, • Enhance cybersecurity measures to protect data
improve production efficiency, and optimize performance to integrity, use advanced security to safeguard proprietary
meet the specialized needs of emerging applications. technologies, and advocate for stronger intellectual
property (IP) laws to deter infringement and protect
• Invest in cutting-edge fabrication methods such as 3D chip
innovation-led competitive advantage.
stacking and accelerate research in emerging fields such as
advanced silicon photonics integration. • Adopt open standards and open-source collaboration to
drive semiconductor innovation.
• Diversify supplier networks across multiple regions while
investing in R&D for alternative materials and technologies.
Implement sustainable manufacturing practices such as
green chemistry and utilize renewable energy sources to
minimize carbon footprint.
Capgemini Research Institute 2025
7
The semiconductor industry in the AI era
Who should read this report and why?
This report will be relevant for decision-makers topics like cybersecurity, softwarization, supply silicon, in-house chip design, sustainability, and supply chain
across the semiconductor ecosystem and its chain resilience, and sustainability—enabling dynamics provide guidance for aligning technology adoption
down-stream industries. Specifically, this will be executives to align their strategies for the future. with organizational goals.
useful for:
2. Leaders in downstream industries: By connecting perspectives from both semiconductor
1. Executives in the semiconductor industry: Professionals in automotive, consumer executives and downstream organizations, this report equips
Integrated device manufacturers, fabless design electronics, retail, telecom, aerospace and stakeholders with the knowledge and strategies needed to
firms, foundries, OSAT companies, material defense, high tech, medical devices/medical thrive in a rapidly evolving landscape.
and subsystem companies, and semiconductor electronics, industrial equipment, financial
capital equipment manufacturers will gain a services, and energy industries will gain a deeper
strategic outlook on industry trends, including understanding of how semiconductor trends
advancements in design, manufacturing, and impact their industries. Insights into custom
packaging. The report also addresses critical
Capgemini Research Institute 2025
8
The semiconductor industry in the AI era
Snapshot of research methodology
1. Survey of 250 semiconductor industry organizations 2. Survey of 800 downstream organizations
6%
11%
3% 1%
Aerospace and defense
8% 11%
Automotive
Integrated device manufacturers 10%
Consumer electronics
10% Fabless design
6%
Energy
Outsourced semiconductor assembly and
test (OSAT) companies Financial services
48%
Foundries 10% High tech
10%
12% Material and subsystem companies Industrial equipment
Semiconductor capital equipment companies Medical devices/medical electronics
Electronic design automation (EDA) companies 10% Retail
13% Telecom
18%
13%
3. Twelve in-depth interviews with executives from the
semiconductor industry and downstream industries.
Capgemini Research Institute 2025
9
The semiconductor industry in the AI era
Definitions – A rich set of flexible and programmable specific purpose, such as image processing or inference,
acceleration engines that offload and improve at a lower monetary and resource cost than a general-
applications performance for AI and ML, purpose processor. ASICs enable ML and other typically
• Neural processing units (NPUs):1 NPU zero-trust security, telecommunications and high-cost functionality in situations where it would
architecture simulates the neural network of the storage, among others. otherwise be impractical. ASICs will not always be the
human brain. It processes large amounts of data appropriate solution, but are worth consideration.
simultaneously, performing trillions of operations • Graphics processing units (GPUs):3 A GPU is an
per second. It uses less power and is far more electronic circuit that can perform mathematical • Softwarization:6 Softwarization is the concept of
efficient than a CPU or GPU, while freeing these calculations at high speed. Computing tasks developing a more standardized, limited set of base
up for other tasks. Combining an NPU with such as graphics rendering, ML, and video chips that can be customized for various industries and
machine learning (ML) offers lightning-fast, high- editing require similar operations on a large solutions by reducing the number of unique hardware
bandwidth AI in real time. dataset. GPUs can perform the same operation designs and instead using software to provide industry-
on multiple data values simultaneously. This specific functionalities – essentially moving the “logic”
• Data processing units (DPUs):2 A DPU is a increases processing efficiency for many from silicon to software.
“system-on-a-chip” (SoC) that combines: compute-intensive tasks.
• Downstream industries: While nearly all industries
– An industry-standard, high-performance, • AI chips:4 AI chips are specialized computing rely on semiconductors for their products or services
software-programmable, multi-core hardware used in the development and and operations, the scope of this research includes
CPU, tightly coupled with the other SoC deployment of AI systems. AI chips are essential automotive, consumer electronics, retail, telecom,
components, for meeting the demand for greater processing aerospace and defense, high tech, medical devices/
– A high-performance network interface power, speed and efficiency. medical electronics, industrial equipment, financial
capable of parsing, processing, and efficiently services, and energy.
transferring data at line rate, or at least • Custom chips:5 Custom or application-specific
network speed, to GPUs and CPUs, integrated circuit (ASIC) chips are designed for a
Capgemini Research Institute 2025
1100
The semiconductor industry in the AI era
01
Organizations
anticipate surging
semiconductor demand
CCaappggeemmiinnii RReesseeaarrcchh IInnssttiittuuttee 22002245
1111
The semiconductor industry in the AI era
Semiconductors are the backbone of the digital world, two years. The main drivers of this increase are the growing
powering smartphones, Gen AI, computers, cars, usage of sophisticated electronics, the rise in electric and
satellites, and virtually every electronic device in use driverless cars, and the development of smart technologies
today. In 2023, nearly 1 trillion semiconductors – more and high-speed internet. Growth in data-driven applications
than 100 times the number of people on the planet – and demand for energy-efficient solutions are also helping to
were sold worldwide.7 Despite a cyclical market decline perpetuate this trend.
in the first half of 2023, worldwide sales recovered in
While the semiconductor industry is cyclical by nature, our
the second half of the year to $527 billion.8 By 2030, it is
research indicates that the semiconductor industry
anticipated that the semiconductor market size will
anticipates demand to increase by 15% by 2026, while
surpass $1 trillion.9
downstream organizations expect an increase of almost 30%
(see Figure 1). While we did not ask the respondents for their
Semiconductor
projections beyond 2026, extrapolating from the current
increase of 15% suggests that the market size could reach
manufacturers and approximately $930 billion by 2030, nearing the $1 trillion
estimate mentioned earlier. The semiconductor industry's
% downstream industries cautious outlook reflects a slow recovery in markets like
29
automotive despite strong AI-driven demand.
expect demand to rise
The semiconductor industry and downstream industries
expected increase in demand for anticipate significant growth in demand over the next
semiconductors by downstream
organizations, to the end of 2026
CCaappggeemmiinnii RReesseeaarrcchh IInnssttiittuuttee 22002245
1122
Gen AI at wTTohhreek :ss Seehmmaiipccioonnngdd tuuhccett oofurr tiinnuddreuu ssottfrr oyy riingn a ttnhhieez aAAtIIi oeenrraas
Figure 1.
