Unnamed: 0
int64 0
354
| Company
stringclasses 13
values | Fichier
stringlengths 7
123
| Content
stringlengths 1.77k
32.8k
|
|---|---|---|---|
232 | capgemini | 2024_11_19_-Capgemini_AI-and-Generative-AI-in-Cybersecurity-report.pdf | Press contact:
Florence Lievre
Tel.: +33 1 47 54 50 71
Email: [email protected]
AI and Gen AI are set to transform cybersecurity for most organizations
While Gen AI heightens vulnerabilities, more than half of organizations also anticipate faster threat
detection and increased accuracy through its use
Paris, November 19, 2024 – The Capgemini Research Institute’s new report, “New defenses, new
threats: What AI and Gen AI bring to cybersecurity”, published today, suggests that while new
cybersecurity risks are emerging, due to the proliferation of AI and generative AI (Gen AI), these
technologies represent a transformative shift in reinforcing cyber-defense strategies for the long
term to predict, detect, and respond to threats. Two thirds of organizations are now prioritizing
AI within their security operations.
According to the report, while AI is considered by organizations as a strategic technology to strengthen their
security strategies, the increased adoption of Gen AI across various industries1 brings heightened
vulnerability. Gen AI introduces three major risk areas for organizations: more sophisticated attacks with
more adversaries, the expansion of the cyber-attack surface, and a growth in vulnerabilities in the entire
lifecycle of custom Gen AI solutions. These risks are also compounded by the misuse of AI and Gen AI by
employees which can significantly increase the risk of data leakage.
Two in three organizations are wary of increased exposure to threats
Almost all organizations surveyed (97%) say they have encountered breaches or security issues related to
the use of Gen AI in the past year. Gen AI also brings additional risks, including hallucinations, biased,
harmful, or inappropriate content generation, and prompt injection attacks2. Two in three organizations
(67%) are worried about data poisoning and the possible leakage of sensitive data through the training
datasets used to train Gen AI models.
Moreover, Gen AI's ability to generate highly realistic synthetic content is posing additional risks: more than
two in five organizations surveyed (43%) said they have suffered financial losses arising from the use of
deepfakes.
Nearly 6 in 10 believe they need to increase their cybersecurity budget to bolster their defenses
consequently.
AI and Gen AI are paramount for detecting and responding to attacks
Surveying 1,000 organizations3 that have either considered AI for cybersecurity or are already using it, the
report finds that most rely on AI to strengthen their data, application and cloud security due to the
technology’s ability to rapidly analyze vast amounts of data, identify patterns and predict potential breaches.
1 Nearly one-quarter (24%) of organizations have enabled Gen AI capabilities in some or most of their functions and locations - Capgemini
Research Institute, “Harnessing the value of generative AI 2nd edition: Top use cases across sectors,” July 2024.
2 This involves using malicious inputs to manipulate AI and Gen AI models, compromising their integrity.
3 1,000 organizations across 12 sectors and 13 countries in Asia Pacific, Europe, and North America, with annual revenues of $1 billion and
over.
Capgemini Press Release
More than 60% of them reported a reduction of at least 5%, in their time-to-detect, and nearly 40% said
their remediation time fell by 5% or more after implementing AI in their security operations centers (SOCs).
Three in five organizations surveyed (61%) believe AI to be essential to effective threat response, enabling
them to implement proactive security strategies against increasingly sophisticated threat actors. In addition,
the same proportion of respondents foresee Gen AI strengthening proactive defense strategies in the long
term, anticipating faster threat detection. Over half of them believe also that the technology will empower
cybersecurity analysts to concentrate more on strategy for combating complex threats.
“The use of AI and Gen AI has so far proved to be a double-edged sword. While it introduces unprecedented
risks, organizations are increasingly relying on AI for faster and more accurate detection of cyber incidents.
AI and Gen AI provide security teams with powerful new tools to mitigate these incidents and transform
their defense strategies. To ensure they represent a net advantage in the face of evolving threat
sophistication, organizations must maintain and prioritize continuous monitoring of the security landscape,
build the necessary data management infrastructure, frameworks and ethical guidelines for AI adoption, and
establish robust employee training and awareness programs,” said Marco Pereira, Global Head
Cybersecurity, Cloud Infrastructure Services, Capgemini.
Methodology
The Capgemini Research Institute surveyed 1,000 organizations that have either considered AI for
cybersecurity or are already using it, across 12 sectors and 13 countries in Asia Pacific, Europe, and North
America. They have annual revenues of $1 billion and over. The global survey took place in May 2024.
Organizations surveyed represent a diverse range of sectors including automotive; consumer products;
retail; banking; insurance; telecom; energy and utilities; aerospace and defense; high-tech; industrial
equipment manufacturing; pharma and healthcare and public sector.
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 fuelled 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 Press Release |
233 | capgemini | 17_10_Generative-AI-at-Work-CRI-Report_Group-press-release-2.pdf | Press contact:
Mollie Mellows
Tel.: + 44 (0) 7342 709384
E-mail: [email protected]
Generative AI expected to accelerate entry-level career progress across
industries
• Employees believe generative AI will facilitate a third (32%) of entry level tasks over the
next 12 months.
• 6 in 10 managers and most employees (71%) expect entry-level roles across functions to
evolve from creation to reviewing generative AI outputs, over the next 3 years.
• Over three-quarters (78%) of leaders and managers predict generative AI will augment
their problem-solving and decision-making in the next three years, and over half think
manager-level positions will evolve towards specialization.
Paris, October 17, 2024 – The Capgemini Research Institute’s new report on generative AI (Gen
AI) in management, ‘Gen AI at work: Shaping the future of organizations’, published today,
suggests that Gen AI could have a positive impact on early-stage careers. In the longer-term,
the report finds that Gen AI has the potential to create new job roles, transform organizational
structures, drive human-AI fusion teams, and make managerial roles more specialist. However,
adoption remains low and nascent. The report finds most employees lack the training they need
to develop Gen AI skills.
Whilst the impact of Gen AI on careers has been hotly debated, this new research finds the majority of
business leaders believe that entry level roles could become more autonomous and evolve into frontline
managerial roles within the next three years. With this in mind, the proportion of managers in teams across
functions could expand from 44% to 53% in the next three years; only 18% of leaders and managers believe
that Gen AI will reduce middle management.
Employees think that, over the next 12 months, generative AI tools could lead to an average time saving of
18% for entry-level workers, implying there could be significant productivity improvements for junior
employees. However, the cost of the Gen AI tool must also be taken into account, cites the report.
Furthermore, 81% of leaders and managers expect new roles such as data curators, AI ethics specialists
and algorithm trainers to emerge at the entry level.
“Generative AI tools are becoming more adept at assisting with complex managerial tasks, which could
challenge the status quo of organizational structure and ways of working,” said Roshan Gya, CEO of
Capgemini Invent and member of the Group Executive Committee. “Generative AI has the potential to shift
from a co-pilot to a co-thinker, capable of strategic collaboration, adding new perspectives and challenging
assumptions. This shift could unlock significant value when tailored to specific business use cases but is
dependent on several factors, including organizations prioritizing building the skills and readiness of
employees, taking proactive steps around talent acquisition and development.”
Potential to redefine management but still a significant gap on actual usage
The report finds that Gen AI is transitioning the view of future leadership and managerial roles toward
becoming more strategic, focusing on decision making and fostering innovation. In fact, many managers
and leaders currently believe that Gen AI tools could act as co-thinkers for them. 65% of the leaders and
Capgemini Press Release
managers surveyed see high potential in Gen AI for complex strategic tasks, and more than half of leaders
believe managers will play a critical role as catalysts of Gen AI-driven change. The technology could also
save leaders and managers up to seven hours each week, with nearly 8 in 10 leaders believing that Gen AI
will positively impact their productivity in the next 12 months.
Gen AI has the potential to amplify the strategic scope of leaderships roles. Currently, managers spend more
than one-third of their time on administrative tasks. However, AI’s ability to automate much of this work
provides opportunities to focus on strategic-planning and problem-solving tasks. In the next three years,
over three-quarters (78%) of leaders and managers expect Gen AI to augment their problem-solving and
decision-making, and over half believe manager-level positions will evolve towards specialization. 57% of
leaders at organizations advanced in their Gen AI implementation already see their roles becoming more
strategic.
While adoption of Gen AI in management has good potential, there is a significant gap between potential
and actual usage. Although 97% of leaders and managers say that they have experimented with Gen AI
tools, only 15% use Gen AI tools at least once a day in their work.
Organizational structures need to transform to enable cohesive human-AI collaboration
For nearly half (46%) of teams, AI is used simply as a tool to enhance existing capabilities and workflows.
However, human-machine partnerships are starting to be embraced. One in three teams are currently using
AI as a ‘team member’, for example by enhancing human performance or using AI agents to complete
predefined tasks without human intervention. According to the research, today AI is used as a supervisor in
only 1% of teams i.e., it is directing, allocating, or prioritizing work for humans. Yet, in the next 12 months,
13% of teams expect to use AI in this role. In an AI-led environment, human judgment is increasingly
important, and the majority of leaders, managers and employees in the research acknowledge this.
Training and managerial guidance required to secure the future of Gen AI at work
Despite the potential of Gen AI to boost productivity across job functions, adoption remains nascent. While
almost two-thirds (64%) of workers already use Gen AI tools for their work, only 20% of employees use
Gen AI tools daily.
Employees also lack proficiency in key skills, with only 16% believing they are getting the support they need
to develop Gen AI skills. Only 13% of employees say they are well-versed in machine conversational skills;
only a third say they can manage Gen AI systemic risks; and less than half claim to have prompt engineering
skills. The report suggests that team members should be equipped with the right AI skills, defining rules and
responsibilities for cohesive human and Gen AI collaboration, ensuring accountability when Gen AI systems
make mistakes, and adapting workflows and processes for the new era of Gen AI.
Report Methodology
Capgemini Research Institute conducted a global quantitative executive survey in May 2024 across 15
different countries and 11 key industries, surveying 1,500 respondents from 500 organizations, with annual
revenue of more than $1 billion. Each unique organization is represented by three executives, one each at
leadership level, middle-management level, and front-line management level (the three respondents can be
from different functions or locations). The report is also based on an entry-level employee survey to take
their perspective on Gen AI adoption by their managers and leaders. The survey targeted 1,000 entry-level
employees from the same 500 organizations as in the executive survey. Hence, overall, each organization,
irrespective of location or function, is represented by five respondents – three executive-level (leaders and
managers) and two entry-level employees. In addition to these executive and entry-level employee surveys,
the report also draws on 15 in-depth interviews with independent experts from various industries across the
globe to validate and substantiate findings. Please note, the study findings reflect the views of the
respondents and are aimed at providing directional guidance.
Capgemini Press Release
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 Press Release |
234 | capgemini | Final-Web-Version-Report-Sustainable-Gen-AI-2.pdf | Developing sustainable Gen AI
Developing
sustainable Gen AI
#GetTheFutureYouWant
2
Developing sustainable Gen AI
Capgemini Research Institute 2025
fo
elbaT
stnetnoc
3
Developing sustainable Gen AI
Capgemini Research Institute 2025
4
Developing sustainable Gen AI
Capgemini Research Institute 2025
evitucexE
yrammus
Generative AI has a significant environmental impact: perform real-time functions, requires an equal or greater
Generative AI (Gen AI) relies on processing huge volumes of amount of energy. Data centers also consume a huge amount
data, which necessitates enormous computational power, of energy and water for cooling purposes. Running an
rendering it an energy-intensive technology. The production inference of 20–50 queries on an LLM uses about 500 ml of
of graphics processing units (GPUs), integral to the water each time.
functioning of Gen AI, requires rare earth metals, the mining
Gen AI has contributed to increased GHG emissions:
of which contributes to greenhouse gas (GHG) emissions.
In our research, we surveyed executives from 2,000
Additionally, the frequent hardware upgrades that Gen AI
organizations that have Gen AI initiatives underway.
requires put a great deal of stress on natural resources, as
well as further polluting the environment. Recent estimates • 48% of executives believe that their use of Gen AI has
suggest that, Gen AI could create between 1.2 to 5.0 million driven a rise in GHG emissions.
metric tons of e-waste by 2030, which is around 1,000 times
• Organizations that currently measure the environmental
more e-waste than was produced in 2023.1
impact of their use of Gen AI project the share of Gen
Estimates suggest that training a GPT-3 model (which AI-driven emissions as a proportion of total organizational
includes 175 billion parameters) consumes an amount of carbon emissions to rise, on average, from 2.6% to 4.8%
electricity equivalent to the annual consumption of 130 over the next two years.
US homes.2 Moving to the next model size up, GPT-4 (with
• 42% of executives have had to relook at their climate goals
1.76 trillion parameters), power consumption of training is
due to Gen AI’s growing footprint.
estimated to be equivalent to yearly power consumption of
5,000 US homes.3 After training, which is a one-time event in
the model’s lifecycle, the inferencing phase, where models
5
Developing sustainable Gen AI
However, many organizations continue to ignore Gen AI’s How to create responsible Gen AI for sustainable
growing carbon footprint: business value?
Just 12% of the executives in our research confirm that their Gen AI has the potential to accelerate sustainable business
organizations measure Gen AI footprint. Furthermore, just priorities, control costs, and drive growth. However,
20% rank “environmental footprint of Gen AI” among the organizations must weigh these benefits against the
top five factors when selecting or building Gen AI models, technology’s environmental costs. We suggest the following
with performance, scalability, and cost dominating their approach to make informed, responsible business decisions:
consideration process. Only 27% of executives say they
• Identify the right technology that addresses your
compare energy consumption levels of Gen AI models.
business needs: It is important to note that Gen AI is
Organizations are currently taking only a partial view of costs,
just one element of the broader tech landscape, and
effectively ignoring the energy costs of model deployment
solving most business problems requires a combination of
and inferencing. As Gen AI models become more complex
different techniques. “Hybrid intelligence” – a convergence
and pervasive, careful management of both financial and
of traditional AI, Gen AI, and technologies such as
environmental costs will be crucial to scalability.
automation, robotic process automation (RPA), etc. –
Organizations are just beginning to incorporate
can unlock new levels of ingenuity and efficiency. Nearly
sustainability measures into the Gen AI lifecycle, with 31%
three-quarters (74%) of executives believe that choosing
of executives saying their organization has taken steps
the appropriate technology (be it AI, Gen AI, analytics, or a
to this end. As many as 74% of executives find measuring
combination of different technologies) that addresses your
Gen AI’s environmental impact challenging due to limited
business needs is crucial to reducing the environmental
transparency from hyperscalers and model providers. They
footprint of Gen AI and harnessing its full potential.
expect the tech sector to lead efforts to normalize and
streamline measurement and transparent reporting of the
environmental impact of Gen AI.
Capgemini Research Institute 2025
evitucexE
yrammus
6
Developing sustainable Gen AI
• Assess and mitigate Gen AI's environmental impact: – Monitor and report your Gen AI footprint: Analysis
and accurate measurement, monitoring, and tracking
are paramount. Organizations should also communicate
– Build the business case for sustainable Gen
their sustainability intentions clearly to stakeholders,
AI: In addition to being more responsible from an
disclosing emissions levels, detailing progress
environmental perspective, prioritizing sustainable Gen
transparently, and setting definite goals. Of those
AI initiatives offers significant business advantages,
measuring the environmental footprint of Gen AI (12%
notably cost reductions and acceleration of work.
of our survey sample), only 28% disclose it, and only 24%
It is important to showcase the business case to
have set targets to reduce it.
top leadership and take into account the incurred
environmental cost. – Implement sustainable practices throughout Gen
AI’s lifecycle:
– Evaluate Gen AI partners and models on
sustainability parameters: More than half (55%) of 1. Hardware-related measures: Prioritize partners
executives believe that including sustainability as a with more energy-efficient and recyclable hardware
key criterion in vendor selection for all Gen AI-related specifically designed for AI/Gen AI.
requirements would reduce environmental footprint. It
2. Model architecture and algorithm-related
is crucial that organizations select the most appropriate
measures: Use smaller, task-specific pre-trained
model. For example, when comparing large and small
models. Consider optimizing model size through
models, organizations must decide on the optimum
techniques such as model compression, pruning,
balance between performance and energy consumption.
quantization, and knowledge distillation to
Decisions on using prebuilt or custom models also
significantly lower cost and energy consumption.
impact computational power and carbon footprint, with
the former demanding greater resources.
Capgemini Research Institute 2025
evitucexE
yrammus
7
Developing sustainable Gen AI
3. Sustainable-infrastructure measures: Consider them at the pilot stage currently. An additional 37% say
providers which use low-carbon energy sources such they have started to explore its potential for sustainability.
as solar, wind, geothermal, and nuclear to power Moreover, two in three (66%) executives say they expect a
AI/Gen AI infrastructure and data centers. Select reduction of more than 10% in GHG emissions in the next
cloud providers that utilize energy-efficient green 3–5 years as an output of Gen AI-led sustainable business
data centers. Choose a region for your server that initiatives. However, this assumption needs to be taken
ensures smaller environmental impact. Additionally, with caution, given the limited number of organizations
consider utilizing edge computing devices to which measure the environmental footprint of their Gen
reduce data transfer and distribute associated AI use.
energy usage.
When it comes to using Gen AI to accelerate progress
4. Sustainable usage: Tracking and quantifying toward business and sustainability goals, it is key to
the carbon footprint of Gen AI applications is identify suitable use cases. Organizations should
critical to eliminating unnecessary usage. Consider carefully identify and prioritize the most appropriate
implementing batch processing and prompt sustainability use cases for Gen AI based on financial and
optimization techniques such as prompt caching or environmental costs and the expected sustainability and
concise chain of thought (CCOT) to ensure efficient business benefits.
processing.
We looked at more than 100 use cases across functions
and sectors and assessed them across two dimensions:
• Tap into Gen AI's potential by investing in the right use the complexity of implementation and potential to create
cases to accelerate sustainable business value: One- a sustainable business impact. A few quick-wins emerged
third (33%) of executives say they have already started from our analysis, including ESG reporting, sustainable
using Gen AI for sustainability initiatives – with half of product design, life cycle assessment (LCA), supplier
Capgemini Research Institute 2025
evitucexE
yrammus
8
Developing sustainable Gen AI
sustainability reporting, virtual assistance, sustainable IT • Govern for sustainable Gen AI: Implementing a
governance, and ESG scenario planning among others. governance model for safe, transparent, sustainable, and
These use cases need to be weighed on environmental ethical usage is also imperative. Most executives (62%)
cost (energy, water, carbon) of using Gen AI and the believe robust guardrails and governance can effectively
business and environmental benefits the technology mitigate Gen AI’s environmental impact. Nearly half
offers. Our analysis also reveals the potential of Gen AI (49%) also rank the lack of clear governance models
to accelerate progress on UN Sustainable Development among the top five challenges of implementing Gen
Goals (SDGs). It should be noted that in many cases AI for sustainability. Partner with technology partners,
organizations plan to deploy a combination of startups, research institutions, sustainability experts, and
technologies to achieve a more comprehensive approach governments to share best practices, develop sustainable
to problem solving and innovation. Gen AI standards, and harness Gen AI to accelerate
sustainable business goals.
• Develop the right data and technology foundations:
Only 37% of executives claim their organization has the
right data-management tools and technologies for Gen
%
AI, and only one-third (33%) evaluate and monitor data 42
quality for Gen AI. Building the right data foundations
and developing the required skillsets are the keys to
deriving maximum benefits from Gen AI. Organizations
can also evaluate the potential of AI agents to create
sustainable business value in ESG reporting and of executives have had to relook at their
compliance-related areas. climate goals due to Gen AI’s growing
environmental footprint.
Capgemini Research Institute 2025
evitucexE
yrammus
9
Developing sustainable Gen AI
Who should read this report and why?
Who? Why?
This report should speak to technology and This report explores the environmental impact, as annual revenue above $1 billion and are already
business leaders across functions, but data, well as the potential of Gen AI to drive sustainable working on Gen AI initiatives. These organizations
digital, and sustainability leaders will find it business value. We analyze organizational awareness are based in 15 countries: US, Canada, Brazil, UK,
particularly helpful. and priorities and look at potential use cases. We France, Germany, Italy, Spain, the Netherlands,
recommend a series of actions for organizations Norway, Sweden, India, Australia, Singapore and
Gen AI models are resource-hungry – with a
to minimize Gen AI’s environmental impact and Japan. The survey spans 12 key industries and
substantial carbon, energy, water, and material
maximize its sustainability potential, while managing sectors: aerospace and defense, automotive,
footprint. It is imperative that, as the technology
cost and performance, and sustaining the impetus of banking and capital markets, consumer products,
develops, it remains within the guardrails of
technological change. energy and utilities, insurance, life sciences,
environmental sustainability.
manufacturing, public sector/government, retail,
The report presents a detailed five-step approach
telecom, and high tech. The report also includes
to developing a sustainable Gen AI. It draws on
qualitative findings from industry leaders.
comprehensive research building on our internal
expertise and a survey of 2,000 senior executives
(director level and above) at organizations that have
Capgemini Research Institute 2025
10
Developing sustainable Gen AI
01
Gen AI has a
significant
environmental impact
Capgemini Research Institute 2025
11
Developing sustainable Gen AI
Generative AI (Gen AI) has rapidly gained traction in the business and consumer world. Our Throughout its lifecycle,
recent report on consumer trends highlights that 58% of consumers in 2024 have replaced
traditional search engines by Gen AI tools for product recommendations – up from 25% in Gen AI has a considerable
2023.4 In the business world as well, Gen AI can replicate – and, in some respects, surpass –
human thought processes, synthesizing tailored content with far-reaching implications for environmental impact
driving innovation, enhancing customer experience (CX), streamlining operational efficiency,
and boosting growth. Our recent research reveals that organizations recognize the vast
potential of Gen AI: 80% have increased investment in the technology in the past 12 months. From manufacturing (encompassing materials and hardware
Moreover, while only 6% had integrated Gen AI across their business functions and locations as impact), model training, and usage (including data centers’
of the end of 2023, that figure had risen to 24% as of October 2024.5 energy, water, and carbon impact) to end-of-life (e-waste),
Gen AI consumes vast quantities of resources and leaves
However, these valuable advantages come at a cost that goes beyond the monetary. It is
notable financial and environmental footprints. Figure 1
important to recognize and address the energy consumption, carbon footprint, water usage,
highlights Gen AI's environmental impact throughout its
and e-waste entailed in the implementation of Gen AI throughout its lifecycle.
lifecycle, based on secondary sources.
%
80
of the organizations have
increased investment in Gen AI
in the past 12 months.
Capgemini Research Institute 2025
12
Developing sustainable Gen AI
Figure 1.
Gen AI’s environmental impact across its lifecycle
Gen AI’s environmental impact across its lifecycle
Manufacturing Model pre-training Model fine-tuning Model reinforcement Model inferencing and End-of-life
carbon footprint footprint footprint learning footprint usage footprint footprint
Materials and hardware Software development and optimization Ongoing usage e-waste
• Around half of the GHG • Training a model of size GPT-4 (1.76 trillion parameters) consumes • In 2022, 60% of Google’s ML energy went to inference, with • Gen AI could create
emissions from producing between 51,772 and 62,319 MWh of electricity – enough to the remaining 40% to training between 1.2 to 5.0 million
the graphics cards power at least 5,000 US homes for a year • Just a single query on ChatGPT consumes almost ten times metric tons of e-waste by
required for Gen AI the energy a Google search requires 2030, which is around
operations come from the 1,000 times more e-waste
• Running an inference of 20-50 queries on an LLM uses about
mining of rare earth than was produced in 2023
500 ml of water
elements
• The International Energy Agency (IEA) forecasts global electricity demand for data centers to more than double, from 460 TWh in 2022 to
1,000 TWh in 2026 (roughly equivalent to the electricity demand of Japan)
• Water consumption at IT infrastructure facilities in Virginia’s ‘data center alley’ in 2023 increased by 69% from 2019 levels
Source: Capgemini Research Institute analysis, Harvard Business Review, "How to make generative AI greener," July 2023, IEA, Electricity 2024: Analysis and forecast to 2026, January 2024, Financial
Times, “US tech groups’ water consumption soars in ‘data centre alley’,” August 2024, Vox, “AI already uses as much energy as a small country. It’s only the beginning,” March 2024, OECD, "How much
water does AI consume? The public deserves to know," November 2023, ARXIV, “The carbon footprint of machine learning training will plateau, then shrink,” April 2022, Frontline Magazine, "E-waste
from AI computers could ‘escalate beyond control’: study,“ October 2024.
Capgemini Research Institute 2025
13
Developing sustainable Gen AI
Gen AI’s hardware requirements put a Gen AI models are energy-hungry where the model is deployed. This demands as much or more
energy than the training phase, with the energy requirement
strain on natural resources and habitats
Srini Koushik, President of AI, Technology, and Sustainability for inferencing expected to increase exponentially as higher
Gen AI is consuming significant amounts of energy and at Rackspace Technology, a US-based multi-cloud solution numbers of people become regular users of Gen AI. As per
resources in data centers around the world. For example, provider across apps, data, and security, says: “As it exists estimates, 60% of Google’s machine learning (ML) energy
it uses thousands of graphics processing units (GPUs). GPU today, AI and sustainability take you in opposite directions. use in 2022 went to inference, with the remaining 40% to
chips require 10–15 times more power to operate than a AI consumes a lot of power, whether it’s training large training.14 According to the International Energy Agency (IEA),
traditional central processing unit (CPU).6 (We note, however, language models (LLMs) or running inference. And this just a single query on ChatGPT consumes 2.9 watt-hours of
that this usage could be at least partially offset by the greater power consumption is growing exponentially.” 11 Larger electricity – almost ten times what a Google search requires.
energy efficiency of GPUs due to their ability to perform models (that include more parameters and therefore require If we assume around 9 billion daily searches (the estimated
many more calculations simultaneously.) 7 more training data) generally consume more energy and total daily searches conducted on Google), running these
generate more carbon in the training process: searches on ChatGPT would require an additional 10 terawatt-
GPU chips and the other hardware that Gen AI requires hours of electricity annually.15 This is equivalent to the annual
• Estimates suggest that training a GPT-3 model with
are often made from copper, cobalt, tungsten, lithium, electricity consumption of 1.5 million EU citizens.16
175 billion parameters consumes nearly 1,300 MWh of
germanium, palladium, lead, chromium, cadmium, mercury,
and other earth metals. Around half of the greenhouse gas electricity, roughly the same amount of power consumed Reinforcement learning methods, such as reinforcement
(GHG) emissions in the production of the graphics cards by 130 US homes in a year.12 learning from human feedback (RLHF) and reinforcement
required for the operation of Gen AI comes from the mining • Moving to the next model size up, GPT-4, with 1.76 trillion learning from AI feedback (RLAIF), also leave a considerable
of earth metals.8 parameters, power consumption of training is estimated at environmental footprint.
between 51,772 and 62,319 MWh – enough to power 5,000
Gen AI typically requires a wider array of hardware than other
US homes for a year (at a conservative estimate).13
types of computing and cycles through that hardware at a
faster rate, requiring more frequent replacement.9 These Following the training process (a one-time event in the
shorter use phases will naturally accelerate the harmful model’s lifecycle) is the inferencing phase – essentially
effects on habitats and more rapidly deplete resources.10
Capgemini Research Institute 2025
14
Developing sustainable Gen AI
Gen AI training and inferencing drive up Figure 2.
energy requirements in data centers Global electricity demand is on a steep upward curve
Global electricity demand is on a steep upward curve
IEA forecasts global electricity demand for data centers to
more than double, from 460 terawatt-hours in 2022 to 1,000
terawatt-hours in 2026 (roughly equivalent to the electricity
demand of Japan), driven primarily by AI (see figure 2).17 1 200
Goldman Sachs estimates that the share of power demand
of data centers will form 3–4% of the global power demand
by 2030.18
This surging electricity demand for AI workloads in data
centers is already impacting the GHG emissions levels of
hyperscalers. Microsoft reported a 31% increase in Scope
3 emissions since 2020, primarily due to the expansion of
data centers.19 Google also reported a 48% increase in GHG
emissions from 2019 levels, owing to rising data-center
energy consumption and supply chain emissions.20 To try to
sate the ever-growing energy demand of their data centers,
organizations are planning to focus on nuclear energy
projects.21 Over the last year, Google Cloud,22 AWS,23 and
Microsoft24 announced plans to use small modular reactors
(SMRs) to power their data centers.
Note: Includes traditional data centers, dedicated AI data centers, and cryptocurrency consumption; excludes demand
from data-transmission networks.
Source: IEA.
Capgemini Research Institute 2025
hWT
Projected electricity demand – data center, AI, and cryptocurrencies
1 000
800
600
400
200
0
2019 2020 2021 2022 2023 2024 2025 2026
Low case Base case High case
15
Developing sustainable Gen AI
Gen AI models are also water-thirsty Widespread Gen AI adoption will see
e-waste levels shoot up
Another consequence of the growth of data center
operations is the huge increase in fresh, clean water Mark Kidd, EVP and General Manager at Iron Mountain
demand to prevent overheating. Water consumption at IT
Data Centers, says: “E-waste is one of the fastest-growing
infrastructure facilities in Virginia’s “data center alley” in
waste streams in the world. Annual e-waste production
2023 increased by 69% from 2019.25 Running an inference of
is on track to reach a staggering 75 million metric tons
20–50 queries on an LLM such as GPT-3 uses about 500 ml
by 2030. Just 17% of global e-waste is documented to be
of water each time.26 Our previous research shows that, as
collected and properly recycled each year.” 32
of 2023, around 11% of consumers had replaced traditional
search engines with Gen AI tools27– a trend that we expect to As the use of Gen AI becomes more widespread and
grow. If GPT-3 took over all 9 billion daily Google searches,28 profound, the e-waste challenge and the cost associated
it would require 4.5 billion liters of water daily to cool the with it will grow correspondingly. The limited lifespan
ensuing data-center operations. This anticipated daily water of Gen AI hardware will fuel this issue. Some estimates
requirement looks quite substantial, considering almost suggest that Gen AI could create between 1.2 to 5.0 million
half the world’s population may face severe water stress as metric tons of e-waste by 2030, which is around 1,000
soon as 2030.29 With Gen AI model training and inferencing times more e-waste than was produced in 2023.33 E-waste
pushing up the focus on nuclear energy, the amount of water largely ends up in landfills, where harmful chemicals
required for cooling nuclear power plants should also be such as mercury, lead, bromine, and arsenic leach out
taken into consideration. Moreover, estimates suggest that from the electronics, polluting soil and consequently
the production of a single microchip, which is extensively endangering the health of wildlife, livestock, and people
used within the Gen AI landscape, requires approximately in the surrounding area. The adoption of circular economy
2,200 gallons (8,328 liters) of ultra-pure water (UPW).30 measures among hardware manufacturers and data center
Approximately 10 million gallons (39 million liters) of ultra- operators is imperative to tackle this.
pure water is used per day by an average chip manufacturing
facility, which is equivalent to water used by 33,000 US
households every day.31
Capgemini Research Institute 2025
16
Developing sustainable Gen AI
Gen AI is one of the Figure 3.
Half of executives agree that Gen AI has increased their organization’s GHG emissions
reasons for the rise in
Half of executives agree that Gen AI has significantly increased their
organization’s GHG emissions
GHG emissions in nearly
Percentage of executives who agree with the statement:
half of organizations
"Gen AI has increased the GHG emissions of our organization,"by sector
In our research, we surveyed executives from 2,000
organizations that are already working on Gen AI initiatives.
A en m via rojo nr mity e n(7 t2 a% l im) o pf a t ch te o e f x Ge ec nu t Aiv I e iss ha ig gr he ee r t th ha at n t th re a ditional 48% 53% 52% 51% 50% 50% 48% 47% 46% 46% 45% 44% 42%
AI models. Nearly half (47%) say their organization’s GHG
emissions have increased in the past 12 months by nearly
6% on average. Executives believe that Gen AI is one of the
reasons for this rise in emissions. As figure 3 shows, a similar
proportion of executives say Gen AI has driven a rise in their
organization’s overall GHG emissions. Global Life sciences Manufacturing Aerospace Telecom Public sector Automotive
and defense
High tech Energy and utilities Retail Consumer products Insurance Banking and
capital markets
Source: Capgemini Research Institute, Gen AI and Sustainability survey, August-September 2024, N = 2,000 executives from
organizations that are working on Gen AI initiatives.
Capgemini Research Institute 2025
17
Developing sustainable Gen AI
Among the organizations that measure the environmental Figure 4.
impact of Gen AI, more than half (51%) say Gen AI is one
Nearly half of executives from advanced organizations say they have had to relook at
of the reasons for the rise of GHG emissions of their
Nearlyt hhaelifr osuf setxaeincaubtiilviteys cforommm aitdmvaennctes dd uoerg taon Gizeant iAonIs say they are reassessing
organizations. They also expect the technology’s emissions as
sustainability commitments due to Gen AI
a proportion of carbon emissions from internal operations to
increase from 2.6% to 4.8% in the next two years.
Percentage of executives who agree with the statement:
"With Gen AI, we have had to relook at our original sustainability commitments/goals,"
Consequently, organizations that have already started
by stage of Gen AI implementation
working with Gen AI are re-evaluating their climate goals;
47%
42% of executives in our research agree. Within organizations 42% 44%
40%
advanced in Gen AI implementation (those implementing
Gen AI across most functions/locations), nearly half (47%)
have had to relook at their sustainability commitments (see
figure 4). Google’s Chief Sustainability Officer, Kate Brandt,
explained in a press interview: “Reaching the net zero goal
by 2030 is extremely ambitious. It will require us to navigate
a lot of uncertainty, including around the future of AI’s
environmental impacts.” 34 All organizations Organizations that have enabled Gen AI
capabilities in some of their functions/locations
Organizations that have begun Organizations that have enabled
%
48 working on some pilots of Gen AI Gen AI capabilities in most/all of
their functions/locations
Source: Capgemini Research Institute, Gen AI and Sustainability survey, August-September 2024, N = 2,000 executives
of the executives say that Gen AI is from organizations that are working on Gen AI initiatives, N = 1,236 executives from organizations working on Gen AI
one of the reasons for rise in GHG pilots, N = 636 executives from organizations enabling Gen AI capabilities in some of their functions/locations and N =
128 executives from organizations enabling Gen AI capabilities in most of their functions/locations.
emissions of their organization.
Capgemini Research Institute 2025
18
Developing sustainable Gen AI
02
The sustainability
of Gen AI remains a
low priority
Capgemini Research Institute 2025
19
Developing sustainable Gen AI
Most organizations
of executives say a lack of
don't measure the 74%
transparency from Gen AI providers
makes measurement challenging
impact of Gen AI
Our research confirms that only a few executives are
currently aware of the extent of the overall environmental
64% cite the complexity of tracking energy
impact of Gen AI. For example, only 28% of executives in
consumption across various applications “First, you must
our research were aware that, on average, a Gen AI query
requires nearly 10 times as much electricity to process as a identify the impact
Google search. Only 31% were aware that training an LLM
so you can track and
at a US-based data center consumes around 700,000 liters
of fresh water. Moreover, only 38% claim to be aware of the
reduce it.”
environmental impact of Gen AI they use. 58% say driving efficiency is more important
than measuring impact
Mauli Tikkiwal, a board member at UK-based Orchard Hill
College and Academy Trust, says: “First, you must identify
the impact so you can track and reduce it.” However, only
one in ten executives (12%) says that their organization
actively measures their Gen AI footprint, while a majority
Mauli Tikkiwal
(82%) plan to start in the next 12–24 months (see figure 5).
They give a range of reasons for this omission: Board member at UK-based
Orchard Hill College and
Academy Trust
Capgemini Research Institute 2025
20
Developing sustainable Gen AI
Figure 5.
OOOnnlynl y1ly2 1% 12 2o%f% oo rogf fao norizgragatianonnizisza mateitoiaosnunsrs em mtheeea aesnsuvurireroe nt htmheee ne tenanlv ivmiriporoancnmt moefe nGnteatnal A li Imimppaacct to of fG Geenn A AII
PPeercrecenntataggee o of fe exexecucutitvieves ss asayiyningg PPeercrecenntataggee o of fe exexecucutitvieves sc ictiitningg t hthee b beeloloww r ereaasosonns sf ofor rn noot t
ththeeiri ro orgrgaanniziazatitoionn m meeaasusureres st htheeiri r mmeeaasusurirningg G Geenn A AI fI ofoootptprirnintt
GGeenn A AI fI ofoootptprirnintt
LaLcakc ko of ft rtarnanspsparaerenncyc yf rforomm h hypypeersrcsaclaelersr/sG/Geenn A AI I
77%% 1122%% mmooddeel pl prorovivdideersr sin in d disicslcolosisningg t hthee e ennvivrioronnmmeenntatla l 7744%%
NNoo a nandd n noot t YeYess
fofoootptprirnint to of fm mooddeelsls
pplalnannniningg t oto
stsatratr t
LaLcakc ko of fa wawaraerenneesss sin in le leadadeersrhshipip t eteamam o of fs usustsatianinababiliitliyt y 6688%%
mmeeasausurirningg
imimppacatc to of fG Geenn A AII
4433%% 3399%% ToTooo c ocommpplelex xt oto m meeasausurere
6644%%
NNoo, p, plalnannniningg t oto NNoo, p, plalnannniningg t oto OOrgrgananiziaztaitoionnala pl priroiortritziaztaitoionn o of fd drirviivningg
5588%%
mmeeasausurere in in t hthee mmeeasausurere in in t hthee eeffifficiceiennciceies st hthrorouugghh G Geenn A AI rI artahtheer rt hthanan it ist s
nneextx t2 244 m moonnththss nneextx t1 122 m moonnththss eennvivrioronnmmeenntatla ilm imppacatct
Source: Capgemini Research Institute, Gen AI and Sustainability Source: Capgemini Research Institute, Gen AI and Sustainability survey, August-September 2024, N = 1,767 executives
survey, August-September 2024, N = 2,000 executives from from organizations that are currently not measuring their Gen AI footprint.
organizations that are working on Gen AI initiatives.
Capgemini Research Institute 2025
21
Developing sustainable Gen AI
Organizations look to the tech sector
to drive sustainable Gen AI
A majority (78%) of the executives in our research Organizations also expect the tech sector to
%
say their organization is using pre-trained Gen AI develop innovative mitigation measures. Eszter 74
models, and only 4% have built their own models Haberl, Sustainability Business Strategy Director
from scratch. Among those using pretrained at India-based auto component manufacturing
models, 63% contract them as a service through company, Motherson Group, says: "The
hyperscale cloud providers, giving rise to a landscape of interaction between generative AI,
reliance on tech providers for measurement and sustainability, and ESG goals is very complex, of executives cite lack of
tracking. However, as figure 5, above, highlights, particularly in manufacturing. While generative transparency in disclosure and
nearly three-quarters (74%) of executives cite AI offers transformative potential for optimizing reporting of Gen AI's environmental
lack of transparency in disclosure and reporting produ |
235 | capgemini | Safeguarding-Europe-s-security-in-the-age-of-AI_Final_digital-version.pdf | Software is eating Defense
Safeguarding
Europe’s Security
in the age of AI
The world is undergoing a
profound geopolitical and
technological transformation.
Artificial Intelligence has
changed defense, and post-
quantum cryptography will be
essential to protecting its future.
This report provides leaders
with a roadmap to navigate this
critical juncture and harness
the potential of technology to
safeguard Europe’s strategic
autonomy and resilience.
Published on the occasion of the Munich
Security Conference 2025
Disclaimer: This report is not an official
publication of the Munich Security Conference
(MSC). The contents of this paper do not purport
to reflect the opinions or views of the MSC and
is meant to provide input to and stimulate the
debate at the MSC.
2 Safeguarding Europe’s Security in the age of AI
Table of contents
FOREWORD........................................................................................4
EXECUTIVE SUMMARY......................................................................5
SUMMARY OF RECOMMENDATIONS...............................................6
EUROPE AT A TECHNOLOGICAL CROSSROADS..............................7
INTRODUCTION.................................................................................8
SECURING AI TO SECURE EUROPE..................................................10
NATO’S VIEW ON ARTIFICIAL INTELLIGENCE ................................17
GLOBAL TRENDS IN AI R&D.............................................................18
SHAPING SECURITY FOR THE QUANTUM AGE..............................20
SECURITY IN THE QUANTUM AGE..................................................21
NATO’S VIEW ON QUANTUM TECHNOLOGIES..............................25
GLOBAL TRENDS IN QUANTUM AND POST-QUANTUM
CRYPTOGRAPHY R&D.....................................................................26
AI AND QUANTUM TECHNOLOGIES AT THE SERVICE
OF DEMOCRATIC STABILITY AND SECURITY.................................28
STRATEGIC RECOMMENDATIONS..................................................29
CONCLUSION...................................................................................35
CONTRIBUTORS..............................................................................36
BIBLIOGRAPHY................................................................................41
END NOTES......................................................................................45
3 Safeguarding Europe’s Security in the age of AI
Foreword
Andreas Conradi,
Head of Defense Europe
& Executive Vice President, Capgemini
National security and defense have never been more Securing the future of our democracies requires proactive
critical in the face of unpredictable, rapidly evolving steps, as conflict will persist, adversaries will mobilize,
threats. And it is the role of cutting-edge technologies, and AI and QC will continue to evolve. To best prepare for
seamlessly integrated across every aspect of defense, that future challenges, we can call on the strategic, innovative,
is enabling the industry to address emerging challenges and technological excellence available in the industry to:
with unmatched agility and precision.
Accelerate innovation and embrace
The concept of ‘software eats defense’ underscores the
transformative technologies like AI and QC
growing importance of software and digital solutions as
strategic drivers of innovation. Governments and armed Strengthen technology sovereignty to create
forces now face not only traditional weaponry but also the the right conditions for technology to thrive
weaponization of technology—tools readily exploited by
adversaries and increasingly embedded within domestic Enhance trust and interoperability by
security infrastructures. Staying ahead and securing maintaining the highest security standards
the future calls for a forward-looking vision where
digital solutions are central to strategic advancements, Act with urgency to ensure readiness for the
enhancing capabilities, resilience, and agility in a rapidly future state of conflict
evolving landscape.
Nowhere is this more evident than in the emergence of
artificial intelligence (AI) and quantum computing (QC).
In a relatively short time, the velocity and volume at which Dr. Benjamin Schulte,
AI processes information have become critical to military
Strategy & Innovation Lead Defense
decision-making, serving as a guiding force in helping
Europe, Capgemini
armed forces navigate the fog of war. Meanwhile, QC is a
rapidly emerging technology capable of obliterating even
the most robust security defenses by today’s standards.
The transformation of defense and security, driven by
new software-defined capabilities, is characterized by
The convergence of these disruptive yet transformative
a critical tension: the imperative for openness and rapid
technologies provide national defense with two invaluable
experimentation with emerging technologies clashes
assets. However, it also presents adversaries with two
with the necessity of securing these advancements
powerful tools for attack—attacks that can unfold quickly,
against adversaries, inherent risks, and new occurring
are difficult to detect, and can cause widespread damage
vulnerabilities. Proactive adoption and continuous
in an instant.
innovation are vital to deter aggression, protect our
democracies, and ensure enduring security in an
To capitalize on the opportunities presented by AI and
increasingly complex world.
QC, while future-proofing defense capabilities, we need
to operate with the most stringent security standards.
While new technologies should be explored and integrated
Understanding the origins and training of AI, as well as
swiftly, safeguarding their integrity and operational
ensuring it remains immune to adversarial influence,
security is equally essential. Against this dynamic backdrop,
is more critical than ever. With the arrival of QC on the
our report examines the transformative potential of AI and
horizon, we should confidently act to secure critical and
quantum technologies in defense, emphasizing how AI can
sensitive information. This should not be measured by
revolutionize decision-making and autonomous operations
today’s encryption standards, but rather with the question:
while quantum advancements promise unbreakable
“Is this ‘quantum safe’?”
encryption and enhanced secure communications.
4 Safeguarding Europe’s Security in the age of AI
Executive summary
Our world is in the midst of profound societal,
technological and geopolitical change.
Our world is in the midst of profound societal, advances, it threatens current cryptographic systems,
technological and geopolitical change. European leaders endangering secure communications, critical infrastructure,
are being required to rethink their strategy to reflect a shift and operational continuity. Unlike AI, PQC may not
to a multipolar world, the re-emergence of high-intensity revolutionize security operations but provides the essential
conflict, and the transformative impact of technology on backbone for safeguarding their integrity. The interplay
security, strategic autonomy, and European resilience. The between these technologies is clear: while AI catalyses
stakes are high, with implications ranging from strategic transformative capabilities, its effectiveness depends on
planning and innovation management to battlefield tactics. the foundational security provided by PQC. Without this
protection, AI’s power to enhance security and resilience
The notion of “software is eating the world” also holds becomes a potential liability.
true for defense and security as the recent events in Ukraine
have demonstrated. Thus, it can be said that “software Europe’s strategic autonomy in a multipolar world will
is eating defense”, with a growing appetite and ever- hinge on its ability to navigate the convergence of AI’s
accelerating pace. Defense innovation historically focused transformative impact and PQC’s protective potential.
on hardware, especially platform centric capabilities such The secure integration of these technologies will ensure
as tanks, aircraft, and ships. seamless coordination among allies and fortify Europe
against hybrid threats and adversarial capabilities. As
Today’s alpha and omega is the interplay of software- digital transformation accelerates, the interplay between
defined and hardware-enabled capabilities, shaping future AI and PQC should be harnessed to strengthen Europe’s
systems, operations, and decision-making. This shift technological sovereignty and resilience.
unlocks unprecedented opportunities and introduces new
vulnerabilities that should be proactively managed. The This report assesses the security implications of AI
transformative power of Artificial Intelligence (AI) and integration and PQC adoption in defense and security,
the emerging threats from quantum computing demand emphasizing their interconnected roles in securing
an urgent, coordinated and strategic response. Without Europe’s strategic future. It concludes with actionable
robust defense, societies face greater risks of manipulation, recommendations for Europe’s political, military, and
threatening stability, sovereignty, and democracy. industrial leaders to:
Artificial Intelligence is transforming the operational
landscape across critical domains, serving as a catalyst for
national security, public safety, infrastructure resilience,
Accelerate innovation and
and crisis management. It enhances decision-making,
situational awareness, and predictive capabilities, reshaping operational integration.
how governments, organizations, and industries address
security challenges. AI brings with it complex challenges
related to data management, supply chains, cybersecurity Strengthen technological sovereignty.
and human oversight, demanding increasing attention to
the secure uses and implementation of the technology.
Enhance trust and interoperability.
In parallel, post-quantum cryptography (PQC) offers
the protection of the digital foundation upon which AI
and other critical systems rely. As quantum computing
5 Safeguarding Europe’s Security in the age of AI
Summary of
recommendations
Accelerate innovation and integration Target Audiences
Adopt a balanced approach between risk-tolerance and ethics
to testing emerging technological solutions
Adapt procurement procedures to the short development
cycle of information technologies
Train and develop AI systems with realistic, high-quality synthetic data
Strengthen technological sovereignty Target Audiences
Increase domestic production of critical components to reduce
external dependencies
Task an EU agency to coordinate and centralize expertise, streamline
adoption, and drive standardization in emerging technologies
Improve the training and anticipate the need for security and defense workforce
in line with the requirements of a rapidly evolving technological landscape
Enhance trust and interoperability Target Audiences
Develop a transatlantic “common data strategy” to facilitate
the sharing of AI training data
Develop a transatlantic shared approach to AI and quantum ethical
development and use
Establish a standardized, robust AI development and management
framework for interoperability between allies
Key Armed forces Policy-makers Defense industry
6 Safeguarding Europe’s Security in the age of AI
Europe at a
technological
crossroads
General (ret.) (OF-9) Eric Autellet
Former Major General of the French
Defense Staff
We stand today at a critical juncture, witnessing a
convergence of societal transformations, technological
breakthroughs and geostrategic changes. Europe’s strategic
context is undergoing a significant transformation and
evolving dynamics which require new mindsets, strategies
and partnerships. While some principles remain unchanged,
tangible aspects like recent technological advances (AI,
cloud computing, big data), changing geopolitical context,
and societal development are transforming the security
and military sphere.
Power and interactions are shifting from global to regional
scales, redefining international relations and prompting
leaders to prioritize regional security,
autonomy and resilience.
The integration of emerging technologies without
strategic foresight risks undermining Europe’s sovereignty,
potentially leading to serious technological and strategic
disruption. A balanced and cohesive approach to
technology deployment is therefore essential.
The increasingly widespread use of digital assets is
now enabling permanent competition, unexpected
confrontations and new ways of fighting. Europe has yet
to master the digital domain, which will be the next arena
for confrontation and war between states. In the near
future, the bloc’s efforts must focus on on mastering new
technologies, in particular the transition to post-quantum
cryptography, ensuring that it is not caught off-guard by
progress in this domain. It is also urgent to cohesively map
European research and development efforts to address
these challenges head-on.
7 Safeguarding Europe’s Security in the age of AI
Introduction
Is software eating defense? AI: transformation or upheaval?
The world is undergoing a profound transformation driven AI is transforming the operational landscape in numerous
by rapid technological change. Technology is reshaping areas, acting as a catalyst for innovation. To enable
society and accelerating change at an unprecedented pace, it to continue being driving force for innovation and
bringing both significant benefits and challenges. As global transformative change in national security, public safety,
power dynamics shift, nations race to secure an edge in infrastructure resilience and crisis management, its uses
this new context of high intensity and high technology. should be secured in the long term.
The military domain is increasingly defined by software.
While tanks, aircraft, and ships were once the core focus However, this technological advance also introduces
of innovation, software now drives transformation, complex challenges ranging from AI-powered cyber-attacks
shaping operational systems, decision-making processes, to algorithmic biases, either inherent to the data used to
and overall defense strategies. train AI systems or maliciously introduced by adversaries
to “poison” the data and render the AI ineffective. Secure
Where should we focus today to secure our way of life in implementation of AI-driven systems is thus fundamental
the future? New challenges arise every day, some of which to mitigate associated risks and ensure that AI improves
may not have even been envisaged just a few years ago. operational efficiency while protecting and being protected
Are we sufficiently aware of these changes, and, more against potential vulnerabilities. These foundational
importantly, able to tackle them effectively? What can transformations are forcing a profound rethink of our
we do today to mitigate future threats? security and defense.
Time to act
This report addresses the strategic, future-defining
It is essential to balance investments
challenges posed by secure AI and PQC for security and
in Gen AI with those in cybersecurity
defense, offering a roadmap for political, military, and
industrial leaders to act decisively to secure Europe’s and quantum technologies to
future. It provides a comprehensive understanding of address current risks effectively.”
current and emerging challenges and presents actionable
recommendations to inspire proactive measures. As Patrice Duboé
Benjamin Franklin aptly noted, “By failing to prepare, Executive Vice President / Chief
you are preparing to fail.” The time to act is now. Technology & Innovation Officer -
Aerospace & Defense, Capgemini
The future of European security
hinges on our mastery of
transformative technologies.
AI and quantum innovations
hould be deployed with precision
and responsibility.”
Dr. Cara Antoine
Chief Technology, Innovation & Portfolio
Officer / Executive Vice President |
Capgemini
8 Safeguarding Europe’s Security in the age of AI
Quantum: the next challenge
The evolution of the AI-augmented battlefield makes
secure communication essential. PQC will play a pivotal
role in securing the digital infrastructure that AI and
digital systems rely on against emerging threats posed
by quantum capabilities. While these applications for
now remain theoretical, they hold the potential to
disrupt secure communications on a massive scale,
rendering current encryption protocols obsolete and
jeopardizing military operations. PQC will provide the
necessary foundation for maintaining the integrity
the digital backbone of future operations.
Quantum computing could render secure communication
impossible overnight. Command would thus no longer
be possible and secure operations would collapse.
Such a catastrophic scenario is not inevitable. PQC can
be deployed now on IT and communications systems,
reducing the threat to data lost now and systems in the
future. PQC operates on classical rather than quantum
computers, and thus provides a practical solution today
to address the significant threats posed by tomorrow’s
quantum and computation power advances.
This report will examine these two key trends, their
potential impact on the future of software-defined
capabilities, and the strategic responses leaders should
consider capitalizing on opportunities while addressing
associated risks. Drawing on qualitative research,
including expert interviews with distinguished defense
professionals from European armed forces, NATO,
the European Defense Agency, and technology and
innovation experts at Capgemini, the report will offer
actionable recommendations for strengthening AI
security and safeguarding quantum technologies
in the years to come.
9 Safeguarding Europe’s Security in the age of AI
Securing AI to
secure Europe
AI is a catalyst for transformation and has led to a
revolution in national security, public safety, critical
infrastructure, and military operations. It enhances
decision-making, situational awareness, and predictive Artificial Intelligence refers to the ability
capabilities, enabling proactive responses to evolving of machines to perform tasks traditionally
threats. AI applications span autonomous systems,
requiring human intelligence, such as recognizing
targeting and decision support, predictive analytics,
patterns, learning from experience, drawing
and cyber defense, among others.
conclusions, making predictions, or generating
The conflicts in Ukraine and the Middle East have recommendations. These applications may guide
highlighted the growing pervasivness of AI and its or alter the behavior of autonomous physical
accelerated integration into a variety of systems systems (like automated vehicles) or operate
and platforms, such as Unmanned Aerial Vehicles
entirely within the digital domain (e.g. ChatGPT),
(UAVs), targeting processes, or the analysis of satellite
with autonomy ranging from partial human
imagery. The conflicts in Ukraine and the Middle East
have showcased the increasing pervasivness of AI and intervention to full independence post-activation.
accelerated integrationinto into a variety of systems
and platforms, such as Unmanned Aerial Vehicles Source: U.S. Department
(UAVs), weapons targeting systems, or the analysis of State (2023)
of satellite imagery. This chapter looks at current
and upcoming applications of AI and focuses on
how to ensure safe and secure uses of AI.
Four elements of AI
Hardware Software
Disks, computers, chips Algorithms, models
Connection Data
Networks Text, figures, images etc
10 Safeguarding Europe’s Security in the age of AI
AI applications in security
and defense
The conflict in Ukraine:
a testbed for AI
Artificial intelligence is already a major tool in a range of
critical security and defense areas. It is already transforming
all operational domains (land, sea, air, space, cyber,
AI is playing a central role in supporting
electromagnetic spectrum) and the way missions are
Ukrainian forces in intelligence, operational
conducted (from anticipation to detection and reaction). By
support and targeting. In the field of counter-
multiplying effects (e.g. swarming) and increasing battlefield
espionage, AI systems, in collaboration
transparency, AI is offering added value across all functions.
with companies such as Palantir, analyze
Its applications span military operations, military support,
vast datasets to identify threats to national
disaster prevention and humanitarian aid, intelligence,
security, flagging suspicious behavior of
homeland security and border management.
Ukrainian citizens or their potential links
with Russia. Moreover, AI is integrated
AI will specially improve decision support in all the areas
with voice translation tools that process
of security and defense. At the strategic level, AI will be
intercepted enemy communications,
able to analyze action plans, issue early warnings and help
extracting actionable intelligence to
produce simulations to guide operational planning. At
anticipate adversary movements.
the operational level, it already processes intelligence
to prioritize and validate targets. At the tactical level,
In the field of operational support, the
AI provides real-time data and actionable intelligence
Operations Centre for Threat Assessment
to optimize immediate responses.
(COTA), leveraging AI, integrates various data
streams, providing real-time information
to guide logistics and strategy. Finally, AI
improves target acquisition, analyzing drone
As AI continues to mature, we and social media data to locate and neutralize
targets of strategic value on a daily basis.
can expect further disruptions in
military operations. The journey
towards AI integration is already
well underway.”
Dr. Bryan Wells
NATO Chief Scientist
11 Safeguarding Europe’s Security in the age of AI
Current, emerging, and future AI applications
Online threat Intelligence Operational Complex
detection processing simulations AI agent decision
and analysis support
Now New Next
Supply chain Predictive Drone swarming Autonomous C2 automation
optimization behavior cyber attack
response
Risks to secure AI uses
and mitigation strategies
AI impacts all operational domains, military functions if
not the very nature of warfare. Securely implementing and
using AI poses specific challenges linked to technology,
people and process. This also requires a clear balance
between the need to rapidly implement AI in the fields
of defense and security while navigating the challenges
related to this implementation. These challenges can be
structured in four main categories: cybersecurity, supply
chain security, data, as well as expertise and human
resources, and require targeted mitigation strategies.
12 Safeguarding Europe’s Security in the age of AI
Challenges and mitigation strategies
Key challenges to the integration of AI in defense and security can be tackled by a
number of mitigation strategies, structured in four main categories: cybersecurity,
supply chain security, data, as well as expertise and human resources.
Domain Challenge Mitigation strategy
System AI is vulnerable to attack • Harmonize security standards to mitigate threats
on AI systems, including model poisoning, oracle attacks,
security
and input perturbation
• Increase collaboration between the public and private
sectors to strengthen AI against adversarial attacks and
improve the cyber security of hardware and software
from the R&D to the implementation phase
Supply chain Strain on the • Invest in developing European semiconductor production
supply chain/disruptions capabilities to reduce external dependency.
security
• Encourage partnerships with European industry,
academia, and research institutions to mitigate risks
from strategic competition
Data • Lack of quality and • Develop a sovereign Cloud for securing sensitive data while
quantity of data guaranteeing compliance with national security protocols
• Difficulty obtaining and • Integrate, in the future, fully homomorphic encryption,
sharing data (encrypted, enabling classified data to be shared and processed
classified, incomplete) securely without decrypting
• Data requiring advanced • Create data centers to meet the sector’s growing
storage and processing computing and storage needs.
capabilities • Use synthetic data in situations where it is impossible to
• Data poisoning obtain data, in particular to anticipate future scenarios
Expertise • Cognitive biases Enhance human oversight and
(e.g. automation bias) expertise through:
and human
• Excessive reliance
resources • (Re)training programs
on AI outputs
• High-level expertise cultivation
• Obligation to respect • Integration of technical specialists
international into military and security operations
humanitarian law
• Maintaining meaningful
human control over the
use of force
• Necessity to trust
the AI system
13 Safeguarding Europe’s Security in the age of AI
Cyber and supply chain security
The secure implementation of AI for defense and security We need to get end users
faces important obstacles in cybersecurity and supply
and operators into capability
chain security. For instance, hostile actors could exploit
development, for clearer
vulnerabilities in AI systems through deceptive data inputs
(“data poisoning”) during the development stage, or by operational needs and agile
targeting the model itself.1 Because these techniques are capability development with direct
not only persistent and evolving threats, but also highly
feedback from the theater. This
sophisticated and difficult to detect, NATO’s AI Strategy
highlights that they put critical infrastructure and sensitive means rethinking how we develop
operations at risk.2 software-defined, hardware-
enabled capabilities.”
Rising global demand for semiconductors and microchips
further strains AI supply chain security, as production is
Dr. Benjamin Schulte
limited by long lead times, complex and capital-intensive
Strategy & Innovation Lead Defense Europe,
design and manufacturing processes, and can be subject
to geopolitical tensions.3 The high costs associated with Capgemini
designing new chips mean that economies of scale are
essential, leading to the concentration of production
between a few leading companies. A few countries
dominate these supply chains, raising concerns about
reliance, strategic leverage, and espionage. The U.S., for
example, has restricted exports of advanced chips and
Europe needs to strengthen its
manufacturing equipment to China,4 highlighting the
importance of controlling critical supply chains for the European champions to remain
secure use of AI in security and defense, especially for digitally sovereign.”
states without independent supply chains of their own.
Dr. Christian Weber
AI systems’ security should be strengthened throughout
Principal, Partner Lead and Client Manager
their lifecycle. This includes fortifying AI against hostile
actors’ attacks and improving the (cyber) security of the Defense, Capgemini Insights & Data Germany
associated hardware and software. Governments are
investing heavily in domestic semiconductor production to
reduce dependence on foreign suppliers.5 The European
Union’s adoption of the €43 billion European Chips Act,
which aims to produce 20% of the world’s semiconductors
by 2030 in the EU, is one such example.6
Cooperation between the public and private sectors is Integrating the results of start-ups
essential to securing AI for security and defense. NATO’s AI into the traditional procurement
strategy highlights partnerships with industry, academia,
and industrial world remains
and research institutions to advance technological
a major Challenge.”
capabilities, safeguard intellectual property, and mitigate
risks from adversarial use or strategic competition.7
Andreas Conradi
These efforts, aligned with the Munich Security
Head of Defense Europe / Executive Vice
Conference’s call for strengthened semiconductor
and AI coordination, aim to foster innovation and ensure President, Capgemini
Europe’s access to vital AI components.8
14 Safeguarding Europe’s Security in the age of AI
Data
Data is another major challenge, first and foremost the The use of synthetically generated data effectively
quantity and quality of available data. Training military AI addresses many challenges. It can fill the gap where
systems relies on accurate, relevant and AI-ready data for real data is unavailable or provide lower-classification
the adequate fulfillment of their functions, but this can data for initial model development, enabling a smoother
be difficult to obtain.9 Furthermore, the vast quantities of transition to higher-classification environments.
data generated (for example by sensors and collaborative
combat operations) require advanced storage and Additionally, synthetic data is often indispensable for
processing capabilities, which are not always available, preparing AI to handle real-world scenarios that have yet to
especially at the edge. These limitations pose significant occur, such as zero-day cybersecurity threats. By simulating
operational challenges on platforms such as submarines, battles or unprecedented situations, synthetic data enables
tanks, and other vehicles where computational resources training for both AI systems and personnel. To be effective,
are highly constrained. Military AI systems depend on however, this data should closely mirror real-world
access to encrypted or classified data, which adds another conditions, requiring a high degree of “equivalence”
layer of complexity and raises the question of who can to ensure reliability.15
access and use this sensitive data. Finally, the risks of data
poisoning and adversarial manipulation—where attackers
corrupt training or test datasets to reduce the performance
of AI models—further raise the stakes because of the
grave consequences that erroneous AI outputs can have
in military settings.10
A commonly used solution to
One solution is to develop sovereign cloud infrastructure11
to secure sensitive defense data in compliance with mitigate paucity of data is the use
national and regional security protocols.12 The future of synthetically generated data.”
integration of fully homomorphic encryption is another
significant step, as it will enable classified data to be shared Dr. Mark Dorn
and processed securely without decryption, protecting Director Defense, Cambridge Consultants
critical information even in cooperative situations.13 The
creation of vast data centers with a capacity in excess of
one gigawatt is another crucial milestone towards meeting
the sector’s growing computing and storage needs. These
facilities would make it possible to process operational data
at an unprecedented scale, while guaranteeing its security
and availability.14 Investing in these strategies could
vastly improve data security management and lay a solid
foundation for the implementation of secure AI across
the defense and security sectors.
15 Safeguarding Europe’s Security in the age of AI
Expertise and human resources
The integration of AI decision-support systems into military
and security applications raises issues around human-
machine interaction. Secure AI implementation requires
high ethical, legal, and human decision-making standards,
which should provide the flexibility required to continually In defense, the trust factor is crucial.
incentivize and nourish innovation rather than stifle it.
Soldiers and military personnel need
Key concerns center on human control over the use of to trust the AI systems they’re using,
force and the mitigation of cognitive biases, such as just as they would trust any other
over-reliance on automated systems while disregarding
tool in combat. That’s why it’s
contradictory information (automation basis), which may
important to invest in AI literacy
distort decision-making.16 Operators may inadvertently
ignore the legal and strategic implications of their decisions and ensure users understand how
and cause errors or unintended outcomes. Integrating AI to use these systems responsibly.”
will also expand internal attack surfaces, as personnel may
misuse AI.17 On the technology end, AI systems can have
Martijn van de Ridder MSc
difficulty adapting to dynamic wartime conditions.18 The
Vice President | Lead Data & AI Defense
U.S. DoD’s Project Maven is a case in point, struggling to
Europe
independently identify an enemy vehicle under different
weather conditions than those it was originally trained
on.19 AI adoption in a military context risks contravening
international humanitarian law, in particular the principles
of distinction, necessity, humanity and proportionality,
which underpin the lawful conduct of hostilities.20
Strategies to enhance huma |
237 | forrester | 786b1a49-7a44-4f8b-94de-2cae08ac8903.pdf | To shareholders and all members of the Forrester community,
Against the backdrop of an uncertain economy and continued layoffs in the tech industry, we continued
our voyage of transitioning clients to Forrester Decisions in 2023. Our target was to migrate two-thirds
of our contract value (CV) to the new platform, and I am happy to report that we achieved that
important milestone. By the end of 2024, our three-year product transition to a single, powerful, and
scalable Forrester Decisions research product will be complete.
We also made progress on two other business imperatives: 1) creating a high-performance sales
organization and 2) capturing opportunities opened by generative artificial intelligence (genAI).
While progress was made, our financial performance did not meet plan. We managed through these
challenges by carefully controlling expenses and staying laser-focused on building a CV growth engine.
A different kind of research and advisory partner
With the advent of Forrester Decisions and its supporting infrastructure of advisory, consulting, and
events, Forrester is filling a unique market gap. We are no longer serving the old research library model
in which mid-level executives in companies received research and built a “library” to answer one-off
questions. And we are not in the consulting business, which focuses on transactions and transitory
relationships.
Forrester is a new type of research and advisory partner serving C-level executives and their teams. We
uniquely help our clients focus on winning, serving, and retaining their customers to drive growth. We
tailor our engagement model to enable our customers to achieve their ongoing initiatives and
outcomes. We are always with our clients — on their side and by their side — through their multi-year
projects and transformations. And we help companies align their marketing, sales, product, technology,
digital, and customer experience functions to operate most efficiently.
Forrester Decisions brings together three powerful research and advisory components:
1. Bold vision to help clients stay ahead of changing customer and market dynamics and better
plan for the future. Forrester Decisions clients get access to customer obsession research,
customer insights, trends and predictions, market forecasts, and technology and service
provider landscapes.
2. Curated tools and frameworks to empower clients to execute on their vision with proven
strategic models and plug-and-play templates. This includes access to key performance
indicators and peer benchmarks, assessments, strategic models, templates, Forrester Wave™
evaluations for specific functions, and certification courses.
3. Hands-on guidance to enable leaders to de-risk decisions, leverage best practices, and set
their teams up for success. These are delivered through guidance sessions with analysts, peer
discussions, and events.
Clients have embraced the new Forrester model. In 2023, we conducted 15,000 client guidance sessions,
driven by initiatives and outcomes that we have recorded for over 85% of clients. Clients report that
Forrester Decisions improves the success rate of transformational initiatives by 26%, accelerates time to
value for transformational initiatives by 50%, and delivers a return on investment of 259%.
Sharpening our go-to-market team
The new Forrester is designed to be sold to C-level executives in user companies larger than $500 million
in revenue and technology vendors and service providers with more than $50 million in revenue. Selling
higher allows us to land and expand across functions. Our goal is to win the leader, win the team, and,
ultimately, to win the organization.
Under the leadership of Nate Swan, Forrester’s chief sales officer, the Forrester sales force is well
positioned to achieve high performance. Nate has brought new leaders into key sales roles, boosted
sales operations and enablement, put a new emphasis on improving sales process and methodology,
increased sales activity and strengthened sales pipelines, and standardized the way that we sell. These
efforts are improving sales efficacy and preparing the company to scale the sales force more quickly.
This experienced, confident, and resolute team is making the changes that will enable the company to
expand contract value.
Additionally, in 2023, we refocused our consulting and events businesses to help drive contract value.
These non-CV businesses are becoming a smaller fraction of our overall revenue mix — in 2023, 70% of
total revenue was in research services, and we expect this share to increase in future years. Our contract
value portfolio is more predictable, scalable, and profitable — and over time, the changing mix will
improve the overall quality of our business.
Seizing the genAI opportunity
Generative AI represents an extraordinary opportunity for Forrester. In meetings with clients, I have
described genAI as the most significant technology change of my lifetime. It represents three
opportunities for Forrester:
1)(cid:3) Research. Forrester was built for exactly these types of moments. When new technology
arrives, large companies need guidance and research to plot the best deployment strategies.
Generative AI was a major focus of our research in 2023, as more than 85 analysts published
reports on how to harness the potential of genAI and provided guidance sessions for clients.
These insights help business and technology leaders separate the massive hype from reality —
and understand how they can leverage AI to augment business practices in ways that were
previously impossible.
2)(cid:3) Product. We believe that genAI will revolutionize the research industry. To meet the genAI
moment, we introduced Forrester’s client-facing generative AI tool, Izola, representing the most
significant step forward in delivering content to our clients since we launched our website 30
years ago. As of April 2024, Izola is available to all Forrester Decisions clients.
3)(cid:3) Operations. In addition to Izola, we have developed four genAI tools that improve internal
efficiency and processes. These systems provide automation for our customer success
managers, account executives, and analysts — enabling them to work more efficiently with
speed.
I am very proud of how quickly Forrester grabbed the genAI opportunity. In a challenging year, we
stayed on offense.
ESG impact
In 2023, we continued to make progress on our environmental, social, and governance (ESG) journey. We
hired our first-ever director of diversity and inclusion to operationalize and build on the strong
foundational work we’ve completed in recent years. We continued to strengthen ties with our local
communities through donations, volunteer work, and outreach.
Outlook for 2024
In a 2023 survey of research and consulting decision-makers that Forrester commissioned from an
independent market research firm, Forrester ranked as one of the most recognizable research and
consulting companies in the world. Our brand had advanced from previous years, and its recognition
and quality was grouped in the top five with much larger players like McKinsey and Bain. Despite our
challenges over the last two years, we continue to punch way above our weight, a good harbinger of the
company’s potential.
I have never been more positive and optimistic about the future of Forrester. We have taken bold action
over the last four years — acquiring SiriusDecisions, focusing the business on net contract value
increase, simplifying our product portfolio, launching our research power platform, Forrester Decisions,
and doubling down on helping our clients use customer obsession to accelerate business growth. It has
been a difficult journey which has been harder than we had expected — but we are confident that we
have made the changes that position Forrester for a better future. The foundational work of 2023 will
enable us to capture a growing share of our substantial total addressable market in 2025 and beyond.
I want to thank all Forresterites who are with me on this journey — your dedication to our clients and
your will to win are inspiring. And I want to thank our investors for their patience and vision as we move
forward — I can assure you that we are working diligently every day to increase shareholder value and
leverage our already strong market position.
Thank(cid:3)you.(cid:3)
(cid:3)
George(cid:3)F.(cid:3)Colony
Form 10-K
2023
UNITED STATES
SECURITIES AND EXCHANGE COMMISSION
Washington,D.C.20549
FORM 10-K
(MarkOne)
☒ ANNUALREPORTPURSUANTTOSECTION13OR15(d)OFTHESECURITIESEXCHANGEACTOF1934
ForthefiscalyearendedDecember31,2023
OR
☐ TRANSITIONREPORTPURSUANTTOSECTION13OR15(d)OFTHESECURITIESEXCHANGEACTOF1934
FORTHETRANSITIONPERIODFROM TO
CommissionFileNumber000-21433
Forrester Research, Inc.
(ExactnameofRegistrantasspecifiedinitsCharter)
Delaware 04-2797789
(Stateorotherjurisdictionof (I.R.S.Employer
incorporationororganization) IdentificationNo.)
60AcornParkDrive
Cambridge,Massachusetts 02140
(Addressofprincipalexecutiveoffices) (ZipCode)
Registrant’stelephonenumber,includingareacode:(617)613-6000
SecuritiesregisteredpursuanttoSection12(b)oftheAct:
Titleofeachclass TradingSymbol(s) Nameofeachexchangeonwhichregistered
CommonStock,$0.01ParValue FORR NasdaqGlobalSelectMarket
SecuritiesregisteredpursuanttoSection12(g)oftheAct:None
IndicatebycheckmarkiftheRegistrantisawell-knownseasonedissuer,asdefinedinRule405oftheSecuritiesAct.YES☐NO☒
IndicatebycheckmarkiftheRegistrantisnotrequiredtofilereportspursuanttoSection13or15(d)oftheAct. YES☐NO☒
IndicatebycheckmarkwhethertheRegistrant:(1)hasfiledallreportsrequiredtobefiledbySection13or15(d)oftheSecuritiesExchangeActof1934duringthe
preceding12months(orforsuchshorterperiodthattheRegistrantwasrequiredtofilesuchreports),and(2)hasbeensubjecttosuchfilingrequirementsforthepast90
days. YES☒NO☐
IndicatebycheckmarkwhethertheRegistranthassubmittedelectronicallyeveryInteractiveDataFilerequiredtobesubmittedpursuanttoRule405ofRegulationS-T
(§232.405ofthischapter)duringthepreceding12months(orforsuchshorterperiodthattheRegistrantwasrequiredtosubmitsuchfiles). YES☒NO☐
Indicatebycheckmarkwhethertheregistrantisalargeacceleratedfiler,anacceleratedfiler,anon-acceleratedfiler,smallerreportingcompany,oranemerginggrowth
company.Seethedefinitionsof“largeacceleratedfiler,”“acceleratedfiler,”“smallerreportingcompany,”and“emerginggrowthcompany”inRule12b-2ofthe
ExchangeAct.
Largeacceleratedfiler ☐ Acceleratedfiler ☒
Non-acceleratedfiler ☐ Smallerreportingcompany ☐
Emerginggrowthcompany ☐
Ifanemerginggrowthcompany,indicatebycheckmarkiftheregistranthaselectednottousetheextendedtransitionperiodforcomplyingwithanyneworrevised
financialaccountingstandardsprovidedpursuanttoSection13(a)oftheExchangeAct. ☐
Indicatebycheckmarkwhethertheregistranthasfiledareportonandattestationtoitsmanagement’sassessmentoftheeffectivenessofitsinternalcontrolover
financialreportingunderSection404(b)oftheSarbanes-OxleyAct(15U.S.C.7262(b))bytheregisteredpublicaccountingfirmthatpreparedorissueditsauditreport.
☒
IfsecuritiesareregisteredpursuanttoSection12(b)oftheAct,indicatebycheckmarkwhetherthefinancialstatementsoftheregistrantincludedinthefilingreflectthe
correctionofanerrortopreviouslyissuedfinancialstatements.☐
Indicatebycheckmarkwhetheranyofthoseerrorcorrectionsarerestatementsthatrequiredarecoveryanalysisofincentive-basedcompensationreceivedbyanyofthe
registrant’sexecutiveo(cid:31)cersduringtherelevantrecoveryperiodpursuantto§240.10D-1(b).☐
IndicatebycheckmarkwhethertheRegistrantisashellcompany(asdefinedinRule12b-2oftheExchangeAct). YES☐NO☒
Theaggregatemarketvalueofthevotingandnon-votingcommonequityheldbynon-affiliatesoftheRegistrant,basedontheclosingpriceofthesharesofcommon
stockonTheNASDAQStockMarketonJune30,2023,wasapproximately$340,000,000.
ThenumberofsharesofRegistrant’sCommonStockoutstandingasofMarch4,2024was19,394,000.
DOCUMENTSINCORPORATEDBYREFERENCE
Portionsoftheregistrant’sProxyStatementrelatedtoits2024AnnualStockholders’MeetingtobefiledsubsequentlyareincorporatedbyreferenceintoPart
IIIofthisForm10-K.
FORRESTERRESEARCH,INC.
INDEXTOFORM10-K
Page
PARTI
Item1. Business 3
Item1A. RiskFactors 7
Item1B. UnresolvedStaffComments 9
Item1C. Cybersecurity 10
Item2. Properties 10
Item3. LegalProceedings 11
Item4. MineSafetyDisclosures 11
PARTII
Item5. MarketforRegistrant’sCommonEquity,RelatedStockholderMatters,andIssuerPurchasesofEquity
Securities 12
Item6. [Reserved] 13
Item7. Management’sDiscussionandAnalysisofFinancialConditionandResultsofOperations 14
Item7A. QuantitativeandQualitativeDisclosuresAboutMarketRisk 23
Item8. ConsolidatedFinancialStatementsandSupplementaryData 24
Item9. ChangesinandDisagreementsWithAccountantsonAccountingandFinancialDisclosure 55
Item9A. ControlsandProcedures 55
Item9B. OtherInformation 55
Item9C. DisclosureRegardingForeignJurisdictionsthatPreventInspections 55
PARTIII
Item10. Directors,ExecutiveOfficers,andCorporateGovernance 56
Item11. ExecutiveCompensation 57
Item12. SecurityOwnershipofCertainBeneficialOwnersandManagementandRelatedStockholderMatters 57
Item13. CertainRelationshipsandRelatedTransactions,andDirectorIndependence 57
Item14. PrincipalAccountantFeesandServices 57
PARTIV
Item15. ExhibitsandFinancialStatementSchedules 58
Item16 Form10-KSummary 58
SIGNATURES 61
2
ThisAnnualReportonForm10-Kcontainsforward-lookingstatementswithinthemeaningofthePrivateSecuritiesLitigation
ReformActof1995.Wordssuchas“expects,”“believes,”“anticipates,”“intends,”“plans,”“estimates,”orsimilarexpressionsare
intendedtoidentifytheseforward-lookingstatements.Referenceismadeinparticulartoourstatementsaboutchangingstakeholder
expectations,migrationofourclientsintoourForresterDecisionsproducts,productdevelopment,holdinghybridevents,possible
acquisitions,futuredividends,futuresharerepurchases,futuregrowthrates,operatingincomeandcashfromoperations,future
deferredrevenue,futurecompliancewithfinancialcovenantsunderourcreditfacility,futureinterestexpense,anticipatedincreases
in,andproductivityof,oursalesforceandheadcount,theadequacyofourcash,andcashflowstosatisfyourworkingcapitaland
capitalexpenditures,andtheanticipatedimpactofaccountingstandards.Thesestatementsarebasedonourcurrentplansand
expectationsandinvolverisksanduncertainties.Importantfactorsthatcouldcauseactualfutureactivitiesandresultsofoperations
tobemateriallydifferentfromthosesetforthintheforward-lookingstatementsarediscussedbelowunder“RiskFactors.”We
undertakenoobligationtoupdatepubliclyanyforward-lookingstatements,whetherasaresultofnewinformation,futureevents,or
otherwise.
PARTI
Item1.Business
General
ForresterResearch,Inc.isaglobalindependentresearchandadvisoryfirm.Wehelpleadersacrosstechnology,customer
experience,marketing,salesandproductfunctionsusecustomerobsessiontoaccelerategrowth.ThroughForrester’sproprietary
research,consulting,andevents,leadersfromaroundtheglobeareempoweredtobeboldatwork,navigatechange,andputtheir
customersatthecenteroftheirleadership,strategy,andoperations.Ouruniqueinsightsaregroundedinannualsurveysofmorethan
700,000consumers,businessleaders,andtechnologyleadersworldwide,rigorousandobjectiveresearchmethodologies,over100
millionreal-timefeedbackvotes,andthesharedwisdomofourclients.
OurcommonstockislistedonNasdaqGlobalSelectMarketunderthesymbol"FORR".
MarketOverview
Webelievethatmarketdynamics—fromempoweredcustomerstotheemergenceofgenerativeAI—havefundamentally
changedbusinessandtechnology.Thesedynamicscontinuetochangestakeholderexpectations.
Consumersandbuyershavenewdemandsandrequirements.Towin,serve,andretaincustomersinthisenvironment,we
believethatcompaniesrequireahigherlevelofcustomerobsession.Customerobsessedfirmsputtheircustomersatthecenteroftheir
leadership,strategy,andoperations.Ourresearchhasshownthatcustomer-obsessedfirmsgrowfasterandaremoreprofitable.
Organizationsandleadersrequireacontinuousstreamofguidanceandanalysistoadapttotheseever-changingbehaviorsand
realities.Webelievethatthereisanincreasingneedforobjectiveexternalsourcesofthisguidanceandanalysis,fuelingwhatwecall
the“goldenageofresearch.”
Forrester’sStrategyandBusinessModel
Thefoundationofourbusinessmodelisourabilitytohelpbusinessandtechnologyleaderstackletheirmostpressingpriorities
anddrivegrowththroughcustomerobsession.Forresterhelpsclientssolveproblems,makedecisions,andtakeactiontodeliver
results.Withourproprietaryresearch,consulting,andevents,ourbusinessmodelprovidesmultiplesourcesofvaluetoourclientsand
createsasystemtoexpandcontractvalue("CV"),whichweviewasourmostsignificantbusinessmetric.
Generallyspeaking,wedefineCVproductsasthoseservicesthatourclientsuseoverayear’stimeandthatarerenewable
periodically,usuallyonanannualbasis.OurCVproductsprimarilyconsistofoursubscriptionresearchproducts,whileournon-CV
businesses,consultingandevents,playcriticalcomplementaryrolesindrivingourCVgrowth.
Withrespecttoourclients,webelievethatithasbecomedifficultforlargecompaniestorunmulti-yearstrategyandchange
managementprojectsontheirownascustomersarechangingfasterandcompetitorsareincreasinglyaggressive.Multi-yearCV
productrelationshipsenableustohelpourclientsformulatetheirvisionforthefutureandthentranslatethoseplansinto
implementationandoutcomesovertime.Forourinvestors,webelievethatCVgrowthwillresultinpredictableandprofitablerevenue
streams.
OurbusinessmodelisbuiltonthepremisethatanincreaseinCVgeneratesmorecashwhichcanthenbeinvestedinimproving
ourgo-to-marketstructure(activitiesincludingsales,product,marketingandacquisitions)andcreatingCVproductsthatclientsrenew
yearafteryear—repeatingthecycleanddrivingthemodelforward.Werefertothismodelasour"CVgrowthengine."
3
OurProductsandServices
Westrivetobeanindispensablesourcethatbusinessandtechnologyleadersacrossfunctions,includingtechnology,customer
experience,digital,marketing,sales,andproduct,worldwideturntoforongoingguidancetoplanandoperatemoreeffectively.
Wedeliverourproductsandservicesgloballythroughthreebusinesssegments–Research,ConsultingandEvents.
Research
Formorethan40years,Forresterhasbeenprovidingobjective,independentanddata-drivenresearchinsightsutilizingboth
qualitativeandquantitativedata.Weadheretorigorous,unbiasedresearchmethodologiesthataretransparentandpubliclyavailable
toensureconsistentresearchqualityacrossmarkets,technologies,andgeographies.
OurprimarysubscriptionresearchservicesincludeForresterDecisions,ForresterResearch,andSiriusDecisionsResearch. This
portfolioofresearchservicesisdesignedtoprovidebusinessandtechnologyleaderswithaprovenpathtogrowththroughcustomer
obsession.Keycontentavailableviaonlineaccessincludes:
• futuretrends,predictions,andmarketforecasts;
• deepconsumerandbusinessbuyerdataandinsights;
• curatedbestpracticemodelsandtoolstorunbusinessfunctions;
• operationalandperformancebenchmarkingdata;and
• technologyandservicemarketlandscapesandvendorevaluations.
Ourresearchservicesalsoincludetimewithouranalyststoapplyresearchtotheircontext.
Launchedin2021,ForresterDecisionsisaportfolioofstandardizedresearchservicescombiningkeyfeaturesofForrester
ResearchwithkeyfeaturesofSiriusDecisionsResearch.WeintendtomigrateourexistingclientsthatpurchaseForresterResearch
andSiriusDecisionsResearchproductstotheForresterDecisionsproducts,andasofJanuary1,2023,ForresterDecisionsbecameour
onlysubscriptionresearchproductavailableformostnewclients.AsofJanuary1,2024,approximately66%ofourCVwas
composedofForresterDecisionsproducts.
Consulting
OurConsultingbusinessincludesconsultingprojectsandadvisoryservices.Wedeliverfocusedinsightsandrecommendations
toassistclientsindevelopingandexecutingtheirtechnologyandbusinessstrategies.Ourconsultingprojectshelpclientswith
challengesaddressedinourpublishedresearch.Ourconsultingprojectsincludeconductingmaturityassessments,prioritizingbest
practices,developingstrategies,buildingbusinesscases,selectingtechnologyvendors,structuringorganizations,anddeveloping
contentmarketingstrategiesandcollateral,andsalestools.ConsultingplaysanimportantroleinsupportingourCVgrowth,aswe
havefoundthatclientsthatpurchaseconsultingprojectsfromusrenewtheirCVcontractsathigherratescomparedtoclientsthatdo
notpurchaseconsulting.
Events
WehostmultipleeventsacrossNorthAmerica,Europe,andtheAsia-Pacificregionthroughouttheyear.ForresterEventsare
thoughtfullydesignedandcuratedexperiencestoprovideclientswithinsightsandactionableadvicetoachieveacceleratedbusiness
growth.ForresterEventsfocusonbusinessimperativesofsignificantinteresttoclients,includingbusiness-to-businessmarketing,
salesandproductleadership,customerexperience,securityandrisk,newtechnologyandinnovation,anddatastrategiesandinsights.
OneoftheprimarypurposesofourEventsbusinessistohelpdriveourCVgrowth,andwehavefoundthatprospectiveclientsthat
haveattendedoneofoureventsconvertintoclientsathigherratescomparedtothosethathavenotattendedanevent.
Weholdallofoureventsashybridevents,consistingofbothin-personandvirtualexperiencesthatallowustoofferadded
attendeebenefitssuchasondemandsessions,morenetworkingopportunitiesandmorecontent,leadingtohigherattendee
engagement.
SalesandMarketing
Webelievewehaveastrongalignmentacrossoursales,marketingandproductfunctions.
WesellourproductsandservicesthroughourdirectsalesforceinvariouslocationsinNorthAmerica,EuropeandtheAsia
Pacificregion.Oursalesorganizationisorganizedintogroupsbasedonclientsize,geography,andmarketpotential.OurPremier
4
groupsfocusonourlargestvendorandenduserclientsacrosstheglobewhileourEmergingandMid-SizeTechgroupfocuseson
smalltomid-sizedvendorclients.OurEuropeanandAsiaPacificgroupsfocusonbothenduserandvendorclientsintheirrespective
geographies.OurInternationalBusinessDevelopmentgroupsellsourproductsandservicesthroughindependentsalesrepresentatives
inselectinternationallocations.Wealsohaveteamsfocusedonnewbusiness,revenuedevelopment,andeventsales.
Weemployed601salespersonnelasofDecember31,2023comparedto709salespersonnelemployedasofDecember31,
2022.
WealsosellselectResearchproductsdirectlyonlinethroughourwebsite.
OurmarketingactivitiesaredesignedtoelevatetheForresterbrand,differentiateandpromoteForrester’sproductsandservices,
improvetheclientexperience,anddrivegrowth.Weachievetheseoutcomesbycombiningthevalueofreputation,demand
generation,customerengagement,andsalesandcustomersuccessenablementprogramstodelivermultichannelcampaignsandhigh-
qualitydigitalexperiences.Ourcustomersuccessorganizationconductspost-saleengagementactivitiesthataredesignedtoalignto
clientoutcomes,acceleratetimetovalue,anddrivehigherretention.
AsofDecember31,2023,ourproductsandservicesweredeliveredtomorethan2,400clientcompanies.Nosingleclient
companyaccountedformorethan4%ofour2023revenues.
PricingandContracts
Wereportourrevenuefromclientcontractsinthreecategoriesofrevenue:(1)research,(2)consulting,and(3)events.We
classifyrevenuefromsubscriptionsto,andlicensesof,ourresearchproductsandservicesasresearchrevenue.Weclassifyrevenue
fromourconsultingprojectsandstandaloneadvisoryservicesasconsultingrevenue.Weclassifyrevenuefromticketstoand
sponsorshipsofeventsaseventsrevenue.
Contractpricingforannualsubscription-basedproductsisprincipallyafunctionofthenumberoflicensedusersattheclient.
Pricingofcontractsisafixedfeefortheconsultingprojectorshorter-termadvisoryservice.Weperiodicallyreviewandincreasethe
listpricesforourproductsandservices.
Wetrackcontractvalueasasignificantbusinessindicator.Contractvalueisdefinedasthevalueattributabletoallofour
recurringresearch-relatedcontracts.Contractvalueiscalculatedastheannualizedvalueofallcontractsineffectataspecificpointin
time,withoutregardtohowmuchrevenuehasalreadybeenrecognized.Contractvaluedecreased4%to$332.1millionat
December31,2023from$345.4millionatDecember31,2022.
Competition
Webelieveourfocusonhelpingbusinessandtechnologyleadersusecustomerobsessiontodrivegrowthsetsusapartfromour
competition.Inaddition,webelievewecompetefavorablydueto:
• ourabilitytoofferforward-lookingresearch,toolsandframeworksaswellashands-onguidance;
• ourfocusonprovidingteamswithinourclients'organizationswiththeconfidencetoexecuteeffectivelywithend-to-end
guidance,valuableknowledge,know-how,andasharedvocabulary;
• ouruseofrigorousresearchmethodologiestoofferobjectiveinsights;and
• ourbrandpromisetobe“onyoursideandbyyourside,”meaningthatwestrivetobeobsessedaboutourclients'needs
andprioritiesandalignedtotheirstrategies.
Ourprincipaldirectcompetitorsincludeotherindependentprovidersofresearchandadvisoryservices,suchasGartner,aswell
asmarketingagencies,generalbusinessconsultingfirms,survey-basedgeneralmarketresearchfirms,providersofpeernetworking
services,anddigitalmediameasurementservices.Inaddition,ourindirectcompetitorsincludetheinternalplanningandmarketing
staffsofourcurrentandprospectiveclients,aswellasotherinformationproviderssuchaselectronicandprintpublishingcompanies.
WealsofacecompetitionfromfreesourcesofinformationavailableontheInternet,suchasGoogle.Ourindirectcompetitorscould
choosetocompetedirectlyagainstusinthefuture.Inaddition,therearerelativelyfewbarrierstoentryintocertainsegmentsofour
market,andnewcompetitorscouldreadilyseektocompeteagainstusinoneormoreofthesemarketsegments.Increasedcompetition
couldadverselyaffectouroperatingresultsthroughpricingpressureandlossofmarketshare.Therecanbenoassurancethatwewill
beabletocontinuetocompetesuccessfullyagainstexistingornewcompetitors.
IntellectualProperty
Ourproprietaryresearch,methodologiesandotherintellectualpropertyplayasignificantroleinthesuccessofourbusiness.We
relyonacombinationofcopyright,trademark,tradesecret,confidentiality,andothercontractualprovisionstoprotectourintellectual
5
property.Weactivelymonitorcompliancebyouremployees,clientsandthirdpartieswithourpoliciesandagreementsrelatingto
confidentiality,ownership,andtheuseandprotectionofForrester’sintellectualproperty.
Employees
Attracting,retaining,anddevelopingthebestandbrightesttalentaroundtheglobeiscriticaltotheongoingsuccessofour
company. AsofDecember31,2023,weemployedatotalof1,744persons.Oftheseemployees,1,257wereintheUnitedStatesand
Canada;282inEurope,MiddleEastandAfrica(“EMEA”);and205intheAsiaPacificregion.
Culture. Ourcultureemphasizescertainkeyvalues—includingclient,courage,collaboration,integrity,andquality—thatwe
believearecriticaltodeliverForrester’suniquevaluepropositionofhelpingbusinessandtechnologyleadersusecustomerobsession
todrivegrowth.Inaddition,weseektofosteraculturewhereemployeescanbecreative,feelsupportedandempowered,andare
encouragedtothinkboldlyaboutnewideas.
DiversityandInclusion(D&I).Wefocusonattracting,hiring,andtheinclusionofallbackgroundsandperspectives,withthe
goalsofimprovingemployeeretentionandengagement,strengtheningthequalityofourresearch,andimprovingclientretentionand
customerexperience.Wefieldregularall-employeesurveystomeasureourprogressagainstourgoals.In2023,inadditiontothe
ongoingtrainingtoequipemployeestoplayanactiveroleinfosteringasafe,respectful,productive,andinclusiveworkenvironment,
examplesofoureffortswithrespecttoD&Iincluded:
• introducinganewD&ILeadershipAdvisoryCounciltohelpaccelerateourD&Igoals;
• increasingemployeeself-identificationwithinhumanresourcesystemprofiles;
• ensuringthatoureventsanddigitalexperiencesareinclusiveandaccessibletoall;and
• ourcontinuationofvariouspartnershipstoattractandaccessmoretalentfromunderrepresentedgroups.