Downstream industries estimate demand for semiconductors to increase at double the rate of the semiconductor industry’s
expectation
Expected semiconductor demand increase in two years to the end of 2026
29%
15%
Semiconductor industry Downstream industries
Source: Capgemini Research Institute, Semiconductor survey, November 2024, N = 190 semiconductor industry organizations
(includes Integrated Device Manufacturers (IDMs),fabless design firms and foundries), N = 800 downstream organizations.
Capgemini Research Institute 2025
13
The semiconductor industry in the AI era
Rapid technological progress is driving demand for more Figure 2.
powerful, efficient, and customized chips, prompting
Nearly three in five semiconductor organizations say that 5G (or other next-gen communication protocols) and Gen AI are
semiconductor manufacturers to invest heavily in R&D. Our
impacting their strategy
research indicates that 58% of semiconductor organizations
believe that 5G or other next-gen communication protocols
are impacting their strategies, while 56% believe that Gen
AI is a strong influence. 5G and other next-generation
Technology domains impacting semiconductor manufacturing strategy
communication protocols are foundational to enabling a wide
range of emerging technologies and markets, including IoT,
autonomous vehicles, AR/VR, and edge computing. They drive
demand for advanced semiconductor solutions that require
higher performance, energy efficiency, and integration.
58% 56%
51%
44% 42%
21%
17% 16%
5G or other next-gen Edge computing AR/VR Blockchain and
communication cryptocurrency
protocols
Generative AI Wearables Quantum Space
computing technology
Source: : Capgemini Research Institute, Semiconductor survey, November 2024, N = 250 semiconductor organizations.
Capgemini Research Institute 2025
14
The semiconductor industry in the AI era
The adoption of AI/Gen
compute engine, and the network subsystem enhances memory solutions such as Micron's HBM3E (one of many
throughput, whether in small or large clusters.” products on the market), which can optimize performance
AI is driving GPU/NPU and reduce CPU offload during AI processing, allowing faster
Qualcomm‘s NPU, designed specifically to take on AI
training and greater responsiveness to queries.12
workloads, is an essential enabler of on-device Gen AI
demand
capabilities. The NPU offers optimal performance, power, and In our research, 58% of semiconductor organizations
space efficiency to handle complex ML operations.10 ChatGPT mentioned they expect higher demand for NPUs to
The adoption of AI and Gen AI is driving demand for model exemplifies the transformative role of GPUs in AI. Leveraging accompany growth in Gen AI adoption, with 57% anticipating
training and inference capabilities, and data centers, while thousands of NVIDIA GPUs, the training and inference increased need for high-performance chips and 56% for
the growth of on-device AI applications further underscores processes for its large language model (LLM) demonstrate memory-intensive chips, signaling a shift toward advanced
the need for specialized semiconductor solutions. Subi the unparalleled efficiency and scalability GPUs bring to AI processing solutions.
Kengeri, VP of AI Systems Solutions at Applied Materials, workloads. This infrastructure supports Gen AI services for
says, "The AI era marks a new wave of growth for the over 100 million users, underscoring the critical contribution
semiconductor industry, propelled by the high returns on of GPU technology to cutting-edge semiconductor
investment generated by AI's economic value. For AI systems, applications in AI-driven innovation.11 A Senior Director at a
the key metric remains Total Cost of Ownership, while for US-based IDM explains, “One of the key advancements in the
Silicon, it is Perf/Watt/$.” semiconductor industry is performance enhancement through
parallel computing, exemplified by GPUs. As AI accelerators
To supply the demand rising from AI/ Gen AI, organizations evolve, various parallelization techniques, including expert
are ramping up their NPU, GPU, and memory capabilities. A parallelism, pipeline parallelism, tensor parallelism, and
Senior Director at a US-based IDM says, “We focus on context parallelism, are being employed. These optimization
optimizing the interplay of compute, memory, and network and architectural innovations are designed to boost overall
components to achieve system-level efficiency as customers throughput, whether for training or inference workloads.”
adopt AI to unlock its benefits. Computational efficiency
minimizes the time required for matrix multiplication, the Furthermore, today's Gen AI models require more data to
memory subsystem ensures data is readily available for the enhance outcomes and exploit new opportunities.
Inferencing LLMs such as ChatGPT benefits from advanced
Capgemini Research Institute 2025
15
The semiconductor industry in the AI era
Figure 3.
Due to Gen AI adoption, nearly three in five semiconductor organizations are seeing increased demand for NPUs, high-performance GPUs, and memory-intensive chips
Areas where organizations anticipate demand for their semiconductor products will be
impacted in the next two years by use of Gen AI applications
58% 57% 56%
40% 39%
16%
NPUs Memory-intensive Custom chips
chips
High performance DPUs Power and MEMS
GPUs chips
Source: Capgemini Research Institute, Semiconductor survey, November 2024, N = 250 semiconductor organizations.
Capgemini Research Institute 2025
16
The semiconductor industry in the AI era
Organizations project high
demand for AI chips and
custom silicon chips
The market is also witnessing a surge in diverse applications
for AI chips and custom-designed chips, underlining their
transformative potential. Intel and AWS announced a multi-
year co-investment in custom chip design that would
encompass Intel products and wafers.13 Sam Geha, EVP of IoT,
Compute and Wireless business at Infineon Technologies,
explains, “Just a few years ago, our role was simply to produce
chips, leaving it to customers to determine how to use them.
Today, however, we are expected to deliver customized
solutions. Software has emerged as a critical differentiator,
particularly as our chips have become increasingly complex.
Beyond general-purpose software, we now provide specialized
solutions for AI and edge AI, enabling customers to effectively
train and deploy models. Alternatively, we can offer services to
manage the training and deployment for them.”
Our research shows that most downstream organizations
foresee heightened demand in the next two years: 88%
anticipate a rise in AI chip needs, 81% foresee heightened
calls for custom chips, and 79% predict an upturn in demand
for memory-intensive chips. As figure 3 shows, according to
39% of semiconductor organizations, Gen AI is expected to
drive demand for custom chips in the next two years.
Capgemini Research Institute 2025
17
The semiconductor industry in the AI era
Figure 4.
Nearly four out of five downstream organizations anticipate increased demand for AI chips, custom silicon chips,
and memory-intensive chips over the next 12 months
Expected demand for chips
88%
81% 79% "The AI era marks a new wave
of growth for the semiconductor
industry, propelled by the high
returns on investment generated by
AI's economic value. For AI systems,
21% the key metric remains Total Cost
of Ownership, while for Silicon, it is
15% 14%
Perf/Watt/$.”
AI chips/chips designed Custom silicon chips Memory-intensive chips
for AI acceleration (custom ASIC) (e.g., HBM, GDDR6)
Subi Kengeri
Proportion of organizations expecting Average expected increase over
VP of AI Systems Solutions,
an increase in demand the next 12 months
Applied Materials
Source: Capgemini Research Institute, Semiconductor survey, November 2024, N = 800 downstream organizations.
Capgemini Research Institute 2025
18
The semiconductor industry in the AI era
Rise of custom silicon
Figure 5.
As AI workloads ramp up energy demand, electricity and Two in three downstream organizations are either already using custom silicon chips or are considering using them in their
infrastructure suppliers must adapt quickly to support the products
growing needs of data centers. Meanwhile, companies such
as NVIDIA with their commanding market share in AI chips,
Usage of custom silicon chips in products
continue to exerts significant influence over pricing. The
need for optimized performance, energy efficiency, and
differentiation in competitive markets such as automotive,
high-tech, and medical devices is driving the use of custom
18%
silicon. Custom chips allow organizations to better meet 56%
specific application demands, reduce costs at scale, and
leverage advances in AI and IoT, fueling widespread adoption
across industries. Yes, we use custom silicon chips in our products
Yes, we are considering using custom silicon chips in our products
According to our research, 56% of downstream organizations
16% No, we may consider using custom silicon chips in the future
are already using custom chips, while 11% are considering
the possibility. A large majority of industrial equipment No, we have no such plans
organizations (85%), and telecom organizations (82%)
currently rely on custom silicon chips. Tech giants Microsoft,
Amazon, and Meta are also developing in-house chips
tailored for AI inferencing. 11%
Source: Capgemini Research Institute, Semiconductor survey, November 2024, N = 800 downstream organizations.
Capgemini Research Institute 2025
1199
The semiconductor industry in the AI era
“Advanced platforms and software are no longer just enablers but critical differentiators in the semiconductor
industry, driving efficiency and scalability in design, manufacturing, and deployment. With the growing complexity of
AI, IoT, and edge computing applications, the ability to integrate domain-specific software with hardware accelerators
will define leadership. To stay competitive, semiconductor players must embrace co-optimization across the stack,
from chip architecture to application interfaces, ensuring they can meet the escalating demands of data-intensive,
low-latency markets.”
Jiani Zhang
EVP, Chief Software Officer,
Capgemini Engineering
Capgemini Research Institute 2025
2200
The semiconductor industry in the AI era
02
Downstream industries
express concerns over
semiconductor supply
CCaappggeemmiinnii RReesseeaarrcchh IInnssttiittuuttee 22002245
21
The semiconductor industry in the AI era
Over half of all
Figure 6.
More than three out of five organizations believe that geopolitical tensions and inadequate fab capacity impact the reliability
downstream organizations
of the semiconductor supply chain
are uncertain that the
semiconductor industry
Factors impacting the reliability of semiconductor supply chain
can cope in 2025
69%
As nations compete for control over vital technologies and 65%
resources, geopolitical tensions continue to impact the 52%
49%
global semiconductor supply chain. The flow of components, 46% 43%
materials, and completed semiconductor products has been
hindered by international trade disputes, export restrictions,
and tariffs. For example, Taiwan’s TSMC, the world's largest
semiconductor manufacturer, with a market share of about
55%, is the producer of the world’s most advanced chips.14
Consumer supply networks that rely on TSMC could be Geopolitical tensions Reliance on Push for sovereignty
seriously disrupted by any military escalation involving China small number of
semiconductor
and Taiwan. Deteriorating US-China ties have also given rise Fab capacity Pandemic Availability of
suppliers
to setbacks in the form of prohibitions on certain products natural resources
and more stringent controls. In 2022, the US introduced
export controls that restrict the People’s Republic of China’s
(PRC’s) ability to obtain advanced computing chips, develop
and maintain supercomputers, and manufacture The The Source: Capgemini Research Institute, Semiconductor survey, November 2024, N = 800 downstream organizations.
Capgemini Research Institute 2025
22
The semiconductor industry in the AI era
advanced semiconductors.15 These rules were revised in The COVID-19 pandemic exposed vulnerabilities in the global
2023, and in December 2024, the US Commerce Department semiconductor supply chain, disrupting logistics, demand,
further expanded the list of Chinese technology companies and production, while heightening concerns over product
subject to export controls and included many that make availability and rising costs. Our research shows that 49%
equipment used to make computer chips, chipmaking tools, of downstream organizations consider this impact to be
and software. China, as a response, announced that it is ongoing, while 47% had to curtail some product/feature
banning exports to the United States of gallium, germanium, launches due to chip shortages during the pandemic.
antimony and other key high-tech materials with potential
Downstream industries share this uncertainty. Around 59%
military applications.16 Our research indicates that 69% of
of downstream organizations believe that suppliers’ ability
downstream organizations believe that geopolitical tensions
to meet their semiconductor demand is an ongoing concern.
significantly impact the reliability of the semiconductor
Similarly, only around one-quarter (26%) feel that supply is
supply chain.
sufficient. This is particularly prominent among sectors such
Additionally, 65% of downstream organizations consider fab as A&D (14%) and organizations headquartered in Sweden
capacity to have a strong impact, while 52% feel that reliance (10%) and the United States (11%).
on a small number of semiconductor suppliers impacts their
reliance on the semiconductor supply chain.