LearningandDevelopment.WehavearobustlearninganddevelopmentprogramandcelebrateandenrichtheForresterculture
throughfrequentrecognitionofachievements.Tokeepemployeesandteamsconnectedandinspiredtodotheirbestworkina
distributedworkenvironment,wehaveenhancedthelearninganddevelopmentopportunitiesforouremployeesacrossabroadrange
ofinitiativesincludingnewhireandonboarding,D&I,andleadershiptraining.
AvailableInformation
ForresterResearchInc.wasincorporatedinMassachusettsonJuly7,1983andreincorporatedinDelawareonFebruary16,
1996.Forrester’scorporateofficesarelocatedinCambridge,Massachusetts.
OurInternetaddressiswww.forrester.com.Wemakeavailablefreeofcharge,onorthroughtheinvestorinformationsectionof
ourwebsite,annualreportsonForm10-K,quarterlyreportsonForm10-Q,currentreportsonForm8-K,andamendmentstothose
reportsfiledorfurnishedpursuanttoSection13(a)or15(d)oftheSecuritiesExchangeActof1934assoonasreasonablypracticable
afterweelectronicallyfilesuchmaterialwith,orfurnishitto,theSEC.TheSECmaintainsaninternetsite(http://www.sec.gov)that
containsreports,proxyandinformationstatementsandotherinformationregardingissuersthatfiledocumentselectronically.
6
Item1A.RiskFactors
Weoperateinarapidlychangingandcompetitiveenvironmentthatinvolvesrisksanduncertainties,certainofwhicharebeyond
ourcontrol.Theserisksanduncertaintiescouldhaveamaterialadverseeffectonourbusinessandourresultsofoperationsand
financialcondition.Theserisksanduncertaintiesinclude,butarenotlimitedto:
RiskFactorsSpecifictoourBusiness
ADeclineinRenewalsorDemandforOurSubscription-BasedResearchServices. Oursuccessdependsinlargepartupon
retaining(onbothaclientcompanyanddollarbasis)andenrichingexistingsubscriptionsforourResearchproductsandservices,
includingthemigrationofourexistingclientsfromourlegacyForresterResearchandSiriusDecisionsproductsintoourForrester
Decisionsportfolioofservices.Futuredeclinesinclientretentionandwalletretention,orfailuretogeneratedemandforandnewsales
ofoursubscription-basedproductsandservices,includingForresterDecisions,duetocompetition,changesinourofferings,or
otherwise,couldhaveanadverseeffectonourresultsofoperationsandfinancialcondition.
DemandforOurConsultingServices. Consultingrevenuescomprised25%ofourtotalrevenuesin2023and28%ofourtotal
revenuesin2022.Consultingengagementsgenerallyareproject-basedandnon-recurring.Adeclineinourabilitytofulfillexistingor
generatenewconsultingengagementscouldhaveanadverseeffectonourresultsofoperationsandfinancialcondition.
OurBusinessMaybeAdverselyAffectedbytheEconomicEnvironment. Ourbusinessisinpartdependentontechnology
spendingandisimpactedbyeconomicconditionssuchasinflation,slowinggrowth,risinginterestrates,threatofrecessionand
supplychainissuesthatmayimpactusandourcustomers.Theeconomicenvironmentmaymateriallyandadverselyaffectdemand
forourproductsandservices.IfconditionsintheUnitedStatesandtheglobaleconomyweretoleadtoadecreaseintechnology
spending,orindemandforourproductsandservices,thiscouldhaveanadverseeffectonourresultsofoperationsandfinancial
condition.AlthoughwedonothaveanyemployeesormaterialclientrelationshipsinRussiaorUkraineandonlyalimitedpresencein
theMiddleEast,ifthecurrentconflictsinUkraineandtheMiddleEastweretoescalateorspreadtootherregions,theremaybe
negativeeffectsonboththeUnitedStatesandtheglobaleconomythatcouldmateriallyandadverselyaffectourbusiness.
OurInternationalOperationsExposeUstoaVarietyofOperationalRiskswhichCouldNegativelyImpactOurResultsof
Operations. AsofDecember31,2023,wehaveclientsinapproximately76countriesandapproximately22%ofourrevenuescome
frominternationalsales.Ouroperatingresultsaresubjecttotherisksinherentininternationalbusinessactivities,includinggeneral
politicalandeconomicconditionsineachcountry,challengesinstaffingandmanagingforeignoperations,changesinregulatory
requirements,compliancewithnumerousforeignlawsandregulations,differencesbetweenU.S.andforeigntaxratesandlaws,
fluctuationsincurrencyexchangerates,difficultyofenforcingclientagreements,collectingaccountsreceivableandprotecting
intellectualpropertyrightsininternationaljurisdictions,andpotentialdisruptionscausedbyforeignwarsandconflicts.Furthermore,
werelyonlocalindependentsalesrepresentativesinsomeinternationallocations.Ifanyofthesearrangementsareterminatedbyour
representativesorus,wemaynotbeabletoreplacethearrangementonbeneficialtermsoronatimelybasis,orclientssourcedbythe
localsalesrepresentativemaynotwanttocontinuetodobusinesswithusorournewrepresentative.
AbilitytoDevelopandOfferNewProductsandServices. Ourfuturesuccesswilldependinpartonourabilitytooffernew
productsandservices.Thesenewproductsandservicesmustsuccessfullygainmarketacceptancebyanticipatingandidentifying
changesinclientrequirementsandchangesinthetechnologyindustryandbyaddressingspecificindustryandbusinessorganization
sectors.Theprocessofinternallyresearching,developing,launching,andgainingclientacceptanceofanewproductorservice,or
assimilatingandmarketinganacquiredproductorservice,isriskyandcostly.Wemaynotbeabletointroducenew,orassimilate
acquired,productsorservicessuccessfully.Ourfailuretodosowouldadverselyaffectourabilitytomaintainacompetitiveposition
inourmarketandcontinuetogrowourbusiness.
TheUseofGenerativeAIinourBusinessandbyOurClientsandCompetitorsCouldNegativelyAffectourBusinessand
Reputation. InOctoberof2023,weintroducedIzola,agenerativeAItoolthatallowsourclientstoqueryourresearchdatabase.We
arealsointheprocessofimplementingvariousothergenerativeAIinitiativeswithinourcompany.Whilewebelievethatgenerative
AItechnologiesoffersignificantopportunities,theyarerapidlyevolvingandtheintegrationofgenerativeAItechnologiesintoour
andourvendors’systems(potentiallywithoutthevendordisclosingsuchusetous |
245 | mckinsey | rewired-in-action-case-collection-2024.pdf | Eyebrow
Title on two lines
Rewired
Lorem ipsum dolor sit amet, consectetur adipiscing elit nullam
rutrum tempus.
in Action
Real-world examples of Digital and AI transformations
and how leading companies succeed
Month Year
Contents
04
Introduction
05
About McKinsey Digital
06
Recipe for capturing value from Digital and AI transformations
08
Lighthouse success stories
0260
CReocnatapc ot fU tsransformation recipe and lessons learned
Eyebrow
Title on two lines
Lorem ipsum dolor sit amet, consectetur adipiscing elit nullam
rutrum tempus.
Month Year
Introduction
When we published Rewired, McKinsey’s in-depth guide to digital and AI transformations, we wanted to show more
examples of how the best companies found success. So, we put together this booklet showcasing companies that have
successfully rewired themselves to turn digital and AI solutions into transformative value.
These companies reflect many of the core lessons of Rewired, including how to align the top team around change that
matters, how to develop technology and data that distributed teams can use to innovate, and how to unlock scale to get the
full financial benefits that are available. By rewiring themselves, these companies have developed the ability to constantly
innovate with digital and AI across the entire business to improve customer experiences and reduce unit costs.
Rewired In Action is a collection of stories highlighting how McKinsey has helped companies get value from their digital and
AI transformations. We hope you find this useful and informative. Please don’t hesitate to reach out to us with any questions.
Robert Levin Johannes-Tobias Lorenz Rodney Zemmel
Senior Partner Senior Partner Senior Partner
Boston Düsseldorf New York
4
About McKinsey Digital
McKinsey Digital is a collection of leaders, experts, and practitioners who help clients create transformative value with
technology. We work with leading companies across the world to drive transformations and build new businesses by
bringing together the capabilities they need. We help our clients create value by harnessing the power of data and artificial
intelligence, modernizing core technology, optimizing and automating operations, building stunning digital experiences, and
developing digital talent and culture.
Our global team includes more than 6,700 strategists, data scientists, designers, architects, product managers, agile
coaches, and software, data, and cloud engineers. Using the latest technology and proven methodologies, we design
digital strategies and build robust software and digital products tailored to our clients’ needs—driving transformations that
accelerate sustainable and inclusive growth.
We’ve served clients in every sector on digital and analytics:
2,000+
companies served on digital and analytics last year.
500+
new businesses built since 2019.
6,700+
practitioners globally, including:
400+ 1,430+ 840+
designers. software and cloud engineers. product owners and agile coaches.
1,500 730+ 1,800+
data scientists. data engineers. integrative digital and analytics
consultant.
5
Recipe for capturing value from digital
and AI transformations
Learnings from serving 2,000+ companies on digital and analytics topics
Strategy
Creating the Transformation Roadmap
Successful transformations start with the CEO and top leadership reimagining their business in the
digital age. The resulting decisions are translated into a detailed strategic roadmap that is both rooted in impact
and clear about the new capabilities needed to deliver it. Leading companies develop transformation roadmaps
focused on business domains that are big enough to generate meaningful value but small enough that it doesn’t
disrupt large parts of the business.
Capabilities
Building Your Talent Bench
You can’t outsource your way to digital excellence. Companies need the capabilities to build and evolve their
proprietary digital solutions, and that requires quality digital and AI talent. Top organizations create a detailed
talent roadmap to hire the best and create an environment where they thrive. This requires understanding what
really motivates top talent and adjusting the company’s culture and approach to excite them.
Adopting a New Operating Model
Building and scaling digital and AI solutions across hundreds of working teams require companies to be much
faster and more flexible in the way they develop technology, so having an agile operating model is critical.
Developing that operating model, however, is perhaps the most complex aspect of a transformation because it
touches the core of the organization and how people work together. It requires determining the right operating
model for you, and building up core capabilities in product management and experience design.
Technology for Speed and Distributed Innovation
The objective for technology is to make it easy for your pods to constantly develop and release digital and AI
innovations to customers and users. Achieving this requires building a distributed technology environment for
easy access to data, applications, and software development tools pods need to rapidly innovate and deliver
secure, high-quality solutions.
Embedded Data Everywhere
The ability of the technology solutions to generate value is dependent on the quality, relevance, and availability
of data. That’s why it’s critical to architect data thoughtfully for easy consumption, reuse, and scaling. The goal
is to have the data teams need so they can use it to make better decisions and build better data-enabled
solutions. The key is to build a set of data products that can be easily consumed by any team or application
across the organization.
Change Management
The Keys to Unlock Adoption and Scaling
Getting customers or business users to adopt solutions as part of their day-to-day activities, and scaling them
across your customer base, markets, or organizational units are often a massive challenge. Companies need
to address the technical, process, and human issues at a sufficiently granular level, have clear KPIs to track
progress, and ensure teams are capturing the value.
6
7
Lighthouse Case Examples
Grupo Mariposa:
Harnessing connected technology in the LatAm food
and beverage market
Latin America, Consumer
Mariposa believed that with the right technology, digital tools, and capabilities, it
could transform the way it serves store owners while overcoming the challenges
of a fragmented LatAm food and beverage market. It created an ecosystem
centered around store owners and powered by AI and digital solutions, such as
conversation commerce, commercial frontline planning, and route optimization.
Charles River Labs:
Accelerating drug development as a digitally-enabled
trusted partner
North America, Life Sciences
Charles River Laboratories, a prominent pre-clinical contract research
organization, embarked on a digital transformation to enable customers to
accelerate development of high-quality medicines for patients. Over a 3-year
journey, they have scaled across all enterprise and stood up an at-scale digital
factory that has dramatically accelerated the speed of delivering new products
and services.
Allianz Direct:
Advancing as Europe’s leading digital insurer
Europe, Insurance
Determined to shape the future of digital insurance and revolutionize the level
of service provided, Allianz Direct embarked on an ambitious journey. They
transformed their processes and leveraged modern technology and advanced
analytics such as AI-based loss assessment and evaluation to become “digitally
unbeatable” in every aspect of their value chain and ensure strong growth for
years to come.
Xcel Energy:
Driving toward net zero with the power of digital
North America, Energy/consumer
Faced with the imperative to replace its aging IT infrastructure and meet
increasing customer demand, Xcel Energy followed a clear roadmap to reform
its technology architecture and use digital to provide affordable, safe and
de-carbonized energy in a highly regulated environment. With initial success
in multiple business units, Xcel Energy continues to scale the program to keep
their plants cost competitive and advance toward zero-carbon baseload.
8
Kiwibank:
Building a better bank for the future of New Zealanders
Asia-Pacific, Banking
With the commitment to provide the highest level of customer service and
grow consistently, Kiwibank, New Zealand’s largest state-owned bank, set a
bold vision for digital transformation and core technology replacement. After
implementing a number of key foundational technology elements, Kiwibank is on
its way to become the top banking choice in the region.
BCP:
Taking banking to new heights on a digital rocketship
Latin America, Banking
BCP, the largest bank in Peru, recognized the potential to enhance customer
experiences and operational efficiency through digital initiatives. Their goals
were twofold: reimagining the customer experience and improving efficiency.
Harnessing new digital techniques, leveraging data and advanced analytics,
adopting new ways of working, and building new capabilities became the path
to realizing their vision of becoming the top bank in Peru.
DBS:
Transforming a banking leader into a technology leader
Asia-Pacific, Banking
In a rapidly changing digital landscape, Singapore-based DBS bank aspired
to transform into a truly digital bank with a clear vision: “Make banking joyful.”
DBS created a best-in-class platform-operating model with joint leadership
between business and technology with a firm focus on customers. It also made
fundamental shifts in its culture and built-up in-house technology capabilities
through innovative recruiting and retention strategies.
Freeport-McMoRan:
Unlocking new mining production through AI
transformation
North America, Mining
Freeport’s expectations for growth required significant capital and lengthy
permitting and construction efforts. Seeking another path, leadership turned
to artificial intelligence (AI) to see if it was possible to get more out of the assets
they already had. By aligning leadership, thoughtfully building out scaling
capabilities, and adopting an agile operating model, Freeport mined AI to drive
new value.
9
Grupo Mariposa:
Harnessing connected technology in the LatAm
food and beverage market
The opportunity The solution
Disrupting a fragmented Creating a new digital ecosystem that puts store owners
CPG market at the center
Guatemala-based Grupo Mariposa traces Grupo Mariposa partnered with McKinsey to create an end-to-end
its roots back to 1885 when it started with ecosystem powered by AI and digital solutions to help overcome the
a single soft drink factory. Today, through challenges of a market with over 3 million points of sale.
its CBC, Bia, and Beliv subsidiaries, it has
At the heart is a new platform powered by advanced analytics and strategic
become a major Latin American food and
partner Yalo, offering “conversational commerce,” which allows store owners
beverage company with operations in more
to conveniently connect with the brand and manage order inventory. The
than 16 countries.
easy-to-use digital tool also gives store owners greater agency by supplying
In Latin America, the food and beverage them with personalized recommendations based on market trends and
ecosystem is highly fragmented, with over supporting them to better serve their customers and rotate inventory. Instead
5 million points of sale. This fragmentation of relying only on a salesperson to tell them what to order, store owners
creates complexity on multiple levels, from (“tiendas”) are advised by digital tools which learn from micro-segments in
sharing data with store owners to planning their own neighborhoods to help them place and track orders digitally.
efficient delivery routes. Companies need
The platform also includes modular solutions for customer service,
to manage the commercial activities of
microloans, loyalty programs, and other services. It enables shopkeepers
thousands of salesforce, merchandisers, and
and sales teams to receive stock-out predictions and order suggestions,
delivery personnel that serve local stores.
place and track orders, and participate in customized loyalty programs. It also
Mariposa believed that with the right helps drive growth by offering business intelligence and knowledge to build
technology, digital tools, and capabilities, it management skills, contributing to the evolution of small shopkeepers into
could transform the way it serves store owners micro-business owners.
while overcoming some of the challenges of
McKinsey and Mariposa built the platform with open technology and
fragmentation. With technology, Mariposa
microservice architecture, enabling future integration with external partners.
aimed to change the beverage ecosystem and
Ultimately, the partners intend to make the platform available to additional
progress on its purpose of becoming the best
companies as a SaaS offering. The vision is to bring lenders, food and
solution for store owners and the first choice
consumer goods suppliers, delivery services, and other companies that
at the point of sale.
create value for store owners into an open ecosystem.
Just as important as the connection with store owners, Mariposa’s
transformation digitizes its go-to-market model. Mariposa co-developed
a proprietary platform with McKinsey to provide digital tools to transform
the commercial frontline roles. This platform is also being marketed to other
CPGs and distributors. It includes an atomic task module that provides each
salesperson with a prioritized list of each day’s tasks. Reps check off tasks on
10
The impact their mobile devices as they finish them., doing away with paperwork. Reps can
also review an AI-generated suggested order list based on consumer behavior
+100k data for each store they visit and place orders digitally. Delivery drivers, whether
in-house or through a vendor, can instantly call up the most efficient route for the
Points of sale that are touched so day’s stops.
far by the digital service channel in
All these technology solutions were made possible by a transformation of
Mariposa’s new ecosystem.
Mariposa’s internal culture. Mariposa integrated digital transformation as part
of its identity and communicated the goals and rationale of the transformation
8-10%
throughout the organization. The company created an implementation playbook
that spelled out the elements of the transformation and then formed a change
Sales uplift from store owners
management committee to oversee the process and support the new agile
who use the solution daily.
operating model. It cemented the changes with a revamped performance
management system and incentives.
5,000+
With these supports in place, Mariposa hired more than 50 highly skilled
digital team members in eight countries across all digital domains. Working in
Sales employees using new digital tools
global agile teams, they helped build tools and capabilities while redesigning
for commercial and deliver management.
and streamlining processes. The final cornerstone of the success was the
engagement of the senior leadership team. Mariposa’s CEO and group president
met weekly with the transformation team for over two years to bring the vision to
action.
“We believe our superpower lies in
fostering strong relationships with SMEs
and harnessing the potential of connected
technology to generate prosperity in Lessons learned
communities. Our ambition is to serve our
clients by providing them with the tools and Align the digital solution to the actual needs of clients
capabilities to make our vision a reality.” (store owners)
– Juan Pablo Mata, CEO of Apex by Grupo Mariposa Mariposa created the new platform with a focus on store owners
Strategic and shifted the go-to-market approach to “pull.” For example,
Roadmap recognizing that storeowners already used WhatsApp, the
“Our biggest learning is that digital
Mariposa team leveraged it for conversational commerce, rather
transformation is not solely about the
than imposing a new platform.
technology. It is about the people – having
the right set of talent, knowing how to recruit Recruit and retain the right tech talent to drive mindset shifts
people, and how to retain them. It’s the people Mariposa focused to attract, recruit, and assemble the right teams
that create the technology that will come and hired over 50 dedicated digital experts in data, technology,
Talent
closer to the business. ” agile, and UX in eight countries. These people became digital
ambassadors to pollinate the culture and mindset shifts
- Alfredo Jose Castañeda, Digital Transformation Leader,
Grupo Mariposa throughout the organization.
Scale the transformation across the organization with
“In the digital transformation process, a
a top-down approach
leader must have two key elements. The first
Along with a clear strategic vision, Mariposa established a change
one is a growth mindset. Once you begin to Adoption
management committee to implement its plan and change
and
understand this different perspective of how
Scaling the operating model. Top leaders consistently and relentlessly
you can bring different capabilities to the
communicated the change plan to internal and external
business, you begin a whole growth mindset
stakeholders, including visits to stores and directly working with
process. And then the second one is servant
salesforce.
leadership. Servant leadership is to accept
that the way that we used to work, it just Video link and case story
doesn’t apply to the modern business”
- Alfredo Jose Castañeda, Digital Transformation Leader,
Grupo Mariposa
11
Charles River Labs:
Accelerating drug development as a digitally-enabled
trusted partner
The opportunity The solution
Accelerating drug Becoming a digitally-enabled trusted partner, putting
development and gaining customers at the core to better serve patients
efficiencies with digital
Charles River Labs set a goal to become a “digitally-enabled trusted partner”
Charles River Laboratories, a prominent pre- that integrates expertise, seamless offerings, and digital delivery to enable
clinical contract research organization (CRO), customers to accelerate the development of high-quality medicines for
plays a pivotal role in the drug development patients. To achieve this, they recognized the need to rethink their approach
ecosystem. It conducts research, in three key areas: customer engagement, internal employee interactions,
development, and safety testing before a life- and their technology foundation.
saving drug ever gets to market. Partnering
With McKinsey’s support, Charles River started with a Digital Diagnostic
with pharma, biopharma, biotech, industry,
that was focused on understanding its starting point for how it engages with
and academia, Charles River contributes to
customers externally and what customers thought about that experience,
over 85 percent of FDA-approved therapies,
internal operations required to deliver customer impact, and its capabilities
supporting companies in bringing novel
on critical enablers of these processes – from technology to talent to
treatments to market.1
data and beyond. Critical to this work was identifying the value at stake,
The world has seen innovation and uncovering customer pain points and unmet needs along the buying journey,
acceleration of therapies like the COVID- and providing digital, design, engineering, and data capabilities. These inputs
19 vaccines in ways that are extraordinary. helped build the business case across multiple initiatives and prioritize where
Historically, these therapies took 5 to 10 to start. Charles River completed this three-month diagnostic phase with
years to develop.2 Yet, during the COVID- a vision for a new digital enterprise and a minimal viable product (MVP) to
19 pandemic, society united to achieve the deliver an online customer engagement and interaction platform.
same feat in just 9 months.3 Charles River
The company spent the next three months mobilizing for the digital business
played a central role in this, which reinforced
build. The team prepared the technical foundations, user stories, and user-
its aspirational goal: what if they were able
tested digital experience to be ready for the first sprint. To ensure they had
to universally subtract a year or more out of
the right set of capabilities going forward, the team set up an “agile talent win
the drug development process? What kind
room” to quickly source new talent and upskill existing employees on agile
of scalability would that take? What kind of
methods, sprint cadence, and ceremonies, as well as the specifics of new
fundamental reimagining of processes would
roles such as product owner, product designer, and scrum master.
it take? To achieve its goals, Charles River
embarked on a transformation into a digital
enterprise, providing their pharmaceutical
clients with expertise, seamless offerings, and
digital delivery.
1. “Biologics Testing Solutions.” Charles River Labs.
2. “Vaccine Research & Development.” Johns Hopkins Coronavirus Resource Center.
3. “FDA Approves First COVID-19 Vaccine.” U.S. Food and Drug Administration, 2021.
12
The impact From there, the “digital factory” was launched—the MVP was called “Apollo,”
representing an entirely new way of engaging digitally with customers and
6 collaborating as an organization, all enabled by a new technical architecture,
including a cloud environment and master data architecture stood up by the team.
Days marketing needs to turnaround
Apollo provides customers the ability to track their projects in one place, and have
a project, down from 28 days.
access to near real-time data, so they know how each project is progressing.
This platform has enabled Charles River to build a richer relationship with clients,
200
becoming a true adviser and thought partner on their drug development journey.
Employees and leaders trained in agile In six short months, the team completed and launched the MVP. Apollo is now
work methods in less than six months. on a scale-up release phase, and in the spirit of lasting change, the company
thinks of it as a “lifestyle rather than a diet.” They continue to deploy the new agile
$100 million+
methodology across the organization to solve problems in innovative ways.
It has been three years and Charles River is successfully scaling across the entire
Annual run-rate impact identified
enterprise, including customer-facing interactions, e-commerce, employee
over an initial three years.
collaboration, lab operations, and automation in finance. They have an at-scale
3 digital factory and have expanded from 3 agile pods to more than 20 across
multiple business units and functions, dramatically accelerating the speed
Months to launch new digital products of delivering new products and services. It also has a best-in-class customer
and services, down from 12 to 18. enablement platform. And at Charles River, they remain customer-focused , using
design thinking in each product and service launched to meet customer and
employee needs. Ultimately, Charles River has successfully shifted from being a
science organization to a science and technology organization and better able to
support patients by accelerating drug development.
“Adopting design thinking and becoming
customer-centric is crucial for reimagination.
We need to start with understanding how Lessons learned
customers work with us, their environment,
Set a North star with customers and patients at the center
challenges, and successful interactions, and
Charles River found inspiration in other companies’
use that information to determine how best
transformations, using them as a North Star. Leveraging insights
to reimagine the process to meet their needs
while achieving our objectives.” Strategic from diverse industries like banking and high tech, they set
Roadmap goals, developed a rapid roadmap, and kept customers’ (and
– Mark Mintz, Corporate Senior Vice President & Chief
their patients’) needs at the center to become a valuable partner,
Information Officer, Charles River Laboratories
gaining a competitive edge.
“One thing the digital transformation has
Creating an unrivaled experience for their digital talent
done for our employees at Charles River is to
Charles River believed that providing a great experience for
allow our employees to become innovators.
clients should also extend to employees, and so the company
They can recognize aspects that could be
Agile aimed to create an outstanding environment for digital, scientific,
optimized, reconsidered from different
Operating
perspectives, and gather feedback from Model and business talent. The organization prioritized exceptional
experience by hosting lively and enjoyable agile meetings,
each other to develop improved tools and
exploring innovative ways to collaborate virtually, and celebrating
processes that significantly impact our
successes.
customers’ experience positively.”
– Pam Walker, Corporate Vice President & Global Head of
Build a digital-first mindset
Operations, Charles River Laboratories
To embrace a digital-first mindset, Charles River established a
“Adopting agile ways of working and design new digital organization with a product-centric agile model. A
Talent
thinking changed the way that we think about transformation office supported the shift to agile practices, while
technology, transforming us from long 12- to a “digital talent win room” facilitated recruitment of new expertise.
18-month deliveries to short, frequent delivery The company also engaged their best business talent, adapting
that is regularly reviewed with customers so roles and establishing external partnerships where needed.
we can quickly pivot based on the value that
we’re creating.”
Video link and case story
– Mark Mintz, Corporate Senior Vice President & Chief
Information Officer, Charles River Laboratories Coming soon
13
Allianz Direct:
Advancing as Europe’s Leading Digital Insurer
The opportunity The solution
Launching a new era for growth Transforming a digital disruptor with state-of-the-art
technology and new ways of working
Allianz Direct, the pan-European digital
insurer of global insurance leader Allianz Allianz Direct had three cornerstone goals: a fully digital business model,
Group, wanted to shape the future of online highly competitive market positioning, and an agile corporate culture,
insurance and provide a new level of service radiating the engineering mindset throughout the organization’s activities.
that could galvanize the organization
With support from McKinsey, Allianz Direct built a state-of-the-art,
and propel it to a new era of growth. To
digital platform that can be scaled across all countries in record time.
outcompete and ensure strong growth well
The platform allows teams to learn from one another as they launch new
into the future, it embarked on a daunting
products, improvements, and plug-and-play software. For customers, the
journey: it would transform ways of working
online experience is easy to use and features many time- and cost-saving
and use modern, cutting-edge technology
innovations with maximum self-service capabilities. In one example,
and advanced analytics to reimagine the end-
Allianz Direct built a flagship service—the “60-second claim”— enabled
to-end user experience, from buying the first
by AI-based loss assessment and evaluation, allowing customers to
product to filing a claim. The North Star was
process a claim in less than a minute by uploading photos and documents.
to become “digitally unbeatable” in all areas of
the value chain and thus Europe’s number one Allianz Direct built momentum in the direct insurance market in Europe
digital insurer. in just a few years by targeting two important market segments: “smart
shoppers” and “price seekers.” The business provides them with the
features they value most, including competitive pricing and a broad online
“With a combination presence.
of technical excellence, All of this was enabled by a foundational change in the organization’s
culture, operational and technical excellence, and a disrupting operating
sophisticated IT, and digital
model. McKinsey helped Allianz Direct create a talent strategy built
marketing capabilities, around hiring the best engineers. This infusion of talent was crucial to
we’ve created a strong building an agile, engineering-focused corporate culture. Today, a third
of Allianz Direct’s employees work in technology or data roles. The Allianz
foundation that will act
team created an operating model based on best-in-class technology
as the innovation engine capabilities and cross-functional agile squads responsible for creating
and marketing insurance products. The result is a highly adaptive and
for the Allianz Group.”
scalable operating model that fosters cross-market collaboration.
– Philipp Kroetz, Chief Executive Officer, Allianz Direct
14
The impact Lessons learned
15% Create a clear roadmap for deploying digital services
Allianz Direct focused its strategic roadmap on a full suite of digital
Year-over-year revenue growth self-service assets (for claims notification, claims management,
momentum (in selected countries). Strategic policy administration) equipped with best-in-class tools such as
Roadmap AI-based loss assessment and claims segmentation.
Work toward rapid implementation
30-50%
Allianz Direct teams worked in biweekly sprints. New products
were tested and implemented immediately whenever practical.
Reduction in costs due to the
Agile More than 40,000 deployments on the platform per year
scalable platform strategy. Operating
Model underline this approach.
Aim for consistency and reusability of digital assets
+90%
By building a platform that could be used across Europe, Allianz
Direct is able to scale its services and continuously improve the
Technology
Customer satisfaction ratings, after customer experience while lowering costs.
reimagining the customer experience.
Make data widely available and easy to use
Allianz Direct committed to instill a data-driven decision-making
culture, so it created easy-to-use dashboards and data-enabled
Data
performance management systems along the full value chain.
“The successful transformation can be attributed to the combination of
technical excellence, sophisticated IT infrastructure, and advanced digital
marketing capabilities, along with robust execution and global delivery in a
complaint way. We dedicated utmost attention, allocating 150% of our focus
to launch and establish our platform as a solid foundation. In addition, we
complemented the approach by emphasizing key aspects such as market
analysis, retail marketing strategies, pricing optimization, efficient damage
management, and streamlined operations to maximize our competitiveness
within the industry.”
- Christoph Weber, Chief Transformation Officer, Allianz Direct
“The most impactful decision was to be stubborn about the outcome, and to
never waiver on what good looks like. And that means you need to invest in the
best technology and in the best people, and be really stubborn about it”
– Philipp Kroetz, CEO, Allianz Direct
“We are disrupting at scale and will continue to work consistently on the
transformation of our business model, always questioning industry standards
and looking beyond our category.”
- Christoph Weber, Chief Transformation Officer, Allianz Direct
Click here to view case story
15
Xcel Energy:
Driving toward net zero with the power of digital
The opportunity The solution
Delivering a tech-enabled, Combining technology and innovation to provide safe,
sustainable future in a highly clean, and reliable energy at an affordable price
regulated environment
Xcel Energy started by developing a path forward and aspirational vision and,
Imagine it’s your first day on the job as chief then worked backward to define a set of technology investments. McKinsey
technology officer (CTO) for one of the brought technical expertise and deep experience with the nuclear power
largest electric and natural gas utilities in sector to help guide the transformation. The work centered on three clear
North America, and suddenly, one of your goals: cost savings through AI and automation; operational excellence and
core systems goes down, leading to a loss of safety; and more efficient regulatory compliance through transparency,
revenue every hour when 5 million customers accelerating to meet its baseload energy needs with zero carbon electricity.
cannot pay their bills. This is what happened
Instead of starting small, Xcel Energy took a bold approach by beginning
to Tim Peterson when he joined Xcel Energy
with one of its most complex and highly regulated domains, nuclear power.
in late 2019 as CTO. Upgrading the utility’s
The utility ini |
246 | mckinsey | generative-ai-and-the-future-of-work-in-america-vf1.pdf | McKinsey Center for Government
Generative
AI and the
future of work
in America
July 2023
Authors
Kweilin Ellingrud
Saurabh Sanghvi
Gurneet Singh Dandona
Anu Madgavkar
Michael Chui
Olivia White
Paige Hasebe
Editor
Lisa Renaud
Cover illustration by Matt Murphy
About the McKinsey Global Institute
The McKinsey Global Institute was established in 1990. Our mission is to provide a fact base to
aid decision making on the economic and business issues most critical to the world’s companies
and policy leaders. We benefit from the full range of McKinsey’s regional, sectoral, and functional
knowledge, skills, and expertise, but editorial direction and decisions are solely the responsibility
of MGI directors and partners.
Our research is grouped into five major themes:
— Productivity and prosperity: Creating and harnessing the world’s assets most productively
— Resources of the world: Building, powering, and feeding the world sustainably
— Human potential: Maximizing and achieving the potential of human talent
— Global connections: Exploring how flows of goods, people, and ideas shape economies
— Technologies and markets of the future: Discussing the next big arenas of value
and competition
We aim for independent and fact-based research. None of our work is commissioned or paid for
by any business, government, or other institution; we share our results publicly free of charge;
and we are entirely funded by the partners of McKinsey. While we engage multiple distinguished
external advisers to contribute to our work, the analyses presented in our publications are MGI’s
alone, and any errors are our own.
You can find out more about MGI and our research at www.mckinsey.com/mgi.
MGI Directors MGI Partners
Sven Smit (chair) Marco Piccitto Michael Chui Jan Mischke
Chris Bradley Olivia White Mekala Krishnan Jeongmin Seong
Kweilin Ellingrud Jonathan Woetzel Anu Madgavkar Tilman Tacke
About the McKinsey Center
for Government
With its independent and analytical approach, the McKinsey Center for Government (MCG) is
a dedicated center of excellence that helps government leaders deliver better outcomes and
experiences for their people.
Backed by a network of global experts, MCG works alongside many of the world’s leading public
sector stakeholders and organizations to enable them to operate at the highest level.
©Eloi Omella/Getty
Contents
At a glance iv
Executive summary 1
Introduction 13
1. A robust recovery marked by job switching
and labor shortages 15
2. Job gains and losses through 2030 23
3. New forces changing labor demand:
Generative AI and federal investment 31
4. Who’s vulnerable? 43
5. Preparing for the future of work 53
Methodology brief 63
Acknowledgments 67
At a glance
— During the pandemic (2019–22), the US labor market saw 8.6 million occupational
shifts, 50 percent more than in the previous three-year period. Most involved people
leaving food services, in-person sales, and office support for different occupations.
— By 2030, activities that account for up to 30 percent of hours currently worked across
the US economy could be automated—a trend accelerated by generative AI. However, we
see generative AI enhancing the way STEM, creative, and business and legal professionals
work rather than eliminating a significant number of jobs outright. Automation’s biggest
effects are likely to hit other job categories. Office support, customer service, and food
service employment could continue to decline.
— Federal investment to address climate and infrastructure, as well as structural shifts,
will also alter labor demand. The net-zero transition will shift employment away from
oil, gas, and automotive manufacturing and into green industries for a modest net gain in
employment. Infrastructure projects will increase demand in construction, which is already
short almost 400,000 workers today. We also see increased demand for healthcare workers
as the population ages, plus gains in transportation services due to e-commerce.
— An additional 12 million occupational transitions may be needed by 2030. As people
leave shrinking occupations, the economy could reweight toward higher-wage jobs. Workers
in lower-wage jobs are up to 14 times more likely to need to change occupations than those in
highest-wage positions, and most will need additional skills to do so successfully. Women are
1.5 times more likely to need to move into new occupations than men.
— The United States will need workforce development on a far larger scale as well as
more expansive hiring approaches from employers. Employers will need to hire for skills
and competencies rather than credentials, recruit from overlooked populations (such as
rural workers and people with disabilities), and deliver training that keeps pace with their
evolving needs.
McKinsey Global Institute | Generative AI and the future of work in America iv
Web 2023
future-of-work
Exhibit 1 of 21
We expect an additional 12 million occupational transitions through 2030.
US job growth, index (0=2016 levels)
40
Resilient and growing
occupations1 30
+17%
9.9M jobs 20
Stalled but rising
occupations² 10
+7%
2.8M jobs
0
–10%
Hit and declining –6.0M jobs
–10
occupations³
2016 2019 2022 2030
Growth • Healthcare demand increase • Investments in • Automation adoption
trajectory as the population ages infrastructure and the • Sustained e-commerce trend
driven by • The push toward digitization net-zero transition • Reduced need for
and technology • Demand for reskilling and customer-facing roles
• Demand for last-mile lifelong learning
delivery
Projected 1M 1M 10M
transitions⁴ From a
to new resilient and
occupations,⁵ growing
2022–30 occupation
to any other
occupation
Occupational 36% of US workers in 2022: 25% of workers: 39% of workers:
categories
• Health professionals • Builders • Production work
within each
profile • Health aides, technicians, • Creatives and arts management • Food services
and wellness
• Property maintenance • Customer service and sales
Occupations • STEM professionals
• Mechanical installation and repair • Office support
where
• Managers
generative AI • Community services
could accelerate • Transportation services • Education and workforce
automation • Business and legal training
significantly professionals
• Agriculture
1Resilient during the pandemic, 2019–22, and expected to grow between 2022 and 2030.
2Stalled during the pandemic, 2019–22, and expected to rise between 2022 and 2030.
3Hit during the pandemic, 2019–22, and continuing to decline between 2022 and 2030.
4Job transitions are defined as jobs in net declining occupations across sectors compared with the 2030 baseline.
5Even in categories that are growing overall, employment may decrease in specific occupations, requiring some workers to find new roles.
Source: O*NET; US Bureau of Labor Statistics; Current Population Survey, US Census Bureau; McKinsey Global Institute analysis
McKinsey & Company
McKinsey Global Institute | Generative AI and the future of work in America v
© Nitat Termmee / Getty
Executive summary
The US labor market is going through a rapid evolution in the way people work and the work
people do. Months after MGI released its last report on the future of work in America, the world
found itself battling a global pandemic.1 Since then, the US job market has come roaring back
from its sudden drop. The nature of work has changed as many workers have stuck with remote
or hybrid models and employers have sped up their adoption of automation technologies. More
recently, the accelerated development of generative AI, with its advanced natural language
capabilities, has extended the possibilities for automation to a much wider set of occupations.
Amid this disruption, workers changed jobs at a remarkable pace—and a subset made bigger
leaps and moved into entirely different occupations. Some 8.6 million occupational shifts took
place from 2019 through 2022. Now even more change is in store. We expect an additional
12 million occupational shifts by 2030. The total number of transitions through 2030 could be
25 percent higher than we projected a little over two years ago.2
Multiple forces are set to fuel growth in certain occupations and erode jobs in others. They
generally fall into three categories: automation, including generative AI; an injection of federal
investment into infrastructure and the net-zero transition; and long-term structural trends such
as aging, continuing investment in technology, and the growth of e-commerce and remote work.
We do not forecast how aggregated employment may be affected by the business cycle in the
short term; instead, we focus on how these forces may reshape the composition of labor demand
over the long term.
Across a majority of occupations (employing 75 percent of the workforce), the pandemic
accelerated trends that could persist through the end of the decade. Occupations that took a hit
during the downturn are likely to continue shrinking over time. These include customer-facing
roles affected by the shift to e-commerce and office support roles that could be eliminated
either by automation or by fewer people coming into physical offices. Declines in food services,
customer service and sales, office support, and production work could account for almost
ten million (more than 84 percent) of the 12 million occupational shifts expected by 2030.
Multiple forces are set to fuel
growth in certain occupations
and erode jobs in others.
1 The future of work in America: People and places, today and tomorrow, McKinsey Global Institute, July 2019.
2 The future of work after COVID-19, McKinsey Global Institute, February 2021.
McKinsey Global Institute | Generative AI and the future of work in America 1
By contrast, occupations in business and legal professions, management, healthcare,
transportation, and STEM were resilient during the pandemic and are poised for continued
growth. These categories are expected to see fewer than one million occupational shifts
by 2030.
For the other categories that account for the remaining one million occupational shifts still to
come, the pandemic was a temporary headwind. Employment in fields like education and training
should rise in the years ahead amid a continuous need for early education and lifelong learning.
Demand for construction workers also stalled during the height of the pandemic but is expected
to rebound strongly.
The changes estimated in our earlier research are happening even faster and on an even bigger
scale than expected. It is becoming even more urgent to solve occupational and geographic
mismatches and connect workers with the training they need to land jobs with better prospects.
The fact that workers have been willing to pivot and change career paths, while a tighter labor
market encouraged companies to hire from broader applicant pools, gives cause for optimism—
but not complacency. The future of work is already here, and it’s moving fast.
In a tighter labor market, workers have been moving
into new roles, accelerating occupational shifts
By the end of 2022, employment had bounced back to its 2019 level. But a great deal was in flux.
Are pandemic-era labor shortages here to stay?
The quits rate soared to new heights during the pandemic, with roughly 48 million Americans
leaving their jobs in 2021 and 51 million in 2022. What people did next is not fully evident from
the data. Some moved into better jobs with higher pay. Others left the labor force, whether out of
discouragement or for personal or health reasons, and it is unclear if or when they will return.
Total employment hit an all-time high after the pandemic, with many employers encountering
hiring difficulties. As of April 2023, some ten million positions remained vacant; labor force
participation had ticked up but was 0.7 percentage point below its prepandemic level. That
translates into roughly 1.9 million workers who are neither employed nor actively looking for jobs.
This erosion comes after an extended 20-year trend of steadily falling participation.
Labor supply may continue to be constrained, given that one in four Americans will be of
retirement age or older by 2030. Without higher participation rates, increased immigration,
or meaningful productivity growth, labor shortages could be a lasting issue as the economy
and the population grow. This remains an open question confronting markets, economists,
and employers.
Workers have shown a willingness
to change career paths, while
a tighter labor market has
encouraged companies to hire
from broader applicant pools.
McKinsey Global Institute | Generative AI and the future of work in America 2
Web 2023
future-of-work
Exhibit 2 and 7 of 21
Exhibit E1
More than 50 percent of recent occupational shifts in the United States involved workers
leaving roles in food services, customer service, office support, and production.
Estimated shifts to another occupation,
by category,¹ 2019–22, %
(XX) — Number of occupational shifts in each
occupational category, 2019–22
Health aides, technicians,
Food services (1.3M)
and wellness (700K)
>75% low-wage jobs
Hit and Resilient >75% low-wage jobs
>75% workers without college degree
declining and growing >75% workers without college degree
Number of shifts Health aides
Top 3 occupations over 2019–22 occupations Food occupations
Nursing assistants 93K
services Business
Fast food and counter workers 529K
Hit during COVID-19 and legal Resilient during COVID-19 Recreation workers 87K
Waiters and waitresses 397K and continuing to 8 professionals and continuing to grow Childcare workers 85K
Cooks 96K decline 16
6 Business and legal professionals (600K)
STEM
professionals <25% low-wage jobs
Customer service and sales (1.3M)
5 25–50% workers without college degree
>75% low-wage jobs
~8.6M total
Project management specialists 110K
>75% workers without college degree
Retail salespersons 447K Customer occupational 7 Others Sales representatives 100K
service and 15 Business operations specialists 38K
Cashiers 158K sales
shifts
Hairdressers, hairstylists, 96K
STEM professionals (400K)
and cosmetologists
5
50% faster rate of change Educators <25% low-wage jobs
Office support (1.2M) than in previous 3 years 25–50% workers without college degree
4
>75% low-wage jobs Computer systems analysts 66K
Builders
>70% workers without college degree 14 3 Computer programmers 56K
Electrical and electronic engineering 21K
Office clerks, general 443K Office 7 Community technologists and technicians
Secretaries and administrative 96K services
support 10
assistants
Others (600K)
First-line supervisors of office and 70K Others
administrative support workers <25% low-wage jobs
Production
Stalled
work 25–50% workers without college degree
but rising
Production work (900K) Light truck drivers 62K
occupations
>75% low-wage jobs Bus drivers, transit and intercity 35K
Stalled during COVID-19 School psychologists 25K
>75% workers without college degree
but starting to rise
Other categories include health professionals, managers, and
Laborers and freight, stock, 126K transportation services.
and material movers
Production helpers 68K
Machinists 66K
Education and workforce training (400K) Builders (300K) Community services (300K) Others (600K)
25–50% low-wage jobs 25–50% low-wage jobs 25–50% low-wage jobs 50–75% low-wage jobs
<25% workers without college degree >75% workers without college degree 50–75% workers without college degree >75% workers without college degree
Note: Figures may not sum to 100%, due to rounding.
1“Occupational shifts” refers to net declines in employment in specific occupations Substitute teachers 154K Carpenters 40K Correctional officers and jailers 65K Maids and housekeeping cleaners 134K
between 2019 and 2022. However, we do not know exactly how individuals moved from
one occupation to another or if they made multiple moves; for that reason, we refer to the Tutors 81K Painters, construction and maintenance 25K Lifeguards, ski patrol, and 36K Coaches and scouts 26K
n chu am nb ge er s o . f occupational shifts rather than specifying the number of workers making those Preschool teachers 25K Drywall and ceiling tile installers 14K other recreational protective Computer, automated teller, 23K
service workers
Source: O*NET; US Bureau of Labor Statistics; Current Population Survey, US Census and office machine repairers
Bureau; McKinsey Global Institute analysis
Rehabilitation counselors 25K
Other categories include agriculture, creatives and art
management, mechanical installation and repair, and
McKinsey & Company property maintenance.
McKinsey Global Institute | Generative AI and the future of work in America 3
The Great Attrition obscured deeper shifts
While most attention was focused on soaring quits rates during the pandemic, something more
structural was also occurring. A subset of people did more than change employers; they moved
into different occupations altogether. Based on net increases and decreases in employment,
some 8.6 million occupational shifts took place from 2019 through 2022—50 percent more than
in the previous three-year period (Exhibit E1).3 While it is impossible to trace individual moves,
many people left their previous roles and landed better-paying jobs in other occupations.
The majority of these shifts came from people leaving jobs in food services, customer service and
sales, office support, and production work (such as manufacturing). At the same time, managerial
and professional roles plus transportation services collectively added close to four million jobs
from 2019 to 2022. Our previous research had anticipated these types of changes over a longer
time frame, but the pandemic suddenly accelerated matters. The past few years have been a
preview of trends we expect to continue through the end of the decade.
More high-wage jobs—and fewer workers taking lower-wage service jobs
Overall employment in low- and middle-wage occupations has fallen from prepandemic levels,
while occupations that pay more than $57,000 annually added about 3.5 million jobs. However, it
is unclear how many higher-paying roles were filled by people who moved up and how many were
filled by new entrants to the labor force. Meanwhile, the number of lower-wage job openings has
not declined. Demand for lower-wage service work remains, but fewer workers are accepting
these roles.
What is clear from the job switching and occupational shifts of the past three years is that the
US labor market accommodated a higher level of dynamic movement. Spiking demand and
labor scarcity forced many employers to consider nontraditional candidates with potential and
train them if they lacked direct experience. While this may not hold in the future, employers and
workers alike can draw on what they have learned about the potential for people to make quick
pivots and add new skills.
Automation and other forces will continue to reshape the labor market
Automation, from industrial robots to automated document processing systems, continues to
be the biggest factor in changing the demand for various occupations. Generative AI is both
accelerating automation and extending it to an entirely new set of occupations. While this
technology is advancing rapidly, other forces are also affecting labor demand. Overall, we expect
significant shifts in the occupational mix in the United States through the end of the decade.
The effects of automation and generative AI
Automation has taken a leap forward with the recent introduction of generative AI tools.
“Generative” refers to the fact that these tools can identify patterns across enormous sets of data
and generate new content—an ability that has often been considered uniquely human. Their most
striking advance is in natural language capabilities, which are required for a large number of work
activities. While ChatGPT is focused on text, other AI systems from major platforms can generate
images, video, and audio.
Although generative AI is still in the early stages, the potential applications for businesses are
significant and wide-ranging. Generative AI can be used to write code, design products, create
marketing content and strategies, streamline operations, analyze legal documents, provide
customer service via chatbots, and even accelerate scientific discovery. It can be used on its own
or with “humans in the loop”; the latter is more likely at present, given its current level of maturity.
3 Measured as net job losses for individual occupations across sectors, net of estimated retirements; derived from US Bureau
of Labor Statistics (BLS) data. An administrative assistant who takes a similar position with another employer has simply
switched jobs and is not part of this analysis. If that person becomes an office manager, they have changed occupations
within the same category (office support). If they become a computer systems analyst, they have moved into a different
occupational category (STEM professionals). The latter two moves are the kind of occupational shifts that we measure. Since
we are unable to trace exactly how individual workers moved, we use net declines as a broad proxy. In our forward-looking
scenario, we refer to people needing to make transitions if demand is projected to decline in their current occupation.
McKinsey Global Institute | Generative AI and the future of work in America 4
All of this means that automation is about to affect a wider set of work activities involving
expertise, interaction with people, and creativity. The timeline for automation adoption could be
sharply accelerated. Without generative AI, our research estimated, automation could take over
tasks accounting for 21.5 percent of the hours worked in the US economy by 2030. With it, that
share has now jumped to 29.5 percent (Exhibit E2).4
4 Note that this is the midpoint, representing the average of a very wide range, from 3.7 to 55.3 percent.
Web 2023
future-of-work
Exhibit 3 and 13 of 21
Exhibit E2
With generative AI added to the picture, 30 percent of hours worked today
could be automated by 2030.
Midpoint automation adoption¹ by 2030 as a share of time spent on work activities, US, %
Automation adoption without Automation adoption with XX — Percentage-point acceleration in
generative AI acceleration generative AI acceleration automation adoption from generative AI
0 10 20 30 40
STEM professionals 16
Education and workforce training 16
Creatives and arts management 15
Business and legal professionals 14
Managers 9
Community services 9
Office support 7
Health professionals 6
Builders 6
Property maintenance 6
Customer service and sales 6
Food services 5
Transportation services 5
Mechanical installation and repair 5
Production work 4
Health aides, technicians, and wellness 4
Agriculture 3
All sectors² 8
¹Midpoint automation adoption is the average of early and late automation adoption scenarios as referenced in The economic potential of generative AI: The next
productivity frontier, McKinsey & Company, June 2023.
²Totals are weighted by 2022 employment in each occupation.
Source: O*NET; US Bureau of Labor Statistics; McKinsey Global Institute analysis
McKinsey & Company
McKinsey Global Institute | Generative AI and the future of work in America 5
Other forces affecting future labor demand
Automation is not occurring in a vacuum, of course. Other trends are affecting the demand for
certain occupations, and we expect the employment mix to change significantly through 2030,
with more healthcare, STEM, and managerial positions and fewer jobs in customer service, office
support, and food services.
— Federal investment: Recent federal legislation is driving momentum and investment in
other areas that will affect jobs.5 Reaching the net-zero emissions goal is one of these
priorities. Some 3.5 million jobs could be displaced through direct and indirect effects across
the economy. But at the macro level, these losses should be more than offset by gains of
4.2 million jobs, primarily led by capital expenditures on renewable energy. The net-zero
transition will likely be a net positive for jobs, but those jobs may be located in different places
and require different skills.
Similarly, major investment in infrastructure projects across the country will bolster
construction jobs, which could see employment growth of 12 percent from 2022 through
2030. However, the sector already had some 383,000 unfilled positions in April 2023. This
shortage will have to be addressed to bring infrastructure projects to life from coast to coast.6
The CHIPS and Science Act is putting additional funding into semiconductor manufacturing
as well as R&D and scientific research.7 This comes at a time when some companies have
been adjusting their supply chains, leading to an uptick in domestic manufacturing. While
manufacturing is likely to boost employment demand overall in the years ahead, the sector is
becoming more high-tech. It will involve fewer traditional production jobs than in the past but
more workers with technical and STEM skills.8
— Other structural trends: At the same time, other trends like rising incomes and education
levels will sustain jobs. An aging population will need more healthcare workers in multiple
roles, while the ongoing process of digitizing the economy will require adding more tech
workers in every sector.
Putting it all together, the mix of jobs is changing, and we
anticipate an additional 12 million occupational shifts
One of the biggest questions of recent months is whether generative AI might wipe out jobs. Our
research does not lead us to that conclusion, although we cannot definitively rule out job losses,
at least in the short term. Technological advances often cause disruption, but historically, they
eventually fuel economic and employment growth.
This research does not predict aggregated future employment levels; instead, we model various
drivers of labor demand to look at how the mix of jobs might change—and those results yield
some gains and some losses.9 In fact, the occupational categories most exposed to generative
AI could continue to add jobs through 2030 (Exhibit E3), although its adoption may slow their
rate of growth. And even as automation takes hold, investment and structural drivers will support
employment. The biggest impact for knowledge workers that we can state with certainty is that
generative AI is likely to significantly change their mix of work activities.
5 While our scenario includes the impact of federal investment in the net-zero transition and infrastructure, it does not include
the full impact of the CHIPS and Science Act and the Inflation Reduction Act, since implementation remained unclear at the
time of this analysis. However, both pieces of legislation point to the possibility of additional upside.
6 Garo Hovnanian, Adi Kumar, and Ryan Luby, “Will a labor crunch derail plans to upgrade US infrastructure?” McKinsey &
Company, October 2022.
7 Note that both the CHIPS and Science Act and the Inflation Reduction Act create room for additional upside in employment.
But since there is still uncertainty about their implementation as of this writing, their effects on jobs are not explicitly
incorporated into our scenario.
8 For more on this topic, see Asutosh Padhi, Gaurav Batra, and Nick Santhanam, The titanium economy: How industrial
technology can create a better, faster, stronger America, Public Affairs, 2022.
9 We rely on employment projections from the US Bureau of Labor Statistics for 2030 employment levels.
McKinsey Global Institute | Generative AI and the future of work in America 6
Web 2023
future-of-work
Exhibit E3
Exhibit 4 and 8 of 21
While STEM, healthcare, builders, and professional fields continue to add jobs,
generative AI could change work activities significantly for many occupations.
Estimated labor demand change and generative AI Midpoint automation Employment,
automation acceleration by occupation, US, 2022–30 adoption¹ by 2030, % absolute
15– 25– 35–
35 25 35 40 5M 10M
Health
professionals
30
Health aides,
technicians,
25 and wellness
STEM
professionals
20 Increasing labor demand Increasing labor demand
and modest change of and high change of work
work activities activities
15
Builders Managers Creatives and
Change
arts management
in labor
demand,² % 10 Transportation
Property
services
maintenance
Business and
legal professionals
5 Mechanical
Community
Agriculture installation services Education and
and repair workforce
training
0
Production 5 10 15 20
work Food
services
–5
Decreasing labor demand
with modest change of
work activities
–10
Customer
service
–15 and sales
Office
support
–20
Increase in automation adoption driven by generative AI acceleration, percentage points
¹Midpoint automation adoption is the average of early and late automation adoption scenarios as referenced in The economic potential of generative AI: The next
productivity frontier, McKinsey & Company, June 2023.
2We consider multiple drivers affecting demand: rising income, aging populations, technology investment, infrastructure investment (including Bipartisan
Infrastructure Law), rising education levels, net-zero transitions, marketization of unpaid work, creation of new occupations, automation (including generative AI),
increased remote working and virtual meetings, and e-commerce and other virtual transactions.
Source: US Bureau of Labor Statistics; Current Population Survey, US Census Bureau; McKinsey Global Institute analysis
McKinsey & Company
McKinsey Global Institute | Generative AI and the future of work in America 7
Resilient and growing occupational categories
The largest future job gains are expected to be in healthcare, an industry that already has an
imbalance, with 1.9 million unfilled openings as of April 2023. We estimate that there could be
demand for 3.5 million more jobs for health aides, health technicians, and wellness workers, plus
an additional two million healthcare professionals.10
By 2030, we further estimate a 23 percent increase in the demand for STEM jobs. Although
layoffs in the tech sector have been making headlines in 2023, this does not change the longer-
term demand for tech talent among companies of all sizes and sectors as the economy continues
to digitize. Employers in banking, insurance, pharmaceuticals, and healthcare, for example,
are undertaking major digital transformations and need tech workers with advanced skills.11
In addition, the transportation services category is expected to see job growth of 9 percent
by 2030.
Declining occupational categories
The biggest future job losses are likely to occur in office support, customer service, and food
services. We estimate that demand for clerks12 could decrease by 1.6 million jobs, in addition to
losses of 830,000 for retail salespersons, 710,000 for administrative assistants, and 630,000
for cashiers. These jobs involve a high share of repetitive tasks, data collection, and elementary
data processing, all activities that automated systems can handle efficiently. Our analysis also
finds a modest decline in production jobs despite an upswing in the overall US manufacturing
sector, which is explained by the fact that the sector increasingly requires fewer traditional
production jobs but more skilled technical and digital roles.13
We estimate that 11.8 million workers currently in occupations with shrinking demand may need
to move into different lines of work by 2030. Roughly nine million of them may wind up moving
into different occupational categories altogether. Considering what has already transpired, that
would bring the total number of occupational transitions through the decade’s end to a level
almost 25 percent higher than our earlier estimates, creating a more pronounced shift in the mix
of jobs across the economy.
Overall, we expect more growth in demand for jobs requiring higher levels of education and skills,
plus declines in roles that typically do not require college degrees (Exhibit E4).
Almost 12 million additional
occupational transitions may be
needed by the end of the decade.
10 Note that registered nurses, nurse practitioners, and nurse anesthetists are in the healthcare professionals category; nurse
midwives and licensed practical and licensed vocational nurses are in the health aides category.
11 Jon Swartz, “As Big Tech cuts workers, other industries are desperate to hire them,” MarketWatch, February 18, 2023; and
Steve Lohr and Tripp Mickle, “As Silicon Valley retrenches, a tech talent shift accelerates,” New York Times, December 29,
2022.
12 Note that clerks include receptionists and information clerks, general office clerks, bookkeeping, accounting, and auditing
clerks, and shipping, receiving, and inventory clerks
13 Building a more competitive US manufacturing sector, McKinsey Global Institute, April 2021.
Mc |
247 | mckinsey | mckinsey-technology-trends-outlook-2023-v5.pdf | Technology Trends
Outlook 2023
July 2023
McKinsey & Company
McKinsey & Company is a global management consulting firm, deeply committed to helping
institutions in the private, public, and social sectors achieve lasting success. For more than
90 years, our primary objective has been to serve as our clients’ most trusted external adviser.
With consultants in more than 100 cities in over 60 markets, across industries and functions, we
bring unparalleled expertise to clients all over the world. We work closely with teams at all levels
of an organization to shape winning strategies, mobilize for change, build capabilities, and drive
successful execution.
Contents
Introduction 4
The AI revolution 11 Cutting-edge engineeering 63
Applied AI 12 Future of mobility 64
Industrializing machine learning 16 Future of bioengineering 69
Generative AI 21 Future of space technologies 74
Building the digital future 26 A sustainable world 79
Next-generation software development 27 Electrification and renewables 80
Trust architectures and digital identity 32 Climate technologies beyond
electrification and renewables 80
Web3 37
Compute and connectivity frontiers 42
Advanced connectivity 43
Immersive-reality technologies 48
Cloud and edge computing 53
Quantum technologies 58
Technology Trends Outlook 2023 3
Introduction
After a tumultuous 2022 for technology investment and New and notable
talent, the first half of 2023 has seen a resurgence of
All of last year’s 14 trends remain on our list, though some
enthusiasm about technology’s potential to catalyze
experienced accelerating momentum and investment,
progress in business and society. Generative AI deserves
while others saw a downshift. One new trend, generative
much of the credit for ushering in this revival, but it
AI, made a loud entrance and has already shown
stands as just one of many advances on the horizon
potential for transformative business impact.
that could drive sustainable, inclusive growth and solve
complex global challenges. This new entrant represents the next frontier of AI.
Building upon existing technologies such as applied
To help executives track the latest developments, the
AI and industrializing machine learning, generative
McKinsey Technology Council has once again identified
AI has high potential and applicability across most
and interpreted the most significant technology trends
industries. Interest in the topic (as gauged by news and
unfolding today. While many trends are in the early
internet searches) increased threefold from 2021 to
stages of adoption and scale, executives can use this
2022. As we recently wrote, generative AI and other
research to plan ahead by developing an understanding
foundational models change the AI game by taking
of potential use cases and pinpointing the critical skills
assistive technology to a new level, reducing application
needed as they hire or upskill talent to bring these
development time, and bringing powerful capabilities
opportunities to fruition.
to nontechnical users. Generative AI is poised to add
Our analysis examines quantitative measures of as much as $4.4 trillion in economic value from a
interest, innovation, and investment to gauge the combination of specific use cases and more diffuse
momentum of each trend. Recognizing the long-term uses—such as assisting with email drafts—that increase
nature and interdependence of these trends, we also productivity. Still, while generative AI can unlock
delve into underlying technologies, uncertainties, and significant value, firms should not underestimate the
questions surrounding each trend. This year, we added economic significance and the growth potential that
an important new dimension for analysis—talent. We underlying AI technologies and industrializing machine
provide data on talent supply-and-demand dynamics learning can bring to various industries.
for the roles of most relevance to each trend. (For more,
please see the sidebar, “Research methodology,”
on page 9.)
About the McKinsey Technology Council
Technology is changing everything in our work and home lives. The McKinsey Technology Council helps
understand what is coming and how it will affect us all—taking a look around the corner toward the futures
that technology change can unlock as well as the tough questions it raises.
We look at a spectrum of technologies, from artificial intelligence to computing to biology, and their
applications across all sectors, from mining to entertainment. We also look at the science, how it
translates into engineering, and when it will accelerate to impact—at scale and around the world.
The McKinsey Technology Council brings together a global group of more than 100 scientists,
entrepreneurs, researchers, and business leaders. We research, debate, inform, and advise, helping
executives from all sectors navigate the fast-changing technology landscape. Together, we are
shaping the future.
—Lareina Yee, senior partner,
McKinsey; chair, McKinsey Technology Council
Technology Trends Outlook 2023 4
++1125%% Investment- in m1ost 4tech tre%nds tightened automotive, chemicals, financial services, and
year over year, but the potential for future life sciences—stand to potentially gain up
growth remains high, as further indicated by to $1.3 trillion in value by 2035. By carefully
the recent rebound in tech valuations. Indeed, assessing the evolving landscape and
tech trends job postings absolute ingvelsotmbeanlt sj orebma pineods sttrionngg sin considering a balanced approach, businesses
2022, at more than $1 trillion combined, can capitalize on both established and
from 2021 to 2022 from 2021 to 2022
indicating great faith in the value potential of emerging technologies to propel innovation
these trends. Trust architectures and digital and achieve sustainable growth.
identity grew the most out of last year’s 14
+12% -14%
−13% trends, increasing by nearly 50 percent as
Tech talent dynamics
security, privacy, and resilience become
increasingly critical across industries. We can’t overstate the importance of
Investment in other trends—such as applied talent as a key source in developing a
tech trends job postings global job postings AI, advanced connectivity, and cloud and competitive edge. A lack of talent is a top
from 2021 to 2022 from 2021 to 2022 edge computing—declined, but that is likely issue constraining growth. There’s a wide gap
due, at least in part, to their maturity. More between the demand for people with the skills
mature technologies can be more sensitive needed to capture value from the tech trends
to short-term budget dynamics than more and available talent: our survey of 3.5 million
nascent technologies with longer investment job postings in these tech trends found that
time horizons, such as climate and mobility many of the skills in greatest demand have
technologies. Also, as some technologies less than half as many qualified practitioners
become more profitable, they can often scale per posting as the global average. Companies
further with lower marginal investment. Given should be on top of the talent market, ready
that these technologies have applications to respond to notable shifts and to deliver a
in most industries, we have little doubt that strong value proposition to the technologists
mainstream adoption will continue to grow. they hope to hire and retain. For instance,
recent layoffs in the tech sector may present
Organizations shouldn’t focus too heavily
a silver lining for other industries that have
on the trends that are garnering the most
struggled to win the attention of attractive
attention. By focusing on only the most hyped
candidates and retain senior tech talent.
trends, they may miss out on the significant
In addition, some of these technologies
value potential of other technologies and
will accelerate the pace of workforce
hinder the chance for purposeful capability
transformation. In the coming decade, 20 to
building. Instead, companies seeking
30 percent of the time that workers spend on
longer-term growth should focus on a
the job could be transformed by automation
portfolio-oriented investment across the
technologies, leading to significant shifts
tech trends most important to their business.
in the skills required to be successful. And
Technologies such as cloud and edge
companies should continue to look at how
computing and the future of bioengineering
they can adjust roles or upskill individuals
have shown steady increases in innovation
to meet their tailored job requirements. Job
and continue to have expanded use cases
postings in fields related to tech trends grew
across industries. In fact, more than 400 edge
at a very healthy 15 percent between 2021 and
use cases across various industries have been
2022, even though global job postings overall
identified, and edge computing is projected to
decreased by 13 percent. Applied AI and next-
win double-digit growth globally over the next
generation software development together
five years. Additionally, nascent technologies,
posted nearly one million jobs between
such as quantum, continue to evolve and
2018 and 2022. Next-generation software
show significant potential for value creation.
development saw the most significant growth
Our updated analysis for 2023 shows that
in number of jobs (Exhibit 1).
the four industries likely to see the earliest
economic impact from quantum computing—
Technology Trends Outlook 2023 5
Web <2023>
<ETxehchibTrietn 1d s-L1>
Exhibit <1> of <3>
Job postings for fields related to tech trends grew by 400,000 between 2021
and 2022, with generative AI growing the fastest.
Tech trend job postings,1 2021–22, thousands
700
600
+6%
500
+29%
400
+12%
300
+16% +15%
200
100
2021 2022
0
Applied AI Next-generation Cloud and edge Trust architectures Future of
software development computing and digital identity mobility
300
200 +27%
100 +8% +7%
+10% +23%
0
Electrification and Climate tech beyond Advanced Immersive-reality Industrializing
renewables electrification and connectivity technologies machine learning
renewables
200
+40% +16% +44% +12%
100
–19%
0
Web3 Future of Future of space Generative AI Quantum
bioengineering technologies technologies
1Out of 150 million surveyed job postings. Job postings are not directly equivalent to numbers of new or existing jobs.
Source: McKinsey’s proprietary Organizational Data Platform, which draws on licensed, de-identified public professional profile data
McKinsey & Company
This bright outlook for practitioners in most fields The talent crunch is particularly pronounced for trends
highlights the challenge facing employers who are such as cloud computing and industrializing machine
struggling to find enough talent to keep up with their learning, which are required across most industries.
demands. The shortage of qualified talent has been a It’s also a major challenge in areas that employ highly
persistent limiting factor in the growth of many high- specialized professionals, such as the future of mobility
tech fields, including AI, quantum technologies, space and quantum computing (Exhibit 2).
technologies, and electrification and renewables.
Technology Trends Outlook 2023 6
Exhibit 2
Most fields related to these tech trends require skills where talent supply is
low, while only a few fields have a talent surplus.
Availability of qualified talent, by skill required per
tech trend,¹ ratio of profiles to job postings Rank 1 2 3
Talent Talent
deficit Machine learning (ML) surplus
Data science
Applied AI
TensorFlow
Amazon Web Continuous
Services integration
Next-generation
software development Infrastructure Cloud
managementAmazon Web Services computing
Cloud and edge computing
Cloud
Risk Regulatory Computer computing
Trust architectures and analysis compliance security
digital identity
Maintenance
Manufacturing
Future of mobility
Automotive industry
Contract management
Electrification and renewables
Photovoltaics Renewable
energy
Climate tech beyond
electrification and renewables Sustainability Energy
Regulatory efficiency
compliance Kubernetes
Advanced connectivity
Telecommunications
Internet of Things
Immersive-reality technologies
Product Computer Graphic
engineering vision design
Industrializing machine learning
PyTorch TensorFlow
ML
Web3
Stakeholder Cloud
management computing Blockchain Molecular
biology
Future of bioengineering
Pharmaceuticals Gene therapy
Remote
Aerospace engineering sensing
Future of space technologies
Aerospace industries
Python ML
Generative AI
Regulatory
Python
compliance
Cloud computing Quantum computing
Quantum technologies
<0.1:1 0.1:1 0.2:1 0.4:1 0.6:1 1:1 2:12 4:1 6:1 8:1
¹The ratio of online profiles claiming each trend’s most needed tech skills to all job postings requiring skill (logarithmic scale).
²Benchmark: 2 profiles with skill per job posting. Average talent supply–demand ratio benchmark based on skills listed for the 20 most common jobs.
Source: McKinsey’s proprietary Organizational Data Platform, which draws on licensed, de-identified public professional profile data
McKinsey & Company
Technology Trends Outlook 2023 7
The 15 tech trends combinations, there’s significant power and potential in
looking across these groupings.
This report lays out considerations for all 15 technology
trends. We grouped them into five broader categories To describe the state of each trend, we developed
to make it easier to consider related trends: the AI scores for innovation (based on patents and research)
revolution, building the digital future, cutting-edge and interest (based on news and web searches). We also
engineering, compute and connectivity frontiers, and a counted investments in relevant technologies and rated
sustainable world. Of course, when considering trend their level of adoption by organizations (Exhibit 3).
Exhibit 3
We described each trend by scoring innovation and interest, and we also
counted investments and rated their level of adoption by organizations.
Innovation, interest, investment, and adoption, by technology trend, 2022
1.0
Adoption rate, score
(0 = no adoption; 5 =
Applied AI mainstream adoption)
0.8 0 1 2 3 4 5
Future of Advanced
0.6 bioengineering connectivity
Innovation,1 score
(0 = lower;
1 = higher) Electrification
Quantum technologies
and renewables
0.4 Industrializing machine learning
Next-generation software development
Cloud and edge computing
Future of
Immersive-reality technologies
mobility
0.2
Climate tech beyond electrification & renewables
Equity investment, $ billion
Trust architectures and digital identity
Future of space tech
Web3
Generative AI
250 150 75
0
0 0.01 0.10 1.00
0 0.2 0.4 0.6 0.8 1.0
Interest,2 score
(0 = lower; 1 = higher)
Note: Innovation and interest scores for the 15 trends are relative to one another. All trends exhibit high levels of innovation and interest compared with
other topics and are also attracting significant investment.
1The innovation score combines the 0–1 scores for patents and research, which are relative to the trends studied. The patents score is based on a measure
of patent filings, and the research score is based on a measure of research publications.
2The interest score combines the 0–1 scores for news and searches, which are relative to the trends studied. The news score is based on a measure of news
publications, and the searches score is based on a measure of search engine queries.
McKinsey & Company
Technology Trends Outlook 2023 8
Research methodology
To assess the development of each technology trend, our team collected data on five tangible
measures of activity: search engine queries, news publications, patents, research publications,
and investment. For each measure, we used a defined set of data sources to find occurrences of
keywords associated with each of the 15 trends, screened those occurrences for valid mentions
of activity, and indexed the resulting numbers of mentions on a 0–1 scoring scale that is relative
to the trends studied. The innovation score combines the patents and research scores; the
interest score combines the news and search scores. (While we recognize that an interest
score can be inflated by deliberate efforts to stimulate news and search activity, we believe that
each score fairly reflects the extent of discussion and debate about a given trend.) Investment
measures the flows of funding from the capital markets into companies linked with the trend.
Data sources for the scores include the following:
— Patents. Data on patent filings are sourced from Google Patents.
— Research. Data on research publications are sourced from the Lens (www.lens.org).
— News. Data on news publications are sourced from Factiva.
— Searches. Data on search engine queries are sourced from Google Trends.
— Investment. Data on private-market and public-market capital raises are sourced
from PitchBook.
— Talent demand. Number of job postings is sourced from McKinsey’s proprietary
Organizational Data Platform, which stores licensed, de-identified data on professional
profiles and job postings. Data is drawn primarily from English-speaking countries.
In addition, we updated the selection and definition of trends from last year’s study to reflect the
evolution of technology trends:
— The generative-AI trend was added since last year’s study.
— We adjusted the definitions of electrification and renewables (previously called future of
clean energy) and climate technologies beyond electrification and renewables (previously
called future of sustainable consumption).
— Data sources were updated. This year, we included only closed deals in PitchBook
data, which revised downward the investment numbers for 2018–22. For future of
space technologies investments, we used research from McKinsey’s Aerospace &
Defense Practice.
Technology Trends Outlook 2023 9
About the authors
Michael Chui Mena Issler Roger Roberts Lareina Yee
McKinsey Global Institute Associate partner, Partner, Senior partner, Bay Area; chair,
Partner, Bay Area Bay Area Bay Area McKinsey Technology Council
The authors wish to thank the following McKinsey colleagues for their contributions to this research:
Bharat Bahl Jonathan DePrizio Naomi Kim Tanya Rodchenko
Soumya Banerjee Ivan Dyakonov Jesse Klempner Lucy Shenton
Arjita Bhan Torgyn Erland Kelly Kochanski Henning Soller
Tanmay Bhatnagar Robin Giesbrecht Matej Macak Naveen Srikakulam
Jim Boehm Carlo Giovine Stephanie Madner Shivam Srivastava
Andreas Breiter Liz Grennan Aishwarya Mohapatra Bhargs Srivathsan
Tom Brennan Ferry Grijpink Timo Möller Erika Stanzl
Ryan Brukardt Harsh Gupta Matt Mrozek Brooke Stokes
Kevin Buehler Martin Harrysson Evan Nazareth Malin Strandell-Jansson
Zina Cole David Harvey Peter Noteboom Daniel Wallance
Santiago Comella-Dorda Kersten Heineke Anna Orthofer Allen Weinberg
Brian Constantine Matt Higginson Katherine Ottenbreit Olivia White
Daniela Cuneo Alharith Hussin Eric Parsonnet Martin Wrulich
Wendy Cyffka Tore Johnston Mark Patel Perez Yeptho
Chris Daehnick Philipp Kampshoff Bruce Philp Matija Zesko
Ian De Bode Hamza Khan Fabian Queder Felix Ziegler
Andrea Del Miglio Nayur Khan Robin Riedel Delphine Zurkiya
They also wish to thank the external members of the McKinsey Technology Council.
Technology Trends Outlook 2023 10
The AI revolution
Technology Trends Outlook 2023 11
Applied AI
The trend—and why it matters McKinsey Global Survey on the state of AI shows that the
proportion of responding organizations adopting AI more
With AI capabilities, such as machine learning (ML),
than doubled from 20 percent in 2017 to 50 percent in 2022.
computer vision, and natural-language processing (NLP),
The 2022 survey also indicated that adopting AI can have
companies in all industries can use data and derive insights
significant financial benefits: 25 percent of respondents
to automate processes, add or augment capabilities, and
attributed 5 percent or more of their companies’ EBIT to AI.
make better decisions. McKinsey research estimates the
However, organizational, technical, ethical, and regulatory
potential economic value at stake from applied AI to be
issues should be resolved before businesses can realize the
$17 trillion to $26 trillion, and the share of companies
technology’s full potential.
pursuing that value has been increasing. The annual
Applied AI Score by vector (0 = lower; 1 = higher)
Scoring the trend
Talent demand News
High innovation and investment scores for
applied AI are commensurate with its large
potential impact. Each year from 2018 to 2022,
applied AI has had the highest innovation scores
of all the trends we studied, and its investment
score also ranks in the top five. Perhaps Equity Searches
unsurprisingly, in 2022, demand for talent in investment 0.2
0.4
applied AI was also highest among all trends. 0.6
0.8
1.0
Adoption rate score, 2022
Patents Research
0 1 2 3 4 5
None Mainstream
1.0
Equity investment, Job postings,
2022, 2021–22, 0
2018 2022
$ billion % difference
104 +6
Talent demand Ratio News Press reports
Industries affected: Aerospace and defense; of actual skilled people featuring trend-
Agriculture; Automotive and assembly; Aviation, to job vacancies related phrases
travel, and logistics; Chemicals; Construction and
Equity investment Searches Search
building materials; Consumer packaged goods;
Private- and public- engine queries for
Education; Electric power, natural gas, and utilities;
market capital raises for terms related to
Financial services; Healthcare systems and
relevant technologies trend
services; Information technology and electronics;
Media and entertainment; Metals and mining; Oil Patents Patent Research Scientific
and gas; Pharmaceuticals and medical products; filings for technologies publications on topics
Public and social sectors; Real estate; Retail related to trend associated with trend
Telecommunications
Technology Trends Outlook 2023 12
Latest developments — Global AI adoption plateaus—for now. While AI adoption
globally is more than double that in 2017, the proportion
These are some recent developments involving applied AI:
of organizations using AI has leveled off to around
— Investment fuels enhanced AI capabilities. Although 50 percent to 60 percent in recent years. However,
investments in AI were down to $104 billion in 2022 companies that have already adopted AI nearly doubled
from a high of $146.8 billion in 2021, they continue to the number of capabilities they use, such as natural-
pace ahead of 2018–20 levels, which averaged language generation or computer vision, from 1.9 in 2018
$73.5 billion. With investments flowing, AI continues to 3.8 in 2022.2
to post state-of-the-art results with continuous
improvements in areas such as model accuracy. For
example, the cost to train image classification systems
has decreased by 63.6 percent, and training times
have improved by 94.4 percent since 2018.1 However,
additional potential for applied AI could be unlocked
by combining it with new emerging AI technology. For
example, the foundation models underlying generative
AI could process large amounts of unstructured
manufacturing data, such as notes and logs, to enrich
current AI solutions that optimize performance.
— Policy makers accelerate regulatory actions to curb
AI misuse. As AI technology advances, so too has its
potential for misuse: the AIAAIC Repository, which
tracks incidents related to the ethical misuse of AI,
algorithms, and automation, indicates that the number
of controversies involving AI has increased by 26 times
since 2012. Algorithmic fairness, bias, and misuse have ‘We haven’t found an industry
become mainstream concerns. An analysis of legislative
or business function that couldn’t
records in 127 countries shows that the number of laws
passed containing the words “artificial intelligence” enhance its performance through
grew from one in 2016 to 37 in 2022. Prompted by the
applying AI. But capturing the
accelerated development of AI by private firms, the
European Union’s AI Act—which regulates foundational value of AI is a journey that
AI models—is nearing law status following parliamentary
requires taking action across
committee approval. Meanwhile, the McKinsey Global
multiple dimensions, from
Survey on the state of AI indicates that there has been
no substantial increase in organizations’ reported talent to technology.’
mitigation of AI-related risks relative to the increase
in AI use. – Michael Chui, partner, Bay Area
1 Daniel Zhang et al., Artificial Intelligence Index Report 2022, AI Index Steering Committee, Stanford Institute for Human-Centered AI, Stanford University,
March 2022.
2 “The state of AI in 2022—and a half decade in review,” QuantumBlack, AI by McKinsey, December 6, 2022.
Technology Trends Outlook 2023 13
Talent market
Applied AI
Demand
Applied AI has seen rapid growth in demand for talent, with job postings more than tripling since 2018. Demand
for data scientists and software engineers grew significantly in 2021 and saw moderate growth in 2022.
Job postings by title, 2018–22, thousands
100 Data scientist
Software engineer
80 Data engineer
Software developer
60
Customer service representative
Project manager
40
Scientist
Product manager
20
0
2018 2022
Skills availability
The demand for practitioners of machine learning, data science, NLP, and some associated tools is high
compared with supply.
Talent availability, % share of postings requiring skill
7
21 3 3 2 2 2
Machine Data TensorFlow Computer Natural-language Deep PyTorch
learning science vision processing learning
Talent availability, ratio of talent to demand
13.2
1.5 1.7 2.0
0.9 0.8 1.3
Machine Data TensorFlow Computer Natural-language Deep PyTorch
learning science vision processing learning
Technology Trends Outlook 2023 14
‘Applied AI has the potential to become more valuable and
useful to companies in combination with generative AI. A key
marker for the future will be how synergies between the two
are captured to maximize value capture across organizations.’
– Carlo Giovine, partner, London
In real life Key uncertainties
Real-world examples involving the use of applied AI The major uncertainties affecting applied AI include
include the following: the following:
— Emirates Team New Zealand dramatically accelerated — Lack of available resources, such as talent and funding,
hydrofoil design and testing by using AI to train a “digital might affect the pipeline of AI applications, despite
twin”—a digital replica of a sailor—to test designs in technical advances in solutions for industrializing ML
a simulated environment. By using the AI “sailor” to and in IT infrastructure.
remove the bottleneck of human sailors performing the
— Cybersecurity and privacy concerns, notably on data
tests, the team reduced costs by 95 percent and was
risks and vulnerabilities, are prevalent—51 percent of
able to test ten times as many designs.
survey respondents cited cybersecurity as a leading
— Freeport-McMoRan deployed a custom-built AI model risk in 2022.
loaded with three years’ worth of operating data to
— Regulation and compliance might affect AI research
optimize production processes and total output at a
and applications.
copper mill. In doing so, it increased production by
10 percent while reducing capital expenditures on a — Ethical considerations—including data governance,
planned expansion. equity, fairness, and “explainability”—surround the
responsible and trustworthy use of AI.
— Telkomsel built a new data analytics platform
supplemented by AI-driven tools to better understand
customers across thousands of microsegments. Using Big questions about the future
9,000 data points per customer across more than
Companies and leaders may want to consider a few
50 models, the company drives personalization by
questions when moving forward with applied AI:
identifying the right way to interact with customers and
offering the most relevant products and services. — How might companies better determine which
AI applications benefit them and their
stakeholders most?
Underlying technologies
— What features make AI trustworthy and responsible
AI comprises several technologies that perform cognitive-
and how should they be integrated into applications?
like tasks. These include the following:
— What checks should companies put in place to guard
— Machine learning (ML). This term refers to models that
against AI-related risks associated with data privacy
make predictions after being trained with data rather
and security, equity, fairness, and compliance?
than following programmed rules.
— How will companies use generative AI in combination
— Computer vision. This type of ML works with visual
with applied AI to maximize potential synergies or
data, such as images, videos, and 3-D signals.
differentiate when it makes sense to use one approach
— Natural-language processing (NLP). This type of ML over the other?
analyzes and generates language-based data, such as
text and speech.
— Deep reinforcement learning. This type of ML uses
artificial neural networks and training through trial and
error to make predictions.
Technology Trends Outlook 2023 15
Industrializing machine learning
The trend—and why it matters solutions, identify and resolve issues in production, and
improve teams’ productivity. Experience suggests that
Industrializing machine learning (ML), commonly referred
organizations that industrialize ML successfully can shorten
to as ML operations, or MLOps, refers to the engineering
the production time frame for ML applications (from proof of
practices needed to scale and sustain ML applications in
concept to product) by about eight to ten times and reduce
an enterprise. These practices are enabled and supported
development resources by up to 40 percent.3 Industrialized
by an ecosystem of technical tools that is rapidly improving,
ML was pioneered by a small number of leading companies,
both in functionality and interoperability. MLOps tools can
but adoption is now spreading as more companies use AI for
help companies transition from pilot projects to viable
a wider range of applications.
business products, accelerate the scale-up of analytics
Industrializing machine Score by vector (0 = lower; 1 = higher)
learning
Talent demand News
Scoring the trend
Scores across news, searches, publications,
and patents increased significantly, while
demand for talent has nearly quadrupled in
the same time frame. These increases suggest Equity Searches
0.1
investment
that the use of methods for industrializing ML 0.2
0.3
could widen in the years ahead 0.4
0.5
Adoption rate score, 2022
Patents Research
0 1 2 3 4 5
None Mainstream 1.0
0.5
0
Equity investment, Job postings, 2018 2022
2022, 2021–22,
$ billion % difference
3 +23
Talent demand Ratio News Press reports
of actual skilled people featuring trend-
to job vacancies related phrases
Industries affected: Aerospace and defense;
Equity investment Searches Search
Automotive and assembly; Electric power,
Private- and public- engine queries for
natural gas, and utilities; Financial services;
market capital raises for terms related to
Information technology and electronics;
relevant technologies trend
Media and entertainment; Metals and
mining; Oil and gas; Pharmaceuticals and Patents Patent Research Scientific
medical products; Telecommunications filings for technologies publications on topics
related to trend associated with trend
3 Based on observations from ML operations deployment in a series of large-scale analytics transformations supported by McKinsey.
Technology Trends Outlook 2023 16
Latest developments
These are some recent developments involving
industrializing ML:
— Companies increasingly commit to industrializing ML.
Investments into companies in the ML industrialization
space reached a high of $4.7 billion in 2021 and
remained strong throughout 2022 at a cumulative
$3.4 billion. With investments flowing, ML decision
makers have also doubled down on their commitments:
85 percent of respondents to a ClearML survey
‘We are at an inflection point
indicated that they had a dedicated MLOps budget in
2022. IDC predicts that 60 percent of enterprises will with artificial intelligence.
have implemented MLOps by 2024. Such investments
Generative AI has captured
could prove wise, as our own research finds that
companies seeing higher returns from AI are more both mainstream and business
likely to engage in ML industrialization.
imaginations. Organizations
— The ecosystem rapidly evolves through acquisitions and
that are willing to continuously
new offerings. The year 2022 was marked by significant
consolidation, partnerships, and new releases. Altair learn and adapt their processes,
acquired RapidMiner, Snowflake acquired Myst AI,
ways of working, and technology
McKinsey acquired Iguazio, and Hewlett Packard
Enterprise acquired Pachyderm. Databricks announc |
249 | mckinsey | EN_McKinsey_GenAI_Implications_Germany_Labor_Market.pdf | Effects of GenAI
on the German
labor market
An opportunity to mitigate skilled labor shortages
November 2023
CONFIDENTIAL AND PROPRIETARY
Any use of this material without specific permission
of McKinsey & Company is strictly prohibited
An opportunity to
mitigate skilled
labor shortages
1
Germany is experiencing Share of businesses affected by skilled labor shortageshave increased 5-fold p.4
since 2009.
skilled labor shortages
General open positions have increased 4-fold since 2004 p.5
2
GenAI can unlock and boost GenAI has the potential to greatly enhance Germany's competitiveness by p.12
boosting productivity growth by an estimated 18%. GenAI can also help
productivity to mitigate
address skilled labor shortages through innovation.
these shortages
This primarily concerns: Professions in workforce training, STEM, and p.15
healthcare that have both the greatest need (represented by the share of job
vacancies per share of employment in an occupation group – of >0.9) and
greatest GenAI potential (>17 pp) for labor shortage mitigation
Greatest profitable effects for a) employees in highly professionalized careers p.14
(e.g., legal and business, 36 pp) and b) higher education (e.g., tertiary p.13
education, 24 pp), as well as c) high-earning employees (e.g., top earners, 12 p.16
pp)
3
Germany boasts the highest number of GenAI startups (>500) in the EU p.20
Germany has a promising
private sector
landscape for GenAI adoption
Germany holds a top-five global ranking in computing power, academic
p.22
publications, and patents, demonstrating its competitiveness in tech and
research.
Germany ranks second among OECD countries in AI skill penetration, with p.23
1.7 out of every 100 workers reporting AI skills, only slightly behind the United
States.
McKinsey & Company 2
The skilled labor shortage in
Germany
Agenda
The potential of GenAI to
increase productivity
The GenAI landscape in Germany
McKinsey & Company 3
At the end of 2022, ~50% of businesses reported they had been
affected by skilled labor shortages, marking a 5x increase since 2009
Manufacturing industry Retail Services Construction Wholesale
Share of enterprises affected by skilled labor shortages according to sectors in Germany,
Reported to the ifo Institute, percent
60
50
40
5x increase
30
20
10
0
2004 06 08 10 12 14 16 18 20 2022
Source: ifo Institute, ifo Konjunkturumfragen McKinsey & Company 4
Skilled labor shortage sentiment is corroborated by reported
open positions quadrupling between 2004 and 2022
Absolute number of open positions in Germany,
Reported to the Federal Employment Agency in Germany, thousands
900
800 4x increase
700
600
500
400
300
200
100
0
2004 06 08 10 12 14 16 18 20 2022
Source: Federal Employment Agency (Bundesagentur für Arbeit), labor market in numbers McKinsey & Company 5
The skilled labor shortage in
Germany
Agenda
The potential of GenAI to
increase productivity
The GenAI landscape in Germany
McKinsey & Company 6
Background: GenAI is the natural evolution of analytical AI,
addressing a novel set of challenges to realize large automation
potential, thus unlocking meaningful productivity potential
Analytical AI Generative AI
Analytical AI algorithms are used GenAI algorithms are used to
to solve analytical tasks faster either create new content on par
and more efficiently than with humans, or greatly enhance
humans — e.g., being able to humans' abilities — e.g.,
classify, predict, cluster generating audio, code, images,
or evaluate data text, and videos
Forecasting Segmenting Conducting Designing Creating Generating
sales customers sentiment concepts marketing or code
analyses social media copy
McKinsey & Company 7
Example: By unlocking productivity potential, GenAI can address skilled
labor shortages in manufacturing, resulting in fewer vacancies due to
more internal task completion
Illustrative – computer engineer
01
Sara’s current job as a computer engineer
Sara is a computer engineer for a manufacturing company who shifts
between 17 unique activities, including testing the performance of
electrical equipment and collaborating with technical personnel. Her
company is struggling to find skilled personnel.
03 Sara’s time rearrangement and productivity
gains
02
Sara’s company adopts new technologies With automation, the resulting free time creates increased
productivity and innovation: Sara can now operate an
Sara's company invests in real-time data analytics and
adjacent workstation, which is underutilized, as her company
machine-learning software to help monitor the computer
has not been able to recruit a suitably skilled new colleague.
systems in the manufacturing plant. The company also
Moreover, she invents a novel solution to a computer-design
purchases several robotics and automation systems to
problem at the plant.
streamline production.
Fewer vacancies and more innovation in 2030 04
Various workflows have been optimized. Thus, numerous positions are now covered
internally where the company had previously struggled to find suitably skilled
colleagues. Moreover, Sara's company has implemented various computer-design
improvements, which speeds up production.
McKinsey & Company 8
Example: By unlocking productivity potentials, GenAI can meet skilled
labor shortages in workforce training resulting in less vacancies and
better apprentice performance and satisfaction
Less vacancies and better apprentice
04
Illustrative – educator and workforce training
performance and satisfaction in 2030
Various individualized courses and modules have been
implemented across the organization's workforce training portfolio.
Hence, now significantly less trainer input is required. Therefore,
the average performance of apprentices has increased, and the
personal satisfaction of apprentices has improved as they now
receive tailored training while having more space for deeper
exchange with trainers on a more personal level..
John's organization has adopted
02
new technologies
03
John's organization invests in educational John's time rearrangement and productivity
generative AI software which can analyze the
gains
needs, constraints, and preferences of each
apprentice, and subsequently offers tailored With automation, the resulting free time creates increased
content and learning styles. Moreover, the productivity and innovation: John can now increase the
new software can create simulation-based number of apprentices under his supervision from 20 to 30
and individualized trainings with much less which is great for the organization, as it has been struggling to
input from John. recruit another workforce trainer. Moreover, he implemented
data-driven informed development conversations and
additionally introduced a new innovative course offering
individualized remote work simulation.
01
John's current job as a workforce trainer
John is a workforce trainer in a vocational school who shifts between 13 unique
activities, including frontal teaching, preparing individual work samples, development
conversations, and assessing the apprentices' individual outputs. His organization
struggles to find skilled trainers.
McKinsey & Company 9
To assess GenAI productivity potentials,
we analyzed around 2,100 distinct work
activities and ~850 professions
ILLUSTRATIVE
Capability requirements
Physical
Answers about
Professions Fine motor skills/dexterity
products and services
Gross motor skills
Navigation
Mobility
Employees in retail and sales Greet customers
Sensory
Sensory perception
Employees in food and
Clean and maintain work areas Cognitive
beverage service
Retrieving information
Recognizing known patterns/categories (supervised learning)
Generating novel patterns/categories
Teachers Demonstrate product features
Logical reasoning/problem solving
Optimizing and planning
Creativity
Health practitioners Process sales and transactions
Articulating/display output
Coordinating with multiple agents
Natural language processing (NLP)
... ...
… … Understanding natural language
… … Generating natural language
Social
~850 ~2,100 activities assessed across all Social and emotional sensing
professions
Social and emotional reasoning
professions
Emotional and social output
Source: National Labor Offices, Occupation Information Network; McKinsey Global Institute analysis McKinsey & Company 10
In Germany, GenAI promises greater productivity potential in
complex processes, such as decision making and collaboration…
With GenAI Without GenAI1
Overall technical automation potential, comparison by midpoint scenarios, percent
Activity groups
55
Decision Applying expertise2
19 +36 pp
making and Disclaimer: Technical automation
collaboration potential implies the availability of
50
Managing3 technological capabilities required to
16 +34 pp
automate a particular work activity, hence,
affecting hours spent on that work activity
Interfacing with 50
stakeholders 25 +25 pp
Data 92
Processing data
management 75 +17 pp
79
Collecting data
65 +14 pp
Physical Performing unpredictable 34
physical work4 +1 pp 33
Performing predictable 70
physical work5 +2 pp 68
1. Previous assessment of work automation before the rise of GenAI, including analytical AI, 3. Managing and developing people
machine leanrning, and deep learning 4. Performing physical activities and operating machinery in unpredictable environments.