%
69
Percentage of downstream
organizations that believe
geopolitical tensions impact the
reliability of the semiconductor
supply chain
Capgemini Research Institute 2025
23
The semiconductor industry in the AI era
GPU computing and AI/ML
Figure 7.
Fewer than three in ten downstream organizations believe chip supply is sufficient
acceleration are the most
relevant advancements
Downstream industries’ perception of chip supply/demand
for downstream
organizations
59%
Semiconductor technology breakthroughs have spurred
47%
innovation in consumer industries, breeding smarter, more
efficient products. AI/ML acceleration and GPU processing
26%
have the potential to revolutionize downstream operations.
GPUs outperform because they provide high throughput
and parallel processing, streamlining real-time inference and
model training, in particular for AI and ML applications.
Suppliers’ ability to meet During the COVID-19 The semiconductor
Our research suggests that 54% of downstream
our semiconductor demand pandemic, we had to industry is supplying chips
organizations (those that rely on fast data processing and
is an ongoing concern curtail some product/feature at a rate sufficient
AI-powered automation) believe that GPU computing and
for our organization launches due to chip shortages for our needs
AI/ ML accelerations are the most relevant advancements for
them. Alessandro Miranda, Senior Director of Radio Access
Network (RAN) Design and Optimization, at ZTE, explains,
“We need specialized hardware and architectures designed
Source: Capgemini Research Institute, Semiconductor survey, November 2024, N = 800 downstream organizations. to accelerate processing and optimize algorithms. Graphics
Capgemini Research Institute 2025
24
The semiconductor industry in the AI era
processing units (GPUs), for instance, which were originally Figure 8.
developed solely for rendering images and videos in video More than half of downstream organizations believe that advancements in GPU computing and AI/ML acceleration can bring
games, have now become the cornerstone of AI processing most value
due to their ability to handle parallel data processing
efficiently.“
Most relevant semiconductor advancements for downstream industries
Dell Technologies showcases the practical application of
GPU computing and AI/ML acceleration by integrating
NVIDIA's AI-ready GPUs, networking solutions, and tools
like AI Enterprise and Omniverse with its own hardware
GPU computing 54%
and expertise. This collaboration offers communications
service providers (CSPs) the tools needed to efficiently run AI and ML acceleration 54%
AI workloads across networks.17
5G/ next-gen communication technologies 49%
Advancements in GHz/watt 47%
NPU computing 27%
DPU 24%
Chiplets 24%
Wearables 19%
Next-generation memory chips (MRAM, ReRAM) 17%
Source: Capgemini Research Institute, Semiconductor survey, November 2024, N = 800 downstream organizations.
Capgemini Research Institute 2025
2255
The semiconductor industry in the AI era
“We need specialized hardware and architectures designed to accelerate processing and optimize algorithms. Graphics
processing units (GPUs), for instance, which were originally developed solely for rendering images and videos in video
games, have now become the cornerstone of AI processing due to their ability to handle parallel data processing
efficiently.”
Alessandro Miranda
Senior Director of Radio Access Network
(RAN) Design and Optimization, ZTE
Capgemini Research Institute 2025
26
The semiconductor industry in the AI era
Downstream
Figure 9.
Nearly half of downstream organizations are looking for enhanced customization and more comprehensive APIs and SDKs
organizations expect
enhanced customization,
Top ranked technology innovations/improvements desired by downstream organizations
more comprehensive APIs
and SDKs, and stronger
Enhanced customization 47%
security
Availability of more comprehensive APIs, SDKs, kits, boards, etc. 47%
Enhanced security 41%
The semiconductor industry has seen significant
enhancements in recent years, particularly in areas such as Higher computation capacity 36%
customization, security, APIs, software development kits AI support 29%
(SDKs), kits, and boards. These advancements are helping
businesses develop more tailored, secure, and efficient Lower power consumption 27%
products. Our research indicates that 47% of downstream Smaller size 25%
organizations have ranked enhanced customization and
Improved integration 25%
availability of more comprehensive APIs, SDKs, kits, boards,
etc., in the top three innovations that they look forward to in Drive customer experience 21%
the semiconductor industry, while 41% have ranked enhanced
security among the top three.
Companies such as Intel18 design custom ASICs for their
CPUs and GPUs, integrating both digital logic and analog Source: Capgemini Research Institute, Semiconductor survey, November 2024, N = 800 downstream organizations.
components to meet the specific performance and power
requirements of their products.
Capgemini Research Institute 2025
27
The semiconductor industry in the AI era
One in three downstream
Figure 10.
More than three out of ten downstream organizations are either already designing their chips in-house or are considering it
organizations is designing
chips in-house |
230 | capgemini | Top-tech-trends-2025_Infographic.pdf | AI is powering
all the top tech
trends of 2025
In 2025, AI is the biggest
current tech influencer
Top five trends for 2025
Industry by Capgemini experts Investors
executives (VCs and private equity)
Generative AI: From copilots to reasoning AI agents
AI & Gen AI in cybersecurity: New defenses, new threats
AI-driven robotics: Blurring the
Multimodal AI Multimodal AI
line between human and machine
The surge in AI is driving
AI in software engineering HealthTech
nuclear resurgence
New-generation supply chains: Agile, greener, and AI-assisted AI in software engineering
Generative AI: From copilots to
reasoning AI agents
At Capgemini, we believe the use of AI agents – autonomous AI systems
capable of independently handling end-to-end tasks and collaborating as
multi-agent systems – will be one of the biggest tech trends for 2025.
2023 - Early generative AI systems 2024-25 – AI agentic systems
<Task description>
interpretation
AI
Agent
• “One-size-fits-all”
tools
• Can’t handle abstraction
• Prone to hallucination and errors
• Does not scale well
• Not sustainable
Database Statistical Computational Network
• Autonomous task execution
• Improved efficiency
• But security challenges in autonomy
Example: ChatGPT by OpenAI, an AI chatbot for Example: DSO Go by Bayer, an AI agent designed to
personalized assistance, customer support, and combine the strengths of a guided conversation in a
content generation. pre-defined pattern with model-based on generative A.
of executives of investors (VCs) that follow the AI
70% and 85% and data tech domain pick
AI agents
as a top three
impactful trend for 2025.