2. Applying expertise to decision making, planning, and creative tasks 5. Performing physical activities and operating machinery in predictable environments
Note: Figures may not sum, because of rounding
McKinsey & Company 11
Source: McKinsey Global Institute analysis
…thus, GenAI makes it possible to contribute significantly to
Germany's competitiveness
With GenAI Without GenAI2
Productivity impact from automation by scenario, 2022-40, CAGR,1 percent
Developed economies Key implications
USA France Austria Global3 Germany for Germany
3.6 3.7 3.7 3.9 Early (vs. late) adoption of
0.7 0.7 0.6 automation potential will lead
0.6
to an additional ~EUR
3.3
2.9 3.0 0.8 3.1 0.9 2,600bn in GDP by 2040
0.6
0.2 0.2 0.6
0.4 0.3 0.6 0.7 Early additional adoption of
Early Late Early Late Early Late GenAIalone can increase
Germany's GDP by
Emerging economies
~EUR 585bn (13%) by 2040
China India Mexico 3.4
GenAI can increase
1.3
3.8 2.6 automation impact on
0.2
0.6 2.3 2.9 productivity growth by
0.5 0.6
~18%, significantly advancing
3.2 0.8 1.1 Germany's competitive
2.3 0.2
0.1 1.8 position
0.7 0.1
0.1
Early Late Early Late Early Late Early Late
Early4 Late4
1. Based on the assumption that automated work hours are reintegrated into work at today's productivity level
2. Previous assessment of work automation before the rise of GenAI
3. Based on 47 countries, representing about 80% of global employment
4. Automation scenarios (early: early adoption of GenAItechnology capabilities; late: late adoption of GenAItechnology capabilities, expert based)
Note: Figures may not sum, because of rounding.
Source: Oxford Economics; The Conference Board Total Economy Database; McKinsey Global Institute analysis McKinsey & Company 12
Education: Greatest labor shortage mitigation potential for tertiary
education level while societally for high school education level
With GenAI Without GenAI1
Key implications
Impact of GenAI on technical automation potential in midpoint scenario, 2023, percent for Germany
Overall technical Additional Population-weighted Additional impact of GenAI is
Education automation potential, automation Share of skilled labor shortage expected to be highest for
level comparison2 potential,2 pp population, % mitigation potential those with tertiary-level
education (24 pp)
Tertiary
60 Example: Computer engineers
education
24 19 Medium (STEM) like Sara or workforce
(Bachelor and
36 trainers like John
above)
The population-weighted
skilled labor shortage
High school 64
mitigation potential is highest
(Diploma or 13 56 Large
for high-school-degree holders
equivalent) 51
(55.9% population share)
Example: Community health
63 care worker or pharmacy
None
9 25 Small technician
(No degree)
54
1. Previous assessment of work automation before the rise of GenAI
2. Based on US extrapolation
Source: StatistischesBundesamt(DeStatis); McKinsey Global Institute analysis McKinsey & Company 13
Professions: GenAI holds the greatest opportunities for workforce
training, business and legal, and STEM
With GenAI Without GenAI1
Impact of GenAI on automation potential sorted by additional GenAI potential, percent
Employment-weighted
Overall technical automation potential, comparison Automation Share of German skilled labor shortage Worldautomation
Professions bymidpoint scenarios, 2023,% potential shift, pp employment % mitigation potential potential shift,pp Key implications
Educator and workforce training 14 54 40 3 Medium 39 for Germany
Business/legal professionals 32 68 36 6 High 30
Greatest skilled labor shortage
51
Creatives and arts management 22 29 1 Medium 25
mitigation potential in Germany in
57
STEM professionals 28 29 7 High 29 the areas of workforce training
Office support 85 22 19 High 21 (40 pp), business and legal
63
(36 pp), and STEM (29 pp)
67
Community services 45 22 6 High 26
49 Employment-weighted labor
Managers 29 20 3 Medium 17
shortage mitigation potential in
Health professionals 45 17 2 Medium 14
28
Germany is largest for
70
Customer service and sales 60 10 9 Medium 12 business and legal, STEM,
Production work 67 75 8 13 Medium 9 office support, and community
Property maintenance 37 7 4 Low 9 services
30
59 Based on relative employment in
Transportation services 53 6 3 Low 7
Germany and the world,
40
Health aides, technicians, and wellness 34 6 9 Medium 9
STEM and community services
Builders 55 5 5 Low 4
50 might profit more in Germany,
Food services 61 66 5 4 Low 8 while workforce training and
68 customer service might profit less
Mechanical installation and repair 63 5 4 Low 6
than the global average
Agriculture 6266 4 1 Low 4
Total 49 65 16 100 12
1.Previous assessment of work automation before the rise of GenAI. | Note: Figures may not sum, because of rounding. McKinsey & Company 14
Source: McKinsey Global Institute analysis
Profession: GenAI has greatest labor shortage mitigation potentials in high
job vacancy concentration areas, such as educator training, STEM, and Health
Job-vacancy concentration2 and corresponding automation shift in Germany
Additional technical automation potential, pp Key implications
40 for Germany
Medium need – high potential High need – high potential
Educator and
workforce training
35
Business/ Highest need (≥0.9) and
legal professionals potential for GenAI for
30
skilled labor shortage
Creatives and arts management STEM professionals
mitigation in workforce
25 training (40 pp), STEM (29
Community services pp), and health
20 Office professions (17 pp)
Managers support
Health professionals
This applies to workforce
15
trainers like John or
Health aides, Customer service Production computer engineers (STEM)
10 technicians, and sales work
like Sara
and wellness Transportation Property maintenance
5 Agriculture services Food services Builders High potential (≥20 pp) but
Mechanical medium need for business,
Medium need – limited potential High need – limited potential
installation and repair creatives, or office support,
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2.0 2.1 2.2 and high needs but limited
potential (≤10 pp) for builders
Share of job vacancies per share of employment in an occupation group2
Note: Figures may not sum because of rounding
1. Previous assessment of work automation before the rise of GenAI. | 2. Share of job vacancies divided by the share of employment within an occupation
group indicating the concentration of open job positions per actual employment
Source: McKinsey Global Institute analysis McKinsey & Company 15
Wages: GenAI is expected to have the biggest labor shortage
mitigating impact in areas with high wages
Largest increase in automation adoption from GenAI
Largest automation adoption without GenAI1
Additional GenAI automation adoption vs. without GenAI per wage group,22030, percentage points
Key implications
USA France Austria Germany for Germany
Developed 15 12 12 In Germany, the
14
13
economies 12 highest wage group
11
10 (quintile) will
9
8 experience the
6 6 greatest additional
5 5
4 automation potential
3 3
(12 pp) and the
corresponding labor
shortage mitigation
6
China India Mexico
potential from GenAI
Largest automation
Emerging
4
adoption without
economies
3 GenAI is highest for
8
7 7 7 7 7 4th wage group
(quintile), with 37%
4
3 3 post-GenAI versus
2 2 2 2 2
1 33% pre-GenAI
Lower Lower Mid Upper Upper Lower Lower Mid Upper Upper Lower Lower Mid Upper Upper Lower Lower Mid Upper Upper
mid mid mid mid mid mid mid mid
1. Previous assessment of work automation before the rise of GenAI
2. Difference between automation potential without GenAIand additional automation potential with GenAI
Source: McKinsey Global Institute analysis McKinsey & Company 16
Upskilling and attracting the right tech talent is the core task of
public and private organizations in mitigating labor shortages
Building on existing capabilities
and competencies Recruiting of new tech talent
1 Upskilling/ • Developing requirements for the • Analyzing the skills and competence
reskilling for AI building and leadership development of profiles of current employees and
roles GenAI core competencies existing open positions that cannot be
• Determining the cohorts with upskilling recruited from the labor market
needs • Establishing employer branding and
• Establishing a boot camp approach to targeted recruiting to attract best-in-class
GenAI training talent
2 Training and • Implementing improvements for the • Identifying the remaining necessary
coaching training program based on early findings qualifications
3 Establishing a • Involving senior management to ensure • Creating a short-term hiring target for
'learning culture' support is provided recruiting the required role profiles
• Defining the behavioral and mindset • Developing a mid-term road-map for
changes required for a learning culture strategic recruiting
• Designing of competency building
initiatives (e.g., on feedback, coaching)
McKinsey & Company 17
The skilled labor shortage in
Germany
Agenda
The potential of GenAI to
increase productivity
The GenAI landscape in Germany
McKinsey & Company 18
Public and private actors must work toward a scenario with both
the right operating environment and the availability of skills
Ensuring availability Key implications
Providing the right environment of skills for Germany
The two key enablers
Procurement
Private German for making use of
sector Government GenAI in Germany are:
The right operating
environment and the
availability of relevant
skills in Germany
R&D Creates and As they both have a
collaboration enforces reinforcing effect
towards the respective
Skills
Data for
counterpart they must
GenAI1
be pursued
simultaneously.
Public policy
Academia
(Germany/EU)
Supports
Quantitative deep dive on every stakeholder (group) on the following pages
1. We apply a GenAIfocus to this framework because GenAIbuilds on the workforce, skillsets, and capabilities, which grew the AI market
Source: Oxford Publication McKinsey & Company 19
Germany is an entrepreneurial but underfunded country with
great potential to becoming a European leader
Private sector
European perspective Leader Emerging
Total funding
Estonia Switzerland UK
of AI companies
proportional
to GDP
Finland
France
Portugal
Iceland Sweden
Belgium Germany
Norway
Austria
Ireland
Denmark
Hungary
The Netherlands
Luxembourg
Lithuania Spain
Malta Greece
Czech Republic Poland
Slovenia
Slovak Republic Russia Italy
Absolute number of AI start-ups
Source: The Global AI, Index, Tortoise AI Report, Tortoise Media McKinsey & Company 20
Germany’s expenditure on R&D is increasing, whereas its
contribution to AI projects has stagnated for years
Private sector Academia
Gross domestic expenditure on R&D, Number of AI projects by global comparison,
% of GDP % total AI projects1
Germany EU-27 Germany Switzerland Austria China EU-27 UK India US
3.5 35
30
3.0
25
20
2.5
15 This is a relative representation;
it doesn't indicate a declining
number in AI projects
10
2.0
5
1.5 0
2011 12 13 14 15 16 17 18 19 20 2021 2011 12 13 14 15 16 17 18 19 20 21 2022
Number of Al projects (i.e., Al-related GitHub "repositories") as a
fractional count based on the share of contributions (i.e.,
"commits") by country and over time
1. fractional count based on contributions
Source: Eurostat.-November 2022; GitHub; Preqin; oecd.ai McKinsey & Company 21
Germany is a leader in technology, but only keeping up
with equally large countries in investments
Private sector Academia
Global perspective – relating to both AI and GenAI in 2022
Absolute
Funding positioned Leader US
private and
public
China
investments UK
in AI
Japan
companies India France
Germany
Spain
Canada Italy
Sweden
South Korea
Australia
The Netherlands
Singapore
Finland
Switzerland
Austria
Mexico Brazil
Aspirational Technology skilled
Uganda Russia
Technology and research1
1. Technology and researchcontains country ranks by theoretical peak computer performance, number of processing cores, number of supercomputers, and maximal LINPACK performance achieved; the country ranks for the number of
conference papers and journal papers; and the country rank for the number of patents
Source: Brookings, 2022 McKinsey & Company 22
Germany has a high proportion of self-
reported AI capabilities compared to the
OECD average, trailing only the US
Skills
Key implications
AI skill penetration of workforce
for Germany
Prevalence of workers with AI skills as self-reported by LinkedIn members from 2015-2022 by country1
United States 2.2 Germany has the 2nd
Germany 1.7
Israel 1.7
second highest AI skill
Canada 1.6
United Kingdom 1.5
penetration (1.5) in its
Korea 1.4
Japan 1.2
workforce, which is only
Turkey 1.2
France 1.1
surpassed by the US with a
OECD average 1.0
Spain 1.0
penetration factor of 2.2
Netherlands 1.0
Italy 0.9
Switzerland 0.9 German workers are 1.7x
Greece 0.9
Australia 0.9 likely to report AI skills
Poland 0.8
Ireland 0.7 than workers in the OECD
Sweden 0.7
Norway 0.7 benchmark
Finland 0.6
Belgium 0.6
Hungary 0.6 Germany is thus in the
Lithuania 0.6
Austria 0.6 group of leading AI
Denmark 0.5
Estonia 0.5 nations, like the US, Israel,
Mexico 0.5
New Zealand 0.4 Canada, and the UK
Slovenia 0.4
Czech Republic 0.4
Portugal 0.4
Luxembourg 0.4
Chile 0.4
Slovak Republic 0.4
1. A Country’s AI skills penetration of 1.5 means that workers in that country are 1.5X more likely to report AI skills than workers in the benchmark
Source: Data from LinkedIn 2015-2022 accessed on Sep 20, 2023; self-reported; OECD.AI (2023) McKinsey & Company 23 |
250 | mckinsey | adopting-ai-at-speed-and-scale-the-4ir-push-to-stay-competitive-v2.pdf | Operations Practice
Adopting AI at speed
and scale: The 4IR push
to stay competitive
AI has brought the Fourth Industrial Revolution to an inflection point,
and manufacturers must choose a path forward: innovate, accelerate,
or follow fast.
by Henry Bristol, Enno de Boer, Dinu de Kroon, Forest Hou, Rahul Shahani, and Federico Torti
February 2024
The world has changed for manufacturers. capabilities they have built to deploy it with both
Preparation for uncertainty has become an industry speed and scale. In this first installment, we’ll
norm, with executives expecting the impact of explore how the maturity of AI marks a 4IR inflection
disruption—whether from geopolitical tensions, point; examine how leading manufacturers are
climate change effects, technology breakthroughs, redefining the leading edge of manufacturing with
or supply chain vulnerabilities—to increase by 15 to this technology; and finally, consider three types
25 percent over the next five years.1 of strategic responses—to innovate, to accelerate,
or to follow fast—that manufacturers will need to
At the same time, advanced manufacturing is now consider as industry becomes more competitive.
flourishing in markets where stagnation had seemed The second and third articles, respectively, will
intractable. Growth in the US manufacturing sector, focus on the at-scale impact of AI within the
for example, had languished at 1.4 percent over manufacturing sector and the essential capabilities
the past two decades. More recently, AI, digital that drive AI adoption.
technologies, sustainable features, and higher skill
have reinvigorated the market: over the past five
years, US industrials companies have generated The S-curves of industrial revolution
total shareholder returns about 400 basis points Global industry transformation has never been
higher than in the previous 15 years. instantaneous. Each “revolutionary shift” saw a lag
period between the introduction of the enabling
The accelerating pace of the Fourth Industrial foundation and widespread adoption. Consider the
Revolution (4IR) can help enable this type of steam engine. The Roman architect Vitruvius made
next-level performance while also increasing mention of a rudimentary steam-powered device
workforce inclusivity and sustainability. The Global as early as 15 BC. Why, therefore, did widespread
Lighthouse Network,2 now in its fifth year, provides adoption take more than 1,800 years? The answer
an expanding pool of examples. In effect, each is simple: steam was neither practical nor cost-
Lighthouse cohort provides a three- to five-year effective until breakthrough engine technologies—
look ahead at the future of operations across the along with the infrastructure of coal supply
value chain. chains—made it so. This tipping point essentially
eliminated the learning curve, allowing the “doing
The most recent cohort affirms a 4IR inflection curve” to steepen. The front-runners had done the
point, marked by two factors. First, machine learning. It wasn’t until the late 18th century that, in
intelligence technologies—AI that, rather than the space of just 20 years, steam engine adoption in
seeking to simulate human intelligence, empowers industry increased from practically nothing to nearly
machines with the specialized intelligence needed 80 percent.
to perform complex tasks in the cyber-physical
world of production—are reaching unprecedented What steam was to the First Industrial Revolution
levels of maturity. Second, leading companies is what AI will be to the fourth. And much as coal
are redefining the concept of a pilot as they scale supply chains and factory infrastructure were the
impact by using entire factories, rather than tipping point that enabled steam power to race
individual use cases, as pilots. up the adoption curve, data collection and data
infrastructure are doing the same in the fourth.
This series of three articles will explore: (1) the Already, some of the world’s leading factories
current status of global manufacturing, with a generate multiple petabytes of data a week. If all
particular focus on (2) what AI looks like among ten million factories in the world operated at this
today’s leading manufacturers, and (3) the level, they would double all human information in
1 “The great acceleration: CIO perspectives on generative AI,” MIT Technology Review, July 18, 2023.
2 The Global Lighthouse Network is a World Economic Forum initiative cofounded with McKinsey and counseled by an advisory board of industry
leaders, including Contemporary Amperex Technology, Foxconn Industrial Internet, Henkel, Johnson & Johnson, Koc Holdings, Siemens, and
Schneider Electric. Lighthouses in the network are designated by an independent panel of experts.
Adopting AI at speed and scale: The 4IR push to stay competitive 2
less than a month (see sidebar “The evolution of Revolution—is that we believe them to be three
the revolutions”). to five years further along 4IR’s adoption curve
than are other manufacturers. Today, they aren’t
focused on piloting use cases. Instead, they’ve built
The adoption S-curve the capabilities to get new use cases right quickly
We can see a pattern as we look back upon each and without trials. For companies with multiple
industrial revolution: it has always taken the shape Lighthouses, entire factories serve as pilots for
of an “S-curve.” The first phase is a learning curve, networkwide deployment at scale. Leaders are now
which tends to be long, and is marked by trial and capturing the value of 4IR technologies ten and
error as early front-runners figure out how to make 50 factories at a time, where others are still working
things work. It then moves on to the next phase, to find value within a single factory.
the “doing” part of the curve. This is when the
foundational technology has been established, Lighthouses are accelerating, and so too is the
and organizations work to deploy it across their maturity chasm they have put between themselves
production networks. Finally, we see an optimization and everyone else. This chasm is evident in the
curve, which is when industries align around what wake of recent disruption and volatility. Consider
works best. New standards and protocols become that 85 percent of Lighthouses saw revenue
ingrained, and costs start to stabilize (Exhibit 1). reductions of less than 10 percent during the
One need only look back at a now-ubiquitous height of the COVID-19 pandemic; this was true
technology, such as the smartphone, to recognize for only 14 percent of other manufacturers.
this three-phase S-curve in play. Lighthouses could react quicker: although they
faced the same supply chain risks, 65 percent
One of the most significant differentiators for of Lighthouses were already dual-sourcing and
members of the Global Lighthouse Network—the increasing inventory by 2022, compared with only
153 factories at the forefront of the Fourth Industrial 24 percent of other companies.
Web <2024>
E<Axdhoipbtiintg 1 AI at speed and scale: the 4IR push to stay competitive >
Exhibit <1> of <4>
From learning to doing: Lighthouses are rapidly climbing the Fourth
Industrial Revolution adoption curve.
Fourth Industrial Revolution (4IR) adoption curve, illustrative
Learning Doing Optimizing
Factory pilots, characterized by Network adoption, where proven Industry new normal, when costs
trial-and-error refinements technologies rapidly scale optimize and standards coalesce
around best-in-class solutions
HIGH
Adoption Scaling
and impact slump Lighthouses, 2024
Pilot
purgatory
Lighthouses, 2018
LOW
Time and investment
Source: Global Lighthouse Network: Adopting AI at speed and scale, World Economic Forum, December 2023
McKinsey & Company
Adopting AI at speed and scale: The 4IR push to stay competitive 3
AI is defining the Fourth “pyramid” of 4IR technologies—it is playing the role
Industrial Revolution of conductor for 4IR technologies, which together
perform a symphony of impact (Exhibit 2).
The true power of AI for the Fourth Industrial
Revolution lies with the fact that it sits at the top of a
Web <2024>
E<Axdhoipbtiintg 2 AI at speed and scale: the 4IR push to stay competitive >
Exhibit <2> of <4>
The full value of Fourth Industrial Revolution technologies comes from a
suite of technology solutions.
Fourth Industrial Revolution (4IR) technology pyramid
Machine intelligence to optimize, augment, or automate
decision making, such as heuristic models, applied AI,
and generative AI
Machine
Digital-worker productivity tools at the operator or
intelligence
process level (eg, augmented/virtual reality, wearables,
and exoskeletons)
Path to Path to Automation and disruption of processes, such as
Worker Production
implementation exponential co-bots and flexible robots, automated guided vehicles
connectivity robotics and
impact and drones, and 3-D printing
and digitization automation
System-level digitization of planning and management
Digital planning and processes, such as the manufacturing execution
management tools system, product life cycle management, customer
relationship management, and other enterprise tools
Connectivity and Underlying data, connectivity, and computing tools
infrastructure tools (eg, cloud and edge hosts, 5G communication, and
data lakes)
McKinsey & Company
The evolution of the revolutions
Late 18th century — Unlock: mass production — Enabling infrastructure:
semiconductors and transistors
— Revolutionary shift: steam power and — Enabling infrastructure: electrical grid
mass production techniques Today
Mid-20th century
— Unlock: mechanization — Revolutionary shift: machine
— Revolutionary shift: programmable
intelligence to make trade-off
— Enabling infrastructure: coal logic and control loops for automation
decisions enabling augmentation
supply chain of robotics and reduction in manual
and optimization
tasks
Late 19th century
— Unlock: advanced manufacturing
— Unlock: standardization and
— Revolutionary shift: electrification,
automation — Enabling infrastructure: big data
assembly line production, new
resources, and synthetic materials
Adopting AI at speed and scale: The 4IR push to stay competitive 4
AI is defining the Fourth Industrial
Revolution, and Lighthouses are
showing us that AI has myriad new use
cases and possibilities for unimaginable
performance improvements.
Consider the example of a rapid changeover at a transparent data connectivity and visualization
production site (Exhibit 3). This requires flexible dashboards, and similar digital-lean solutions.
robotics to handle different products, automated These use cases took much longer for Lighthouses
guided vehicles to move materials and parts, 3-D to implement than they do today; where most
printing to customize line fixtures, and wearable Lighthouses say it took ten to 20 months average
technology to keep managers and technicians time to implement their first five use cases,
informed with real-time data. What orchestrates 75 percent say they can now do it in less than six
this complex interplay of elements, each of which is months. Even more impressively, 30 percent claim
individually complex? The answer: AI. they can do it in less than three months.
But AI needs terabytes of data generated by This is because for early use cases, factories first
and collected from a broad range of sources: had to rewire their data collection and connectivity
enterprise systems, machine sensors, connectivity layers, design tech stacks that added to or
infrastructure, and human workers. That’s why the upgraded legacy infrastructure, train their people
most advanced front-runners are ahead. They had on how to use advanced new tools, and reorganize
the foresight to make investments and take on risks themselves to deploy digital solutions quickly
involved in building the data foundations that are and with strong feedback. Once built, these
needed to power AI technologies and unlock their capabilities became the foundation for the rapid
potential impact (see sidebar “Understanding AI: deployment of new use cases. One Lighthouse, for
How it actually works). example, says it was able to implement a gen-AI-
based technician adviser in just days and weeks,
With AI, machine intelligence can orchestrate highly not months and years.
complex technologies for rapid solutions.
Ongoing challenges see Lighthouses move from
pilot purgatory to scaling slump. At the factory
Capabilities to tackle the scaling slump level, capabilities are a solved problem, at least
AI is defining the Fourth Industrial Revolution, and for Lighthouses. Now, many Lighthouses are
Lighthouses are showing us that AI has myriad in the valley that follows the “false peak” of the
new use cases and possibilities for unimaginable learning curve, stalling on impact while they rewire
performance improvements. themselves for network scale. This isn’t easy. Taking
a technology that works in one place and extending
it across an entire production network introduces
Use cases inform capabilities, and capabilities
massive new challenges: data, technology, talent,
lead scale. Back in 2018, cutting-edge use cases
and organizational challenges that exist at a single
looked like localized applications of advanced
site are not the same at the macro level—and neither
analytics and autonomous vehicles, or radically
Adopting AI at speed and scale: The 4IR push to stay competitive 5
Web <2024>
E<Axdhoipbtiintg 3 AI at speed and scale: the 4IR push to stay competitive >
Exhibit <3B> of <4>
With AI, machine intelligence can orchestrate highly complex technologies
for rapid solutions.
Implementation of a “one click” changeover
1
3
5
2
4
1 2 3 4 5
Machine intelligence Wearables and devices Flexible robotics Integrated systems, Cloud and edge hosts
“conducts” an give critical alerts to are designed to be including the connect all systems
orchestra of Fourth technicians based on easily reprogrammed manufacturing wirelessly and execute
Industrial Revolution real-time data so they can handle execution system and critical computations
technologies, multiple, diverse distributed control
maintaining system, products system, inform and
timing, sequencing, control all machines
and responsiveness needed to make the
next product
McKinsey & Company
Adopting AI at speed and scale: The 4IR push to stay competitive 6
are the solutions. For that gen-AI-based technician to integrate ChatGPT into Bing mere months after
adviser to see the light of two dozen factories, those its launch.
factories must first be ready to receive it.
And like both the steam engine in the First
Those that overcome the scaling slump can Industrial Revolution and AI technologies in tech
define entire industries. This is because they and banking, we expect 4IR’s breakthrough
set standards. The adoption of steam engines in technologies to catapult from single digit
manufacturing is an early example. For a more to widespread adoption within the decade.
modern-day example: consider Toyota, which Lighthouses are leading the way. Already, AI-based
managed to scale its production system at the use cases make up over 60 percent of the use
macro level. Shortly after, lean manufacturing and cases presented by new Lighthouse applicants, up
Six Sigma became standard fare for companies from just 11 percent in 2019.
all across the world—with the accompanying
emergence of new standards, protocols,
certifications, and regulatory measures. The Lighthouses are accelerating the
innovations became institutionalized; Toyota leading edge: How will you respond?
defined the new normal, which is reflective of Although it may be some time before gen AI and
the optimization phase of the adoption curve. In other highly advanced emerging technologies
other sectors outside of manufacturing, such as see networkwide adoption in manufacturing,
tech and banking, AI is already at this stage, so Lighthouses are already achieving factory-scale
the conversations are focused on standards and adoption. All of the newly recognized Lighthouses
regulatory compliance. These sectors can deploy have at least one gen AI pilot in process, and
AI at scale, fast—for example, Microsoft was able several have implemented, tested, and iterated
Understanding AI: How it actually works
Before exploring in detail how A subset field in AI, machine learning, large language models (with hundreds
Lighthouses have been rapidly adopting began developing traction by the 1980s. of billions of neurons) that can learn
analytic AI and piloting its emergent It focused on teaching machines to especially abstract patterns (exhibit).
offshoot, generative AI, it’s first crucial learn relationships hidden in data and
Each of these breakthroughs has
to gather a basic understanding of what to build approximate models of real
followed its own accelerating adoption
underpins AI. Everyone is talking about it, systems. Within two decades, a branch
curve. Today, both machine learning and
which raises a worthy question: Do of machine learning called deep learning
deep learning techniques, excluding
you actually know the fundamentals of emerged, as “neural networks” became
generative adversarial networks (gen
how AI works? popular methods to model real systems
AI) and encompassing such methods as
by mimicking how the human brain
Pioneered in the 1950s, AI now refers to gradient and adaptive boosting, random
works, with millions of computational
the broad field of developing machines, forests, convolutional and recurrent
“neurons.” 2017 saw the popularization
applications, and tools that approximate neural networks, decision trees, support
of transformers and the advent of
human behavior, including all aspects vector machine algorithms, and more
generative adversarial networks, a type
of perceiving, reasoning, learning, and are collectively referred to as analytic
of deep learning known as generative
problem solving. The first instances AI. This family of technologies has seen
AI (gen AI), which has enabled use of
included statistical analyses and rapid maturity and pace of adoption
exceptionally large neural nets called
predictions enabled by early computers. by Lighthouses.
Adopting AI at speed and scale: The 4IR push to stay competitive 7
Understanding AI: How it actually works (continued)
The next evolution: Gen AI unstructured data sets, like a human brain. the database, often referred to as the large
Gen AI is projected to add between It leverages a “transformer” architecture language model or foundation model.3
$2.6 trillion and $4.4 trillion in annual value to generate “embeddings”—an approach
To pick or generate sequences of tokens,
to the global economy1—nearly a quarter initially designed for natural language
one deep learning model predicts
of which could be captured by productivity processing tasks. Embeddings are
subsequent tokens, while another
improvements of up to two times and task massive vectors representing hundreds
analyzes and scores the selection—which
automations of nearly 70 percent across of thousands of parameters for any given
is exactly why gen AI is often referred
activities related to manufacturing and “token,” or piece of information. (For text-
to as a generative adversarial network.
supply chains.2 That’s why it makes sense based models, a token might be as small as
This unique approach is what enables
that, as of mid-2023, nearly one-third the prefix “un.”) It can predict or “generate”
gen AI to begin to process troves of
of all companies surveyed said that they content by identifying a probability that
unstructured data to emulate true human
have implemented gen AI in at least one any one token will sequentially follow
reasoning and connection, synthesize
business function. another token. This probability calculation
insights, generate content, and generally
accounts for the proximity of that token’s
Gen AI’s differentiating factor is that it can “humanize” user interactions.
vector embedding with others stored in
pay attention to patterns across immense
Web <2024>
E<Axdhoipbtiintg AI at speed and scale: the 4IR push to stay competitive >
Sidebar <1> of <1>
Generative AI is the next new frontier of a long AI journey.
Artificial intelligence timeline
1950s 1960s 1970s 1980s 1990s 2000s 2010s 2020s
Artificial intelligence Machine learning Deep learning Generative AI
The broad field of Major approach to achieve Branch of machine Branch of deep learning that uses
developing machines AI by teaching machines learning that uses “neural exceptionally large neural networks
that can replicate human to learn relationships networks” to model real called large language models (LLMs)
behavior, encompassing hidden in data and build systems by mimicking (with hundreds of billions of neurons)
perceiving, reasoning, approximate models of how the human brain that can learn especially abstract pat-
learning, and problem real systems works, utilizing millions terns; applying these LLMs to interpret
solving of computational “neurons” and create text, images, video, and data
has become known as generative AI
McKinsey & Company
1 “The economic potential of generative AI: The next productivity frontier,” McKinsey, June 14, 2023.
2 “The great acceleration: CIO perspectives on generative AI,” MIT Technology Review, July 18, 2023.
3 “Embeddings: The Language of LLMs and GenAI,” Medium, October 4, 2023.
gen AI use cases for impact in as little as days or They’ve coupled these enablers with clear business
weeks—not months or years. Again, they’re able to strategy and effective change management.
do so because they’ve already built the necessary Notably, Lighthouses avoid the trap of investing
enablers: solid data and tech infrastructure, a strong in technology for its own sake, instead ensuring
talent base, and agile operating models. that every use case presents clear business value.
Adopting AI at speed and scale: The 4IR push to stay competitive 8
Paradoxically, they’ve surged ahead by taking a This moment compels a decision:
patient, measured approach—typically between Lead, innovate, or follow fast
ten and 20 months for use case implementation,
What does an inflection point mean for
with an ROI period of approximately two and a half
manufacturers? It compels a crucial decision
years. This takes patience and a steady hand, but
about how to respond. The global front-runners—
the returns are worth it. Lighthouses’ 4IR use cases
Lighthouses—along with an increasing number of
have averaged between two and three times ROI
other leading organizations are pushing through the
within three years and between four and five times
scaling slump, achieving network-level impact. What
ROI within five years.
will other companies do?
As Lighthouses continue to push through the
There is more than one intelligent response. An
scaling slump, their ability to implement use cases
organization might choose to be a network-level
more rapidly is improving. The most recent three
digital innovator. This is an industry pathfinder
Lighthouse cohorts implemented use cases
that takes a risk on—and proves, at the factory
26 percent faster than the first three did, and
level—the next big thing. This is the path that
75 percent of Lighthouses report that they’re able
many Lighthouses have taken. But there’s also a
to deploy a new advanced use case in under six
smart path forward for the industry accelerator.
months; 30 percent can do so in less than three
This company focuses on network-level impact,
months. Technology adoption is self-perpetuating.
changing the landscape for an entire industry.
The more companies progress, the faster they
Finally, there’s great opportunity that lies with
progress. This also provides them with added
being a fast follower. This company embraces
agility and ability to respond to disruption—a major
the playbook already written by the innovators
factor in the expanding performance gap between
and accelerators, capturing value while skipping
leaders and laggards.3
the costs and tribulations of the learning curve
altogether (Exhibit 4).
Web <2024>
E<Axdhoipbtiintg 4 AI at speed and scale: the 4IR push to stay competitive >
Exhibit <4> of <4>
As Fourth Industrial Revolution technologies accelerate and Lighthouses
tackle the scaling slump, others will need to take a strategic response.
Response archetypes
Innovator securing Accelerator achieving Fast follower rapidly Laggard falling behind on
competitive advantage by competitive advantage deploying off-the-shelf digitization and potentially
piloting new technologies with speed and scale of solutions once proven to losing relevance as a
and proving their impact technology adoption, with be impactful, scalable, and competitive manufacturer
production network–level cost effective
impact
McKinsey & Company
3 Global Lighthouse Network 2023 Research Survey, August 2023.
Adopting AI at speed and scale: The 4IR push to stay competitive 9
Each of these three responses can comprise an The next two articles in this series will explore what
Find more content like this on the
intelligent strategic response. This exciting juncture advanced AI and gen AI look like among today’s
McKinsey Insights App
of the Fourth Industrial Revolution provides a leading manufacturers, and the capabilities that
momentous opportunity for manufacturers to are powering Lighthouses to scale advanced
choose a course of action. There’s freedom in that, technologies across full production networks.
and it means that companies can take the approach
that best suits their circumstances and business This article originally appeared in the Global
needs. But make no mistake: this inflection point Lighthouse Network whitepaper Adopting AI at
also means that inaction is a sure-fire path to failure. speed and scale, published on December 14, 2023.
Scan • Download • Personalize
Henry Bristol is a consultant in McKinsey’s Dallas office, Enno de Boer is a senior partner in the New Jersey office, Dinu de
Kroon is a partner in the Zurich office, Forest Hou is a partner in the Shanghai office, and Rahul Shahani is a partner in the
New York office. Federico Torti is the advanced manufacturing and supply chains initiatives lead at the World Economic Forum.
Designed by McKinsey Global Publishing
Copyright © 2024 McKinsey & Company. All rights reserved.
Adopting AI at speed and scale: The 4IR push to stay competitive 10 |
251 | mckinsey | technologys-generational-moment-with-generative-ai-a-cio-and-cto-guide.pdf | Technology’s generational
moment with generative AI:
A CIO and CTO guide
CIOs and CTOs can take nine actions to reimagine business and technology
with generative AI.
This article is a collaborative effort by Aamer Baig, Sven Blumberg, Eva Li, Douglas Merrill, Adi Pradhan,
Megha Sinha, Alexander Sukharevsky, and Stephen Xu, representing views from McKinsey Digital.
© Getty Images
July 2023
Hardly a day goes by without some new CTO, the generative AI boom presents a unique
business-busting development related to opportunity to apply those lessons to guide the
generative AI surfacing in the media. The C-suite in turning the promise of generative AI
excitement is well deserved—McKinsey research into sustainable value for the business.
estimates that generative AI could add the
equivalent of $2.6 trillion to $4.4 trillion of value Through conversations with dozens of tech
annually.¹ leaders and an analysis of generative AI initiatives
at more than 50 companies (including our own),
CIOs and chief technology officers (CTOs) have a we have identified nine actions all technology
critical role in capturing that value, but it’s worth leaders can take to create value, orchestrate
remembering we’ve seen this movie before. New technology and data, scale solutions, and
technologies emerged—the internet, mobile, manage risk for generative AI (see sidebar, “A
social media—that set off a melee of experiments quick primer on key terms”):
and pilots, though significant business value
often proved harder to come by. Many of the 1. Move quickly to determine the company’s
lessons learned from those developments still posture for the adoption of generative AI,
apply, especially when it comes to getting past and develop practical communications to, and
the pilot stage to reach scale. For the CIO and appropriate access for, employees.
A quick primer on key terms
Generative AI is a type of AI that can create new content (text, code, images, video) using patterns it has learned by training on exten-
sive (public) data with machine learning (ML) techniques.
Foundation models (FMs) are deep learning models trained on vast quantities of unstructured, unlabeled data that can be used for
a wide range of tasks out of the box or adapted to specific tasks through fine-tuning. Examples of these models are GPT-4, PaLM 2,
DALL·E 2, and Stable Diffusion.
Large language models (LLMs) make up a class of foundation models that can process massive amounts of unstructured text and
learn the relationships between words or portions of words, known as tokens. This enables LLMs to generate natural-language text,
performing tasks such as summarization or knowledge extraction. Cohere Command is one type of LLM; LaMDA is the LLM behind
Bard.
Fine-tuning is the process of adapting a pretrained foundation model to perform better in a specific task. This entails a relatively short
period of training on a labeled data set, which is much smaller than the data set the model was initially trained on. This additional training
allows the model to learn and adapt to the nuances, terminology, and specific patterns found in the smaller data set.
Prompt engineering refers to the process of designing, refining, and optimizing input prompts to guide a generative AI model toward
producing desired (that is, accurate) outputs.
Learn more about generative AI in our explainer “What is generative AI” on McKinsey.com.
1 “The economic potential of generative AI: The next productivity frontier,” McKinsey, June 14, 2023.
2 Technology’s generational moment with generative AI: A CIO and CTO guide
2. Reimagine the business and identify use 1. Determine the company’s posture for
cases that build value through improved the adoption of generative AI
productivity, growth, and new business As use of generative AI becomes increasingly
models. Develop a “financial AI” (FinAI) widespread, we have seen CIOs and CTOs respond
capability that can estimate the true costs and by blocking employee access to publicly available
returns of generative AI. applications to limit risk. In doing so, these
companies risk missing out on opportunities for
3. Reimagine the technology function, and focus innovation, with some employees even perceiving
on quickly building generative AI capabilities in these moves as limiting their ability to build
software development, accelerating technical important new skills.
debt reduction, and dramatically reducing
manual effort in IT operations. Instead, CIOs and CTOs should work with risk
leaders to balance the real need for risk mitigation
4. Take advantage of existing services or adapt with the importance of building generative AI
open-source generative AI models to develop skills in the business. This requires establishing
proprietary capabilities (building and operating the company’s posture regarding generative AI
your own generative AI models can cost tens by building consensus around the levels of risk
to hundreds of millions of dollars, at least in the with which the business is comfortable and how
near term). generative AI fits into the business’s overall strategy.
This step allows the business to quickly determine
5. Upgrade your enterprise technology company-wide policies and guidelines.
architecture to integrate and manage
generative AI models and orchestrate how Once policies are clearly defined, leaders should
they operate with each other and existing AI and communicate them to the business, with the CIO
machine learning (ML) models, applications, and and CTO providing the organization with appropriate
data sources. access and user-friendly guidelines. Some
companies have rolled out firmwide communications
6. Develop a data architecture to enable access about generative AI, provided broad access to
to quality data by processing both structured generative AI for specific user groups, created pop-
and unstructured data sources. ups that warn users any time they input internal
data into a model, and built a guidelines page that
7. Create a centralized, cross-functional appears each time users access a publicly available
generative AI platform team to provide generative AI service.
approved models to product and application
teams on demand.
2. Identify use cases that build value
8. Invest in upskilling key roles—software through improved productivity,
developers, data engineers, MLOps engineers, growth, and new business models
and security experts—as well as the broader
CIOs and CTOs should be the antidote to the “death
nontech workforce. But you need to tailor the
by use case” frenzy that we already see in many
training programs by roles and proficiency
companies. They can be most helpful by working
levels due to the varying impact of generative AI.
with the CEO, CFO, and other business leaders
to think through how generative AI challenges
9. Evaluate the new risk landscape and
existing business models, opens doors to new ones,
establish ongoing mitigation practices to
and creates new sources of value. With a deep
address models, data, and policies.
understanding of the technical possibilities, the
Technology’s generational moment with generative AI: A CIO and CTO guide 3
CIO and CTO should identify the most valuable — Software development: McKinsey research
opportunities and issues across the company that shows generative AI coding support can help
can benefit from generative AI—and those that software engineers develop code 35 to 45
can’t. In some cases, generative AI is not the best percent faster, refactor code 20 to 30 percent
option. faster, and perform code documentation 45
to 50 percent faster.³ Generative AI can also
McKinsey research, for example, shows generative automate the testing process and simulate
AI can lift productivity for certain marketing use edge cases, allowing teams to develop
cases (for example, by analyzing unstructured more-resilient software prior to release, and
and abstract data for customer preference) by accelerate the onboarding of new developers
roughly 10 percent and customer support (for (for example, by asking generative AI questions
example, through intelligent bots) by up to 40 about a code base). Capturing these benefits
percent.² The CIO and CTO can be particularly will require extensive training (see more in
helpful in developing a perspective on how best action 8) and automation of integration and
to cluster use cases either by domain (such as deployment pipelines through DevSecOps
customer journey or business process) or use case practices to manage the surge in code volume.
type (such as creative content creation or virtual
agents) so that generative AI will have the most — Technical debt: Technical debt can account for
value. Identifying opportunities won’t be the most 20 to 40 percent of technology budgets and
strategic task—there are many generative AI use significantly slow the pace of development.⁴
cases out there—but, given initial limitations of CIOs and CTOs should review their tech-debt
talent and capabilities, the CIO and CTO will need balance sheets to determine how generative
to provide feasibility and resource estimates to AI capabilities such as code refactoring,
help the business sequence generative AI priorities. code translation, and automated test-case
generation can accelerate the reduction of
Providing this level of counsel requires tech leaders technical debt.
to work with the business to develop a FinAI
capability to estimate the true costs and returns on — IT operations (ITOps): CIOs and CTOs will
generative AI initiatives. Cost calculations can be need to review their ITOps productivity efforts
particularly complex because the unit economics to determine how generative AI can accelerate
must account for multiple model and vendor costs, processes. Generative AI’s capabilities are
model interactions (where a query might require particularly helpful in automating such tasks
input from multiple models, each with its own fee), as password resets, status requests, or
ongoing usage fees, and human oversight costs. basic diagnostics through self-serve agents;
accelerating triage and resolution through
improved routing; surfacing useful context,
3. Reimagine the technology function such as topic or priority, and generating
Generative AI has the potential to completely suggested responses; improving observability
remake how the tech function works. CIOs and through analysis of vast streams of logs to
CTOs need to make a comprehensive review of identify events that truly require attention; and
the potential impact of generative AI on all areas developing documentation, such as standard
of tech, but it’s important to take action quickly to operating procedures, incident postmortems,
build experience and expertise. There are three or performance reports.
areas where they can focus their initial energies:
2 Ibid.
3 Begum Karaci Deniz, Martin Harrysson, Alharith Hussin, and Shivam Srivastava, “Unleashing developer productivity with generative AI,”
McKinsey, June 27, 2023.
4 Vishal Dalal, Krish Krishnakanthan, Björn Münstermann, and Rob Patenge, “Tech debt: Reclaiming tech equity,” McKinsey, October 6, 2020.
4 Technology’s generational moment with generative AI: A CIO and CTO guide
4. Take advantage of existing services infrastructure, reducing the need for data
or adapt open-source generative AI transfers. The other approach is to “bring
models data to the model,” where an organization can
A variation of the classic “rent, buy, or build” decision aggregate its data and deploy a copy of the large
exists when it comes to strategies for developing model on cloud infrastructure. Both approaches
generative AI capabilities. The basic rule holds true: achieve the goal of providing access to the
a company should invest in a generative AI capability foundation models, and choosing between them
where it can create a proprietary advantage for the will come down to the organization’s workload
business and access existing services for those that footprint.
are more like commodities.
— Maker—builds a foundation model to address
The CIO and CTO can think through the implications a discrete business case. Building a foundation
of these options as three archetypes: model is expensive and complex, requiring
huge volumes of data, deep expertise, and
— Taker—uses publicly available models through massive compute power. This option requires
a chat interface or an API, with little or no a substantial one-off investment—tens or
customization. Good examples include off- even hundreds of millions of dollars—to build
the-shelf solutions to generate code (such the model and train it. The cost depends on
as GitHub Copilot) or to assist designers with various factors, such as training infrastructure,
image generation and editing (such as Adobe model architecture choice, number of model
Firefly). This is the simplest archetype in terms of parameters, data size, and expert resources.
both engineering and infrastructure needs and
is generally the fastest to get up and running. Each archetype has its own costs that tech
These models are essentially commodities that leaders will need to consider (Exhibit 1). While new
rely on feeding data in the form of prompts to the developments, such as efficient model training
public model. approaches and lower graphics processing unit
(GPU) compute costs over time, are driving costs
— Shaper—integrates models with internal data down, the inherent complexity of the Maker
and systems to generate more customized archetype means that few organizations will adopt
results. One example is a model that supports it in the short term. Instead, most will turn to some
sales deals by connecting generative AI combination of Taker, to quickly access a commodity
tools to customer relationship management service, and Shaper, to build a proprietary capability
(CRM) and financial systems to incorporate on top of foundation models.
customers’ prior sales and engagement
history. Another is fine-tuning the model with
5. Upgrade your enterprise technology
internal company documents and chat history
to act as an assistant to a customer support architecture to integrate and manage
agent. For companies that are looking to generative AI models
scale generative AI capabilities, develop more Organizations will use many generative AI models
proprietary capabilities, or meet higher security of varying size, complexity, and capability. To
or compliance needs, the Shaper archetype is generate value, these models need to be able to
appropriate. work both together and with the business’s existing
systems or applications. For this reason, building a
There are two common approaches for separate tech stack for generative AI creates more
integrating data with generative AI models in complexities than it solves. As an example, we can
this archetype. One is to “bring the model to look at a consumer querying customer service at a
the data,” where the model is hosted on the travel company to resolve a booking issue (Exhibit
organization’s infrastructure, either on-premises 2). In interacting with the customer, the generative
or in the cloud environment. Cohere, for example, AI model needs to access multiple applications and
deploys foundation models on clients’ cloud data sources.
Technology’s generational moment with generative AI: A CIO and CTO guide 5
Exhibit 1
Each archetype has its own costs.
Archetype Example use Estimated total cost of ownership
cases
— Off-the-shelf ~ $0.5 million to $2.0 million, one-time
coding assistant
— Off-the-shelf coding assistant: ~$0.5 million for integration. Costs include a team of 6 working for 3 to
for software
4 months.
developers
— General-purpose customer service chatbot: ~$2.0 million for building plug-in layer on top of 3rd-party
model API. Costs include a team of 8 working for 9 months.
— General-purpose
Taker customer ~ $0.5 million, recurring annually
service chatbot
— Model inference:
with prompt
engineering only • Off-the-shelf coding assistant: ~$0.2 million annually per 1,000 daily users
and text chat only
• General-purpose customer service chatbot: ~$0.2 million annually, assuming 1,000 customer chats
per day and 10,000 tokens per chat
— Plug-in-layer maintenance: up to ~$0.2 million annually, assuming 10% of development cost.
— Customer ~ $2.0 million to $10.0 million, one-time unless model is fine-tuned further
service chatbot
— Data and model pipeline building: ~$0.5 million. Costs include 5 to 6 machine learning engineers and
fine-tuned with
data engineers working for 16 to 20 weeks to collect and label data and perform data ETL.¹
sector-specific
knowledge and — Model fine-tuning²: ~$0.1 million to $6.0 million per training run³
chat history
• Lower end: costs include compute and 2 data scientists working for 2 months
Shaper
• Upper end: compute based on public closed-source model fine-tuning cost
— Plug-in-layer building: ~$1.0 million to $3.0 million. Costs include a team of 6 to 8 working for 6 to 12
months.
~ 0.5 million to $1.0 million, recurring annually
— Model inference: up to ~$0.5 million recurring annually. Assume 1,000 chats daily with both audio and
texts.
— Model maintenance: ~$0.5 million. Assume $100,000 to $250,000 annually for MLOps platform⁴ and
1 machine learning engineer spending 50% to 100% of their time monitoring model performance.
— Plug-in-layer maintenance: up to ~$0.3 million recurring annually, assuming 10% of development cost.
— Foundation ~ $5.0 million to $200.0 million, one-time unless model is fine-tuned or retrained
model trained
— Model development: ~$0.5 million. Costs include 4 data scientists spending 3 to 4 months on model
for assisting in
design, development, and evaluation leveraging existing research.
patient diagnosis
— Data and model pipeline: ~$0.5 million to $1.0 million. Costs include 6 to 8 machine learning engineers
and data engineers working for ~12 weeks to collect data and perform data ETL.¹
Maker — Model training⁵: ~$4.0 million to $200.0 million per training run.³ Costs include compute and labor cost
of 4 to 6 data scientists working for 3 to 6 months.
— Plug-in-layer building: ~$1.0 million to $3.0 million. Costs include a team of 6 to 8 working 6 to 12
months.
~ $1.0 million to $5.0 million, recurring annually
— Model inference: ~$0.1 million to $1.0 million annually per 1,000 users. Assume each physician sees 20
to 25 patients per day and patient speaks for 6 to 25 minutes per visit.
— Model maintenance: ~$1.0 million to $4.0 million recurring annually. Assume $250,000 annually for
MLOps platform⁴ and 3 to 5 machine learning engineers to monitor model performance.
— Plug-in-layer maintenance: up to ~$0.3 million recurring annually, assuming 10% of development cost.
Note: Through engineering optimizations, the economics of generative AI are evolving rapidly, and these are high-level estimates based on total cost of ownership
(resources, model training, etc) as of mid-2023.
1 Extract, transform, and load.
² Model is fine-tuned on data set consisting of ~100,000 pages of sector-specific documents and 5 years of chat history from ~1,000 customer representatives, which is
~48 billion tokens. Lower end cost consists of 1% parameters retrained on open-source models (eg, LLaMA) and upper end on closed-source models. Chatbot can be
accessed via both text and audio.
³ Model is optimized after each training run based on use of hyperparameters, data set, and model architecture. Model may be refreshed periodically when needed (eg, with
fresh data).
⁴ Gilad Shaham, “Build or buy your MLOps platform: Main considerations,” LinkedIn, November 3, 2021.
5 Model is trained on 65 billion to 1 trillion parameters and data set of 1.2 to 2.4 trillion tokens. The tool can be accessed via both text and audio.
6 Technology’s generational moment with generative AI: A CIO and CTO guide
Exhibit 2
Generative AI is integrated at key touchpoints to enable a tailored
Generative AI is integrated at key touchpoints to enable a tailored
ccuussttoommere jro juorunrenye.y.
Illustrative customer journey using travel agent bot API calls
Cus- Customer logs in and requests Customer reviews Customer requests Customer completes book-
tomer to change booking options live agent ing change and drops off
Disagrees
Inter- Chatbot Chatbot Chatbot Chatbot Agent Agent
action activated communi- re- pings picks up inputs
cates sponds cus- case new solu-
message tomer and tion for
Selects
and support provides review/
option
options new feedback
solution to model
Genera- Model Model Model Model Model
tive AI receives checks explains instructs instructs
model user booking issue and booking customer
request policy and gives system to support
and pulls sees cus- alternate complete system to
user info in tomer can- options task assign
prompt not make agent
change
Back- Log-in authentifi- Booking Workflow Workflow
end apps cation, model/cus- modification management management for
tomer info access policy for booking live agent
authorization management assignment
Data
source
Customer ID data Customer history Policy data Booking system Agent assignment
data data data
Infra-
structure Cloud/on-premises infrastructure and compute
and
compute
McKinsey & Company
Technology’s generational moment with generative AI: A CIO and CTO guide 7
For the Taker archetype, this level of coordination — Model hub, which contains trained and
isn’t necessary. But for companies looking to approved models that can be provisioned on
scale the advantages of generative AI as Shapers demand and acts as a repository for model
or Makers, CIOs and CTOs need to upgrade their checkpoints, weights, and parameters.
technology architecture. The prime goal is to
integrate generative AI models into internal systems — Prompt library, which contains optimized
and enterprise applications and to build pipelines to instructions for the generative AI models,
various data sources. Ultimately, it’s the maturity of including prompt versioning as models are
the business’s enterprise technology architecture updated.
that allows it to integrate and scale its generative AI
capabilities. — MLOps platform, including upgraded MLOps
capabilities, to account for the complexity of
Recent advances in integration and orchestration generative AI models. MLOps pipelines, for
frameworks, such as LangChain and LlamaIndex, example, will need to include instrumentation
have significantly reduced the effort required to to measure task-specific performance, such
connect different generative AI models with other as measuring a model’s ability to retrieve the
applications and data sources. Several integration right knowledge.
patterns are also emerging, including those that
enable models to call APIs when responding to In evolving the architecture, CIOs and CTOs will
a user query—GPT-4, for example, can invoke need to navigate a rapidly growing ecosystem
functions—and provide contextual data from an of generative AI providers and tooling. Cloud
external data set as part of a user query, a technique providers provide extensive access to at-scale
known as retrieval augmented generation. Tech hardware and foundation models, as well as a
leaders will need to define reference architectures proliferating set of services. MLOps and model
and standard integration patterns for their hub providers, meanwhile, offer the tools,
organization (such as standard API formats and technologies, and practices to adapt a foundation
parameters that identify the user and the model model and deploy it into production, while
invoking the API). other companies provide applications directly
accessed by users built on top of foundation
There are five key elements that need to be models to perform specific tasks. CIOs and CTOs
incorporated into the technology architecture to will need to assess how these various capabilities
integrate generative AI effectively (Exhibit 3): are assembled and integrated to deploy and
operate generative AI models.
— Context management and caching to
provide models with relevant information from
enterprise data sources. Access to relevant 6. Develop a data architecture to
data at the right time is what allows the model to enable access to quality data
understand the context and produce compelling The ability of a business to generate and
outputs. Caching stores results to frequently scale value, including cost reductions and
asked questions to enable faster and cheaper improved data and knowledge protections, from
responses. generative AI models will depend on how well it
takes advantage of its own data. Creating that
— Policy management to ensure appropriate advantage relies on a data architecture that
access to enterprise data assets. This control connects generative AI models to internal data
ensures that HR’s generative AI models that sources, which provide context or help fine-tune
include employee compensation details, for the models to create more relevant outputs.
example, cannot be accessed by the rest of the
organization.
8 Technology’s generational moment with generative AI: A CIO and CTO guide
Exhibit 3
The tech stack for generative AI is emerging.
The tech stack for generative AI is emerging.
Illustrative generative AI tech stack
Users
Apps Models Data
Tooling Infrastructure
Apps-as-a- Data sources Experience layer Policy
service with Embeddings, DTC² or B2B applications (eg, Jasper) management
embedded unstructured
Role-based
foundation data,
access
models analytical
API gateway control and
End-user- data, trans- content-
facing actional data based policies
applications to secure
and founda- Context management and caching enterprise
tion models data assets
User and task context retrieved from enterprise data
accessed
sources to prompt generative AI models, cache for
through a
common requests
browser
interface as
SaaS¹ (eg,
Midjourney)
Data Model hub Prompt
platforms library
Platforms that allow users to share
Vector models and data sets (eg, Hugging Face)
databases,
data
warehouse,
data lake Closed-source Open-/closed-source
foundation foundation models
models Trained model that is made
API-based, pre- accessible (eg, BLOOM)
trained models
(eg, GPT-4)
MLOps platform
Existing enterprise platforms
(eg, ERP,³ CRM⁴)
Cloud or on-premises infrastructure QA and
and compute hardware observability
QA model
outputs (eg,
checks for bias)
1Software as a service.
2Direct to consumer.
3Enterprise resource planning.
4Customer relationship management.
McKinsey & Company
Technology’s generational moment with generative AI: A CIO and CTO guide 9
In this context, CIOs, CTOs, and chief data officers with internal systems, enterprise applications,
need to work closely together to do the following: and tools, and also develops and implements
standardized approaches to manage risk, such as
— Categorize and organize data so it can be used responsible AI frameworks.
by generative AI models. Tech leaders will need
to develop a comprehensive data architecture CIOs and CTOs need to ensure that the platform
that encompasses both structured and team is staffed with people who have the right
unstructured data sources. This requires putting skills. This team requires a senior technical leader
in place standards and guidelines to optimize who acts as the general manager. Key roles include
data for generative AI use—for example, by software engineers to integrate generative AI
augmenting training data with synthetic samples models into existing systems, applications, and
to improve diversity and size; converting media tools; data engineers to build pipelines that
types into standardized data formats; adding connect models to various systems of record and
metadata to improve traceability and data data sources; data scientists to select models and
quality; and updating data. engineer prompts; MLOps engineers to manage
deployment and monitoring of multiple models and
— Ensure existing infrastructure or cloud services model versions; ML engineers to fine-tune models
can support the storage and handling of the with new data sources; and risk experts to manage
vast volumes of data needed for generative AI security issues such as data leakage, access
applications. controls, output accuracy, and bias. The exact
composition of the platform team will depend on
— Prioritize the development of data pipelines to the use cases being served across the enterprise. In
connect generative AI models to relevant data some instances, such as creating a customer-facing
sources that provide “contextual understanding.” chatbot, strong product management and user
Emerging approaches include the use of vector experience (UX) resources will be required.
databases to store and retrieve embeddings
(specially formatted knowledge) as input for Realistically, the platform team will need to work
generative AI models as well as in-context initially on a narrow set of priority use cases,
learning approaches, such as “few shot gradually expanding the scope of their work as they
prompting,” where models are provided with build reusable capabilities and learn what works
examples of good answers. best. Technology leaders should work closely with
business leads to evaluate which business cases to
fund and support.
7. Create a centralized, cross-functional
generative AI platform team
8. Tailor upskilling programs by roles
Most tech organizations are on a journey to a
product and platform operating model. CIOs and and proficiency levels
CTOs need to integrate generative AI capabilities Generative AI has the potential to massively
into this operating model to build on the existing lift employees’ productivity and augment their
infrastructure and help to rapidly scale adoption capabilities. But the benefits are unevenly
of generative AI. The first step is setting up a distributed depending on roles and skill levels,
generative AI platform team whose core focus is requiring leaders to rethink how to build the actual
developing and maintaining a platform service skills people need.
where approved generative AI models can be
provisioned on demand for use by product and Our latest empirical research using the generative
application teams. The platform team also defines AI tool GitHub Copilot, for example, helped software
protocols for how generative AI models integrate engineers write code 35 to 45 percent faster.⁵ The
5 “Unleashing developer productivity with generative AI,” June 27, 2023.
10 Technology’s generational moment with generative AI: A CIO and CTO guide
benefits, however, varied. Highly skilled developers Beyond training up tech talent, the CIO and CTO
saw gains of up to 50 to 80 percent, while junior can play an important role in building generative
developers experienced a 7 to 10 percent decline in AI skills among nontech talent as well. Besides
speed. That’s because the output of the generative understanding how to use generative AI tools for
AI tools requires engineers to critique, validate, such basic tasks as email generation and task
and improve the code, which inexperienced management, people across the business will
software engineers struggle to do. Conversely, in need to become comfortable using an array of
less technical roles, such as customer service, capabilities to improve performance and outputs.
generative AI helps low-skill workers significantly, The CIO and CTO can help adapt academy models
with productivity increasing by 14 percent and staff to provide this training and corresponding
turnover dropping as well, according to one study.⁶ certifications.
These disparities underscore the need for The decreasing value of inexperienced engineers
technology leaders, working with the chief human should accelerate the move away from a classic
resources officer (CHRO), to rethink their talent talent pyramid, where the greatest number of
management strategy to build the workforce of the people are at a junior level, to a structure more
future. Hiring a core set of top generative AI talent like a diamond, where the bulk of the technical
will be important, and, given the increasing scarcity workforce is made up of experienced people.
and strategic importance of that talent, tech Practically speaking, that will mean building the
leaders should put in place retention mechanisms, skills of junior employees as quickly as possible
such as competitive salaries and opportunities while reducing roles dedicated to low-complexity
to be involved in important strategic work for the manual tasks (such as writing unit tests).
business.
Tech leaders, however, cannot stop at hiring. 9. Evaluate the new risk landscape
Because nearly every existing role will be affected and establish ongoing mitigation
by generative AI, a crucial focus should be on practices
upskilling people based on a clear view of what Generative AI presents a fresh set of ethica |
252 | mckinsey | generative-ai-and-the-future-of-hr_final.pdf | People & Organizational Performance Practice
Generative AI and
the future of HR
A chatbot may not take your job—but it will almost certainly change it.
Here’s how to start thinking about putting gen AI to work for you.
June 2023
Generative AI: It’s powerful. It’s accessible. And in. Lean forward and figure out how to use it in a way
it’s poised to change the way we work. On this that’s productive and safe.
episode of the McKinsey Talks Talent podcast,
talent leaders Bryan Hancock and Bill Schaninger Lucia Rahilly: The immediacy of the use cases
talk with McKinsey Technology Council chair feels so novel and so lightning fast. Explain what
Lareina Yee and global editorial director Lucia generative AI is, so we’re working from a common
Rahilly about the promise and pitfalls of using gen AI definition of that term.
in HR—from recruiting to performance management
to chatbot-enabled professional growth. An edited Lareina Yee: Generative AI is a technology that
version of their discussion follows. prompts the next best answer. A lot of people have
used ChatGPT to summarize information, to draft
a response to something, by pulling together an
What’s so different—and so disruptive enormous amount of public data. But there’s also
Lucia Rahilly: There has been so much buzz in amazing imaging. I might want a song, audio, video,
recent months about generative AI and tools like or code. Code is a huge example. It’s amazing the
ChatGPT. Many people seem to be ricocheting range of things that generative AI can do in the
between wonder at the potential of these tools world, and it’s just getting started.
and fear of their inherent risks. Lareina, what’s
different about generative AI, and what’s behind its Bryan Hancock: I asked ChatGPT about myself,
disruptive potential? and it accurately reported that I do a lot of work
on talent. However, it inaccurately reported that I
Lareina Yee: A couple of things stand out about went to Cornell because it assumed that Cornell
generative AI. In November 2022, OpenAI released was the most appropriate answer based on my
ChatGPT 3.5, and within five days, there were a background instead of the University of Virginia—
million users. So the speed of adoption was unlike where I did go. I thought it was very interesting that
anything we’ve seen. you don’t necessarily get what’s right but rather
what’s logical.
For me, what was most profound about that moment
was that anyone—of any age, any education level, Lareina Yee: In some ways, that emulates how
any country—could go onto GPT, query a question we think. I’m not suggesting it’s thinking the way
or two, and find something practical or fun, like a humans do, but in many ways, we use shortcuts
poem or an essay. There was an experience there and cues to make assumptions. That is kind of why
that was accessible to everybody. We’ve seen a lot people say, “Gosh, it feels really clever.” But to your
of advancement in the technology since then, and point, Bryan, it’s not 100 percent accurate. There’s a
it’s only been a couple of months. great term for that: “hallucinating.”
A second super-interesting thing is you don’t
What gen AI means for recruiters . . .
need to be a computer scientist to leverage the
technology—it can be used in all types of jobs. Lucia Rahilly: We’ll talk more about some of the
OpenAI’s research estimates that 80 percent of risks, but let’s turn to what these kinds of generative
jobs can incorporate generative AI technology and AI capabilities mean for talent in particular. Do
capabilities into activities that happen today in work. you expect generative AI to reshape or alter the
That is a profound impact on talent and jobs, and it’s recruiting process in any meaningful way?
different than how we’ve talked about it before.
Bryan Hancock: I think it’ll reshape recruiting in two
In some ways, the genie is out of the bottle. It’s meaningful ways. The first is helping managers write
probably not the best strategy to try to put it back better job requirements. Generative technology
2 Generative AI and the future of HR
can actually pull on the skills that are required to be in the recruiting process. Does generative AI have
successful in the job. That’s not to say managers a role in accelerating that shift from credentials like
don’t need to check the end product. They’ll need college degrees to the skills that candidates are
to be that human in the loop to make sure the actually capable of contributing to the workplace?
job requirement is a good one. But gen AI can
dramatically improve speed and quality. Lareina Yee: I’m optimistic it can. One thing this
technology does extremely well is tagging—the
The other application in recruiting is candidate ability to tag unstructured data for words. There are
personalization. Right now, if you’re an organization a lot of businesses that are thinking about applying
with tens of thousands of applicants, you may or that to e-commerce, to different types of retail
may not have super customized ways of reaching experiences. But you could also apply it to talent
out to the people who have applied. With generative acquisition or looking for capabilities. Now you don’t
AI, you can include much more personalization need to look for a credential or a degree. You could
about the candidate, the job, and what other jobs look for keywords in terms of capabilities and skills.
may be available if there’s a reason the applicant
isn’t a fit. All those things are made immensely Looking at social media, how do people talk about
easier and faster through generative AI. certain capabilities? You may find there are better
words to associate with those who have those skills.
Bill Schaninger: The best application of gen AI Think of a world where you want to be able to find
is in large skill pools where you’re trying to fill candidates who have amazing experience from
a reasonably well-known job. We need a more learning on the job but don’t have PhDs or college
productive and efficient way to navigate all the degrees. I’m optimistic that this could open more
profiles coming through. Where it makes me a little doors for folks like that.
anxious is anytime it’s a novel job—a new role—or
even, in US law, a job that’s changed more than 25 Bill Schaninger: This is an interesting trade-off in
percent or 33 percent. In those cases, you have to the business world, which likes proprietary data sets
go back and revalidate the criterion by which you and grouping of profiles. The real power might be,
would judge people in or out of the pool. “How much can you get in the public domain until
you start bumping up against paywalls?”
The challenge with validation is you need a
performance criterion to regress against and Long ago, when LinkedIn was bought, the APIs got
say, “What’s the difference?” In some cases, that limited to job titles—not necessarily all the spec
means figuring out how to get that criterion out of that was underneath it. There is power in these
a data lake without encroaching on other people’s pools—in particular, in profiles of jobs—because
proprietary performance data. If you say, “Well, then you can go look at tasks and skills. I’d imagine
we’re only going to use our data as the employer,” there’s going to be a race here toward figuring
then you are only basing the criterion off people out how we can piece these together to form the
you’ve already hired. And to validate, you have to ontological cloud, if you will, of “these 17 things
look at the people you didn’t hire. describe this skill.” Because it really is about skills
and not credentials.
So it doesn’t mean the technology can’t be used. It
just means there’s probably a little bit more front-
end work on applying it to novel jobs and a wide- . . . And what it means for
open opportunity for the big skill pools. professional growth
Bryan Hancock: You can also think about this
Lucia Rahilly: We talk a lot about having over- as aiding a skill-based transition not just from
indexed on credentials and under-indexed on skills the employer’s perspective but from the
Generative AI and the future of HR 3
candidate’s or employee’s perspective. In the provide much more transparency; you can actually
current world, if you’re somebody who may have see how close you are to a lot of things. I love it for
some skills but don’t have a very clear view of what the employee experience part. I get anxious about
your career opportunities might be, you are highly the selection part just because we’re still not sure
dependent on a manager or somebody taking about what’s in the data lake and how good people
an interest in you and helping to navigate you to are at prompting the AI.
“nontraditional” paths.
Lareina Yee: Right. It’s great to give you some
But in a world of generative AI, you could have a options, but it’s not an answer or a recommendation
conversation with a very intelligent chatbot and engine. Your judgment matters.
say, “Hey, here are my skills and experiences. What
jobs could be open to me?” And it could come back Bryan Hancock: Another thing we’re seeing is that
and say, “Well, most people with your skill profile do ChatGPT—and generative AI more broadly—can
these things, but some do A, B, C,” with “C” being be particularly good at getting new workers more
coding. And then, you could say, “Tell me what quickly up to speed.
these jobs in coding would be,” and it could pull a
job description for a coder that is not just geared There’s interesting research that Erik Brynjolfsson
toward an IT person but translated into words you at Stanford, along with others from MIT, have
understand. Then you could say, “OK, this is great. recently come out with, which looks at call-center
I’m interested. What learning experiences do I workers. They found that generative AI functionality
need?” And generative AI could tell you what those wasn’t all that helpful for the most experienced
learning experiences are. representatives. It was incredibly helpful with
new folks because they were able to get that
So for somebody who has the innate ability but not institutional knowledge much more quickly. It was
the visibility, generative AI can illuminate a range of at their fingertips. They could ask a question and
career paths and start helping people understand get the answer. So the productivity of new folks was
how to get there. dramatically higher. Generative AI really gets you
80–90 percent of the way to full proficiency.
Lareina Yee: Imagine I’m ten years into my
career and I’m feeling a little stuck. What if I had a Lareina Yee: Bryan, I love that, and I share
professional development AI assistant that helped the optimism.
me think through questions like, “What type of job
should I seek? What are the types of roles within my
company? How do I think about them?” and “What What’s new for the performance review
classes would I take?” as opposed to waiting for
Bryan Hancock: One of my personal favorite uses
someone to reskill me—which sounds awful. How
for generative AI on the people front is actually for
do I take the initiative ten years into my career to
performance reviews. Hear me out: I don’t want
build the skill sets and understand the range of jobs
generative AI actually generating somebody’s
available for my capabilities? That would be so cool.
performance review. That needs the human in the
loop, needs human judgment, needs empathy.
Bill Schaninger: Depending on the regulatory
environment you’re in, you’re not allowed to make
But let me use this example of what I do as a
any selection decision without a human being
McKinsey evaluator: I get written feedback from
involved. This is particularly true in the EU. It’s a nice
15 to 20 individuals. They enter it into a digital
way of augmenting human work but not cutting out
system. I’ve got long-form feedback. I look at
the decision making. On the employee side, it should
upward feedback scores that include written
4 Generative AI and the future of HR
commentary as well as specific number-based that. I think there’s a lot that enhances what we’ve
scores. I look at how often people were actually been trying to do so laboriously for years.
deployed on engagements. I look at compliance-
related measures. Did they turn in their stuff Bill Schaninger: We talk about putting the manager
on time? A whole range of things. For me, as an back in performance management. Every time you
evaluator, getting to a first draft is an incredibly talk to somebody about something good or bad, log
arduous process. I take pride in the time and the it away. That way, at the end of the year, it’s more of
thoughtfulness that goes into it. an aggregation and synthesis, and it’s not a surprise
to anyone. But that requires regular entry. So while
But what if I could hit a button and get a draft? I love what you’re describing, it’s not the tech that
When I have each of the conversations with the does that; it’s the people committing to the common
15 people that best know the person I’m evaluating, data capture and the common approaches that
what if I had a draft I was already working from? It’s enable it.
not a replacement for going through everything, but
that initial synthesis would help me get more quickly Bryan Hancock: Your point is well-taken. Then, as
to what I really need to probe for that person’s an evaluator, I apply my human judgment.
development and growth.
Bill Schaninger: The normative data is nice.
I’m excited about that use case because it When we get our sponsorship and mentorship
eliminates a lot of work. At first, many people would data at McKinsey, we see how we compare to
think, “I’d never want generative AI anywhere near other partners in a given region. If you don’t have a
performance reviews.” But it’s exciting if we think reference point, though, how would you know what
of this as a productivity aid or as something that “good” actually is? When you get the normative
helps us be even better. data, you can start getting some guidance. I like all
that, and it’s all enabled by huge amounts of data.
Lareina Yee: Now let’s talk about the employee
he’s evaluating. The employee gets the feedback, If this enables a more robust and wholesome view
and Bryan probably wrote it clearly, and he of actual performance, it makes it a whole lot easier
delivered it with empathy, so the person is feeling, to have a difficult performance conversation. We
“OK, I’ve got some strengths, and I’ve got some need to put the manager back in performance
development needs.” management. But can we make it easier on
managers so they can spend the time managing
But what if I, as the employee, can query, instead of scribbling out a schedule or knitting
“Who are five success models with my strengths together 15 data points?
and weaknesses, and what have they gone on to
do? How can I visualize my career development?
How can I continue to work on it?” I could also Bias and other risks
have an assistant that helps me map my Lucia Rahilly: Let’s talk a bit more about some of
professional development. In that way, when we the risks. Generative AI learns based on historical
check in a year later, I’ve really improved and data, and historical patterns of data reflect historical
increased my aspirations. biases. By relying on generative-AI-driven tools,
what’s the risk we are inadvertently propagating
What if Bill is someone I should model myself on? these inherited biases?
Instead of Bryan having to introduce me to Bill,
generative AI helps me realize that I’ve got the Lareina Yee: Certainly, today, generative AI can
makings of a Bill Schaninger. I can be inspired by amplify bias.
Generative AI and the future of HR 5
Let’s say I’m recruiting, and I describe some The other thing is there’s a real opportunity for what
different qualifications. I’m looking at urban we typically call “change management.” If you don’t
centers of talent, and I decide I’d like to look for think through how the technology changes the job,
basketball captains; or perhaps, instead, I say that workflow, or collaboration model, then you’re not
lacrosse captains are desirable. These are team necessarily directing that additional time toward
sports with captains and leadership, so in some something that’s more value added. You need to
way that makes sense. think about how it affects the rest of the workday
and workweek.
But if you look at demographics, who plays
basketball in cities is very different from who plays Bill Schaninger: In many cases, we’d like to blame
lacrosse. And so, by emphasizing lacrosse, you the technology and not highlight the poor problem
will typically get more young White male leaders, solving that happened just before implementing
whereas if you chose basketball, you might find it. Getting a better, shinier tool that’s faster and
more African Americans or Latinos. What about more expansive doesn’t relieve you of the burden of
softball, where we see women? What happens if, thinking things through.
instead, we select a whole set of sports? Even then,
just the selection of the sports as a filter could Lareina Yee: The bigger thing to call out here is that
amplify bias in the questioning. I think the power of three of us have spent this time thinking about all
the question is on us as humans. the positive intentions and the ways we can use this
for good. But there are probably people who are
Bryan Hancock: Of course there are also thinking about this technology and asking, “How
intellectual property concerns. can I use this for harm?” Traditionally, this is why
government regulation, policy, and international
But I also think there’s a risk of us all becoming less standards play a fundamental role in our society. I
interesting. If you are somebody in a creative field don’t think you can completely leave it to the private
and you leverage generative AI to get your output sector to self-regulate.
up from six articles a week to 12, you’re spending
less time per article. You may need to do that to get
to publication in time, but that also means you’re Preparing for the inevitable
not spending as much time in the shower, on a Lucia Rahilly: A big concern for people is that these
run, or in the car thinking about the articles. Your kinds of tools will eliminate their job or—potentially
productivity will go up, but you may not necessarily even worse—become their bosses. What do you
have as much time for creative thinking. We think people can do now to prepare for the changes
know that the most creative thoughts come from that are coming with generative AI?
downtime—when you’re doing something else and
letting your mind wander. Bill Schaninger: I would try to make it easier for
them to learn and play with it. This is better than
This risk of being less interesting is important, and continuing to try to resist it. I don’t think we should
one that we may not have fully thought through yet. become beholden to these fears.
Lareina Yee: Precisely. There are a lot of risks. Let’s Lucia Rahilly: And assuming HR and talent
also think about leaders who are implementing processes become increasingly automated,
this technology. Often people had a workflow how can leaders ensure that generative AI doesn’t
where they would think about a technology and the get in the way of what Bryan called “the human in
business return on investment, and only at the end the loop?”
would ask, “Are there any risks we should worry
about?” I would strongly recommend that you think Lareina Yee: Leaders have a huge role to play in
about risk up front in the workflow design. two ways. One is to modernize and leapfrog their
6 Generative AI and the future of HR
own talent capabilities within their functions. And get managers more consistently up to the level
Find more content like this on the
second, if 80 percent of their workforce is shifting, of performance that HR leaders have always
McKinsey Insights App
they play a huge role in how that happens and how wanted them to achieve instead of working on
it affects employees at their companies. I think administrative tasks. I hope that HR would view
leaders have a huge voice at the table. this as an opportunity to routinize and get rid of
the work that they don’t have to do. Then for the
Bryan Hancock: It’s a tremendous opportunity for work that they do have to do, they can use this
HR to increase access to opportunities for huge technology to find a way to get better answers
swaths of their workforce. It’s an opportunity to more quickly.
Scan • Download • Personalize
Bryan Hancock is a partner in McKinsey’s Washington, DC, office; Bill Schaninger is a senior partner emeritus in the
Philadelphia office; and Lareina Yee is a senior partner in the Bay Area office. Lucia Rahilly is the global editorial director and
deputy publisher of McKinsey Global Publishing and is based in the New York office.
Designed by McKinsey Global Publishing
Copyright © 2023 McKinsey & Company. All rights reserved.
Generative AI and the future of HR 7 |
253 | mckinsey | whats-the-future-of-generative-ai-an-early-view-in-15-charts.pdf | McKinsey Explainers
What’s the future
of generative AI?
An early view in
15 charts
Generative AI has hit the ground running—so fast that it
can feel hard to keep up. Here’s a quick take pulled from our
top articles and reports on the subject.
August 2023
Since the release of ChatGPT in November 2022, it’s been all over the McKinsey research has found that
headlines, and businesses are racing to capture its value. Within the
technology’s first few months, McKinsey research found that generative generative AI features stand to add
AI (gen AI) features stand to add up to $4.4 trillion to the global
economy—annually. up to $4.4 trillion to the global
The articles and reports we’ve published in this time frame examine
economy—annually.
questions such as these:
— What will the technology be good at, and how quickly?
— What types of jobs will gen AI most affect?
— Which industries stand to gain the most?
— What activities will deliver the most value for organizations?
— How do—and will—workers feel about the technology?
— What safeguards are needed to ensure responsible use of gen AI?
In this visual Explainer, we’ve compiled all the answers we have so far—
in 15 McKinsey charts. We expect this space to evolve rapidly and will
continue to roll out our research as that happens. To stay up to date on
this topic, register for our email alerts on “artificial intelligence” here.
What’s the future of generative AI? An early view in 15 charts 2
Web <2023>
<Future of GenAI>
Exhibit <1> of <15>
Gen AI finds its legs Generative AI has been evolving at a rapid pace.
The advanced machine learning that powers Timeline of major large language model (LLM) developments following ChatGPT’s launch
gen AI–enabled products has been decades
Nov 2022 Dec Jan 2023 Feb Mar Apr
in the making. But since ChatGPT came off the
starting block in late 2022, new iterations of gen
1 2 3 4 5 6 7 8 9 10 11 12 13
AI technology have been released several times
a month. In March 2023 alone, there were six 1 Nov 30, 2022: OpenAI’s 4 Feb 2, 2023: Amazon’s 7 Mar 7: Salesforce 10 Mar 16: Microsoft
major steps forward, including new customer ChatGPT, powered by multimodal-CoT model announces Einstein GPT announces the integration
GPT-3.5 (an improved incorporates “chain-of- (leveraging OpenAI’s of GPT-4 into its
relationship management solutions and support version of its 2020 GPT-3 thought prompting,” in models), the first Office 365 suite,
for the financial services industry. release), becomes the first which the model explains generative AI technology potentially enabling broad
widely used text- its reasoning, and for customer relationship productivity increases
generating product, outperforms GPT-3.5 on management
Source: What every CEO should know about gaining a record several benchmarks 11 Mar 21: Google releases
100 million users in 8 Mar 13: OpenAI releases Bard, an AI chatbot based
generative AI
2 months 5 Feb 24: As a smaller GPT-4, which offers on the LaMDA family
model, Meta’s LLaMA is significant improvements of LLMs
2 Dec 12: Cohere releases more efficient to use than in accuracy and
the first LLM that some other models but hallucinations mitigation, 12 Mar 30: Bloomberg
supports more than continues to perform claiming 40% announces an LLM
100 languages, making it well on some tasks com- improvement vs GPT-3.5 trained on financial data
available on its enterprise pared with other models to support natural-
AI platform 9 Mar 14: Anthropic language tasks in the
6 Feb 27: Microsoft introduces Claude, an AI financial industry
3 Dec 26: LLMs such as introduces Kosmos-1, assistant trained using a
Google’s Med-PaLM are a multimodal LLM that method called 13 Apr 13: Amazon
trained for specific use can respond to image and “constitutional AI,” which announces Bedrock, the
cases and domains, such audio prompts in addition aims to reduce the first fully managed service
as clinical knowledge to natural language likelihood of harmful that makes models
outputs available via API from
multiple providers in
addition to Amazon’s own
Titan LLMs
McKinsey & Company
What’s the future of generative AI? An early view in 15 charts 3
Web <2023>
<Future of GenAI>
Exhibit <2> of <15>
The road to human-level Due to generative AI, experts assess that technology could achieve human-
performance just got shorter level performance in some capabilities sooner than previously thought.
For most of the technical capabilities shown Estimated range for technology to achieve human-level performance, by technical capability
in this chart, gen AI will perform at a median
Post-recent generative AI developments (2023)¹ Pre-generative AI (2017)¹
level of human performance by the end of this
Median Top quartile Median Top quartile
decade. And its performance will compete with
the top 25 percent of people completing any 2010 2020 2030 2040 2050 2060 2070 2080
and all of these tasks before 2040. In some
Coordination with multiple agents
cases, that’s 40 years faster than experts
Creativity
previously thought.
Logical reasoning and problem solving
Source: The economic potential of generative
Natural-language generation
AI: The next productivity frontier
Natural-language understanding
Output articulation and presentation
Generating novel patterns and categories
Sensory perception
Social and emotional output
Social and emotional reasoning
Social and emotional sensing
¹Comparison made on the business-related tasks required from human workers.
Source: McKinsey Global Institute occupation database; McKinsey analysis
McKinsey & Company
What’s the future of generative AI? An early view in 15 charts 4
Web <2023>
<Future of GenAI>
Exhibit <3> of <15>
And automation of knowledge Advances in technical capabilities could have the most impact on activities
work is now in sight performed by educators, professionals, and creatives.
Impact of generative AI on technical automation Without generative AI¹
Previous waves of automation technology
potential in midpoint scenario, 2023 With generative AI
mostly affected physical work activities, but
gen AI is likely to have the biggest impact Overall technical automation potential, Share of global
Occupation group comparison in midpoint scenarios, 2023, % employment,2 %
on knowledge work—especially activities
involving decision making and collaboration. Educator and workforce training 15 54 4
Professionals in fields such as education, law,
Business and legal 32
technology, and the arts are likely to see parts professionals 62 5
of their jobs automated sooner than previously 28
STEM professionals 57 3
expected. This is because of generative AI’s
39
ability to predict patterns in natural language Community services 65 3
and use it dynamically.
28
Creatives and arts management 53 1
Source: The economic potential of generative 66
Office support 87 9
AI: The next productivity frontier
27
Managers 44 3
29
Health professionals 43 2
45
Customer service and sales 57 10
29
Property maintenance 38 4
Health aides, technicians, 34
and wellness 43 3
73
Production work 82 12
70
Food services 78 5
42
Transportation services 49 3
Mechanical installation 61
and repair 67 4
59
Agriculture 63 21
49
Builders 53 7
51
Overall 100
63
Note: Figures may not sum, because of rounding.
¹Previous assessment of work automation before the rise of generative AI.
2Includes data from 47 countries, representing about 80% of employment across the world.
Source: McKinsey Global Institute analysis
McKinsey & Company
What’s the future of generative AI? An early view in 15 charts 5
Web <2023>
<Future of GenAI>
Exhibit <4> of <15>
Apps keep proliferating to There are many applications of generative AI across modalities.
address specific use cases
Generative AI use cases, nonexhaustive
Gen AI tools can already create most types of Modality Application Example use cases
written, image, video, audio, and coded content.
Text Content writing
And businesses are developing applications to
address use cases across all these areas. In the Chatbots or assistants
near future, we expect applications that target Search
specific industries and functions will provide
Analysis and synthesis
more value than those that are more general.
Code Code generation
Source: Exploring opportunities in the
generative AI value chain Application prototype
and design
Data set generation
Image Stock image generator
Image editor
Audio Text to voice generation
Sound creation
Audio editing
3-D 3-D object generation
or other
Product design and
discovery
Video Video creation
Video editing
Voice translation
and adjustments
Face swaps and
adjustments
McKinsey & Company What’s the future of generative AI? An early view in 15 charts 6
Generative AI will affect business functions differently across industries.
Some industries will gain more
Generative AI productivity
t G
o
dh
f
ie
f
f
fn aa
e
cA rn etI o’ ns ro
t
sp b,t r useh
u
sc ice i ns
h
er e
a
s
s im
s
s
t
fp
h
ua nec cmt
t
w
ii
oxi
n
l al
s
nd
,
de ap
sim
e wn
p
ed
o
l
lro atn
a
s
n
a
tc
h
v
e
ea or si
f
ce aty
le
i Lm owp ia mc pt a b cty business functio Hn igs h¹ impact MarketingC au ns dto sm ae ler
s
operatiP or nS o so df utS cwu ta p Rrp e &l y e
D
nc gh ia ni en
e
a rn ind
g
operatiR onis skS t ar nat de lg ey
g
a an ld finaC ncT oa erl pe on rt
a
a ten d
IT
o 2rganization
Total, % of
of an industry’s revenue. Nearly all industries industry Total, 760– 340– 230– 580– 290– 180– 120– 40– 60–
revenue $ billion 1,200 470 420 1,200 550 260 260 50 90
will see the most significant gains from
Administrative and
deployment of the technology in their marketing professional services 0.9–1.4 150–250
and sales functions. But high tech and banking Advanced electronics
and semiconductors 1.3–2.3 100–170
will see even more impact via gen AI’s potential
to accelerate software development. Advanced manufacturing3 1.4–2.4 170–290
Agriculture 0.6–1.0 40–70
Source: The economic potential of generative
Banking 2.8–4.7 200–340
AI: The next productivity frontier
Basic materials 0.7– 1.2 120–200
Chemical 0.8–1.3 80–140
Construction 0.7–1.2 90–150
Consumer packaged goods 1.4–2.3 160–270
Education 2.2–4.0 120–230
Energy 1.0– 1.6 150–240
Healthcare 1.8–3.2 150–260
High tech 4.8–9.3 240–460
Insurance 1.8– 2.8 50–70
Media and entertainment 1.8– 3.1 80–130
Pharmaceuticals and
2.6–4.5 60–110
medical products
Public and social sector 0.5–0.9 70–110
Real estate 1.0–1.7 110–180
Retail4 1.2–1.9 240–390
Telecommunications 2.3–3.7 60–100
Travel, transport, and logistics 1.2–2.0 180–300
2,600–4,400
Note: Figures may not sum to 100%, because of rounding. 1Excludes implementation costs (eg, training, licenses). 2Excluding software engineering. 3Includes aero-
space, defense, and auto manufacturing. 4Including auto retail. Source: Comparative Industry Service (CIS), IHS Markit; Oxford Economics; McKinsey Corporate and
Business Functions database; McKinsey Manufacturing and Supply Chain 360; McKinsey Sales Navigator; Ignite, a McKinsey database; McKinsey analysis
McKinsey & Company What’s the future of generative AI? An early view in 15 charts 7
Web <2023>
<Future of GenAI>
Exhibit <6> of <15>
So understanding the use Generative AI could deliver significant value when deployed in some use
cases that will deliver the most cases across a selection of top industries.
value to your industry is key
Selected examples of key use cases for main Value potential High
of function for
functional value drivers (nonexhaustive)
the industry Low
Our report, The economic potential of
generative AI: The next productivity frontier, Total value
potential Value
contains spotlight sections detailing how to per industry, potential,
identify the use cases with the highest value $ billion (% as % of Product R&D,
of industry operating software Customer Marketing Other
potential in the banking, life sciences, and retail revenue) profits1 engineering operations and sales functions
and consumer-packaged-goods industries.
Banking 200–340 9–15 Legacy code Customer Custom retail Risk model
These provide a good framework for assessing (3–5%) conversion emergency banking offers documentation
interactive voice
your own industry. Optimize migration Push personalized Create model
response (IVR)
of legacy marketing and sales documentation,
frameworks with Partially automate, content tailored for and scan for
Source: The economic potential of generative natural-language accelerate, and each client of the missing
translation enhance resolution bank based on documentation
AI: The next productivity frontier
capabilities rate of customer profile and history and relevant
emergencies through (eg, personalized regulatory
generative nudges), and updates
AI–enhanced IVR generate alternatives
interactions (eg, for for A/B testing
credit card losses)
Retail 400–660 27–44 Consumer research Augmented Assist copy writing Procurement
and (1–2%) reality–assisted for marketing suppliers
Accelerate consumer
consumer customer support content creation process
research by testing
packaged enhancement
scenarios, and Rapidly inform the Accelerate writing of
goods2
enhance customer workforce in real copy for marketing Draft playbooks
targeting by creating time about the status content and for negotiating
“synthetic customers” of products and advertising scripts with suppliers
to practice with consumer
preferences
Pharma 60–110 15–25 Research and Customer Generate content Contract
and (3–5%) drug discovery documentation for commercial generation
medical generation representatives
Accelerate the Draft legal
products
selection of proteins Draft medication Prepare scripts for documents
and molecules best instructions and risk interactions with incorporating
suited as candidates notices for drug physicians specific
for new drug resale regulatory
formulation requirements
¹Operating profit based on average profitability of selected industries in the 2020–22 period.
2Includes auto retail.
McKinsey & Company
What’s the future of generative AI? An early view in 15 charts 8
Web <2023>
<Future of GenAI>
Exhibit <7> of <15>
Despite gen AI’s commercial Commercial leaders are already leveraging generative AI use cases—but
promise, most organizations most feel the technology is underutilized.
aren’t using it yet
Reported use of technology at organization¹ and level at which respondents think it should be used,2
% of respondents at commercially leading organizations
When we asked marketing and sales leaders
Machine learning Generative AI
how much they thought their organization
should be using gen AI or machine learning Almost always
for commercial activities, 90 percent thought 15 Often
20
it should be at least “often.” That’s hardly 25 Sometimes
surprising, given that marketing and sales is Rarely
40
Almost never
the area with the most potential for impact, 20
as we saw earlier. But 60 percent said their 20
organizations rarely or never do this.
Source: AI-powered marketing and sales reach
new heights with generative AI
40 65
50 55
25
10 10
5
Currently Should Currently Should
use use use use
1Senior executives in significant global B2B and B2C sales and marketing organizations across a wide range of industries and company maturity levels were
asked: To what extent is your organization using machine learning/generative AI solutions?
2Q: How much do you think your organization should be using machine learning/generative AI solutions?
McKinsey & Company
What’s the future of generative AI? An early view in 15 charts 9
Web <2023>
<Future of GenAI>
Exhibit <8> of <15>
Marketing and sales leaders Commercial leaders are cautiously optimistic about generative AI use
are most enthusiastic about cases, anticipating moderate to significant impact.
three use cases
Share of respondents estimating the impact of generative AI on use case as “significant” or
“very significant,”1 % of respondents at commercially leading organizations
Our research found that marketing and sales
leaders anticipated at least moderate impact Lead identification
(real time, based on customer trends) 60
from each gen AI use case we suggested.
They were most enthusiastic about lead Marketing optimization
(A/B testing, SEO strategies) 55
identification, marketing optimization, and
personalized outreach. Personalized outreach
(chatbots, virtual assistants) 53
Source: AI-powered marketing and sales reach Dynamic content
(websites, marketing collateral) 50
new heights with generative AI
Up-/cross-selling recs
(via usage patterns, support tickets) 50
Success analytics
(continuous churn modeling) 45
Marketing analytics
(dynamic audience targeting) 45
Dynamic customer-journey mapping
(identifying critical touchpoints) 45
Automated marketing workflows
(nurturing campaigns) 35
Sales analytics
(predictive pricing, negotiation) 30
Sales coaching
(hyperpersonalized training) 25
1Senior executives in significant global B2B and B2C sales and marketing organizations across a wide range of industries and company maturity levels were
asked: Please share your estimated ROI/impact these tools would have if implemented in your organization.
Source: McKinsey analysis
McKinsey & Company
What’s the future of generative AI? An early view in 15 charts 10
Web <2023>
<Future of GenAI>
Exhibit <9> of <15>
Software engineering, the Generative AI can increase developer speed, but less so for complex tasks.
other big value driver for many
Reduction in task completion Code Code Code High-complexity
industries, could get much
time using generative AI,1 % documentation generation refactoring tasks
more efficient
When we had 40 of McKinsey’s own
<10
developers test generative AI–based tools,
we found impressive speed gains for many
common developer tasks. Documenting
code functionality for maintainability (which
considers how easily code can be improved) can
be completed in half the time, writing new code
in nearly half the time, and optimizing existing 20–30
code (called code refactoring) in nearly two-
thirds the time.
Source: Unleashing developer productivity with 35–45
generative AI
45–50
1Compared with task completion without the use of generative AI.
Source: McKinsey analysis
McKinsey & Company
What’s the future of generative AI? An early view in 15 charts 11
Web <2023>
<Future of GenAI>
Exhibit <10> of <15>
And gen AI assistance could Generative AI tools have potential to improve the developer experience.
make for happier developers
Developer sentiments, Strongly Somewhat Neither agree Somewhat Strongly
% of respondents agree agree nor disagree disagree disagree
Our research found that equipping developers
with the tools they need to be their most Able to focus on satisfying
productive also significantly improved Felt happy and meaningful work Was in a ‘flow’ state
their experience, which in turn could help
Without With Without With Without With
companies retain their best talent. Developers generative AI generative AI generative AI generative AI generative AI generative AI
using generative AI–based tools were more
than twice as likely to report overall happiness,
fulfillment, and a state of flow. They attributed
44
this to the tools’ ability to automate grunt work 50 31
that kept them from more satisfying tasks
and to put information at their fingertips faster 20 25
15
than a search for solutions across different
online platforms.
30 38 30 56 30 50
Source: Unleashing developer productivity with
generative AI
POSITIVE POSITIVE
25 13 30 20
6
NEGATIVE 13 NEGATIVE
20
15
30
5 5
Note: Figures may not sum to 100%, because of rounding.
McKinsey & Company
What’s the future of generative AI? An early view in 15 charts 12
Web <2023>
<State of AI 2023>
Exhibit <1PDF> of <11>
Momentum among workers for Respondents across regions, industries, and seniority levels say they are
using gen AI tools is building already using generative AI tools.
A new McKinsey survey shows that the vast Reported exposure to generative AI tools, % of respondents
majority of workers—in a variety of industries
Regularly use Regularly use for work Regularly use Have tried at No Don’t
and geographic locations—have tried for work and outside of work outside of work least once exposure know
generative AI tools at least once, whether in or
outside work. That’s pretty rapid adoption less By office location Asia–Pacific 4 18 19 36 19 3
than one year in. One surprising result is that Developing markets 9 11 20 34 23 3
baby boomers report using gen AI tools for work Europe 10 14 11 45 15 6
more than millennials.
Greater China 9 10 18 46 14 3
North America 6 22 13 38 19 3
Source: The state of AI in 2023: Generative AI’s
By industry Advanced industries 5 11 16 47 15 5
breakout year
Business, legal, and professional services 7 16 13 41 21 2
Consumer goods/retail 7 11 12 40 26 4
Energy and materials 6 8 15 50 19 3
Financial services 8 16 18 41 14 4
Healthcare, pharma, and medical products 6 10 17 44 15 7
Technology, media, and telecom 14 19 17 37 9 3
By job title C-suite executives 8 16 13 42 18 2
Senior managers 10 14 16 42 15 3
Midlevel managers 7 16 20 35 19 4
By age Born in 1964 or earlier 6 17 21 30 18 9
Born 1965–80 7 18 18 37 17 3
Born 1981–96 5 22 24 36 11 3
By gender identity Men 8 16 16 37 19 4
Women 12 15 6 46 18 3
Note: Figures may not sum to 100%, because of rounding. In Asia–Pacific, n = 164; in Europe, n = 515; in North America, n = 392; in Greater China (includes
Hong Kong and Taiwan), n = 337; and in developing markets (includes India, Latin America, and Middle East and North Africa), n = 276. For advanced industries
(includes automotive and assembly, aerospace and defense, advanced electronics, and semiconductors), n = 96; for business, legal, and professional services,
n = 215; for consumer goods and retail, n = 128; for energy and materials, n = 96; for financial services, n = 248; for healthcare, pharma, and medical products,
n = 130; and for technology, media, and telecom, n = 244. For C-suite respondents, n = 541; for senior managers, n = 437; and for middle managers, n = 339.
For respondents born in 1964 or earlier, n = 143; for respondents born between 1965 and 1980, n = 268; and for respondents born between 1981 and 1996,
n = 80. Age details were not available for all respondents. For respondents identifying as men, n = 1,025; for respondents identifying as women, n = 156.
The survey sample also included respondents who identified as “nonbinary” or “other” but not a large enough number to be statistically meaningful.
Source: McKinsey Global Survey on AI, 1,684 participants at all levels of the organization, April 11–21, 2023
McKinsey & Company
What’s the future of generative AI? An early view in 15 charts 13
Web <2023>
<Future of GenAI>
Exhibit <12> of <15>
But organizations still Job postings for fields related to tech trends grew by 400,000 between
need more gen AI–literate 2021 and 2022, with generative AI growing the fastest.
employees
Tech trend job postings, 2021–22,1 thousands
As organizations begin to set gen AI goals, 700
they’re also developing the need for more 600
+6%
gen AI–literate workers. As generative and
500
other applied AI tools begin delivering value to +29%
400
early adopters, the gap between supply and +12%
demand for skilled workers remains wide. To 300
+16% +15%
stay on top of the talent market, organizations 200
should develop excellent talent management
100
capabilities, delivering rewarding working 2021 2022
0
experiences to the gen AI–literate workers they Applied AI Next-generation Cloud and edge Trust architectures Future of
hire and hope to retain. software development computing and digital identity mobility
300
Source: McKinsey Technology Trends Outlook
2023 200 +27%
100 +8% +7%
+10% +23%
0
Electrification and Climate tech beyond Advanced Immersive-reality Industrializing
renewables electrification and connectivity technologies machine learning
renewables
200
+40% +16% +44% +12%
100
–19%
0
Web3 Future of Future of space Generative AI Quantum
bioengineering technologies technologies
1Out of 150 million surveyed job postings. Job postings are not directly equivalent to numbers of new or existing jobs.
Source: McKinsey’s proprietary Organizational Data Platform, which draws on licensed, de-identified public professional profile data
McKinsey & Company
What’s the future of generative AI? An early view in 15 charts 14
Web <2023>
<Future of GenAI>
Exhibit <13> of <15>
Organizations should proceed Inaccuracy, cybersecurity, and intellectual property infringement are the
with caution most-cited risks of generative AI adoption.
Generative AI–related risks that organizations consider relevant and are working to mitigate,
The possibilities of gen AI are thrilling to many.
% of respondents1
But like any new technology, gen AI doesn’t
come without potential risks. For one thing, gen Organization considers risk relevant Organization working to mitigate risk
AI has been known to produce content that’s Inaccuracy 56 32
biased, factually wrong, or illegally scraped from
Cybersecurity 53 38
a copyrighted source. Before adopting gen AI
Intellectual property infringement 46 25
tools wholesale, organizations should reckon
Regulatory compliance 45 28
with the reputational and legal risks to which
Explainability 39 18
they may become exposed. One way to mitigate
Personal/individual privacy 39 20
the risk? Keep a human in the loop; that is, make
Workforce/labor displacement 34 13
sure a real human checks any gen AI output
Equity and fairness 31 16
before it’s published or used.
Organizational reputation 29 16
National security 14 4
Source: The state of AI in 2023: Generative AI’s
Physical safety 11 6
breakout year
Environmental impact 11 5
Political stability 10 2
None of the above 1 8
1Asked only of respondents whose organizations have adopted Al in at least 1 function. For both risks considered relevant and risks mitigated, n = 913.
Source: McKinsey Global Survey on AI, 1,684 participants at all levels of the organization, April 11–21, 2023
McKinsey & Company
What’s the future of generative AI? An early view in 15 charts 15
Web <2023>
<Future of GenAI>
Exhibit <14> of <15>
Gen AI could ultimately boost Generative AI could contribute to productivity growth if labor hours can be
global GDP redeployed effectively.
McKinsey has found that gen AI could Productivity impact from automation by scenario, 2022–40, CAGR,¹ %
substantially increase labor productivity
Without generative AI² Additional with generative AI
across the economy. To reap the benefits
of this productivity boost, however, workers Global³ Developed economies
whose jobs are affected will need to shift to Japan Germany France United States
other work activities that allow them to at least 4.2
3.9
match their 2022 productivity levels. If workers 0.6 3.7 3.6
0.6
3.3
are supported in learning new skills and, in 0.7 0.7
0.6
some cases, changing occupations, stronger
global GDP growth could translate to a more
sustainable, inclusive world. 2.6 3.7 1.6 0.1 3.4 1.3 0.2 3.0 2.9
0.8 0.6
0.2 1.4 1.1 0.2 0.3
Source: The economic potential of generative 0.1 0.6 0.4
AI: The next productivity frontier Early Late Early Late Early Late Early Late Early Late
Emerging economies
China Mexico India South Africa
3.8
0.6
2.9
0.6 2.3 2.3
0.5 0.5
3.2
0.8 2.3
0.1 1.8 1.7
0.7 0.0 0.0 0.0
Early Late Early Late Early Late Early Late
Note: Figures may not sum, because of rounding.
1Based on the assumption that automated work hours are reintegrated in work at productivity level of today.
2Previous assessment of work automation before the rise of generative AI.
3Based on 47 countries, representing about 80% of world employment.
Source: Conference Board Total Economy database; Oxford Economics; McKinsey Global Institute analysis
McKinsey & Company
What’s the future of generative AI? An early view in 15 charts 16
Web <2023>
<Future of GenAI>
Exhibit <15> of <15>
Gen AI represents just a small Generative AI could create additional value potential above what could be
piece of the value potential unlocked by other AI and analytics.
from AI
AI’s potential impact on the global economy, $ trillion
Gen AI is a big step forward, but traditional 17.1–25.6
advanced analytics and machine learning
13.6–22.1
continue to account for the lion’s share of
6.1–7.9
task optimization, and they continue to find 2.6–4.4
11.0–17.7
new applications in a wide variety of sectors.
~15–40% ~35–70%
Organizations undergoing digital and AI
incremental incremental
transformations would do well to keep an eye on
economic impact economic impact
gen AI, but not to the exclusion of other AI tools.
Just because they’re not making headlines
doesn’t mean they can’t be put to work to deliver
increased productivity—and, ultimately, value.
Source: The economic potential of generative Advanced analytics, New generative Total use All worker productivity Total AI
traditional machine AI use cases case–driven enabled by generative AI, economic
AI: The next productivity frontier
learning, and deep potential including in use cases2 potential2
learning1
1Updated use case estimates from "Notes from the AI frontier: Applications and value of deep learning,” McKinsey Global Institute, April 17, 2018.
2The range of potential value from the combined impact of new generative AI use cases and the increased worker productivity they could enable is $6.1 trillion
to $7.9 trillion, including revenue impacts conservatively translated into productivity impact as difference between total impact and cost-isolated impact.
McKinsey & Company
What’s the future of generative AI? An early view in 15 charts 17
Designed by McKinsey Global Publishing
Copyright © 2021 McKinsey & Company. All rights reserved.
Find more content like this on the
McKinsey Insights App
Scan • Download • Personalize
Designed by McKinsey Global Publishing
Copyright © 2023 McKinsey & Company. All rights reserved. |
255 | mckinsey | how-ai-is-transforming-strategy-development_final.pdf | Strategy & Corporate Finance Practice
How AI is transforming
strategy development
Artificial intelligence is set to revolutionize strategy activities. But as AI
adoption spreads, strategists will need proprietary data, creativity, and new
skills to develop unique options.
This article is a collaborative effort by Alexander D’Amico, Bruce Delteil, and Eric Hazan, with Andrea Tricoli
and Antoine Montard, representing views from McKinsey’s Strategy & Corporate Finance Practice.
February 2025
At its core, strategy entails deriving insights from facts and data, developing real options based
on those insights, making hard-to-reverse choices, and executing initiatives that convert those
choices into value. Data analytics has assisted in this work for several decades, but never before
has technology been able to not only augment and partially automate inputs into strategy but
also combine them into complex analyses. In time, it may even recommend viable strategies.
Artificial intelligence and generative AI have the potential to transform how strategists work
by strengthening and accelerating activities such as analysis and insight generation while
mitigating challenges posed by human biases and the social side of strategy. Building on the
recent explosion in data and earlier AI advances that produced dramatic improvements in
forecasting accuracy, the latest tools are making deriving insights much easier and cheaper.
The impact we are seeing in client organizations and in our own work as strategists leads us
to view this moment as a new inflection point in strategy design—potentially on par with the
creation of core strategic frameworks in the 1970s and ’80s.
While AI won’t change the need for leaders to demonstrate strategic courage by committing
to big moves, we expect that the technology will, in time, enhance every phase of strategy
development, from design through mobilization and execution. Today, the technology delivers
the greatest benefits in the design phase, by helping organizations assess their starting point in
the context of industry and market dynamics. They can use it to size potential markets, analyze
competitor moves, and estimate the value of different strategic initiatives across multiple
scenarios. But that’s just the beginning: Strategy requires mobilizing the organization, ensuring
the right allocation of resources, and monitoring execution. In all these tasks, AI can play a role.
Emerging roles for AI in strategy
Human judgment remains essential to crafting the strategic vision, which combines the
organization’s ambition with a view of how to realize it. However, AI can accelerate and bring
greater rigor to the work of strategy teams. Even in these early days, we see five roles for AI:
researcher, interpreter, thought partner, simulator, and communicator. Each of these roles can
come into play at various steps across the different phases of strategy development (table):
We expect that AI will, in time,
enhance every phase of strategy
development, from design through
mobilization and execution.
How AI is transforming strategy development 2
Table
AI can assist strategists in all stages of the strategy development process.
Illustrative use cases
Core use case Supplementary use case
Thought
Researcher Interpreter partner Simulator Communicator
Design the Align on the strategic
strategy challenge
Assess through multiple
lenses
Explore value-creating
big moves
Commit to a bold
strategy
Mobilize the Empower and engage
organization
Translate strategy into
concrete initiatives
Prioritize and reallocate
resources
Govern and rewire
plans and budgets
Execute, Force execution
monitor, momentum
and review
Drive and support
performance
Test assumptions
and adapt
Launch the next S-curve
— Researcher. Strategists spend significant time gathering and enriching data from numerous
sources. AI’s ability to summarize and create meaningful connections across all data sets
can significantly enhance these efforts. For example, an AI-powered engine that identifies
potential M&A targets can pinpoint under-the-radar assets that fit a company’s strategic
thesis, enhancing what today is often a serendipitous process relying on executives’ and their
intermediaries’ market knowledge. One such tool can scan public information on more than
40 million companies across various languages and create a short list of relevant targets in
How AI is transforming strategy development 3
minutes. While AI is more thorough and faster than humans, strategists still need to pose the
right questions to generate the distinctive insights they seek.
— Interpreter. To turn data analytics into useful insights, strategists need to interpret how
the findings can advance their goals. For example, a search for growth opportunities often
entails looking into adjacencies. Those expansion ideas can come from many places, such
as reviews of competitors’ moves or a deep understanding of customers’ emerging needs. AI
tools can facilitate this discovery process by converting data from a disparate set of inputs—
such as annual reports, patents, customer reviews, and purchasing data—into “growth scans.”
These scans summarize the most frequently pursued adjacencies and then interpret and
score their fit with the company’s strategy. The resulting perspective can help strategists
narrow down options, find precedents or benchmarks for actions under consideration, and
uncover fresh ideas.
Another area where AI is already acting as an interpreter is trend monitoring. Strategists
need to keep tabs on changes in major trends when developing their options and reviewing
their assumptions. A gen-AI-powered engine can read massive amounts of information
and disaggregate trends into their component patterns and then interpret whether those
patterns suggest a trend is accelerating, maturing, or subsiding. For example, an organization
seeking to understand the demand for sustainable building materials can monitor interest
from architects, patent volumes, and competitors’ mentions long before those signals
translate into sales volumes.
— Thought partner. AI can also serve as a brainstorming partner, speeding up idea generation
and countering business leaders’ potential biases or blind spots. Gen AI in particular
can help strategists avoid common pitfalls by assessing their plans against established
frameworks. For example, a team can pressure test a strategy—both before and during its
execution—by leveraging gen AI to play a challenger role to highlight potential hidden pitfalls
or management blind spots.
— Simulator. Before committing to a strategic course, strategists consider the impact of
multiple market scenarios based on macroeconomic conditions, potential competitor moves,
and stakeholder reactions, among other factors. AI can make this scenario analysis much
more rigorous through advanced modeling capabilities and tactical game and simulation
applications. This capability can also be valuable during the strategy’s execution, with AI
monitoring early signals from the market, simulating their impact, and alerting the team when
it might be prudent to change course.
— Communicator. A clear narrative of the strategic path and objective and their implications
for the organization and its stakeholders is essential to mobilizing action. Gen AI’s ability to
summarize concepts in different formats has been among the technology’s most popular
applications since ChatGPT was launched. Strategists can use gen AI tools to make their
narratives more compelling to different audiences with different levels of expertise (such
as regional markets, regulators, or analysts) and in different formats (briefs, talking points,
or, most recently, podcasts1). AI can also monitor whether external communications are
consistent across different channels.
1 Google Labs, “NotebookLM now lets you listen to a conversation about your sources,” blog entry by Biao Wang,
September 11, 2024.
How AI is transforming strategy development 4
To see how these five applications can work in practice, consider the case of a Southeast Asian
regional bank that wanted to expand to a new segment or geography. The strategy team used
its AI model to analyze the business context and promising trends in the industry and region.
The tool generated interactive reports that allowed the strategists to fine-tune their follow-up
research. Based on this work, the strategy team decided to focus on opportunities in the digital
financial ecosystem (particularly peer-to-peer payments) and microcredit.
Next, the team asked AI to provide recommendations on the most promising adjacencies for
growth investments. Based on an analysis of information from banks around the world, the
tool created a graph of close and synergistic business segments. Management prioritized a
few for deeper analysis—for example, a cross-border digital offering across the region or the
microcredit segment in Vietnam—and built hypotheses on their potential growth trajectories.
To learn more about each segment, they asked AI, “Who are my competitors in each market,
and what are their value propositions?” Some of the markets were unfamiliar to the leaders,
so the strategy team posed questions such as, “We are considering entering the Vietnamese
banking market. What are the risks that have emerged in the past? Are there examples of failed
attempts (with sources)?”
The team also considered inorganic options such as partnerships and M&A. Based on an AI scan,
they short-listed a few small and medium-size businesses with the technology the company
needed to support its digital ambition. Gen AI also helped them build initial due diligence profiles
to support potential outreach.
Finally, as hypotheses solidified into concrete strategic options, AI helped the strategists
simulate the resulting P&L and growth projections. Additionally, the tool utilized internal data,
such as management reports on the bank’s earlier expansion into another country, to help
management understand the strengths and weaknesses of their execution capabilities.
Numerous organizations have started building tools to make such scenarios a reality, with some
developing proprietary AI agents to simulate reasoning or perform complex research tasks.
However, even those earlier in their AI journey can start exploring some of the roles that AI can
play. As technology advances, strategists who build the skills to develop unique applications for
AI models will gain a critical insights edge over competitors.
Considerations for strategy leaders deploying AI
While the journey of the Southeast Asian bank is compelling, strategists should be mindful of
several challenges in deploying AI. Generative AI presents well-documented risks, ranging from
model bias (historical training data can lead AI to overemphasize certain types of customers,
for example) to reduced explainability (failure to offer a logical foundation for the analysis) to
hallucinations (constructing credible-sounding but false content). The good news is that each of
these pitfalls is being addressed. For example, AI can help police itself: A “critic agent” can check
the work done by other AI applications and flag when the content might be incorrect or directly
instruct a reworking of the task in question.
Beyond these well-understood risks, gen AI presents five additional considerations for
strategists. First, it elevates the importance of access to proprietary data. Gen AI is accelerating
a long-term trend: the democratization of insights. It has never been easier to leverage off-
the-shelf tools to rapidly generate insights that are the building blocks of any strategy. As the
adoption of AI models spreads, so do the consequences of relying on commoditized insights.
How AI is transforming strategy development 5
After all, companies that use generic inputs will produce generic outputs, which lead to generic
strategies that, almost by definition, lead to generic performance or worse. As a result, the
importance of curating proprietary data ecosystems (more on these below) that incorporate
quantitative and qualitative inputs will only increase.
Second, the proliferation of data and insights elevates the importance of separating signal from
noise. This has long been a challenge, but gen AI compounds it. We believe that as the technology
matures, it will be able to effectively pull out the signals that matter, but it is not there yet.
Third, as the ease of insight generation grows, so does the value of executive-level synthesis.
Business leaders—particularly those charged with making strategic decisions—cannot operate
effectively if they are buried in data, even if that data is nothing but signal. As with gen AI’s
growing ability to separate signal from noise, the technology is getting better at synthesis, but
in the near term, strategy leaders need to own that task.
Fourth, AI reinforces the importance of the processes that organizations follow to develop their
strategies. Our research shows that the quality of the process is far more important to strategies’
success than the quality of insights. High-quality processes include, but are not limited to, the
development and examination of strategic alternatives, properly accounting for uncertainty,
pushing to make bold commitments, and, most importantly, taking steps to remove bias from
decisions. Fortunately, as gen AI speeds up the development of insights, it leaves more time for
strategy teams to hone best-in-class processes.
Finally, to successfully leverage gen AI, the strategy function needs to invest in technology
for creating and accessing ecosystems of proprietary data sources. The ecosystem approach
removes the need for companies to internally generate or own the full gamut of proprietary data.
Instead, they build networks of sources that they can seamlessly tap into using technology. In
addition, strategists will need to identify (and often customize) gen AI tools that can effectively
serve as researchers, simulators, interpreters, thought partners, and communicators.
Moving forward
So where do you begin? We recommend three near-term steps:
— Get smart. The strategist of tomorrow needs to understand how AI works. How does a
word prediction engine manipulate complex concepts and information? How are insights
generated from the information included in models and prompts? Those who gain this
expertise will be able to contribute to creating the tools their work requires, such as running
complex simulations on how markets and competitive landscapes will evolve. Individuals with
such skills will be highly sought after, making their retention a management priority.
— Start building today. AI is here to stay, and finding the right way to apply it to strategy
development is essential. Strategy teams should familiarize themselves with the possibilities
AI offers, from helping in their research and insight generation to identifying potential risks.
Teams that explore how the available tools can assist in these tasks will better understand
what other tools they will need to build or invest in to meet their specific needs. From an
organizational perspective, leaders need to help strategy teams gain access to expertise in
data science, data engineering, and large language models. This can be done by embedding
technology experts into strategy teams or by providing strategists access to them through
centers of excellence.
How AI is transforming strategy development 6
— Develop your proprietary insights ecosystem. Even with state-of-the-art capabilities, AI
Find more content like this on the
models will be limited to interpreting existing data—they cannot generate new signals.
McKinsey Insights App
For example, AI won’t replace the insights from ethnographic research or the direct input
from customers. Indeed, such proprietary information will become even more critical to
generating unique insights as external data grows more affordable and accessible to all
market participants. To gain an edge, strategists will need to expand their exposure to
different domains by connecting with innovators and stakeholders within and outside their
organizations. Strategists’ core focus will increasingly become developing hypotheses,
testing and learning from them, and maintaining the AI and data infrastructure that enable
the conversion of insights into a competitive advantage.
Scan • Download • Personalize
Artificial intelligence can’t—and, we believe, won’t—replace human logic and interpretation in a
complex domain, such as strategy. However, the technology can provide faster, more objective
answers that can significantly augment our decision prowess. Through the various roles AI can
already play, from researcher to thought partner to simulator, we are starting to see how these
tools may, in time, redefine strategists’ roles and help companies make strategic decisions. By
making the strategy development process more efficient while allowing the space for creativity
and breakthrough ideas that help leaders define the consequent bold moves, AI can deliver the
competitive edge needed to beat the market.
Alexander D’Amico is a senior partner in McKinsey’s Connecticut office, Bruce Delteil is a partner in the Hanoi
office, Eric Hazan is a senior partner emeritus in the Paris office, Andrea Tricoli is an associate partner in the
London office, and Antoine Montard is a client capabilities director in the Lisbon office.
This article was edited by Joanna Pachner, an executive editor in the Toronto office.
Designed by McKinsey Global Publishing
Copyright © 2025 McKinsey & Company. All rights reserved.
How AI is transforming strategy development 7 |
258 | mckinsey | capturing-the-generative-ai-opportunity-for-the-dutch-labor-market.pdf | Capturing the generative
AI opportunity for the
Dutch labor market
Dutch businesses can embrace generative AI to speed up automation,
increase productivity, and ease labor market tightness. What is needed
to accelerate adoption and capture the benefits?
This article is a collaborative effort by Ashley van Heteren, Eva Beekman, and Ferry Grijpink, with Just van
der Wolf and Wouter Kokx, representing views from McKinsey’s Amsterdam office.
November 2024
The Dutch labor market is strong and evolving: work in Europe, we estimate that automation will likely
labor force participation1 is notably high, and affect 15 percent of total full-time-equivalent work hours
unemployment levels are historically low.2 But in the Netherlands by 2030—a 50 percent increase over
ongoing trends, including an aging population a scenario without gen AI.6 This represents a significant
and declining productivity growth, are putting the opportunity for Dutch businesses to tackle the
labor market under pressure.3 Recently, McKinsey challenges posed by the evolving and tight labor market,
projected that labor market tightness could triple by especially in areas in which increased automation
2030 if the Netherlands maintains current levels of can help relieve labor shortages. However, a wide
growth in GDP (1.6 percent CAGR) and productivity deployment of gen AI is expected to require extensive
(0.4 percent CAGR).4 re- and upskilling programs, including for jobs that were
previously unaffected by automation. It may also require
While traditional automation solutions to increase regulation7 to mitigate its potential risks,8 such as data
labor productivity, such as document processing privacy, intellectual property risk, and fairness.
systems, have played a significant role in addressing
this challenge, most have been limited to processing In this article, we focus on how gen AI can
structured data. Generative AI (gen AI), however, influence the future of work in the Netherlands
unlocks a new domain of automation with its ability to across sectors and organizations. We discuss how
process unstructured data such as natural language expected adoption speed varies by sector and by
and images.5 It therefore broadens the spectrum of sector composition in terms of shares of small and
automation potential to more occupations, including medium-size enterprises (SMEs), independents,
knowledge work and customer service, and holds the and corporations. We also explore actions public
potential to boost productivity and economic growth and private leaders could take to accelerate gen AI
for the Netherlands. adoption in the Netherlands. This perspective takes a
longer-term view of the singular effects that might be
Following the base case scenario of automation of directly brought about by gen AI; it does not forecast
current activities (“slower scenario”) from a recent aggregated employment effects that can be brought
McKinsey Global Institute (MGI) report on the future of about by the business cycle in the short term.
1 Share of employed people to the total population aged 20 to 64.
2 “Tension in the labor market,” Centraal Bureau voor de Statistiek (Statistics Netherlands), accessed August 18, 2024.
3 “Dashboard spanningsindicator” (“Dashboard voltage indicator”), UWV, accessed October 25, 2024.
4 Netherlands advanced: Building a future labor market that works, McKinsey, June 18, 2024.
5 “A new future of work: The race to deploy AI and raise skills in Europe and beyond,” McKinsey Global Institute (MGI), May 21, 2024.
6 Ibid.
7 “As gen AI advances, regulators—and risk functions—rush to keep pace,” McKinsey, December 21, 2023.
8 “Implementing generative AI with speed and safety,” McKinsey Quarterly, March 13, 2024.
Methodology
The following data used in this article is as a third “slower” scenario, the 25 economic potential of generative AI: The
directly sourced from a 2024 McKinsey percent point between the early and next productivity frontier and Generative AI
report, Netherlands advanced: Building a late scenarios and the future of work in America—as well as
future labor market that works1: a 2024 report, A new future of work: The race
— impact of automation on productivity to deploy AI and raise skills in Europe and
— employment projections
beyond. A full description of the methodolo-
Both this article and the Netherlands ad- gy and data used is included in the technical
— scenarios for automation adoption;
vanced report draw on the methodology and appendixes of those reports.2
two scenarios to bookend the model,
findings of two 2023 McKinsey reports—The
the “late” and “early” scenarios as well
1 Netherlands advanced: Building a future labor market that works, McKinsey, June 18, 2024.
2 The economic potential of generative AI: The next productivity frontier, McKinsey, June 14, 2023; “Generative AI and the future of work in America,” MGI, July 26, 2023; “A
new future of work: The race to deploy AI and raise skills in Europe and beyond,” MGI, May 21, 2024.
Capturing the generative AI opportunity for the Dutch labor market 2
Gen AI creates new opportunities sidebar, “Methodology”). Of course, the exact
spanning sectors and organizations automation path for gen AI is subject to considerable
uncertainty. The extent to which this potential is
The introduction of gen AI and the public
realized will depend on the ability of Dutch companies
breakthrough of OpenAI’s ChatGPT in late 2022 have
to innovate, the capability of workers to re- and
accelerated the automation potential of activities that
upskill, and the support of policy makers.
involve communication, documentation, or interaction
with people. Gen AI—in particular, applications that
In the rest of this section, we dig into the precise
are open or that use publicly available data—has also
factors affecting uptake as well as the sectors in
become available to a wider group of organizations.
which gen AI has the most promise for addressing
The range of work activities suitable for automation
labor market challenges.
has expanded to include those requiring subject
matter expertise, interpersonal interaction, and
Impact from gen AI will depend
creativity. Consequently, the timeline for automation
on sector composition
adoption could accelerate significantly, reaching
While our models reveal potential for gen AI to
previously unaffected jobs, such as those in
relieve labor market tensions, we also acknowledge
professional services.
the challenges involved in gen AI reaching its full
potential in the Netherlands.
Our modeling suggests that even in the “slower
scenario,” adoption of gen AI could reduce the
Not all Dutch businesses are preparing to adopt
total number of hours needed to perform current
analytical AI (that is, AI methods preceding generative
workforce activities by 15 percent (Exhibit 1; see
AI). AWVN (General Employers Association of the
Web <2024>
E<MxChKib24it9 1087_Gen AI impact on NL labor>
Exhibit <1> of <4>
Generative AI is expected to accelerate automation adoption in 2030 by 50
percent in all scenarios.
Automation of current work activities, % of full-time-equivalent hours expected to be automated
2030 automation potential, Netherlands, %
Description1 Without generative AI With generative AI
Fastest pace of automation
Early scenario development and adoption 35 50
25% point between early
Slower scenario and late scenarios 10 15
Slowest pace of automation
Late scenario development and adoption 2 3
Note: The range of scenarios represents uncertainty regarding the availability of technical capabilities, based on interviews with experts and survey responses.
1Scenarios for productivity growth in the Netherlands based on “A new future of work: The race to deploy AI and raise skills in Europe and beyond," McKinsey
Global Institute (MGI), May 21, 2024.
Source: MGI analysis
McKinsey & Company
Capturing the generative AI opportunity for the Dutch labor market 3
Netherlands) reported that 40 percent of Dutch and 46 percent in the United States.10 Sectors
companies are not yet using AI in their businesses dominated by small companies might be slower
because of a lack of knowledge, safety and privacy to embrace new automation opportunities. For
concerns, or perceived irrelevance.9 This finding example, when it came to adoption of digital sales
illustrates the types of challenges the Netherlands technologies, in 2019, the top 10 percent of largest
may face in rapidly adopting gen AI technologies. companies captured 60 to 95 percent of digital
revenues.11 This is expected to affect the rate and
The speed of gen AI adoption at a country level speed at which gen AI might accelerate automation in
is determined by multiple factors, including the the Netherlands.
economic maturity of a country, overall sector
readiness to embrace new technologies, and Large companies in the Netherlands are adopting
crucially, sector composition in terms of shares of analytical AI technologies faster than smaller
SMEs, independents, and corporations (Exhibit companies. Centraal Bureau voor de Statistiek (CBS)
2). A relatively high percentage of workers in the reported in 2020 that 48 percent of companies with
Netherlands (about 65 percent) are employed by 500 or more employees were using one or more AI
SMEs, compared with 57 percent in Germany, 54 technologies, compared with only 8 to 13 percent for
percent in the United Kingdom, 52 percent in France, companies with ten to 50 employees.12
9 Sandra Olsthoorn, “Werkgevers nog huiverig voor inzet AI” (“Employers are still hesitant about using AI”), Het Financieele Dagblad, June 10, 2024.
10 Statista defines SMEs as companies with fewer than 250 employees; US County Business Pattern 2021 defines SMEs as companies with
fewer than 500 employees; “Enterprises by business size,” OECD, accessed August 18, 2024.
11 F or more, see Jacques Bughin, Tanguy Catlin, and James Manyika, “Twenty-five years of digitization: Ten insights into how to play it right,” MGI,
May 21, 2019.
12 “ICT use in companies; company size, 2020,” CBS, updated April 22, 2022.
Web <2024>
<MCK249087_Gen AI impact on NL labor>
Exhibit 2
Exhibit <2> of <4>
Small companies are expected to adopt automation more slowly than larger
organizations.
Potential High Medium Low
Existing Expected
Up- and
Investment availability automation
reskilling
Archetypes Workers in Netherlands, million capacity + + of data and = adoption by
capabilities
infrastructure 2030
CCoorrppoorraattiioonnss 3.3 Corporations
((>>550000 ppeerrssoonnss)) (>500 persons)
LLaarrggeerr SSMMEEss1¹ Larger SMEs1
1.6 (101–500
((110011––550000 ppeerrssoonnss))
persons)
Small SMEs1
SSmmaallll SSMMEEss1¹
2.3 (2–100
((22––110000 ppeerrssoonnss))
persons)
IInnddeeppeennddeennttss Independents (1
1.2
((11 ppeerrssoonn)) person)
PPuubblliicc sseeccttoorr 0.6 Public sector
Note: Intensity based on qualitative scoring criteria.
¹Small and medium-size enterprises.
Source: Centraal Bureau voor Statistiek (CBS); McKinsey analysis
McKinsey & Company
Capturing the generative AI opportunity for the Dutch labor market 4
As with the adoption of analytical AI, the smaller- Challenges to automation adoption are not equal
company archetype could capture impact from gen for all sectors in the Netherlands, because the
AI more slowly in the next five years for the following share of SMEs varies widely across sectors (Exhibit
three reasons: 3). For example, the agriculture sector could
potentially benefit from using analytical AI and gen
Limited investment capacity. Smaller companies usually AI to improve efficiency for farms and agriculture
have lower investment capacity, constraining their companies, such as by improving on-farm decision
ability to acquire new capabilities, tools, or resources. making with camera images.15 However, the high
For example, Dutch SMEs invest about 1 percent of percentage of independent and smaller companies
profits in R&D compared with about 5 percent for larger in the sector along with factors such as plot size or
companies.13 They might lack the scale to manage these specific legislation has thus far led to lower adoption
solutions systematically, such as keeping marketing of automation. In fact, McKinsey research shows
content generator inputs up to date. However, many that only 33 percent of Dutch farmers, who farm on
gen AI solutions—especially those integrated in existing smaller plots than farmers in neighboring countries,
software packages such as Adobe and Microsoft use at least one agricultural technology, compared
Office—are already available for many SMEs. with 45 percent in Germany and 51 percent in France.
Lower up- and reskilling capabilities. Introducing Additionally, we expect that the public sector will
gen AI tools into existing operating models requires experience slower adoption of gen AI. While most
significant change management of IT-related subsectors in the public sector have a high share of
processes, including upskilling and reskilling enterprises with more than 500 employees (about
programs to help employees use gen AI tools 50 to 80 percent in government, healthcare, and
effectively. Smaller companies often lack the scale to education) and may have more scale to enable gen AI
benefit from designing and running such programs upskilling, public acceptance and different regulatory
and are half as likely to provide formal upskilling requirements can potentially slow the adoption of
programs compared with large corporates.14 gen AI in this domain. McKinsey research also shows
Rather than formal in-person training programs, that the overall investment in analytical and gen AI is
these organizations might rely more on digital and lower in public sectors than in other sectors.16
self-organized training, which could provide a less
effective learning environment.
Gen AI holds the greatest potential
Less robust existing data and infrastructure. The to address Dutch labor market
low-complexity technological landscapes of challenges in five sectors
smaller companies typically do not justify extensive Although various interventions can address labor
investment. Consequently, the technology is generally market challenges, our models show that gen AI
less mature, and existing data and infrastructure are can be a major productivity booster, particularly in a
generally less available. SMEs may, for example, use handful of sectors (Exhibit 4). A previous McKinsey
customer databases manually and not directly link report, Netherlands advanced: Building a future labor
databases to marketing and sales systems. All of market that works,17 estimated that roughly one-
this limits the pace at which gen AI solutions can be third of the necessary productivity improvements to
integrated with existing systems and data. address labor market tightness in the Netherlands can
be achieved through automation powered by gen AI.
13 R&D expenses versus profits for companies with fewer than 250 employees and for companies with more than 250 employees. See “Bedrijven;
arbeid, financiele gegevens, bedrijfsgrootte, bedrijfstak” (“Companies; employment, financials, company size, industry”), CBS, October 10, 2024;
“Research en development; personeel, uitgaven, bedrijfsgrootte, bedrijfstak” (“Research and development; personnel, expenditure, company
size, industry”), CBS, August 30, 2024.
14 For more, see “A microscope on small businesses: Spotting opportunities to boost productivity,” McKinsey Global Institute, May 2, 2024.
15 “From bytes to bushels: How gen AI can shape the future of agriculture,” McKinsey, June 20, 2024.
16 “The state of AI in early 2024: Gen AI adoption spikes and starts to generate value,” McKinsey, May 30, 2024.
17 Netherlands advanced: Building a future labor market that works, McKinsey, June 18, 2024.
Capturing the generative AI opportunity for the Dutch labor market 5
Web <2024>
<MCK249087_Gen AI impact on NL labor>
Exhibit 3
Exhibit <3> of <4>
Expected automation adoption varies across sectors based on business
archetypes.
Employment per business archetype, %
Expected automation
adoption
Business archetypes:
due to distribution of
Independents Small SMEs1 Larger SMEs1 Corporations ⚫ Public sector business archetypes
Real estate and rental
7 21 19 53
and leasing
Transportation and 10 25 21 44 Medium to(cid:22)
warehousing
high
Finance and insurance 8 40 12 40
Medium-to-high share of
Wholesale trade 6 43 17 33
large corporations
expected to positively
Manufacturing 5 34 32 29 contribute to automation
adoption
Information 23 28 23 26
Professional, scientific,
43 20 17 20
and technical services
Construction 31 29 19 20
Arts, entertainment, and
57 13 17 13
recreation
Other services 46 27 15 13 Low
Accommodation and food
7 74 10 10
services Public sector affiliation or
low share of large
Agriculture, forestry,
fishing, and hunting 52 32 10 5 corporations expected to
negatively contribute to
⚫ Administrative and
1 17 82 automation adoption
support and government
⚫ Healthcare and social
9 10 14 67
assistance
⚫ Educational services 13 1 32 54
⚫ Utilities 3 26 34 37
Note: Figures may not sum to 100%, because of rounding.
1Small and medium-size enterprises.
Source: Centraal Bureau voor de Statistiek (CBS); McKinsey analysis
McKinsey & Company
Capturing the generative AI opportunity for the Dutch labor market 6
Web <2024>
<MCK249087_Gen AI impact on NL labor>
Exhibit <4> of <4>
Exhibit 4
Generative AI can boost productivity across sectors in the Netherlands.
Expected automation adoption by 2030¹ Jobs, 2022
Medium to high; less orchestration required 1 million
Low; more orchestration required
100,000
Labor market tension per sector, Q4 2023
14
Below average Above average
12 Finance and
insurance
10
Very tight
8
Manufacturing IT
Health care and
social assistance Professional, scientific,
Accommodation and and technical services
6 food services
Wholesale Construction
trade Educational services
4 Transportation
and warehousing Administrative
and support and
Tight
Agriculture, forestry, government
fishing and hunting
2
Not tight
0
0 2 4 6 8 10
Automation accelerated with generative AI,² % of current FTE³ hours in 2030
1Based on business archetype and share of employment in large organizations.
2Slower scenario: the 25% point between the early and late “bookend” scenarios, assuming 10% automation by 2030 without generative AI.
3Full-time equivalent.
Source: Centraal Bureau voor de Statistiek (CBS); UWV; McKinsey analysis
McKinsey & Company
Capturing the generative AI opportunity for the Dutch labor market 7
While gen AI is a game changer in speeding up benefit significantly from gen AI tools, particularly
automation for some sectors, implementing this in automating daily tasks, and its high level of
technology may face delays due to sector nature digitalization can facilitate the integration of gen AI
and composition. Sectors may require varying levels into primary processes and technology development.
of innovation orchestration and support to see For example, ING partnered with McKinsey to
positive results. develop and deploy a gen AI–powered chatbot.19
Gen AI can also speed up credit and underwriting
This orchestration could include four elements. applications by automating document and image
The first, raising awareness, could involve helping processing and generating credit memos.
sectors understand relevant high-impact gen AI
use cases and their feasibility. Next, technology The IT sector can benefit greatly from gen AI’s ability
such as cost-effective, scalable, and secure gen AI to automate and accelerate software development
solutions should be available, along with support processes.20 Gen AI–powered coding assistance,
for AI literacy training. Third, support to establish automated testing, and intelligent architecture-
pilots and measure impact on labor tightness could design recommendations help IT companies speed
serve as proof points for broader adoption and up feature delivery and reduce time to market. AFAS,
scaling. Last, cross-sector partnerships could be for example, has developed an API integration with
established to scale impact, pool funding, and share OpenAI, giving companies access to advanced AI
knowledge—for instance, across public sector, models for generating business information.21 This
industry, and educational institutions. Below, we helps automate tasks such as reporting, business
explore five sectors with high potential for gen data analysis, and document creation. Gen AI can
AI automation impact: finance and insurance; help minimize incident and problem resolution times
IT; administrative and support and government; by creating tighter feedback loops, linking user
educational services; and professional, scientific, behaviors to feature changes, and collecting real-
and technical services. time feedback. Delft University of Technology’s AI for
Software Engineering Lab (AI4SE) explores these
Two sectors likely to self-propel opportunities.22
adoption of gen AI
The finance and insurance sector and the IT sector Orchestration can accelerate gen AI
have the highest potential for gen AI automation adoption in three additional sectors
and some of the highest labor market tension. Gen Our models suggest that most sectors would benefit
AI innovation in these sectors will enhance their from orchestration to accelerate gen AI adoption,
productivity.18 This, combined with their higher share including the public sector and sectors with a higher
of large corporations and large SMEs (approximately share of smaller companies. Here we explore three
50 percent of all businesses), provides business sectors that will likely require orchestration for
incentives and scale to encourage adoption without gen AI adoption and where the potential for gen AI
significant orchestration. impact is highest—including the administrative and
support and government sector, the educational
The finance and insurance sector is expected services sector, and the professional, scientific, and
to adopt gen AI because of strong commercial technical services sector.
incentives to substantially automate operations
and address labor shortages. This sector stands to In the administrative and support and government
sector, many traditional tasks are highly suited for
18 Our insights are based solely on sector composition; other factors may influence the actual potential for gen AI adoption.
19 “Banking on innovation: How ING uses generative AI to put people first,” McKinsey, accessed October 25, 2024.
20 “Unleashing developer productivity with generative AI,” McKinsey, June 27, 2023.
21 “AFAS & AI: hoe gaan we bij AFAS om met AI?” (“AFAS and AI: How do we deal with AI at AFAS?”), AFAS, accessed November 26, 2024.
22 “TU Delft and JetBrains are launching new ICAI lab AI for Software Engineering,” TU Delft, October 12, 2023.
Capturing the generative AI opportunity for the Dutch labor market 8
As gen AI becomes increasingly
integrated into various sectors, the
demand for specialized skills and
expertise in AI is expected to grow.
automation using gen AI, such as case handling Gen AI will create new roles and
and call center operation. In educational services, occupational categories
early gen AI applications already focus on adaptive As gen AI becomes increasingly integrated into
tutoring and service chatbots. These tools can various sectors, the demand for specialized
enhance learning experiences and alleviate labor skills and expertise in AI is expected to grow. For
shortages by supplementing teacher-led instruction. example, we expect increased demand in three
However, both of these subsectors are in the public occupational categories in the Dutch labor market:
sector and are expected to adopt gen AI more slowly,
as discussed above. In some cases, however, the Gen AI practitioners. Gen AI specialists—including
government has played a leading role in accelerating prompt and agent engineers or AI content
adoption to enhance service delivery and support. auditors—form a new subexpertise within the AI
For example, in the administrative and support and playing field. Globally, these roles have grown
government sector, Rijksdienst voor Ondernemend rapidly24 across sectors that implement gen AI
Nederland (RVO) launched AskSenna,23 an AI-driven in their daily practice, especially in IT functions.
tool designed to assist start-ups and early-stage Additionally, the surge in gen AI will drive demand
companies by providing instant answers to regulatory for related software and data engineering support.
and business-related queries. For example, the Dutch company Weaviate helps
companies structure their data to facilitate the
In the professional, scientific, and technical services development of gen AI use cases.25
sector, including consulting, solutions are being
developed to use gen AI for advanced search, Gen AI researchers. Gen AI has created new
synthesis tasks, and virtual coaching. However, given opportunities in research positions, within both
the high percentage of independents and SMEs in academia and enterprises. For instance, Philips
the sector (80 percent), we expect slower holistic is developing gen AI applications to improve
adoption of gen AI solutions and therefore a slower clinical decisions, diagnosis, and workflow.26
impact on labor tightness. This sector may benefit The Dutch start-up Cradle uses gen AI to predict
from orchestration that is particularly targeted to protein properties that could accelerate vaccine
help smaller enterprises understand the specific development.27
applications of gen AI relevant to their business,
learn approaches to successfully implement those Semiconductor, software, and other engineers. As
applications, and improve AI literacy. gen AI continues to grow, semiconductor-related
23 For more, see the AskSenna website.
24 “Generative AI demand soars 1,800% for US employers,” Lightcast, October 19, 2023; Ted Liu and Kelly Monahan, “2024 in-demand skills:
Unprecedented growth in AI and emergent skills for uniquely human work,” Upwork, March 19, 2024.
25 For more, see the Weaviate website.
26 Vidya Sagar, “Philips partners with AWS to develop generative AI applications,” NS Medical Devices, April 18, 2023.
27 For more, see the Cradle website.
28 “Mogelijke uitbreiding ASML op Brainport Industries Campus in Eindhoven” (“Possible expansion of ASML at Brainport Industries Campus in
Eindhoven”), Eindhoven, April 22, 2024; 2023 annual report, ASML, February 14, 2024.
Capturing the generative AI opportunity for the Dutch labor market 9
jobs that enable this technology will likewise expand, sector could facilitate this process—for example, as
including semiconductor engineers to provide Techniek Nederland has been doing since 2013 by
computing power, software engineers to build front- reskilling individuals from various backgrounds to
end solutions, and a wide range of other roles. This become installation engineers.30
presents an opportunity for the Dutch semiconductor
industry, and companies such as ASML, ASM, Besi, Granular insights such as regional job gain and loss
and NXP Semiconductors are positioned to grow analysis are crucial to understand a company’s
substantially because of the expected increase in reskilling needs and make informed decisions.
semiconductor demand driven by the growth of end To assist Dutch businesses in identifying their
applications including gen AI. For example, in April automation potential and its workforce impact,
2024, ASML and the local municipality Eindhoven public sector agencies and businesses could
signed a letter of intent to expand ASML’s facilities to consider developing tools that map different job
accommodate an additional 20,000 employees—a types to their expected automation potential.
50 percent increase in growth, which could be partly Such tools could provide businesses and local
driven by gen AI.28 governments with valuable insights and guidance
for transitioning to gen AI–driven processes,
We have previously emphasized the importance helping them make informed decisions and support
of developing soft and hard skills to keep pace employees throughout their journey.
and enable career advancement.29 This ongoing
development can enhance productivity in current Orchestrating sectors with high degrees
positions and prepare individuals for the high- of SMEs and labor tightness
demand jobs created by and for gen AI. In private sectors such as the professional,
scientific, and technical services sector, in which
both automation potential through gen AI and the
Three moves could accelerate proportion of SMEs are high, orchestration can
gen AI adoption and manage its accelerate gen AI implementation. Larger tech
effect on the Dutch workforce corporations, universities, public sector agencies
Three actions by public and private stakeholders could such as UWV, and sectoral employer organizations
accelerate gen AI adoption in the Netherlands and will could potentially facilitate this.
likely have a positive effect on the Dutch workforce.
Furthermore, Dutch companies of all sizes have
Preparing for granular upskilling opportunities to engage in the gen AI–based
and reskilling needs automation market. For example, banks could
Increased automation requires upskilling workers to collaborate with AI or software companies to
use new gen AI tools—monitoring service chatbots or create specific propositions or loans for gen
using copilots to write marketing content, for example. AI development and help disseminate these
Companies will need to develop training programs throughout the sector. The Nederlandse AI
as these solutions are implemented. And because Coalitie (NL AIC), a public–private partnership,
demand for some professions may decline, workers aims to promote the adoption and ethical use of AI
may need to reskill, sometimes across sectors. This technologies across sectors.31
would necessitate greater orchestration. The public
29 Netherlands advanced, June 18, 2024.
30 For more, see the Techniek Nederland, Techniekpact, and Mensen Maken de Transitie websites.
31 For more information, see the Nederlandse AI Coalitie website.
Capturing the generative AI opportunity for the Dutch labor market 10
Making bold investments to lead the education, and the Innovation Center for Artificial
Find more content like this on the
Netherlands’ gen AI transition Intelligence. In the private sector, many large
McKinsey Insights App
Strategic investments by both public and private corporations are investing significantly in developing
organizations could accelerate the adoption of gen innovative technology. For example, ASML and
AI and alleviate labor shortages in the Netherlands.32 Philips launched DeepTechXL, a private investment
Such funding for long-term innovation will be crucial, fund to finance and guide high-tech start-ups.34
especially in sectors that are strategically important
to the economy, such as manufacturing (including the
semiconductor industry), healthcare, construction,
and education. For example, NFI, SURF, and TNO By embracing collaboration on cutting-edge
Scan • Download • Personalize
have received €13.5 million to develop a Dutch gen technology opportunities such as gen AI, the
AI model that could accelerate development and Netherlands can position itself to build an
adoption across sectors.33 increasingly thriving business ecosystem. A proactive
approach can accelerate innovation and economic
The Netherlands already hosts a few AI funds, such growth as well as ensure the workforce is well
as NL AIC for responsible AI, AI growth fund AiNed, prepared to adapt to the changing landscape and
the Nationaal Onderwijslab AI (National Education flourish in a future shaped by AI.
Lab AI) established by Radboud University for AI in
32 “Time to place our bets: Europe’s AI opportunity,” MGI, October 1, 2024.
33 “The Netherlands starts construction of GPT-NL as its own AI language model,” TNO, November 2, 2023.
34 Heiko Jessayan, “DeepTechXL haalt €110 mln op door inleg van ASLM en pensioenfonds PME” (“DeepTechXL raises €110 million through
contributions from ASLM and pension fund PME”), Het Financieele Dagblad, March 14, 2024.
Ashley van Heteren, Eva Beekman, and Ferry Grijpink are partners in McKinsey’s Amsterdam office, where Just van der Wolf
is an AI expert and Wouter Kokx is an associate partner.
The authors wish to thank Alexander Veldhuijzen, Dieuwert Inia, Gurneet Singh Dandona, Hagar Heijmans, Joris van Niel,
Lex van der Vegt, Marc de Jong, Michael Chui, Reinout Goedvolk, and Sven Smit for their contributions to this article.
Copyright © 2024 McKinsey & Company. All rights reserved.
Capturing the generative AI opportunity for the Dutch labor market 11 |
259 | mckinsey | a-new-future-of-work-the-race-to-deploy-ai-and-raise-skills-in-europe-and-beyond.pdf | A
new
future
of
work:
The
race
to
deploy
AI
and
raise
skills
in
Europe
and
beyond
A new future of work:
The race to deploy
AI and raise skills in
Europe and beyond
Authors
Eric Hazan
Anu Madgavkar
Michael Chui
Sven Smit
Dana Maor
Gurneet Singh Dandona
Roland Huyghues-Despointes
May 2024
About the McKinsey
Global Institute
The McKinsey Global Institute was established in 1990. Our mission is to provide a fact
base to aid decision making on the economic and business issues most critical to the world’s
companies and policy leaders. We benefit from the full range of McKinsey’s regional, sectoral,
and functional knowledge, skills, and expertise, but editorial direction and decisions are solely
the responsibility of MGI directors and partners.
Our research is grouped into five major themes:
— Productivity and prosperity: Creating and harnessing the world’s assets most productively
— Resources of the world: Building, powering, and feeding the world sustainably
— Human potential: Maximizing and achieving the potential of human talent
— Global connections: Exploring how flows of goods, services, people, capital, and ideas
shape economies
— Technologies and markets of the future: Discussing the next big arenas of value and
competition
We aim for independent and fact-based research. None of our work is commissioned or paid
for by any business, government, or other institution; we share our results publicly free of
charge; and we are entirely funded by the partners of McKinsey. While we engage multiple
distinguished external advisers to contribute to our work, the analyses presented in our
publications are MGI’s alone, and any errors are our own.
You can find out more about MGI and our research at www.mckinsey.com/mgi.
MGI Directors MGI Partners
Sven Smit (chair) Michael Chui
Chris Bradley Mekala Krishnan
Kweilin Ellingrud Anu Madgavkar
Sylvain Johansson Jan Mischke
Olivia White Jeongmin Seong
Tilman Tacke
A new future of work: The race to deploy AI and raise skills in Europe and beyond ii
Contents
At a glance Spotlight: Manufacturing
3 40
Context: Labor shortages Spotlight: Healthcare
and a slowdown in 42
productivity growth
4 Implications for
the workforce
Potential for accelerated 44
work transitions ahead
10 Enhancing productivity and
human capital in a time of
The varied geography of technological ferment
labor market disruptions 52
22
Technical appendix
New skills for a new era 60
26
Acknowledgments
Spotlight: Wholesale and 65
retail trade
36
Spotlight: Financial services
38
A new future of work: The race to deploy AI and raise skills in Europe and beyond 1
A new future of work: The race to deploy AI and raise skills in Europe and beyond 2
At a glance
Amid tightening labor markets and a slowdown in productivity growth, Europe and the
United States face shifts in labor demand, spurred by AI and automation. Our updated
modeling of the future of work finds that demand for workers in STEM-related, healthcare,
and other high-skill professions would rise while demand for occupations such as office
workers, production workers, and customer service representatives would decline. By
2030, in a midpoint adoption scenario, up to 30 percent of current hours worked could be
automated, accelerated by generative AI. Efforts to achieve net-zero emissions, an aging
workforce, and growth in e-commerce as well as infrastructure and technology spending and
overall economic growth could also shift employment demand.
By 2030, Europe could require up to 12 million occupational transitions, double the
prepandemic pace. In the United States, required transitions could reach almost 12 million,
in line with the prepandemic norm. Both regions navigated even higher levels of labor market
shifts at the height of the COVID-19 period, suggesting that they can handle this scale of
future job transitions. The pace of occupational change is broadly similar among countries in
Europe, although the specific mix reflects their economic variations.
Businesses will need a major skills upgrade. Demand for technological and social and
emotional skills could rise as demand for physical and manual and higher cognitive skills
stabilizes. Surveyed executives in Europe and the United States expressed a need not just
for advanced IT and data analytics but also for critical thinking, creativity, and teaching and
training—skills they report as currently being in short supply. Companies plan to focus on
retraining workers, in addition to hiring or subcontracting, to meet skill needs.
Workers with lower wages face challenges of redeployment as demand reweights
toward occupations with higher wages in both Europe and the United States.
Occupations with lower wages are likely to see reductions in demand, and workers will need
to acquire new skills to transition to better-paying work. If that doesn’t happen, there is a risk
of a more polarized labor market, with more higher-wage jobs than workers and too many
workers for existing lower-wage jobs.
Choices made today could revive productivity growth while creating better societal
outcomes. Embracing the path of accelerated technology adoption with proactive worker
redeployment could help Europe achieve an annual productivity growth rate of up to
3 percent through 2030. However, slow adoption and slow redeployment would limit that to
0.3 percent, closer to today’s level of productivity growth in Western Europe. Slow worker
redeployment would leave millions unable to participate productively in the future of work.
A new future of work: The race to deploy AI and raise skills in Europe and beyond 3
1
Context: Labor shortages
and a slowdown in
productivity growth
This report focuses on labor markets in Europe and the United States, looking at the
next few years to 2030. Technology and other factors will spur changes in the pattern of
labor demand, but these expected shifts need to be taken in the context of deep-seated
labor market changes already under way. Our study focuses on nine major economies in the
European Union along with the United Kingdom (which we refer to collectively in this report as
“Europe”), in comparison with the United States.
Structural shifts in labor markets have been ongoing for decades, including the very long-
term decline in the share of employment in agriculture, industry, and mining in favor of
services (Exhibit 1). More recently, labor markets were buffeted by pandemic shocks that
propelled not only faster shifts in hiring needs and more job switching but also new employee
preferences such as hybrid work. While COVID-19 exacerbated labor market tightening,
Europe’s high employment rate, a rapidly aging population, and a steady fall in working
hours make continuing shortages of workers and skills a persistent challenge for the future.
The burning question that remains is this: to what extent can the forthcoming technological
disruption solve labor market challenges in Europe?
A new future of work: The race to deploy AI and raise skills in Europe and beyond 4
Web <2024>
E<MxhCKib2i4t2 1172 VivaTech 2024>
Exhibit <1> of <16>
Employment in Europe and the United States has shifted toward
service sectors.
Share of total employment by sector, Europe1 and US, 1850–2022, %
Europe
100
Construction
Transportation
Agriculture
80
Manufacturing
Mining
Utilities
Household work
60
Trade (retail and wholesale)
Professional services
40 Business and repair services
Telecommunications
Healthcare
20 Entertainment
Financial services
Education
Government
0
1850 1900 1950 2000 2022
US
100
Construction
Transportation
Agriculture
Manufacturing
80 Mining
Utilities
Household work2
Trade (retail and wholesale)
60
Professional services
40 Business and repair services
Telecommunications
Healthcare
20 Entertainment
Financial services
Education
0 Government
1850 1900 1950 2000 2022
1Includes Czech Republic, Denmark, France, Germany, Italy, Netherlands, Poland, Spain, Sweden, and United Kingdom.
2Increase from 1850 to 1860 in US primarily due to changes in how unpaid labor was tracked.
Source: Eurostat; Integrated Public Use Microdata Series USA, 2017; Ivan T. Berend, An Economic History of Twentieth-Century Europe, Cambridge University
Press, October 2016; US Bureau of Labor Statistics
McKinsey & Company
A new future of work: The race to deploy AI and raise skills in Europe and beyond 5
Europe’s future of work unfolds amid labor shortages
and a slowdown in productivity growth
In both Europe and the United States, labor market tightness has been on the rise, with
unfilled positions on the rise in both regions and unemployment at historically low levels.1 As
populations age on both sides of the Atlantic and the number of hours worked per worker
falls, particularly in Europe, labor market tightness is not likely to resolve naturally. In this
context, employers are increasingly competing for talent.
The pandemic had additional lasting impacts on workplaces, notably the increased adoption
of hybrid work. While about 90 percent of the working population was working fully on-site
in 2018, that number dropped to some 60 percent between 2021 and 2022. Since then,
the number has stabilized. However, only 40 percent of the 72 minutes saved daily from
not having to commute is allocated to work, with the rest mostly allocated to leisure and
caregiving.2 The overall impact on productivity is still being debated.3
Overall, in the global economy, productivity is crucial for remaining competitive.4 When a
company becomes more productive, it can produce more or higher-quality goods or services
with the same amount of resources. This often leads to lower production costs, allowing
companies to remain competitive or even expand. As a result, they may need to hire more
workers to meet the increased demand for their products or services. Also, increased
productivity in one sector can stimulate job growth in related industries; it boosts innovation
and leads to the creation of new job roles in areas such as research and development,
engineering, and information technology. Increased productivity would help address labor
market challenges, enabling employers to produce more even in tight talent markets, driving
economic growth, and creating better-paying jobs with opportunities to build human capital.
Yet Europe has experienced a long-term productivity slowdown, with productivity growth
almost steadily decreasing since the 1960s (Exhibit 2).5 Alongside its divergence in
productivity growth relative to the United States, Europe’s competitiveness is also waning.
The issues appear to be systemic rather than cyclical. European companies lag behind US
peers on multiple key metrics, such as return on invested capital, revenue growth, capital
expenditure, and R&D. Initial delays in Europe in technology development and adoption
help explain this gap, as Europe did not benefit from the information communications and
technology–driven productivity advancements that have occurred in the United States since
the 1990s. Our previous research indicates that Europe lags behind in eight out of ten key
cross-sector technologies where “winner takes most” effects are common, widening the gap
between the two regions.6 The two areas in which European companies still have an edge are
cleantech and next-gen materials, the research found.
1 In third quarter 2023, the unemployment rate stood at 6.0 percent in Europe and 3.7 percent in the United States,
compared with a peak of 11.5 percent in Europe in 1994 and 7.5 percent in the United States in 1992. For detailed data, see
“Unemployment Statistics,” Eurostat, March 2024; “Job Vacancies,” Eurostat, March 2024; and “Job Openings and Labor
Turnover,” US Bureau of Labor Statistics, March 2024.
2 Cevat Giray Aksoy et al., Time savings when working from home, National Bureau of Economic Research working paper,
number 30866, January 2023.
3 Several studies have associated remote work with productivity decreases ranging from 8 to 19 percent, whereas some
reports show a reduction of 4 percent for individual employees. Conversely, other research indicates productivity
improvements of 10 percent and more when switching to hybrid work. See, for example, Michael Gibbs, Friederike Mengel,
and Christoph Siemroth, Work from home & productivity: Evidence from personnel & analytics data on IT professionals,
Becker Friedman Institute for Economics at the University of Chicago working paper, number 2021-56, July 2021;
Natalia Emanuel and Emma Harrington, Working remotely? Selection, treatment, and the market provision of remote
work, Federal Reserve Bank of New York staff reports, number 1061, May 2023; Marta Angelici and Paola Profeta,
Smart-working: Work flexibility without constraints,” CESifo working paper, number 8165, March 2020.
4 Assuming constant exchange rates.
5 “Investing in productivity growth,” McKinsey Global Institute, March 1, 2024.
6 “Securing Europe’s competitiveness, addressing its technology gap,” McKinsey Global Institute, September 22, 2022.
A new future of work: The race to deploy AI and raise skills in Europe and beyond 6
Web <2024>
E<MxhCKib2i4t2 2172 VivaTech 2024>
Exhibit <2> of <16>
European and US productivity growth decreased seven and three
percentage points, respectively, between 1950 and 2022.
Labor productivity growth (annual change in GDP per hours worked), % year over year
Second industrialization Post-war boom Era of contention Era of markets
Electrification, mass production, Continued urbanization Energy crises and Integration of Pre- and
and industrialization and infrastructure stagflation GVCs1; ICT2 post-GFC3
build-out revolution slowdown
8
United States
Europe
6
4
2
0
−2
−4
−6
1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010 2020
Note: Productivity is defined as GDP per hour worked, in 2010 dollars, as measured by purchasing-power parity. Calculated using a Hodrick-Prescott filter (λ =
6.25). Europe is calculated using the simple average of France, Germany, Italy, Spain, Sweden, and the United Kingdom. The remaining ten European countries
in our analysis were excluded because of data availability issues.
1Global value chains.
2Information and communication technology.
3Global financial crisis.
Source: Antonin Bergeaud, Gilbert Cette, and Rémy Lecat, “Productivity trends in advanced countries between 1890 and 2012,” The Review of Income and
Wealth, September 2016, Volume 62, Number 3; McKinsey Global Institute analysis
McKinsey & Company
Automation technology has the potential to revive productivity growth, allowing economies
to solve most of today’s labor market challenges. However, Europe and the United States are
not on the same trajectory for capturing this productivity growth: most AI-related innovations
are developed in the United States. There are fears in both regions that the adoption of
these technologies could prove disruptive to labor markets and exacerbate the challenges
of both finding requisite skills in the workforce and enabling workers to move from declining
occupations into rising ones.
Workers navigated major changes in demand for work during COVID-19, which resulted in a
temporary surge in occupational transitions—a sign that labor markets could successfully
adjust to rapid and heightened shifts in the pattern of employment demand. In Europe, some
3 percent of the working population voluntarily or involuntarily exited their occupational
categories between 2019 and 2022, more than triple the historical average. In the
period between 2019 to 2022, 5.5 percent of the US working population was affected by
occupational shifts, 1.5 times the historical average.7 The occupational shifts in both Europe
and the United States have subsequently returned to their historical rate, although some
professions continue to be affected, including food service.
7 Estimates based on US Bureau of Labor Statistics data.
A new future of work: The race to deploy AI and raise skills in Europe and beyond 7
Now, as Europe looks ahead, automation, AI, and other trends present opportunities for
higher productivity growth but with faster occupational transitions. Business leaders and
policy makers will face critical choices on how much to embrace technological change and
investment while training and redeploying workers into the jobs of the future. These choices
will determine whether Europe’s countries, companies, and labor force can derive the full
productivity and human capital benefits of the future of work.
Business leaders and policy
makers will face critical
choices on how much to
embrace technological change
and investment while training
and redeploying workers
into the jobs of the future.
A new future of work: The race to deploy AI and raise skills in Europe and beyond 8
A new future of work: The race to deploy AI and raise skills in Europe and beyond 9
2
Potential for accelerated
work transitions ahead
Demand for labor will continue to evolve over time, affected by structural trends at play in
Europe and the United States. Foremost among these is the expected advancements in
technology, especially AI, which could accelerate productivity growth and alter labor demand.
Structural factors such as the aging workforce and rising healthcare needs, particularly
pronounced in Europe, and additional priorities such as climate change will also reshape
demand for workers. Additionally, some trends that were boosted by the pandemic are likely
to endure, including the growth in e-commerce and the switch to remote work.
These trends represent opportunity for productivity growth but also underscore the need for
workers to transition from declining occupations to rising ones. In Europe, by our estimates, a
faster technology adoption scenario could be associated with productivity growth of roughly
2 to 3 percent per year, requiring some 12 million occupational transitions, or roughly double
the pace of occupational shifts in the prepandemic period. In the United States, with its
more dynamic labor market, the trend would be closer to the historical norm, but automation
adoption could accelerate further after 2030 in both regions. While the scale of occupational
transitions may appear daunting, both Europe and the United States navigated even higher
levels of labor market shifts during the pandemic, signaling the potential to handle future
transitions as well.
In this chapter, we outline how demand for labor could evolve and require accelerated
occupational transitions in the coming years, considering a range of scenarios to reflect
the uncertainties around pace of technology adoption (see Box 1, “Our methodology for
estimating occupational transitions”).
A new future of work: The race to deploy AI and raise skills in Europe and beyond 10
Box 1
Our methodology for estimating occupational transitions
We used methodology consistent with other A critical driver of occupational transitions is the
McKinsey Global Institute reports on the future rate at which automation, AI, and generative AI (gen
of work, dating back to 2017, to model trends of AI) will be adopted (exhibit). Two scenarios are used
job changes at the level of occupations, activities, to bookend the work-automation model: “late” and
and skills.1 For this report, we focused our analysis “early.” The “early” scenario flexes all parameters
on the 2022–30 period. We also considered how to the extremes of plausible assumptions, resulting
automation adoption could evolve beyond 2030 to in the fastest pace of automation development
2035.2 The drivers of the model have been updated and adoption, and the “late” scenario flexes all
accordingly. parameters in the opposite direction. The reality is
likely to fall somewhere between the two.4
Our model differentiates between employment
demand and occupational transitions. For the For this report, we have modeled region-specific
first, it estimates net changes in employment scenarios:
demand by sector and occupation; for the second,
— For Europe, we modeled two outcomes: a
it estimates the net decline in occupations across
“faster” scenario and a “slower” one. For the
sectors compared with the 2030 baseline. When
faster scenario, we use the midpoint—the
counting transitions, we do not include gains in this
arithmetical average between our late and
calculation to avoid double counting.
early scenarios. For the slower scenario, we
In this report, we focus our analysis on Europe use a “mid late” trajectory, an arithmetical
and the United States. For Europe, we included average between a late adoption scenario
ten countries: nine EU members that together and the midpoint scenario. We model this
represent 75 percent of the European working slower, mid-late scenario for Europe because
population—the Czech Republic, Denmark, France, achieving the faster, midpoint scenario by
Germany, Italy, Netherlands, Poland, Spain, and 2030 would require an occupational transition
Sweden—and the United Kingdom. In this report, rate significantly higher than seen in Europe’s
numbers referring to “Europe” correspond to the recent prepandemic past.
total estimates for these ten focus countries, which
— For the United States, we use the midpoint
were analyzed individually. Numbers have not
scenario, based on our earlier research. This
been extrapolated to the full European working
is an arithmetical average between our late
population. For the United States, we build on
and early scenarios of automation technology
estimates published in our 2023 report Generative
adoption.
AI and the future of work in America.3
We also estimate the productivity effects of
To understand the impact of automation and
automation, using GDP per full-time-equivalent
overall potential changes in demand in each
(FTE) employee as the measure of productivity. We
occupation, we included multiple drivers in our
first calculated automation displacement under
modeling: automation adoption, net-zero transition,
different scenarios by multiplying the projected
e-commerce growth, remote work adoption,
number of FTEs by the estimated automation
increases in income, aging populations, technology
adoption rate for each occupation in each country.
investments, infrastructure investments,
We considered only job activities that are available
marketization of unpaid work, new jobs, and
and well defined as of the date of this report. Also,
increased educational levels.
to be conservative, we assumed automation has a
labor substitution effect but no other performance
1 The modeling examines more than 850 unique occupations, more than 2,000 different activities, and 18 technical capabilities for each
activity. We also leveraged the framework devised in MGI’s 2018 report Skill shift: Automation and the future of the workforce. For
more detail, see the technical appendixes in A future that works: Automation, employment, productivity, McKinsey Global Institute,
January 2017.
2 For 2035, we modeled only the potential automation adoption rates for each occupation, not the occupational transitions required.
3 For more, see “Generative AI and the future of work in America,” McKinsey Global Institute, July 26, 2023.
4 “The economic potential of generative AI: The next productivity frontier,” McKinsey, June 14, 2023.
A new future of work: The race to deploy AI and raise skills in Europe and beyond 11
gains. We assumed that workers displaced which scenario evolves. Second, labor demand
by automation rejoin the workforce at 2022 could shift based on macroeconomic shifts in
productivity levels, net of automation. consumption due to changes in prices and costs,
which our model does not account for. Indeed, as
Our main sources of data are national and regional
automation increases productivity and income and
labor surveys. For the United States, we used data
lowers costs and the prices of goods and services,
from the Current Population Survey, conducted by
it could shift consumption, and thus labor demand,
the US Census Bureau for the US Bureau of Labor
in unanticipated ways. In the literature, this
Statistics. For Europe, we used data from the
specific impact of automation has been framed as
Labor Force Survey carried out by the European
the “deflationist” nature of technology adoption.
Commission and local labor agencies’ data.
Rapid adoption of technology could therefore
As described in chapter 4, we also conducted
establish a new equilibrium of demand. Third, the
a survey of more than 1,100 executives in five
shifts we model are the ones broadly anticipated
countries.
given the underlying base and current momentum
Our model has some important uncertainties of economies. We do not model changes in
and limitations. First, structural attributes— industrial production, trade, or labor migration that
such as management–employee relations, the may be driven by geopolitical, climatic, or social
regulatory and investment framework, and current factors, for example.
AI and innovation momentum—would affect
Web <2024>
E<MxhCKib2i4t2 172 VivaTech 2024-BOX>
Exhibit <B1> of <16>
Europe has varying automation adoption scenarios through 2030.
Automation of current work activities, % of working hours modeled to be automated, with generative AI
acceleration, Europe1 and the US, 2022–80
Early scenario Europe1 Late scenario Europe1 Faster scenario Europe1
Slower scenario Europe1 Midpoint scenario US
100
80
60
40
20
0
2022 2030 2040 2050 2060 2070 2080
Note: The range of scenarios represents uncertainty regarding the availability of technical capabilities, based on interviews with experts and survey responses.
The early scenario makes more-aggressive assumptions for all key model parameters (technical potential, integration timeline, economic feasibility, and regula-
tory and public adoption). The “faster” or midpoint adoption scenario is the average between the early and late scenarios. The “slower” scenario is the average
between the late scenario and the midpoint scenario.
1Includes Czech Republic, Denmark, France, Germany, Italy, Netherlands, Poland, Spain, Sweden, and United Kingdom.
Source: Eurostat; Occupational Information Network; Oxford Economics; US Bureau of Labor Statistics; national statistical agencies of the European countries
considered; McKinsey Global Institute analysis
McKinsey & Company
A new future of work: The race to deploy AI and raise skills in Europe and beyond 12
As technology reshapes work, demand is changing
for a wide range of occupations
Our analysis suggests that demand for some occupations could grow sharply by 2030. In our
faster, midpoint technology adoption scenario, demand for STEM and health professionals
would grow by 17 to 30 percent between 2022 and 2030, adding seven million positions in
Europe and an additional seven million in the United States. Despite the surge in tech sector
layoffs in 2023 and the potential of generative AI (gen AI) to augment tasks such as coding,
the broader, long-term demand for tech talent could remain robust across businesses of
every size and sector in an increasingly digital economy (Exhibit 3). Similarly, demand for
health aides, technicians, and wellness workers could continue growing by 25 to 30 percent
between 2022 and 2030, adding 3.3 million positions in Europe and 3.5 million in the
United States.
By contrast, demand for workers in food services, production work, customer services,
sales, and office support—all of which declined over the 2012–22 period—could continue to
decline until 2030.8 These jobs involve a high share of repetitive tasks, data collection, and
elementary data processing—all activities that automated systems can handle efficiently. In
all, our analysis suggests that this could lead to decreases in demand for these positions of
between 300,000 and 5.0 million positions in Europe and 0.1 million to 3.7 million positions in
the United States.
Demand for other occupations would remain in line with overall demand growth. This includes
positions for educators and workforce trainers in Europe and includes businesses and
legal professionals, as well as community services workers, in the United States. Demand
for occupations such as management, construction, creative and arts management, and
transportation services is expected to increase by about 8 to 9 percent.
Our analysis highlights some differences between Europe and the United States in the
occupations with growing or diminishing demand. Those differences are a result of the
differences in occupational composition between the two regions, as well as cultural
specificities. For example, the greater share of public employment in Europe, especially in
administrative activities, may reduce the impact of the expected disruption on these workers
for the coming years. Understanding the nuances of how this might play out and who might be
affected is critical to ensuring a smooth transition for individuals and businesses alike.
8 Examples here include cashiers, call-center representatives, tellers, and guest service agents.
A new future of work: The race to deploy AI and raise skills in Europe and beyond 13
Web <2024>
E<MxhCKib2i4t2 3172 VivaTech 2024>
Exhibit <3> of <16>
Demand for healthcare and STEM roles could grow, while demand for office
support and customer service roles could decline.
Net expected change in labor demand, Europe1 and US, faster/midpoint scenario,1 2022–30
Europe2 US
Employ- Employ-
ment ment
Employment change vs 2022, change vs Employment change vs 2022, change vs
Occupational category million 2022, % million 2022, %
Health aides, techni-
3.3 25.2 3.5 29.7
cians, and wellness
STEM professionals 2.3 16.7 1.8 23.1
Health professionals 1.5 23.6 2.0 30.1
Managers 1.1 9.1 1.1 11.3
Business or legal pro-
1.0 6.9 1.1 6.6
fessionals
Builders 0.7 6.9 0.8 11.9
Transportation services 0.5 7.9 0.5 9.5
Property maintenance 0.4 5.3 0.5 10.3
Creatives and arts man-
0.4 8.6 0.2 10.7
agement
Community services 0.3 3.5 0.4 6.6
Educator and workforce
0.2 1.6 0.3 2.6
training
Mechanical installation
0.1 1.2 0.5 7.0
and repair
Agriculture −0.2 –3.8 0 2.3
Food services −0.3 –3.3 −0.3 –1.9
Production work −0.9 –5.3 −0.1 –0.7
Customer service and
−1.7 –12.1 −2.0 –13.4
sales
Office support −5.0 –18.3 −3.7 –18.5
1For Europe, we used the “faster” scenario, which corresponds to the “midpoint” scenario in the United States. The “faster” or midpoint adoption scenario is the
average between the early and late scenarios. The “slower” scenario is the average between the late scenario and the midpoint scenario.
2Includes Czech Republic, Denmark, France, Germany, Italy, Netherlands, Poland, Spain, Sweden, and United Kingdom.
Source: Eurostat; Occupational Information Network; Oxford Economics; US Bureau of Labor Statistics; national statistical agencies of the European countries
considered; McKinsey Global Institute analysis
McKinsey & Company
A new future of work: The race to deploy AI and raise skills in Europe and beyond 14
Some 12 million occupational transitions may be needed
in both Europe and the United States by 2030
Our analysis finds that in our faster automation adoption scenario, some 12.0 million
occupational transitions would be needed by 2030 in the ten European countries, affecting
6.5 percent of the current employed workforce.9 Under the slower scenario, the number of
occupational transitions needed would amount to 8.5 million in Europe, affecting 4.6 percent
of the current employed workforce. In the United States, the figures for the midpoint scenario
we use (which corresponds to the faster European scenario) are 11.8 million occupational
shifts, affecting 7.5 percent of the current employed workforce.
The range of outcomes for Europe from the two scenarios reflects different potential for the
number of work hours that could be automated, thereby affecting both potential productivity
gains and the number of occupational transitions that might be needed. A failure to achieve
the faster-paced adoption model would mean fewer occupational transitions are needed. But
it would also mean failing to achieve some significant productivity gains in the period to 2030.
Occupational transitions would need to roughly double in Europe
but return to their historical level in the United States
The pace of change in required occupational transitions is uneven between Europe and the
United States. Europe could experience a stark acceleration in the pace of occupational
change needed in both the faster and slower scenarios, with the number rising to between 1.1
million and 1.5 million occupational transitions annually between 2022 and 2030. That is 1.6
to 2.2 times the historical 2016–19 rate, before the COVID-19 pandemic, indicating a potential
doubling of this measure of change in the European employment market. By contrast, in
the United States, the number of occupational transitions needed annually between 20 |
260 | mckinsey | beyond-the-hype-capturing-the-potential-of-ai-and-gen-ai-in-tmt.pdf | Beyond the hype:
Capturing the potential
of AI and gen AI in tech,
media, and telecom
February 2024
Beyond the hype:
Capturing the potential
of AI and gen AI in tech,
media, and telecom
February 2024
Contents
Introduction: The promise and the challenge of generative AI 2
State of the Art
4
The economic potential of generative AI 5
Making the most of the generative AI opportunity: Six questions for CEOs 33
Sector View: Telecom Operators
38
The AI-native telco: Radical transformation to thrive in turbulent times 39
How generative AI could revitalize profitability for telcos 48
Generative AI use cases: A guide to developing the telco of the future 60
Tech talent in transition: Seven technology trends reshaping telcos 70
Deploying Gen AI
81
The organization of the future: Enabled by gen AI, driven by people 82
The data dividend: Fueling generative AI 91
Technology’s generational moment with generative AI: A CIO and CTO guide 101
As gen AI advances, regulators—and risk functions—rush to keep pace 113
What the Future Holds
119
Six major gen AI trends that will shape 2024’s agenda 120
Appendix: Generative AI solutions in action 125
Glossary 127
Beyond the hype: Capturing the potential of AI and gen AI in TMT 1
Introduction: The promise and
the challenge of generative AI
The emergence of generative AI (gen AI) presents both a challenge and a significant opportunity for leaders looking
to steer their organizations into the future. How big is the opportunity? McKinsey research estimates that gen AI
could add to the economy between $2.6 trillion and $4.4 trillion annually while increasing the impact of all artificial
intelligence by 15 to 40 percent. In the technology, media, and telecommunications (TMT) space, new gen AI use
cases are expected to unleash between $380 billion and $690 billion in impact—$60 billion to $100 billion in
telecommunications, $80 billion to $130 billion in media, and about $240 billion to $460 billion in high tech. In
fact, it seems possible that within the next three years, anything not connected to AI will be considered obsolete or
ineffective.
Some leaders are moving to seize the moment and implement gen AI in their organizations at scale, but others remain
in the pilot stage, and some have yet to decide what to do. If companies are to remain competitive and relevant in the
coming years, it is essential that executives understand the potential impact of gen AI and develop the strategies
necessary to incorporate it into their operations. Such strategies would involve an AI-native transformation, focused
on building and managing the adoption of gen AI. McKinsey has conducted extensive research into how to embed
gen AI to ensure that the technology delivers meaningful value. We’ve also spent much of the past year working with
clients to create and then implement gen AI road maps. That combination of research and hands-on experience has
allowed us to identify more than 100 gen AI use cases in TMT across seven business domains.1
Our experience working with clients already indicates the potential for telcos to achieve significant impact with
gen AI across all key functions. The largest share of total impact will likely be in customer care and sales, which
together would account for approximately 70 percent of total impact; network operations, IT, and support functions
would round out the rest. The technology already is showing meaningful impact in enhancing interactions between
employees and customers: the personalization of products and campaigns, improvements in sales effectiveness, and
a reduction in time to market can spark a potential revenue increase of 3 to 5 percent. Customer care interactions—
where as much as 50 percent of activity could be automated—have potential for a 30 to 45 percent increase in
productivity while improving the customer experience and customer satisfaction scores. On the labor side, up to 70
percent of repetitive work activities could be automated via gen AI to improve productivity. There is also potential for
new efficiencies in knowledge search, validation, and synthesis, where some 60 percent of activity has the potential
for automation. And gen AI tools could boost developer productivity by 20 to 45 percent.
These areas provide rich soil for use cases. More challenging will be to go from sketching a road map to building
proofs of concept to scaling successfully and capturing impact. Years of experience in designing and implementing
digital transformations have taught us a lot, but gen AI’s nature and speed of disruption are creating a new layer of
uncertainty.
Becoming an AI-native organization at scale involves making the most of technology, data, and governance. Success
follows when leaders embrace an operating model that leverages the strengths of both humans and machines; is
rooted in agility, flexibility, and continuous learning; and is supported by strong data and analytics talent. Another
condition of success is to invest in data quality and quantity, focusing on the data life cycle to ensure high-quality
information for training the gen AI model. Building capabilities into the data architecture, such as vector databases
and data pre- and post-processing pipelines, will enable the development of use cases. Talent, data, technology,
governance—none of these can be an afterthought.
¹ Marketing and digital, sales and channels, customer care, customer strategy, support, additional areas, and new businesses.
Beyond the hype: Capturing the potential of AI and gen AI in TMT 2
Successful implementations share a clear vision and decisive approach. We advise that financial plans maintain or
increase gen AI budgets over the next year. These budgets should include resources dedicated to gen AI for the shaping
and crafting of bespoke solutions (for example, training large language models with telco-specific data, rather than
implementing off-the-shelf ones) or partnerships with IT vendors to accelerate the timeline for implementation.
The AI journey has been shown to contain many challenges and learning opportunities, such as preparing and shifting
an organization’s culture, finding data sets of significant size, and addressing the interpretability of the outputs provided
by models. Leaders should expect such daunting challenges as a shortage of talent, lack of organizational commitment
and prioritization (including among C-level executives), and difficulties in justifying ROI for certain business cases, all
amid a changing regulatory and ethics landscape that creates further uncertainty. But daunting does not have to mean
impossible. Developing a system of protocols and guardrails (such as building “moderation” models to check outputs
for different risks and ensure users receive consistent responses) will be a crucial step toward mitigating the new risks
introduced by gen AI. Another key will be change management—involving end users in the model development process
and deeply embedding technology into their operations.
This collection presents McKinsey’s top insights on gen AI, providing a detailed examination of this technology’s
transformative potential for organizations. It offers top management guidance on how to prepare for the implementation
of gen AI and explores the implications of gen AI’s use by the TMT industries, especially telecommunications. The
collection covers the essential requirements for deploying gen AI, including organizational readiness, data management,
and technological considerations. It also emphasizes the importance of effectively managing risks associated with gen
AI implementation. Furthermore, this compilation offers an overview of the future developments and advancements
expected in the field of generative AI.
Gen AI will continue to evolve. New capabilities, such as the ability to analyze and comprehend images or audio, and an
expanding ecosystem with marketplaces for GPT (generative pretrained transformers), are constantly emerging. For
leaders, the stakes are high. But so are the opportunities. The next move from TMT players will define how they move
from isolated cases to implementations at scale, from hype to impact.
Alex Singla Alexander Sukharevsky Brendan Gaffey Noshir Kaka
Senior Partner Senior Partner Senior Partner Senior Partner
Managing Partner Managing Partner Global Leader Global Leader
QuantumBlack QuantumBlack TMT Practice TMT Practice
AI by McKinsey AI by McKinsey
Peter Dahlström Andrea Travasoni Venkat Atluri
Senior Partner Senior Partner Senior Partner
Europe Leader Global Leader Global Leader
TMT Practice Telecom Operators Telecom Operators
TMT Practice TMT Practice
Tomás Lajous Benjamim Vieira Víctor García de la Torre
Senior Partner Senior Partner Associate Partner
AI and Gen AI Leader Digital and Analytics Leader TMT Practice
TMT Practice TMT Practice
Beyond the hype: Capturing the potential of AI and gen AI in TMT 3
1
State of
the art
Beyond the hype: Capturing the potential of AI and gen AI in TMT 4
The
economic
potential
of
generative
AI
The economic
potential of
generative AI
The next productivity frontier
June 2023
Authors
Michael Chui
Eric Hazan
Roger Roberts
Alex Singla
Kate Smaje
Alexander Sukharevsky
Lareina Yee
Rodney Zemmel
1
Generative AI as a
technology catalyst
To grasp what lies ahead requires an understanding of the breakthroughs that have enabled the rise
of generative AI, which were decades in the making. ChatGPT, GitHub Copilot, Stable Diff usion, and
other generative AI tools that have captured current public attention are the result of signifi cant levels
of investment in recent years that have helped advance machine learning and deep learning. This
investment undergirds the AI applications embedded in many of the products and services we use
every day.
But because AI has permeated our lives incrementally—through everything from the tech powering
our smartphones to autonomous-driving features on cars to the tools retailers use to surprise and
delight consumers—its progress was almost imperceptible. Clear milestones, such as when AlphaGo,
an AI-based program developed by DeepMind, defeated a world champion Go player in 2016, were
celebrated but then quickly faded from the public’s consciousness.
ChatGPT and its competitors have captured the imagination of people around the world in a way
AlphaGo did not, thanks to their broad utility—almost anyone can use them to communicate and
create—and preternatural ability to have a conversation with a user. The latest generative AI
applications can perform a range of routine tasks, such as the reorganization and classifi cation
This article is excerpted from the full McKinsey report, The economic potential of generative AI: The
next productivity frontier. To read the full report, including details about the research, appendix, and
acknowledgements, visit mck.co/genai.
The economic potential of generative AI: The next productivity frontier 6
of data. But it is their ability to write text, compose music, and create digital art that has garnered
headlines and persuaded consumers and households to experiment on their own. As a result, a
broader set of stakeholders are grappling with generative AI’s impact on business and society but
without much context to help them make sense of it.
How did we get here? Gradually, then all of a sudden
For the purposes of this report, we define generative AI as applications typically built using foundation
models. These models contain expansive artificial neural networks inspired by the billions of neurons
connected in the human brain. Foundation models are part of what is called deep learning, a term
that alludes to the many deep layers within neural networks. Deep learning has powered many of
the recent advances in AI, but the foundation models powering generative AI applications are a step
change evolution within deep learning. Unlike previous deep learning models, they can process
extremely large and varied sets of unstructured data and perform more than one task.
Foundation models have enabled new capabilities and vastly improved existing ones across a broad
range of modalities, including images, video, audio, and computer code. AI trained on these models
can perform several functions; it can classify, edit, summarize, answer questions, and draft new
content, among other tasks.
Continued innovation will also bring new challenges. For example, the computational power required
to train generative AI with hundreds of billions of parameters threatens to become a bottleneck in
development.¹ Further, there’s a significant move—spearheaded by the open-source community and
spreading to the leaders of generative AI companies themselves—to make AI more responsible, which
could increase its costs.
Nonetheless, funding for generative AI, though still a fraction of total investments in artificial
intelligence, is significant and growing rapidly—reaching a total of $12 billion in the first five months
of 2023 alone. Venture capital and other private external investments in generative AI increased by
an average compound growth rate of 74 percent annually from 2017 to 2022. During the same period,
investments in artificial intelligence overall rose annually by 29 percent, albeit from a higher base.
The rush to throw money at all things generative AI reflects how quickly its capabilities have
developed. ChatGPT was released in November 2022. Four months later, OpenAI released a new
large language model, or LLM, called GPT-4 with markedly improved capabilities.² Similarly, by May
2023, Anthropic’s generative AI, Claude, was able to process 100,000 tokens of text, equal to about
75,000 words in a minute—the length of the average novel—compared with roughly 9,000 tokens
when it was introduced in March 2023.³ And in May 2023, Google announced several new features
powered by generative AI, including Search Generative Experience and a new LLM called PaLM 2 that
will power its Bard chatbot, among other Google products.⁴
From a geographic perspective, external private investment in generative AI, mostly from tech
giants and venture capital firms, is largely concentrated in North America, reflecting the continent’s
current domination of the overall AI investment landscape. Generative AI–related companies based
in the United States raised about $8 billion from 2020 to 2022, accounting for 75 percent of total
investments in such companies during that period.⁵
Generative AI has stunned and excited the world with its potential for reshaping how knowledge work
gets done in industries and business functions across the entire economy. Across functions such
as sales and marketing, customer operations, and software development, it is poised to transform
roles and boost performance. In the process, it could unlock trillions of dollars in value across sectors
from banking to life sciences. We have used two overlapping lenses in this report to understand
The economic potential of generative AI: The next productivity frontier 7
2
Generative AI use
cases across functions
and industries
the potential for generative AI to create value for companies and alter the workforce. The
following sections share our initial findings.
The economic potential of generative AI: The next productivity frontier 8
Exhibit 1
The potential impact of generative AI can be evaluated through two lenses.
Lens 1 Lens 2
Total economic Labor productivity potential
potential of 60-plus across ~2,100 detailed work
organizational use activities performed by
cases1 global workforce
Cost impacts
of use cases
Revenue
impacts of
use cases1
1For quantitative analysis, revenue impacts were recast as productivity increases on the corresponding spend in order to maintain comparability with cost
impacts and not to assume additional growth in any particular market.
McKinsey & Company
Generative AI is a step change in the evolution of artifi cial intelligence. As companies
rush to adapt and implement it, understanding the technology’s potential to deliver value
to the economy and society at large will help shape critical decisions. We have used two
complementary lenses to determine where generative AI with its current capabilities could
deliver the biggest value and how big that value could be (Exhibit 1).
The fi rst lens scans use cases for generative AI that organizations could adopt. We defi ne
a “use case” as a targeted application of generative AI to a specifi c business challenge,
resulting in one or more measurable outcomes. For example, a use case in marketing is the
application of generative AI to generate creative content such as personalized emails, the
measurable outcomes of which potentially include reductions in the cost of generating such
content and increases in revenue from the enhanced eff ectiveness of higher-quality content
at scale. We identifi ed 63 generative AI use cases spanning 16 business functions that could
deliver total value in the range of $2.6 trillion to $4.4 trillion in economic benefi ts annually
when applied across industries.
That would add 15 to 40 percent to the $11.0 trillion to $17.7 trillion of economic value that we
now estimate nongenerative artifi cial intelligence and analytics could unlock. (Our previous
estimate from 2017 was that AI could deliver $9.5 trillion to $15.4 trillion in economic value.)