AI & Gen AI in cybersecurity:
New defenses, new threats
While Gen AI offers transformative potential to enhance security measures, malicious
actors have quickly recognized its capacity for evil, employing it for sophisticated
attacks that target both human vulnerabilities and machine defenses
Industry executives ranked AI & Gen AI in cybersecurity as the top tech trend of 2025
and investors ranked it third overall
Governments across the world are responding to threats with stricter laws:
In August 2024, the Singapore government launched In 2024, the EU promulgated its Cyber Resilience Act
Operational Technology (OT) Cybersecurity Masterplan (CRA), requiring manufacturers to embed enhanced
to enhance the security and resilience industrial control cybersecurity measures across a broad range of everyday
systems and its technologies. hardware and software products.
AI-driven robotics: Blurring the line
between human and machine
LLMs are transforming robotic capabilities and have accelerated the
development of next-gen robotics to handle complex, interconnected tasks,
enhancing operational efficiency, personalizing customer experiences,
and improving decision-making across industries.
MMiiccrroossoofftt iinnaauugguurraatteedd iittss fifirrsstt AAII aanndd
rroobboottiiccss RR&&DD cceenntteerr iinn TTookkyyoo,, JJaappaann..
ooff iinnvveessttoorrss (rVanCkse adnd private equity) ranked
NNVVIIDDIIAA iiss ppllaannnniinngg ttoo llaauunncchh iittss
AAII--ppoowweerreedd rroobboottiiccss
8899%% AAII--ppoowweerreedd hhuummaannooiidd rroobboott JJeettssoonn TThhoorr
iinn tthhee fifirrsstt hhaallff ooff 22002255..
aammoonngg tthhee ttoopp tthhrreeee ttrreennddss foofr 2 2002255 i nin t thhee
iinndduussttrryy aanndd eennggiinneeeerriinngg ddoommaaiinn..
OOppeennAAII--bbaacckkeedd 11XX TTeecchhnnoollooggiieess
iinnttrroodduucceedd tthhee NNEEOO BBeettaa AAII hhuummaannooiidd
rroobboott ffoorr hhoouusseehhoolldd cchhoorreess..
The surge in AI is driving
nuclear resurgence
The energy sector is transforming at an unprecedented pace, driven by mounting pressure to respond to the
climate crisis and supported by innovation across sectors, from renewables and biofuels to low-carbon
hydrogen and beyond. Nuclear energy stands out as a focal point for 2025
SMRs and AMRs will lead the way for new nuclear as they are poised for
rapid industrialization
Small Modular Reactors (SMR’s) Advanced modular reactors (AMR’s)
Offer safer, scalable, and Also known as Generation IV
cost-effective alternatives to reactors, use innovative fuels,
traditional reactors, using an coolants, and technology to
established fuel supply chain generate low-carbon electricity,
without needing ultra-heavy and are intrinsically safe,
forging capacity. compact, and portable.
Key activities in this space:
Google announced plans to Meta has announced a planned Amazon has signed agreements
purchase electricity RFI for 1-4 GW of new nuclear. to support the development of
generated using SMRs. nuclear energy projects
New-generation supply chain:
Agile, greener, and AI-assisted
By harnessing cutting-edge technologies such as digital twin and AI-powered algorithms in their supply chains,
businesses can simulate various scenarios to optimize operations for agility and resilience. Sustainable supply
chains and product passports enable transparency and accountability in sourcing and production
General Motors (GM) integrates
sustainability into its supply chain
of CTOs, heads of innovation/CIOs, and through the BrightDrop platform for EV
heads of R&D, engineering, and product logistics, sustainable sourcing practices,
85%
agree that ‘new-generation supply chain’ and advancements in EV technologies.
is among the top three technology
trends for 2025. Pfizer uses AI to optimize its supply chain,
enhance drug development, clinical trials,
and vaccine distribution
By 2030, several groundbreaking trends
are poised to revolutionize our lives.
Programmable Quantum Genome Artificial general
Hyperconnectivity
new materials computing therapy intelligence
New materials Quantum computing This involves AGI can understand, Offers seamless
engineered to uses the unique modifying an learn, and apply combination of
change their properties of tiny individual’s genetic knowledge across a terrestrial and
properties, such as particles to solve material to treat or wide range of tasks non-terrestrial
shape or color, at problems much prevent disease, at a human level, networks to facilitate
molecular assembly faster than classical potentially offering enabling and communication and
level in response to computers can, cures for genetic allowing machines to collaboration on a
external stimuli or helping with tasks disorders and potentially perform global scale,
programming such as encryption, personalized many intellectual enhancing
optimization, and medicine tailored to tasks that a human connectivity and
simulations an individual could do integration across
platforms and devices
Download report Subscribe to our research
the Capgemini Group. Copyright © 2025 Capgemini. All rights reserved. |
231 | capgemini | INVENT-PoV_GenAI-Mobility_EN.pdf | Generative AI for
sustainable mobility
Introduction 3
A few reminders about generative AI 4
Our vision of GenAI use cases
in relation to mobility 7
Generative AI for operational efficiency 8
Revolutionizing the traveler experience 10
Interview client 12
Conclusion 14
Our expertise 15
Auteurs & contributeurs 17
2
Introduction
Transport decarbonization, which is essential to the fight against
climate change, will not only require the electrification of vehicles,
but also a substantial shift in traffic towards collective, shared and/or
active mobility. But facilitating and accelerating this modal shift is
not just a question of developing networks or increasing financial
incentives: travelers also need to be attracted by an offer that is
clear, appropriate, practical and economical. From the moment they
choose their route to the moment they arrive at their destination,
their experience must be as personalized and contextualized as
possible, ensuring maximum fluidity before, during and after their
journey. This seamless process is likely to create a positive perception
of the transport modes used, the brand and its offering, and thus a
strong motivation to repeat
the experience and
recommend it.
70%
1 Attracting and retaining new
customers so that they abandon
their private car is one of the
main challenges facing public
transport authorities (PTA) and
operators (PTO) today.