Our second lens complements the fi rst by analyzing generative AI’s potential impact on
the work activities required in some 850 occupations. We modeled scenarios to estimate
when generative AI could perform each of more than 2,100 “detailed work activities”—
The economic potential of generative AI: The next productivity frontier 9
such as “communicating with others about operational plans or activities”—that make up
those occupations across the world economy. This enables us to estimate how the current
capabilities of generative AI could aff ect labor productivity across all work currently done by
the global workforce.
Some of this impact will overlap with cost reductions in the use case analysis described
above, which we assume are the result of improved labor productivity. Netting out this
Exhibit 2
Generative AI could create additional value potential above what
could be unlocked by other AI and analytics.
AI’s potential impact on the global economy, $ trillion
17.1–25.6
13.6–22.1
6.1–7.9
2.6–4.4
11.0–17.7
~15–40% ~35–70%
incremental incremental
economic impact economic impact
Advanced analytics, New generative Total use All worker productivity Total AI
traditional machine AI use cases case-driven enabled by generative economic
learning, and deep potential AI, including in use potential
learning1 cases
1Updated use case estimates from "Notes from the AI frontier: Applications and value of deep learning,” McKinsey Global Institute, April 17, 2018.
McKinsey & Company
The economic potential of generative AI: The next productivity frontier 10
overlap, the total economic benefits of generative AI—including the major use cases we
explored and the myriad increases in productivity that are likely to materialize when the
technology is applied across knowledge workers’ activities—amounts to $6.1 trillion to
$7.9 trillion annually (Exhibit 2).
While generative AI is an exciting and rapidly advancing technology, the other applications of
AI discussed in our previous report continue to account for the majority of the overall potential
value of AI. Traditional advanced-analytics and machine learning algorithms are highly
Box 1
How we estimated the value potential of generative AI use cases
To assess the potential value of generative AI, a customer service use case but not in a use
we updated a proprietary McKinsey database of case optimizing a logistics network, where value
potential AI use cases and drew on the experience primarily arises from quantitative analysis.
of more than 100 experts in industries and their
We then estimated the potential annual value
business functions.1 Our updates examined
of these generative AI use cases if they were
use cases of generative AI—specifically, how
adopted across the entire economy. For use
generative AI techniques (primarily transformer-
cases aimed at increasing revenue, such as some
based neural networks) can be used to solve
of those in sales and marketing, we estimated
problems not well addressed by previous
the economy-wide value generative AI could
technologies.
deliver by increasing the productivity of sales and
We analyzed only use cases for which generative marketing expenditures.
AI could deliver a significant improvement in the
Our estimates are based on the structure of the
outputs that drive key value. In particular, our
global economy in 2022 and do not consider the
estimates of the primary value the technology
value generative AI could create if it produced
could unlock do not include use cases for which
entirely new product or service categories.
the sole benefit would be its ability to use natural
language. For example, natural-language
capabilities would be the key driver of value in
1 “Notes from the AI frontier: Applications and value of deep learning,” McKinsey Global Institute, April 17, 2018.
effective at performing numerical and optimization tasks such as predictive modeling, and
they continue to find new applications in a wide range of industries. However, as generative AI
continues to develop and mature, it has the potential to open wholly new frontiers in creativity
and innovation. It has already expanded the possibilities of what AI overall can achieve (see
Box 1, “How we estimated the value potential of generative AI use cases”).
In this chapter, we highlight the value potential of generative AI across two dimensions:
business function and modality.
The economic potential of generative AI: The next productivity frontier 11
Value potential by function
While generative AI could have an impact on most business functions, a few stand out when
measured by the technology’s impact as a share of functional cost (Exhibit 3). Our analysis
of 16 business functions identifi ed just four—customer operations, marketing and sales,
software engineering, and research and development—that could account for approximately
75 percent of the total annual value from generative AI use cases.
Web <2023>
E<Vxihvaitbeicth 3 full report>
Exhibit <3> of <16>
Using generative AI in just a few functions could drive most of the technology’s
impact across potential corporate use cases.
Represent ~75% of total annual impact of generative AI
500
Sales
Software engineering
Marketing
(for corporate IT)
Software engineering
(for product development)
400
Customer operations
Product R&D1
300
Impact, $ billion
Supply chain
200
Manufacturing
Finance Risk and compliance
Talent and organization (incl HR)
100
Procurement management
Corporate IT1 Legal
Strategy
Pricing
0
0 10 20 30 40
Impact as a percentage of functional spend, %
Note: Impact is averaged.
¹Excluding software engineering.
Source: Comparative Industry Service (CIS), IHS Markit; Oxford Economics; McKinsey Corporate and Business Functions database; McKinsey Manufacturing
and Supply Chain 360; McKinsey Sales Navigator; Ignite, a McKinsey database; McKinsey analysis
McKinsey & Company
Notably, the potential value of using generative AI for several functions that were prominent
in our previous sizing of AI use cases, including manufacturing and supply chain functions,
is now much lower.⁶ This is largely explained by the nature of generative AI use cases, which
exclude most of the numerical and optimization applications that were the main value drivers
for previous applications of AI.
The economic potential of generative AI: The next productivity frontier 12
Generative AI as a virtual expert
In addition to the potential value generative AI can deliver in function-specific use cases,
the technology could drive value across an entire organization by revolutionizing internal
knowledge management systems. Generative AI’s impressive command of natural-language
processing can help employees retrieve stored internal knowledge by formulating queries
in the same way they might ask a human a question and engage in continuing dialogue. This
could empower teams to quickly access relevant information, enabling them to rapidly make
better-informed decisions and develop effective strategies.
In 2012, the McKinsey Global Institute (MGI) estimated that knowledge workers spent about
a fifth of their time, or one day each workweek, searching for and gathering information. If
generative AI could take on such tasks, increasing the efficiency and effectiveness of the
workers doing them, the benefits would be huge. Such virtual expertise could rapidly “read”
vast libraries of corporate information stored in natural language and quickly scan source
material in dialogue with a human who helps fine-tune and tailor its research, a more scalable
solution than hiring a team of human experts for the task.
Following are examples of how generative AI could produce operational benefits as a virtual
expert in a handful of use cases.
In addition to the potential
value generative AI can
deliver in function-specific
use cases, the technology
could drive value across
an entire organization
by revolutionizing
internal knowledge
management systems.
The economic potential of generative AI: The next productivity frontier 13
Customer operations
Generative AI has the potential to revolutionize the entire customer operations function,
improving the customer experience and agent productivity through digital self-service
and enhancing and augmenting agent skills. The technology has already gained traction
in customer service because of its ability to automate interactions with customers using
natural language. Research found that at one company with 5,000 customer service
agents, the application of generative AI increased issue resolution by 14 percent an hour and
reduced the time spent handling an issue by 9 percent.⁷ It also reduced agent attrition and
requests to speak to a manager by 25 percent. Crucially, productivity and quality of service
improved most among less-experienced agents, while the AI assistant did not increase—
and sometimes decreased—the productivity and quality metrics of more highly skilled
agents. This is because AI assistance helped less-experienced agents communicate using
techniques similar to those of their higher-skilled counterparts.
The following are examples of the operational improvements generative AI can have for
specific use cases:
— Customer self-service. Generative AI–fueled chatbots can give immediate and
personalized responses to complex customer inquiries regardless of the language or
location of the customer. By improving the quality and effectiveness of interactions via
automated channels, generative AI could automate responses to a higher percentage of
customer inquiries, enabling customer care teams to take on inquiries that can only be
resolved by a human agent. Our research found that roughly half of customer contacts
made by banking, telecommunications, and utilities companies in North America are
already handled by machines, including but not exclusively AI. We estimate that generative
AI could further reduce the volume of human-serviced contacts by up to 50 percent,
depending on a company’s existing level of automation.
— Resolution during initial contact. Generative AI can instantly retrieve data a company
has on a specific customer, which can help a human customer service representative more
successfully answer questions and resolve issues during an initial interaction.
— Reduced response time. Generative AI can cut the time a human sales representative
spends responding to a customer by providing assistance in real time and recommending
next steps.
— Increased sales. Because of its ability to rapidly process data on customers and their
browsing histories, the technology can identify product suggestions and deals tailored
to customer preferences. Additionally, generative AI can enhance quality assurance and
coaching by gathering insights from customer conversations, determining what could be
done better, and coaching agents.
We estimate that applying generative AI to customer care functions could increase
productivity at a value ranging from 30 to 45 percent of current function costs.
Our analysis captures only the direct impact generative AI might have on the productivity of
customer operations. It does not account for potential knock-on effects the technology may
have on customer satisfaction and retention arising from an improved experience, including
better understanding of the customer’s context that can assist human agents in providing
more personalized help and recommendations.
The economic potential of generative AI: The next productivity frontier 14
Marketing and sales
Generative AI has taken hold rapidly in marketing and sales functions, in which text-based
communications and personalization at scale are driving forces. The technology can create
personalized messages tailored to individual customer interests, preferences, and behaviors,
as well as do tasks such as producing first drafts of brand advertising, headlines, slogans,
social media posts, and product descriptions.
However, introducing generative AI to marketing functions requires careful consideration.
For one thing, using mathematical models trained on publicly available data without
sufficient safeguards against plagiarism, copyright violations, and branding recognition risks
infringing on intellectual property rights. A virtual try-on application may produce biased
representations of certain demographics because of limited or biased training data. Thus,
significant human oversight is required for conceptual and strategic thinking specific to each
company’s needs.
Potential operational benefits from using generative AI for marketing include the following:
— Efficient and effective content creation. Generative AI could significantly reduce the
time required for ideation and content drafting, saving valuable time and effort. It can also
facilitate consistency across different pieces of content, ensuring a uniform brand voice,
writing style, and format. Team members can collaborate via generative AI, which can
integrate their ideas into a single cohesive piece. This would allow teams to significantly
enhance personalization of marketing messages aimed at different customer segments,
geographies, and demographics. Mass email campaigns can be instantly translated into
as many languages as needed, with different imagery and messaging depending on the
audience. Generative AI’s ability to produce content with varying specifications could
increase customer value, attraction, conversion, and retention over a lifetime and at a
scale beyond what is currently possible through traditional techniques.
— Enhanced use of data. Generative AI could help marketing functions overcome the
challenges of unstructured, inconsistent, and disconnected data—for example, from
different databases—by interpreting abstract data sources such as text, image, and
varying structures. It can help marketers better use data such as territory performance,
synthesized customer feedback, and customer behavior to generate data-informed
marketing strategies such as targeted customer profiles and channel recommendations.
Such tools could identify and synthesize trends, key drivers, and market and product
opportunities from unstructured data such as social media, news, academic research, and
customer feedback.
— SEO optimization. Generative AI can help marketers achieve higher conversion and
lower cost through search engine optimization (SEO) for marketing and sales technical
components such as page titles, image tags, and URLs. It can synthesize key SEO tokens,
support specialists in SEO digital content creation, and distribute targeted content to
customers.
— Product discovery and search personalization. With generative AI, product discovery
and search can be personalized with multimodal inputs from text, images and speech, and
deep understanding of customer profiles. For example, technology can leverage individual
user preferences, behavior, and purchase history to help customers discover the most
The economic potential of generative AI: The next productivity frontier 15
relevant products and generate personalized product descriptions. This would allow
CPG, travel, and retail companies to improve their e-commerce sales by achieving higher
website conversion rates.
We estimate that generative AI could increase the productivity of the marketing function with
a value between 5 and 15 percent of total marketing spending.
Our analysis of the potential use of generative AI in marketing doesn’t account for knock-on
effects beyond the direct impacts on productivity. Generative AI–e |
261 | mckinsey | mckinsey-technology-trends-outlook-2024.pdf | Technology Trends
Outlook 2024
July 2024
McKinsey & Company
McKinsey & Company is a global management consulting firm, deeply committed to helping
institutions in the private, public, and social sectors achieve lasting success. For more than
90 years, our primary objective has been to serve as our clients’ most trusted external adviser.
With consultants in more than 100 cities in over 60 markets, across industries and functions, we
bring unparalleled expertise to clients all over the world. We work closely with teams at all levels
of an organization to shape winning strategies, mobilize for change, build capabilities, and drive
successful execution.
Contents
Insights across trends 4
The AI revolution 13 Cutting-edge engineering 65
Generative AI 14 Future of robotics 66
Applied AI 20 Future of mobility 71
Industrializing machine learning 25 Future of bioengineering 77
Future of space technologies 82
Building the digital future 30 A sustainable world 87
Next-generation software develo pment 31 Electrification and renewables 88
Digital trust and cybersecurity 36 Climate technologies beyond
electrification and renewables 94
Compute and connectivity frontiers 43
Advanced connectivity 44
Immersive-reality technologies 49
Cloud and edge computing 54
Quantum technologies 59
Technology Trends Outlook 2024 3
Insights across trends
Despite challenging overall market conditions in New and notable
2023, continuing investments in frontier technologies
The two trends that stood out in 2023 were gen AI and
promise substantial future growth in enterprise adoption.
electrification and renewables. Gen AI has seen a spike
Generative AI (gen AI) has been a standout trend since
of almost 700 percent in Google searches from 2022
2022, with the extraordinary uptick in interest and
to 2023, along with a notable jump in job postings and
investment in this technology unlocking innovative
investments. The pace of technology innovation has
possibilities across interconnected trends such as
been remarkable. Over the course of 2023 and 2024,
robotics and immersive reality. While the macroeconomic
the size of the prompts that large language models
environment with elevated interest rates has affected
(LLMs) can process, known as “context windows,” spiked
equity capital investment and hiring, underlying
from 100,000 to two million tokens. This is roughly the
indicators—including optimism, innovation, and longer-
difference between adding one research paper to a
term talent needs—reflect a positive long-term trajectory
model prompt and adding about 20 novels to it. And the
in the 15 technology trends we analyzed.
modalities that gen AI can process have continued to
These are among the findings in the latest McKinsey increase, from text summarization and image generation
Technology Trends Outlook, in which the McKinsey to advanced capabilities in video, images, audio, and text.
Technology Council identified the most significant This has catalyzed a surge in investments and innovation
technology trends unfolding today (to know more about aimed at advancing more powerful and efficient
the Council, see the sidebar “About the McKinsey computing systems.
Technology Council”). This research is intended to help
The large foundation models that power generative
executives plan ahead by developing an understanding
AI, such as LLMs, are being integrated into various
of potential use cases, sources of value, adoption drivers,
enterprise software tools and are also being employed
and the critical skills needed to bring these opportunities
for diverse purposes such as powering customer-facing
to fruition.
chatbots, generating ad campaigns, accelerating
Our analysis examines quantitative measures of drug discovery, and more. We expect this expansion
interest, innovation, investment, and talent to gauge the to continue, pushing the boundaries of AI capabilities.
momentum of each trend. Recognizing the long-term Senior leaders’ awareness of gen AI innovation has
nature and interdependence of these trends, we also increased interest, investment, and innovation in
delve into the underlying technologies, uncertainties, AI technologies and other trends, such as robotics,
and questions surrounding each trend. (For more about which is a new addition to our trends analysis this year.
new developments in our research, please see the Advancements in AI are ushering in a new era of more
sidebar “What’s new in this year’s analysis” on page 9; for capable robots, spurring greater innovation and a wider
more about the research itself, please see the sidebar range of deployments.
“Research methodology” on pages 10–11.)
About the McKinsey Technology Council
Technology is a catalyst for new opportunities, from inventing new products and services,
expanding the productivity frontier and capturing more value in our day-to-day work. The
McKinsey Technology Council helps business leaders understand frontier technologies and the
potential application to their businesses.
We look at a spectrum of technologies, from generative AI, machine learning, and quantum
computing to space technologies that are shaping new opportunities and applications. The McKinsey
Technology Council convenes a global group of more than 100 scientists, entrepreneurs, researchers,
and business leaders. We research, debate, and advise executives from all industries as they navigate the
fast-changing technology landscape.
—Lareina Yee, senior partner, McKinsey; chair, McKinsey Technology Council
Technology Trends Outlook 2024 4
−26% Electrification and renewables was the other learning solutions. Applied AI and
trend that bucked the economic headwinds, industrializing machine learning, boosted by
posting the highest investment and interest the widening interest in gen AI, have seen
scores among all the trends we evaluated. the most significant uptick in innovation,
tech trends job postings Job postings for this sector also showed a reflected in the surge in publications and
modest increase. patents from 2022 to 2023. Meanwhile,
from 2022 to 2023
electrification and renewable-energy
Although many trends faced declines in
technologies continue to capture high
investment and hiring in 2023, the long-term
interest, reflected in news mentions and
−17% outlook remains positive. This optimism is
web searches. Their popularity is fueled
supported by the continued longer-term
by a surge in global renewable capacity,
growth in job postings for the analyzed
their crucial roles in global decarbonization
trends (up 8 percent from 2021 to 2023)
efforts, and heightened energy security
global job postings and enterprises’ continued innovation and
needs amid geopolitical tensions and
from 2022 to 2023 heightened interest in harnessing these
energy crises.
technologies, particularly for future growth.
The talent environment largely echoed the
In 2023, technology equity investments
investment picture in tech trends in 2023.
+8% fell by 30 to 40 percent to approximately
The technology sector faced significant
$570 billion due to rising financing costs
layoffs, particularly among large technology
and a cautious near-term growth outlook,
companies, with job postings related to
prompting investors to favor technologies
the tech trends we studied declining by
tech trends job postings with strong revenue and margin potential.
26 percent—a steeper drop than the
This approach aligns with the strategic
from 2021 to 2023
17 percent decrease in global job postings
perspective leading companies are
overall. The greater decline in demand for
adopting, in which they recognize that
tech-trends-related talent may have been
fully adopting and scaling cutting-edge
fueled by technology companies’ cost
technologies is a long-term endeavor. This
reduction efforts amid decreasing revenue
recognition is evident when companies
growth projections. Despite this reduction,
diversify their investments across a
the trends with robust investment and
portfolio of several technologies, selectively
innovation, such as generative AI, not only
intensifying their focus on areas most likely
maintained but also increased their job
to push technological boundaries forward.
postings, reflecting a strong demand for
While many technologies have maintained
new and advanced skills. Electrification and
cautious investment profiles over the past
renewables was the other trend that saw
year, gen AI saw a sevenfold increase
positive job growth, partially due to public
in investments, driven by substantial
sector support for infrastructure spending.
advancements in text, image, and video
generation. Even with the short-term vicissitudes in
talent demand, our analysis of 4.3 million
Despite an overall downturn in private
job postings across our 15 tech trends
equity investment, the pace of innovation
underscored a wide skills gap. Compared
has not slowed. Innovation has accelerated
with the global average, fewer than half of
in the three trends that are part of the “AI
potential candidates have the high-demand
revolution” group: generative AI, applied AI,
tech skills specified in job postings. Despite
and industrializing machine learning. Gen
the year-on-year decreases for job postings
AI creates new content from unstructured
in many trends from 2022 to 2023, the
data (such as text and images), applied
number of tech-related job postings in 2023
AI leverages machine learning models
still represented an 8 percent increase from
for analytical and predictive tasks, and
2021, suggesting the potential for longer-
industrializing machine learning accelerates
term growth (Exhibit 1).
and derisks the development of machine
Technology Trends Outlook 2024 5
Exhibit 1
Despite a one-year drop in job postings, demand for jobs in many
technology trends has increased over two years.
Annual change in tech trend job postings, 2021–23, millions of postings¹
AI revolution Building the digital future Compute and connectivity Cutting-edge engineering A sustainable world
2021 2022 2023
1.4 1.4
1.2 1.2
1.0 1.0
0.8 +52% –37% 0.8
change
0.6 0.6
+33% –29%
0.4 0.4
+34% –11% –5%
+55% +1%
0.2 +72% 0.2
2021 2023
0 0
Next-generation Applied AI Climate technologies Future of mobility Electrification
software beyond electrification and renewables
development and renewables
0.6 0.6
0.4 0.4
++4499%% ––3344%% ++3399%% ––3388%% +32% –24% +55% –36%
0.2 ++7777%% ––3366%% 0.2
0 0
Digital trust and Cloud and edge Industrializing Advanced Immersive-reality
cybersecurity computing machine learning connectivity technologies
+6% –23% +29% –9% +110% +111% +29% –20% +44% –17%
0.2 0.2
0 0
Future of Future of space Generative Future of Quantum
bioengineering technologies AI robotics technologies
Cumulative change in tech trend job postings, 2021–23, millions of postings¹
2021 2023
1.0 1.0
–5% change
0.8 0.8
–6% +20% +48% +73%
0.6 0.6
0.4 0.4
0.2 0.2
2021 2023
0 0
Next-generation Applied AI Climate technologies Future of mobility Electrification
software beyond electrification and renewables
development and renewables
0.4 0.4
+0% –1%
0.2 –1% –14% +14% 0.2
0 0
Digital trust and Cloud and edge Industrializing Advanced Immersive-reality
cybersecurity computing machine learning connectivity technologies
–18% +18% +341% +3% +19%
0.2 0.2
0 0
Future of Future of space Generative Future of Quantum
bioengineering technologies AI robotics technologies
1Out of 130 million surveyed job postings (extrapolated Jan–Oct 2023). Job postings are not directly equivalent to numbers of new or existing jobs.
Source: McKinsey’s proprietary Organizational Data Platform, which draws on licensed, de-identified public professional profile data
McKinsey & Company
Technology Trends Outlook 2024 6
Enterprise technology
and space. Factors that could affect the adoption of
adoption momentum
these technologies include high costs, specialized
applications, and balancing the breadth of technology
The trajectory of enterprise technology adoption
investments against focusing on a select few that may
is often described as an S-curve that traces the
offer substantial first-mover advantages.
following pattern: technical innovation and exploration,
experimenting with the technology, initial pilots in As technologies gain traction and move beyond
the business, scaling the impact throughout the experimenting, adoption rates start accelerating, and
business, and eventual fully scaled adoption (Exhibit companies invest more in piloting and scaling. We see
2). This pattern is evident in this year’s survey analysis this shift in a number of trends, such as next-generation
of enterprise adoption conducted across our 15 software development and electrification. Gen AI’s rapid
technologies. Adoption levels vary across different advancement leads among trends analyzed, with about
industries and company sizes, as does the perceived a quarter of respondents self-reporting that they are
progress toward adoption. scaling its use. More mature technologies, like cloud
and edge computing and advanced connectivity,
We see that the technologies in the S-curve’s early
continued their rapid pace of adoption, serving
stages of innovation and experimenting are either
as enablers for the adoption of other emerging
on the leading edge of progress, such as quantum
technologies as well (Exhibit 3).
technologies and robotics, or are more relevant to
a specific set of industries, such as bioengineering
Web <2024>
<ETxehchibTrietn 2ds-L0>
Exhibit <3> of <3>
Technologies progress through different stages, with some at the leading
edge of innovation and others approaching large-scale adoption.
Adoption curve of technology trends, adoption score
Higher adoption
5 Fully scaled
4
Advanced connectivity
Applied AI
Cloud and edge computing
Generative AI
3
4 Scaling
Digital trust and cybersecurity
Electrification and renewables
Industrializing machine learning
Adoption
Next-gen software development
2
Climate technologies beyond
3 Piloting
electrification and renewables
Future of bioengineering¹
Future of mobility¹
2 Experimenting
1 Frontier Future of robotics¹
innovation Immersive-reality technologies
1
Future of space technologies¹
Quantum technologies
Lower adoption
¹Trend is more relevant to certain industries, resulting in lower overall adoption across industries compared with adoption within relevant industries.
Source: McKinsey technology adoption survey data; McKinsey analysis
McKinsey & Company
Technology Trends Outlook 2024 7
Web <2024>
<ETxehchibTrietn 3ds-L1>
Exhibit <2> of <3>
More-mature technologies are more widely adopted, often serving as
enablers for more-nascent technologies.
Self-reported adoption level by tech trend, 2023,1 % of respondents
Not investing Experimenting Piloting Scaling Fully scaled
Cloud and edge computing 25 14 13 26 22
Advanced connectivity 33 14 16 20 17
Generative AI2 26 18 20 26 10
Applied AI 26 18 21 24 11
Next-generation software development 37 14 18 23 8
Digital trust and cybersecurity 37 18 15 20 10
Electrification and renewables 37 17 19 20 7
Industrializing machine learning 37 16 20 19 8
Future of mobility 45 18 16 16 5
Climate technologies beyond electrification and
46 16 18 15 5
renewables
Immersive-reality technologies 43 18 20 15 4
Future of bioengineering 50 17 15 15 3
Future of robotics 41 22 19 13 5
Quantum technologies 47 18 20 15
Future of space technologies 57 15 13 12 3
1Respondents may interpret these categories differently based on their organizations. As such, the results should be considered as indicative of organizations’
self-assessments, rather than precise measurements. 2For a deeper look at our AI-related trends, see “The state of AI in early 2024: Gen AI adoption spikes
and starts to generate value,” McKinsey, May 30, 2024.
Source: McKinsey technology adoption survey data
McKinsey & Company
The process of scaling technology adoption also the external ecosystem conditions to ensure the
requires a conducive external ecosystem where user successful integration of new technologies into
trust and readiness, business model economics, their business models. Executives should monitor
regulatory environments, and talent availability play ecosystem conditions that can affect their prioritized
crucial roles. Since these ecosystem factors vary by use cases to make decisions about the appropriate
geography and industry, we see different adoption investment levels while navigating uncertainties and
scenarios playing out. For instance, while the leading budgetary constraints on the way to full adoption (see
banks in Latin America are on par with their North the “Adoption developments across the globe” sections
American counterparts in deploying gen AI use cases, within each trend that showcase examples of adoption
the adoption of robotics in manufacturing sectors varies dimensions for the trends or particular use cases therein
significantly due to differing labor costs affecting the that executives should monitor). Across the board,
business case for automation. leaders who take a long-term view—building up their
talent, testing and learning where impact can be found,
As executives navigate these complexities, they
and reimagining the businesses for the future—can
should align their long-term technology adoption
potentially break out ahead of the pack.
strategies with both their internal capacities and
Technology Trends Outlook 2024 8
The 15 tech trends
What’s new in this year’s analysis
This report lays out considerations for all 15 technology
This year, we reflected the shifts in the
trends. For easier consideration of related trends,
technology landscape with two changes on the
we grouped them into five broader categories: the AI
list of trends: digital trust and cybersecurity
revolution, building the digital future, compute and
(integrating what we had previously described
connectivity frontiers, cutting-edge engineering, and a
as Web3 and trust architectures) and the future
sustainable world. Of course, there’s significant power
of robotics. Robotics technologies’ synergy
and potential in looking across these groupings when
with AI is paving the way for groundbreaking
considering trend combinations.
innovations and operational shifts across the
To describe the state of each trend, we developed scores economic and workforce landscapes. We also
for innovation (based on patents and research) and deployed a survey to measure adoption levels
interest (based on news and web searches). We also across trends.
sized investments in relevant technologies and rated
their level of adoption by organizations (Exhibit 4).
Exhibit 4
Each trend is scored based on its level of innovation, interest, investment,
and adoption.
Innovation, interest, investment, and adoption, by technology trend, 2023
1.0
Adoption level, score
Applied AI (1 = frontier innovation;
5 = fully scaled)
0.8 1 2 3 4 5
Industrializing machine learning
Advanced connectivity
0.6
Future of bioengineering
Innovation,1 score
(0 = lower; Next-generation software development
1 = higher)
Cloud and edge computing
0.4 Immersive-reality technologies
Electrification/
Future of renewables
mobility
Climate technologies beyond electrification and renewables
0.2
Digital trust and cybersecurity Equity investment, $ billion
Future of robotics
Future of space technologies
Generative AI
Quantum technologies
250 150 75 20
0
0 1.00
0 0.2 0.4 0.6 0.8 1.0
Interest,2 score
(0 = lower; 1 = higher)
Note: Innovation and interest scores for the 15 trends are relative to one another. All 15 trends exhibit high levels of innovation and interest compared with
other topics and are also attracting significant investment.
1The innovation score combines the 0–1 scores for patents and research, which are relative to the trends studied. The patents score is based on a measure
of patent filings, and the research score is based on a measure of research publications.
2The interest score combines the 0–1 scores for news and searches, which are relative to the trends studied. The news score is based on a measure of news
publications, and the searches score is based on a measure of search engine queries.
McKinsey & Company
Technology Trends Outlook 2024 9
Research methodology
To assess the development of each technology trend, our team collected data on five tangible
measures of activity: search engine queries, news publications, patents, research publications,
and investment. For each measure, we used a defined set of data sources to find occurrences of
keywords associated with each of the 15 trends, screened those occurrences for valid mentions
of activity, and indexed the resulting numbers of mentions on a 0–1 scoring scale that is relative
to the trends studied. The innovation score combines the patents and research scores; the
interest score combines the news and search scores. (While we recognize that an interest
score can be inflated by deliberate efforts to stimulate news and search activity, we believe that
each score fairly reflects the extent of discussion and debate about a given trend.) Investment
measures the flows of funding from the capital markets into companies linked with the trend.
Data sources for the scores include the following:
— Patents. Data on patent filings are sourced from Google Patents, where the data highlight
the number of granted patents.
— Research. Data on research publications are sourced from Lens.
— News. Data on news publications are sourced from Factiva.
— Searches. Data on search engine queries are sourced from Google Trends.
— Investment. Data on private-market and public-market capital raises (venture capital and
corporate and strategic M&A, including joint ventures), private equity (including buyouts and
private investment in public equity), and public investments (including IPOs) are sourced from
PitchBook.
— Talent demand. Number of job postings is sourced from McKinsey’s proprietary
Organizational Data Platform, which stores licensed, de-identified data on professional
profiles and job postings. Data are drawn primarily from English-speaking countries.
In addition, we updated the selection and definition of trends from last year’s report to reflect
the evolution of technology trends:
— The future of robotics trend was added since last year’s publication.
— Data sources and keywords were updated. For data on the future of space technologies
investments, we used research from McKinsey’s Aerospace & Defense Practice.
Technology Trends Outlook 2024 10
Research methodology
(continued)
Finally, we used survey data to calculate the enterprise-wide adoption scores for each trend:
— Survey scope. The survey included approximately 1,000 respondents from 50 countries.
— Geographical coverage. Survey representation was balanced across Africa, Asia, Europe,
Latin America, the Middle East, and North America.
— Company size. Size categories, based on annual revenue, included small companies
($10 million to $50 million), medium-size companies ($50 million to $1 billion), and large
companies (greater than $1 billion).
— Respondent profile. The survey was targeted to senior-level professionals knowledgeable
in technology, who reported their perception of the extent to which their organizations were
using the technologies.
— Survey method. The survey was conducted online to enhance reach and accessibility.
— Question types. The survey employed multiple-choice and open-ended questions for
comprehensive insights.
— Definition of enterprise-wide adoption scores:
• 1: Frontier innovation. This technology is still nascent, with few organizations investing in or
applying it. It is largely untested and unproven in a business context.
• 2: Experimentation. Organizations are testing the functionality and viability of the
technology with a small-scale prototype, typically done without a strong focus on a near-
term ROI. Few companies are scaling or have fully scaled the technology.
• 3: Piloting. Organizations are implementing the technology for the first few business use
cases. It may be used in pilot projects or limited deployments to test its feasibility and
effectiveness.
• 4: Scaling. Organizations are in the process of scaling the deployment and adoption of the
technology across the enterprise. The technology is being scaled by a significant number
of companies.
• 5: Fully scaled. Organizations have fully deployed and integrated the technology across
the enterprise. It has become the standard and is being used at a large scale as companies
have recognized the value and benefits of the technology.
Technology Trends Outlook 2024 11
About the authors
Lareina Yee Michael Chui Roger Roberts Mena Issler
Senior partner, Bay Area; chair, McKinsey Global Institute Partner, Associate partner,
McKinsey Technology Council partner, Bay Area Bay Area Bay Area
The authors wish to thank the following McKinsey colleagues for their contributions to this research:
Aakanksha Srinivasan Carlo Giovine Joshua Katz Noah Furlonge-Walker
Ahsan Saeed Celine Crenshaw Julia Perry Obi Ezekoye
Alex Arutyunyants Daniel Herde Julian Sevillano Paolo Spranzi
Alex Singla Daniel Wallance Justin Greis Pepe Cafferata
Alex Zhang David Harvey Kersten Heineke Robin Riedel
Alizee Acket-Goemaere Delphine Zurkiya Kitti Lakner Ryan Brukardt
An Yan Diego Hernandez Diaz Kristen Jennings Samuel Musmanno
Anass Bensrhir Douglas Merrill Liz Grennan Santiago Comella-Dorda
Andrea Del Miglio Elisa Becker-Foss Luke Thomas Sebastian Mayer
Andreas Breiter Emma Parry Maria Pogosyan Shakeel Kalidas
Ani Kelkar Eric Hazan Mark Patel Sharmila Bhide
Anna Massey Erika Stanzl Martin Harrysson Stephen Xu
Anna Orthofer Everett Santana Martin Wrulich Tanmay Bhatnagar
Arjit Mehta Giacomo Gatto Martina Gschwendtner Thomas Hundertmark
Arjita Bhan Grace W Chen Massimo Mazza Tinan Goli
Asaf Somekh Hamza Khan Matej Macak Tom Brennan
Begum Ortaoglu Harshit Jain Matt Higginson Tom Levin-Reid
Benjamin Braverman Helen Wu Matt Linderman Tony Hansen
Bharat Bahl Henning Soller Matteo Cutrera Vinayak HV
Bharath Aiyer Ian de Bode Mellen Masea Yaron Haviv
Bhargs Srivathsan Jackson Pentz Michiel Nivard Yvonne Ferrier
Brian Constantine Jeffrey Caso Mike Westover Zina Cole
Brooke Stokes Jesse Klempner Musa Bilal
Bryan Richardson Jim Boehm Nicolas Bellemans
We appreciate the contributions of members of QuantumBlack, AI by McKinsey, to the insights on the AI-related trends.
They also wish to thank the external members of the McKinsey Technology Council for their insights and perspectives,
including Ajay Agrawal, Azeem Azhar, Ben Lorica, Benedict Evans, John Martinis, and Jordan Jacobs.
Technology Trends Outlook 2024 12
The AI revolution
Technology Trends Outlook 2024 13
Generative AI
The trend—and why it matters revolutionized, with models such as Suno creating original pieces
in various styles.
Generative AI (gen AI) has been making significant strides,
pushing the boundaries of machine capabilities. Gen AI models Gen AI has sparked widespread interest, with individuals and
are trained on vast, diverse data sets. They take unstructured organizations across different regions and industries exploring its
data, such as text, as inputs and produce unique outputs—also potential. According to the latest McKinsey Global Survey on the
in the form of unstructured data—ranging from text and code to state of AI, 65 percent of respondents say their organizations are
images, music, and 3D models. regularly using gen AI in at least one business function, up from
one-third last year,1 and gen AI use cases have the potential to
Over the past year, we’ve seen remarkable advancements in
generate an annual value of $2.6 trillion to $4.4 trillion.2
this field, with text generation models such as OpenAI’s GPT-4,
Anthropic’s Claude, and Google’s Gemini producing content that However, it’s important to recognize the risks that accompany the
mimics human-generated responses, as well as with image- use of this powerful technology, including bias, misinformation, and
generation tools such as DALL-E 3 and Midjourney creating deepfakes. As we progress through 2024 and beyond, we anticipate
photorealistic images from text descriptions. OpenAI’s recent organizations investing in the risk mitigation, operating model,
launch of Sora, a text-to-video generator, further showcases talent, and technological capabilities required to scale gen AI.
the technology’s potential. Even music composition is being
1 “The state of AI in early 2024: Gen AI adoption spikes and starts to generate value,” McKinsey, May 30, 2024.
2 The economic potential of generative AI: The next productivity frontier, McKinsey, June 14, 2023.
THE AI REVOLUTION Score by vector (0 = lower; 1 = higher)
Generative AI
Talent demand News
Scoring the trend
Gen AI saw a surge in 2023, driven by ChatGPT’s
late-2022 launch, alongside earlier models such as
DALL-E 2 and Stable Diffusion. Gen AI saw significant
growth from 2022 to 2023 across each quantitative
dimension, such as a sevenfold increase in the number Equity Searches
of searches and investments, reflecting a strong sense investment 0.2
0.4
of excitement about the trend.
0.6
0.8
Adoption score, 2023 1.0
Frontier Fully
innovation scaled
Patents Research
1 2 3 4 5
1.0
Equity investment, Job postings,
2023, 2022–23, 0
2019 2023
$ billion % difference
$36 +111%
Industries affected: Aerospace and defense;
Talent demand Ratio News Press reports
Agriculture; Automotive and assembly; Aviation,
of skilled people featuring trend-
travel, and logistics; Business, legal, and profes-
to job vacancies related phrases
sional services; Chemicals; Construction and
building materials; Consumer packaged goods; Equity investment Searches Search
Education; Electric power, natural gas, and utilities;
Private- and public- engine queries for
Financial services; Healthcare systems and
market capital raises for terms related to
services; Information technology and electronics;
relevant technologies trend
Media and entertainment; Metals and mining; Oil
and gas; Pharmaceuticals and medical products; Patents Patent Research Scientific
Public and social sectors; Real estate; Retail; filings for technologies publications on topics
Semiconductors; Telecommunications
related to trend associated with trend
Technology Trends Outlook 2024 14
Latest developments — LLMs are increasingly being embedded into various
enterprise tools. We are witnessing a significant uptick
Gen AI is a fast-growing and constantly innovating trend,
in the integration of LLMs into various enterprise
with recent developments including the following:
tools. This surge is fueled by the growing demand for
— Multimodal generative models are on the rise. As automation, efficiency, personalized user experiences,
gen AI continues to evolve and gain more attention in and the capacity to decipher complex patterns that
various industries, it’s becoming increasingly clear that can lead to actionable insights. Consequently, a rising
multimodality will play a pivotal role. By combining text, number of vendors are choosing to integrate LLMs into
images, sounds, and videos, AI models can generate their applications and tools. This trend is especially
outputs applicable across a wide range of industries prominent in the marketing and customer care domains,
and business functions. This pursuit of multimodality with Salesforce Einstein and ServiceNow serving as
is intensifying across leading players such as OpenAI prime examples.
and Google (with its Lumiere AI web app). For example,
— The multiagent approach has gained significant traction
Google’s Gemini showcases a powerful multimodal
with the rapid development of LLMs and continued
system capable of processing information in various
innovation. Companies now recognize the benefits
formats, including text, code, tables, images, and
of employing multiple language models that work in
even audio.
harmony rather than relying on a single model. This
— Powerful open-source models are challenging their approach offers a fresh perspective on tackling complex
closed-source counterparts in performance and challenges by leveraging the capabilities of multiple AI
developer adoption. While significant investments agents, each specializing in different domains, to solve
are encouraging the development of proprietary large a single problem collaboratively. By working together,
language models (LLMs), such as GPT-4 with vision these agents can not only accelerate problem-solving
(GPT-4V), the AI community is also witnessing a surge in but also leverage varied perspectives and expertise to
open-source models, such as Llama 3. This momentum deliver more effective and efficient solutions. Some
is fueled by the enthusiasm of developers and users of the tools using this approach tend to be unstable,
who welcome the unprecedented access to build but as models improve, their throughput s |
262 | mckinsey | the-economic-potential-of-generative-ai-the-next-productivity-frontier.pdf | The
economic
potential
of
generative
AI
The economic
potential of
generative AI
The next productivity frontier
June 2023
Authors
Michael Chui
Eric Hazan
Roger Roberts
Alex Singla
Kate Smaje
Alex Sukharevsky
Lareina Yee
Rodney Zemmel
ii The economic potential of generative AI: The next productivity frontier
Contents
Key insights Spotlight: Pharmaceuticals
3 and medical products
30
Chapter 1: Generative AI
as a technology catalyst Chapter 3: The generative
4 AI future of work: Impacts
on work activities, economic
Glossary growth, and productivity
6 32
Chapter 2: Generative AI use Chapter 4: Considerations
cases across functions and for businesses and society
industries 48
8
Appendix
Spotlight: Retail and 53
consumer packaged goods
27
Spotlight: Banking
28
The economic potential of generative AI: The next productivity frontier 1
2 The economic potential of generative AI: The next productivity frontier
Key insights
1. Generative AI’s impact on equal to an additional $200 billion 6. Generative AI can substantially
productivity could add trillions to $340 billion annually if the use increase labor productivity across
of dollars in value to the global cases were fully implemented. In the economy, but that will require
economy. Our latest research retail and consumer packaged investments to support workers
estimates that generative AI could goods, the potential impact is also as they shift work activities or
add the equivalent of $2.6 trillion significant at $400 billion to $660 change jobs. Generative AI could
to $4.4 trillion annually across the billion a year. enable labor productivity growth
63 use cases we analyzed—by of 0.1 to 0.6 percent annually
comparison, the United Kingdom’s 4. Generative AI has the potential through 2040, depending on the
entire GDP in 2021 was $3.1 trillion. to change the anatomy of work, rate of technology adoption and
This would increase the impact of augmenting the capabilities of redeployment of worker time
all artificial intelligence by 15 to individual workers by automating into other activities. Combining
40 percent. This estimate would some of their individual activities. generative AI with all other
roughly double if we include the Current generative AI and other technologies, work automation
impact of embedding generative AI technologies have the potential to could add 0.5 to 3.4 percentage
into software that is currently used automate work activities that absorb points annually to productivity
for other tasks beyond those use 60 to 70 percent of employees’ time growth. However, workers will need
cases. today. In contrast, we previously support in learning new skills, and
estimated that technology has the some will change occupations. If
2. About 75 percent of the value that potential to automate half of the worker transitions and other risks
generative AI use cases could time employees spend working.1 can be managed, generative AI
deliver falls across four areas: The acceleration in the potential for could contribute substantively to
Customer operations, marketing technical automation is largely due economic growth and support a
and sales, software engineering, to generative AI’s increased ability more sustainable, inclusive world.
and R&D. Across 16 business to understand natural language,
functions, we examined 63 use which is required for work activities 7. The era of generative AI is just
cases in which the technology that account for 25 percent of total beginning. Excitement over this
can address specific business work time. Thus, generative AI has technology is palpable, and early
challenges in ways that produce more impact on knowledge work pilots are compelling. But a full
one or more measurable outcomes. associated with occupations that realization of the technology’s
Examples include generative AI’s have higher wages and educational benefits will take time, and leaders
ability to support interactions requirements than on other types in business and society still
with customers, generate creative of work. have considerable challenges to
content for marketing and sales, address. These include managing
and draft computer code based on 5. The pace of workforce the risks inherent in generative
natural-language prompts, among transformation is likely to AI, determining what new skills
many other tasks. accelerate, given increases in the and capabilities the workforce will
potential for technical automation. need, and rethinking core business
3. Generative AI will have a significant Our updated adoption scenarios, processes such as retraining and
impact across all industry sectors. including technology development, developing new skills.
Banking, high tech, and life economic feasibility, and diffusion
sciences are among the industries timelines, lead to estimates that
that could see the biggest impact half of today’s work activities could
as a percentage of their revenues be automated between 2030 and
from generative AI. Across the 2060, with a midpoint in 2045, or
banking industry, for example, the roughly a decade earlier than in our
technology could deliver value previous estimates.
The economic potential of generative AI: The next productivity frontier 3
1
Generative AI as a
technology catalyst
To grasp what lies ahead requires an understanding of the breakthroughs that have enabled
the rise of generative AI, which were decades in the making. ChatGPT, GitHub Copilot, Stable
Diffusion, and other generative AI tools that have captured current public attention are the
result of significant levels of investment in recent years that have helped advance machine
learning and deep learning. This investment undergirds the AI applications embedded in many
of the products and services we use every day.
But because AI has permeated our lives incrementally—through everything from the tech
powering our smartphones to autonomous-driving features on cars to the tools retailers use
to surprise and delight consumers—its progress was almost imperceptible. Clear milestones,
such as when AlphaGo, an AI-based program developed by DeepMind, defeated a world
champion Go player in 2016, were celebrated but then quickly faded from the public’s
consciousness.
ChatGPT and its competitors have captured the imagination of people around the world
in a way AlphaGo did not, thanks to their broad utility—almost anyone can use them to
communicate and create—and preternatural ability to have a conversation with a user.
The latest generative AI applications can perform a range of routine tasks, such as the
reorganization and classification of data. But it is their ability to write text, compose music,
and create digital art that has garnered headlines and persuaded consumers and households
to experiment on their own. As a result, a broader set of stakeholders are grappling with
generative AI’s impact on business and society but without much context to help them make
sense of it.
4 The economic potential of generative AI: The next productivity frontier
How did we get here? Gradually, then all of a sudden
For the purposes of this report, we define generative AI as applications typically built using
foundation models. These models contain expansive artificial neural networks inspired by the
billions of neurons connected in the human brain. Foundation models are part of what is called
deep learning, a term that alludes to the many deep layers within neural networks. Deep
learning has powered many of the recent advances in AI, but the foundation models powering
generative AI applications are a step change evolution within deep learning. Unlike previous
deep learning models, they can process extremely large and varied sets of unstructured data
and perform more than one task.
Foundation models have enabled new capabilities and vastly improved existing ones across
a broad range of modalities, including images, video, audio, and computer code. AI trained
on these models can perform several functions; it can classify, edit, summarize, answer
questions, and draft new content, among other tasks.
Continued innovation will also bring new challenges. For example, the computational power
required to train generative AI with hundreds of billions of parameters threatens to become a
bottleneck in development.2 Further, there’s a significant move—spearheaded by the open-
source community and spreading to the leaders of generative AI companies themselves—to
make AI more responsible, which could increase its costs.
Nonetheless, funding for generative AI, though still a fraction of total investments in artificial
intelligence, is significant and growing rapidly—reaching a total of $12 billion in the first five
months of 2023 alone. Venture capital and other private external investments in generative
AI increased by an average compound growth rate of 74 percent annually from 2017 to 2022.
During the same period, investments in artificial intelligence overall rose annually by 29
percent, albeit from a higher base.
The rush to throw money at all things generative AI reflects how quickly its capabilities have
developed. ChatGPT was released in November 2022. Four months later, OpenAI released
a new large language model, or LLM, called GPT-4 with markedly improved capabilities.3
Similarly, by May 2023, Anthropic’s generative AI, Claude, was able to process 100,000
tokens of text, equal to about 75,000 words in a minute—the length of the average novel—
compared with roughly 9,000 tokens when it was introduced in March 2023.4 And in May
2023, Google announced several new features powered by generative AI, including Search
Generative Experience and a new LLM called PaLM 2 that will power its Bard chatbot, among
other Google products.5
From a geographic perspective, external private investment in generative AI, mostly from
tech giants and venture capital firms, is largely concentrated in North America, reflecting the
continent’s current domination of the overall AI investment landscape. Generative AI–related
companies based in the United States raised about $8 billion from 2020 to 2022, accounting
for 75 percent of total investments in such companies during that period.6
Generative AI has stunned and excited the world with its potential for reshaping how
knowledge work gets done in industries and business functions across the entire economy.
Across functions such as sales and marketing, customer operations, and software
development, it is poised to transform roles and boost performance. In the process, it could
unlock trillions of dollars in value across sectors from banking to life sciences. We have used
two overlapping lenses in this report to understand the potential for generative AI to create
value for companies and alter the workforce. The following sections share our initial findings.
The economic potential of generative AI: The next productivity frontier 5
Glossary
Application programming interface (API) is a way to programmatically access (usually
external) models, data sets, or other pieces of software.
Artificial intelligence (AI) is the ability of software to perform tasks that traditionally require
human intelligence.
Artificial neural networks (ANNs) are composed of interconnected layers of software-based
calculators known as “neurons.” These networks can absorb vast amounts of input data and
process that data through multiple layers that extract and learn the data’s features.
Deep learning is a subset of machine learning that uses deep neural networks, which are
layers of connected “neurons” whose connections have parameters or weights that can be
trained. It is especially effective at learning from unstructured data such as images, text, and
audio.
Early and late scenarios are the extreme scenarios of our work-automation model. The
“earliest” scenario flexes all parameters to the extremes of plausible assumptions, resulting in
faster automation development and adoption, and the “latest” scenario flexes all parameters
in the opposite direction. The reality is likely to fall somewhere between the two.
Fine-tuning is the process of adapting a pretrained foundation model to perform better in
a specific task. This entails a relatively short period of training on a labeled data set, which
is much smaller than the data set the model was initially trained on. This additional training
allows the model to learn and adapt to the nuances, terminology, and specific patterns found
in the smaller data set.
Foundation models (FM) are deep learning models trained on vast quantities of
unstructured, unlabeled data that can be used for a wide range of tasks out of the box or
adapted to specific tasks through fine-tuning. Examples of these models are GPT-4, PaLM,
DALL·E 2, and Stable Diffusion.
Generative AI is AI that is typically built using foundation models and has capabilities that
earlier AI did not have, such as the ability to generate content. Foundation models can also
be used for nongenerative purposes (for example, classifying user sentiment as negative or
positive based on call transcripts) while offering significant improvement over earlier models.
For simplicity, when we refer to generative AI in this article, we include all foundation model
use cases.
Graphics processing units (GPUs) are computer chips that were originally developed for
producing computer graphics (such as for video games) and are also useful for deep learning
applications. In contrast, traditional machine learning and other analyses usually run on
central processing units (CPUs), normally referred to as a computer’s “processor.”
Large language models (LLMs) make up a class of foundation models that can process
massive amounts of unstructured text and learn the relationships between words or portions
of words, known as tokens. This enables LLMs to generate natural-language text, performing
tasks such as summarization or knowledge extraction. GPT-4 (which underlies ChatGPT) and
LaMDA (the model behind Bard) are examples of LLMs.
6 The economic potential of generative AI: The next productivity frontier
Machine learning (ML) is a subset of AI in which a model gains capabilities after it is trained
on, or shown, many example data points. Machine learning algorithms detect patterns and
learn how to make predictions and recommendations by processing data and experiences,
rather than by receiving explicit programming instruction. The algorithms also adapt and can
become more effective in response to new data and experiences.
Modality is a high-level data category such as numbers, text, images, video, and audio.
Productivity from labor is the ratio of GDP to total hours worked in the economy. Labor
productivity growth comes from increases in the amount of capital available to each worker,
the education and experience of the workforce, and improvements in technology.
Prompt engineering refers to the process of designing, refining, and optimizing input
prompts to guide a generative AI model toward producing desired (that is, accurate) outputs.
Self-attention, sometimes called intra-attention, is a mechanism that aims to mimic cognitive
attention, relating different positions of a single sequence to compute a representation of the
sequence.
Structured data are tabular data (for example, organized in tables, databases, or
spreadsheets) that can be used to train some machine learning models effectively.
Transformers are a relatively new neural network architecture that relies on self-attention
mechanisms to transform a sequence of inputs into a sequence of outputs while focusing its
attention on important parts of the context around the inputs. Transformers do not rely on
convolutions or recurrent neural networks.
Technical automation potential refers to the share of the worktime that could be automated.
We assessed the technical potential for automation across the global economy through
an analysis of the component activities of each occupation. We used databases published
by institutions including the World Bank and the US Bureau of Labor Statistics to break
down about 850 occupations into approximately 2,100 activities, and we determined the
performance capabilities needed for each activity based on how humans currently perform
them.
Use cases are targeted applications to a specific business challenge that produces one
or more measurable outcomes. For example, in marketing, generative AI could be used to
generate creative content such as personalized emails.
Unstructured data lack a consistent format or structure (for example, text, images, and audio
files) and typically require more advanced techniques to extract insights.
The economic potential of generative AI: The next productivity frontier 7
2
Generative AI use
cases across functions
and industries
Generative AI is a step change in the evolution of artificial intelligence. As companies
rush to adapt and implement it, understanding the technology’s potential to deliver value
to the economy and society at large will help shape critical decisions. We have used two
complementary lenses to determine where generative AI with its current capabilities could
deliver the biggest value and how big that value could be (Exhibit 1).
8 The economic potential of generative AI: The next productivity frontier
Exhibit 1
The potential impact of generative AI can be evaluated through two lenses.
Lens 1 Lens 2
Total economic Labor productivity potential
potential of 60-plus across ~2,100 detailed work
organizational use activities performed by
cases1 global workforce
Cost impacts
of use cases
Revenue
impacts of
use cases1
1For quantitative analysis, revenue impacts were recast as productivity increases on the corresponding spend in order to maintain comparability with cost
impacts and not to assume additional growth in any particular market.
McKinsey & Company
The first lens scans use cases for generative AI that organizations could adopt. We define
a “use case” as a targeted application of generative AI to a specific business challenge,
resulting in one or more measurable outcomes. For example, a use case in marketing is the
application of generative AI to generate creative content such as personalized emails, the
measurable outcomes of which potentially include reductions in the cost of generating such
content and increases in revenue from the enhanced effectiveness of higher-quality content
at scale. We identified 63 generative AI use cases spanning 16 business functions that could
deliver total value in the range of $2.6 trillion to $4.4 trillion in economic benefits annually
when applied across industries.
That would add 15 to 40 percent to the $11.0 trillion to $17.7 trillion of economic value that we
now estimate nongenerative artificial intelligence and analytics could unlock. (Our previous
estimate from 2017 was that AI could deliver $9.5 trillion to $15.4 trillion in economic value.)
Our second lens complements the first by analyzing generative AI’s potential impact on
the work activities required in some 850 occupations. We modeled scenarios to estimate
when generative AI could perform each of more than 2,100 “detailed work activities”—
such as “communicating with others about operational plans or activities”—that make up
those occupations across the world economy. This enables us to estimate how the current
capabilities of generative AI could affect labor productivity across all work currently done by
the global workforce.
The economic potential of generative AI: The next productivity frontier 9
Some of this impact will overlap with cost reductions in the use case analysis described
above, which we assume are the result of improved labor productivity. Netting out this
overlap, the total economic benefits of generative AI—including the major use cases we
explored and the myriad increases in productivity that are likely to materialize when the
technology is applied across knowledge workers’ activities—amounts to $6.1 trillion to
$7.9 trillion annually (Exhibit 2).
Exhibit 2
Generative AI could create additional value potential above what
could be unlocked by other AI and analytics.
AI’s potential impact on the global economy, $ trillion
17.1–25.6
13.6–22.1
6.1–7.9
2.6–4.4
11.0–17.7
~15–40% ~35–70%
incremental incremental
economic impact economic impact
Advanced analytics, New generative Total use All worker productivity Total AI
traditional machine AI use cases case-driven enabled by generative economic
learning, and deep potential AI, including in use potential
learning1 cases
1Updated use case estimates from "Notes from the AI frontier: Applications and value of deep learning,” McKinsey Global Institute, April 17, 2018.
McKinsey & Company
10 The economic potential of generative AI: The next productivity frontier
While generative AI is an exciting and rapidly advancing technology, the other applications of
AI discussed in our previous report continue to account for the majority of the overall potential
value of AI. Traditional advanced-analytics and machine learning algorithms are highly
effective at performing numerical and optimization tasks such as predictive modeling, and
they continue to find new applications in a wide range of industries. However, as generative AI
continues to develop and mature, it has the potential to open wholly new frontiers in creativity
and innovation. It has already expanded the possibilities of what AI overall can achieve (please
see Box 1, “How we estimated the value potential of generative AI use cases”).
Box 1
How we estimated the value potential of generative AI use cases
To assess the potential value of generative AI, a customer service use case but not in a use
we updated a proprietary McKinsey database of case optimizing a logistics network, where value
potential AI use cases and drew on the experience primarily arises from quantitative analysis.
of more than 100 experts in industries and their
We then estimated the potential annual value
business functions.1 Our updates examined
of these generative AI use cases if they were
use cases of generative AI—specifically, how
adopted across the entire economy. For use
generative AI techniques (primarily transformer-
cases aimed at increasing revenue, such as some
based neural networks) can be used to solve
of those in sales and marketing, we estimated
problems not well addressed by previous
the economy-wide value generative AI could
technologies.
deliver by increasing the productivity of sales and
We analyzed only use cases for which generative marketing expenditures.
AI could deliver a significant improvement in the
Our estimates are based on the structure of the
outputs that drive key value. In particular, our
global economy in 2022 and do not consider the
estimates of the primary value the technology
value generative AI could create if it produced
could unlock do not include use cases for which
entirely new product or service categories.
the sole benefit would be its ability to use natural
language. For example, natural-language
capabilities would be the key driver of value in
1 “Notes from the AI frontier: Applications and value of deep learning,” McKinsey Global Institute, April 17, 2018.
In this chapter, we highlight the value potential of generative AI across two dimensions:
business function and modality.
The economic potential of generative AI: The next productivity frontier 11
Value potential by function
While generative AI could have an impact on most business functions, a few stand out when
measured by the technology’s impact as a share of functional cost (Exhibit 3). Our analysis
of 16 business functions identified just four—customer operations, marketing and sales,
software engineering, and research and development—that could account for approximately
75 percent of the total annual value from generative AI use cases.
Web <2023>
E<Vxihvaitbeicth 3 full report>
Exhibit <3> of <16>
Using generative AI in just a few functions could drive most of the technology’s
impact across potential corporate use cases.
Represent ~75% of total annual impact of generative AI
500
Sales
Software engineering
Marketing
(for corporate IT)
Software engineering
(for product development)
400
Customer operations
Product R&D1
300
Impact, $ billion
Supply chain
200
Manufacturing
Finance Risk and compliance
Talent and organization (incl HR)
100
Procurement management
Corporate IT1 Legal
Strategy
Pricing
0
0 10 20 30 40
Impact as a percentage of functional spend, %
Note: Impact is averaged.
¹Excluding software engineering.
Source: Comparative Industry Service (CIS), IHS Markit; Oxford Economics; McKinsey Corporate and Business Functions database; McKinsey Manufacturing
and Supply Chain 360; McKinsey Sales Navigator; Ignite, a McKinsey database; McKinsey analysis
McKinsey & Company
Notably, the potential value of using generative AI for several functions that were prominent
in our previous sizing of AI use cases, including manufacturing and supply chain functions,
is now much lower.7 This is largely explained by the nature of generative AI use cases, which
exclude most of the numerical and optimization applications that were the main value drivers
for previous applications of AI.
12 The economic potential of generative AI: The next productivity frontier
Generative AI as a virtual expert
In addition to the potential value generative AI can deliver in function-specific use cases,
the technology could drive value across an entire organization by revolutionizing internal
knowledge management systems. Generative AI’s impressive command of natural-language
processing can help employees retrieve stored internal knowledge by formulating queries
in the same way they might ask a human a question and engage in continuing dialogue. This
could empower teams to quickly access relevant information, enabling them to rapidly make
better-informed decisions and develop effective strategies.
In 2012, the McKinsey Global Institute (MGI) estimated that knowledge workers spent about
a fifth of their time, or one day each work week, searching for and gathering information. If
generative AI could take on such tasks, increasing the efficiency and effectiveness of the
workers doing them, the benefits would be huge. Such virtual expertise could rapidly “read”
vast libraries of corporate information stored in natural language and quickly scan source
material in dialogue with a human who helps fine-tune and tailor its research, a more scalable
solution than hiring a team of human experts for the task.
Following are examples of how generative AI could produce operational benefits as a virtual
expert in a handful of use cases.
In addition to the potential
value generative AI can
deliver in specific use
cases, the technology
could drive value across
an entire organization
by revolutionizing
internal knowledge
management systems.
The economic potential of generative AI: The next productivity frontier 13
How customer operations
could be transformed
Customer self-service interactions
Customer interacts with a humanlike chatbot that
delivers immediate, personalized responses to
complex inquiries, ensuring a consistent brand
voice regardless of customer language or location.
Customer–agent interactions
Human agent uses AI-developed call scripts and
receives real-time assistance and suggestions for
responses during phone conversations, instantly
accessing relevant customer data for tailored and
real-time information delivery.
Agent self-improvement
Agent receives a summarization of the conversation in
a few succinct points to create a record of customer
complaints and actions taken.
Agent uses automated, personalized insights generated
by AI, including tailored follow-up messages or
personalized coaching suggestions.
14 The economic potential of generative AI: The next productivity frontier
Customer operations
Generative AI has the potential to revolutionize the entire customer operations function,
improving the customer experience and agent productivity through digital self-service
and enhancing and augmenting agent skills. The technology has already gained traction
in customer service because of its ability to automate interactions with customers using
natural language. Research found that at one company with 5,000 customer service
agents, the application of generative AI increased issue resolution by 14 percent an hour and
reduced the time spent handling an issue by 9 percent.8 It also reduced agent attrition and
requests to speak to a manager by 25 percent. Crucially, productivity and quality of service
improved most among less-experienced agents, while the AI assistant did not increase—
and sometimes decreased—the productivity and quality metrics of more highly skilled
agents. This is because AI assistance helped less-experienced agents communicate using
techniques similar to those of their higher-skilled counterparts.
The following are examples of the operational improvements generative AI can have for
specific use cases:
— Customer self-service. Generative AI–fueled chatbots can give immediate and
personalized responses to complex customer inquiries regardless of the language or
location of the customer. By improving the quality and effectiveness of interactions via
automated channels, generative AI could automate responses to a higher percentage of
customer inquiries, enabling customer care teams to take on inquiries that can only be
resolved by a human agent. Our research found that roughly half of customer contacts
made by banking, telecommunications, and utilities companies in North America are
already handled by machines, including but not exclusively AI. We estimate that generative
AI could further reduce the volume of human-serviced contacts by up to 50 percent,
depending on a company’s existing level of automation.
— Resolution during initial contact. Generative AI can instantly retrieve data a company
has on a specific customer, which can help a human customer service representative more
successfully answer questions and resolve issues during an initial interaction.
— Reduced response time. Generative AI can cut the time a human sales representative
spends responding to a customer by providing assistance in real time and recommending
next steps.
— Increased sales. Because of its ability to rapidly process data on customers and their
browsing histories, the technology can identify product suggestions and deals tailored
to customer preferences. Additionally, generative AI can enhance quality assurance and
coaching by gathering insights from customer conversations, determining what could be
done better, and coaching agents.
We estimate that applying generative AI to customer care functions could increase
productivity at a value ranging from 30 to 45 percent of current function costs.
Our analysis captures only the direct impact generative AI might have on the productivity of
customer operations. It does not account for potential knock-on effects the technology may
have on customer satisfaction and retention arising from an improved experience, including
better understanding of the customer’s context that can assist human agents in providing
more personalized help and recommendations.
The economic potential of generative AI: The next productivity frontier 15
How marketing and sales
could be transformed
Strategization
Sales and marketing professionals efficiently
gather market trends and customer information
from unstructured data sources (for example,
social media, news, research, product information,
and customer feedback) and draft effective
marketing and sales communications.
Awareness
Customers see campaigns tailored
to their segment, language, and
demographic.
Consideration
Customers can access comprehensive information,
comparisons, and dynamic recommendations, such as
personal “try ons.”
16 The economic potential of generative AI: The next productivity frontier
Conversion
Virtual sales representatives enabled by generative
AI emulate humanlike qualities—such as empathy,
personalized communication, and natural language
processing—to build trust and rapport with
customers.
Retention
Customers are more likely to be retained with
customized messages and rewards, and they can
interact with AI-powered customer-support chatbots
that manage the relationship proactively, with fewer
escalations to human agents.
Marketing and sales
Generative AI has taken hold rapidly in marketing and sales functions, in which text-based
communications and personalization at scale are driving forces. The technology can create
personalized messages tailored to individual customer interests, preferences, and behaviors,
as well as do tasks such as producing first drafts of brand advertising, headlines, slogans,
social media posts, and product descriptions.
However, introducing generative AI to marketing functions requires careful cons |
263 | accenture | Accenture-Mondelez-International-Abridged-Transcript-FINAL.pdf | HOW MONDELĒZ
INTERNATIONAL USES
DATA AND AI TO
TRANSFORM THEIR
ENTERPRISE
VIDEO TRANSCRIPT
Venky Rao (00:23): Javier Polit (01:22):
What was the impetus for Mondelēz to The starting point is really spending time with
becoming more of a data driven data led the business, Venky, and really understanding,
company? when I first joined, I was listening and learning.
I'm still learning. I've only been with the
Javier Polit (00:28): company two and a half years or so now, but
I think it all starts with our goal at Mondelēz really understanding what the pain points were.
International is to be the leader of snacking. And And, and it really wasn't about just filling one
we need an even stronger growth strategy to hole. It was really trying to understand the
keep up with the pace, and even influence, our holistic opportunities that we had. And once we
consumer demand and our consumer behavior. had that defined, it was building that vision and
So, we really started to focus on a relentless that strategy and making certain that you got
consumer centricity in making certain that we support of the strategy by the C-suite, which
started to aggregate 360-degree insights of, of we did. And the executive team and the board
our consumers. And the time was right because were all behind us, and we started
we had been preparing from a business communicating that strategy to the enterprise.
perspective and also from a technology And that required a lot of work for us to do and
perspective. We had the right foundation in say we need to start investing in our people
place. The company was on a cloud strategy elevating capabilities, looking at the strategic
when I joined here, multi-cloud strategy, we partners that we were going to use, right?
brought in the Google Cloud platform. So, we Besides the Accenture’s and, and the
had that behind us, and the team was doing Microsoft’s and the Googles of the world and
some great work before I joined, and we finished other strategic partners. How we're going to
that work and, it gave us really time to pivot and firmly have the conversation that we had with
really start focusing on data and AI. you and all the other partners, bring your best
to us as we're continuing to try to be the best
Venky Rao (01:18): that we can and leverage partners as we're
How do you start a journey like that? Where do trying to build capability inside the enterprise
you start from?
and, and driving change in the enterprise as so that the whole enterprise feels good about
well. From a behavioral and work perspective. the work that you're trying to drive. And they
understand that there's a sense of importance
Venky Rao (02:23):
and urgency to what you're doing. And, and
You know, there is this old saying, you know,
when I talk about communicating a lot of, I have
what you track and what you measure actually
about 30 touch points with my organization on
gets done. So, when you get on a
an annual basis, and we talk about these things.
transformation journey like this, especially in this
And then the last thing is making certain that
space of AI, AI enabled, which is all new, how
do you start measuring success? you have a core strategic central AI data
science team that's really helping the
Javier Polit (02:37): organization. You can't have these silos in the
Yeah. Well, you know, you can't manage what enterprise where they're going on and building
you don't measure. We've all heard that, that their own data science and data strategies
nomenclature, I can tell you that we've had
without understanding that there's a holistic data
some really, really good maturity here over the
driven strategy that all that data needs to come
last 18 to 24 months in regards to how we track,
together, and somebody needs to be the
how we measure the ROI’s on work that we're
steward of that. And monitoring is the data
delivering to the, to the business and the value
inside the enterprise is data outside the
based on the business case that we initially put
together for the business. And, and through that enterprise what data needs to ingress or egress
work, what we're seeing now in all our business from different sources? And you just can't have
reviews that we have conversation around that working in a silo. So, I would say it's
digital comes up the work that's being driven probably those three dimensions.
around digital. And with that we talk about data
science, and we talk about the AI work that
Venky Rao (04:39):
that's being done, right? You know, you set a
So Javier, how do you see talent and tech
transformation strategy and a vision, and you
working together to achieve the Mondelēz
say, okay, it's a three-year horizon. I always say
vision?
that after the second year, you start figuring out
what your next three-year horizon's going to be.
So, it's, it's something that is just never done. Javier Polit (04:45):
Venky it’s just continuous work. Yeah, there, there's a lot of dimensions to that.
And I will tell you that, you know, we win with
Venky Rao (03:25): our people. Our people are our greatest asset.
Now, having said that, what are the most And we invest in our people in many different
important factors in making a transformation
ways and our people are critical to anything we
successful?
change or anything we make, you know, our
success is possible because it's 79,000
Javier Polit (03:31):
incredible colleagues that we have around the
Well, I think when, when you think about a
world. And some of the things that we're doing
transformation in any large enterprise, and I've
right now is as we continue to drive the
had the opportunity to do this a couple times, is
you need to have the right sponsorship. You importance of being data driven enterprise and
know, once you develop that strategy and that have an innovative culture, we're able to make
vision, making certain that the board, the those pivots and become a dynamic
executive team is behind it, and then you need organization. . We talk about being a dynamic
to communicate as much as possible and learning organization, right?
communicate that strategy and what you're
trying to do, and communicate the sponsorship
Where we, we are not a knowing culture, we're a and, and a brilliant one at that. So Javier, where
learning culture and we want to continue to do you see the whole AI Adoption space
innovate and take risk. And I think, you know, all evolving to?
that's done through sound leadership, but, but
it's also having the right partners at the table, Javier Polit (07:13):
right? And we firmly encourage our partners,
When you think about the evolution of AI today,
whether it's Accenture or whether it's Google or
companies are using narrow AI, right? It's taking
whether it's Microsoft and many others, to bring
the ability to have a human process be
the best that you have. And we've had those
conducted through AI with greater efficiency.
conversations too, bring the best to us and make
And you have companies that are adopting that
certain that we could really partner and do some
really industry leading things, right? So, it's really well, that are the 12% AI achievers and those
not something that you could do on your own, that are falling, following and, and a little bit
but you have to have a pool of experts inside the behind. And then the next level of AI is general
enterprise as well as the experts that your AI or human AI, where you have artificial
partners bring as well. intelligence that can basically do what a human
thinks. And the more complicated AI that's going
Venky Rao (06:01):
to be happening in the future, and it's happening
So Javier, how would you assess, CPG industry
in different parts of the world today, is super AI,
in terms of AI maturity compared to other
where AI can now do things better and in a
industries? I mean, especially, I know that you
smarter way than humans can. So, it's going to
look at the tech sector quite a bit, and, and you
take a lot of inspiration from how some of the big be an evolving space. We'll have to see how
technology companies operate. But how do you those technologies, when they come to be
see that evolving in the CPG industry? commonplace are going to be leveraged in in
different industries.
And, and they're already starting to be used in, in
Javier Polit (06:21): certain industries.
Yeah, you know, we're continuously doing
industry sensing in that space and see how we
Venky Rao (08:05):
match up to other CPG companies or fast-
So in closing, any thoughts, Javier as we wrap
moving consumer goods companies. But I think
up.
it's fair to say that the tech sector is still far
ahead. But I would also say in the same breath
that I think that the gap is narrowing and Javier Polit (08:08):
especially I think what's, what's helped us narrow You know, there’s probably an abundance of
that gap that that gap is companies really thoughts and because I think we'd all agree that
advancing their digital roadmaps in the digital it's a complicated space, but I think there's
plans, right? So, I think, there's enormous room
maybe six pillars of an AI strategy, right? And I
for growth in AI Adoption and AI Adoption across
always say start with the business value, right?
all industries. Every company's a tech company.
Define the trap business value and recognize the
We've heard that phrase. I always try to extend it
leverage that you need to unlock that growth for
and say, every company's a tech company and if
the business. And when you think about
you don't conduct yourself as such, you're just
not going to be successful. algorithms, which are the critical algorithms that
are going to solve the business value that was
Venky Rao (07:04): defined by the business, and when you think
Absolutely. And that's such a spot-on answer about algorithms today, it's a complicated world,
right? We need to make certain that they're
designed to scale and that they're unbiased
because we hear a lot of algorithms are being
defined with bias now. And we have to be very
cautious about that. And then you have to think
about data, right? Because you understand the
business now, you, you're built, you're defining
the algorithms are going to support that
business value that you're trying to capture.
And you got to look at the data and have a clear
first, second-, and third-party data strategy,
right? And make certain that you have a life
cycle around that data that to create signals of
value for the enterprise. The fourth area that I
would say is a platform strategy, making certain
that you have the right ecosystem, and we
talked about that earlier, making certain that you
have the right foundation of capabilities to
create and be able to manage inside the
enterprise. And then the ability to execute that
strategy, right? How should our enterprises be
organized to be able to execute that strategy
across the enterprise? And that means different
teams and different responsibilities and different
ways of working and different behaviors in the
enterprise. And then the greatest investment is,
is the sixht piece of this is focus on your talent
and the culture that you're building and how
you're going to continue to retain, attract and
engage those resources that are helping you
bring this value to life and this distinctive
capability that you're building in your enterprise.
Venky Rao (09:56):
Javier, that was an outstanding response and a
very, very good framework for everybody to
follow, right? So, thank you so much.
Javier Polit (10:02):
Venky that was a pleasure, thank you for the
time. Thank you for the partnership.
Copyright © 2023 Accenture
All rights reserved.
Accenture and its logo
are registered trademarks
of Accenture. |
264 | accenture | Accenture-Video-Transcript-Semi-Value-Chain-New-Approach-Gen.pdf | GENERATIVE AI’S ROLE
IN THE
SEMICONDUCTOR
VIDEO TRANSCRIPT
Speaker B: Miles, we're seeing companies, as
Speaker A: Hello everyone, my name is Padam
you mentioned, really introduced off the shelf
and I lead data and AI in high tech industry at
solutions that enable back office productivity like
Accenture. Welcome to our live discussion on
Copilot and things of that nature. But what's
transformative impact of generative AI in
really going to catapult companies into really the
semiconductor industry. I'm joined by Miles, an
competitive space getting products to market
expert in semiconductor applications, and today
faster is utilizing Gen AI functionality across the
we are driving into how generative AI is
entire product design lifecycle. So it's not only
revolutionizing the sector.
bringing products to market faster, but it's how
can we use not only LLMs but LMMs in the
Speaker B: Yeah, thanks Padam, thanks for
design process to create more efficient and
having me. I lead our Gen AI and AI for or semi
optimized products for their customers? So far, I
globally. Let's just hop right into it. Padam, could
was hoping you could talk a little bit about how
you start us off by sharing some of the insights
Gen AI is transforming supply chain in the
that you're seeing in the semiconductor space?
semiconductor industry.
Speaker A: There's a lot happening in
Speaker A: Supply chain is such an important
semiconductor industry today. We're seeing
area for high tech industry, but also a very
semiconductor industry enabling the rest of the
complex one. Generative AI can automate
world to leverage generative AI. But these
interactions with your third party suppliers. It
companies are focusing on complex challenges
enables real time substitution based on
beyond just the basic needs for their own needs.
inventory and supplier availability. Think about
There's a significant push towards enhancing
disruptions that could happen with a supplier.
productivity and speeding up market entry.
Are you able to provide new options to your
Strategic adoption of these new technologies is
customers in a much more autonomous
a key in driving new innovation. And speaking of
manner? And this automation improves
innovation, Miles, how is generative AI
procurement processes and overall supply
influencing the R & D process within this
chain.
industry?
Copyright © 2024 Accenture
All rights reserved.
Accenture, its logo, and High
Performance Delivered are
trademarks of Accenture.
GENERATIVE AI’S ROLE
IN THE
SEMICONDUCTOR
VIDEO TRANSCRIPT
Speaker A: I'll talk about three things. One of
the main challenges is skill gap and resource
Speaker B: So moving on to manufacturing.
limitations. Addressing the talent gap, especially
Padam, what advancements are being driven in
in technology engineering but also in
this area for Gen AI?
governance, is crucial for adopting next gen
capabilities. The second one is the fear of
Speaker A: A lot has happened in the
unknown, especially with hallucinations and
manufacturing space over the last two decades,
data privacy is huge as well. And the third
but I believe Gen AI is pivotal in predictive
important thing is building strong data
analytics. To improve yield and throughput even
foundation. Deploying responsible AI and
further, we can use synthetic data to improve
governance in that strong data foundation and
model accuracy and do better prediction and
strategically aligning to business value is very,
managed effects. This significantly enhances the
very important. It's clear that generative AI is a
manufacturing processes and efficiencies. So
powerful driver of innovation in semiconductor
you are producing less crap and building more
industry. For our viewers, we hope this
products and higher quality products. Miles, lets
discussion inspires you to explore the
go a step further. Do you want to talk a little bit
possibilities generative AI can offer. And thank
about the improvements Gen AI is bringing
you for joining us today.
about in testing and quality assurance.
Speaker B: The fabs across the world are
collecting a whole host of metrology data.
Metrology data is limited in terms of the quantity
that we're getting, but what companies can do is
they can create synthetic data in order to
leverage more efficient ways to look at defects in
product nonconformity for things that they're
producing now. Taking that metrology data that
exists, we can use that as input into future
capabilities or future LMM capabilities that I
mentioned earlier in order to design new and
more efficient products and to make sure that
those products are reaching the market not only
faster, but they're able to meet new customer
demands. Before we wrap up, Padam, could you
highlight some of the challenges that the
semiconductor industry face in Gen AI adoption? |
268 | accenture | Accenture-Accelerating-The-UKs-Generative-AI-Reinvention.pdf | Contents
Executive summary 4
The generative AI opportunity 7
The UK’s progress 18
The five imperatives to accelerate the UK’s reinvention through generative AI 24
Imperative 1: Lead with value 27
Imperative 2: Understand and develop an AI-enabled, secure digital core 31
Imperative 3: Reinvent talent and ways of working 36
Imperative 4: Close the gap on responsible AI 40
Imperative 5: Drive continuous reinvention 44
Generating growth: How generative AI can power the UK’s reinvention Copyright © 2024 Accenture. All rights reserved. 2
Preface
The UK is now a nation of potential Realising the value of generative AI (gen AI) will not come too quickly to job displacement, rather than how to use the
easy. When I speak to CEOs and leaders across industries, it’s technology to amplify human abilities and brilliance. Creating
innovators. Large Language Models
clear progress has been made. But getting to scale is proving an environment for humans and machine to best work
(LLMs), like ChatGPT, put advanced
a challenge. Based on current executive choices, the UK could together is no longer a choice; it is a responsibility. A landmark
leave nearly half a trillion pounds in economic value on the shift in digital training will be crucial to achieving this, with
skills at everyone’s fingertips. If
table in the next 15 years. executives anticipating the need to reskill 20 million people.
government and business leaders
At Accenture, we’ve been working with clients to navigate In the right hands, gen AI is a catalyst for reinvention. So,
can harness this, it could lead to a
this complex terrain—and of course, Accenture is on its own whether you’re just starting out or already on your AI journey,
new era of growth. journey of reinvention with AI, too. Though each organisation’s this report offers the formula to deploy gen AI successfully,
journey is unique, one thing is clear: those at the forefront are responsibly and with real impact.
shifting from productivity-focused use cases to positioning
We’ve already had a glimpse of how gen AI may change how
gen AI at the heart of their growth strategies.
we live and work: a future reimagined. Now all of us—from
The good news is this: The foundations clients have been employees to business leaders to government—hold the
building through their digital transformations are exactly responsibility to translate the promise into reality and deliver
what is needed to scale gen AI. With this technology, we can broad-based growth for the UK.
now complete the sentence of the digital age. Organisations
that can connect modernised tech foundations to a forward-
thinking vision for the future will emerge as true leaders.
Shaheen Sayed
CEO, Accenture, UK,
At the heart of this is taking a people-centric approach.
Ireland and Africa
Our analysis shows this will lead to greater economic upside.
We need to shift our thinking from AI to IA—Intelligence
Augmentation. The conversation on talent often moves
Generating growth: How generative AI can power the UK’s reinvention Copyright © 2024 Accenture. All rights reserved. 3
Executive
Fixing the triple fracture Make AI a multiplier Based on delivering more than 1,000 global
gen AI projects, we see a formula for success
The UK has built strong foundations for the age Closing these gaps requires a people-centric
summary
emerging. In this report, we outline the
of AI—but the cracks are beginning to show. approach. Three quarters of the nation’s
five imperatives behind that formula and
workforce could see at least a third of their
A delivery gap is opening as organisations
how it can accelerate the UK’s AI-powered
working hours enhanced by the current
struggle to move their use of gen AI beyond
reinvention: lead with value; understand
Gen AI presents a bigger state of the technology. Our economic
proofs of concept. Of the organisations that
and develop an AI-enabled, secure digital
modelling forecasts that when employees
opportunity for the UK than
have piloted gen AI, most (89%) have not
core; reinvent talent and ways of working;
are empowered to innovate and identify new
any other G7 country. scaled its use across their business. close the gap on responsible AI; and, drive
opportunities, financial gains are greatest. Yet,
continuous reinvention.
Many workers still lack even basic digital skills three out of five executives are prioritising
The technology could almost double the
and access to the training needed to develop investments in process automation that
The elements of the formula are mutually
UK’s long-term growth rate over the next 15
them, signalling an inhibiting skills gap. cut costs in the short-term over ones that
reinforcing, so shouldn’t be applied in
years (to 2038) and deliver a larger economic
Around 20 million people—62% of today’s transform people’s roles for the long-term.
isolation. Strategic alignment between
impact compared to the other
workforce—need reskilling. Executives report
technology, talent, governance and value
There is a real optimism among UK workers
22 countries we analysed.
that less than half (43%) of their workforce
roadmaps is essential. Our modelling
about the impact of AI. Three times as
is confident in the digital fundamentals
But there is no guarantee the full potential estimates an organisation is four times more
many people think gen AI will accelerate,
required for work. A surge in digital skills
for productivity and growth will be realised. likely to succeed in scaling the use of gen AI
rather than decelerate, their career
training is needed, and urgently.
A people-centric approach is required that if coordinated action is taken towards the five
progression. Many are moving ahead of their
puts the emphasis on using gen AI to amplify imperatives simultaneously.
Finally, a trust gap is emerging between organisations: half of the people using gen AI
human abilities. Too few organisations are
employees and executives, impeding at work are self-starters who are using tools
taking this approach today. Without strategic
adoption. Only a third (33%) of people expect they procured themselves. More needs to be
intervention, £485 billion in economic value
business leaders to be responsible and make done to harness this enthusiasm. Over the past 18 months,
could be left untapped by 2038—an amount
the right decisions to ensure gen AI has a gen AI has captured
A formula for success
equivalent to double the country’s current
positive impact on the UK, and even fewer
imaginations; now, with this
Nearly one in 10 (9%) organisations are using
annual healthcare spending.1
(27%) trust the government to do so.
formula, it can deliver results
gen AI at scale, so we know it can be done.
What should public and private sector
leaders do over the next 12 months to put
their organisations—and the UK economy—at
the forefront?
Generating growth: How generative AI can power the UK’s reinvention Copyright © 2024 Accenture. All rights reserved. 4
Authors
This report was a
collaborative effort
between our Data and
AI team based in the
UK, supported by our
Chris Lane Mark Farbrace Joe Hildebrand Suhail Kapoor
research team: Data & AI Lead—UK, Gen AI Lead—UK, Managing Director, Manging Director, Data
Ireland and Africa, Ireland and Africa, Gen AI for Human & AI—UK & Ireland,
Accenture Accenture Potential—UK & Ireland, Accenture
Accenture
Nitya Langley Kayur Rughani Ali Shah Bella Thornely
Managing Director, Managing Director, Responsible AI Senior Manager, Data
Data & AI—UK & Data & AI—UK & Lead—UK & Ireland, & AI—UK & Ireland,
Ireland, Accenture Ireland, Accenture Accenture Accenture
Generating growth: How generative AI can power the UK’s reinvention Copyright © 2024 Accenture. All rights reserved. 5
About the research
We took a multi-pronged approach to researching the UK’s
gen AI-powered reinvention. The report is based on:
Economic modelling to forecast the potential impact of gen AI on productivity and growth for the
economy, organisations and people. We mapped out the future growth trajectories under three different
AI deployment scenarios: aggressive, cautious and our proposed people-centric approach.
Surveys conducted with 3,752 employees and 1,085 executives from public and private sector
organisations in the UK. The samples covered 19 industries and included different demographic
groups by geography, company size and socioeconomic background. The employee survey looked
at UK workers’ experiences with gen AI. The executive survey looked at leaders’ perceptions of the AI
ecosystem, their investments in gen AI and their AI strategy. The surveys were conducted in July and
August 2024.
Interviews, client experience and case studies, drawing on insights with leaders from across the AI
ecosystem, including large technology providers, industry, government and civil society.
Further details on the research approach can be found at the end of the report.
Generating growth: How generative AI can power the UK’s reinvention Copyright © 2024 Accenture. All rights reserved. 6
The generative AI
opportunity: For people,
organisations and the
economy
Generating growth: How generative AI can power the UK’s reinvention Copyright © 2024 Accenture. All rights reserved. 7
The gen AI state of play
Figure 1. Welcome to the age of generative AI
Analyse Simulate
Scenario Optimise
Segment Recommend
Generating growth: How generative AI can power the UK’s reinvention Copyright © 2024 Accenture. All rights reserved. 8
Right here, right now
This is a pivotal moment. Gen AI is going to have a 140 languages.3 Wayve, a company developing
profound impact on how we work and live, more so autonomous driving technology, has pioneered a
than any other recent technological advancement. vision-language-action model (VLAM) that explains
It has the power to reshape industries and multiply to passengers how its AI ‘thinks’ and makes driving
workforce capabilities. The steps individuals, business decisions, increasing transparency and user trust.4
leaders and policy makers take now, will set the
These startups are not examples from Silicon Valley
trajectory for the UK in the years—and even decades—
or Shenzhen—they’re based here in the UK. In fact,
to come.
around one in four gen AI startups in Europe are based
We are only at the start of the S-curve (a model in London.5 These organisations—alongside the almost
showing a technology’s adoption from slow growth to 200,000 UK residents we estimate have AI skills—form
rapid rise and eventual saturation), but the potential part of an AI ecosystem that most executives (68%)
is already evident. Drug design and development surveyed describe as advanced or world-leading (see
company Exscientia has cut drug discovery times by Figure 2).
70%.2 AI video communications platform Synthesia
enables anyone to change written content into
studio-quality videos, voiced by AI avatars in over
Generating growth: How generative AI can power the UK’s reinvention Copyright © 2024 Accenture. All rights reserved. 9
Figure 2. The UK’s AI ecosystem has strong foundations
State of the UK’s AI ecosystem Strengths of the UK’s AI ecosystem Net Availability of AI skills in the UK
% respondents1 % respondents2 +/- # people reporting skills on LinkedIn3
Research
13 World leading 5 71 +66
institutions
Computing
10 61 +51
infrastructure
Talent pool 13 57 +44
55 Advanced 193,146
Regulatory
18 47 +29
environment
2.8x
Access to
21 45 +24
funding 69,156
10 Somewhat developed
Cost of doing
25 41 +16
5 Underdeveloped business
16 Don’t know
Weakness Strength
2023 2024
(As of July) (As August)
1. Respondents were asked: How would you describe the UK’s AI ecosystem? AI ecosystem was defined as: the network of organisations, resources and stakeholders involved in
the development of AI technologies, including government entities, companies, research institutions and support structures such as funding infrastructure, regulatory
frameworks and talent pools that collectively contribute to the growth and development of AI. Accenture UK AI business leader survey, fielded July-August 2024.
2. Respondents were asked: Would you consider each of the following as either a strength or a weakness of the UK's AI ecosystem? Data for “neither strength nor weakness” is
not shown. Accenture UK AI business leader survey, fielded July-August 2024.
3. Accenture UK Tech Talent Tracker based on data from LinkedIn Professional Network.
Generating growth: How generative AI can power the UK’s reinvention Copyright © 2024 Accenture. All rights reserved. 10
Impact from
self to society
The UK’s strong foundations position it to become a global leader in
the gen AI era. With a high share of services and knowledge-based
industries—sectors where the technology can have the greatest
impact—the potential benefits for the UK’s economy, organisations and
people are significant.
Generating growth: How generative AI can power the UK’s reinvention Copyright © 2024 Accenture. All rights reserved. 11
Our research brings into
view the size of the prize:
For the economy Figure 3. T he UK has more to gain from generative AI than any other G7 nation
We model that gen AI could:
Annual GDP gains in 2038 compared with non gen AI baseline
• Add up to £736 billion to annual UK GDP
%
in 2038—this amounts to a 23% increase
to the baseline forecast for 2038.
+23% +22%
• Shift average annual real GDP growth for
+21% +21% +20% +20%
2023–2038 from a baseline forecast of +18%
1.6% to 3.0%, representing an 89% boost
to the UK’s long-term growth rate.
Gen AI is forecast to benefit the UK economy
UK Germany France Canada Italy Japan USA
more than that of any other G7 nation
(see Figure 3).
Source: Accenture Research, simulated GDP growth under three scenarios. GDP gains shown
for People-Centric scenario. Oxford Economics GDP forecast used as the baseline.
Generating growth: How generative AI can power the UK’s reinvention Copyright © 2024 Accenture. All rights reserved. 12
For organisations
The more than 5 million businesses that make up UK plc, alongside public sector organisations
that deliver citizen services and shape the corporate environment, are the agents for creating
the conditions for growth.6
A double-digit productivity uplift could be achieved across the private and public sectors,
based on the current state of the technology. The sectors that are amongst the most important
to the UK’s economy, such as financial services, have the most to gain (see Figure 4).
If the productivity benefits are translated into cost savings, the gains could be substantial.
Across all industries analysed, total annual savings could reach £166.7 billion if the full potential
of today’s technology to automate and augment work is realised.
We estimate that
Nowhere is this opportunity bigger than in the public sector. We estimate that 47% of working
hours in the UK public sector (excluding healthcare) could be enhanced by gen AI (either
47
through automation or augmentation). This translates into a potential productivity gain of 14–
%
20% that, if realised, could result in £34.4 billion in annual savings, equivalent to more than the
annual expenditure on primary school education.7
of working hours in the UK public sector
(excluding healthcare) could be enhanced
by gen AI (either through automation
or augmentation).
Generating growth: How generative AI can power the UK’s reinvention Copyright © 2024 Accenture. All rights reserved. 13
Figure 4. Potentialproductivitygainscouldbe30%+acrossthefinancialservicesandtechsectors
Productivity gains from gen AI exposure %
Modelled range*
Software & Platforms £17.6
Capital Markets £9.7
Banking £12.7
Insurance £3.4
High Tech £1.9
Communications & Media £13.0
Life Sciences £1.3
Public Service £34.4
Travel £2.0
Retail £11.3
Aerospace & Defence £3.2
Industrial £25.4
Natural Resources £2.0
Consumer Goods & Services £5.4
Utilities £4.0
Health £13.1
Automotive £5.1
Chemicals £0.9
Energy £0.3
5% 10% 15% 20% 25% 30% 35% Mid-point
cost savings (£bn)
Source: Accenture Research based on ONS and US O*net. Lower and upper bound based on potential hours saved by occupation valued at annual occupation headcount.
Generating growth: How generative AI can power the UK’s reinvention Copyright © 2024 Accenture. All rights reserved. 14
Gen AI isn’t just a productivity play—it creates new avenues for using gen AI. Our analysis revealed that companies actively Gen AI can close the capability gaps that typically favour
growth. In the second quarter of 2024, AI startups attracted pursuing this strategy delivered a 10.7 percentage point total large companies. Today, platforms like Jasper enable SMBs
31% of all venture funding in the UK. This is nearly three times return to shareholder (TRS) premium in 2023 compared to to access services such as marketing content creation at low
the amount compared to the same period in 2022, before those that did not, even after controlling for company size, cost. In the future, agentic AI—autonomous systems that make
ChatGPT’s public release.8 headquarters location and industry.10 independent decisions and take actions to achieve specific
goals—may enable SMBs to autonomously run entire business
A significant proportion of the growth opportunity comes Gen AI’s performance premium offers both opportunities and
processes. Startups are already developing customisable AI
from the build out of AI’s foundations. In the race for AI risks for incumbent organisations. Leveraging it effectively—
agents capable of handling customer inquiries, managing
supremacy, leading technology companies are building by tapping into unique data sets from existing customer
workflows and resolving issues across multiple channels.
infrastructure akin to the expansion of the electric grid in the relationships, for instance—can provide an ‘intelligence
early 20th century. Just as electricity transformed industries advantage’ that boosts returns. Failing to do so, however,
and powered global economies, gen AI is poised to drive could leave you vulnerable to disruption from a new
the next wave of innovation. Analysts estimate over a trillion generation of AI-first companies. Survey responses from small
dollars will be spent globally on AI infrastructure over the next and medium-sized businesses (SMBs) hint at future trends.
five years, as companies compete to ‘own the grid’ of this new While SMBs only slightly trail large companies in AI adoption,
technological era.9 their satisfaction levels are notably higher. In fact, 86% of
SMB executives are satisfied with their return on investment
Over time, the effective use of gen AI will become an
(ROI), compared to 75% of executives from large corporations.
increasingly important source of competitive advantage.
This may reflect SMBs’ agility and fewer legacy constraints.
We analysed earnings calls from 1,300 global companies
For instance, SMB leaders are less likely than those in large
with revenues exceeding £750 million to assess the extent
multinationals to cite their technology platform as a significant
to which they cited efforts to build competitive advantage
barrier to scaling gen AI.
Generating growth: How generative AI can power the UK’s reinvention Copyright © 2024 Accenture. All rights reserved. 15
For people
By harnessing individual human potential, organisations will realise the most benefits. Gen AI could also help address talent gaps. Investment in vocational training in the UK is 20%
below the peak seen in the early 2000s.13 This is contributing to a shortage of people in areas
No current technology has the potential to have a bigger impact on our working lives than AI.
such as health and social work.14 Gen AI can help address these skill shortages quickly. One
Three in four people in the UK could have at least a third of their working hours enabled by the
interviewee described how a gen AI tool streamlined onboarding for new carers, enabling them
technology, either through automation or augmentation.
to reach the top 20% of performers within six weeks. Workers recognise this potential—over
Automation will save people time, taking tedious tasks off human three times as many survey respondents expect gen AI to accelerate rather than slow their
hands. Our modelling suggests the average UK worker could save 18% of their working career progression.
hours spent on routine activities. A doctor, for example, could save five hours a week while a
As people spend more time doing work they enjoy and doing it well, gen AI could help in a
commercial sales rep could save twelve hours a week.
more profound sense by improving the overall experience of work. In an experiment with our
The time saved could be reinvested in the higher-value work people enjoy doing. Creativity own sales team, we found that gen AI didn’t just result in marked increases in productivity but
is the most underutilised skill in the UK: 26% of people we surveyed say they aren’t currently also grew peoples’ confidence (+34%) and their belief they were making a meaningful impact
applying their creativity at work. With gen AI, the average UK worker could increase the time (+31%).15 Gen AI added to their job satisfaction rather than subtracted.
they spend on creative tasks by 13%.11
We see similar findings in our survey. UK workers recognise gen AI will be important to their
The benefit of augmentation will be accrued not just in time but in quality. Imaging tools, for productivity and problem-solving. But they also anticipate the technology will benefit their
example, could help medical teams make quicker and more accurate diagnoses. Early pilots autonomy and sense of purpose (see Figure 5). Familiarity with the tools reduces anxiety, as
shows that AI could help the National Health Service (NHS) almost halve diagnosis times for employees recognise how the technology complements their existing skills and helps them
stroke patients.12 perform tasks more effectively. Daily ‘power users’ of the technology were more than twice as
likely to expect gen AI to be important to both their creativity and fulfilment from work, relative
to irregular users.
Generating growth: How generative AI can power the UK’s reinvention Copyright © 2024 Accenture. All rights reserved. 16
Figure 5. PeopleanticipatebroadbenefitsfromgenAI—theirexpectationsincreaseastheyusethetoolsmore
Workers’ level of gen AI use (of those with access to the tools), % respondents1
31 56 13
Irregular users Light users Power users
Share of workers that anticipate gen AI will be important to their work experience,
% respondents by level of gen AI use1
90 90
84 84
81
80 79
74
71
68
66
64 64 63
59
56
54
53
50
40
35
33
21 21
Productivity Problem-solving Learning Creativity Autonomy Well-being Fulfillment Purpose
1. Irregular users are respondents who never or rarely use the gen AI tools available to them. Light users use the tools often (at least once a week) or sometimes (once a month). Power users use the tools every day.
Source: Accenture UK AI employee survey, fielded July-August 2024.
Generating growth: How generative AI can power the UK’s reinvention Copyright © 2024 Accenture. All rights reserved. 17
The UK’s
progress
Generating growth: How generative AI can power the UK’s reinvention Copyright © 2024 Accenture. All rights reserved. 18
Mind the gap
While the UK has laid strong foundations for the AI Figure 6. B ased on the decisions being made today, the UK is running closest to our
opportunity, it is not yet positioned to fully realise low-end economic scenario, potentially leaving £485bn in value on the table
its potential.
Our survey of business leaders examined UK economic growth simulation, 2023-38
GDP in £ billions (constant prices)
which of the three economic growth
Scenario GDP gain GDP CAGR GDP gain
scenarios we modelled aligns most closely
vs. baseline premium as a share
with the UK’s current trajectory. by 2038 vs. 1.58% of baseline
baseline
4,000
In our most optimistic, ‘people-centric’ scenario, People-centric £736bn +1.4pp +22.8%
organisations harness gen AI to automate
£485bn
routine tasks, redirecting the time saved into Cautious £561bn +1.1pp +17.4%
higher-value activities. In contrast, in our
3,500
‘aggressive’ scenario, companies prioritise Aggressive £251bn +0.5pp +7.8%
cost-cutting, with workers finding themselves
in less dynamic roles (or unemployed) after Baseline
being displaced, which stifles growth and
3,000
exacerbates inequality (see ‘Further details on
the research’ for more on these scenarios).
On current trends, the UK is leaning toward the
lower end of the growth spectrum—closest to our 2,500
2023 2026 2029 2032 2035 2038
‘aggressive’ scenario—potentially leaving £485
billion in economic value untapped (see Figure 6).
Source: Accenture Research, simulated GDP growth under three scenarios. Oxford Economics GDP forecast used as the baseline.
Exchange rate is based on the period average (USD per Pound), Oxford Economics
Generating growth: How generative AI can power the UK’s reinvention Copyright © 2024 Accenture. All rights reserved. 19
Triple fracture
What is contributing to the lost potential? These trends are mirrored among workers. Although 43% of
UK employees have access to gen AI tools to support their
63%
We identified three points of tension where gen AI
work, only 19% use them at least once a week. Even fewer
deployment is strained:
(7%) are applying the tools to critical decision-making or high-
impact analysis.
A delivery gap
of London-based employees
In 2024, survey respondents expect gen AI to account for We see regional disparity in levels of gen AI deployment.
10% of their technology spend, rising to 15% in 2025. That Organisations outside London plan to invest a third less in have access to gen AI tools;
investment has yet to translate into scaled deployment. While the technology. While 63% of London-based employees have
only 38% of employees
79% of executives report their organisations have at least access to gen AI tools, only 38% of employees elsewhere in
elsewhere in the UK do.
piloted gen AI in one or more parts of their business, only 9% the UK do. And business leaders in the capital report using the
have scaled the technology (with use cases in production technology in around 50% more of their business operations.
in more than half of their business functions). Many lack the
Given gen AI’s potential to drive productivity and growth,
foundations needed to scale. Fewer than one in four (24%)
there is an urgent need to level up its adoption nationwide.
executives, for example, feel confident that their organisation’s
technology capabilities meet the requirements to successfully
A skills gap
scale gen AI.
A landmark shift in digital skills training is essential to unlock
Where gen AI is being implemented, the focus tends to be on the benefits of gen AI. The executives we surveyed estimate
the bottom line. Three out of five executives are prioritising that 62% of their workforces will require reskilling—equivalent
investments in process automation that cut costs over to roughly 20 million people (see Figure 7). For some, this will
initiatives that augment people’s roles and transform how involve developing technical skills such as AI engineering. For
they work. most, it will focus on training to collaborate with AI systems.
Generating growth: How generative AI can power the UK’s reinvention Copyright © 2024 Accenture. All rights reserved. 20
Figure 7. Executivesestimatethat62%oftheirworkforceswillrequirereskillingduetogenAI
Expectations for how gen AI will change roles at organisations by the UK, by select region
% of current job roles1
Jobs to be transitioned: Requiring reskilling / upskilling for new roles Jobs with significant enhancement: requiring substantial reskilling/ upskilling
Jobs with some enhancement: requiring some reskilling / upskilling Jobs not impacted: No reskilling/ upskilling required
22 18 18 15 16 15 15 13 13 17
17
20 21 22 20 24
62%
23 22
23
23
23
20
17
24 23 20
23
22
24 25
30 34 38 41 41 43 46 46 48 38
London North-West West Midlands East of England South-East East Midlands Scotland South-West Yorkshire UK
and the Humber
4.78 3.57 2.85 3.12 4.77 2.41 2.89 2.60 2.64 33.09
Employment mn
1. Respondents were asked: How, if at all, do you expect generative AI to change job roles at your organisation? (Please estimate what proportion of current job roles you expect to fall into the
following categories by distributing 100% points across the options listed. Not all regions are included due to insufficient sample size.
Source: Accenture UK AI employee survey, fielded July-August 2024. ONS current employment levels. Apr-Jun 2024.
Generating growth: How generative AI can power the UK’s reinvention Copyright © 2024 Accenture. All rights reserved. 21
Despite leaders expecting gen AI Many (40%) say they are pushed to use new
deployment to require a significant technology they haven’t been trained on.
uplift in skills, less than half (45%) report
Regional disparities can be seen here too.
that their organisations have increased
Londoners are both more likely to have
training on either gen AI fundamentals
access to training opportunities and are
or technical skills in the past year.
more willing to pursue them. Over the past
Many workers still lack even basic digital skills 12 months, more than 60% of organisations
or access to the training needed to develop in London have increased training on gen AI
them. Executives estimate that less than skills, compared to only 40% in other regions.
half (43%) of their workforce is confident And when considering the potential impact
in the digital fundamentals required for of gen AI on their work, 64% of London-
work. At the same time, nearly one in five based employees are likely to consider
workers (17%) report not having received reskilling, compared to less than half (46%)
any digital skills training in the last two years. of those based outside the capital.
Generating growth: How generative AI can power the UK’s reinvention Copyright © 2024 Accenture. All rights reserved. 22
A trust gap
As we highlighted in our previous (27%) or business leaders (33%) will make Figure 8. Employeesandexecutivesarenotaligned
research report, Work, workforce, the right decisions to ensure gen AI has a on the long-term societal impact of gen AI
workers: Reinvented in the age of positive impact on the UK. Trust levels in the
generative AI, transparency and trust government range from as low as 17% in the
Expectations about the outcomes of the
are required for people to adopt gen South West to as high as 42% in London.
widespread use of gen AI in the UK
AI tools. That research revealed a trust
Expectations around the value gen AI can
gap between workers and leaders. % of executive and employee respondents
deliver—whether in boosting economic
Decrease Increase Net +ve Net -ve
Almost a year later, we find the trust gap growth, equality or employment—differ
Productivity
persists. Trust and user acceptance was significantly between employees and
Executives 10% 67% +56
the fourth most common barrier cited by leadership (see Figure 8). This disparity Employees 12% 47% +35
organisations to scaling the use of the highlights concerns about social inclusion
Economic growth
technology, behind data security and privacy and employee rights, underscoring the trust
Executives 11% 60% +49
concerns, quality and accuracy concerns gap. If not addressed, these issues could Employees 15% 30% +15
and the cost of implementation. Few undermine the potential benefits of gen AI.
Digital inclusion
workers have confidence that government
Executives 17% 51% +33
Employees 24% 30% +5
Economic equality
Executives 24% 39% +15
Employees 34% 16% -18
Social mobility
Executives 24% 39% +15
Employees 30% 17% -13
Source: Accenture UK AI business leader and employee surveys, fielded July-August 2024.
Generating growth: How generative AI can power the UK’s reinvention Copyright © 2024 Accenture. All rights reserved. 23
The five imperatives
to accelerate the UK’s
reinvention through
generative AI
Generating growth: How generative AI can power the UK’s reinvention Copyright © 2024 Accenture. All rights reserved. 24
A formula
for success
What could be done to get
the UK’s gen AI-led
reinvention back on track?
Based on our experience of delivering over
1,000 gen AI projects globally, we see a
formula emerging for how organisations can
responsibly scale the use of gen AI:
’Moving from discrete interventions
Imperative 1: Imperative 2: Imperative 3: Imperative 4: Imperative 5:
to real innovation requires a much
bolder and more holistic approach. Lead with value Understand and Reinvent talent Close the gap on Drive continuous
While many early adopters are develop an AI- and ways of responsible AI: reinvention:
Shift the focus from
focusing on building a technol |
269 | accenture | Accenture-Competitive-Switzerland-2024-Report.pdf | Playing the Long Game
Can Switzerland lead
the way in generative AI?
Competitive Switzerland Innovation. Insights. Actions
Contents
Executive Summary 3
01 | Generative AI Can Have a Profound Impact on 5
the Swiss Workforce and Economy
02 | Barriers to Unlocking the Full Potential of generative AI 16
03 | Final Considerations 37
Methodology Appendix 45
References 49
Playing the Long Game Can Switzerland lead the way in generative AI? 2
Executive Summary
Disruption provides opportunity
Half of Swiss executives believe their companies are well-prepared to harness
the opportunities presented by technological disruptions, with generative AI
(gen AI) being a primary driver today. Our research indicates that Switzerland
is third worldwide in terms of the impact of generative AI on work time.
This technology could significantly boost the Swiss economy, potentially
unlocking an additional CHF 92 billion of economic value by 2030 under
what we refer to as a “people-centric” scenario.
Switzerland will need to take a people-centric approach
To seize the economic opportunity, the Swiss workforce needs to be prepared.
Accenture’s analysis of the Swiss workforce and its tasks reveals that 45% of
work time in Switzerland is highly likely to be impacted by gen AI. This transition,
however, is viewed not as a threat, but as an opportunity to enhance productivity,
particularly within financial services.
Revenue opportunities, rather than mere productivity increases
Generative AI has the potential to do more than just boost productivity.
In Switzerland, 91% of executives believe gen AI will have a greater impact
on revenue growth than reducing costs. The optimistic outlook bodes well
for Swiss companies, as evidenced by the proactive steps taken by Helvetia,
Roche, Novartis, Givaudan, ABB, and Swisscom.
Switzerland is a top innovative country, having been ranked first in the WIPO
Global Innovation Index for the last 13 years. Its talent is recognized, too:
Switzerland has ranked first in the INSEAD Global Talent Competitiveness
Index for the last ten years.
Playing the Long Game Can Switzerland lead the way in generative AI? 3
Challenges remain
To become a global leader in generative AI, Switzerland needs to address
key challenges in three areas: enterprise, workforce, and regulatory readiness.
Top Swiss companies have room to increase their use of AI. Only a small
portion of Swiss companies are currently scaling gen AI initiatives and expect
to take longer compared to what global peers believe. Swiss workers,
meanwhile, are highly open to gen AI. They recognize its value and are willing
to acquire new skills. Still, while their optimism is evident, they maintain
a cautious stance on job security, work quality, and overall well-being.
The regulatory focus on AI has dramatically increased globally in the last
decade. Swiss policies largely align with the OECD principles, although not
completely. This has led to an ongoing and dynamic public debate within
the federal parliament, as shown by how frequently generative AI was
discussed between 2019 and February 2024.
Clear action points for Swiss companies
Five imperatives for Swiss companies to scale gen AI throughout their
organization are: leading with value; understanding and developing an
AI-enabled secure digital core; reinventing talent and ways of working;
Switzerland
closing the gap on responsible AI; and driving continuous reinvention.
truly has
Ideas for Swiss policymakers
Starting from a position of strength, there are some further ideas to allow a world of
the country to capture the full benefits of gen AI, including defining a
strategic vision for gen AI, fostering international collaboration, enhancing opportunity
role transition mechanisms, strengthening the dialogs and oversight on
ahead!
the AI/gen AI revolution, and supporting gen AI literacy in society.
Playing the Long Game Can Switzerland lead the way in generative AI? 4
01
Generative AI
Can Have a Profound
Impact on the
Swiss Workforce
and Economy
Playing the Long Game Can Switzerland lead the way in generative AI? 5
Disruption is on the rise Disruption has become a prominent force in today’s business landscape,
compelling organizations to adapt and reinvent themselves to stay competitive.
The Accenture Pulse of Change Index measures disruption across various
and is driving the need for
categories, including consumer, social, geopolitical, economic, climate, talent,
and technology categories. The indexed scores from 2019 to 2023 reveal a
reinvention
significant increase in technology disruption, primarily driven by advancements in
generative AI. This highlights the urgency for businesses to embrace innovation
and reinvent their strategies to navigate the evolving landscape effectively.
Accenture Pulse of Change Index
Indexed score, 2019–23
Overall Consumer Geopolitical Economic Climate Talent Technology
and Social
+17%
+34% +41%
+33% +88%
+13%
+20%
19 23 19 23 19 23 19 23 19 23 19 23 19 23
Source: Accenture Pulse of Change Index 2024. Overall measure of disruption is based on the average of the six sub-components, each of which are based on a set of indexed scores of a set of indicators.
Playing the Long Game Can Switzerland lead the way in generative AI? 6
Preparing for disruption is crucial for organizations to thrive in an ever-
changing environment. The data shows that only about half of Swiss
Only one in two Swiss executives feels prepared for
executives feel “very prepared” to tackle the multifaceted disruption ahead,
the multifaceted disruption ahead, with technology
particularly in technology. This indicates a need for organizations to enhance
readiness trailing behind.
their readiness by investing in technology adoption, upskilling employees,
and fostering a culture of innovation.
Level of preparation of companies for different types of disruption
(% of respondents saying they are “very prepared”)
Consumer Geopolitical Economic Climate Talent Technology
and Social
+2 pp –11 pp
–6 pp –2 pp –1 pp +7 pp
55%
51% 53% 53%
49%
47% 46%
43% 45% 43% 47% 42%
Global Switzerland Global Switzerland Global Switzerland Global Switzerland Global Switzerland Global Switzerland
Source: Accenture Pulse of Change 2024. Global N = 3450; Switzerland N = 100
Playing the Long Game Can Switzerland lead the way in generative AI? 7
Proportion of working hours in scope to be either automated or augmented by gen AI
Working hours in scope for automation augmentation
Switzerland is
UK 24.0 23.4 47.5
the third most
Canada 22.8 22.9 45.7
exposed country
Switzerland 23.2 22.0 45.2
to generative AI,
Germany 23.6 21.5 45.1
with 45% of work
Australia 23.2 21.7 44.9
time highly likely
Japan 24.9 19.4 44.4
France 22.5 21.5 44.1 to be impacted.
US 22.8 21.2 44.0
Sweden 21.3 22.2 43.5
Italy 22.3 20.3 42.6
Norway 21.2 21.4 42.6
Finland 21.3 21.2 42.5
Denmark 21.1 21.0 42.1
Argentina 23.7 18.3 42.0
Spain 21.5 20.4 41.9
Mexico 23.1 18.0 41.2
Chile 21.7 18.7 40.5
Brazil 21.8 17.4 39.2
Colombia 21.4 17.6 39.0
South Africa 21.4 17.3 38.7
Saudi Arabia 21.6 15.5 37.2
China 17.6 15.0 32.5
India 16.5 15.2 31.7
Note: Estimates are based on Human+Machine identification of work tasks exposure to impact of generative AI.
For details see the methodological notes in the appendix.
Source: Accenture Research based on National Statistical Institutes and O*Net.
Playing the Long Game Can Switzerland lead the way in generative AI? 8
Work roles will be transformed
in different ways and to different
degrees, depending on the specific
nature of tasks and the time that
each task takes.
Playing the Long Game Can Switzerland lead the way in generative AI? 9
)krow
egdelwonk(
laitnetop
noitatnemgua
rehgiH
Exposure to generative AI for top 20 occupations in Switzerland
Percentage of working time by role
45%
Software and
Applications
40% Developers
and Analysts
Shop
Administrative
35% and Specialized
Salespersons
Nursing and Secretaries
Midwifery Sales, Marketing,
30% Profes s ionals AO st sh oe cr iH ate ealth Managing Directors a Mn ad n aD ge ev re slopment a MB nu ads ni n aAe gds ems
r
siS ne isr tv ri ac te ios n
Professionals and Chief Executives
25% Physical and
Science
E Tn ecg hin ne ice ir ain ng
s
Other
S
C ul pe pric oa rtl G Oe ffn ice er a Cl
lerks
20% Domestic, Workers
Hotel, and Office
Cleaners and Helpers
15% Personal Ca re Waiters and
Worke rs in Primary Bartenders Manufacturing, Material Recording
Health Services School and Early Mining, Construction, and Transport Clerks
10% Childhood Teachers and Distribution
Managers
Building Finishers
Building and
5% and Related Trades
Housekeeping Building Frame
Workers
Supervisors and Related
0% Trades Workers
5% 10% 15% 20% 25% 30% 35% 40% 45% 50%
Higher automation potential (process work)
150K
Bubble size
75K
Number of employees
Note: Estimates are based on Human+Machine identification of work tasks exposure to the impact of generative AI. For details, see the methodology notes in the appendix. 10K in Switzerland
Source: Accenture, Federal Statistical Office of Switzerland, and O*Net.
Playing the Long Game Can Switzerland lead the way in generative AI? 10
Case Study
The office clerk’s apprenticeship
Focusing on the worker of the future
Switzerland revamps its commercial apprenticeship program
In the wake of rapidly advancing artificial intelligence, traditional models of tests are still necessary, that knowledge is immediately taken further, interwoven
vocational training face unprecedented challenges, with companies potentially into professional situations, practiced with fictitious customer conversations, and
preferring to turn to generative AI rather than hiring apprentices. consolidated in training units and practical assignments.
The concept of apprenticeship needs to undergo a profound transformation, Emphasis is put on imbuing a digital mindset and shifting towards skills machines
as this development risks marginalizing the human element of learning and cannot easily replace, such as critical thinking, creativity, social skills, and working
innovation. In the long term, this shift could lead to a widening skills gap, where independently. Moreover, trainees are prepared for the modern agile business
the workforce lacks essential problem-solving and creative thinking skills. environment, learning to adapt to changes quickly and fluently.
Switzerland has taken a proactive stance to address this problem by rejuvenating Final examinations are taken by future learners through hands-on tasks that
the basic vocational training programs such as the “Kauffrau/Kaufmann EFZ” reflect the most essential professional situations and combine basic knowledge
and “Kauffrau/Kaufmann EBA” programs. Both have been updated to optimally in German, English, IT, and economics, including accounting, general education,
prepare young businesspeople for the changing world of work and the and education in professional practice.3
future of commerce.1 The commercial apprenticeship (KV) in Switzerland is
notably popular, with 12,000 apprenticeships annually accounting for almost This initiative is just an example, but as generative AI evolves and matures, it’s
a fifth of all young professionals starting their training across roughly 250 critical to keep updating apprenticeship programs at scale, to keep them relevant
recognized professions.2 and to adequately prepare the workforce to work effectively alongside AI,
leveraging these technologies to augment human capabilities rather than replace
Schools and companies are equally involved, with trainees benefiting from them. In doing so, society can foster a more adaptable, skilled, and resilient
learning at school, at companies, and through inter-company courses. What’s workforce that is prepared to meet the challenges of the future.
more, teaching centers on a hands-on approach. For example, while vocabulary
Playing the Long Game Can Switzerland lead the way in generative AI? 11
Productivity improvement driven by gen AI by industry in Switzerland
% of time
0% 5% 10% 15% 20% 25% 30% 35%
Gen AI has the potential to
Banking 22.3% 30.4%
increase productivity by double
Insurance 19.9% 28.5%
digits in all industries, with
Capital Markets 20.4% 28.3% financial services leading
Software & Platform 20.7% 28% the pack. Additionally, Swiss
High Tech 17.3% 24.4% executives see it as a driver
of new revenues.
Comms & Media 16.8% 23.6%
Life Sciences 13.8% 20.5%
While industries in the financial services
Automotive 14.3% 20.3%
sector – capital markets, insurance, and
Retail 13.3% 19.7% banking – have the highest exposure to
gen AI, no Swiss industry is likely to remain
Public Service 13.1% 19.1%
untouched.
Energy 13.0% 19.1%
Chemicals 12.4% 18.0% To put this into context, this is a similar
increase in productivity to what Switzerland
Consumer Goods & Services 11.8% 17.9%
witnessed in the aftermath of the Internet
Natural Resources 12.3% 17.6%
revolution, from 2000 to 2006–09, and the
Utilities 11.8% 17.5% same increase seen in the last 11–16 years.
Industrial 12.2% 17.5%
Travel 11.2% 16.4%
Health 9.4% 15.5%
13% 19%
Note: Estimates are based on Human+Machine identification of work tasks exposure to impact of generative AI.
For details see the methodological notes in the appendix.
Source: Accenture Research based on National Statistical Institutes and O*Net.
Playing the Long Game Can Switzerland lead the way in generative AI? 12
Do you view generative AI as being more beneficial to
revenue growth or cost reduction for your organization?
91% of Swiss executives
% of Swiss companies’ respondents
(76% globally) believe generative AI
will be more beneficial for revenue
growth than cost reduction. This
Cost reduction
highlights a significant optimism
9%
about leveraging generative AI
for strategic advantages, beyond
automation.
Swiss executives
agree that
91%
generative AI will
be a key driver
Revenue growth
for revenue
growth over cost
reduction.
Source: Accenture Pulse of Change, Nov. 2023; n (Switzerland) = 100; companies’ press releases
Playing the Long Game Can Switzerland lead the way in generative AI? 13
Examples of top
Novartis
Swiss players turning
launched the Generative Chemistry (GenChem) initiative, revolutionizing
to generative AI to drug discovery. Using advanced AI, GenChem designs molecule structures
increase top line growth and identifies potential new medicines. This approach speeds up the
discovery of top-quality molecules and enhances their developmental
success rates. With the support of over 250 data scientists, key research
areas, from target identification to predictive biomarkers, are optimized.
Givaudan
has launched a generative AI creation assistant, a proprietary AI model,
trained on the company’s knowledge and data to support the creativity of
perfumers and flavorists.
Swisscom and NVIDIA
have joined forces with an investment of CHF 100 million to spearhead the
development of generative AI supercomputers in Switzerland and Italy. This
collaboration aims to establish a Trusted AI Factory, focusing on creating
secure, sovereign gen AI solutions.
Playing the Long Game Can Switzerland lead the way in generative AI? 14
Economic growth simulation, Switzerland 2023–2038 Impact of generative AI on Swiss GDP, by scenario
GDP in billion CHF (2015 constant prices and exchange rate) GDP added against baseline by 2030 and annual growth 2023–2030,
Baseline: Oxford Economics. Simulations for three scenarios. billion CHF and %
1,100 Annual growth rate 3.9% 2.5% 2.3%
2023-30
1,000 131
900
CHF 92 billion
additional value
800 unlocked by 2030
48
39
700
600
22 24 26 28 30 32 34 36 38 People-centric Cautious Aggressive
People-centric Cautious Aggressive Baseline
Note: Higher-quality jobs defined as those with higher Net Better Off score.
Source: Accenture Research analysis. See methodology slide for further details.
By adopting generative AI in a people-centric and responsible manner across
the Swiss economy, significant benefits are anticipated by 2030. Focusing By embracing a responsible, people-centric approach
on enhancing rather than replacing jobs could yield an extra CHF 92 billion
to generative AI, the Swiss economy stands to unlock
in economic value, contributing to an additional annual growth rate of 1.6%.
additional economic value of CHF 92 billion by 2030.
The people-centric scenario assumes a moderate adoption pace, achieving
full integration within ten years with no effects on unemployment. It also
emphasizes active support for workers transitioning to higher-quality jobs,
ensuring that the adoption of generative AI enhances, rather than undermines,
employment quality.
Playing the Long Game Can Switzerland lead the way in generative AI? 15
02
Barriers to Unlocking
the Full Potential of
generative AI
Playing the Long Game Can Switzerland lead the way in generative AI? 16
Switzerland’s tech competitive advantage
By several measures, Switzerland is in an excellent
position to be leading the generative AI wave
1st 1st 5th
In the WIPO Global Innovation In the INSEAD Global Talent In the IMD World Digital
Index in the last 13 years4 Competitiveness Index in Competitive Ranking in 20236
the last 10 years5
Playing the Long Game Can Switzerland lead the way in generative AI? 17
Large existing talent pool Strong tech infrastructure
Switzerland is recognized for its robust talent pool, particularly in tech. The Swiss National Supercomputing Centre (CSCS) provides strong
Almost 6% of its employment base are ICT specialists7, 1.5 times the computing capacity via several supercomputers (e.g., Piz Daint exceeding
European average. 25 petaflops) for detailed and complex simulation across various fields11.
CSCS will also house the new Alps supercomputer (10,000 GPUs NVIDIA),
According to the latest Global Talent Competitiveness Index, Switzerland, which will be launched in spring 2024 thanks to the efforts of ETH Zurich and
along with the US and Singapore, is at the top of the ranking in attracting EPFL to develop open-source models for generative AI12.
and retaining skilled professionals.
High focus on R&D expenditure
Switzerland has consistently been among the countries with the highest R&D
expenditure as a proportion of GDP. The latest data indicates that the private
sector plays a significant role, contributing to over two-thirds of the R&D
expenditure, which amounts to over 3% of the GDP8, more than 1% higher
than the European average. Switzerland’s leadership in patents also highlights
the commitment to innovation. In 2023, it led the world in patent applications
per million residents, filing twice as many as Sweden, the second-ranked
country, and nearly eight times more than the United States9, with significant
contributions across various industries such as pharmaceuticals, consumer
goods, and high tech.
World-class research institutions
Nine Swiss institutions appear in the latest ranking of the 500 best global
universities, with ETH named the top university in continental Europe10. In 2023,
approximately 2,000 patents were published in Switzerland, with around
1,200 patents granted. It ranked first in applications per million residents, with
a large margin over other innovative countries. Swiss companies filed almost
seven times as many patent applications per million inhabitants last year as
companies in the United States, with significant contributions across a range
of industries such as pharmaceuticals, consumer goods, and high tech.
Playing the Long Game Can Switzerland lead the way in generative AI? 18
Enterprise
readiness
To seize the full
potential of generative
AI and capitalize on its
Regulatory
opportunities, Switzerland readiness
needs to address
three key challenges
Workforce
readiness
Playing the Long Game Can Switzerland lead the way in generative AI? 19
Room for improvement in terms of AI readiness
Our AI index reveals that top Swiss companies have the potential to enhance
their AI readiness. While there are some pioneers in different aspects of
AI, many companies have room for improvement, particularly when it comes
to AI talent and responsible AI.
Difficulty in scaling
A survey13 reveals that 62% of Swiss companies have implemented AI to some
degree. However, the challenge lies in expanding the technology throughout
Enterprise
the entire organization. Our pulse survey14 indicates that only 2% of Swiss
companies are currently scaling gen AI initiatives and expect to take longer
readiness
compared to what global peers believe. Also, only 7%, half the global average,
are “extremely confident” they have the right data strategy and core digital
capabilities in place to effectively leverage generative AI.
Difficult AI governance
To leverage gen AI for specific use cases, businesses might need to feed
sensitive data into these models. This requires businesses to implement strong
safeguards to protect sensitive information and prevent unauthorized access or
breaches that could compromise privacy and trust. 52% of Swiss organizations
lack clear AI workplace policies, indicating a critical need for guidelines15,
and only 4% have progressed from designing or initiating the scaling up of a
responsible data and AI model to fully integrating one into their enterprise16.
Playing the Long Game Can Switzerland lead the way in generative AI? 20
Lack of digital skills in the workforce
An Adecco survey shows that Swiss workers are trailing their peers in the acquisition
of digital skills17 (e.g., artificial intelligence, machine learning, data analytics,
data mining, design thinking, digital design, digital marketing, programming,
data analysis), and instead are more focused on job-specific and functional skills
(e.g., accounting, marketing, finance, human resources, analysis, IT).
Unseen use of AI
Our Accenture global workforce survey shows that 85% of Swiss workers
Workforce
already utilize generative AI in their jobs in various ways18. This trend reflects
the increasing integration of generative AI into everyday tasks and highlights a
readiness
possible disconnect in formal training and understanding of these technologies’
capabilities and ethical implications among the workforce.
Trust deficit
Based on our workforce survey, 50% of Swiss workers are concerned about the
quality of gen AI output, in line with the global sample19. Additionally, 48% fear job
displacement due to generative AI. This skepticism extends beyond the workplace
into other aspects of Swiss daily life. A study by the University of Zurich’s Research
Center for the Public Sphere and Society (fög) showed that merely 29% of
participants would engage with news articles authored entirely by AI, in stark
contrast to the 84% who would opt for content crafted by journalists20.
Playing the Long Game Can Switzerland lead the way in generative AI? 21
Rapid technological advancement
The pace at which AI and gen AI technologies evolve far exceeds the speed
at which regulatory frameworks can be developed and implemented, leading
to a perpetual catch-up scenario for regulators.
Social expectations and ethical implications
Gen AI raises complex ethical and social questions, connected to bias,
privacy, and the potential for job displacement. Developing regulations
that effectively address these concerns without stiffing innovation is a
Regulatory
delicate balance. In particular, the expected workforce shift will likely lead
to substantial changes in job roles within the next few years, adding societal
readiness pressure regarding skill training and policy development.
International coordination
The global nature of gen AI development and deployment necessitates
international collaboration and harmonization of regulatory standards.
Switzerland’s active participation in international discussions and bodies, such
as the Council of Europe’s Committee on Artificial Intelligence, highlights the
importance of global cooperation. However, aligning international norms with
national regulations presents a challenge.
Playing the Long Game Can Switzerland lead the way in generative AI? 22
Enterprise readiness Workforce readiness Regulatory readiness
Our outside-in analysis indicates that top
Swiss companies have room to bring their
use of AI to a higher level in some areas.
Playing the Long Game Can Switzerland lead the way in generative AI? 23
Enterprise readiness Workforce readiness Regulatory readiness
Swiss companies AI index
Our AI index, an outside-in analysis that measures a company’s
(Average percentile rank vs global industry peers, 23 Swiss players, 2023) level of advancement in its AI journey, highlights the presence of
pioneering companies that excel. However, a wide interquartile range
100 across several indices suggests that many firms still have significant
untapped potential to harness. The companies analyzed exhibit a
90
robust tech foundation with a significant median score, indicating
a well-established IT infrastructure and a willingness to embrace
80
emerging technologies, which is critical for AI development. In
strategic communications, Swiss companies demonstrate a proactive
70
approach to AI, with 35% mentioning AI at least once in
a strategic context during their earnings calls.
60
50 In responsible AI, Swiss companies exhibit a dynamic range, with
some demonstrating commendable ethical AI practices while others
40 have yet to reach such standards. This variance presents a dual
challenge and opportunity – encouraging a universal commitment
30 to ethical AI can propel Swiss firms to the forefront of responsible
innovation and serve as a beacon for global standards. The index
20 highlights a critical gap in AI talent. While there are standout players,
the data reveals that on average, only 4% of job postings mention AI
10
skills. This indicates a potential shortfall in the required
skill sets for advancing AI technology.
0
Tech Responsible AI Board tech
foundation AI patents quotient
AI strategic AI AI Workforce
This graph shows how Swiss companies compare with global
mentions talents VC – M&A quality
industry peers across various AI index pillars. For example, Swiss
companies have a median score of 43.5 in the AI talents pillar,
Top quartile Bottom quartile Median indicating they surpass 43.5% of their worldwide competitors in
Note: Strategic mention is defined as any reference to AI-related terminology during earnings calls that link AI their ability to attract and retain AI talents.
to one or more specific categories: future trends, strategy, investment, use cases, risk, and human capital.
Source: AI Index, Accenture Research.
Playing the Long Game Can Switzerland lead the way in generative AI? 24
Enterprise readiness Workforce readiness Regulatory readiness
Expected / actual timing to fully scale up generative AI enterprise-wide
% of respondents
C-suite leaders are less confident and
Global 9% 69% 20% 3% more conservative in their timelines for
scaling generative AI enterprise-wide
compared to their global counterparts.
Switzerland 2% 61% 36% 1%
Only 2% of Swiss companies say they are currently
scaling up generative AI enterprise-wide and a
significantly higher percentage of Swiss leaders
The organization has the right data strategy and core digital capabilities than those globally expect this integration to take
(e.g., the use of structured, unstructured, and synthetic data) in place to between 12 and 18 months. Furthermore, fewer
effectively leverage generative AI Swiss leaders are “extremely confident” in their
data strategy and digital capabilities to effectively
% of respondents saying “extremely confident”
leverage generative AI than leaders globally.
Global 13%
Switzerland 7%
Currently scaling up generative AI enterprise-wide Within 12 months
12 to 18 months Longer than 18 months
Source: Accenture Pulse of Change, March 2024. n (Global) = 2,800, n (Switzerland) = 100
Playing the Long Game Can Switzerland lead the way in generative AI? 25
Enterprise readiness Workforce readiness Regulatory readiness
Swiss employees’ view on generative AI
% of respondents
Swiss workers
Swiss employees are highly receptive to generative AI ... but while their optimism is evident, they maintain a
are highly open
technology, recognizing its value and showing a cautious stance on job security, work quality, and overall
to generative AI,
willingness to acquire new skills ... well-being.
but companies
should address
their remaining
see value in working say it could add to their concerns: stress,
92% 54%
with gen AI stress and burnout
job displacement,
and output
accuracy.
want to learn new are concerned about job
93% 48%
gen AI skills displacement
are already using gen AI at work are concerned about accuracy
85% 50%
in some fashion of tool output
… and while only 30% of Swiss organizations are currently reskilling their workforce to meet growth goals,
91% recognize the necessity to revise their reskilling strategies in response to generative AI.
Source: Accenture Change workforce survey, Oct.–Nov. 2023, n (Switzerland) = 250; Accenture Pulse of Change, March 2024. n (Global) = 2800, n (Switzerland ) = 100
Playing the Long Game Can Switzerland lead the way in generative AI? 26
Enterprise readiness Workforce readiness Regulatory readiness
Number of global AI-related policy initiatives over time
0 50 100 150 200 250
The regulatory focus on AI has dramatically
2005 or
27 increased at a global level in the last decade.
before
06 1
As of 2023, the OECD counted more than 1,000 AI-related policies
07 1
globally, reaching a notable peak in 2019. Since 2019, there has
10 2 been a slight slowdown, but still a high level of activity in AI policy
introductions, indicating sustained interest and investment.
11 5
12 4
National AI policies are the most widespread type of AI-related policies
13 8 (70%), with all the 71 countries analyzed placing a strong emphasis on
developing national policies addressing AI, directly or indirectly.
14 13
Over
1,000 15 12 27% of all policies focus on trustworthy, human-centric AI, with greater
AI-related 16 37 emphasis in APAC and North America on this theme.
policies
17 57
Only 10% of overall policies focus on AI coordination and monitoring,
18 141
suggesting that some countries might be weaving these efforts into
19 220 broader digital governance strategies, indicating a holistic approach
to technology policy, or that some countries may have a decentralized
20 191
approach, focusing on individual policy initiatives without an overarching
21 136 coordination mechanism.
22 85
23 64
European Lighthouse on Swiss
secure and safe AI Supercomputer
Source: Accenture Research on OECD AI Policy observatory.
Data related to 71 countries analyzed (including European Union)
Digital society
initiative
Interdepartmental
working group on AI
Guidelines
on AI
Digital
Switzerland
Playing the Long Game Can Switzerland lead the way in generative AI? 27
Countries have diverse approaches to regulating AI:
some approach this horizontally, while others see more
benefit in industry-specific regulation
Market-driven approach
A regulatory strategy emphasizing innovation and economic growth by minimizing
government intervention in the development and application of AI technologies.
Risk-based approach
A regulatory strategy that focuses on identifying, assessing, and mitigating
potential risks associated with AI technologies to protect consumers and society.
Horizontal approach
The regulatory framework covers a broad range of issues, from AI development
to economic impact, in one document.
Vertical approach
The approach implements various regulations focused on different aspects,
or types of AI.
nevird-tekraM
hcaorppa
Vertical
approach
desab-ksiR hcaorppa
Enterprise readiness Workforce readiness Regulatory readiness
Horizontal
Source: Accenture Research on HSBC, AI Regulation, Assessing impact on companies, Feb. 2024; Accenture Research analysis approach
Playing the Long Game Can Switzerland lead the way in generative AI? 28
Enterprise readiness Workforce readiness Regulatory readiness
Qualitative classification of selected countries’ AI-related policies approach
Final text of the EU AI Act approved Bill No. 2338/2023 has the goal 2023 pro-innovation approach to
in March to provide risk-based of establishing detailed rules, AI regulation white paper 2023.
classification to ensure safety and principles, and guidelines for the
compliance with fundamental rights. development and application of AI Five cross-sectoral principles for
in the country. regulating AI on a non-statutory basis
The AI Act also applies to providers and to be applied by different sector
developers outside of the EU whose AI regulators.
systems affect EU individuals. AI laws are distributed across federal
agencies and state-level regulations
(e.g., California on gen AI). AI policy in 2019 and the Model AI
Artif |
271 | accenture | Accenture-Banking-Top-10-Trends-2024.pdf | Banking on AI
Banking Top 10 Trends for 2024
Introduction
The Digital Age revolutionized
banking; expect even more
from the Age of AI
A quarter of a century ago we stood on the interactions, today deal with only a tiny As we enter the Age of AI,
threshold of the Digital Age. Amazon had just proportion. The use of cash declined as new
many bankers feel the
made the bold decision to broaden its sales ways of paying emerged. With technology
same sense of awe that
catalog beyond books, Google was launched having become a critical differentiator, and
to help us find our way around a rapidly with almost $550 billion2 invested in the their counterparts did a
expanding internet, and we were blissfully fintech sector since 2010 alone, the industry quarter of a century ago.
unaware that the dot-com bubble was about experienced an influx of digital-native
to burst. A few years earlier, expecting digital to competitors. These included both agile
displace our industry’s incumbents, Bill Gates start-ups and bigtechs with deep pockets,
famously declared: “The world needs banking, huge customer bases, troves of data and
but it does not need banks.”1 unmatched technological expertise.
Digital didn’t disappoint us. The past 25 years Yet despite their best efforts, no fintech has
saw a revolution in how companies work and managed to break into the global top-250 list
the products and services they offer. Banks of banks by assets.3 It appears the world does
changed fundamentally. Their branches, need banks after all.
which used to handle virtually all customer
Banking on AI | Banking Top 10 Trends for 2024 2
Introduction
Today, we again stand on the verge of transformational change.
The ability to process and analyze vast stores of data, the enabling
power of cloud, and the rapid maturation of artificial intelligence
are combining to create a wealth of opportunities for enhancement
and innovation across organizations’ operations, workforce,
products and experiences.
As we enter the Age of AI, many bankers feel the same sense of
awe that their counterparts did a quarter of a century ago. They
know that, as with digitalization, very little will remain untouched.
These technologies are unlikely to change what banking does,
but they will dramatically transform how it does it.
Each of the trends we describe in this report is either caused or
amplified by AI. We, together with most bankers today, are peering
into the future: trying to figure out what this technology holds for
the industry. We are confident the Age of AI will change banking
and many other industries; exactly how, we will only know in
retrospect. However, it is we who get to choose where and how
we will use AI. The challenge is to ensure it’s a force for good that
benefits all humankind.
Banking on AI | Banking Top 10 Trends for 2024 3
Introduction
Our Top 10 Banking Trends.
01 02 03 04 05
The rise of Capturing All the risk A whole The power
gen AI the digital we cannot see new way of of pricing
dividend working
06 07 08 09 10
Time Regulation From The key Beyond
to think recalibrated technology to to the core Six Sigma
cloud first engineering
Banking on AI | Banking Top 10 Trends for 2024 4
Trend: 1
The rise
of gen AI
Banks are likely to benefit more from generative
AI than any other industry. Our analysis of operational
efficiency indicates a potential to boost productivity by
22-30%,4 while a further study found that revenue could be
increased by 6%.5 To achieve these improvements, however,
it will be necessary not only to utilize the cloud and data
effectively, but also to fundamentally rethink work and talent.
Trend 1 | The rise of gen AI
“AI will fundamentally
transform everything, from Sweeping statements like this are usually given little credence in the sober
world of banking. But that was before generative AI came along. Suddenly
business to science to
all bets are off, and bankers throughout the industry are wondering
6
society itself.” whether there is any part of the business that won’t sooner or later be
affected, if not actually transformed.
With good reason. We recently analyzed 19,265 tasks across 900 job
families in 19 industries, using data from the US Bureau of Labor Statistics
and others. The study included a breakdown of the time spent on each task
and an assessment of the potential for automation and augmentation by
generative AI. We concluded that banking is likely to be more extensively
impacted than any other industry, with almost three-quarters of all work
being well-suited to automation or augmentation (Figure 1).
Banking on AI | Banking Top 10 Trends for 2024 6
Trend 1 | The rise of gen AI
Figure 1. Banking is likely to be more profoundly impacted by gen AI than any other industry.
Work time distribution by industry and potential impact of LLMs.
Higher potential Higher potential Low potential for
for automation for augmentation automation or augmentation
Banking 39% 34% 27%
Insurance 33% 37% 30%
Capital Markets 32% 37% 31%
Software & Platforms 31% 37% 32%
Health 42% 25% 33%
Communications & Media 34% 31% 35%
Retail 36% 28% 36%
Life Sciences 34% 29% 37%
High Tech 31% 31% 38%
Travel 35% 27% 38%
Automotive 34% 27% 39%
Public Service 34% 26% 40%
Energy 35% 23% 42%
Utilities 34% 23% 43%
Industrial 33% 24% 43%
Consumer Goods & Services 32% 24% 44% Note: Weighted by employment levels
in the US in 2022. Estimates are based
Aerospace & Defense 30% 26% 44% on human + machine identification
of the exposure of work tasks to the
Chemicals 31% 22% 47% impact of generative AI.
Source: Accenture Research based
Natural Resources 31% 19% 50%
on US BLS and O*Net.
Banking on AI | Banking Top 10 Trends for 2024 7
Trend 1 | The rise of gen AI
AI has of course been around for a long time; most tech
Banking 22% 30%
historians credit the English mathematician and cryptanalyst
Insurance 20% 28%
Alan Turing with having developed the concept in 1950. What is
new is that cloud-based generative AI engines have reached the Capital Markets 19% 28%
point where they are surpassing human capabilities in important
Software & Platforms 19% 27%
respects. These progressively adaptive engines are advancing
Communications & Media 14% 20%
at an unprecedented speed, arousing both wonder and alarm
in most parts of business and society. Life Sciences 14% 20%
High Tech 14% 20%
Within months of the launch of ChatGPT at the end of 2022, early Figure 2. Banks can improve
Retail 13% 19%
adopters in the banking industry were exploring the most promising their productivity by up to 30%
use cases. Today, little more than a year later, virtually every bank Public Service 13% 18% by adopting generative AI.
has a generative AI strategy of some description and is running a
Travel 12% 17%
Potential hours saved by industry,
variety of proofs of concept. Many are reporting impressive results.
Energy 12% 16% valuated at US annual occupation
The next 12 months will see scaled adoption across multiple parts
of the organization, with the more ambitious banks using it as the Utilities 11% 16% headcount and wages of 2022.
US value only.
foundation for what we call Total Enterprise Reinvention.
Aerospace & Defense 11% 16%
Note: Estimates are based on human
Our analysis indicates that there are hundreds of use cases for Health 10% 16% + machine identification of work tasks
exposure to the impact of generative AI.
generative AI in banking. Productivity is the most obvious benefit. Industrial 11% 15%
Source: Accenture Research based on
As Figure 2 shows, there is greater potential to boost output in US BLS and O*Net data.
Automotive 11% 15%
banking than in any other industry.
Chemicals 10% 14%
Consumer Goods & Services 9% 13%
Natural Resources 9% 12%
Banking on AI | Banking Top 10 Trends for 2024 8
Trend 1 | The rise of gen AI
These gains are being realized in a wide Functions other than sales, marketing and
variety of areas, from due diligence and risk
customer interaction that are likely to receive
and compliance to legal contract generation
and code writing. However, we believe early attention are risk management and
the most significant financial impact will
compliance, technology, HR and legal.
be in helping banks increase revenue. Our
models show that by pairing AI with people
to offer personalized wealth advisory, guide Generative AI offers CEOs the chance to reshape their
commercial relationship conversations, tailor bank, empower their people, amplify their productivity and
products for individual customers, enhance increase profitability. But most executives recognize that it
the quality of contact center interactions, cannot do this on its own; to realize its full potential it needs
and streamline their product application and to work in tandem with human ingenuity. For this reason
onboarding processes, banks can improve alone, any AI strategy needs to have the workforce at its
their revenue by 6% or more within three core. The successful deployment of AI not only demands
years.7 a set of skills that few banks have in sufficient numbers,
but also requires significant changes in what people do
and how they do their work. Banks that manage this aspect
effectively will have a big advantage as they explore and
unravel the exciting possibilities of AI.
Banking on AI | Banking Top 10 Trends for 2024 9
Trend: 2
Capturing the
digital dividend
While most banks have mastered digital, its focus—more
often than not—has been on servicing. Turning even a
modest number of digital interactions into opportunities
holds immense potential. To do that, banks will need to find
ways to have meaningful conversations with customers
across digital channels. AI may hold the key.
Trend 2 | Capturing the digital dividend
After a quarter of a century of digitalizing their operations, channels,
and experiences, with a strong focus on servicing journeys, banks
can congratulate themselves for having mastered digital.
Virtually every bank has a mobile app that half of which are from their primary bank— BBVA is one bank that has succeeded at this.
works effectively: it manages the majority of 73% acquired at least one financial services By 2017 it was using its digital channels for most
customer interactions, is typically rated well product from a new provider in the past of its customer servicing, but for only 25% of
over 4 out of 5 by customers and—together 12 months.9 product sales.* Five years later the picture had
with digital enhancements elsewhere in the changed: 61% of its sales were closed on the
organization—continues to deliver big efficiency Digitalization has improved banks’ ability bank’s digital channels, and its cost-to-income
gains and convenience for customers. to solve customers’ most basic needs, but ratio had fallen from approximately 50% to
conversations about their financial aspirations 43% (see also page 13).
Yet there have been unwelcome side-effects. and how the bank can help them achieve their
By shifting customer engagement out of the goals have become increasingly rare. Yet the To increase their percentage of digital sales,
branch and onto their digital channels, banks’ goal of increasing the proportion of digital banks are getting better at personalizing their
experiences have become functionally correct sales depends on it. interactions. Like many service providers, Bank
but emotionally void. And at the same time as of America asks customers for feedback every
their personal connection with customers has The good news is that customers still trust time they engage with the organization. It now
weakened, so has banks’ ability to differentiate banks and are sending them clear signals of has more than 50 million responses. But instead
themselves: Accenture’s Life Trends 2024 what they want. To capture the full potential of just aggregating that data to gain a better
survey8 found that 42% of consumers find it of digital, banks need to improve their ability understanding of its customer base as a whole,
hard to distinguish between financial services to respond to these signals. This includes the bank’s primary aim is to focus on individual
brands. In the process, customer loyalty shifting their thinking about digital from customers: how they feel, what they want, and
has weakened. The average consumer “servicing” to “conversations”. how their experiences could be improved.10
has 6.3 financial services products, only
* M easured by the percentage of total lifetime economic value of all products sold. Banking on AI | Banking Top 10 Trends for 2024 11
Trend 2 | Capturing the digital dividend
Currently, as our 2022 analysis of 41 leading banks to gain a better understanding of each Bank customers, in the
banks across ten markets shows, less than customer’s circumstances, and to reach out
past 12 months, used
15% of them provide comprehensive rewards proactively with empathy, timely advice and
branches more than any
for customers who increase the number relevant offers. We call this approach ‘life-
of products and services they use or the centricity’. When you feel recognized and other channel to open
transactions they conduct with the bank. appreciated, why would you buy elsewhere?
accounts, get advice and
The ability to treat each customer as an
acquire new products.
individual can make a big difference to As banks commit to having conversations
both the customer and the bank, but too with customers, the logic of life-centricity Almost 2 out of 3 turn
often personalization goes little further rather than product-centricity becomes more to branches to solve
than delivering banner advertisements. compelling, and we expect to see corporate
specific and complicated
structures changing to reflect this. This will
problems.
In 2024, a growing number of banks will have many benefits, for both parties. When
seek to realize a greater return on their the banking app—consumers’ second-most
Source: Accenture Global Banking
investment in digital by using their vast important consumer technology after their
Consumer Study, 2023.
stores of customer data and advanced car8—becomes more than just a means of
analytics and AI capabilities to move beyond checking account balances and making
basic demographic segmentation and start payments but provides a steady flow of
treating customers as individuals. This will valuable, tailored advice and propositions,
not only make customers feel more special, the relationship between the two becomes
increasing their loyalty. It will also allow these more trusting, durable and productive.
Banking on AI | Banking Top 10 Trends for 2024 12
Trend 2 | Capturing the digital dividend
BBVA is a good example of a bank that has transformed its
operating model to (among other things) develop an end-to-end
personalization capability, optimize its customer experiences,
and improve the effectiveness of its customer acquisition and
cross-selling. Just one of the metrics it has announced is a 30%
improvement in its conversion rate for auto-loan sales.11
The ultimate objective is to offer the same authentic, personal
experience through digital channels as banks have always done
face-to-face in their branches. Commerzbank believes its new
mobile virtual assistant will do this, enabling private and small-
business customers to have natural and engaging conversations
on general topics as well as for financial advice.12 By combining the
convenience and efficiency of digital with the contextual relevance
that comes from a deeper and more timely understanding of each
customer, banks will be able to shift a growing proportion of their
sales to digital while simultaneously reinforcing trust and loyalty.
This is the digital dividend they have been
pursuing for so long.
Banking on AI | Banking Top 10 Trends for 2024 13
Trend: 3
All the risk
we cannot see
In 2024, banks will be confronted by a variety of
risks: some familiar, others less predictable. We
have identified five that we think deserve attention.
Planning for the unplanned will pay dividends.
Trend 3 | All the risk we cannot see
With hindsight, all risks are obvious.
Yet as we entered 2023, no one foresaw
that a bank failure in California would
escalate into a regional banking panic
and ultimately lead to the merger of
Switzerland’s last two major banks.
Given the far-reaching consequences of events such as these,
banks need to improve their planning for risks we cannot always
see. This is especially true as stability continues to elude the
industry and the markets it serves. In our latest Risk Survey,
72% of senior banking risk professionals said their organization’s
risk management capabilities and processes have failed to
keep pace with the rapidly changing risk landscape.13
Banking on AI | Banking Top 10 Trends for 2024 15
Trend 3 | All the risk we cannot see
It’s obviously impossible to know 01
exactly what risks 2024 will bring,
but here are a few ideas to get the Banks have invested vast amounts
in bolstering their cyber defences.
conversation started:
However, in November last year, a ransomware attack on the US
subsidiary of the Chinese bank ICBC nearly crashed the US 30-year
Treasury auction and forced participants to trade by using USB pen
drives.14 The advent of generative AI has handed hackers another potent
weapon, enabling them to attack all of banks’ surfaces with deep fakes
that can deceive voice analysis and other defences, amplify phishing
attacks, and create much more complex and elusive viruses. In 2024, as
the likelihood of such attacks succeeding edges toward the inevitable,
banks will shift the focus of their strategies from prevention to resilience.
They too will use generative AI—not only to detect attacks but also to
increase the frequency, depth and scope of their scenario planning,
and to look not only at the immediate implications of a cyber breach
but also the second- and third-order effects—and how they should
prepare and respond.
Banking on AI | Banking Top 10 Trends for 2024 16
Trend 3 | All the risk we cannot see
02
Almost 17 years of near-zero rates has
Figure 3. The average house price has risen more than
caused house prices to rise strongly.
personal disposable income.
There is a growing risk of stressed customers defaulting on their
160
mortgages as rates remain high and salary increases fail to offset Evolution of house prices and personal disposable
consumer price inflation. In a sample of Western markets, the rise in income across selected major economies*
150
the price of houses has significantly exceeded the average growth in Indexed: 2013 Q1 = 100
household disposable income since 2015 (Figure 3). As rates remain House price
140 index
elevated and low pre-Covid mortgages roll off, the risk of stressed
consumers defaulting rises, even where unemployment is low.
130
The question then is: will governments allow large numbers of
120
employed but hard-pressed home-owners to lose their properties or
will we see some interesting public/private partnerships—the Canadian Personal disposable
income index
government is already talking about interventions to help citizens 110
crushed by rising rates.15 In our 2023 Global Risk Survey, only 35% of
172 banking executives said their organization is fully able to assess 100
2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023
the risks associated with interest rate increases.16 This alone suggests
a low level of readiness to intervene if the situation turns ugly. *Overall indices calculated as simple averages of house price and personal disposable
income indices for: Australia, Belgium, Canada, Germany, Spain, France, UK, Italy,
Switzerland, Netherlands and US
Source: Accenture Research based on Federal Reserve Bank of Dallas
Banking on AI | Banking Top 10 Trends for 2024 17
Trend 3 | All the risk we cannot see
03
Figure 4. Commercial real estate exposure constitutes a significant share of
GDP and of banks’ and other financial institutions’ balance sheets.
The status of commercial real estate (CRE)
CRE debt as % of GDP Bank loan exposure to CRE
is similarly precarious. % of total assets, Dec 2022
18%
A lot has been written about it recently, and the bankruptcies 12% 0% 2% 4% 6% 8% 10% 12%
of Signa Development17 and WeWork have highlighted
Sweden
what may be the most publicized risk in waiting. As with
mortgages, 15 years of near-zero rates followed by a sudden USA
rise, combined with a shift to work-from-home, has left many US Europe
Norway
commercial property developers and real-estate owners
Owners of Netherlands
in a perilous position. It is a global risk, and CRE debt and
CRE debt Others
equity are held not only by banks but also by other players US only Germany
throughout the financial industry—often beyond the scope 13%
Belgium
of regulators (see Figure 4). Commercial
mortage-backed 14% 38% Banks Australia
securities
Italy
15%
Insurance Spain
21%
France
Agencies and government-
sponsored entities
Source: Accenture Research based on IMF: Global Financial
Stability Report, October 2023, and Reserve Bank of Australia:
Financial Stability Risks from Commercial Real Estate.
Banking on AI | Banking Top 10 Trends for 2024 18
Trend 3 | All the risk we cannot see
04
Figure 5. Non-bank financial institutions hold nearly
60% of the private sector’s total global financial assets.
$ trillions. Financial assets held by central banks and $422
The rise in shadow banking.
$396
public financial institutions are excluded.
$361
In the aftermath of the 2008 Financial Crisis, off-balance-sheet
$332 $335
lending became a priority for regulators, who introduced $315
waves of Basel regulations as well as many local measures. $294
$284
This caused banks to dial back their risk. But the question is: $263
$251
has that risk gone, or have we just moved it out of sight? Banks $224 $236
$209 $210
hold less than 50% of financial assets (Figure 5) and the share 57%
56%
of US non-bank mortgage origination has ballooned from
56%
9% in 2010 to 62% in 2022.18 Is anyone monitoring that risk, 55% 56% 55%
54%
54%
and what would the inevitable fallout be for banks, insurance
53%
51%
companies and pension funds should this turn bad? 50% 49%
48% 50% Banks
Non-bank
financial
institutions
2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021
Source: Accenture Research analysis based on Financial Stability Board, “Global Monitoring Report
on Non-Bank Financial Intermediation 2022”
Banking on AI | Banking Top 10 Trends for 2024 19
Trend 3 | All the risk we cannot see
05
Our aim is not to be a banking Nostradamus,
implying that we can see and evaluate all major
China’s growing involvement in the
risks. We’re simply making the point that banks
economies of most countries, and its
face a large and varied array of risks, some of
concerted effort to attract foreign
which have been publicly scrutinized while
investors, is another risk that
others are hidden in plain sight. Many have the
warrants scrutiny.
potential to cause extensive damage. To protect
The government has worked hard in recent years to
themselves and their customers, banks need to
strengthen its regulatory regime, but the fact that its
improve the frequency, depth and scope of their
residential property sector in particular is so heavily
leveraged, and that developers like Evergrande were scenario planning, using real-time data.
allowed to run up liabilities of approximately US$300
billion,19 show that the risk is very real. If the mounting
We believe that in 2024, these scenarios
debt burden is a bubble, and if the authorities fail to deal
will inform more board conversations and
with the threat, the fall-out for global banks as well as
economies worldwide could be severe. guide more strategic decisions.
Banking on AI | Banking Top 10 Trends for 2024 20
Trend: 4
A whole new
way of working
The way banks work is about to change radically. New
skills, approaches and mindsets will be needed, not only
in IT but—more critically—in every function and level of the
bank. The challenge is way bigger than recruitment alone
can solve. An entirely new strategy is called for.
Trend 4 | A whole new way of working
The digitalization of banks’ operations over the
past 25 years caused an escalation in what was
commonly dubbed ‘the war on talent’.
There is no doubt competition for The challenge goes beyond this, however, and of possibilities for banks to generate new
high-end technical skills will intensify in is different than during the Digital Age. With value for customers, more rewarding
2024 as every financial institution, and digital, banks hired specialist teams to develop work for employees, and growth for the
indeed every organization on the planet, their online and mobile banking applications. organization. To seize this opportunity,
advances its strategy to capitalize on AI, Because AI will impact nearly every job in leaders need to reimagine the future of
cloud, and data analytics. every bank, recruitment simply won’t work. human + machine work, starting with a
Banks will need to create a culture of curiosity, blank slate. They are starting to think about
Some leading banks, including Lloyds Banking receptiveness to change and continuous how generative AI should be integrated
Group20 and Banco Santander,21 are investing development—one that encourages and into every role and function, and how their
heavily in their captive IT organizations. They enables all employees to reinvent their workforces and culture will change as
are recruiting and training aggressively to roles and, indeed, themselves. the technology automates much of the
acquire the experts they need as they scale necessary work and elevates human skills
the roll-out of AI. However, demand is likely The Digital Age saw IT teams designing and such as strategic and creative thinking,
to greatly exceed their availability. In addition, building websites and mobile apps, but it judgement and relationship building.
the most talented among them will prefer to barely changed the work that most banking
work for firms that can offer careers leading professionals did. Generative AI, on the other
to leadership roles. Most banks will therefore hand, will change what people do and how
need an alternative approach. they do it. In the process it will open a world
Banking on AI | Banking Top 10 Trends for 2024 22
Trend 4 | A whole new way of working
Our 2022 Future of Work survey22 found new human roles that include the introduction, Only
that only 26% of bank CEOs had a future- management and governance of this innovation.
ready strategy that was holistically focused Less obvious, but just as important, is how 26%
on changing how, why and where their people will work alongside the machines to
people work. This is sure to change swiftly preserve the human face of the bank: be
as organizations develop ambitious plans available to customers, maintain relationships,
around AI. It is important that this strategy and show genuine empathy as they help
of bank CEOs have a
concentrates not only on the necessary to address their concerns.
future-ready strategy.
changes in roles, tasks and skills, but also
on how generative AI is likely to change It is only when the human + machine workforce
the soul of the organization. is expanded and enhanced in such a holistic and
human-centric way, and when HR and change
We have been warning for years that banks, professionals are fully involved in shaping the
in their well-intentioned drive to digitalize, transformation, that the full potential of
have become remote, impersonal and generative AI will be within banks’ reach.
undifferentiated. Generative AI could
exacerbate that. As banks define the
objectives of their generative AI
transformation, they are envisaging
Banking on AI | Banking Top 10 Trends for 2024 23
Trend 4 | A whole new way of working
OCBC putting gen AI to work
Singapore’s OCBC Bank, a generative AI trailblazer, has
completed a six-month trial of an intelligent chatbot and is
now rolling it out to all its 30,000 employees to help them
write, translate, research and innovate. Participants said
they were able, on average, to do their work 50% faster—
which included the time taken to check the accuracy of the
bot’s output. An earlier trial, to develop code, summarize
documents, transcribe calls and create an internal
knowledge base, boosted productivity by a similar amount.
The bank currently uses AI to make more than four million
decisions daily in risk management, customer services and
sales—and expects this to increase to 10 million by 2025.23
Banking on AI | Banking Top 10 Trends for 2024 24
Trend: 5
The power
of pricing
Banks have always known that optimized pricing can have
a huge impact on their top and bottom lines. This year, they
are starting to combine intuition with generative AI and more
current and comprehensive data to turbo-charge scenario
planning and move closer to personalized pricing.
Trend 5 | The power of pricing
Every businessperson knows that a small
change in price can have an oversized effect
on demand, revenue and income.
In banking, all things being equal, a Despite years of talk about “hyper- smaller and smaller groups to find the perfect
1% increase in revenue translates into a personalization”, banks’ pricing has always solution—similar to how Isaac Newton used
~40 bps improvement in pre-tax ROE. A been characterized more by consistency and calculus to measure the area under a curve.
1% improvement in cost, however, only simplicity than the ability and willingness of Unfortunately, until now, banks haven’t been
improves ROE by ~25 bps.24 individual customers to pay. What’s more, able to approximate Newton’s precision as he
with interest rates having been stuck virtually conceived of infinitely smaller spatial figures.
The challenge, however, has always been at zero for the past 15 years, there was little This has meant that, for many customers, their
to predict the impact of a price change on benefit to be gained by improving prices were wide of the mark.
revenue. Economists can plot graphs showing the sensitivity of pricing.
the price elasticity of demand, but they In the future, AI will play a major role in bringing
can seldom take account of all the relevant In 2024 we will see the beginnings of a pricing to perfection. It will consider thousands
variables and offer more than an averaged change in all this; a different approach to of variables to rapidly come up with a perfect
view of a customer base or market. Which pricing and sales that could be one of the price for retail and commercial customers—
means that a banker who sets a price will hope most important contributions of generative either individuals or small segments with very
that it works for most customers but will know AI to corporate profitability—as well as similar needs. It will measure the outcome,
that for a significant proportion it is too high, customer value. In theory there is a perfect feed it back into its calculations along with
and there’s a risk of attrition, while for another price for each combination of customer, competitive data and other changes, and
group it is less than they would be willing to product, and channel. Ideally, banks would adjust in real time.
pay, which represents a revenue forfeit. like to price customers in increasingly
Banking on AI | Banking Top 10 Trends for 2024 26
Trend 5 | The power of pricing
The new prices can be delivered health, and then shares the value this creates is that the benefits are mostly passed back to the
automatically to all customers, together through personalized interest rates and other customer. In this case, the race to perfection will
with tailored incentives for saving more or rewards. “It’s simple,” the bank states. “We initially advantage the early adopters and ultimately
subscribing to more products. These could believe that we’ll do well when our clients the banking customers. However this plays out,
be promoted through personalized marketing do well, and society will benefit too.”25 pricing is likely to receive a lot more attention
scripts, also crafted by generative AI. With as generative AI matures.
millions of iterations, and the ability to learn Dynamic pricing has always been possible,
from each, banks should soon be able to but it has mostly depended on intuition.
zero in on the perfect price. In the future, banks will price their services
with a greater understanding of how each
They will also be able to execute their business variable affects the outcome in relation to
strategies with more precision: set prices that each customer. Some may use the ability to
find the ideal balance between profit, growth maximize short-term profits, while others will Dynamic pricing has
and customer value, and between short-term test innovations and drive growth; another always been possible, but
and longer-term objectives. B |
272 | accenture | Accenture-POV-Reinventing-Life-Sciences-Age-of-Gen-AI-28Aug2024.pdf | Reinventing life
sciences in the age
of generative AI
Contents
04 07 10 41
19
Executive Supercharging AI is revolutionizing Five C-suite Start your
summary science with intelligent the pharmaceutical imperatives reinvention journey
technologies: value chain
This is not your typical
technology upgrade
IA
evitareneg
fo
ega
eht
ni
secneics
efil
gnitnevnieR
Contents Executive summary Supercharging science Revolutionizing the value chain Five C-Suite imperatives Start your reinvention journey 2
AAuutthhoorrss
Petra Jantzer Selen Karaca-Griffin Kailash Swarna Tracy Ring Jen Spada
Senior Managing Director – Thought Leadership Principal Director – Managing Director – Chief Data Officer and Global Generative Managing Director –
Global Life Sciences Lead, Products and Life Sciences, Global Clinical Lead, AI Lead – Applied Intelligence Products, Global Generative AI Strategy Lead,
Accenture Accenture Research Accenture Life Sciences Accenture Life Sciences Accenture
Petra Jantzer is the Global Lead for Selen Karaca-Griffin is the Global Research Kailash is the Global Life Sciences Clinical Tracy Ring serves as the Chief Data Jennifer is a Managing Director in
Accenture Life Sciences and the senior Lead for Accenture Products and Life Development Lead at Accenture, driving Officer and Global Generative AI Lead Accenture's Life Sciences practice,
Client Account Lead for one of the world’s Sciences, leading a team of 30+ researchers digital transformation strategies to boost for Accenture's Data & AI Life Sciences spearheading the life sciences innovation
leading pharmaceutical companies. globally. She is responsible for developing R&D productivity for global life sciences division. With more than two decades of team to develop cutting-edge solutions
She was previously the industry leader for the industry’s thought leadership agenda, clients. With over 20 years in the field, experience, she has crafted AI strategies that advance client initiatives. With a robust
R&D and Europe Life Sciences and a former which includes scientific innovation, science he has extensive experience across for numerous organizations, orchestrated background in the industry, her expertise
partner at McKinsey. Petra has 20+ years and technology convergence, digital health, drug discovery, translational sciences, large-scale transformation deals, and spans commercial transformation, strategy,
of industry experience, holds a Ph.D. market disruptions and their impact on the clinical development, pharmacovigilance, facilitated extensive platform ecosystem operations, patient services and digital
in tumor immunology and specializes in future of industries. She is based in Boston, regulatory affairs and commercialization. partnerships. In her current role, Tracy marketing. Additionally, Jennifer directs
cross-functional transformation programs. Massachusetts. Selen holds two BS degrees Kailash also engages in research at MIT's leverages generative AI to revolutionize the Accenture’s patient research activities.
She is president and co-founder of in molecular biology and chemistry, an Sloan School of Management, focusing life sciences sector, providing guidance to She earned her Bachelor of Science in
Advance – a cross-industry association in MS degree in biotechnology, and an MBA on financial engineering to enhance R&D CDOs and Chief Analytics/AI Officers on biological engineering and a MBA from
Switzerland dedicated to driving gender from Babson College. She also serves as decisions. He holds an MBA from MIT and a maximizing strategic value and impact in Cornell University.
equality in business. a Biotechnology Industrial Advisory Board Ph.D. in physical chemistry from Oklahoma commercial, R&D, supply chain and other
Member for Northeastern University. State University. enabling domains through a Responsible AI
framework. She specializes in implementing
AI solutions for regulatory submissions,
early-stage drug development, and
advancing the future of commercial and
intelligent supply chains.
IA
evitareneg
fo
ega
eht
ni
secneics
efil
gnitnevnieR
Contents Executive summary Supercharging science Revolutionizing the value chain Five C-Suite imperatives Start your reinvention journey 3
EXECUTIVE SUMMARY
Where could reinvention take your business?
What if your What if your company What if your company What if your company What if you could
company could could dramatically could rapidly optimize could dynamically use everything you
develop novel compress R&D manufacturing anticipate market know about all of your
medicines for timelines and recipes and facilitate shocks and customers — from
previously reduce the cost agile, resilient black-swan events patients to providers
undruggable targets of developing a and sustainable and respond with to payers — to truly
and address currently medicine from end-to-end supply minimal disruptions meet them where
untreatable illnesses? billions to millions chains of new to patients? they are with speed
of dollars? modalities for and efficiency?
better competitive
advantage?
Organizations are achieving exactly these kinds of breakthroughs by using intelligent
technologies such as classical and generative artificial intelligence (AI) reinventing themselves.
IA
evitareneg
fo
ega
eht
ni
secneics
efil
gnitnevnieR
Contents Executive summary Supercharging science Revolutionizing the value chain Five C-Suite imperatives Start your reinvention journey 4
In 2023, we presented Total Enterprise Reinvention for Biopharma, The Why
a strategy of changing every part of a business, adopting change
AI is supercharging science and reinventing
at scale and generating innovation, resilience and value. When business — this isn’t your typical technology upgrade.
companies engage in Total Enterprise Reinvention, we wrote,
they commit to creating a strong digital core on which they
can essentially turn “change” into a capability, such that any
transformative effort in any area of the business builds on and
The What
contributes to other efforts. The result — demonstrated by the few
AI is revolutionizing the value chain, offering strategic
leading companies we identified as “Reinventors” — is sustainable,
opportunities to generate significant value if workflows and
accelerated and efficient growth.
processes are consistently reinvented end to end.
At that time, we forecast that companies embracing the
transformative power of technology, data and AI to drive reinvention
would be ahead of the curve in the next decade and beyond.
The How
This year, the growing impact of disruptive technologies such as
Five C-suite imperatives will help you reinvent
generative AI has made it even clearer that continuous reinvention is
your business and pull ahead of the pack.
becoming the default strategy for the world's leading organizations.
In fact, our research found that the competitive edge belongs to
Reinventors, who not only define the new performance frontier
for their industries but also enjoy the largest financial benefits. In
this report, we present our recommended approach to continuous
reinvention in the era of generative AI: The Why, What and How.
IA
evitareneg
fo
ega
eht
ni
secneics
efil
gnitnevnieR
Contents Executive summary Supercharging science Revolutionizing the value chain Five C-Suite imperatives Start your reinvention journey 5
About the research
We took a comprehensive approach to study the topic of Total Enterprise Reinvention.
This report is based on:
• A multi-year survey of over 3,000 • Our annual life sciences CEO Imperatives
executives across 19 industries and 10 Research, which identifies critical
countries. Respondents were asked about disruptions and key priorities based on
their organization’s approach to business qualitative interviews with the CEOs of
transformation and reinvention strategy, the top 40 life sciences companies by
as well as about their specific programs revenue. We validate these trends in our
and success factors. The surveys were annual CEO roundtable, where industry
conducted in November 2022 and CEOs gather to discuss the industry’s
October to November 2023. In this report, most pressing issues and promising
we provide comparisons between the two, opportunities. The CEO roundtable was
focusing on new insights gained from the held at the 2024 meeting of the World
most recent responses. Economic Forum in Davos, where we
gathered industry C-suite leaders to
• The annual Pulse of Change Index that
discuss the impact of classical and
quantifies the level of change affecting
generative AI on the life sciences industry.
businesses globally, caused by six major
factors: technology, talent, economic, • Collaboration with our innovation strategy
geopolitical, climate and consumer experts and subject matter advisors to
and social. The index provides context ideate, shape and push the boundaries
supporting the need for reinvention. of our thinking on reinvention in the age
of generative AI. We then tested our
approach with multiple clients.
IA
evitareneg
fo
ega
eht
ni
secneics
efil
gnitnevnieR
Contents Executive summary Supercharging science Revolutionizing the value chain Five C-Suite imperatives Start your reinvention journey 6
7
IA
evitareneg
fo
ega
eht
ni
noitnevnieR
Supercharging science with
intelligent technologies:
This is not your typical technology upgrade
IA
evitareneg
fo
ega
eht
ni
secneics
efil
gnitnevnieR
7
IA
evitareneg
fo
ega
eht
ni
secneics
efil
gnitnevnieR
The biopharma industry is on the brink of a Consider these challenges at a high level:
groundbreaking revolution, propelled by the
• Lengthy and costly drug development: • Low growth due to patent expirations as well as
remarkable potential of intelligent technologies
The average time to bring a new medicine to market government and private market pressures:
such as classical and generative AI and next
is 10–12 years, with costs exceeding $2.6 billion. The top 20 biopharma companies (with some
generation computing. These technology
Approximately 90% of drug candidates fail during exceptions) are experiencing a low-growth period, with
advancements promise to deliver breakthrough
discovery and development, and R&D productivity has an average revenue CAGR of 4% over the next five years.2
treatments and life-changing medicines at an
remained stagnant over the past decade.1 This anemic growth is attributed to factors such as
unparalleled pace, addressing the industry’s
patent expirations, pricing pressures from governments
most pressing challenges head-on.
• Increasing complexity in manufacturing and
(e.g., Inflation Reduction Act in the US) and private
commercialization:
market forces leading to net price decreases.
As scientific progress leads to new modalities
and personalized treatments, the complexity of • High cost of capital:
manufacturing and commercializing these therapies The persistent high cost of capital is compelling CFOs
increases. Many new modalities are launched with to explore ways to enhance profitability. In addition, it
unsustainable supply chains that require years puts pressure on leaders to invest in programs that can
or even decades to optimize. generate returns faster.
IA
evitareneg
fo
ega
eht
ni
secneics
efil
gnitnevnieR
Contents Executive summary Supercharging science Revolutionizing the value chain Five C-Suite imperatives Start your reinvention journey 8
Intelligent technologies are set to transform every aspect
of the life sciences industry, from drug discovery, clinical
development and patient care to manufacturing and the reliable
supply of complex medicines. This shift promises to usher in an
era of unprecedented innovation and efficiency and will drive
better outcomes for patients.
But this opportunity hinges on a critical caveat: Unlike previous
technology transformations, a purposeful shift to intelligent
technologies requires companies to embrace and deeply embed a
culture of continuous reinvention across the enterprise.
According to our research, life sciences is one of the top two
industries most actively pursuing reinvention, the other being
software and platforms.3 Our annual CEO priorities research confirms
that harnessing intelligent technologies for business transformation
is the top priority for CEOs in 2024. This marks the first time in a
decade that technology has been identified as a standalone priority,
underscoring its pivotal role in addressing the industry’s challenges.4
IA
evitareneg
fo
ega
eht
ni
secneics
efil
gnitnevnieR
Contents Executive summary Supercharging science Revolutionizing the value chain Five C-Suite imperatives Start your reinvention journey 9
ht
ni
secneics
efil
gnitnevnieR
AI is revolutionizing the
pharmaceutical value chain
10
IA
evitareneg
fo
ega
eht
ni
secneics
efil
gnitnevnieR
1100
IA
evitareneg
fo
ega
eht
ni
secneics
efil
gnitnevnieR
To start, consider the effect intelligent technologies
are already having along the biopharma value chain:
• More than 50 drug candidates discovered • Intelligent technologies are helping leaders
with AI are now progressing through clinical better allocate capital for manufacturing,
pipelines. Molecules are being designed at supply chain and commercialization.
a fraction of the time previously required, Historical sales data, prescription patterns,
and for several targets once considered epidemiology and target population
undruggable.5 information improve forecasting. This
can inform commercial and medical team
• The analysis of historical data, literature,
positioning as well as the timing and
real-world evidence and simulations of
location of new manufacturing sites.
multiple trial scenarios all enable companies
to optimize clinical trial protocols and These are just a few examples of how intelligent
resource allocation. This ultimately shortens technologies are driving meaningful and
trial duration and reduces costs. positive changes in the biopharma industry.
A more comprehensive view of the “strategic
• Companies are leveraging advanced
bets” across the biopharma value chain is
analytics of complex chemistry and biology
shown in Figure 1. These strategic bets confer a
in recipes to achieve up to a 90% decrease
significant competitive advantage at each step
in waste production and energy and water
of the value chain. In the following section, we
consumption, while improving consistency
explore each possibility in turn.
and speed in the manufacturing of life-
saving drugs.6
IA
evitareneg
fo
ega
eht
ni
secneics
efil
gnitnevnieR
Contents Executive summary Supercharging science Revolutionizing the value chain Five C-Suite imperatives Start your reinvention journey 11
ga
eht
ni
secneics
efil
gnitnevnieR
Figure 1
Illustrative strategic bets in biopharmaceutical value chain
Process development,
Value Pre-discovery Discovery research Preclinical Clinical, Commercialization Enterprise functions
manufacturing,
chain & translational Regulatory, Safety
quality & supply chain
Basic research into Novel target discovery Prediction & optimization Optimize trial & protocol Accelerated & accurate Predictive brand & Dynamic portfolio
disease biology & new approaches to of PKPD/ADME properties design with simulation product launch portfolio strategy management and
corporate strategy
structural biology
Basic research into Site enablement Predictive manufacturing Dynamic access
Discovery, refinement &
treatment modalities and optimization process robustness optimization Proactive risk
Accelerated target development of novel
management and
validation through biomarkers
crisis mitigation
optimized pharmacology
Basic research into safety Clinical data Autonomous Real-time content
and efficacy in humans management demand sensing & supply chain & review
Expanding biobanking
SC orchestration Enhanced corporate
Modality selection
to leverage emerging
brand and reputation
& optimization
Systems approach to technology for multi-omics
Hyper personalized
Regulatory
disease and target
submissions Predictive asset engagement
modeling
Design & synthesis of maintenance Strategic location
clinic-ready molecules Integrating internal & planning
Strategic external clinical data Real-time data analysis Democratized insights
bets into early discovery & and safety monitoring Optimized quality & & recommendations
translational science real-time batch release Strategic E2E employee
Optimize developability
value planning
& manufacturability
CMC regulatory filing
Improved customer
Optimize discovery and patient experience Optimized knowledge
portfolio for PTRS management & learning
Recipe development,
scaling & optimization
Resilient, sustainable,
agile supply networks
for all modalities
Accelerated post-
approval process &
product optimization
Source: Accenture 2024.
IA
evitareneg
fo
ega
eht
ni
secneics
efil
gnitnevnieR
Contents Executive summary Supercharging science Revolutionizing the value chain Five C-Suite imperatives Start your reinvention journey 12
Research and Development However, the industry struggles with fragmentation
(Pre-discovery, discovery research, across different functional areas of the value chain,
pre-clinical & translational, clinical, resulting in numerous handoffs, non-standard
regulatory, safety) application of technology solutions and persistent
data silos. To overcome these inefficiencies, there is
The future of R&D hinges on using intelligent
a critical need to establish a common language for
technologies to dramatically improve cycle times,
effective collaboration among scientists, engineers and
success rates and efficiency. By accelerating discovery
marketers.
through in silico methods and using AI for tasks
By optimizing processes and responding more adeptly
ranging from molecule generation, optimization of
to market demands, companies can make full use
lead compounds, biomarker discovery and patient
of their organizational capabilities. Take the recent
stratification, companies can enhance their clinical
repurposing of diabetes drugs for obesity treatment.
success rates and expedite the entire R&D process.
Companies used extensive safety data collected over
AI’s ability to predict off-target effects, optimize drug
a decade for one disease and efficiently used market
safety profiles and incorporate digital tools for remote
signals to enhance drug development for a different
monitoring and patient retention presents a significant
disease. Such strategic shifts promise to mitigate high
leap towards increasing investigational new drug
attrition rates and reduce lengthy, costly processes
approval rates and reducing trial durations.
currently burdening the industry.
IA
evitareneg
fo
ega
eht
ni
secneics
efil
gnitnevnieR
Contents Executive summary Supercharging science Revolutionizing the value chain Five C-Suite imperatives Start your reinvention journey 13
Process Development, Manufacturing, Quality
• Companies can demonstrate a deep understanding
After a drug has been discovered and moved into clinical
and control over complex drug production through
trials, companies should develop a consistent and
a combination of scientific data and AI, expediting
scalable recipe to support supply as quickly as possible.
regulatory filing and approval.
However, R&D efforts are surfacing increasingly specific
and complex drugs that require increasingly complicated,
• Recipe tech transfer and knowledge management
biology-based recipes. Faster clinical trials mean less
between sites in the supply chain (internal sites or
time for recipe development teams to optimize these
external contract manufacturers) will be accelerated
complex recipes and scale them to the appropriate level.
and less subject to the risk of unexpected quality issues,
Intelligent technologies will help move some of this
ensuring supply chain “resilience by design.”
recipe development from slow and repetitive wet lab
experimentation to the in silico space, ensuring that:
• All markets rapidly adopt significant post-approval
recipe improvements, like new automation or process
• Drugs progress through clinical trials to commercial
sensing tech.
production without being hindered by recipe
development and scaling issues.
IA
evitareneg
fo
ega
eht
ni
secneics
efil
gnitnevnieR
Contents Executive summary Supercharging science Revolutionizing the value chain Five C-Suite imperatives Start your reinvention journey 14
eht
ni
secneics
efil
gnitnevnieR
Supply Chain
of supply nodes — can have extremely long-term This in turn can lead to complex post-approval change
Classical and generative AI give companies the
impacts on commercial supply chain agility, controls that must be managed by regulatory affairs for
opportunity to redesign their supply chain and operations
sustainability and resilience. all markets as well as a complex proliferation of product
end-to-end, enhancing resilience, agility and sustainability.
variants that must be managed by supply chain planners.
When combined with both internal and third-party data,
Even in the actual manufacturing of the product,
With a coordinated, connected data fabric and improved
AI can give companies a unified view of demand. This
complexity increases depending on how many sites are
standardization, all parties — from commercial supply
allows them to not only understand but control the supply-
involved in the process and how well they stay harmonized
chain to recipe-development teams — can work together
side complexity of novel treatments. AI can also generate
with each other and their colleagues in R&D who are
more effectively. Such integration also better enables AI
scenarios and automate responses to many potential
developing new drugs that will move to those nodes over
tools to support collaboration, oversight and enterprise-
disruptions. This approach helps improve production
time. Each site manages its own production process,
wide continuous improvement. Without such coordination,
processes for maximum yield and highest quality.
equipment, asset management, quality control, operations
fragmentation occurs and opportunities are missed.
technology (and sometimes IT) system landscape and
Traditionally, companies complete the design of their
continuous improvement programs at individual nodes.
supply chain and operations in silos long before the
Local site level continuous improvements or corrective and
product enters the commercial supply chain. Decisions
preventative actions can lead to divergent evolutions of
during development — regarding manufacturing recipes
the recipes for a product.
(bill of materials, equipment), formulation, packaging,
release, shipping methods, CMC filing strategy and choice
IA
evitareneg
fo
ega
eht
ni
secneics
efil
gnitnevnieR
Contents Executive summary Supercharging science Revolutionizing the value chain Five C-Suite imperatives Start your reinvention journey 15
Commercialization
AI is transforming commercialization, offering significant
In marketing, ongoing advancements powered by AI
improvements across access strategies, marketing and
will accelerate original and derivative content creation
customer engagement.
including images, copy and animation. These technologies
can facilitate dynamic marketing material assembly within
For example, AI significantly bolsters access strategies by
regulatory constraints. Early applications of generative AI to
using advanced data for deal modeling and enhanced payer
medical loss ratio processes are already bolstering human
contract negotiations. Generative AI can help simulate
reviewers’ work without taking them out of the loop.
complex payer negotiations and streamline decision-making
processes. Integrating disparate data sources facilitates
Generative AI is also transforming customer engagement.
contract performance monitoring. Using AI can enhance
Using personal large language models like assistants
oversight and minimize rebate leakage, improving revenue.
better prepares field teams for customer interactions.
These “assistants” access data, provide insights, simulate
conversations and analyze customer contexts for more
productive engagement. The era of the “bionic rep” is here.
IA
evitareneg
fo
ega
eht
ni
secneics
efil
gnitnevnieR
Contents Executive summary Supercharging science Revolutionizing the value chain Five C-Suite imperatives Start your reinvention journey 16
Figure 2
Opportunities are evident in each area of the value chain. Based on our research and client experience, these strategic bets represent considerable potential value if
the workflows and processes are reinvented end-to-end.
The size of the opportunity if the workflows and processes are reinvented end-to-end.
Pre-discovery & Preclinical, Product development, Commercialization Enterprise
discovery research translational, clinical, manufacturing, quality functions
regulatory & safety & supply chain
Accelerate timelines by Accelerate timelines Lower supply chain risk Optimize patient and Lower costs and
almost 3 years per by 1.5 years per and get critical customer engagement to increase efficiency
successful drug successful drug medicines into the accelerate time to peak
hands of patients faster sales while effectively
managing costs
Discover better drug
candidates (e.g., for
undruggable targets)
1-3%
10 - 30%
Revenue uplift
(product availability) Acceleration in time
to peak sales
3 to 5%
$0.3 - 1.5B $0.2 - 0.8B
10 to 15%
Production &
Revenue upside per Revenue upside per Fulfillment costs Commercial costs
successful drug successful drug
10 to 15% 20 to 25%
$600 - 800M $300 - 400M
30% +
Working capital Script conversion
Costs per successful drug Costs per successful drug reduction (inventory) and adherence Corporate function cost
Source: Accenture Research, 2024. See methodology section.
IA
evitareneg
fo
ega
eht
ni
secneics
efil
gnitnevnieR
Contents Executive summary Supercharging science Revolutionizing the value chain Five C-Suite imperatives Start your reinvention journey 17
The full potential lies in connection.
Leaders need to think in terms of value streams.
The transformative power of intelligent technologies Understanding that opportunity is critical. Yet many life It’s time to think in terms of value streams — such as
such as generative AI on individual parts of the sciences C-suite executives in our recent pulse survey7 accelerating time to clinic or accelerating time to market
biopharma value chain are undeniable. However, remain focused on individual use cases rather than — rather than small pilot projects. (We will cover this idea
while those effects are exciting in and of themselves, end-to-end processes and capabilities. Consider how: in more detail in the section titled “lead with value”).
leaders will need to bridge functional silos to reap the
It’s time to capture the benefits of connecting deep
full benefits of these technologies.
2/3 2/3
functional areas of expertise. All functions should align
Fundamentally, generative AI empowers by know which have outlined potential their reinvention efforts to these cross-functional value
democratizing access to information, accelerating areas they want impacts of generative streams to ensure that their reinvention is comprehensive
its flow throughout the organization. Adopting to prioritize but AI but say that further and delivers value to patients, the entire enterprise and the
generative AI thus presents an opportunity to foster do not have an analysis is required healthcare system.
better collaboration and ultimately deliver value implementation plan to fully articulate
It’s time to develop end-to-end capabilities. This means
across the entire value chain — where the impact on business value
rethinking many processes and integrating intelligent
the whole is greater than the sum of its parts.
technologies into all aspects of the workflows of that
capability. It also means developing the skills needed to
use AI effectively.
Five C-suite
imperatives to help
reinvent your business
IA
evitareneg
fo
ega
eht
ni
secneics
efil
gnitnevnieR
19
IA
evitareneg
fo
ega
eht
ni
secneics
efil
gnitnevnieR
01/
Five C-suite
Lead with value
imperatives
02/
Over the past several months, Accenture has
engaged in numerous discussions with clients
Reinvent talent and ways of working
regarding the impact of generative AI. We have
also undertaken more than 1,000 generative
AI-focused projects, many in collaboration with
03/
leading biopharmaceutical companies. Drawing
from these experiences and our analysis of
Understand and develop an AI-enabled secure digital core
industry leaders, we have identified five key
imperatives for CEOs who are committed to
capitalizing on the opportunities presented by
04/
intelligent technologies.
Close the gap on responsible AI
05/
Drive and support continuous reinvention
IA
evitareneg
fo
ega
eht
ni
secneics
efil
gnitnevnieR
Contents Executive summary Supercharging science Revolutionizing the value chain Five C-Suite imperatives Start your reinvention journey 20
01/
Lead with value
Rather than focusing solely on technology, companies should
prioritize efforts to understand how intelligent technologies
can fundamentally redefine processes and capabilities. Leading
with value means not only seeking cost saving opportunities
but also driving systematic acceleration, innovation and
growth. By taking a strategic view, companies can move
away from low-value proofs-of-concept and embrace the full
potential of intelligent technologies.
To be able to do that, biopharma companies must focus on five
large investment areas to create value at scale. We call these
investment areas “value streams.”
IA
evitareneg
fo
ega
eht
ni
secneics
efil
gnitnevnieR
Contents Executive summary Supercharging science Revolutionizing the value chain Five C-Suite imperatives Start your reinvention journey 21
Accelerating time to clinic Making medicines more accessible
Fundamentally, these value streams represent
the company’s objectives and are inherently
Select and validate novel targets and deliver human- Improve physical access to and affordability of
cross-functional. For instance, making medicines
ready molecules to the clinic at twice the speed and novel medicines to address unmet medical needs
more accessible requires early discussions during
half the cost by using in silico methods, reducing globally. One aspect of this includes supporting
clinical trial design, among clinical, market access
wet-lab experiments and using AI for predicting off novel complex modalities (for example, cell and
and manufacturing teams. These discussions
target effects, thereby optimizing differentiated gene therapies, antibody drug conjugates and
should cover cost of goods sold implications,
efficacy and reducing safety liabilities. messenger RNA) by lowering cost of goods sold
potential manufacturing challenges and impacts
and capital investment in the supply chain while
on reimbursement. Finally, enhancing patient
maximizing agility and quality. Another aspect
access to therapy necessitates collaborative
Accelerating time to market includes an improved financial coverage and more
efforts from manufacturing, supply chain and
affordable pricing to ensure more populations can
commercial teams. Design and execute patient-centric trials and efficient,
afford therapy.
well-controlled and well-understood manufacturing
The integration of value streams across
processes that maximize efficacy, safety and
different biopharmaceutical functions is
consistent sustainable quality to drive differentiated
Establishing end-to-end
illustrated in Figure 3.
regulatory approvals in a third less time.
insights and feedback loops
Share insights across the organization to enable
Maximizing the value faster information flow, better planning and reporting
proposition of medicines — and keep all stakeholders in the loop. Creating and
sharing actionable insights across the value chain for
Prove the health and economic outcomes to
the life of a molecule will allow teams to anticipate
maximize patient benefit and make the case for
and solve for bottlenecks at every step through
physicians to prescribe and for payers to cover
dynamic planning, portfolio prioritization, patient
the medicine globally.
impact, capital allocation and reporting to investors.