Technological innovations, in
of respondents to a survey of the
particular generative AI, will
Autonomy community of experts
have a decisive role to play in
believe that LLMs will help meeting this challenge. We are
reduce the number of privately convinced that this form of
owned cars and increase modal artificial intelligence,
shift by enabling mobility popularized in 2022 by ChatGPT,
operators to offer more is capable of removing a number
of major obstacles to the
convenient solutions.
accessibility and attractiveness
of sustainable mobility. This is
also the opinion of a majority of
the members of the Autonomy
community, 70% of whom
believe that it will help to reduce the number of privately owned cars and
increase modal shift by enabling mobility operators to offer more
practical solutions (although 33% see it more as a fad than a genuine
lever for change)1.
As suggested by the first use cases that have been implemented or are
still being studied, and as we will see in this document, generative AI will
have a very significant impact on two key areas: operational efficiency
and the traveler experience, with, as a result, significant benefits for
mobility companies, their employees, their customers… and the
environment.
1 Survey addressed to members of the Autonomy community in January 2024.
3
A few reminders about
generative AI
Generative AI, or GenAI, refers to a form of artificial intelligence capable, as its
name suggests, of generating text, images, video, audio, or a mixture of these
media.
It is based on statistical models trained on an extremely large body of data,
including computer code, equivalent to several billion pages of documents, images
and thousands of hours of video or audio. On a basis of a written query (prompt),
GenAI is able to generate original and unique content, which is similar – but not
identical – to the content it has assimilated.
Text / Speech Code Image / Video
Summarising a text/
Autofill code Increasing resolution
a conversation
Modifying/styling
Translating Translating a code
an image
Analysing a corpus/ Optimising an Translating an image
a conversation existing code into a photo
understanding a question/ Documenting an Detecting an anomaly
an instruction existing code in an image/video
Re(writing) according Writing code based Creating a 2D image based
to instructions on instructions on instructions
Describing an Testing an existing code
Transforming 2D into 3D
image/video by units
Answering questions Orchestrating
in a factual manner a workflow
Solving logical/
mathematical problems
Mature Emerging Arising
Figure 1-Maturity of fields of application depending on the nature of the data supplied to GenAI.
4
Generative AI can be applied to any
type of data (text, image, video,
sound, code, etc.) both as input and
output to the query, so the fields of
application are theoretically
innumerable: creating articles,
personalizing content, producing
computer code, generating data sets,
correcting images, 3D animation, and
many more.
Large Language Models (LLMs), a
sub-category of GenAI specializing in
language, are now the most mature,
robust and widely used GenAIs. They
are first trained to predict the next
word in a given sequence of words.
They are then specialized to perform
tasks other than their primary function.
In this way, they excel at understanding
and producing text across a fairly wide
range of applications: classification,
research, synthesis, conversation,
translation, writing, etc.
On the other hand, because of their
probabilistic nature, and despite their
often impressive performance, LLMs
have intrinsic limitations that should
not be overlooked:
• Bias: the model depends on the training
data set from which it draws its
inspiration. It mechanically reproduces
any weaknesses, such as biases,
stereotypes, prejudices, errors,
obsolescence, etc.
• Reliability: as models predict the next
word in a sequence based on the
previous words, they can generate
information that seems logical in
context, but which is not actually true or
accurate. There is therefore never
absolute certainty that the answer is
accurate, relevant or appropriate. This
can even lead to gross factual errors
called hallucinations.
5
To overcome these limitations as
much as possible, it is essential to
implement techniques to control
the model according to the
intended use case, and to set up a
certain number of safeguards:
ensuring the quality of the data;
improving the relevance, accuracy
and way of processing the query
(prompt-engineering); forcing the
model to respond only on the basis
of sources provided and traceable in
the response (context-engineering);
and possibly – although, in practice,
much more complex – adapting the
model (fine-tuning). In all cases,
automated and human controls
need to be put in place before the
solutions are deployed on a large
scale.
All this means that LLM
performance can be significantly
improved for the application in
question. Even so, they will not be
able to perform tasks completely
autonomously, without user
supervision. The user will have to
treat them as an assistant,
constantly keeping a critical eye on
them.
Finally, whatever the project is, we
must not forget to consider the
issues of security, compliance and
environmental footprint that
generative AI and LLMs sometimes
raise so acutely.
6
Our vision of GenAI use cases
in relation to mobility
Operational efficiency Passenger experience
Technical documentation
Situational analysis Customer feedback
analysis
Customer complaint
Documentaion production Conversational tools
management
Personalising and
Knowledge capitalisation
Instant translation contextualising passenger
and transfer
journeys and information
Support functions
Synthesis of financial or Audits and verification of Follow-up of
HR documentation procedures medical visits
Figure 2 - Major GenAI use cases in the mobility sector.
7
Generative AI for
operational efficiency
In the mobility and transport sector, operational efficiency is the sinews of
war. In all sector businesses, improving efficiency in the field means helping to
provide a service that is less costly, more reliable, more responsive and more
resilient in the face of unforeseen events, and that offers travelers a more
satisfactory experience. In this area, generative AI is a powerful lever for
improvement, as numerous use cases already illustrate.
Analysis of technical and legal
documentation
The mobility and transport sector operates and orchestrates equipment of
extremely varied nature, technology and era. It is also a particularly regulated
sector, governed by complex provisions and standards that frequently evolve. As a
result, the technical and legal documentation is abundant, fluctuating and
heterogeneous and needs to be constantly taken into account. Particularly
well-suited to enriched and personalized documentary research, generative AI can
bring considerable time savings and improve accuracy when it comes to obtaining
the right information, so helping you to make the right decision in real time.
Production of reporting
documents
In all professions, employees produce numerous reports to provide information on
their activity, share their experiences and alert others to any difficulties or
incidents they may have encountered. Generative AI can relieve this valuable but
time-consuming task, and improve its quality, by assisting employees with data
entry, pointing out missing information and even, in the future, directly
transcribing voice recordings. It can also facilitate the use of these reports by
making unsuspected comparisons, detecting imperceptible similarities, and
proposing new categorizations. In this way, generative AI will be able to carry out
root cause analyses and suggest ways of resolving problems much more quickly.
8
Situational analysis
The new generative AIs are multimodal, meaning that they can process all kinds of media
simultaneously, text and images for example. They have this fine-tuned ability to describe
images and therefore to depict a context. Integrated into surveillance systems, they can be
used to detect problem situations that require both object or person detection and context
analysis. For example, they can be used to detect high-risk situations (such as illness or
aggression, or crowds of people) and assist a security guard to intervene more quickly. It is
also possible to characterize damage or quality defects, warn of the presence of obstacles
on or near tracks (intruders, vegetation, landslides, etc.), check the cleanliness of premises
and equipment to specify cleaning operations, or certify compliance with safety procedures.