IA
evitareneg
fo
ega
eht
ni
secneics
efil
gnitnevnieR
Contents Executive summary Supercharging science Revolutionizing the value chain Five C-Suite imperatives Start your reinvention journey 22
Figure 3
Illustrative value streams in biopharmaceutival value chain:
* Value streams. Source: Accenture 2024.
IA
evitareneg
fo
ega
eht
ni
secneics
efil
gnitnevnieR
Contents Executive summary Supercharging science Revolutionizing the value chain Five C-Suite imperatives Start your reinvention journey 23
Case study
Lead with value: A modern approach to commercialization
As companies go through their reinvention journey, it is operational appro |
273 | accenture | Accenture-UNGC-GenAI-Global-Goals-Report.pdf | GEN AI FOR THE
GLOBAL GOALS
The Private Sector’s Guide to Accelerating Sustainable
Development with Technology
The 2030 Agenda — our global blueprint
for peace and prosperity on a healthy
planet — is in deep trouble. AI could help
to turn that around. It could supercharge
climate action and efforts to achieve
the 17 Sustainable Development Goals
by 2030. But all this depends on AI
technologies being harnessed responsibly
and made accessible to all."
António Guterres
United Nations
Secretary-General
2 United Nations Global Compact | Accenture Gen AI for the Global Goals 3
CONTENTS
Welcome 6
Introduction 11
What is Gen AI? 15
Why is Gen AI exciting for business? 18
Looking Forward 47
What’s the catch? 21
Resource 1: Playbook for implementing
The private sector’s leading role in Gen AI responsibly 54
sustainable development 22
Resource 2: Using Gen AI to support
Using Gen AI to advance the your sustainable development ambition 56
Sustainable Development Goals 25
Resource 3: The role of each
Operational Efficiency 28 business function 58
Sustainable Value Chain 30 Resource 4: Broader ecosystem
advocacy recommendations 60
Innovation 32
Appendix 65
Communication and Reporting 34
References 69
Mitigating the sustainable
development risks of Gen AI 37 Acknowledgments 72
4 United Nations Global Compact | Accenture
FOREWORD: UNITED NATIONS
THE TEN PRINCIPLES OF THE
GLOBAL COMPACT
UNITED NATIONS GLOBAL COMPACT
HUMAN RIGHTS
Sanda Ojiambo
1. Businesses should support and respect the protection of
Assistant Secretary General and CEO,
internationally proclaimed human rights; and
United Nations Global Compact
2. make sure that they are not complicit in human rights abuses.
The world is wavering on the 2030 Agenda for Anchored in international standards, the Ten
Sustainable Development and achieving the Principles of the UN Global Compact are the guiding
LABOUR
Sustainable Development Goals (SDGs or Global framework to ensure core business models are
Goals). Increased geopolitical tensions, inequalities, principles-based, and they can also be applied to
and climate change impacts have hindered progress guide companies towards responsible AI models as 3. Businesses should uphold the freedom of association and the
and added to the complexity of the sustainability the private sector makes this technological leap. effective recognition of the right to collective bargaining;
landscape.
This report acknowledges and complements 4. the elimination of all forms of forced and compulsory labour;
Gen AI for the Global Goals outlines the private existing UN-level efforts towards in-depth analysis
5. the effective abolition of child labour; and
sector’s opportunity to use Generative AI (Gen and recommendations on the governance of AI for
AI) as an accelerator for SDG action. The private humanity and the Global Digital Compact. Drawn
6. the elimination of discrimination in respect of employment
sector’s access to capital, wealth of data, and ability from a multi-stakeholder consultative process, the
and occupation.
to act quickly across geographies creates a unique report highlights examples of tangible actions being
opportunity for impact. Yet, the private sector taken today to help private sector leaders consider
must pay special attention to the unique risks how they can support the advancement of the SDGs
ENVIRONMENT
of an explosion in Gen AI usage. with Gen AI in their strategies and operations.
Indeed, the UN Secretary General’s high-level In compiling this report, we are grateful to many 7. Businesses should support a precautionary approach to
multistakeholder advisory body on AI interim report colleagues at the UN Global Compact and our environmental challenges;
notes that AI applications could potentially be a collaborators at Accenture for their insights and
game changer in helping to meet the SDGs, but also contributions. We would also like to express our 8. undertake initiatives to promote greater environmental
that AI poses risks to cyber security, privacy, and appreciation to the business leaders and other responsibility; and
cultural diversity. stakeholders who were critical to the development
of this report. 9. encourage the development and diffusion of environmentally
To this end, several global initiatives are underway friendly technologies.
that aim to provide the necessary frameworks As we approach 2030, the stakes are high if we
for responsible investment, development and want to secure a prosperous future for people and
deployment of AI models, including Gen AI. the planet outlined in the SDGs. It is time for the ANTI-CORRUPTION
Collectively, these initiatives call on stakeholders private sector to take bold, ambitious action to
across the AI value chain to adhere to long-standing, move us forward faster. 10. Businesses should work against corruption in all its forms,
internationally agreed principles and standards for including extortion and bribery.
responsible, rights-based conduct.
For nearly 25 years, the UN Global Compact has
been the call to companies to align their operations
and strategies with its Ten Principles covering human
The Ten Principles of the United Nations Global Compact are derived from: the Universal Declaration of Human Rights,
rights, labour, environment, and anti-corruption. the International Labour Organization’s Declaration on Fundamental Principles and Rights at Work, the Rio Declaration
on Environment and Development, and the United Nations Convention Against Corruption
6 United Nations Global Compact | Accenture Gen AI for the Global Goals 7
FOREWORD: ACCENTURE
The promise of technology to unlock change continues
to inspire the private sector towards groundbreaking
innovation, and the monumental advancements brought
to us through the revolutionary technology of Gen AI
are no different. Gen AI is rapidly transforming daily
operations and productivity across the private sector.
Stephanie Jamison Accenture research shows that 97% of executives
Global Resources Industry believe Gen AI will transform their industry over the
Practice Chair and Global
next three to five years. Despite exciting growth, we
Sustainability Services
Lead, Accenture are still at the early stages of this technology; we must
continue to learn and evolve our approach to mitigate
risk, starting with intentional design when identifying
and refining use cases.
Gen AI isn’t just about increasing productivity. It has the
potential to revolutionize how we approach sustainable
development and offers new opportunities to drive
Arnab Chakraborty progress forward. At this early stage, business leaders
Chief Responsible AI have a unique opportunity to chart the course for Gen
Officer, Accenture
AI’s impact on people and the planet. With the SDGs
as our North Star, we can consider how the private
sector can use Gen AI to support our global push for
sustainable development.
This report shares key use cases across Gen AI for
sustainable development—empowering teams towards
operational efficiency, sustainable supply chains,
Louise James
innovation, and clear communication and reporting.
Global Co-Lead, Accenture
However, private sector leaders must balance the
Development partnerships
upside against the unique risks Gen AI introduces.
This report outlines findings and best practices from
our extensive experience developing and deploying
Gen AI both internally and with our clients. By following
this guidance, we can achieve the promise of Gen AI to
accelerate progress towards the SDGs.
We are grateful to the UN Global Compact for our
long-standing partnership and to its teams for their
insightful collaboration throughout this exciting and
critical work. We look forward to our continued work
together as we tackle the global issues behind the SDGs.
8 United Nations Global Compact | Accenture Gen AI for the Global Goals 9
INTRODUCTION
Gen AI for the Global Goals 11
INTRODUCTION
INTRODUCTION
Global challenges, including ongoing and reignited geo-political The reason for this interest? Gen AI can facilitate unprecedented
conflicts, the climate crisis, high inflation, and lingering effects of access to hyper-specific, tailored information, accelerate innovation
“We believe in the potential The value of data became
the COVID-19 pandemic, have converged to significantly hinder through cross-disciplinary thinking, and increase productivity
of this technology and think relevant even before we
progress on sustainable development. We are currently on track to help businesses overcome the converging headwinds and
if it’s implemented with started talking about Gen AI.
the appropriate guardrails to meet only 17% of the Sustainable Development Goals (SDG)1 complex problems which make sustainable development Early on, with machine
and principles, Gen AI can targets by 2030 . [1] All 17 SDGs, such as Gender Equality and progress so challenging. learning, we saw how data
directly impact sustainable Climate Action, are complex and require multiple stakeholder could improve our service
development in a range of collaboration. As global leaders juggle multiple issues concurrently, Imagine tackling multiple SDGs by applying Gen AI through and delivery times. Now,
areas, including increasing progress on sustainable development is becoming even more targeted actions across the agricultural value chain. At the start, leveraging AI with all this
access to clean water and challenging, widening the gap between action and goals. Gen AI can help farmers better forecast weather and crop yield, standardized and codified
sanitation, reducing hunger develop and optimize biological pest control methods, predict data brings significantly
and poverty, enabling At the same time, advances in technology across fields ranging soil erosion and suggest mitigation measures, and help with enhanced value. We have a
affordable clean energy, from computing to medicine and beyond are transforming our sustainable crop breeding. Next, Gen AI can help with the robust data storage network
building sustainable cities and societies and economies. The rise of Artificial Intelligence has agricultural supply chain, from optimizing supply chain logistics and history to leverage.”
communities, and addressing
had a particularly wide impact, with machine learning powering to forecasting demand to better manage food spoilage, helping Beatriz Tumoine, Global Social
overall climate action.”
analysis, decision making, and resource optimization across bridge the gap between the one billion meals of edible food Impact Director, Cemex
Greg Ulrich, Chief AI and Data sectors and company sizes. In fact, nearly 75% of large companies wasted each day and the 783 million people affected by hunger
Officer, Mastercard have already integrated AI into their business strategies .[2] each year .[6] Gen AI can also help workers along the agricultural
supply chains by identifying high risks for human rights violations,
Generative AI (Gen AI) in particular has captured the attention providing tailored educational and training programs, and acting
of the private sector due to its potential to unlock new business as a sustainability knowledge disseminator. Finally, Gen AI can
models and technologies. An overwhelming majority (97%) of help consumers better manage food waste, helping divert from
executives believe Gen AI will transform their industry and play landfills worldwide and promoting circular economy practices.
a major role in their strategies over the next three to five years .[3] Each of these applications represent an opportunity to create
Of these executives, 31% have already made significant investments business value while acting as an accelerator for impact across
in related initiatives, and 99% plan to amplify their investments .[3] the SDGs.
As a result, global investments in AI are projected to reach $200
billion by 2025 [4] , while the market for Gen AI could reach $1.3 Yet, we are still in the nascent stages of the Gen AI revolution,
trillion by 2032. [5] ironing out wrinkles in the technology and increasing our
understanding of the related environmental and social risks it
brings. Failing to manage these tradeoffs of Gen AI use could lead
to the technology causing more harm than good.
The world is at a critical juncture. Gen AI, if implemented
responsibly, has the potential to accelerate the private sector’s
progress on sustainable development and help bridge the gap
to 2030. With this report, the UN Global Compact gives private
sector leaders the tools to develop and deploy Gen AI responsibly
and to use Gen AI to advance sustainable development.
1. The SDGs are a set of 17 global objectives that aim
to end poverty, protect the planet, and ensure peace
and prosperity for all.
12 United Nations Global Compact | Accenture Gen AI for the Global Goals 13
INTRODUCTION
WHAT IS GEN AI?
Artificial Intelligence is a machine-based system that can
replicate human thinking, converting various inputs into outputs
ranging from predictions or recommendations to content.[ 7]
Gen AI is a type of artificial intelligence which can generate new
content beyond what it has already been exposed to.[ 8] It does this
by identifying and replicating patterns in existing text, images,
or other data to create realistic new data. Common consumer
Gen AI products include GPT-4/4o, Gemini, Claude, and Midjourney.
While most of the world’s attention is currently directed at Large
Language Models (LLMs), which use large text databases to
mimic all kinds of human language, models have been created
to generate anything from protein structures to memes.
General purpose “foundation models” (trained on large and
broad data sets) are the core of the Gen AI ecosystem. These
models can be tuned and supplemented with proprietary data
to create use-specific Gen AI applications. Applications and
foundation models typically rely on cloud providers for the
computational infrastructure needed for training and inference.2
In turn, these cloud providers rely on hardware providers for the
actual computers running the calculations, especially graphics
processing units (GPUs).
GEN AI APPLICATIONS
GEN
AI ATIONS
Provide applications that customize foundation
APPLIC m spo ed ce ifils
c
u bs ui sn ig
n
ea sd sd pit rio on ba lel mda sta and tuning to solve
AL
N
DATIO FOUNDATIONAL MODELS
N
OU Provide models, trained on diverse sets of data
F DELS
(often the open web), that can be leveraged to
O
M
develop custom Gen AI applications
UCTURE
INFRASTRUCTURE
ASTR
Provide infrastructure to host, compute, and store
INFR
Gen AI workloads using purpose-built hardware
(e.g., GPUs) through cloud providers or onsite
2. Training is the set-up of a model while
Figure 1: Gen AI Tech Stack inference is the use of a finished model.
14 United Nations Global Compact | Accenture Gen AI for the Global Goals 15
INTRODUCTION
It’s hard to manage or improve what you The biggest benefit Gen AI can deliver
can’t measure. When you layer Gen AI is contextual, localized strategy. This
on top of existing data, you can unlock can help deliver contextual and specific
insights and uncover unbelievably actions and recommendations, helping
powerful opportunities.” unlock unprecedented SDG action.”
Emilio Tenuta, Senior Vice President and Gagandeep K. Bhullar, Founder and CEO,
Chief Sustainability Officer, Ecolab SuperHumanRace
For someone who has been working Gen AI models are becoming more
on income inequality for my entire life, powerful and knowledgeable, with the
seeing an opportunity to train people ability to solve tasks we previously
quickly to help them create wealth is couldn’t imagine. The speed at which
just incredibly exciting.” this technology is developing is
astonishing and incredibly exciting.”
Shamina Singh, Founder and President of Mastercard’s
Center for Inclusive Growth and EVP, Sustainability,
Hilda Kosorus, Director of Data and AI Center
Mastercard
of Excellence for EMEA, Crayon
The greatest potential of Gen AI is having
a collective intelligence just a prompt
away and embedding that in business
processes to allow companies to make
better decisions.”
Vikram Nagendra, Director of Corporate Sustainability, SAP
16 United Nations Global Compact | Accenture Gen AI for the Global Goals 17
INTRODUCTION
WHY IS GEN AI EXCITING
FOR BUSINESS?
Gen AI’s potential to create business value comes from its three
foundational capabilities: acting as a Data Miner, an Insight
“In the past 30 years, there is
Navigator, or a Knowledge Amplifier. When combined with other
no single technology except
business capabilities, Gen AI can help companies lower costs
for AI that I have been able
through increased operational efficiency, streamline management to stand up in front of CEOs
of complex value chains, increase revenue through innovative and credibly and authentically
new offerings, and simplify reporting and compliance. When say that it will have a material
companies use Gen AI responsibly to achieve these ends they can positive impact on every part
unlock business value while advancing sustainable development. of their enterprise.”
Julie Sweet, CEO and Chair,
Imagine if businesses used Gen AI to tackle the logistical and Accenture
analytical barriers to developing a truly circular economy. FOUNDATIONAL CAPABILITIES
R&D teams could use Gen AI to accelerate development of
replacements for resource intensive and environmentally
OF GEN AI
degrading materials. Design teams could use a Gen AI assistant
to help embed circular principles across product and service
systems, starting with sustainable material selection and
advancing through designing for extended product use and new
business models. Gen AI can also help logistics teams optimize
DATA MINER Gen AI surpasses traditional analytics tools by extracting valuable insights
transportation and inventories across forward distribution
from unlabeled and unstructured data such as text, images, video, or audio,
channels and manage the increased operational complexity of
with the potential to link unstructured qualitative data with structured
reverse logistics networks. Once products reach customers,
quantitative data. As an example, Gen AI could provide deeper insights into
Gen AI can improve services that facilitate asset sharing or
market sentiment and investment trends by analyzing unstructured data
help guide customers and technicians through repairs to extend
like filings, reports, news articles, or internal communications.[9]
product life. When life-extension is no longer an option, Gen AI
can help recovery and recycling vendors to more effectively
separate valuable materials from waste streams for recovery.
Gen AI can also help businesses learn from best practices,
INSIGHT Interpreting data to drive decision-making is not always intuitive,
improving communication with value chain partners, regulators,
requiring specially trained employees and a deep familiarity with the
NAVIGATOR
and consumers to drive ecosystem-wide change. By tackling
process or context of the decision in question. Gen AI can help employees
these challenges, businesses can take the next steps towards
apply technical knowledge to analyze complex data and provide
decoupling growth from resource use, creating value while
recommendations, predictions, or explanations for businesses to act
tackling SDGs like climate action, responsible consumption
upon. For example, Gen AI can support technicians during infrastructure
and production, and affordable and clean energy.
maintenance by providing interactive guidance generated from preventative
maintenance systems and the technician’s live observations.[10]
KNOWLEDGE Gen AI tools can empower the workforce by functioning as capable
and customizable search engines, communication coaches, or virtual
AMPLIFIER
assistants. For instance, Gen AI can be used to help draft memos
and presentations or generate training plans to upskill employees
for incoming regulations.
18 United Nations Global Compact | Accenture Gen AI for the Global Goals 19
INTRODUCTION
WHAT’S THE CATCH?
Gen AI is an exciting advancement, but poses a number of
user and external risks that require careful consideration and
It is important for companies
management. User risks may include biased outputs and factual
to consider how they are
errors, opaque processes, and the opportunity for misuse.
getting the best information
External risks include increased resource use across energy, out of Gen AI. What is your
water, and infrastructure and the potential to transform society by governance system to ensure
shifting the job market and spreading misinformation. The adoption you have checks and balances
of broader AI technologies has been uneven, with businesses in around unintentional outputs?
advanced economies accounting for the majority of capability Do you have transparency and
development.[11] Not all regions and countries have equal access an understanding of the data
to the infrastructure, training, and data required to take advantage being fed into the system?”
of Gen AI’s benefits, which could widen the existing digital divide.
Brigid Evans, Director of Global
The new and rapidly changing Gen AI landscape only adds Policy, Pearson
uncertainty to these risks.
The UN Global Compact has and continues to advocate for a
principles-based approach to responsible business, considering Each time we evaluate a
human rights, the environment, labour, and anti-corruption.3 use case, we consider if it’s
Given the scale of global investment in Gen AI, it is imperative necessary to use Gen AI or if
that we monitor its development and implementation to maximize a traditional digital application
benefits while avoiding further negative effects on the SDGs. or AI could suffice.”
The UN Global Compact hopes this report can serve as a guide to Giulia Brandetti, Head of Data
the private sector in how to responsibly apply Gen AI, as well as Governance and Resource
how to leverage it as a tool to accelerate sustainable development. Allocation, Enel Group
3. See The Ten Principles of the UN Global Compact
for more detail.
20 United Nations Global Compact | Accenture Gen AI for the Global Goals 21
INTRODUCTION
THE PRIVATE SECTOR’S
LEADING ROLE IN SUSTAINABLE
DEVELOPMENT
When thinking about the SDGs, we need to think
about where we can accelerate action and create
a flywheel effect, and how Gen AI can support that.”
The private sector, responsible for more than 60% While companies face pressure to move quickly with
Shamina Singh, Founder and President of Mastercard’s Center
of global GDP4, is the largest player in production of Gen AI, they also have a responsibility to start small
for Inclusive Growth and EVP, Sustainability, Mastercard
goods and services worldwide .[12] As a driving force and move safely. Gen AI should always be developed
behind innovation and the explosion of Gen AI, the with humans in the loop, meaning that people are in
private sector has a unique opportunity to lead the charge of (and accountable for) reviews to ensure the
way in harnessing this technology for sustainable safe and responsible use of this technology. Due to its
development. By prioritizing the SDGs throughout central role in sustainable development, the private
the use of Gen AI (as described in Resources 1-4), sector should go beyond responsible implementation
the private sector can drive positive impact and and leverage technologies like Gen AI to quickly close
advance the SDGs globally. the gap between intent and action on SDGs. Gen AI’s ability to scale information and analytics
can help us get farther faster on global issues.”
The UN Global Compact challenges companies Recognizing these responsibilities and the challenge
developing, deploying, and using Gen AI to work of navigating emerging technologies, this report lays Márcia Balisciano, Chief Sustainability Officer, RELX Group
towards two key objectives when it comes to its out how to achieve these two objectives through
use, shown in figure 2. actionable insights and recommendations.
SUSTAINABLE
By increasing productivity, the private sector has
GEN AI
unlocked tremendous economic growth, but this has
come at a cost. This is where AI can really step in and
MEANS ENDS play a positive role — at the intersection of maintaining
economic growth and sustainable development."
DEVELOP AND DEPLOY USE GEN AI TO ADVANCE
Vikram Nagendra, Director of Corporate Sustainability, SAP
GEN AI RESPONSIBLY SUSTAINABLE DEVELOPMENT
Companies must ensure that the means Companies must also consider the ends
used to develop and deploy Gen AI are for which Gen AI is deployed. The private
ethical and transparent. The private sector can accelerate and amplify
sector can do this through adopting both SDG action through applying Gen AI in
responsible processes and governance. sustainable development action areas.
Providing cited sources throughout a Gen AI
response helps to increase traceability and trust.”
Emma Grande, Director of ESG Strategy and Engagement, Salesforce
Figure 2: Key Objectives for the Private Sector
4. Additionally, more than 80% of production in low and
on Gen AI and Sustainable Development middle income countries is private sector driven.[12]
22 United Nations Global Compact | Accenture Gen AI for the Global Goals 23
USING GEN AI
TO ADVANCE
THE SUSTAINABLE
DEVELOPMENT
GOALS
Gen AI for the Global Goals 25
USING GEN AI TO ADVANCE THE SUSTAINABLE DEVELOPMENT GOALS
USING GEN AI TO ADVANCE THE
SUSTAINABLE DEVELOPMENT
GOALS
• Resource Optimization • Lifecycle Assessment
Three key elements underpin the successful and responsible use
• Efficient Code • Responsible Sourcing
of Gen AI. First, companies must ensure they clearly understand
With Gen AI, the goal is not
the problem they are solving and agree that Gen AI is an appropriate • Worker Effectiveness • Supplier Engagement
to replace human work but
solution relative to the tradeoffs. Second, they must ready their to supercharge it.”
people to use Gen AI responsibly by supporting them with the
Emma Grande, Director of
appropriate digital, data, and AI literacy training. Finally, companies
ESG Strategy and Engagement,
need to set up the right governance structures to maintain safety Salesforce
and accountability. After building the foundation, Gen AI’s ability
to act as a Data Miner, Insight Navigator, and Knowledge Amplifier
OPERATIONAL SUSTAINABLE
can be unleashed to help support sustainable development action
EFFICIENCY VALUE CHAIN
and accelerate progress towards the SDGs.
These foundational capabilities can be applied to existing
FOUNDATIONAL
technologies and business operations to accelerate sustainable CAPABILITIES
In the last year, we have
development across four use case categories, shown below.
addressed a number of DATA MINER
The following examples illustrate how businesses can — and
low-hanging fruits with Gen
already are — using Gen AI to advance their sustainability AI. Moving forward, from a INSIGHT NAVIGATOR
journeys. As the technology is so new to business, existing maturity cycle perspective,
KNOWLEDGE
case studies largely represent the low-hanging fruit, offloading we will see more high-value-
administrative work and democratizing access to information. added cases.” AMPLIFIER
Yet even these initial examples can have significant positive
Vikram Nagendra, Director of
effects on the private sector’s ability to make progress on Corporate Sustainability, SAP
sustainable development. As Gen AI improves, we can expect COMMUNICATION
INNOVATION
radical changes in the pace of innovation and level of impact of
AND REPORTING
this transformative technology, potentially impacting sustainable
development in ways that are yet to be imagined.5
The use cases outlined here are just the beginning of Gen AI’s
ability to reshape the way businesses operate globally. Gen AI
is positioned to play a pivotal role in advancing sustainable
development towards the SDGs. By responsibly integrating Gen AI • Sustainability Reporting • Green Finance
into daily operations, companies can drive positive change and
• Marketing Sustainability • Sustainable Product and Service Design
progress towards their SDGs while achieving their business goals.
• Boosting Collaboration • Cutting Edge Research
5. At the time of writing, business use of Gen AI is Figure 3: Use Cases of Gen AI for Sustainable Development
so early stage that most companies are working to
validate the exact operational impacts before public
disclosure. Several business leaders we interviewed
indicated promising initial results. Also note that the
rapid and concentrated development of Gen AI in a few
countries means that these case studies skew towards
large companies in the Global North.
26 United Nations Global Compact | Accenture Gen AI for the Global Goals 27
USING GEN AI TO ADVANCE THE SUSTAINABLE DEVELOPMENT GOALS
OPERATIONAL EFFICIENCY
Companies need to manage a finite number of resources efficiently to operate within financial and
planetary boundaries to drive consistent and sustainable returns. Opportunities for Gen AI to increase
efficiencies exist across a variety of operational capabilities, such as resource optimization, worker
CASE STUDIES
effectiveness, and code efficiency. Of course, businesses must consider the resource costs of Gen AI
adoption and usage.
SUPERHUMANRACE SIEMENS
Resource Optimization: Minimizing the resource requirements
to achieve business outcomes represents a dual opportunity SuperHumanRace set out to improve maternal Siemens deployed the Siemens Industrial Copilot,
Our goal is to increase our
for companies to lower both costs and environmental impact. health in India, prioritizing the states with the a Gen AI solution developed in partnership with
use of technologies that
The private sector can layer Gen AI’s foundational capabilities augment the capabilities of poorest outcomes. The company developed an Microsoft, on a Schaeffler manufacturing line,
on top of existing analytics and AI technologies to help employees our colleagues, enhancing app designed to provide doctors with personalized showcasing the power of Gen AI to increase
optimize the use of resources from computing power to shipping our efficiency and productivity recommendations for maternal health. Utilizing industrial efficiency and operations solutions.
networks. For example, companies can use Gen AI to upgrade while ensuring a human is Gen AI alongside machine modeling, the
a machine-learning-powered predictive analytics system into always the final decision-maker.” app leverages a large data set on maternal The Siemens Industrial Copilot has been
a prescriptive maintenance system that generates instructions health trends, interventions, and permutations instrumental in assisting Schaeffler’s automation
Michela Buzzichelli, Head of Data
and recommendations for workers .[10] Science and AI at Enel Global ICT, of high-risk pregnancies to deliver tailored engineers in generating code for programmable
Enel Group recommendations to each patient. logic controllers (PLCs). PLCs are the brains
Worker Effectiveness: Effective training and tools are critical which control factory machines; one in three
to supporting employees across their roles. Traditional training The app generates questionnaires for doctors runs on a Siemens device. By using natural
methods often fall short in providing effective, individually tailored that are tailored to the pregnancy stage, medical language inputs to develop code, the time, effort,
learning environments, while inflexible tools lack the adaptability condition, and risk factors of each patient. By and probability of errors in the coding process
to support decision-making across an employee’s responsibilities. integrating existing machine learning models with have been significantly reduced. This has not
With Gen AI, we could
Workers can use Gen AI as a powerful professional educational Gen AI, the app identifies patterns in patient data only decreased human effort on repetitive tasks
maximize the productivity
tool, personalizing learning on sustainable development topics time we’re getting back from to provide contextual descriptions of risk factors but also allowed engineering resources to focus
to each employee’s role, native language, and region-specific our workers and partners to tailored to patients. SuperHumanRace offers on higher-value work. In addition, it has the
regulations or policies. Furthermore, Gen AI can support create more opportunities in AI-enabled suggestions that link information potential to empower less-experienced shop-floor
identifying and designing specific sustainable development service of people on the planet.” with specific actions, such as recommended employees to transition into engineering roles,
training or courses relevant to a co |
274 | Autres | Artificial intelligence index report.pdf | Artificial
Intelligence
Index Report
2023
Artificial Intelligence
Index Report 2023
Introduction to the
AI Index Report 2023
Welcome to the sixth edition of the AI Index Report! This year, the report introduces more original data than any
previous edition, including a new chapter on AI public opinion, a more thorough technical performance chapter,
original analysis about large language and multimodal models, detailed trends in global AI legislation records,
a study of the environmental impact of AI systems, and more.
The AI Index Report tracks, collates, distills, and visualizes data related to artificial intelligence. Our mission is
to provide unbiased, rigorously vetted, broadly sourced data in order for policymakers, researchers, executives,
journalists, and the general public to develop a more thorough and nuanced understanding of the complex field of
AI. The report aims to be the world’s most credible and authoritative source for data and insights about AI.
From the Co-Directors
AI has moved into its era of deployment; throughout 2022 and the beginning of 2023, new large-scale AI models
have been released every month. These models, such as ChatGPT, Stable Diffusion, Whisper, and DALL-E 2, are
capable of an increasingly broad range of tasks, from text manipulation and analysis, to image generation, to
unprecedentedly good speech recognition. These systems demonstrate capabilities in question answering and the
generation of text, image, and code unimagined a decade ago, and they outperform the state of the art on many
benchmarks, old and new. However, they are prone to hallucination, routinely biased, and can be tricked into
serving nefarious aims, highlighting the complicated ethical challenges associated with their deployment.
Although 2022 was the first year in a decade where private AI investment decreased, AI is still a topic of great
interest to policymakers, industry leaders, researchers, and the public. Policymakers are talking about AI more
than ever before. Industry leaders that have integrated AI into their businesses are seeing tangible cost and
revenue benefits. The number of AI publications and collaborations continues to increase. And the public is
forming sharper opinions about AI and which elements they like or dislike.
AI will continue to improve and, as such, become a greater part of all our lives. Given the increased presence of
this technology and its potential for massive disruption, we should all begin thinking more critically about how
exactly we want AI to be developed and deployed. We should also ask questions about who is deploying it—as
our analysis shows, AI is increasingly defined by the actions of a small set of private sector actors, rather than a
broader range of societal actors. This year’s AI Index paints a picture of where we are so far with AI, in order to
highlight what might await us in the future.
Jack Clark and Ray Perrault
Artificial Intelligence
Index Report 2023
Top Ten Takeaways
1 I ndustry races ahead of academia. 4 The world’s best new scientist … AI?
Until 2014, most significant machine learning AI models are starting to rapidly accelerate
models were released by academia. Since then, scientific progress and in 2022 were used to aid
industry has taken over. In 2022, there were 32 hydrogen fusion, improve the efficiency of matrix
significant industry-produced machine learning manipulation, and generate new antibodies.
models compared to just three produced by
5 The number of incidents concerning
academia. Building state-of-the-art AI systems
the misuse of AI is rapidly rising.
increasingly requires large amounts of data, computer
According to the AIAAIC database, which tracks
power, and money—resources that industry actors
incidents related to the ethical misuse of AI, the
inherently possess in greater amounts compared to
number of AI incidents and controversies has
nonprofits and academia.
increased 26 times since 2012. Some notable incidents
2 Performance saturation on in 2022 included a deepfake video of Ukrainian
traditional benchmarks. President Volodymyr Zelenskyy surrendering and
AI continued to post state-of-the-art results, but U.S. prisons using call-monitoring technology on their
year-over-year improvement on many benchmarks inmates. This growth is evidence of both greater use of
continues to be marginal. Moreover, the speed at AI technologies and awareness of misuse possibilities.
which benchmark saturation is being reached is
6 The demand for AI-related
increasing. However, new, more comprehensive
professional skills is increasing across
benchmarking suites such as BIG-bench and HELM
virtually every American industrial sector.
are being released.
Across every sector in the United States for which
3 AI is both helping and there is data (with the exception of agriculture,
harming the environment.
forestry, fishing, and hunting), the number of AI-
New research suggests that AI systems can have related job postings has increased on average from
serious environmental impacts. According to
1.7% in 2021 to 1.9% in 2022. Employers in the United
Luccioni et al., 2022, BLOOM’s training run
States are increasingly looking for workers with AI-
emitted 25 times more carbon than a single air
related skills.
traveler on a one-way trip from New York to
San Francisco. Still, new reinforcement learning
models like BCOOLER show that AI systems
can be used to optimize energy usage.
Artificial Intelligence
Index Report 2023
Top Ten Takeaways (cont’d)
7 For the first time in the last decade, 10 Chinese citizens are among those
year-over-year private investment who feel the most positively about
in AI decreased. AI products and services. Americans …
Global AI private investment was $91.9 billion in not so much.
2022, which represented a 26.7% decrease since In a 2022 IPSOS survey, 78% of Chinese respondents
2021. The total number of AI-related funding events (the highest proportion of surveyed countries) agreed
as well as the number of newly funded AI companies with the statement that products and services using
likewise decreased. Still, during the last decade as a AI have more benefits than drawbacks. After Chinese
whole, AI investment has significantly increased. In
respondents, those from Saudi Arabia (76%) and India
2022 the amount of private investment in AI was 18
(71%) felt the most positive about AI products. Only
times greater than it was in 2013.
35% of sampled Americans (among the lowest of
8 While the proportion of companies surveyed countries) agreed that products and services
adopting AI has plateaued, the using AI had more benefits than drawbacks.
companies that have adopted AI
continue to pull ahead.
The proportion of companies adopting AI in 2022
has more than doubled since 2017, though it has
plateaued in recent years between 50% and 60%,
according to the results of McKinsey’s annual
research survey. Organizations that have adopted
AI report realizing meaningful cost decreases and
revenue increases.
9 Policymaker interest in AI
is on the rise.
An AI Index analysis of the legislative records of 127
countries shows that the number of bills containing
“artificial intelligence” that were passed into law
grew from just 1 in 2016 to 37 in 2022. An analysis
of the parliamentary records on AI in 81 countries
likewise shows that mentions of AI in global
legislative proceedings have increased nearly
6.5 times since 2016.
Artificial Intelligence
Index Report 2023
Steering Committee
Co-directors
Jack Clark Raymond Perrault
Anthropic, OECD SRI International
Members
Erik Brynjolfsson Katrina Ligett Juan Carlos Niebles Yoav Shoham
Stanford University Hebrew University Stanford University, (Founding Director)
Salesforce Stanford University,
John Etchemendy Terah Lyons AI21 Labs
Stanford University Vanessa Parli
James Manyika
Stanford University Russell Wald
Google,
Stanford University
University of Oxford
Staff and Researchers
Research Manager and Editor in Chief Research Associate
Nestor Maslej Loredana Fattorini
Stanford University Stanford University
Affiliated Researchers
Elif Kiesow Cortez Helen Ngo Robi Rahman Alexandra Rome
Stanford Law School Hugging Face Data Scientist Freelance Researcher
Research Fellow
Graduate Researcher
Han Bai
Stanford University
Undergraduate Researchers
Vania Siddhartha Mena Naima Sukrut Stone Lucy Elizabeth
Chow Javvaji Hassan Patel Oak Yang Zimmerman Zhu
Stanford Stanford Stanford Stanford Stanford Stanford Stanford Stanford
University University University University University University University University
Artificial Intelligence
Index Report 2023
How to Cite This Report
Nestor Maslej, Loredana Fattorini, Erik Brynjolfsson, John Etchemendy, Katrina Ligett, Terah Lyons,
James Manyika, Helen Ngo, Juan Carlos Niebles, Vanessa Parli, Yoav Shoham, Russell Wald, Jack Clark,
and Raymond Perrault, “The AI Index 2023 Annual Report,” AI Index Steering Committee,
Institute for Human-Centered AI, Stanford University, Stanford, CA, April 2023.
The AI Index 2023 Annual Report by Stanford University is licensed under
Attribution-NoDerivatives 4.0 International.
Public Data and Tools
The AI Index 2023 Report is supplemented by raw data and an interactive tool.
We invite each reader to use the data and the tool in a way most relevant to their work and interests.
Raw data and charts: The public data and Global AI Vibrancy Tool: Compare up to
high-resolution images of all the charts 30 countries across 21 indicators. The Global AI
in the report are available on Google Drive. Vibrancy tool will be updated in the latter half of 2023.
AI Index and Stanford HAI
The AI Index is an independent initiative at the
Stanford Institute for Human-Centered Artificial Intelligence (HAI).
The AI Index was conceived within the One Hundred Year Study on AI (AI100).
We welcome feedback and new ideas for next year.
Contact us at [email protected].
Artificial Intelligence
Index Report 2023
Supporting Partners
Analytics and Research Partners
Artificial Intelligence
Index Report 2023
Contributors
We want to acknowledge the following individuals by chapter and section for their contributions of data,
analysis, advice, and expert commentary included in the AI Index 2023 Report:
Research and Development
Sara Abdulla, Catherine Aiken, Luis Aranda, Peter Cihon, Jack Clark, Loredana Fattorini, Nestor Maslej,
Besher Massri, Vanessa Parli, Naima Patel, Ray Perrault, Robi Rahman, Alexandra Rome, Kevin Xu
Technical Performance
Jack Clark, Loredana Fattorini, Siddhartha Javvaji, Katrina Ligett, Nestor Maslej, Juan Carlos Niebles,
Sukrut Oak, Vanessa Parli, Ray Perrault, Robi Rahman, Alexandra Rome, Yoav Shoham, Elizabeth Zhu
Technical AI Ethics
Jack Clark, Loredana Fattorini, Katrina Ligett, Nestor Maslej, Helen Ngo, Sukrut Oak, Vanessa Parli,
Ray Perrault, Alexandra Rome, Elizabeth Zhu, Lucy Zimmerman
Economy
Susanne Bieller, Erik Brynjolfsson, Vania Chow, Jack Clark, Natalia Dorogi, Murat Erer, Loredana Fattorini,
Akash Kaura, James Manyika, Nestor Maslej, Layla O’Kane, Vanessa Parli, Ray Perrault, Brittany Presten,
Alexandra Rome, Nicole Seredenko, Bledi Taska, Bill Valle, Casey Weston
Education
Han Bai, Betsy Bizot, Jack Clark, John Etchemendy, Loredana Fattorini, Katrina Ligett, Nestor Maslej,
Vanessa Parli, Ray Perrault, Sean Roberts, Alexandra Rome
Policy and Governance
Meghan Anand, Han Bai, Vania Chow, Jack Clark, Elif Kiesow Cortez, Rebecca DeCrescenzo, Loredana Fattorini,
Taehwa Hong, Joe Hsu, Kai Kato, Terah Lyons, Nestor Maslej, Alistair Murray, Vanessa Parli, Ray Perrault, Alexandra Rome,
Sarah Smedley, Russell Wald, Brian Williams, Catherina Xu, Stone Yang, Katie Yoon, Daniel Zhang
Diversity
Han Bai, Betsy Bizot, Jack Clark, Loredana Fattorini, Nezihe Merve Gürel, Mena Hassan, Katrina Ligett,
Nestor Maslej, Vanessa Parli, Ray Perrault, Sean Roberts, Alexandra Rome, Sarah Tan, Lucy Zimmerman
Public Opinion
Jack Clark, Loredana Fattorini, Mena Hassan, Nestor Maslej, Vanessa Parli, Ray Perrault,
Alexandra Rome, Nicole Seredenko, Bill Valle, Lucy Zimmerman
Conference Attendance
Terri Auricchio (ICML), Lee Campbell (ICLR), Cassio de Campos (UAI), Meredith Ellison (AAAI), Nicole Finn (CVPR),
Vasant Gajanan (AAAI), Katja Hofmann (ICLR), Gerhard Lakemeyer (KR), Seth Lazar (FAccT), Shugen Ma (IROS),
Becky Obbema (NeurIPS), Vesna Sabljakovic-Fritz (IJCAI), Csaba Szepesvari (ICML), Matthew Taylor (AAMAS),
Sylvie Thiebaux (ICAPS), Pradeep Varakantham (ICAPS)
Artificial Intelligence
Index Report 2023
We thank the following organizations and individuals who provided
data for inclusion in the AI Index 2023 Report:
Organizations
Code.org Lightcast
Sean Roberts Layla O’Kane, Bledi Taska
Center for Security and LinkedIn
Emerging Technology, Murat Erer, Akash Kaura,
Georgetown University Casey Weston
Sara Abdulla, Catherine Aiken
McKinsey & Company
Computing Research Natalia Dorogi, Brittany Presten
Association
Betsy Bizot NetBase Quid
Nicole Seredenko, Bill Valle
GitHub
Peter Cihon, Kevin Xu OECD.AI Policy Observatory
Luis Aranda, Besher Massri
Govini
Rebecca DeCrescenzo, Women in Machine Learning
Joe Hsu, Sarah Smedley Nezihe Merve Gürel, Sarah Tan
We also would like to thank Jeanina Casusi, Nancy King, Shana Lynch, Jonathan Mindes,
Michi Turner, and Madeleine Wright for their help in preparing this report, and Joe Hinman and
Santanu Mukherjee for their help in maintaining the AI Index website.
Artificial Intelligence
Index Report 2023
Table of Contents
Report Highlights 11
Chapter 1 Research and Development 20
Chapter 2 Technical Performance 69
Chapter 3 Technical AI Ethics 125
Chapter 4 The Economy 168
Chapter 5 Education 234
Chapter 6 Policy and Governance 263
Chapter 7 Diversity 296
Chapter 8 Public Opinion 319
Appendix 344
ACCESS THE PUBLIC DATA
Artificial Intelligence
Index Report 2023
Report Highlights
Chapter 1: Research and Development
The United States and China had the greatest number of cross-country collaborations in AI
publications from 2010 to 2021, although the pace of collaboration has slowed. The number of AI
research collaborations between the United States and China increased roughly 4 times since 2010,
and was 2.5 times greater than the collaboration totals of the next nearest country pair, the United
Kingdom and China. However the total number of U.S.-China collaborations only increased by 2.1%
from 2020 to 2021, the smallest year-over-year growth rate since 2010.
AI research is on the rise, across the board. The total number of AI publications has more than
doubled since 2010. The specific AI topics that continue dominating research include pattern
recognition, machine learning, and computer vision.
China continues to lead in total AI journal, conference, and repository publications.
The United States is still ahead in terms of AI conference and repository citations, but those
leads are slowly eroding. Still, the majority of the world’s large language and multimodal models
(54% in 2022) are produced by American institutions.
Industry races ahead of academia. Until 2014, most significant machine learning models were
released by academia. Since then, industry has taken over. In 2022, there were 32 significant
industry-produced machine learning models compared to just three produced by academia.
Building state-of-the-art AI systems increasingly requires large amounts of data, computer power,
and money—resources that industry actors inherently possess in greater amounts compared to
nonprofits and academia.
Large language models are getting bigger and more expensive. GPT-2, released in 2019,
considered by many to be the first large language model, had 1.5 billion parameters and cost an
estimated $50,000 USD to train. PaLM, one of the flagship large language models launched in 2022,
had 540 billion parameters and cost an estimated $8 million USD—PaLM was around 360 times
larger than GPT-2 and cost 160 times more. It’s not just PaLM: Across the board, large language and
multimodal models are becoming larger and pricier.
Artificial Intelligence
Index Report 2023
Chapter 2: Technical Performance
Performance saturation on traditional benchmarks. AI continued to post state-of-the-art results,
but year-over-year improvement on many benchmarks continues to be marginal. Moreover,
the speed at which benchmark saturation is being reached is increasing. However, new, more
comprehensive benchmarking suites such as BIG-bench and HELM are being released.
Generative AI breaks into the public consciousness. 2022 saw the release of text-to-image
models like DALL-E 2 and Stable Diffusion, text-to-video systems like Make-A-Video, and chatbots
like ChatGPT. Still, these systems can be prone to hallucination, confidently outputting incoherent or
untrue responses, making it hard to rely on them for critical applications.
AI systems become more flexible. Traditionally AI systems have performed well on narrow tasks
but have struggled across broader tasks. Recently released models challenge that trend; BEiT-3,
PaLI, and Gato, among others, are single AI systems increasingly capable of navigating multiple tasks
(for example, vision, language).
Capable language models still struggle with reasoning. Language models continued to improve
their generative capabilities, but new research suggests that they still struggle with complex
planning tasks.
AI is both helping and harming the environment. New research suggests that AI systems can have
serious environmental impacts. According to Luccioni et al., 2022, BLOOM’s training run emitted 25
times more carbon than a single air traveler on a one-way trip from New York to San Francisco. Still,
new reinforcement learning models like BCOOLER show that AI systems can be used to optimize
energy usage.
The world’s best new scientist … AI? AI models are starting to rapidly accelerate scientific
progress and in 2022 were used to aid hydrogen fusion, improve the efficiency of matrix
manipulation, and generate new antibodies.
AI starts to build better AI. Nvidia used an AI reinforcement learning agent to improve the design
of the chips that power AI systems. Similarly, Google recently used one of its language models,
PaLM, to suggest ways to improve the very same model. Self-improving AI learning will accelerate
AI progress.
Artificial Intelligence
Index Report 2023
Chapter 3: Technical AI Ethics
The effects of model scale on bias and toxicity are confounded by training data and mitigation
methods. In the past year, several institutions have built their own large models trained on
proprietary data—and while large models are still toxic and biased, new evidence suggests that
these issues can be somewhat mitigated after training larger models with instruction-tuning.
Generative models have arrived and so have their ethical problems. In 2022, generative models
became part of the zeitgeist. These models are capable but also come with ethical challenges. Text-
to-image generators are routinely biased along gender dimensions, and chatbots like ChatGPT can
be tricked into serving nefarious aims.
The number of incidents concerning the misuse of AI is rapidly rising. According to the AIAAIC
database, which tracks incidents related to the ethical misuse of AI, the number of AI incidents
and controversies has increased 26 times since 2012. Some notable incidents in 2022 included a
deepfake video of Ukrainian President Volodymyr Zelenskyy surrendering and U.S. prisons using
call-monitoring technology on their inmates. This growth is evidence of both greater use of AI
technologies and awareness of misuse possibilities.
Fairer models may not be less biased. Extensive analysis of language models suggests that while there
is a clear correlation between performance and fairness, fairness and bias can be at odds: Language
models which perform better on certain fairness benchmarks tend to have worse gender bias.
Interest in AI ethics continues to skyrocket. The number of accepted submissions to FAccT, a
leading AI ethics conference, has more than doubled since 2021 and increased by a factor of 10 since
2018. 2022 also saw more submissions than ever from industry actors.
Automated fact-checking with natural language processing isn’t so straightforward after all.
While several benchmarks have been developed for automated fact-checking, researchers find that
11 of 16 of such datasets rely on evidence “leaked” from fact-checking reports which did not exist at
the time of the claim surfacing.
Artificial Intelligence
Index Report 2023
Chapter 4: The Economy
The demand for AI-related professional skills is increasing across virtually every American
industrial sector. Across every sector in the United States for which there is data (with the exception
of agriculture, forestry, fishing, and hunting), the number of AI-related job postings has increased on
average from 1.7% in 2021 to 1.9% in 2022. Employers in the United States are increasingly looking for
workers with AI-related skills.
For the first time in the last decade, year-over-year private investment in AI decreased.
Global AI private investment was $91.9 billion in 2022, which represented a 26.7% decrease since 2021.
The total number of AI-related funding events as well as the number of newly funded AI companies
likewise decreased. Still, during the last decade as a whole, AI investment has significantly increased.
In 2022 the amount of private investment in AI was 18 times greater than it was in 2013.
Once again, the United States leads in investment in AI. The U.S. led the world in terms of total
amount of AI private investment. In 2022, the $47.4 billion invested in the U.S. was roughly 3.5 times
the amount invested in the next highest country, China ($13.4 billion). The U.S. also continues to lead in
terms of total number of newly funded AI companies, seeing 1.9 times more than the European Union
and the United Kingdom combined, and 3.4 times more than China.
In 2022, the AI focus area with the most investment was medical and healthcare ($6.1 billion);
followed by data management, processing, and cloud ($5.9 billion); and Fintech ($5.5 billion).
However, mirroring the broader trend in AI private investment, most AI focus areas saw less
investment in 2022 than in 2021. In the last year, the three largest AI private investment events were:
(1) a $2.5 billion funding event for GAC Aion New Energy Automobile, a Chinese manufacturer of
electric vehicles; (2) a $1.5 billion Series E funding round for Anduril Industries, a U.S. defense products
company that builds technology for military agencies and border surveillance; and (3) a $1.2 billion
investment in Celonis, a business-data consulting company based in Germany.
While the proportion of companies adopting AI has plateaued, the companies that have adopted
AI continue to pull ahead. The proportion of companies adopting AI in 2022 has more than doubled
since 2017, though it has plateaued in recent years between 50% and 60%, according to the results of
McKinsey’s annual research survey. Organizations that have adopted AI report realizing meaningful
cost decreases and revenue increases.
Artificial Intelligence
Index Report 2023
Chapter 4: The Economy (cont’d)
AI is being deployed by businesses in multifaceted ways. The AI capabilities most likely to have
been embedded in businesses include robotic process automation (39%), computer vision (34%), NL
text understanding (33%), and virtual agents (33%). Moreover, the most commonly adopted AI use
case in 2022 was service operations optimization (24%), followed by the creation of new AI-based
products (20%), customer segmentation (19%), customer service analytics (19%), and new AI-based
enhancement of products (19%).
AI tools like Copilot are tangibly helping workers. Results of a GitHub survey on the use of Copilot,
a text-to-code AI system, find that 88% of surveyed respondents feel more productive when using
the system, 74% feel they are able to focus on more satisfying work, and 88% feel they are able to
complete tasks more quickly.
China dominates industrial robot installations. In 2013, China overtook Japan as the nation installing
the most industrial robots. Since then, the gap between the total number of industrial robots installed
by China and the next-nearest nation has widened. In 2021, China installed more industrial robots than
the rest of the world combined.
Artificial Intelligence
Index Report 2023
Chapter 5: Education
More and more AI specialization. The proportion of new computer science PhD graduates from
U.S. universities who specialized in AI jumped to 19.1% in 2021, from 14.9% in 2020 and 10.2% in 2010.
New AI PhDs increasingly head to industry. In 2011, roughly the same proportion of new AI PhD
graduates took jobs in industry (40.9%) as opposed to academia (41.6%). Since then, however, a
majority of AI PhDs have headed to industry. In 2021, 65.4% of AI PhDs took jobs in industry, more
than double the 28.2% who took jobs in academia.
New North American CS, CE, and information faculty hires stayed flat. In the last decade,
the total number of new North American computer science (CS), computer engineering (CE),
and information faculty hires has decreased: There were 710 total hires in 2021 compared to
733 in 2012. Similarly, the total number of tenure-track hires peaked in 2019 at 422 and then
dropped to 324 in 2021.
The gap in external research funding for private versus public American CS departments
continues to widen. In 2011, the median amount of total expenditure from external sources for
computing research was roughly the same for private and public CS departments in the United
States. Since then, the gap has widened, with private U.S. CS departments receiving millions more
in additional funding than public universities. In 2021, the median expenditure for private universities
was $9.7 million, compared to $5.7 million for public universities.
Interest in K–12 AI and computer science education grows in both the United States and the
rest of the world. In 2021, a total of 181,040 AP computer science exams were taken by American
students, a 1.0% increase from the previous year. Since 2007, the number of AP computer science
exams has increased ninefold. As of 2021, 11 countries, including Belgium, China, and South Korea,
have officially endorsed and implemented a K–12 AI curriculum.
Artificial Intelligence
Index Report 2023
Chapter 6: Policy and Governance
Policymaker interest in AI is on the rise. An AI Index analysis of the legislative records of 127
countries shows that the number of bills containing “artificial intelligence” that were passed into law
grew from just 1 in 2016 to 37 in 2022. An analysis of the parliamentary records on AI in 81 countries
likewise shows that mentions of AI in global legislative proceedings have increased nearly 6.5 times
since 2016.
From talk to enactment—the U.S. passed more AI bills than ever before. In 2021, only 2% of
all federal AI bills in the United States were passed into law. This number jumped to 10% in 2022.
Similarly, last year 35% of all state-level AI bills were passed into law.
When it comes to AI, policymakers have a lot of thoughts. A qualitative analysis of the
parliamentary proceedings of a diverse group of nations reveals that policymakers think about
AI from a wide range of perspectives. For example, in 2022, legislators in the United Kingdom
discussed the risks of AI-led automation; those in Japan considered the necessity of safeguarding
human rights in the face of AI; and those in Zambia looked at the possibility of using AI for
weather forecasting.
The U.S. government continues to increase spending on AI. Since 2017, the amount of U.S.
government AI-related contract spending has increased roughly 2.5 times.
The legal world is waking up to AI. In 2022, there were 110 AI-related legal cases in United
States state and federal courts, roughly seven times more than in 2016. The majority of these cases
originated in California, New York, and Illinois, and concerned issues relating to civil, intellectual
property, and contract law.
Artificial Intelligence
Index Report 2023
Chapter 7: Diversity
North American bachelor’s, master’s, and PhD-level computer science students are becoming
more ethnically diverse. Although white students are still the most represented ethnicity among
new resident bachelor’s, master’s, and PhD-level computer science graduates, students from other
ethnic backgrounds (for example, Asian, Hispanic, and Black or African American) are becoming
increasingly more represented. For example, in 2011, 71.9% of new resident CS bachelor’s graduates
were white. In 2021, that number dropped to 46.7%.
New AI PhDs are still overwhelmingly male. In 2021, 78.7% of new AI PhDs were male.
Only 21.3% were female, a 3.2 percentage point increase from 2011. There continues to be a gender
imbalance in higher-level AI education.
Women make up an increasingly greater share of CS, CE, and information faculty hires.
Since 2017, the proportion of new female CS, CE, and information faculty hires has increased from
24.9% to 30.2%. Still, most CS, CE, and information faculty in North American universities are male
(75.9%). As of 2021, only 0.1% of CS, CE, and information faculty identify as nonbinary.
American K–12 computer science education has become more diverse, in terms of both gender
and ethnicity. The share of AP computer science exams taken by female students increased from
16.8% in 2007 to 30.6% in 2021. Year over year, the share of Asian, Hispanic/Latino/Latina, and
Black/African American students taking AP computer science has likewise increased.
Artificial Intelligence
Index Report 2023
Chapter 8: Public Opinion
Chinese citizens are among those who feel the most positively about AI products and services.
Americans … not so much. In a 2022 IPSOS survey, 78% of Chinese respondents (the highest
proportion of surveyed countries) agreed with the statement that products and services using AI
have more benefits than drawbacks. After Chinese respondents, those from Saudi Arabia (76%) and
India (71%) felt the most positive about AI products. Only 35% of sampled Americans (among the
lowest of surveyed countries) agreed that products and services using AI had more benefits than
drawbacks.
Men tend to feel more positively about AI products and services than women. Men are also
more likely than women to believe that AI will mostly help rather than harm. According to the
2022 IPSOS survey, men are more likely than women to report that AI products and services make
their lives easier, trust companies that use AI, and feel that AI products and services have more
benefits than drawbacks. A 2021 survey by Gallup and Lloyd’s Register Foundation likewise revealed
that men are more likely than women to agree with the statement that AI will mostly help rather than
harm their country in the next 20 years.
People across the world and especially America remain unconvinced by self-driving cars. In
a global survey, only 27% of respondents reported feeling safe in a self-driving car. Similarly, Pew
Research suggests that only 26% of Americans feel that driverless passenger vehicles are a good
idea for society.
Different causes for excitement and concern. Among a sample of surveyed Americans, those
who report feeling excited about AI are most excited about the potential to make life and society
better (31%) and to save time and make things more efficient (13%). Those who report feeling more
concerned worry about the loss of human jobs (19%); surveillance, hacking, and digital privacy (16%);
and the lack of human connection (12%).
NLP researchers … have some strong opinions as well. According to a survey widely distributed to
NLP researchers, 77% either agreed or weakly agreed that private AI firms have too much influence,
41% said that NLP should be regulated, and 73% felt that AI could soon lead to revolutionary societal
change. These were some of the many strong opinions held by the NLP research community.
Artificial Intelligence
Index Report 2023
Artificial Intelligence
Index Report 2023
CHAPTER 1:
Research and
Development
Table of Contents Chapter 1 Preview 20
Artificial Intelligence
Index Report 2023
CHAPTER 1 PREVIEW:
Research and Development
Overview 22 Computer Vision 46
Chapter Highlights 23 Natural Language Processing 47
Speech Recognition 48
1.1 Publications 24
Overview 24 1.2 Trends in Significant
Machine Learning Systems 49
Total Number of AI Publications 24
General Machine Learning Systems 49
By Type of Publication 25
System Types 49
By Field of Study 26
Sector Analysis 50
By Sector 27
National Affiliation 51
Cross-Country Collaboration 29
Systems 51
Cross-Sector Collaboration 31
Authorship 53
AI Journal Publications 32
Parameter Trends 54
Overview 32
Compute Trends 56
By Region 33
La |
275 | bcg | the-c-suites-ai-agenda-slideshow-jan-2024-new.pdf | BCG AI RADAR
From Potential to Profit with GenAI
JANUARY 2024
Survey of 1,406 executives provides insights into AI and GenAI sentiment in 2024
Executive roles Respondents from 50 markets (the 13 markets in green have >25 respondents) Industries and key functions
Norway Sweden TMT 252
14% CEO Netherlands
Canada Denmark
UK
Germany
Belgium Austria Consumer 177
France
Switzerland
Spain Azerbaijan
US Italy
Turkey Japan
14% CFO Portugal Malta Greece Israel Pakistan Industrial goods 169
Morocco Egypt UAE Bangladesh
Mexico
Saudi Qatar India Hong Kong Manufacturing 164
Arabia Thailand
Nigeria Philippines
14% CIO Togo Ethiopia
Colombia Sri Lanka Malaysia Financial institutions 156
Kenya
Singapore
Tanzania Indonesia
10% COO Brazil Angola Health care 138
Botswana
Australia
Chile Energy 81
10% CTO South Africa
Argentina
Transportation 68
18% CXO1
Public sector 63
Company revenue
Insurance 59
9% 24% 18% 18% 32%
20% Other2 Travel/tourism 41
$101M–$500M $501M–$1B $1B–$2B $2B–$5B >$5B Marketing 38
Source: BCG AI Radar (2024); n = 1,406 in 50 markets.
Note: Because of rounding, not all percentage totals add up to 100%. TMT = technology, media, and telecommunucations.
1“CxO” represents executives who directly report to the CEO (e.g., CMO, CSO, CISO).
2“Other” executive titles include chair of the board and president.
71%
of executives surveyed say that they plan
to increase tech investments in 2024—an
11-point jump from 2023
Generative AI
will revolutionize
the world—and
executives want
to capitalize
89%
rank AI and GenAI as a top-three tech
priority for 2024, and 51% put it at the top of
their list (cybersecurity and cloud computing
are the other two top priorities)
Source: BCG AI Radar (2024); n = 1,406 in 50 markets.
A global wave of rising tech and AI/GenAI investment
Executives planning to increase Executives planning to increase
71% 85%
their tech investment in 2024 their AI/GenAI investment in 2024
overall overall
Middle East 85% Middle East 93%
Asia-Pacific 80% Europe 86%
Africa 77% Asia-Pacific 85%
Europe 68% North America 85%
North America 65% Africa 82%
South America 63% South America 75%
Source: : BCG AI Radar (2024); n = 1,406 in 50 markets.
Note: In Asia-Pacific, n = 308; in North America, n = 303; in Europe, n = 647; in the Middle East, n = 28; in South America, n = 51; in Africa, n = 69.
However, most organizations are Top three reasons for dissatisfaction
not doing enough to realize the
benefits of the technology.
1
Lack of talent and skills
66%
Unclear AI and GenAI roadmap
2
and investment priorities
of executives are ambivalent or
outright dissatisfied with their
organization’s progress on AI and
No strategy for responsible
3
generative AI so far.
AI and GenAI
Source: BCG AI Radar (2024); n = 1,406 in 50 markets. For executives reporting dissatisfaction, n = 310.
62%
46%
say their firms are
still waiting to see
how AI-specific
regulations develop
Executives
of their workforce,
on average, will need
across the board to undergo upskilling
in the next three
years due to GenAI
face pressing
challenges
6%
of companies have
managed to train
more than 25% of
their people on
GenAI tools so far
Source: BCG AI Radar (2024); n = 1,406 in 50 markets.
Executives who report that more than 25% of
their workers have trained on GenAI tools
Middle East 11%
North America 8%
Executives worldwide must
boost upskilling, as Europe,
Africa, and South America Asia-Pacific 7%
are falling behind.
Europe 5%
Africa 3%
South
2%
America 6%
Source: BCG AI Radar (2024); n = 1,406 in 50 markets.
overall
Note: In Asia-Pacific, n = 308; in North America, n = 303; in Europe, n = 647;
in the Middle East, n = 28; in South America, n = 51; in Africa, n = 69.
9 0%
are either waiting for These are the observers. They
GenAI to move beyond the are opting for a wait-and-see
hype or experimenting in approach.
small ways.
That’s not an option with
generative AI.
Source: BCG AI Radar (2024); n = 1,406 in 50 markets.
Winners invest for productivity and topline growth.
1
They target 10%+ productivity gains and reinvest for revenue uplift.
Winners are upskilling systematically.
2
They are scaling their learning muscle—and that extends to
Winners are
executives as well.
acting now—
Winners are vigilant about cost of use.
here’s how they’re 3
They understand that cost of use has long-term implications and
must command attention now.
staying ahead
Winners build strategic relationships.
4
They develop an ecosystem of partners to manage complex and rapidly
evolving challenges.
Winners implement responsible AI (RAI) principles.
5
They put RAI on the CEO agenda and proactively plan for emerging
Source: BCG AI Radar (2024); n = 1,406 in 50 markets. policies and regulations.
Percentage of companies expecting cost savings
in 2024
One of the biggest benefits
that GenAI promises is
1.3x 1.5x
productivity gains.
The potential benefit is
even greater for companies
that invest more—they’re 70%
67%
54%
46%
1.5x more likely to
anticipate upward of 10%
in cost savings.
Any cost savings More than 10% cost savings1
All companies Companies expecting to invest more than
$50 million in AI/GenAI in 20242
Source: BCG AI Radar (2024); n = 1,406 in 50 markets.
1Of companies expecting cost savings.
2For companies expecting to invest more than $50 million, n = 122.
Key goals for growth with AI and GenAI investments
1.3x
1.4x
The key is to invest
in productivity—and
topline growth. 80%
66%
62%
48%
Expand market access Build business adjacencies
All companies Companies expecting to invest more than
$50 million in AI/GenAI in 20241
Source: BCG AI Radar (2024); n = 1,406 in 50 markets.
1For companies expecting to invest more than $50 million, n = 122.
The imperative to
provide GenAI training
is clear.
Executives believe that
46%
of workers, on average, Overwhelming majorities
will need to be reskilled believe that GenAI will
in the next three years. create new roles (81%) and
require significant change
management (74%).
Source: BCG AI Radar (2024); n = 1,406 in 50 markets.
Companies with more than 25% of their
workforce trained on GenAI tools
3.5x
Companies that invest
more are ahead on
reskilling their workers—
and on building their
learning muscle at scale.
21%
6%
All companies Companies expecting
to invest more than
$50 million in
Source: BCG AI Radar (2024); n = 1,406 in 50 markets. AI/GenAI in 20241
1For companies expecting to invest more than $50 million, n = 122.
Confidence in the executive team’s GenAI proficiency
The need to upskill
extends to the C-suite.
Completely confident 1%
59%
Very confident 11%
Confident 29%
of leaders surveyed say
they have limited or no
Limited confidence 40%
confidence in their
executive team’s
No confidence 19%
proficiency in GenAI.
Source: BCG AI Radar (2024); n = 1,406 in 50 markets.
Most important consideration when choosing
an AI and GenAI solution
Cost of use, which
IP and data protection 39%
has serious long-term
implications, is not
commanding the
Quality and performance 32%
attention it should
Cost 19%
Source: BCG AI Radar (2024); n = 1,406 in 50 markets.
Potential partners seen as a trusted source
of information
Winners are
Big tech platforms 71%
building strategic
relationships
with an evolving
Software providers 49%
ecosystem of
partners
GenAI companies 38%
Source: BCG AI Radar (2024); n = 1,406 in 50 markets.
Companies investing more are getting a head start
Company is already
preparing for AI-specific 38% 72%
regulations
The sheer speed of
GenAI adoption makes
RAI more important than
Company has
ever, and organizations guardrails in place for 50% 68%
using AI/GenAI at work
must be proactive in
addressing this.
14% 27%
CEO is in charge of RAI
All companies Companies expecting to invest more than $50 million in AI/GenAI in 20241
Source: BCG AI Radar (2024); n = 1,406 in 50 markets.
1For companies expecting to invest more than $50 million, n = 122.
Deploy GenAI in everyday
tasks to realize 10% to 20%
productivity potential.
Three value plays
Reshape critical functions for
to maximize GenAI’s
30% to 50% enhancement in
efficiency and effectiveness.
potential
Invent new GenAI business
models to build a long-term
competitive advantage. |
276 | bcg | BCG-Executive-Perspectives-Future-of-Data-Management-with-AI-EP9-10Dec2024.pdf | Executive
Perspectives
The Future of Data Management
with AI
December 2024
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
most recent learning in a new series designed to help CEOs navigate AI. With and value of
AI at an inflection point, the focus in 2024 is on turning AI’s potential into
the future of data
real profit.
management with AI
In this edition, we discuss the future of data management, and the role AI will play in
fundamentally transforming the function. We address key questions on the minds of
leaders:
• How do I keep pace with growing data regulations?
• How can I unlock cost advantages while improving my data quality?
.d
e
v
re
• How can I improve my data team’s experience and generate more enthusiasm
se
r
sth
g
around data management? ir
llA
.p
u
• How do I get started…and how do I get this right? o rG
g
n
itlu
s
n
o
C
n
This document is a guide for CEOs and technology leaders to cut through o ts
o
B
y
the hype around AI in data management and understand what creates value b
4
2
0
now and in the future. 2 ©
th
g
iry
p
1 o
C
Executive summary | GenAI will industrialize the use of data, improving quality,
expanding and simplifying access, and increasing productivity
Data management, a manual and tedious job, is overwhelmed with growing unstructured data
Acting fast is key
(>10x in 10yrs), higher quality bars, and tighter regulatory oversight
to tackle rising
complexity and
Economics are turning out to be even more challenging, with data costs projected to grow 80% from
costs
2023 to 2028, and with hidden costs further fueling data cost growth
GenAI can support GenAI simplifies and augments data management tasks, accelerating time to value. Coupled
the needs of your with the right tooling, GenAI offers the potential to automate and expedite key data tasks, improving
data team… data quality and unlocking efficiencies
.d
e
v
AI vision for data management is to drive competitive advantage through improved data quality, re
se
expanded coverage, self-service analytics, and automated workflows, transforming roles and
r
sth
g
democratizing data access with scalable, secure, and compliant solutions
ir
llA
.p
u
…by reshaping o rG
g
n
your data function Five key drivers ensure a successful transformation: itlu
s
n
for innovation and • Adopt an AI vision to drive sequential building of your data capability o
C
n
o
growth • Rewire your end-to-end data management workflows to unlock efficiency gains ts
o
B
y
• Employ a product-centric data operating model b
4
2
0
• Ensure a robust data governance across the life cycle 2
©
th
• Invest in key partnerships to accelerate capability builds g iry
p
2 o
C
Have we started leveraging AI to reshape our
data function?
Do we have a refreshed data strategy, enabled by
GenAI, that is aligned with business outcomes?
How is our data function lined up to respond to the
growing data needs of the business?
2
Key questions
How did our direct and indirect data costs change
CDO/CIOs
in the past years?
should answer
How are we embedding GenAI data considerations
.d
e
v
re
in technology and operating model priorities? se
r
sth
g
ir
llA
.p
How did we adapt our data-related people, processes, u
o
rG
g
n
and organization to expanding needs of AI? itlu
s
n
o
C
n
o
ts
How do we manage data-related risks and ensure o B
y
b
4
2
adherence to evolving regulatory requirements? 0
2
©
th
g
iry
p
3 o
C
Today's data management is already burdened
by three key areas of friction along the data journey
Accessing data Understanding data Governing & monitoring data
• Processes today are more difficult to • Data annotation is a labor-intensive • Data usage monitoring is not actively
manage due to a significant increase in process and, even if automated, requires performed
access groups human-in-the-loop
• Regulations frequently change, making
• Organizations include multiple levels of • Data annotation is the primary enabler compliance challenging
approval - delays of up to six weeks to the remainder of the data journey
• Setting policies requires alignment
• Changing policies and numerous rules • Data stewards are either not qualified or across stakeholders with conflicting
require consistent management not accountable priorities
..dd
ee
• Concerns are increasing around vv rree
ssee
Intellectual Property (IP) rights and
rr
sstthh
gg
third-party data
iirr
llllAA
..pp
uu
oo
rrGG
gg
nn
iittlluu
ss
nn
60% of a data scientist's time is spent Due to the manual and time-consuming nature Although General Data Protection Regulation oo CC
nn
oo
waiting for data of the job, we're seeing highest churns for data (GDPR) allows customers to request data ttss
oo
BB
yy
custodians and steward`s deletion, for legal purposes we need to keep data bb
- BU CDO, Global Energy Company 44
22
00
for 6 years in case of litigation 22
- Executive, GSIB1 ©©
tthh
gg
- CDO, Global P&C insurer iirryy
pp
1.Global systematically important bank 4 oo
CC
In addition, exponential growth in the volume of unstructured and multimodal
data in the past decade has further raised the bar on data management
Data is growing exponentially; This growing volume with AI will increase demand for
~90% unstructured data by 2025 data management activities
Global data in • Identification of source and history of data
181 Data provenance
zettabytes1 • Authenticity of model data for intended use
147 10% Where did data come from?
~14x • Third-party training data underlying models
Structured
64 90% • Training data accuracy for desired output
Data classification
• Quality and consistency of labeled data
Unstructured2 How is the data labeled?
13 • Reduced training, improved model performance
2014 2020 2023 2025E
Data lineage • Traceability of data transformation
What is the sequence of processing • Reproduction of calculated results .d
Mobile, real-time data and IOT sensors steps? • Interpretation of the data used for model e v re
creating large amounts of data
se
r
sth
g
Data quality, metadata • Greater quality of model outputs
ir
llA
Synthetic data generation via GenAI .p
will drive strong growth of data volumes completeness • Higher model performance without bias u o rG
How accurate is my data? • Management of data drift and concept drift g n
itlu
s
n
o
Decisions enabled by AI drive companies C
Regulatory compliance • Multiple data regulations globally (e.g., EU AI Act) n
to collect more data than ever o ts
Is this usage ethical & within • Higher quality bar imposed by regulators o B
y
b
regulations? • Vagueness of regulations on unstructured data 4
2
With AI, data inputs becoming 0 2
©
'multimodal', widening tappable landscape th
g
iry
1. A zettabyte is 1 billion terabytes 2. Unstructured data includes information that is not stored in a structured database format including audio, video, emails, customer reviews, etc. 5 p o
C
Source: IDC; Seagate; Statista estimates
Higher standards imply increased data management direct
and hidden costs, which are forecast to rise in the coming years
Around 80% growth estimated for data management Also, hidden costs of data management will drive
costs in the next 5 years1 further increases across the board
2023-28
CAGR Manual interventions driving human cost
181
(e.g., BU analyst effort on data modifications)
CAGR
13% 42 Software 12%
Delayed analytics, use-case time to value
(e.g., opportunity cost of delayed business decisions)
25 Hardware 9%
Indexed to
100 100 Remediation to regulatory inspection .d
e
v
re
24 47 Internal people 15% (e.g., issue identifications and corrections) se r sth
g
17
ir
llA
Fines and security breaches (e.g., due to
.p
u
o
rG
24 noncompliance with regulatory guidelines) g
n
itlu
68 Services 14% s n
o
C
n
36 o
Overspending on technology and engineering ts
o
B
(e.g., overlapping data management tools) y b
4
2
2023 2028E Overall: 13% 0 2
©
th
g
1. Incurred by IT function, with existing processes & tech iry
p
Source: IDC Semiannual Software Tracker; IDC Worldwide ICT Spending Guide; WEF Future of Jobs Report; Economic Intelligence Unit, Gartner Forecast Analysis: Data and Analytics Services; 6 o C
BCG publication ‘A New Architecture to Manage Data Costs and Complexity’
In this challenging environment, GenAI can help simplify data management
and accelerate time to value across the data management value chain
CLEAN MATCH
and refine data data through identifying similar
or related data
GenAI can interpret
and create new
GENERATE ENHANCE
content, implying
new data data traceability
potential to augment
.d
e
or automate v re
se
many key data
IDENTIFY ACCELERATE r
sth
g
ir
llA
types of data compliance and risk management .p
management tasks u
o
and infer metadata
rG
g
n
itlu
s
n
o
C
AUGMENT n
o
ts
o
B
y
b
data analytics and 4 2
0
2
insight generation ©
th
g
iry
p
7 o
C
Client example | A build-out of GenAI-led metadata labeling and lineage annotation
capabilities enabled significant productivity gains
A global financial We helped the client by focusing five key impact levers… …achieving tangible results
institution client aspires
• Help accelerate identification of potential gaps/risks
to accelerate its data
(e.g., code/data duplication) through lineage capturing Human acceptance
management and
70%+
of "LLM out-of-box"
Enhance data • Enable data-related roles to focus on “value-added”
governance controls at scale tasks (e.g., review of outputs) business description
transformation journey
• Streamline continued monitoring and refresh of
data estate, with less manual intervention
Accuracy
•1 Current data governance
in PII1 tagging and data
only covers small portion 90%+
Improve accuracy & • Improve accuracy and ensure comprehensive coverage lineage captured post-
of data estate (only
focusing on critical data coverage of output of lineage and business metadata human validation
assets)
.d
e
v
•2 Heavily manual Drive efficiency to • Augment productivity for data-related roles Productivity boost
re
se
processes are needed to accelerate (e.g., data steward, central data governance function) to accelerate coverage
r
sth
g
generate business coverage • Boost productivity by up to 50% for critical manual tasks 50%+ of data under ir llA
metadata (weeks per data
governance
.p
u
source) with limited o
rG
alignment between Unlock additional g n
enterprise and BU
use cases and
• Create development and deployment patterns for itlu
additional use cases Reduction s n
o
processes C
•3 Low efficiency of E2E in compliance timeline n
o
data lineage and current 2-5 Yr. for high-impact data ts o B
tooling cannot generate Increase colleague • Minimize repetitive manual tasks and improve working assets y b 4
cross-system lineage satisfaction experience for targeted users 2 0
2
©
th
g
iry
p
1. PII = Personal identifiable information 8 o
C
To control rising data-related challenges, it is imperative
for companies to work toward the AI reshape vision
Empower data management to drive competitive advantage through improved data quality,
AI vision for
expanded coverage, self-service analytics, and automated workflows, transforming roles
data mgmt.
and democratizing data access with scalable, secure, and compliant solutions
FROM … TO …
Data management generally considered an Empowered data management function transforming
afterthought in management priorities data as a driver of competitive advantage for organization
Data management focus restricted to critical Expanded data management coverage, to significantly
.d
e
areas (e.g., data under regulatory purview) enhance data quality and utilization across functions v
re
se
r
sth
g
Siloed, centralized approach to data Departments empowered with self-service analytics ir
llA
management hindering access to data and insights, unlocking data potential across organization
.p
u
o
rG
g
n
itlu
Nature of data management work perceived Reinvented roles with engaging, joyful, and strategic s
n
o
C
as mundane and unappealing responsibilities, adding visible value to organization n
o
ts
o
B
y
b
Core data activities (collection, cleaning, etc.) Core data activities automated and streamlined with 4
2
0
2
manual, restricted to data engineers/stewards business owner involvement ©
th
g
iry
p
9 o
C
Five key drivers can ensure that companies are on a path
in line with the vision for data management
Translate your AI-driven data management
vision to drive sequential build
Support your AI-enabled vision for data by starting
Accelerate
with building a platform and foundational capabilities,
ensuring business alignment and enabling more AI data governance
advanced capabilities
Fundamentally transform AI data governance from the
sidelines of IT into a core, daily business practice —
Rewire your E2E data
GenAI-led embedding standards, control, and governance culture
management workflows
across business units
“RESHAPE”
Fully rewire data management workflows,
minimizing manual iterative loops via of data
.d
automated processes and AI interfaces to management e v re
accelerate time to value se
r
sth
g
Employ a product-centric model
Invest in key partnerships to
ir
llA
.p
u
o
address capability gaps rG
to data, as part of org-wide
g
n
platform operating model Identify and drive partnerships with technology
itlu
s
n
o
providers to address gaps and accelerate capability C
n
Establish efficient and effective ways of working to o
builds across the data value chain ts o
drive faster throughput via transforming from a B
y
b
reactive, service-based data model to a more proactive 4 2
0
2
‘business partnership’ product model ©
th
g
iry
p
10 o
C
Value-centric approach starts and delivers value early from AI,
and matures organizational capabilities and data/tech foundations as you go
Plan your journey Experiment with AI Mature your foundations Consolidate and scale
Define vision, business Resolve initial issues and Establish teams, expand data Accelerate delivery of
opportunity, and key pain points, train core platforms, achieve economies next wave of capability
workflows impacted teams, activate key squads of scale builds
.d
e
v
re
se
r
sth
g
ir
llA
.p
u
o
rG
Value creation through successive waves of capabilities g
n
itlu
s
n
o
C
Governance coverage of data domains n o
ts
o
B
y
b
4
2
Democratizing access and ease of data use across organization 0 2
©
th
g
iry
p
11 o
C
But to unlock value from data as capabilities are deployed, it’s key to sequence
initiatives by building capabilities in waves
Start with fundamental Build advanced capabilities Activate ‘insight generating’
1 2 3
‘no-regret’ initiatives for regulatory compliance capabilities for business
Metadata and lineage annotation are High-quality metadata and lineage High-quality metadata, lineage, MDM,
logical starting points, to accurately enable consistent tagging of data sets, and synthetic data will better support
describe/catalog data consistently, and implying better discoverability and discovery and utilization, and
are critical for other GenAI capabilities to improved document cataloging & generate better insights for business
be effective (e.g., data quality related) mgmt. operations for better compliance (e.g., through augmented analytics)
Policy compliance mgmt.
Augmented analytics
.d
e
Document creation & mgmt. Document creation & mgmt. v re
se
MDM augmentation MDM augmentation
r
sth
g
ir
llA
.p
Enhanced data mining Enhanced data mining u
o
rG
g
n
Synthetic data generation Synthetic data generation itlu
s
n
o
C
Data quality management Data quality management Data quality management n o
ts
o
B
y
Metadata labeling Metadata labeling Metadata labeling b
4
2
0
2
©
Lineage annotation Lineage annotation Lineage annotation th
g
iry
p
12 o
C
Directly enable Augment
AI is anticipated to have the largest impact on data analytics and
data management workflows, necessitating focus across people and processes
In which of the following IT processes and workflows do you anticipate
AI technologies having a transformative impact? [Multi-select]
70%
% of respondents
> 40%
30-40% of AI
< 30%
transformation
effort should be
Application performance invested in
.d
e
v
re
monitoring se
people and r sth
g
Project and portfolio Software development life
Data analytics
processes
ir
llA
.p u
management cycle o
Drive change management
rG
g
n
and other processes related itlu
IT infrastructure (data center, Data management s n
to people o
IT service management C
networks) operations workflows n
o
ts
o
B
Example to follow
y
b
IT asset management 4 2
0
2
and maintenance ©
th
g
iry
p
13 o
C
Source: BCG Build for the Future 2024 survey (n=1,000 respondents)
TODAY: Enrich metadata manually TOMORROW: ~50% effort reduction
against registry in metadata labeling with GenAI
Human intensive
Illustrative example Input data & associated Input data & associated
metadata info metadata info
Automated
Human to identify data domain GenAI to identify data domain
and business process associated and business process associated
Human to identify proper data GenAI to route to proper data
registry to look up registry and validate
Data registry
.d
e
Look up in registry, identify GenAI to create anomaly list v re
Data registry se
anomalies, and gaps of metadata and recommended changes
r
sth
g
ir
llA
.p
u
o
rG
Human to remediate identified Human to review and g n
itlu
anomalies and enrich metadata accept/reject s n
o
C
n
o
ts
o
B
y
b
Data storage Data storage 4 2 0
2
©
th
g
GenAI covers steps that take ~50% of the effort
iry
p
14 o
C
In addition to process, key data management roles are being transformed
to be more interesting, interactive, and productive with GenAI
i ii iii iv v
Chief data Data governance Data domain Data Data
officer (CDO) office owner (DDO) steward custodian
GenAI
Low Medium Medium Very High High
impact
Drives enterprise-wide data Operationally supports rollout of Is responsible for a specific data Operationally supports DDOs Determines strategic direction
strategy and culture, champions enterprise-level data governance domain (global/ regional) with set of data families for data platforms
data governance and and data culture
evangelizes org on data Refines data domain taxonomy Proposes remediation actions Ensures implementation of data
Prepares requirements to roll- and glossary and roadmap strategy on data platforms and
Typical
Coordinates prioritization of out governance of data other relevant IT systems
response
actions / data quality domains, including taxonomy, Produces data domain Implements data governance
remediation plans identifying roles in organization heatmap, defines data quality policies and processes Manages IT architecture for
targets and measures of data platforms
.d
Trains employees on data data quality Aligns with data custodian on IT e v
re
management roles needs
se
r
sth
g
ir
llA
.p
u
o
How Augments and facilitates Allows employees to explore Augments data domain creation Augments data glossary, Automates tagging and labeling, rG
g
GenAI improvements to data supply policies and data management, and maintenance of taxonomy, dictionary, ontology generation monitors compliance of data n itlu
chain (e.g., data quality through chatbots, improving ontology, and glossary; and coherence; automates data strategy on data platform and s n
drives o
C
assessment and efficiency and reducing ad hoc automates reporting (e.g., map creation and data glossary other systems n
value o
remediation planning) support and training heatmap and data quality) updates
ts
o
B
y
b
4
2
0
2
©
As technology enables bionic features, the burden of these roles will be reduced, enabling them to scale to higher-value tasks
th
g
iry
p
15 o
Source: BCG Marketing Org & Op Benchmarks C
A move from today's siloed structure toward a product-centric approach
will improve data accessibility and consumption
TODAY TOMORROW
Business communicates consistent Data product manager understands
Business Business
needs to internal data scientist needs of the business
Data
scientist Data product team
Data scientist needs data engineer
to construct models
Data DDP1 framework Data product manager builds backlog and
engineer user stories with data product team
Data engineer constructs models
in data lakehouse
Data product team sources,
Data Enabled by
cleanses, and documents data
scientist
Data scientist requests additional tools changes in .d e
v
re
Infra and
from infrastructure team approach to Data
p
lt ae ta fom
rm
se tl ef- ap mro v ti os i to rn ans st fo oo rl ms f dro am
ta
the se r
sth
g
Ops
Data scientist transforms data to be
gd oa vt ea r,
n
r ao nle cs e,
, PM notifies the business team about new data
ir
llA
.p u
leveraged by business team o rG
and tech product releases with new features g
Business Platform n itlu
architecture team s n
Independent governance team enforces Data product team enforces quality and o
C
n
data quality and policies security policies o ts
o
B
y
b
4
2
0
2
“Service request fulfillment” model Business partnership “product” model ©
th
g
iry
p
16 o
1. DDP = data and digital platform. Please refer to "The Future of the AI-Driven Tech and People Stack" Executive Perspective for more details C
Establish a centralized system to fundamentally transform AI data governance
through integrated standards and control across business units
Federated data governance organization to ensure
standards, control, and governance culture across BUs
Governance council, formed of data domain owners,
1
Illustrative
steers strategic priorities and establishes global policies
1 Governance council
(including responsible use of data), embedding standards
2 Data domain forum across business units
Data domain
2 Data domain forum sets the overall strategy and
3
m priorities for the data domain to ensure that the domain's
Data domain ownership
a data product portfolio meets consumer requirements
e
t
e
c 3 Data domain ownership ensures that global
n .d
a
n
Data stewardship priorities are enforced in the data domain and data quality e v re
se
r
e
issues are escalated to the council, instilling data r
sth
v
o
g
e r
4
Data product Data product 4
Dgo ave tr ana pn rc oe
d
c uul ct tur oe win
n
B eU rs
ship manages the data
g ir
llA
.p
u o rG
o ownership ownership g n
C product's E2E life cycle and ensures data availability,data itlu
s
n
usage,and consumer satisfaction o
C
n
o
ts
o
Data custodianship 5 Core governance team drafts policies for the B y
b
4
governance council and supports data domains in the 2
5 0
2
©
implementation of data governance capabilities
th
g
iry
p
17 o
C
Partnerships with technology providers are key to address gaps
and accelerate capability development across the data value chain
Sample processes Partners/Tools1
Data quality automation
AI tools identify and correct errors, inconsistencies, and inaccuracies in data, as well as enrich data
with additional information
CLEAN
Data anonymization
Anonymize data to protect privacy while preserving utility and integrity
Synthetic data generation
Generate synthetic data that resembles real data to protect privacy and facilitate safe data sharing
Code generation
GENERATE
Automatic functional code generation, optimization, and rewriting using trained LLMs results in
rapid prototyping and speeds up development cycles
.d
Data labeling e
v
re
AI automates process of labeling data/suggesting answers for human review and verification
se
r
sth
g
IDENTIFY Data classification
ir
llA
.p
Automate the process of classifying data into different categories such as sensitivity, type u o
rG
g
n
MDM automation itlu
s
n
Auto MDM (master data management) powered by AI intelligently identifies and reconciles data, o
C
reducing manual effort and improving data quality n o
ts
o
MATCH
Data lineage B
y
b
Trace the path and transformations of data from its source to destination, ensuring transparency 4 2
0
2
and accountability in data handling, which helps in verifying data quality and compliance ©
As of Q3 2024
th
g
iry
1. Non-exhaustive 18 p o
C
Source: IDC; Gartner; secondary research
How to get started | Four key activities can activate
the building of strong foundations and accelerate time to value
Develop core capabilities & strategies Extend capabilities across data Fully integrate AI into data
to effect a data-driven transformation management opportunities management at org-wide level
1 Develop your data strategy: Adopt a value-centric strategy to prioritize opportunities and align business outcomes,
data platforms, and assets to unlock outcomes over time
Assess existing setup: Review how your data function is positioned along the five drivers, understanding impact on
2
workflows and key AI opportunities that map to it
.d
Focus on fundamental capabilities: Introduce and pilot AI with lineage annotation, metadata labeling and data e v
3 re
se
quality management, leveraging third-party vendors as needed r sth
g
ir
llA
Fill talent gaps: Upskill existing workforce and recruit new data-related talent that is needed to support transformation, .p
u
4 o
rG
while protecting upskilling time required
g
n
itlu
s
n
o
C
n
o
ts
o
Supported by embedding an AI culture organization-wide, driven by leadership, to empower B
y
b
4
2
teams in leveraging AI for sustainable and scalable growth, and to boost adoption 0
2
©
th
g
iry
p
19 o
C
NAMR
BCG experts |
Key contacts Dylan Vladimir David Amanda Matthew
Bolden Lukic Martin Luther Kropp
for AI data
Sesh Julie Beth Djon Steve
management
Iyer Bedard Viner Kleine Mills
transformation
Benjamin Renee Helen
Daniel Bo
Rehberg Laverdiere Han
Martines Xu
Vikram Tauseef
Sivakumar Charanya
EMESA APAC
.d
e
v
Nicolas Jessica Marc Jeff Romain de
re
se
de Bellefonds Apotheker Schuuring Walters Laubier r sth
g
ir
llA
.p
u
Dan Andrej Marcus Julian Aparna o rG
Sack Levin Wittig King Kapoor g
n
itlu
s
n
o
C
n
Robert Lucas Akira Nipun o ts
o
Xu Quarta Abe Kalra B
y
b
4
2
0
2
©
th
g
iry
p
20 o
C |
277 | bcg | BCG-Executive-Perspectives-Unlocking-Impact-from-AI-HR-EP1-30July2024.pdf | Executive
Perspectives
Unlocking Impact from GenAI
Human Resources
July 2024
1
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
Introduction In this BCG
Executive Perspective,
we show you how to
As part of our ongoing series of C-suite conversations on AI, we are sharing our most
recent learnings in a series designed to help navigate the rapidly changing world of
leverage AI to transform
AI. After working with over 1,000 clients in the past year, we've found that AI is at an
and create value in HR
inflection point: in 2024, the focus is on turning AI's potential into real profit.
In this edition, we discuss the future of human resources (HR) and the role AI
will play in turbocharging the function's capacity to deliver on unprecedented
demands.We address key questions on the minds of HR leaders, including:
• What will my HR organization look like – both how we are structured and what
tools and skillsets are required?
• How can we achieve near-term performance gains with AI and GenAI while
building the necessary capabilities?
• Given the sensitivity of our work, how can we proactively address ethical and
employee/candidate experience risks?
• How can we drive adoption, engagement, and adherence to capture value?
This document is a guide for CEOs and CHROs to cut through the hype
around AI in HR and understand what creates value now and in the future.