These capabilities could be combined with traditional computer vision-type AI for greater
efficiency and to limit the large energy and environmental impact of generative AI.
Feedback from experience
SNCF Réseau, National Company of the French Railways,
improves document-retrieval for its customers
With the opening of rail traffic to competition, SNCF Réseau will
be approached by a growing number of operators to handle
regulatory technical issues. To provide these customers with a
fast and relevant response, Capgemini has helped SNCF Réseau
to develop the demonstrator of a search assistant based on
generative AI. This solution takes the form of a conversational
agent, whose ergonomics and path have been optimized to offer
users a simple, fluid and personalized experience. Queried in
natural language via this interface, possibly in several languages,
the generative AI engine, owned by Capgemini, draws its
information from a database of technical documents that has
been compiled and qualified in advance. One of the special
features of this model is that it displays the sources that support
the users’ response, which reinforces the users’ confidence and
enables them, if necessary, to deepen, validate or share their
research. Finally, metrics have been put in place to monitor the
model’s performance and ensure that it durably meets the
expectations of SNCF Réseau’s customers.
« The implementation of an initial solution based on generative AI
will enable our sales forces, and soon our customers, to save time,
whether they are railway companies or public transport
authorities. The solution we have developed will enable them to
find all the information they need in the regulatory railway
documentation (the Network Reference Document) in just one
minute. Generative AI opens up new creative possibilities for
simplifying the day-to-day lives of actors involved in the rail
industry. »
Olivia Fischer, Head of Markets, Offer and Customer
Experience at SNCF Réseau
9
Revolutionizing the
traveler experience
Transport may be collective, but the experience is individual. All travelers have
their own itinerary, their own needs, their own control of the offer and the
tools at their disposal. And conditions are constantly changing, so no two
itineraries are ever the same. So how can we offer everyone a satisfactory
experience when no two travelers expect exactly the same information, at the
same time and in the same form? Generative AI can help solve this complex
equation of the traveler experience. In push or pull mode, it can provide each
customer with precisely the information they need, when they need it, and on the
channel that suits them best. In this way, it can help to deliver the personalized,
contextualized and optimized experience that is likely to convince as many
passengers as possible to opt for greener mobility.
Conversational tools
Generative AI makes it possible to set up conversational agents that are much
more advanced than they are today, and capable of communicating in a language
comparable to the language of humans. At a time when the search for information
(a fare, a timetable, an itinerary, a possible connection, etc.) occupies a
predominant place in the customer journey, the possibility of a conversational
interface will be a huge advantage. Without necessarily having to change the
underlying algorithms of existing chatbots, this will enable passengers to easily
express their needs, constraints and criteria, without having to go through a
multitude of screens and filters. The tool will also be able to add personalized
recommendations to the response, depending on the profile (foreign visitor,
person with reduced mobility, cyclist, etc.) and suggestions (directing passengers
towards more sustainable solutions, offering custom subscription packages, etc.).
70% of the members of the Autonomy community believe that conversational
tools will make it easier to take account of the diversity of needs, thereby creating
more inclusive mobility.
70% of respondents to a survey of Autonomy
2
community of experts believe that
conversational tools will make it possible
to take better account of the diversity of
needs and create more inclusive mobility.
2 Survey addressed to members of the Autonomy community in January 2024.
10
Customer feedback
Listening to and taking into account the “voice of the customer” is essential for identifying
the problems encountered, the needs and the expectations of travelers, and therefore
improving the experience on an ongoing basis. Today, however, this is a fairly laborious
process, both for passengers who want to express their opinions and for the staff
responsible for processing them. Generative AI can considerably help both: the travelers, by
enabling them to express themselves in natural language, or even orally, and the teams, by
automatically sorting, categorizing, and qualifying the opinions gathered. AI is capable of
identifying the key points despite the diversity of formulations, and even of detecting irony.
It can then offer an immediate, targeted and personalized response to each individual,
create regular summary reports to measure and monitor customer satisfaction; and finally –
in the longer term – detect and escalate similar and recurring problems.
Instant translation
Thanks to its translation capabilities, generative AI can remove the language barrier which,
for foreign tourists, is often the main obstacle to a positive transport experience. In
anticipation of an influx of visitors of all nationalities at the Paris Olympic and Paralympic
Games, Paris public transport operator RATP and National Company of the French Railways
SNCF are preparing several systems. One of these systems, which is currently being tested,
will make it possible to instantly translate the audio announcements broadcast in stations
into several languages and then to pronounce them using a synthetic voice. Another solution
will provide agents with a specialized instant translation application: the queries, expressed
by passengers in their own language, will first be translated into French for the agent,
who will be able to formulate their response in French before it is in turn translated into
the passenger’s language. This will result in smoother and more efficient exchanges,
for both staff and customers, and will improve the traveler/visitor experience.
1111
Customer interview
Mathilde Villeneuve
Project Director at the RATP
Data Factory
Can you tell us about RATP’s approach to
generative AI?
RATP is taking a pragmatic, value-driven approach by integrating
generative AI into its sustainable mobility strategy. This initiative
explores two major areas: improving the quality of working life and
efficiency of its agents, and developing solutions tailored to its
business units and strategic challenges. Generative AI solutions are
becoming an essential pillar of RATP’s toolbox, aimed at accelerating
the exploitation of data and improving operational efficiency..
Which ones have you identified?
Several priority themes have already been identified, such as increased
control of mobility needs in order to plan transport solutions,
improving operational performance and service rendered to
passengers (e.g. incident analysis, chatbot assisting agents in stations),
and improving the quality of working life for agents (e.g. control of
purchasing processes in the context of public procurement contracts).
RATP is industrializing a first use case: a virtual assistant for station
agents. Station agents are the first point of contact for passengers and
station guards. They need to be particularly versatile in order to
answer all the questions asked by travelers, ensure compliance with rail
safety standards and implement the requirements of public transport
1122
authorities. This system enables agents to be more efficient in carrying
out their daily tasks, such as providing clear and precise information to
passengers on fares, refunds and access procedures. As well as
optimizing work processes and improving service quality, this virtual
assistant saves time for RATP’s 5,500 agents.
How is RATP taking up this solution?
RATP is adopting an approach based on creating value for users by
rapidly industrializing practical applications for its business units. It is
focused on supporting business units in identifying relevant use cases,
and on involving and training users from the earliest stages of
development. This user-oriented approach also means that the risks
and limitations associated with the use of generative AI (impact on
employment, algorithmic biases, organizational changes, etc.) can be
accelerated and taken into account, while keeping the human element
at the center of the loop.
RATP, aware that the democratization of generative AI within a large
company requires an iterative and collaborative approach, is building
its approach by mobilizing all the necessary skills internally and via its
partners (data/AI expertise, cloud provider, etc.). To achieve its
objectives, RATP relies on its data platforms/AI. It is gradually putting in
place a technical basis that will enable it to control and rationalize
practices and solutions, secure the data and company’s know-how, and
guarantee the ethical and reliable use of these technologies
(transparency of algorithms, evaluation of models, governance of
technologies, etc.).
1133
Conclusion
Even as the transport actors imagine and develop
their first applications of generative AI, the
extremely rapid progress of the technology will very
quickly enable them to envisage other use cases,
with an even greater impact. For example, the ability
to bring more energy- and data-efficient language
models to mobile terminals (smartphones, tablets,
etc.) will improve the performance and experience of
applications, for example for train crews or network
maintenance agents. Another expected advance is
that multimodal models, capable of processing text,
images and/or sound at the same time, will open a
vast field of possibilities, enabling, for example, the
creation of composite reports combining photos and
audio commentary.
However, whether for these applications of the future
or those already under development, we must always
bear in mind the limitations of generative AI in general
and LLMs in particular. The purpose of these tools is to
assist, help and accelerate, but not to carry out tasks or
IMAGE make decisions without human approval. The users,
whether agents or travelers, must therefore be made
aware of the fact that they will need to systematically
check the information communicated to them. When
they are implemented, technological solutions must
always be accompanied by a framework for use and
appropriate communication. Despite these precautions,
the gains in terms of time, ergonomics and experience
are usually considerable, at least if the product has been
properly designed and optimized by specialists.
The mobility and transport sector is still in its early
stages in terms of the use of generative AI, but the first
use cases suggest its immense potential for improving
both operational efficiency and the passenger
experience. Progress in these two key areas will help to
make green mobility more pleasant, more accessible,
and more attractive, thereby encouraging modal shift.
Generative AI is emerging as a major instrument in the
transformation towards sustainable mobility, and public
transport operators and authorities alike must seize its
tremendous potential without further delay.
14
Finally, of course, all these
benefits depend on users
adopting the solutions.
For example, there is no
43%
guarantee that customers will
accept this technology, even if it
meets their needs.
Will LLMs
The experts in the Autonomy
community are very divided on replace mobility
the subject: 57% of them think applications?
that travelers will be searching
for their itinerary using an LLM
in the future, compared with
43% who believe that 57%
traditional applications still
have a bright future ahead of
them.
Major educational efforts are
therefore essential if generative
AI is to become the powerful
Yes No
accelerator of sustainable
mobility envisaged in this report. AI will enable us to identify We will still use mobility
the best itinerary by having apps to choose and book
a spoken conversation. an itinerary.
Figure 3 - Survey addressed to members of the Autonomy community in January 2024.
Our expertise
Within Capgemini Invent, our R&I Lab is organized
around research and innovation programs applied to
our customers’ challenges, including AI and mobility,
drawing on expertise and best practices in research
and innovation brought by Quantmetry, an
acquisition by Capgemini Invent back in 2023.
15
Capgemini authors
Mehdi Essaidi,
Vice-President Smart Mobility,
Capgemini Invent
Lucile Ramackers,
Senior Manager Sustainable Mobility,
Capgemini Invent
Toscane Berberian,
Senior Mobility Consultant,
Capgemini Invent
Alexandre Lapene,
Data Science Director,
Generative AI Specialist,
Capgemini Invent
Capgemini sponsor
Alex Marandon,
Vice-President & Global Head of Invent
Generative AI Accelerator,
Capgemini Invent
Capgemini contributors
Philippe Cordier,
Chief Data Scientist and Vice-President of
Artificial Intelligence and Data Engineering,
Capgemini Invent
Farès Goucha,
Rail Industry Director,
Capgemini Invent
Sophie Poulin,
Automotive, Mobility, Transport & Travel
Client Director,
frog part of Capgemini Invent
Hugo Cascarigny,
Vice-President, Data & Analytics Intelligent
Industry,
Capgemini Invent
Autonomy contributor
Ross Douglas,
CEO,
Autonomy Paris
17
About Capgemini Invent
As the digital innovation, design and transformation brand of the Capgemini Group,
Capgemini Invent enables CxOs to envision and shape the future of their businesses.
Located in over 30 studios and more than 60 offices around the world, it comprises
a 12,500+ strong team of strategists, data scientists, product and experience
designers, brand experts and technologists who develop new digital services,
products, experiences and business models for sustainable growth.
Capgemini Invent is an integral part of Capgemini, a global business and technology
transformation partner, helping organizations to accelerate their dual transition to
a digital and sustainable world, while creating tangible impact for enterprises and
society. It is a responsible and diverse group of 340,000 team members in more than
50 countries. With its strong over 55-year heritage, Capgemini is trusted by its
clients to unlock the value of technology to address the entire breadth of their
business needs. It delivers end-to-end services and solutions leveraging strengths
from strategy and design to engineering, all fueled by its market leading capabilities
in AI, cloud and data, combined with its deep industry expertise and partner
ecosystem. The Group reported 2023 global revenues of €22.5 billion.
Get the future you want
Visit us at www.capgemini.com/invent
About Autonomy
Autonomy, based in Paris produces an annual Expo, Global Decarbonization Expo
which is focused on Commercial and Industrial solar, battery storage autonomous
and electric vehicles. In addition to the expo, Autonomy produces bespoke summits
on specific subjects in key cities and content on decarbonizing the economy.
Visit us at www.autonomy.paris
Copyright © 2024 Capgemini. All rights reserved. |
Subsets and Splits