1
2
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
Executive summary | Unlocking impact from GenAI in human resources
The changing nature of work is placing unprecedented demands on HR, e.g., organization-wide upskilling and
GenAI will behavioral change
enable HR to To meet these demands, HR can leverage Generative AI (GenAI) to become more productive, effective, and engaged
deliver against (e.g., ~20-40% productivity improvement)
new demands
In the near term, many leaders are starting with significant opportunities inr ecruiting (e.g., 20-25% near-term cost
reduction) and admin
HR organizations are investing in key enablers to re-imagine the function, including:
Foundational • Org and op model: Reorienting around employee experience, with new roles to shape and govern GenAI
investments will • Talent and skills: Up/re-skilling and hiring specialists to address 55-75% skill disruption in key HR roles
be required to • Data, tech, and partnerships: Preparing data and infrastructure, partnering to assemble portfolio of tools
capture value • Risk and responsible AI: Addressing potential bias and ensuring compliance with regulatory requirements
• Change management: Taking a science-backed approach to change behaviors and drive adoption
Most leaders are already taking action with GenAI in HR (e.g., among enterprises already deploying GenAI,
70-80% are using it in HR)
HR leaders must
act now
To get started, HR leaders must build integrated implementation roadmaps, upskill fellow leaders, and
prepare HR data and guardrails to ensure reliability, compliance, and sustained value capture
3
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
HR organizations face unprecedented demands
as the future of work unfolds
Key trends in the future of work Requirements of HR
Personalized strategies to
• Dynamic, more competitive rewards/benefits
attract and retain top talent in • Personalized L&D1 pathways and career journeys
a cost-constrained environment • Deep focus on DE&I2 – on people and tech fronts
• Up-/re-skilling of non-tech talent to address GenAI
Skill disruption and human- disruption (e.g., 86% expect need in near-term)
machine teaming • Continual re-design of ways of working, teams, roles
• Behavioral change and human-machine trust
• Dynamic talent planning to address technical skill
Rising need for – and lagging gaps via hiring and re-skilling (e.g., to address >3x
supply of – tech talent increase in demand for data scientists in past 5 years;
future demand may vary with GenAI)
1. Learning & Development; 2. Diversity, equity & inclusion; Source: BCG skills disruption index; How to Attract, Develop, and Retain AI Talent (2023); BCG experience
4
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
To deliver against these demands, HR of the …leveraging GenAI for
future must be fundamentally different… step-changes in…
Business
partners
Productivity
20-40% 90%+
Increase for HRBPs and Boost for some
HRBP
CHRO recruiters administrative workflows
Talent ecosystem
Strategic consultants and
The CEO’s strategic
The skills portfolio manager
organization architects
partner for a future-
ready organization
Speed and effectiveness
10X 50%
HR
function
Work rhythms
of the Product owners
The conductors of Faster content Decreased time to hire
learning and innovation future Guardians of AI- creation
enabled employee
experience
Engagement
HRIS
3X 25%
HR-IT experts of the new
Shared services
strategy and digital function
Ethics and bias
Short-term heroes of Growth in employee Rise in HR retention
Strategy and digital The champions of integrity standardization, long-
engagement
4
function of the future and tech bias prevention term role divergence
Source: BCG workforce diagnostic; BCG 'toil v. joy' diagnostic; BCG experience
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
5
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
E2E process re-imagination is critical for HR to break the historical
compromise between productivity and employee engagement
Anticipate Attract Develop Engage
HR strategy Recruiting HR admin/ Compensation Learning and Performance and Employee Employee
and planning and resourcing shared services and benefits development career mgmt. engagement relations
From: Manual data Losing top Siloed teams Information stored Time-intensive Highly reliant on Reliant on Manual and
producing static candidates due to providing in different places, content creation, human opinion, employees to time-intensive
results lengthy process fragmented time-intensive to fragmented often providing raise issues/ documentation
and human bias service and get answers employee feedback too late concerns and reporting
long wait times experience
To: Dynamic Humans Streamlined, Chatbots and self- Personalized Delivering real- Predicting issues Automating
forecasting of augmented with faster support service tools to L&D journey and time feedback by monitoring and admin to enable
future workforce automated from 'hire to retire' help employees clear career path and objective analyzing time for human
needs and sourcing support, find information to increase performance employee engagement
re-designing freeing up time quickly engagement and insights analysis sentiment around inquiries
workflows retention
Productivity
increase
Engagement
increase
Productivity increase: <10% ~10-25% ~25-50% >50%
Engagement increase: Low Medium High
Source: BCG workforce diagnostic; BCG 'toil vs. joy' diagnostic; BCG experience
6
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
Sensitivity and risk
Productivity impact
Talent/L&D
Early adopters of Recruiting Interactive Engagement
Content Admin workforce Employee
creation HR chatbot planning listening
GenAI in HR are
5%
12%
starting with low-
28%
25%
18% 40%
hanging fruits that
present lower risk and
44%
offer higher near-term 36%
70% 70%
productivity gains
28%
24%
Examples follow
We are not considering it
We are considering it
We are implementing, piloting, or scaling
1. Have you started experimenting on the following applications?
Source: BCG survey of CHROs or direct reports to CHRO, n=64 6
7
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
Example 1 | Recruiting - Multi-year journey re-imagines recruitment
as part of broader professional services HR transformation
Where did they start? What are they doing? Value identified
Professional services firm facing significant From 2020-2022: 75%+ solution adoption to date driving
changes in demand for talent, including: • Diagnostic including identification of impact at scale, including:
• Greater quantity pain points and differentiators
• Broader range of skillsets • Partner diligence and selection
Decrease in time and
20-25%
• Increased diversity • Future architecture definition
expense
• Experimentation and pilots, including
In parallel, candidates and recruiters workflow re-design and standardization,
Overall recruiting cost
dissatisfied with talent acquisition tools, training, RAI guardrails, change mgmt 20-25%
reduction
putting brand and ability to capture top-tier
talent at risk From 2023-2024:
Lift in offer conversion
• E2E scaling using agile approach (e.g., 10-15%
per recruiter
Firm had formulated an organization-wide 2-week sprints, dedicated product owners)
recruiting vision and made progress on – Consolidation from many complex
quick wins, e.g., virtual recruiting tools workflows to three global standards Plus higher data fidelity, global KPI
– Streamlining from 5x ATS tools to 1 standardization, and better recruiter +
As next step, desire to pursue E2E digital • Continued RAI guardrail refinement candidate experience
transformation featuring GenAI to • Iteration of candidate communications to
accelerate performance, unlock time for ensure ongoing transparency Lessons learned informing broader go-
deeper human thinking and engagement forward HR opportunities
Source: BCG experience
8
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
Example 1 | Recruiting – E2E re-design unlocks time
for more strategic, engaging work
Recruitment and Plan hiring Attract and source Screen Facilitate Hire and onboard Avg. duration and
resourcing journey needs candidates candidates interviews candidates engagement1
Manually assess Manually review Negotiate and Illustrative
Lead interviews for
and forecast talent Consult hiring resumes and cover extend offers
each candidate
needs manager and other letters
stakeholders on 25-30%
Select priority Summarize and Iterate onboarding
Current profile requirements
Align stakeholders candidates score each process as needed
state on forecast candidate post- 30-35%
Schedule interviews interview
Manually develop,
Oversee onboarding
review, and post Conduct phone Send candidate 35-45%
for new hires
job ads screenings follow-ups
Up to 50% reduced
GenAI enables GenAI generates
GenAI automates time to hire
GenAI creates video/text-based personalized
AI predicts resume synthesis,
targeted job interviews compensation and
workforce need with highlighting
postings onboarding plans
human involvement competency based
Future state HR selects
on specific skills
HR review and candidates from HR welcomes, 10-15%
with GenAI HR verifies with HR develops
approves posts shortlist, meets oversees 35-45%
stakeholders shortlist based on
priority candidates onboarding
GenAI output 45-50%
Engagement level: Low Medium High
1. Based on BCG 'toil v. joy' diagnostic responses to: How much do you agree with this statement "I enjoy this task"? From (1) - Strongly disagree to (5) - Strongly
agree; Source: BCG workforce diagnostic; BCG 'toil v. joy' diagnostic; BCG experience
9
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
Global airline carrier enhances
Example 2 | HR admin -
service speed and quality with GenAI in shared services
Where did they start? What are they doing? Value identified
Global airline carrier running organization- Conducted holistic assessment across In HR, two priority opportunities
wide improvement program as part of path corporate functions including: • Shared services center (focus of this
to recovery from COVID-19, which significantly • Activity and time allocation example)
impacted flight crew staff: • Potential GenAI impact, including • Recruiting and hiring
productivity, cost, engagement, and skills
Impact on morale disruption Of requests eliminated
55-60% (e.g., due to errors,
Aligned focus opportunities: Finance and HR
missing information)
Prioritized shortlist of workflows based on
High 20-40% Faster response time for
pain points, and assessed data capabilities
demands High remaining requests
for transversal use across organization
on employees attrition
Detailed target state and business cases for < 1 yr Breakeven point
priority opportunities, including cost savings,
Seeking to break this cycle by re-
investment, and skill requirements
establishing employee satisfaction and
operational stability, exploring how GenAI … Plus more efficient, consistent, and
Developed implementation roadmap
can help achieve these strategic goals higher-quality responses
including quick wins to fund the journey and
long-term investments
Source: BCG experience
10
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
Streamlined shared services workflow
Example 2 | HR admin -
enables faster, higher-quality service
Shared services Initiate Validate Gather Create Send Avgerage response
request process requests requests information responses responses time
Employees send From From HR admin
question through original other personalizes
Up to 3
email requester employees template and
manually writes up days
Current
a response, HR admin sends
state HR admin HR admin sometimes response
Ask for
conducts a looks up resulting in
input if
discrepancies
completeness relevant
needed between different
check policies
employees 20-40% reduction
in response time
with current tech
Employee sends Employee interacts
question through with chatbot in real-
GenAI chatbot Chatbot engages time; chatbot redirects HR admin creates
with questioner, to HR admin as response using
Future state HR admin sends
requests missing needed for complex chatbot inputs and
response
with GenAI info or redirects to inquiries checks for ~2 days
correct completeness
HR admin human
department as
validation
needed
(during training only)
1 Source: BCG workforce diagnostic; BCG experience
11
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
To unlock value, HR leaders are investing in foundational enablers
Org design Talent Data, tech, and Risk and Change
and op model and skills partnerships responsible AI management
Restructuring HR, Developing Evolving data to fully Ensuring all Taking a science-
including new roles professional skillsets leverage GenAI and solutions are backed approach to
to address in HR to execute expanding compliant with change behaviors
governance, data against new roles ecosystem of GenAI regulations, and drive adoption,
maintenance, and and requirements partners to assemble especially those that engagement, and
bias tool portfolio are HR-focused adherence
12
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
Today's siloed HR organization
Org design and op model |
will evolve to re-orient around the employee experience
CHRO
Co-leads transformation office to define case for change,
drive organization-wide initiatives, and ensure ROI
North
Star HR
Talent Work HRBP Product HR digital Ethics and
Strategic
ecosystem rhythms owners and data bias
alignment
Unified E2E talent Owners of highly Managerial Specialized, tech- Strategic, data- Governing roles that
around acquisition and personalized, coaches working oriented oriented professionals work across HR,
management behavioral science- shoulder-to- professionals who who link data flows legal, and IT to
employee
professionals informed, "in the shoulder with shape solutions to of business strategy proactively identify
experience, trained in dynamic workflow" L&D business to drive increase to dynamic strategic and mitigate risk,
"segment-of-one" journeys and change, continually engagement and workforce planning oversee compliance,
including
skills analysis human/GenAI redesign work and retention and advise on HR and monitor models/
new roles collaboration team structures for (e.g. well-being, technology, including algorithms for bias
models optimal compensation) data maintenance
that shape
performance
and govern
GenAI
Shared services
Critical forcing mechanism for short-term process/tool standardization, mid-to long-term evolution based on organizational strategy
Existing HR role New HR role Degree of change: Low Med-low Medium High
Source: BCG experience
13
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
Up/re-skilling and new hires will be required
Talent and skills |
to meet the demands of these highly evolved and net new roles
Across three HR roles with the highest business impact,
deep up/re-skilling required amid 55-75% skill disruption …plus hiring of new profiles
Not exhaustive
Extent of skill disruption and example upskilling required for top 3 roles by business impact For new roles to govern,
shape, and build GenAI…
HRBP L&D Recruiter
• RAI/ethics experts
• HR/IT experts
45% 45%
65%
10% 10% For evolved roles requiring
more specialization,
10%
45% 45%
advanced degree holders in
25%
topics including…
• Behavioral science
Example • Interpersonal • Behavioral • Talent sourcing
upskilling communication science • Relationship • Data science
required • Problem solving • L&D strategy building • Programming
High: Upskilling required Moderate: Less important Low: Keep and adapt
Source: BCG analysis LightCast skill taxonomy; BCG experience 13
14
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
For HR organizations lagging in data foundations,
Data, tech, & partnerships |
critical to address upfront and factor into GenAI investment priorities
HR organizations commonly need to Opportunity to invest in data readiness
overcome 3 barriers to prepare for GenAI and GenAI solutions in parallel
Data silos
1 Example GenAI opportunities to pursue in
HR data often housed across different, siloed parallel to HR data preparation:
systems – need to centralize in unified platform
• Recruiting and onboarding content: Writing
job descriptions, marketing emails, personalized
Data quality and inconsistency
2 onboarding; relies on existing job descriptions/
materials
Disparate systems often lead to inconsistent
data entries and formats – critical to standardize • Recruiting admin: Automating scheduling, and
and establish data governance generating reminders and follow-up
communications; does not require employee data
3 Legacy systems • L&D recommendations and content: Suggesting
trainings and developing content; requires basic
HR often using legacy systems that do not
employee data, training records, and existing
integrate with GenAI tools – upgrades required
content that is easier to clean and maintain
to enable data integration and security
15
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
~80-85% of HR organizations are exclusively
Data, tech, & partnerships |
buying or assembling built + bought solutions
Spectrum of partner solutions available based on Five guiding questions for HR
organizational needs and priorities GenAI partnership decisions
Core systems (e.g., CRM) Point solutions Compatibility
How well does the solution integrate with existing
Description Integrated platforms with GenAI in Specialized GenAI-powered tools HR systems and data?
their product roadmaps designed for specific HR activities
Functionality
Can I afford to wait for my core system provider(s)
Advantages • Enterprise-grade technologies • Leveraging pockets of innovation
to add GenAI functionalities? Or do I need to act
• Streamlined integration that are fast to market
now (i.e., build)?
• Able to achieve step changes in • Typically, flexible pricing
capabilities (e.g., via acquisition) • Often faster to implement
Data security/compliance
Does the solution provide adequate protections
Disadvantages • Typically, higher cost and longer • Across E2E HR requirements, may for sensitive HR data? Comply with data privacy
time to implement not be cost-effective
and security regulations?
• Risk of lower quality in some • Risk of data fragmentation
capabilities vs. point solutions • Risk of partner being acquired by
Scalability
• May "lock" into one partnership lean core players
Can the solution scale with our evolving needs?
Examples • Human capital management • Skills-based talent development Cost/ROI
platform tools What are the initial and ongoing costs, and
expected ROI?
Source: BCG survey of CHROs or direct reports to CHRO (n=64); BCG experience 15
16
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
Many emerging GenAI regulations explicitly
Risk and responsible AI |
address HR given the high sensitivity around its data and responsibilities
Not exhaustive
July 2023 May 2024 May 2024 Expected 2024
Geo
NYC AI Regulations EU Artificial Colorado Artificial EU, US,
Regulation (Local Law 144 Intelligence Act3 Intelligence Act Cross-border
of 2021)1 (CAIA)4
• Regulates how GenAI is • Given that GenAI tech for HR orgs will be required to: • More regulations are
used in hiring and HR falls into highest risk expected targeting HR
• Enact and report on risk-
promotion decisions category, HR orgs will • HR must monitor ongoing
management policy to • HR must conduct annual need to meet stringent commissions & regulations on
Potential
govern GenAI tool use
audit for potential bias requirements (e.g., data the horizon including
implications
• Conduct annual impact
in automated employment governance, transparent No Robot Bosses Act
for HR
assessments to identify
decision tools and provide candidate comms, (prohibits sole reliance on
algorithmic bias
subsequent notice2 compliance obligations, automated hiring decisions),
• Alert candidates of AI
incident reporting) AI Pact (network to support
use in hiring decisions
organizations’ AI compliance)
Disclaimer: For informational purposes only - does not constitute legal advice; 1. In effect as of Jan 1, 2023; 2. Defined as computer-based tools that use AI, machine learning, statistical modeling, or
data analytics to help employers make employment decisions; 3. Passed March, 2024, going into effect Summer 2024; 4. Signed May 2024, going into effect February 2026;
Source: City of New York; State of Colorado; European Commission; U.S. Congress; BCG analysis
17
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
…Underscoring the importance of HR compliance
and proactive risk mitigation
Not exhaustive
Change Governance Data security Regulatory Candidate
management vs. speed and model bias conformity experience
Addressing resistance Balancing compliance Avoiding exposure of Setting up to comply Minimizing
HR must to changes in ways with momentum sensitive data and with new HR-focused performance issues
proactively of working biased outputs laws/regulations during rollout
anticipate
many
• Upskill early and • Provide clear • Create new role(s) • Ensure proactive • Start small with
potential continuously, starting decision-making responsible for monitoring of GenAI- pilots to minimize the
with leadership model on E2E work systematic model specific regulations to impact of the first risks,
re-design while oversight, including inform partnerships, rollout
• Build personalized
including… anticipating tradeoffs bias prevention models in use, and change journeys with • Prioritize
solution design
tailored behavioral • Establish clear, • Prioritize red- transparency where
science interventions streamlined teaming and other • Develop ongoing possible during pilot
governance processes stress tests to regulatory training and rollout, fostering
• Identify and measure
with a focus on actively address risk to ensure leadership spirit of co-creation
adoption and value to
prioritizing security, and workforce are and ongoing
track progress, inform • Consider dedicated
action, and informed communication
solution iteration HR-IT roles for data
productivity gains
oversight
Additional detail follows
Source: BCG experience
18
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
To ensure adoption, engagement, and adherence,
Change management |
a proactive change program grounded in behavioral science is required
Four levers to catalyze the flywheel of behavioral change
Personalized change journeys Closed feedback loops
2x adoption rate with personalized Always-on change monitoring, e.g., to
user journeys that leverage inform feedback loops and respond to
interventions based on behavioral 90% of employees seeking regular
insights, pain points, and psychological leadership communication
Accuracy Trust
traits
GenAI
Nudges Co-creation
adoption
Subtly guiding behavior without For example, collaborating with
forbidding options or significantly recruiters to re-design talent
changing economic incentives; e.g., acquisition processes and develop
changing default options, highlighting supporting set of GenAI solutions
Usage
peer benchmarks to boost desired
behavior 55-60%
Source: Behavioral Science Lab; BCG workforce diagnostic; BCG experience
Three steps for HR to begin the GenAI journey
Build an integrated GenAI roadmap, grounded in
strategic HR goals, and collaborate with leadership to
chart the enterprise-wide transformation journey
Upskill leaders in HR and across the organization,
including ongoing hands-on experimentation (e.g., everyday
tools, custom GPTs, and agents)
Prepare data (e.g., mapping data sources, implementing
centralized HRIS platform, standardizing formats) and
develop guardrails to ensure reliability and compliance
19
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
20
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
4202
©
thgirypoC
Key contacts for HR AI transformation
BCG Experts |
NAMR
Allison David Julie Renee Dylan Bill
Bailey Martin Bedard Laverdiere Bolden Beaver
Frank Tauseef Julia Kristy Sesh Matthew
Breitling Charanya Dhar Ellmer Iyer Kropp
Juliana Vladimir Rajiv Nithya Charles
Lisi Lukic Shenoy Vaduganathan Westrin
EMESA
Vinciane Jens Jaap Nicolas Nina Erik
Beauchene Baier Backx de Bellefonds Kataeva Lenhard
Julia Tom Dan Ben Marc Sebastian
Madden Martin Sack Shuttleworth Schuuring Ullrich
APAC
Ashish Sreyssha Sagar Chris Fang Jeffrey
Garg George Goel Mattey Ruan Walters |
278 | bcg | ai-radar-2025-slideshow-jan-2025-r.pdf | BCG AI RADAR
From Potential to Profit:
Closing the AI Impact Gap
JANUARY 2025
SURVEY METHODOLOGY
Global research of 1,803 C-level executives on AI in 2025
Executive roles Company revenue Markets Industries and key functions
US 214
TMT 323
14% 14% India 200
Financial institutions 256
Germany 198
5%
UK 182
Consumer/retail 250
France 171
31%
Italy 102 Healthcare/medical 192
CEO
30%
Singapore 101
CSO
<$500M Transportation/travel/tourism 145
Brazil 87
CxO1
17% $500M–<$1B
Japan 82
CTO Energy 144
$1B–<$2B
CIO UAE 81
8% $2B–<$5B
Manufacturing 142
CDO2 Spain 79
>$5B
16%
9% C-suite3 Nigeria 65
Industrial goods 110
Indonesia 64
7%
Australia 46 Public sector 110
Saudi Arabia 45
Insurance 49
22% Greater China 38
27%
South Africa 28 Marketing 41
Morocco 12
Real estate 41
Qatar 8
Source: BCG AI Radar 2025 Survey
Note: Revenue thresholds for survey inclusion: $500+ million USD (US, Europe, Japan, Australia); $100+ million USD (rest of APAC, Middle East, Africa). Survey conducted September to December 2024.
1CxO represents other executives who directly report to the CEO (e.g., CMO, CFO, COO, etc.). 2Includes CDO and CAIO. 3C-suite includes EVP, SVP, VP, Chairman, President.
Where is the value in AI?
GenAI investments are projected to increase by
60% in the next 3 years
+60%
AI ambitions are
growing alongside +30%
investments
2023 2024 2027
Source: BCG IT Spend Survey 2024.
One in three companies across all markets are planning to spend
$25 million+ on AI in 2025
How much are you planning to invest in AI in 2025?
Japan 53% 26% 10% 11%
US 59% 23% 9% 9%
Singapore 63% 25% 6% 6%
UK 65% 18% 10% 7%
France 69% 17% 6% 8%
Up to $25M
Germany 69% 18% 9% 4% $26M–$50M
$51M–$100M
India 71% 15% 8% 6%
>$100M
UAE 78% 15% 6% 1%
Spain 81% 6% 5% 8%
Italy 83% 12% 3% 2%
Brazil 86% 12% 2%
Global 69% 18% 7% 6% One in three companies
Source: BCG AI Radar 2025 Survey (n=1,803).
… but, only
75% 25%
of executives rank AI/GenAI as a top three strategic priority … of executives are
seeing significant
value from AI
Source: BCG Radar 2025 Survery (n=1,803).
What are leading companies doing differently?
Deploy AI in everyday tasks
to realize 10% to 20%
productivity potential
Three value plays to
Reshape critical functions for
30% to 50% enhancement in
maximize AI potential
efficiency and effectiveness
Invent new products and
services to build long-term
competitive advantage
Source: BCG analysis.
They focus +80% of their AI investments in reshaping
critical functions and inventing new products and services
Leading
18%
Deploy
companies go
Individual-productivity focused
well beyond 40%
Reshape
Process-level productivity aimed
deploy …
at reshaping critical functions
Invent
Company-level innovation
42%
core to the business
+80%
Source: BCG Build for the Future 2024 Global Study (merged with Digital
Acceleration Index), (n=1,000).
Share of AI investments in Deploy, Reshape,
and Invent initiatives
… but most
companies aren’t
yet prioritizing 27% Deploy
Individual-productivity focused
investments in
44%
Reshape
higher-impact Process-level productivity aimed
at reshaping critical functions
56%
plays
Invent
Company-level innovation
core to the business
29%
Source: BCG AI Radar 2025 Survey (n=1,803).
Leading companies extract greater value by
focusing their AI investments
40%
In reality, most
6.1
companies go broad
2.1x and dilute efforts
3.5
across multiple
pilots, seeing lower
ROI as a result
AI use cases More ROI for AI
prioritized anticipated
Leading companies
Other companies
Source: BCG Build for the Future 2024 Global Study (merged with Digital Acceleration Index), (n=1,000).
Yet 60% of companies are failing to define and monitor any
Leading
financial KPIs related to AI value creation
companies set
How is your organization tracking value creation from AI?
clear goals and
track top- and
32% 28% 16% 24%
bottom-line
impact
60%
Not tracking yet Operational only Financial only Operational and financial
Source: BCG AI Radar 2025 Survey (n=1,803).
Leaders follow the 10-20-70 principle to create value
10%
… but
Algorithms
2 in 3
20%
Technology
companies struggle to:
· Reimagine workflows and drive
incentives, culture, and change
70%
People and processes
· Hire AI talent and upskill workforce
Source: BCG AI Radar 2025 Survey (n=1,803).
Note: AI talent refers to AI specialists (i.e., data scientists, ML ops engineers) and non–specialists (i.e., upskilled talent leveraging AI tools).
Data privacy and security 66%
AI risks that
companies must
Lack of control or understanding
48%
of AI decisions
navigate
Regulatory challenges
44%
and compliance
Source: BCG AI Radar 2025 Survey (n=1,803).
Note: Percentages correspond to share of executives ranking risk in their top 3.
76%
Cybersecurity
remains critical
Recognize that their AI cybersecurity
measures need further improvements
Source: BCG AI Radar 2025 Survey (n=1,803).
2025: The year of AI agents?
What an agent is
What is an
agent?
Memory Reasoning Systems
Remembering across tasks Decomposing a problem Accessing external
Simply put,
and changing states and planning actions systems on your behalf
it’s an AI that
What an agent does
has learned to
use tools
Observes Plans Acts
Collect and process data Evaluate possible actions Execute by leveraging internal
from environment and prioritize toward a goal or external tools/systems
Source: BCG Analysis
67% However, agents require
deep reimagination of
work and are not a silver
bullet for impact
15%
are considering autonomous agents as
part of their AI transformation
Source: BCG AI Radar 2025 Survey (n=1,803).
Optimism around AI agents is consistent across geographies
Role that companies see for AI agents moving into 2025
74%
US 37% 37%
72%
Japan 26% 46%
UAE 32% 40% 72%
Singapore 31% 40% 71%
70%
India 34% 36%
68%
UK 27% 41%
France 34% 33% 67%
63%
Germany 30% 33%
Central or
complementary role
Spain 38% 24% 62%
Exploring
Brazil 31% 30% 61%
Italy 18% 38% 56%
Global 32% 35% 67%
Source: BCG AI Radar 2025 Survey (n=1,803).
Unlocking new potential to reshape processes and services
Agents deliver up to 3x more productivity and speed benefits compared to
traditional assistants
Breaking down silos
The biggest opportunity is seamless enterprise collaboration through zero-touch
services, advanced planning, and automated customer 360 activation
AI agents:
Key leadership
Managing the risks of more complexity
AI agents are more complex than assistants, requiring robust testing and
priorities optimization to manage operational and cyber risks effectively
Cutting through the hype
Mislabeling and overhype of “agents” will dilute trust and lead to unmet
expectations; leaders must clarify capabilities and set realistic goals
Moving forward
Success lies in targeted, high-impact applications, focused on practical design,
Source: BCG analysis.
seamless integration, and data quality over hype
With AI agents on the rise, who will hold
the power: humans or AI?
Executives see talent and AI as complementary
Human-centered AI-focused
14% 64% 22%
Prioritizing AI and humans working AI taking the lead,
human talent, side by side but humans retaining
using AI only oversight
when necessary
Source: BCG AI Radar 2025 Survey (n=1,803).
How do you expect the workforce in your organization to change?
Increase headcount,
More FTEs 8%
adding new skills
Less than 10% of
executives expect
a decrease in More productivity and
Existing workforce 68% upskilling of existing talent
to meet AI needs
headcount due
to AI automation
Restructure workforce
with new roles to replace
Net neutral 17%
redundant ones
Decrease headcount
Fewer FTEs 7%
due to AI automation
Source: BCG AI Radar 2025 (n=1,803).
~70% of the companies have trained
less than 1 in 4 of their workforce
AI upskilling is
accelerating,
but the work is
71%
94%
not over
Companies with more
than 25% of their workforce
trained on AI/GenAI tools
29%
6%
2023 2024
Source: BCG AI Radar 2025 Survey (n=1,803).
Companies with more than one-quarter of their workforce trained
on AI/GenAI tools
Singapore 44%
Singapore and
Japan 38%
Japan lead in
Germany 30%
AI/GenAI
Spain 29%
upskilling; Brazil France 29%
UK 29%
and Italy are
US 29%
falling behind
UAE 27%
India 26%
Italy 20%
29%
Brazil 20%
globally
Source: BCG AI Radar 2025 Survey (n=1,803).
Strategic playbook for CEOs
Breaking through AI’s imagination gap
Rethink what is possible with AI and business transformation
Targeting and prioritizing AI efforts
Focus on a few transformative opportunities in core functions
Leading companies
maximize value
Putting AI at the service of enterprise ambition
Define and track clear KPIs
creation by...
Leading the cultural and organizational change
Lean in personally and drive the change
Preparing for what’s next
Anticipate AI’s next value play and accompanying risks
Source: BCG analysis. |
279 | bcg | 24-013_d9b45b68-9e74-42d6-a1c6-c72fb70c7282.pdf | Working Paper 24-013
Navigating the Jagged Technological
Frontier: Field Experimental Evidence
of the Effects of AI on Knowledge
Worker Productivity and Quality
Fabrizio Dell'Acqua Saran Rajendran
Edward McFowland III Lisa Krayer
Ethan Mollick François Candelon
Hila Lifshitz-Assaf Karim R. Lakhani
Katherine C. Kellogg
Navigating the Jagged Technological
Frontier: Field Experimental Evidence
of the Effects of AI on Knowledge
Worker Productivity and Quality
Fabrizio Dell'Acqua Saran Rajendran
Harvard Business School Boston Consulting Group
Edward McFowland III Lisa Krayer
Harvard Business School Boston Consulting Group
Ethan Mollick François Candelon
The Wharton School Boston Consulting Group
Hila Lifshitz-Assaf Karim R. Lakhani
Warwick Business School Harvard Business School
Katherine C. Kellogg
MIT Sloan School of Management
Working Paper 24-013
Copyright © 2023 by Fabrizio Dell’Acqua, Edward McFowland III, Ethan Mollick, Hila Lifshitz-Assaf, Katherine
C. Kellogg, Saran Rajendran, Lisa Krayer, François Candelon, and Karim R. Lakhani.
Working papers are in draft form. This working paper is distributed for purposes of comment and discussion only.
It may not be reproduced without permission of the copyright holder. Copies of working papers are available from
the author.
We thank Michael Bervell, John Cheng, Pallavi Deshpande, Maxim Ledovskiy, John Kalil, Kelly Kung, Rick
Lacerda, MarcAntonio Awada, Paula Marin Sariego, Rafael Noriega, Alejandro Ortega, Rahul Phanse, Quoc-Anh
Nguyen, Nitya Rajgopal, Ogbemi Rewane, Kyle Schirmann, Andrew Seo, Tanay Tiwari, Elliot Tobin, Lebo
Nthoiwa, Patrick Healy, Saud Almutairi, Steven Randazzo, Anahita Sahu, Aaron Zheng, and Yogesh Kumaar for
helpful research assistance. We thank Kevin Dai for outstanding support with data and visualizations. For helpful
feedback, we thank Maxime Courtaux, Clement Dumas, Gaurav Jha, Jesse Li, Max Männig, Michael Menietti,
Rachel Mural, Zahra Rasouli, Esther Yoon, Leonid Zhukov, and David Zuluaga Martínez. Lakhani would like to
thank Martha Wells, Anne Leckie, Iain Banks, and Alastair Reynolds for inspiring AI futures. We used Poe,
Claude, and ChatGPT for light copyediting and graphics creations. Lakhani is an advisor to Boston Consulting
Group on AI Strategy and learning engagement.
Funding for this research was provided in part by Harvard Business School.
Navigating the Jagged Technological Frontier:
Field Experimental Evidence of the Effects of AI on
Knowledge Worker Productivity and Quality*
Fabrizio Dell’Acqua1, Edward McFowland III1, Ethan Mollick2, Hila
Lifshitz-Assaf1,3, Katherine C. Kellogg4, Saran Rajendran5, Lisa Krayer5,
François Candelon5, and Karim R. Lakhani1
1Digital Data Design Institute, Harvard Business School; 2The Wharton
School, University of Pennsylvania; 3Warwick Business School, Artificial
Intelligence Innovation Network; 4MIT Sloan School of Management;
5Boston Consulting Group, Henderson Institute
September 22, 2023
*Fabrizio Dell’Acqua ([email protected]), Edward McFowland III ([email protected]),
Ethan Mollick ([email protected]), Hila Lifshitz-Assaf ([email protected]),
Katherine C. Kellogg ([email protected]), Saran Rajendran ([email protected]), Lisa
Krayer ([email protected]), François Candelon ([email protected]), Karim R. Lakhani
([email protected]). We thank Michael Bervell, John Cheng, Pallavi Deshpande, Maxim Ledovskiy, John
Kalil, Kelly Kung, Rick Lacerda, MarcAntonio Awada, Paula Marin Sariego, Rafael Noriega, Alejandro
Ortega,RahulPhanse,Quoc-AnhNguyen,NityaRajgopal,OgbemiRewane,KyleSchirmann,AndrewSeo,
TanayTiwari,ElliotTobin,LeboNthoiwa,PatrickHealy,SaudAlmutairi,StevenRandazzo,AnahitaSahu,
Aaron Zheng, and Yogesh Kumaar for helpful research assistance. We thank Kevin Dai for outstanding
supportwithdataandvisualizations. Forhelpfulfeedback,wethankMaximeCourtaux,ClementDumas,
Gaurav Jha, Jesse Li, Max Männig, Michael Menietti, Rachel Mural, Zahra Rasouli, Esther Yoon, Leonid
Zhukov, and David Zuluaga Martínez. Lakhani would like to thank Martha Wells, Anne Leckie, Iain
Banks, and Alastair Reynolds for inspiring AI futures. We used Poe, Claude, and ChatGPT for light
copyeditingandgraphicscreations. LakhaniisanadvisortoBostonConsultingGrouponAIStrategyand
learningengagement. Allerrorsareourown.
1
Abstract
The public release of Large Language Models (LLMs) has sparked tremendous
interestinhowhumanswilluseArtificialIntelligence(AI)toaccomplishavarietyof
tasks. In our study conducted with Boston Consulting Group, a global management
consulting firm, we examine the performance implications of AI on realistic,
complex, and knowledge-intensive tasks. The pre-registered experiment involved
758 consultants comprising about 7% of the individual contributor-level consultants
at the company. After establishing a performance baseline on a similar task, subjects
were randomly assigned to one of three conditions: no AI access, GPT-4 AI access,
or GPT-4 AI access with a prompt engineering overview. We suggest that the
capabilities of AI create a “jagged technological frontier” where some tasks are
easily done by AI, while others, though seemingly similar in difficulty level, are
outside the current capability of AI. For each one of a set of 18 realistic consulting
tasks within the frontier of AI capabilities, consultants using AI were significantly
moreproductive(theycompleted12.2%moretasksonaverage,andcompletedtasks
25.1% more quickly), and produced significantly higher quality results (more than
40% higher quality compared to a control group). Consultants across the skills
distribution benefited significantly from having AI augmentation, with those below
theaverageperformancethreshold increasingby43%andthose above increasingby
17% compared to their own scores. For a task selected to be outside the frontier,
however,consultantsusingAIwere19percentagepointslesslikelytoproducecorrect
solutionscomparedtothosewithoutAI.Further,ouranalysisshowstheemergenceof
twodistinctivepatternsofsuccessfulAIusebyhumansalongaspectrumofhuman-
AI integration. One set of consultants acted as “Centaurs,” like the mythical half-
horse/half-humancreature,dividinganddelegatingtheirsolution-creationactivities
to the AI or to themselves. Another set of consultants acted more like “Cyborgs,”
completely integrating their task flow with the AI and continually interacting with
thetechnology.
2
1 Introduction
The capabilities of Artificial Intelligence to produce human-like work have improved
rapidly,especiallysincethereleaseofOpenAI’sChatGPT,oneofseveralLargeLanguage
Models (LLMs) that are widely available for public use. As AI capabilities overlap more
with those of humans, the integration of human work with AI poses new fundamental
challenges and opportunities, in particular in knowledge-intensive domains. In this
paper, we examine this issue using randomized controlled field experiments with highly
skilled professional workers. Our results demonstrate that AI capabilities cover an
expanding, but uneven, set of knowledge work we call a "jagged technological frontier.”
Within this growing frontier, AI can complement or even displace human work; outside
of the frontier, AI output is inaccurate, less useful, and degrades human performance.
However, because the capabilities of AI are rapidly evolving and poorly understood,
it can be hard for professionals to grasp exactly what the boundary of this frontier
might be at a given. We find that professionals who skillfully navigate this frontier gain
large productivity benefits when working with the AI, while AI can actually decrease
performancewhenusedforworkoutsideofthefrontier.
Though LLMs are new, the impact of other, earlier forms of AI have been the subject
ofconsiderablescholarlydiscussion(e.g.,Brynjolfssonetal.(2018);FurmanandSeamans
(2019);Puranam(2021)). BecauseofthelimitationsoftheseearlierformsofAI,nonroutine
tasks that were difficult to codify seemed protected from automation (Autor et al., 2003;
Acemoglu and Restrepo, 2019), especially as previous waves of technology had mostly
automatedlower-skilledoccupations(GoldinandKatz,1998). ThereleaseofChatGPTin
November, 2022 changed both the nature and urgency of the discussion. LLMs proved
unexpectedly capable at creative, analytical, and writing tasks, including scoring at top
levelsatgraduateandprofessionalexaminations(Girotraetal.,2023;Geerlingetal.,2023;
Kung et al., 2023; Boussioux et al., 2023). This represented an entirely new category of
automation, one whose abilities overlapped with the most creative, most educated, and
mosthighlypaidworkers(Eloundouetal.,2023).
Studies of previous generations of AI (Brynjolfsson et al., 2023) and controlled
1
experiments on the impact of recently released LLMs (e.g., Noy and Zhang (2023); Choi
and Schwarcz (2023)) suggest that these systems can have a large impact on worker
performance. In our study, we focus on complex tasks, selected by industry experts to
replicate real-world workflows as experienced by knowledge workers. Most knowledge
work includes this sort of flow, a set of interdependent tasks, some of which may be
good fit for current AI, while some are not. We examine both kinds of tasks, and build
on recent studies to suggest ways of understanding the rapidly evolving impact of AI on
knowledgeworkers,underwhichcircumstancesorganizationsmaybenefit,andhowthis
mightchangeasthetechnologyadvances.
This is important because understanding the implications of LLMs for the work
of organizations and individuals has taken on urgency among scholars, workers,
companies, and even governments (Agrawal et al., 2018; Iansiti and Lakhani, 2020; Berg
etal.,2023). PreviousformsofAIledtoconsiderabledebateintheliteraturearoundhow
and whether professionals should adopt AI for knowledge work (Anthony et al., 2023)
and the potential impact this might have on organizations (Raisch and Krakowski, 2021;
Glaeser et al., 2021; Brynjolfsson et al., 2021). Some scholars focused on the potential
for AI to help professionals improve their effectiveness and efficiency (DeStefano et al.,
2022;Balakrishnanetal.,2022;ValentineandHinds,2023). Othersdemonstratedthat,for
critical tasks, it can be risky for professionals to use AI (Lebovitz et al., 2021), especially
black-boxed (e.g., Lebovitz et al. (2022); Waardenburg et al. (2022)), and showed how
professionals are struggling to use it effectively (Pachidi et al., 2021; Van den Broek et al.,
2021). Finally, another group of researchers argued that the “algorithmic management”
affordedbyAIcancreatenegativepersonalimpactsforprofessionals(Kelloggetal.,2020;
Möhlmann et al., 2021; Tong et al., 2021) and raise accountability and ethical questions
(Choudhuryetal.,2020;Cowgilletal.,2020;Rahmanetal.,2024). Yet,mostofthestudies
predate ChatGPT, and investigate forms of AI designed to produce discrete predictions
basedonpastdata. ThesesystemsarequitedifferentfromLLMs.
Specifically, outside of their technical differences from previous forms of machine
learning, there are three aspects of LLMs that suggest they will have a much more
rapid, and widespread, impact on work. The first is that LLMs have surprising
2
capabilities that they were not specifically created to have, and ones that are growing
rapidly over time as model size and quality improve. Trained as general models,
LLMsnonethelessdemonstratespecialistknowledgeandabilitiesaspartoftheirtraining
processandduringnormaluse(Singhaletal.,2022;Boikoetal.,2023). Whileconsiderable
debate remains on the concept of emergent capabilities from a technological perspective
(Schaeffer et al., 2023), the effective capabilities of AIs are novel and unexpected, widely
applicable, and are increasing greatly in short time spans. Recent work has shown
that AI performs at a high level in professional contexts ranging from medicine to law
(Ali et al., 2023; Lee et al., 2023), and beats humans on many measures of innovation
(Boussioux et al., 2023; Girotra et al., 2023). And, while score performance on various
standardized academic tests is an imperfect measure of LLM capabilities, it has been
increasingsubstantiallywitheachgenerationofAImodels(OpenAI,2023).
The general ability of LLMs to solve domain-specific problems leads to the second
differentiating factor of LLMs compared to previous approaches to AI: their ability to
directly increase the performance of workers who use these systems, without the need
for substantial organizational or technological investment. Early studies of the new
generation of LLMs suggest direct performance increases from using AI, especially for
writing tasks (Noy and Zhang, 2023) and programming (Peng et al., 2023), as well as
for ideation and creative work (Boussioux et al., 2023; Girotra et al., 2023). As a result,
the effects of AI are expected to be higher on the most creative, highly paid, and highly
educatedworkers(Eloundouetal.,2023;Feltenetal.,2023)
The final relevant characteristic of LLMs is their relative opacity. This extends to the
failurepointsofAImodels,whichincludeatendencytoproduceincorrect,butplausible,
results (hallucinations or confabulations), and to make other types of errors, including in
mathandwhenprovidingcitations. TheadvantagesofAI,whilesubstantial,aresimilarly
unclear to users. It performs well at some jobs, and fails in other circumstances in ways
thataredifficulttopredictinadvance. Contributingfurthertotheopacityisthatthebest
ways to use these AI systems are not provided by their developers and appear to be best
learnedviaongoingusertrial-and-errorandthesharingofexperiencesandheuristicsvia
variousonlineforumslikeusergroups,hackathons,TwitterfeedsandYouTubechannels.
3
Taken together, these three factors – the surprising abilities of LLMs, their ability to
do real work with virtually no technical skill required of the user, and their opacity and
unclear failure points – suggest that the value and downsides of AI may be difficult for
workers and organizations to grasp. Some unexpected tasks (like idea generation) are
easy for AIs, while other tasks that seem to be easy for machines to do (like basic math)
are challenges for some LLMs. This creates a “jagged Frontier,” where tasks that appear
to be of similar difficulty may either be performed better or worse by humans using
AI. Due to the “jagged” nature of the frontier, the same knowledge workflow of tasks
can have tasks on both sides of the frontier, see Figure 1. The future of understanding
how AI impacts work involves understanding how human interaction with AI changes
depending on where tasks are placed on this frontier, and how the frontier will change
over time. Investigating how humans navigate this jagged frontier, and the subsequent
performanceimplications,isthefocusofourwork.
Wecollaboratedwithaglobalmanagementconsultingfirm(BostonConsultingGroup
- BCG) and advised them on designing, developing, and executing two pre-registered
randomized experiments to assess the impact of AI on high humancapital professionals.
Subsequently, the author team received the data that the company collected for the
purpose of this experiment and conducted the analysis presented in this paper. The
studywasstructuredinthreephases: aninitialdemographicandpsychologicalprofiling,
a main experimental phase involving multiple task completions, and a concluding
interview segment. We tested two distinct tasks: one situated outside the frontier of
AI capabilities and the other within its bounds. The experiment aimed to understand
how AI integration might reshape the traditional workflows of these high human capital
professionals.
OurresultsshowthatthisgenerationofLLMsarehighlycapableofcausingsignificant
increases in quality and productivity, or even completely automating some tasks, but the
actual tasks that AI can do are surprising and not immediately obvious to individuals or
even to producers of LLMs themselves. Because this frontier is expanding and changing,
the overall results suggest that AI will have a large impact on work, one which will
increasewithLLMcapabilities,butwheretheimpactsoccurwillbeuneven.
4
2 Methods
We collected data from two randomized experiments to assess the causal impact of AI,
specifically GPT-4 – the most capable of the AI models at the time of the experiments
(Spring 2023) – on high human capital professionals working traditionally without AI.1
We pre-registered our study, detailing the design structure, the experimental conditions,
thedependentvariables,andourmainanalyticalapproaches.2 Ouraimwastodetermine
how introducing this AI into the tasks of highly-skilled knowledge workers might
augment,disrupt,orinfluencetheirtraditionalworkflows.
BCG individual contributor consultants around the world were offered the
opportunity to spend 5 hours working on this experiment to evaluate the impact of AI
ontheiractivities. Approximately7%ofBCG’sglobalindividualcontributorconsultants’
cohortengagedinandcompletedtheexperiment.
Theexperimentwasstructuredintothreedistinctphases. Initially,consultantsentered
the study by completing a survey that captured their demographic and psychological
profiles, as well as details about their role within the company. A few weeks after
enrolling, participants received a link to complete the main experimental phase. This
phase commenced with a pre-task survey, followed by the tasks detailed subsequently,
and concluded with a post-task survey. In the final phase, participants were interviewed
tosharetheirexperiencesandperspectivesontheroleofAIintheirprofession.
In the first phase, we administered an enrollment survey to gather information about
potential participants.3 This survey captured details such as office location, internal
affiliation, and tenure at BCG. Participants also completed psychological assessments,
specifically providing insights into their Big 5 personality traits (Soto and John, 2017),
innovativeness (Agarwal and Prasad, 1998), self-perceived creativity (Miron-Spektor
et al., 2004), and paradox mindset (Miron-Spektor et al., 2018). Furthermore, the
survey included a short section on their reading habits (including their familiarity with
1TheprojecthasreceivedIRBapproval,IRB23-0392.
2Pre-registrationcompletedonOpenScienceFoundation,osf.io/ytaev. Thepre-registrationisavailable
fromtheauthorsuponrequestandwillbemadepubliclyavailableafterarticleacceptanceoraftertheOSF
embargoperiodhaspassed,whichevercomesfirst
3Outofthe852consultantswhorespondedtothesurvey,758-about89%-completedtheexperiment.
5
AI characters in fiction), and demographic details like gender, native language, and
educational background. We utilized these data for stratified random assignment and
ascontrolsinourregressionmodels,asdescribedbelow.
The study encompassed 758 strategy consultants, each of whom completed the initial
survey and experimental tasks. Each participant was assigned to one of two distinct
experiments. Stratification of participants was based on multiple criteria, both between
experimentsandacrossexperimentaltreatments. Thesecriteriaincludedgender,location,
tenure at BCG, individual openness to innovation, and native English-speaking status.
This information was collected with the survey administered during phase one, a few
weeksbeforethemainexperiment.
In order to ensure genuine engagement and effort from participants, we incentivized
their performance in the experiment. Participants who diligently participated in all
aspects of the experiment were honored with an "office contribution" recognition,
carryingfinancialimplicationsrelatedtotheirannualbonuses. Furthermore,torecognize
and reward excellence, the top 20% of participants received additional recognition, and
the top 5% was also awarded with a small gift. Executives at BCG reported that the
recognition received by top participants was important because it was shared with the
committeethatoverseestheircareerdevelopmentandperformanceassessments.
Subjects were allocated to one of two distinct experiments, each involving a
unique type of task, with no overlap between the groups. Both tasks were designed
in collaboration with multiple people at BCG to represent some of the typical job
activities encountered by individual contributor consultants. Approximately half of the
participants (385 consultants) tackled a series of tasks where they were prompted to
conceptualize and develop new product ideas, focusing on aspects such as creativity,
analytical skills, persuasiveness and writing skills. The other half (373 consultants)
engaged in business problem-solving tasks using quantitative data, customer and
company interviews, and including a persuasive writing component. Both sets of tasks
were developed to be realistic, and were designed with the input of professionals in
the respective sectors. A senior level executive at the company commented on these
tasks being “very much in line with part of the daily activities” of the subjects involved.
6
Notably,someformsofthesetasksarealsousedbythecompanytoscreenjobapplicants,
typically from elite academic backgrounds (including Ph.D.s), for their highly-coveted
positions.
Both experiments followed a consistent structure. Initially, participants undertook a
task without the aid of AI, establishing a baseline for performance and enabling within-
subject analyses. Following this, participants were randomly assigned to one of three
conditions to assess the influence of AI on their tasks, with these conditions being
consistent across both experiments. The first group (a control condition) proceeded
without AI support; the second (“GPT Only”) had the assistance of an AI tool based
on GPT-4; and the third (“GPT + Overview”) not only utilized the same AI tool but
also benefited from supplementary prompt engineering overview, which increased their
familiarity with AI. These materials included instructional videos and documents that
outlinedandillustratedeffectiveusagestrategies.
Ratherthanrelyingonself-reportedmetricsorindirectindicators,wedirectlyassessed
participants’ skills through a task that closely mirrored the main experiment. In both
experiments, we employ an assessment task that, while different from the experimental
task, is highly comparable, ensuring a precise evaluation of skills for this specific task
type.4 Our findings indicate that performance in the assessment task is a predictor of
performance in the experimental task, allowing us to study the differential effects of
introducingAItoparticipantsofdifferentskilllevels.
Each task assigned to participants came with a specific time allocation. In the
experiment using a task inside the frontier, the assessment task duration was set for 30
minutes, while the subsequent one was allotted 90 minutes. Conversely, in the outside-
the-frontierexperiment,boththefirstandsecondtasksweredesignated60minuteseach,
though participants could complete them earlier if they finished ahead of time. It is
importanttonotethatforthetaskinsidethefrontier,participantswererequiredtoremain
on the task’s page for the entire duration of the task, and could not complete the exercise
earlier. This approach ensured that our analysis for the inside-the-frontier tasks focused
4Dell’Acqua et al. (2023) adopts a comparable experimental framework to evaluate subjects’
competencies.
7
chiefly on the qualitative differences, rather than any timing improvements brought
aboutbyusingAI.Thesetimeframeswereautomaticallyenforced,withtheexperimental
systemadvancingtothenextquestiononcethestipulatedtimeforataskelapsed.
In every experimental task, subjects assigned to the AI conditions had access to
a company platform. This platform, developed using the OpenAI API, facilitated an
interactive experience with OpenAI’s GPT-4, mirroring the dynamics of ChatGPT. It
enabled the collection of all participants’ prompts and AI’s corresponding responses,
providing a comprehensive view into the collaborative behaviors between subjects and
AI.Allsubjectsusedthesameversionofthetool,accessingGPT-4asavailableattheend
ofApril,2023,andusingdefaultsystempromptsandtemperature.
Asidefromthethematicdifferences,thetasksdifferedinanotherkeyway. Whileboth
were designed to be comparably complex and realistic, the first task was selected to be
withinthepotentialtechnologicalfrontierofGPT-4. Thesecondexperimentwasdesigned
sothatGPT-4wouldmakeanerrorwhenconductingtheanalysis,ensuringtheworkfell
justoutsidethefrontier.
3 Results
3.1 Quality and Productivity Booster - Inside the Frontier
The inside-the-frontier experiment focused on creative product innovation and
development. The initial assessment task asked participants to brainstorm innovative
beverage concepts. From their set of ideas, they identified the most viable option and
devisedacomprehensiveplanforitsmarketdebut. Afterthistask,subjectsmovedtothe
mainexperimentalphaseandthecontexttransitionedtothemainexperimentaltask.
In this experimental task, participants were tasked with conceptualizing a footwear
idea for niche markets and delineating every step involved, from prototype description
to market segmentation to entering the market. An executive from a leading global
footwearcompanyverifiedthatthetaskdesigncoveredtheentireprocesstheircompany
8
typically goes through, from ideation to product launch.5 Participants responded to
a total of 18 tasks (or as many as they could within the given time frame). These
tasks spanned various domains. Specifically, they can be categorized into four types:
creativity (e.g., “Propose at least 10 ideas for a new shoe targeting an underserved
market or sport.”), analytical thinking (e.g., “Segment the footwear industry market
based on users.”), writing proficiency (e.g., “Draft a press release marketing copy for
your product.”), and persuasiveness (e.g., “Pen an inspirational memo to employees
detailing why your product would outshine competitors.”). This allowed us to collect
comprehensiveassessmentsofquality. AlltasksanddetailsarereportedinAppendixA.
In the experiment, the primary outcome variable is the quality of the subjects’
responses. To quantify this quality, we employed a set of human graders to evaluate
each question that participants didn’t leave unanswered.6 Each response was evaluated
by two human graders. We then calculated the mean grade assigned by humans to each
question. This gave us 18 dependent variables (one per each question). We subsequently
averaged these scores across all questions to derive a composite “Quality” score, which
we use in our main analyses. As an additional assessment, we also utilized GPT-4, to
independentlyscorethesubjects’responses. Similarlytothehumangrades,weproduced
ascoreforeachoneofthe18questions,andthenacomposite“Quality(GPT)”score.
Figure 2 uses the composite human grader score and visually represents the
performance distribution across the three experimental groups, with the average score
plotted on the y-axis. A comparison of the dashed lines and the overall distributions of
the experimental conditions clearly illustrates the significant performance enhancements
associatedwiththeuseofGPT-4. BothAIconditionsshowclearsuperiorperformanceto
thecontrolgroupnotusingGPT-4.
Table 1 presents the results of the analyses using response quality as the dependent
variable and highlights the performance implications of using AI. Columns 1, 2, and
3 utilize human-generated grades as the dependent variable, while Column 4 uses the
5Theexecutivenotedtheonlystepmissingfromthisexercisewasanevaluationofhowthenewproduct
integrateswiththecompany’sexistingproductlines. Asourexperimentusedafictionalcompany,wedid
notrequireparticipantstopresenttheirproductsuggestionsinrelationtoexistingones.
6GraderswerefromBCG,orMBAstudentsatatopprogram.
9
composite grades generated by GPT-4. Across all specifications, both treatments — GPT
+OverviewandGPTOnly—demonstratepositiveeffects. InColumn1,GPT+Overview
leads to a 1.75 increase in scores over the control mean of 4.1, which is a 42.5% increase;
GPT Only led to a 1.56, or 38% increase. Notably, Columns 2, 3, and 4 incorporate
performance metrics from the assessment task and the treatment coefficients they report
remainveryconsistent. Column4usesGPTscoresasthedependentvariable,andshows
coefficientsof1.34fortheGPT+Overviewtreatmentand1.21fortheGPTOnlytreatment
over the control group, which are equal respectively to 18.6% and 16.8% increases in
performance.7
The beneficial impacts of using AI remain consistent across all our specifications. We
mergedourAItreatmentsandusedallourpre-registeredqualityvariablesasdependent
variables. This included individual grades for each question as evaluated by humans, as
wellasgradesevaluatedbyGPT-4,basedonthethreespecificationsoutlinedinColumns
1-3 of Table 1. This resulted in a comprehensive set of 108 regressions. All of these
regressions showed a significant effect of introducing AI on consultants’ performance.
Figures3and4show54oftheseregressionseach. Additionally,threedashedlinesreport
the average effects of each regression. The mean effect size when comparing subjects
using AI with a control with no GPT-4 access is 1.69 (a 46.6% increase over the control
mean)whenusinghumanevaluationsand1.36(20.2%)whenusingGPT-4evaluations.
Another key observation from the table is the differential impact of the two AI
treatments. Specifically, the GPT + Overview treatment consistently exhibits a more
pronounced positive effect compared to the GPT Only treatment. The bottom of the
table displays a p-value that tests whether the effects of receiving GPT + Overview were
equivalent to those of being assigned to GPT Only, showing this value to be below
or around the conventional 5% threshold in all specifications. This underscores the
importance of the added overview in enhancing the efficacy of AI assistance. However,
we should note that the overview increased “retainment” (i.e., copying and pasting the
GPT-4 output), and retainment itself was associated with better performance.8 The table
7These percentage improvements are relatively lower also because GPT-4 tends to be a more lenient
graderandscoresourcontrolbaselinehigher.
8AppendixCprovidesfurtherdetails.
10
also highlights various other factors, such as gender, native English proficiency, tenure,
location,andtechopenness,andtheirinfluenceontheoutcomes.9
Table 2 presents the results related to the percentage of task completion by subjects,
which is the dependent variable in this analysis. Across Columns 1, 2, and 3, both
treatments — GPT + Overview and GPT Only — demonstrate a positive effect on task
completion. Onaverage,thesecoefficientsindicatea12.2%increaseincompletionrates.10
The control group completed on average 82% of their tasks, while the GPT + Overview
condition completed about 93% and GPT Only about 91%. Column 2 incorporates the
performance metric from the assessment, and Column 3 further extends the analysis
by including the same set of controls as in Table 1. The coefficients suggest that the
integrationofAItoolsenhancestherateoftaskcompletionverysignificantly,atthesame
timeasitincreasesquality.
Figure 5 presents an important trend: the most significant beneficiaries of using AI
are the bottom-half-skill subjects, consistent with findings from Noy and Zhang (2023)
and Choi and Schwarcz (2023).11 By segmenting subjects exposed to one of the two AI
conditions into two distinct categories — top-half-skill performers (those ranking in the
top 50% on the assessment task) and bottom-half-skill performers (those in the bottom
50%)—weobservedperformanceenhancementsintheexperimentaltaskforbothgroups
when leveraging GPT-4. When comparing the two groups, though, we see the bottom-
half-skillperformersexhibitedthemostsubstantialsurgeinperformance,43%,compared
to the top-half-skill subjects, 17%. Note that the top-half-skill performers also receive a
significantboost,althoughnotasmuchasthebottom-half-skillperformers.
For the task inside the frontier, we did not allow any subjects to complete the task
before the allotted time was over. Instead, their final question was an especially long
9Weemploybinaryvariablesforallthesefactors. "Female"issetto1ifasubjectidentifiesasfemaleand
0otherwise. "EnglishNative"is1ifasubjectclaimsnativeproficiencyinEnglishand0otherwise(nearly
everysubjectindicateseitherNativeorAdvancedproficiencyinEnglish). "LowTenure"is1ifasubjecthas
beenwithBCGforayearorless,and0otherwise. "Location"is1ifasubject’sofficeislocatedinEuropeor
theMiddleEast,and0otherwise. Lastly,"TechOpenness"is1ifthesubjectexpressedahigherreceptivity
totechnologyintheirenrollmentsurvey,and0otherwise.
10When directly comparing the two AI treatments at the bottom of the table, the difference in their
impactsisnotstatisticallysignificant.
11Itisimportanttonotethat"higher-skill"and"lower-skill"herearerelative. Alltheseconsu |
280 | bcg | 24-013_d9b45b68-9e74-42d6-a1c6-c72fb70c7282.pdf | Working Paper 24-013
Navigating the Jagged Technological
Frontier: Field Experimental Evidence
of the Effects of AI on Knowledge
Worker Productivity and Quality
Fabrizio Dell'Acqua Saran Rajendran
Edward McFowland III Lisa Krayer
Ethan Mollick François Candelon
Hila Lifshitz-Assaf Karim R. Lakhani
Katherine C. Kellogg
Navigating the Jagged Technological
Frontier: Field Experimental Evidence
of the Effects of AI on Knowledge
Worker Productivity and Quality
Fabrizio Dell'Acqua Saran Rajendran
Harvard Business School Boston Consulting Group
Edward McFowland III Lisa Krayer
Harvard Business School Boston Consulting Group
Ethan Mollick François Candelon
The Wharton School Boston Consulting Group
Hila Lifshitz-Assaf Karim R. Lakhani
Warwick Business School Harvard Business School
Katherine C. Kellogg
MIT Sloan School of Management
Working Paper 24-013
Copyright © 2023 by Fabrizio Dell’Acqua, Edward McFowland III, Ethan Mollick, Hila Lifshitz-Assaf, Katherine
C. Kellogg, Saran Rajendran, Lisa Krayer, François Candelon, and Karim R. Lakhani.
Working papers are in draft form. This working paper is distributed for purposes of comment and discussion only.
It may not be reproduced without permission of the copyright holder. Copies of working papers are available from
the author.
We thank Michael Bervell, John Cheng, Pallavi Deshpande, Maxim Ledovskiy, John Kalil, Kelly Kung, Rick
Lacerda, MarcAntonio Awada, Paula Marin Sariego, Rafael Noriega, Alejandro Ortega, Rahul Phanse, Quoc-Anh
Nguyen, Nitya Rajgopal, Ogbemi Rewane, Kyle Schirmann, Andrew Seo, Tanay Tiwari, Elliot Tobin, Lebo
Nthoiwa, Patrick Healy, Saud Almutairi, Steven Randazzo, Anahita Sahu, Aaron Zheng, and Yogesh Kumaar for
helpful research assistance. We thank Kevin Dai for outstanding support with data and visualizations. For helpful
feedback, we thank Maxime Courtaux, Clement Dumas, Gaurav Jha, Jesse Li, Max Männig, Michael Menietti,
Rachel Mural, Zahra Rasouli, Esther Yoon, Leonid Zhukov, and David Zuluaga Martínez. Lakhani would like to
thank Martha Wells, Anne Leckie, Iain Banks, and Alastair Reynolds for inspiring AI futures. We used Poe,
Claude, and ChatGPT for light copyediting and graphics creations. Lakhani is an advisor to Boston Consulting
Group on AI Strategy and learning engagement.
Funding for this research was provided in part by Harvard Business School.
Navigating the Jagged Technological Frontier:
Field Experimental Evidence of the Effects of AI on
Knowledge Worker Productivity and Quality*
Fabrizio Dell’Acqua1, Edward McFowland III1, Ethan Mollick2, Hila
Lifshitz-Assaf1,3, Katherine C. Kellogg4, Saran Rajendran5, Lisa Krayer5,
François Candelon5, and Karim R. Lakhani1
1Digital Data Design Institute, Harvard Business School; 2The Wharton
School, University of Pennsylvania; 3Warwick Business School, Artificial
Intelligence Innovation Network; 4MIT Sloan School of Management;
5Boston Consulting Group, Henderson Institute
September 22, 2023
*Fabrizio Dell’Acqua ([email protected]), Edward McFowland III ([email protected]),
Ethan Mollick ([email protected]), Hila Lifshitz-Assaf ([email protected]),
Katherine C. Kellogg ([email protected]), Saran Rajendran ([email protected]), Lisa
Krayer ([email protected]), François Candelon ([email protected]), Karim R. Lakhani
([email protected]). We thank Michael Bervell, John Cheng, Pallavi Deshpande, Maxim Ledovskiy, John
Kalil, Kelly Kung, Rick Lacerda, MarcAntonio Awada, Paula Marin Sariego, Rafael Noriega, Alejandro
Ortega,RahulPhanse,Quoc-AnhNguyen,NityaRajgopal,OgbemiRewane,KyleSchirmann,AndrewSeo,
TanayTiwari,ElliotTobin,LeboNthoiwa,PatrickHealy,SaudAlmutairi,StevenRandazzo,AnahitaSahu,
Aaron Zheng, and Yogesh Kumaar for helpful research assistance. We thank Kevin Dai for outstanding
supportwithdataandvisualizations. Forhelpfulfeedback,wethankMaximeCourtaux,ClementDumas,
Gaurav Jha, Jesse Li, Max Männig, Michael Menietti, Rachel Mural, Zahra Rasouli, Esther Yoon, Leonid
Zhukov, and David Zuluaga Martínez. Lakhani would like to thank Martha Wells, Anne Leckie, Iain
Banks, and Alastair Reynolds for inspiring AI futures. We used Poe, Claude, and ChatGPT for light
copyeditingandgraphicscreations. LakhaniisanadvisortoBostonConsultingGrouponAIStrategyand
learningengagement. Allerrorsareourown.
1
Abstract
The public release of Large Language Models (LLMs) has sparked tremendous
interestinhowhumanswilluseArtificialIntelligence(AI)toaccomplishavarietyof
tasks. In our study conducted with Boston Consulting Group, a global management
consulting firm, we examine the performance implications of AI on realistic,
complex, and knowledge-intensive tasks. The pre-registered experiment involved
758 consultants comprising about 7% of the individual contributor-level consultants
at the company. After establishing a performance baseline on a similar task, subjects
were randomly assigned to one of three conditions: no AI access, GPT-4 AI access,
or GPT-4 AI access with a prompt engineering overview. We suggest that the
capabilities of AI create a “jagged technological frontier” where some tasks are
easily done by AI, while others, though seemingly similar in difficulty level, are
outside the current capability of AI. For each one of a set of 18 realistic consulting
tasks within the frontier of AI capabilities, consultants using AI were significantly
moreproductive(theycompleted12.2%moretasksonaverage,andcompletedtasks
25.1% more quickly), and produced significantly higher quality results (more than
40% higher quality compared to a control group). Consultants across the skills
distribution benefited significantly from having AI augmentation, with those below
theaverageperformancethreshold increasingby43%andthose above increasingby
17% compared to their own scores. For a task selected to be outside the frontier,
however,consultantsusingAIwere19percentagepointslesslikelytoproducecorrect
solutionscomparedtothosewithoutAI.Further,ouranalysisshowstheemergenceof
twodistinctivepatternsofsuccessfulAIusebyhumansalongaspectrumofhuman-
AI integration. One set of consultants acted as “Centaurs,” like the mythical half-
horse/half-humancreature,dividinganddelegatingtheirsolution-creationactivities
to the AI or to themselves. Another set of consultants acted more like “Cyborgs,”
completely integrating their task flow with the AI and continually interacting with
thetechnology.
2
1 Introduction
The capabilities of Artificial Intelligence to produce human-like work have improved
rapidly,especiallysincethereleaseofOpenAI’sChatGPT,oneofseveralLargeLanguage
Models (LLMs) that are widely available for public use. As AI capabilities overlap more
with those of humans, the integration of human work with AI poses new fundamental
challenges and opportunities, in particular in knowledge-intensive domains. In this
paper, we examine this issue using randomized controlled field experiments with highly
skilled professional workers. Our results demonstrate that AI capabilities cover an
expanding, but uneven, set of knowledge work we call a "jagged technological frontier.”
Within this growing frontier, AI can complement or even displace human work; outside
of the frontier, AI output is inaccurate, less useful, and degrades human performance.
However, because the capabilities of AI are rapidly evolving and poorly understood,
it can be hard for professionals to grasp exactly what the boundary of this frontier
might be at a given. We find that professionals who skillfully navigate this frontier gain
large productivity benefits when working with the AI, while AI can actually decrease
performancewhenusedforworkoutsideofthefrontier.
Though LLMs are new, the impact of other, earlier forms of AI have been the subject
ofconsiderablescholarlydiscussion(e.g.,Brynjolfssonetal.(2018);FurmanandSeamans
(2019);Puranam(2021)). BecauseofthelimitationsoftheseearlierformsofAI,nonroutine
tasks that were difficult to codify seemed protected from automation (Autor et al., 2003;
Acemoglu and Restrepo, 2019), especially as previous waves of technology had mostly
automatedlower-skilledoccupations(GoldinandKatz,1998). ThereleaseofChatGPTin
November, 2022 changed both the nature and urgency of the discussion. LLMs proved
unexpectedly capable at creative, analytical, and writing tasks, including scoring at top
levelsatgraduateandprofessionalexaminations(Girotraetal.,2023;Geerlingetal.,2023;
Kung et al., 2023; Boussioux et al., 2023). This represented an entirely new category of
automation, one whose abilities overlapped with the most creative, most educated, and
mosthighlypaidworkers(Eloundouetal.,2023).
Studies of previous generations of AI (Brynjolfsson et al., 2023) and controlled
1
experiments on the impact of recently released LLMs (e.g., Noy and Zhang (2023); Choi
and Schwarcz (2023)) suggest that these systems can have a large impact on worker
performance. In our study, we focus on complex tasks, selected by industry experts to
replicate real-world workflows as experienced by knowledge workers. Most knowledge
work includes this sort of flow, a set of interdependent tasks, some of which may be
good fit for current AI, while some are not. We examine both kinds of tasks, and build
on recent studies to suggest ways of understanding the rapidly evolving impact of AI on
knowledgeworkers,underwhichcircumstancesorganizationsmaybenefit,andhowthis
mightchangeasthetechnologyadvances.
This is important because understanding the implications of LLMs for the work
of organizations and individuals has taken on urgency among scholars, workers,
companies, and even governments (Agrawal et al., 2018; Iansiti and Lakhani, 2020; Berg
etal.,2023). PreviousformsofAIledtoconsiderabledebateintheliteraturearoundhow
and whether professionals should adopt AI for knowledge work (Anthony et al., 2023)
and the potential impact this might have on organizations (Raisch and Krakowski, 2021;
Glaeser et al., 2021; Brynjolfsson et al., 2021). Some scholars focused on the potential
for AI to help professionals improve their effectiveness and efficiency (DeStefano et al.,
2022;Balakrishnanetal.,2022;ValentineandHinds,2023). Othersdemonstratedthat,for
critical tasks, it can be risky for professionals to use AI (Lebovitz et al., 2021), especially
black-boxed (e.g., Lebovitz et al. (2022); Waardenburg et al. (2022)), and showed how
professionals are struggling to use it effectively (Pachidi et al., 2021; Van den Broek et al.,
2021). Finally, another group of researchers argued that the “algorithmic management”
affordedbyAIcancreatenegativepersonalimpactsforprofessionals(Kelloggetal.,2020;
Möhlmann et al., 2021; Tong et al., 2021) and raise accountability and ethical questions
(Choudhuryetal.,2020;Cowgilletal.,2020;Rahmanetal.,2024). Yet,mostofthestudies
predate ChatGPT, and investigate forms of AI designed to produce discrete predictions
basedonpastdata. ThesesystemsarequitedifferentfromLLMs.
Specifically, outside of their technical differences from previous forms of machine
learning, there are three aspects of LLMs that suggest they will have a much more
rapid, and widespread, impact on work. The first is that LLMs have surprising
2
capabilities that they were not specifically created to have, and ones that are growing
rapidly over time as model size and quality improve. Trained as general models,
LLMsnonethelessdemonstratespecialistknowledgeandabilitiesaspartoftheirtraining
processandduringnormaluse(Singhaletal.,2022;Boikoetal.,2023). Whileconsiderable
debate remains on the concept of emergent capabilities from a technological perspective
(Schaeffer et al., 2023), the effective capabilities of AIs are novel and unexpected, widely
applicable, and are increasing greatly in short time spans. Recent work has shown
that AI performs at a high level in professional contexts ranging from medicine to law
(Ali et al., 2023; Lee et al., 2023), and beats humans on many measures of innovation
(Boussioux et al., 2023; Girotra et al., 2023). And, while score performance on various
standardized academic tests is an imperfect measure of LLM capabilities, it has been
increasingsubstantiallywitheachgenerationofAImodels(OpenAI,2023).
The general ability of LLMs to solve domain-specific problems leads to the second
differentiating factor of LLMs compared to previous approaches to AI: their ability to
directly increase the performance of workers who use these systems, without the need
for substantial organizational or technological investment. Early studies of the new
generation of LLMs suggest direct performance increases from using AI, especially for
writing tasks (Noy and Zhang, 2023) and programming (Peng et al., 2023), as well as
for ideation and creative work (Boussioux et al., 2023; Girotra et al., 2023). As a result,
the effects of AI are expected to be higher on the most creative, highly paid, and highly
educatedworkers(Eloundouetal.,2023;Feltenetal.,2023)
The final relevant characteristic of LLMs is their relative opacity. This extends to the
failurepointsofAImodels,whichincludeatendencytoproduceincorrect,butplausible,
results (hallucinations or confabulations), and to make other types of errors, including in
mathandwhenprovidingcitations. TheadvantagesofAI,whilesubstantial,aresimilarly
unclear to users. It performs well at some jobs, and fails in other circumstances in ways
thataredifficulttopredictinadvance. Contributingfurthertotheopacityisthatthebest
ways to use these AI systems are not provided by their developers and appear to be best
learnedviaongoingusertrial-and-errorandthesharingofexperiencesandheuristicsvia
variousonlineforumslikeusergroups,hackathons,TwitterfeedsandYouTubechannels.
3
Taken together, these three factors – the surprising abilities of LLMs, their ability to
do real work with virtually no technical skill required of the user, and their opacity and
unclear failure points – suggest that the value and downsides of AI may be difficult for
workers and organizations to grasp. Some unexpected tasks (like idea generation) are
easy for AIs, while other tasks that seem to be easy for machines to do (like basic math)
are challenges for some LLMs. This creates a “jagged Frontier,” where tasks that appear
to be of similar difficulty may either be performed better or worse by humans using
AI. Due to the “jagged” nature of the frontier, the same knowledge workflow of tasks
can have tasks on both sides of the frontier, see Figure 1. The future of understanding
how AI impacts work involves understanding how human interaction with AI changes
depending on where tasks are placed on this frontier, and how the frontier will change
over time. Investigating how humans navigate this jagged frontier, and the subsequent
performanceimplications,isthefocusofourwork.
Wecollaboratedwithaglobalmanagementconsultingfirm(BostonConsultingGroup
- BCG) and advised them on designing, developing, and executing two pre-registered
randomized experiments to assess the impact of AI on high humancapital professionals.
Subsequently, the author team received the data that the company collected for the
purpose of this experiment and conducted the analysis presented in this paper. The
studywasstructuredinthreephases: aninitialdemographicandpsychologicalprofiling,
a main experimental phase involving multiple task completions, and a concluding
interview segment. We tested two distinct tasks: one situated outside the frontier of
AI capabilities and the other within its bounds. The experiment aimed to understand
how AI integration might reshape the traditional workflows of these high human capital
professionals.
OurresultsshowthatthisgenerationofLLMsarehighlycapableofcausingsignificant
increases in quality and productivity, or even completely automating some tasks, but the
actual tasks that AI can do are surprising and not immediately obvious to individuals or
even to producers of LLMs themselves. Because this frontier is expanding and changing,
the overall results suggest that AI will have a large impact on work, one which will
increasewithLLMcapabilities,butwheretheimpactsoccurwillbeuneven.
4
2 Methods
We collected data from two randomized experiments to assess the causal impact of AI,
specifically GPT-4 – the most capable of the AI models at the time of the experiments
(Spring 2023) – on high human capital professionals working traditionally without AI.1
We pre-registered our study, detailing the design structure, the experimental conditions,
thedependentvariables,andourmainanalyticalapproaches.2 Ouraimwastodetermine
how introducing this AI into the tasks of highly-skilled knowledge workers might
augment,disrupt,orinfluencetheirtraditionalworkflows.
BCG individual contributor consultants around the world were offered the
opportunity to spend 5 hours working on this experiment to evaluate the impact of AI
ontheiractivities. Approximately7%ofBCG’sglobalindividualcontributorconsultants’
cohortengagedinandcompletedtheexperiment.
Theexperimentwasstructuredintothreedistinctphases. Initially,consultantsentered
the study by completing a survey that captured their demographic and psychological
profiles, as well as details about their role within the company. A few weeks after
enrolling, participants received a link to complete the main experimental phase. This
phase commenced with a pre-task survey, followed by the tasks detailed subsequently,
and concluded with a post-task survey. In the final phase, participants were interviewed
tosharetheirexperiencesandperspectivesontheroleofAIintheirprofession.
In the first phase, we administered an enrollment survey to gather information about
potential participants.3 This survey captured details such as office location, internal
affiliation, and tenure at BCG. Participants also completed psychological assessments,
specifically providing insights into their Big 5 personality traits (Soto and John, 2017),
innovativeness (Agarwal and Prasad, 1998), self-perceived creativity (Miron-Spektor
et al., 2004), and paradox mindset (Miron-Spektor et al., 2018). Furthermore, the
survey included a short section on their reading habits (including their familiarity with
1TheprojecthasreceivedIRBapproval,IRB23-0392.
2Pre-registrationcompletedonOpenScienceFoundation,osf.io/ytaev. Thepre-registrationisavailable
fromtheauthorsuponrequestandwillbemadepubliclyavailableafterarticleacceptanceoraftertheOSF
embargoperiodhaspassed,whichevercomesfirst
3Outofthe852consultantswhorespondedtothesurvey,758-about89%-completedtheexperiment.
5
AI characters in fiction), and demographic details like gender, native language, and
educational background. We utilized these data for stratified random assignment and
ascontrolsinourregressionmodels,asdescribedbelow.
The study encompassed 758 strategy consultants, each of whom completed the initial
survey and experimental tasks. Each participant was assigned to one of two distinct
experiments. Stratification of participants was based on multiple criteria, both between
experimentsandacrossexperimentaltreatments. Thesecriteriaincludedgender,location,
tenure at BCG, individual openness to innovation, and native English-speaking status.
This information was collected with the survey administered during phase one, a few
weeksbeforethemainexperiment.
In order to ensure genuine engagement and effort from participants, we incentivized
their performance in the experiment. Participants who diligently participated in all
aspects of the experiment were honored with an "office contribution" recognition,
carryingfinancialimplicationsrelatedtotheirannualbonuses. Furthermore,torecognize
and reward excellence, the top 20% of participants received additional recognition, and
the top 5% was also awarded with a small gift. Executives at BCG reported that the
recognition received by top participants was important because it was shared with the
committeethatoverseestheircareerdevelopmentandperformanceassessments.
Subjects were allocated to one of two distinct experiments, each involving a
unique type of task, with no overlap between the groups. Both tasks were designed
in collaboration with multiple people at BCG to represent some of the typical job
activities encountered by individual contributor consultants. Approximately half of the
participants (385 consultants) tackled a series of tasks where they were prompted to
conceptualize and develop new product ideas, focusing on aspects such as creativity,
analytical skills, persuasiveness and writing skills. The other half (373 consultants)
engaged in business problem-solving tasks using quantitative data, customer and
company interviews, and including a persuasive writing component. Both sets of tasks
were developed to be realistic, and were designed with the input of professionals in
the respective sectors. A senior level executive at the company commented on these
tasks being “very much in line with part of the daily activities” of the subjects involved.
6
Notably,someformsofthesetasksarealsousedbythecompanytoscreenjobapplicants,
typically from elite academic backgrounds (including Ph.D.s), for their highly-coveted
positions.
Both experiments followed a consistent structure. Initially, participants undertook a
task without the aid of AI, establishing a baseline for performance and enabling within-
subject analyses. Following this, participants were randomly assigned to one of three
conditions to assess the influence of AI on their tasks, with these conditions being
consistent across both experiments. The first group (a control condition) proceeded
without AI support; the second (“GPT Only”) had the assistance of an AI tool based
on GPT-4; and the third (“GPT + Overview”) not only utilized the same AI tool but
also benefited from supplementary prompt engineering overview, which increased their
familiarity with AI. These materials included instructional videos and documents that
outlinedandillustratedeffectiveusagestrategies.
Ratherthanrelyingonself-reportedmetricsorindirectindicators,wedirectlyassessed
participants’ skills through a task that closely mirrored the main experiment. In both
experiments, we employ an assessment task that, while different from the experimental
task, is highly comparable, ensuring a precise evaluation of skills for this specific task
type.4 Our findings indicate that performance in the assessment task is a predictor of
performance in the experimental task, allowing us to study the differential effects of
introducingAItoparticipantsofdifferentskilllevels.
Each task assigned to participants came with a specific time allocation. In the
experiment using a task inside the frontier, the assessment task duration was set for 30
minutes, while the subsequent one was allotted 90 minutes. Conversely, in the outside-
the-frontierexperiment,boththefirstandsecondtasksweredesignated60minuteseach,
though participants could complete them earlier if they finished ahead of time. It is
importanttonotethatforthetaskinsidethefrontier,participantswererequiredtoremain
on the task’s page for the entire duration of the task, and could not complete the exercise
earlier. This approach ensured that our analysis for the inside-the-frontier tasks focused
4Dell’Acqua et al. (2023) adopts a comparable experimental framework to evaluate subjects’
competencies.
7
chiefly on the qualitative differences, rather than any timing improvements brought
aboutbyusingAI.Thesetimeframeswereautomaticallyenforced,withtheexperimental
systemadvancingtothenextquestiononcethestipulatedtimeforataskelapsed.
In every experimental task, subjects assigned to the AI conditions had access to
a company platform. This platform, developed using the OpenAI API, facilitated an
interactive experience with OpenAI’s GPT-4, mirroring the dynamics of ChatGPT. It
enabled the collection of all participants’ prompts and AI’s corresponding responses,
providing a comprehensive view into the collaborative behaviors between subjects and
AI.Allsubjectsusedthesameversionofthetool,accessingGPT-4asavailableattheend
ofApril,2023,andusingdefaultsystempromptsandtemperature.
Asidefromthethematicdifferences,thetasksdifferedinanotherkeyway. Whileboth
were designed to be comparably complex and realistic, the first task was selected to be
withinthepotentialtechnologicalfrontierofGPT-4. Thesecondexperimentwasdesigned
sothatGPT-4wouldmakeanerrorwhenconductingtheanalysis,ensuringtheworkfell
justoutsidethefrontier.
3 Results
3.1 Quality and Productivity Booster - Inside the Frontier
The inside-the-frontier experiment focused on creative product innovation and
development. The initial assessment task asked participants to brainstorm innovative
beverage concepts. From their set of ideas, they identified the most viable option and
devisedacomprehensiveplanforitsmarketdebut. Afterthistask,subjectsmovedtothe
mainexperimentalphaseandthecontexttransitionedtothemainexperimentaltask.
In this experimental task, participants were tasked with conceptualizing a footwear
idea for niche markets and delineating every step involved, from prototype description
to market segmentation to entering the market. An executive from a leading global
footwearcompanyverifiedthatthetaskdesigncoveredtheentireprocesstheircompany
8
typically goes through, from ideation to product launch.5 Participants responded to
a total of 18 tasks (or as many as they could within the given time frame). These
tasks spanned various domains. Specifically, they can be categorized into four types:
creativity (e.g., “Propose at least 10 ideas for a new shoe targeting an underserved
market or sport.”), analytical thinking (e.g., “Segment the footwear industry market
based on users.”), writing proficiency (e.g., “Draft a press release marketing copy for
your product.”), and persuasiveness (e.g., “Pen an inspirational memo to employees
detailing why your product would outshine competitors.”). This allowed us to collect
comprehensiveassessmentsofquality. AlltasksanddetailsarereportedinAppendixA.
In the experiment, the primary outcome variable is the quality of the subjects’
responses. To quantify this quality, we employed a set of human graders to evaluate
each question that participants didn’t leave unanswered.6 Each response was evaluated
by two human graders. We then calculated the mean grade assigned by humans to each
question. This gave us 18 dependent variables (one per each question). We subsequently
averaged these scores across all questions to derive a composite “Quality” score, which
we use in our main analyses. As an additional assessment, we also utilized GPT-4, to
independentlyscorethesubjects’responses. Similarlytothehumangrades,weproduced
ascoreforeachoneofthe18questions,andthenacomposite“Quality(GPT)”score.
Figure 2 uses the composite human grader score and visually represents the
performance distribution across the three experimental groups, with the average score
plotted on the y-axis. A comparison of the dashed lines and the overall distributions of
the experimental conditions clearly illustrates the significant performance enhancements
associatedwiththeuseofGPT-4. BothAIconditionsshowclearsuperiorperformanceto
thecontrolgroupnotusingGPT-4.
Table 1 presents the results of the analyses using response quality as the dependent
variable and highlights the performance implications of using AI. Columns 1, 2, and
3 utilize human-generated grades as the dependent variable, while Column 4 uses the
5Theexecutivenotedtheonlystepmissingfromthisexercisewasanevaluationofhowthenewproduct
integrateswiththecompany’sexistingproductlines. Asourexperimentusedafictionalcompany,wedid
notrequireparticipantstopresenttheirproductsuggestionsinrelationtoexistingones.
6GraderswerefromBCG,orMBAstudentsatatopprogram.
9
composite grades generated by GPT-4. Across all specifications, both treatments — GPT
+OverviewandGPTOnly—demonstratepositiveeffects. InColumn1,GPT+Overview
leads to a 1.75 increase in scores over the control mean of 4.1, which is a 42.5% increase;
GPT Only led to a 1.56, or 38% increase. Notably, Columns 2, 3, and 4 incorporate
performance metrics from the assessment task and the treatment coefficients they report
remainveryconsistent. Column4usesGPTscoresasthedependentvariable,andshows
coefficientsof1.34fortheGPT+Overviewtreatmentand1.21fortheGPTOnlytreatment
over the control group, which are equal respectively to 18.6% and 16.8% increases in
performance.7
The beneficial impacts of using AI remain consistent across all our specifications. We
mergedourAItreatmentsandusedallourpre-registeredqualityvariablesasdependent
variables. This included individual grades for each question as evaluated by humans, as
wellasgradesevaluatedbyGPT-4,basedonthethreespecificationsoutlinedinColumns
1-3 of Table 1. This resulted in a comprehensive set of 108 regressions. All of these
regressions showed a significant effect of introducing AI on consultants’ performance.
Figures3and4show54oftheseregressionseach. Additionally,threedashedlinesreport
the average effects of each regression. The mean effect size when comparing subjects
using AI with a control with no GPT-4 access is 1.69 (a 46.6% increase over the control
mean)whenusinghumanevaluationsand1.36(20.2%)whenusingGPT-4evaluations.
Another key observation from the table is the differential impact of the two AI
treatments. Specifically, the GPT + Overview treatment consistently exhibits a more
pronounced positive effect compared to the GPT Only treatment. The bottom of the
table displays a p-value that tests whether the effects of receiving GPT + Overview were
equivalent to those of being assigned to GPT Only, showing this value to be below
or around the conventional 5% threshold in all specifications. This underscores the
importance of the added overview in enhancing the efficacy of AI assistance. However,
we should note that the overview increased “retainment” (i.e., copying and pasting the
GPT-4 output), and retainment itself was associated with better performance.8 The table
7These percentage improvements are relatively lower also because GPT-4 tends to be a more lenient
graderandscoresourcontrolbaselinehigher.
8AppendixCprovidesfurtherdetails.
10
also highlights various other factors, such as gender, native English proficiency, tenure,
location,andtechopenness,andtheirinfluenceontheoutcomes.9
Table 2 presents the results related to the percentage of task completion by subjects,
which is the dependent variable in this analysis. Across Columns 1, 2, and 3, both
treatments — GPT + Overview and GPT Only — demonstrate a positive effect on task
completion. Onaverage,thesecoefficientsindicatea12.2%increaseincompletionrates.10
The control group completed on average 82% of their tasks, while the GPT + Overview
condition completed about 93% and GPT Only about 91%. Column 2 incorporates the
performance metric from the assessment, and Column 3 further extends the analysis
by including the same set of controls as in Table 1. The coefficients suggest that the
integrationofAItoolsenhancestherateoftaskcompletionverysignificantly,atthesame
timeasitincreasesquality.
Figure 5 presents an important trend: the most significant beneficiaries of using AI
are the bottom-half-skill subjects, consistent with findings from Noy and Zhang (2023)
and Choi and Schwarcz (2023).11 By segmenting subjects exposed to one of the two AI
conditions into two distinct categories — top-half-skill performers (those ranking in the
top 50% on the assessment task) and bottom-half-skill performers (those in the bottom
50%)—weobservedperformanceenhancementsintheexperimentaltaskforbothgroups
when leveraging GPT-4. When comparing the two groups, though, we see the bottom-
half-skillperformersexhibitedthemostsubstantialsurgeinperformance,43%,compared
to the top-half-skill subjects, 17%. Note that the top-half-skill performers also receive a
significantboost,althoughnotasmuchasthebottom-half-skillperformers.
For the task inside the frontier, we did not allow any subjects to complete the task
before the allotted time was over. Instead, their final question was an especially long
9Weemploybinaryvariablesforallthesefactors. "Female"issetto1ifasubjectidentifiesasfemaleand
0otherwise. "EnglishNative"is1ifasubjectclaimsnativeproficiencyinEnglishand0otherwise(nearly
everysubjectindicateseitherNativeorAdvancedproficiencyinEnglish). "LowTenure"is1ifasubjecthas
beenwithBCGforayearorless,and0otherwise. "Location"is1ifasubject’sofficeislocatedinEuropeor
theMiddleEast,and0otherwise. Lastly,"TechOpenness"is1ifthesubjectexpressedahigherreceptivity
totechnologyintheirenrollmentsurvey,and0otherwise.
10When directly comparing the two AI treatments at the bottom of the table, the difference in their
impactsisnotstatisticallysignificant.
11Itisimportanttonotethat"higher-skill"and"lower-skill"herearerelative. Alltheseconsu |
281 | bcg | bcg-accelerating-climate-action-with-ai-nov-2023-rev.pdf | Accelerating Climate
Action with AI
November 2023
By Amane Dannouni, Stefan A. Deutscher, Ghita Dezzaz, Adam Elman, Antonia Gawel,
Marsden Hanna, Andrew Hyland, Amjad Kharij, Hamid Maher, David Patterson,
Edmond Rhys Jones, Juliet Rothenberg, Hamza Tber, Maud Texier, and Ali Ziat
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.
CCoonntteennttss
01 Foreword 28 AI for Climate:
A Summary of Critical Policy
Outcomes
02 E xecutive Summary
41 Endnotes
05 The Climate Action Imperative
and the Promise of AI
43 About the Authors
09 How AI Can Help
Accelerate Climate Action 45 Acknowledgements
22 Navigating AI’s Potential Risks 47 References
Foreword
This report aims to provide policymakers, corporate decision
makers, and climate leaders with a clear and concise understanding
of the role that artificial intelligence (AI) can play in climate action.
More specifically, its goals are to highlight AI’s significant This work draws on interviews with a range of climate
potential to help address our environmental challenges, change and AI experts, builds on previous research from
to shed light on climate-relevant AI risks, and to offer organizations including Climate Change AI and the AI for
policymakers a streamlined framework for desirable the Planet Alliance, and leverages BCG’s analysis and client
policy outcomes. experience as well as Google’s technical and operational
expertise—and its experience in developing solutions.
Throughout the report, we share examples of successful
early applications of AI for climate and of instances in
which policymakers have already taken the initiative to
enable, promote, or guide the use of AI for climate action
across sectors.
1 ACCELERATING CLIMATE ACTION WITH AI
Executive Summary
Accelerating climate action is imperative, as we are While AI is only just starting to be applied to climate
on a path to fall short of the Paris Agreement’s goal challenges, leading-edge organizations and use cases
to keep warming under 1.5° Celsius. are already delivering results—and demonstrating the
promise of AI for climate—along three dimensions.
• The United Nations Intergovernmental Panel on Climate
Change (IPCC) estimates that, based on action to date, • Information. AI-curated information sources are aiding
the world will likely see warming of 2.8°C with cata- nations in shaping their climate strategy—and in re-
strophic consequences. sponding to emergencies such as wildfires.
• The IPCC forecasts that in order to meet the 1.5°C goal, • Prediction. AI’s predictive power is helping save lives by
the world will need to reduce emissions—from the base- offering advance warning of floods.
line of 2010 levels—by 43% by 2030.
• Optimization. AI applications are enabling organiza-
By scaling currently proven applications and tions to understand and reduce their Scope 1, 2, and 3
technology, artificial intelligence (AI) has the carbon footprints.1
potential to unlock insights that could help mitigate
5% to 10% of global greenhouse gas (GHG) emissions AI also poses risks that must be considered and
by 2030—and significantly bolster climate-related managed thoughtfully to ensure its use has a net
adaptation and resilience initiatives. positive impact on climate.
• 87% of executives view AI as having the potential to • Energy-Related GHG Emissions. A 2022 paper in
address climate issues. Nature Climate Change estimates that cloud and
hyperscale data centers are responsible for 0.1%–0.2%
• AI’s positive impact will be multiplied should it contrib- of global GHG emissions and that roughly 25% of data
ute to scientific breakthroughs that open new pathways center workloads are related to machine learning (ML).
for climate action. Yet, newer and more complex AI models may require
more energy. At present, robust forecasts for AI’s future
AI can contribute to climate action by reducing energy requirements remain elusive given uncertain
emissions, guiding adaptations to unavoidable adoption rates and the broad spectrum of potential
climate change impacts, and providing foundational technical advancements with the potential to decrease
capabilities that enable climate action. AI’s energy intensity. Nonetheless, AI providers are
already striving to enhance energy efficiency and
• Mitigation. Helping with both the reduction and remov- integrate clean energy sources.
al of emissions—and with the underlying measurement
needed to size the challenge and track progress • Water Use. Water-based cooling remains the most
energy-efficient option for data centers, and its overall
• Adaptation and Resilience. Aiding countries, regions, impact on water consumption is low. In 2016 in the US,
cities, citizens, and businesses in forecasting climate- data centers were estimated to have used less than
related hazards, developing plans to address them, and 0.02% of the country’s water consumption for cooling.
responding in real time to crises Nevertheless, in some cases, water-based cooling can
put pressure on local water resources. Data center oper-
• Foundational Capabilities. Enabling climate-related ators have begun to address this issue by providing more
modeling, research into climate economics, and new disclosure, exploring new cooling techniques, and invest-
approaches to climate education and supporting break- ing in replenishment initiatives.
throughs in fundamental research
BOSTON CONSULTING GROUP GOOGLE 2
• Waste. While data centers currently account for only a Policymakers have a critical oversight role to play in
small fraction of the world’s e-waste challenge, there is maximizing the benefits from AI-driven climate
an opportunity for tech firms to build on early circularity action while minimizing its risks. Critical policy
successes and take a more thoughtful approach embrac- outcomes to pursue include the following:
ing more recycling and reuse.
• Enabling AI for climate progress by encouraging data
• Other Potential Risks. AI applications should be sharing, ensuring affordable technology access, building
sustainable and equitable by intention. AI can be applied awareness, and investing in talent
to both climate-friendly and climate-unfriendly appli-
cations, can narrow or widen disparities between the • Accelerating the deployment of AI for climate by defining
Global North and the Global South, and can be trained public and private sector priorities, delivering on public
on data sets that reflect the world’s diversity. Leaders sector use cases, and encouraging private sector action
and model builders need to be mindful in their design
choices. • Promoting environmentally and socially responsible
deployment of AI
3 ACCELERATING CLIMATE ACTION WITH AI
How We Define Artificial Intelligence
According to the Massachusetts Institute of Technology, • delivering improved predictions (predictive use cases),
AI is defined as the ability of computers to imitate human and
cognitive functions such as learning and problem-solving,
using math and logic to simulate the process of reasoning • suggesting optimization moves and recommendations
that helps humans learn from new information and make to reach targets (prescriptive use cases).
decisions.
These goals can be attained by applying wide range of
For the purposes of this report, we are using a broader techniques including those in the table below--all of which
definition of AI that comprises a set of mathematical and we include in this report’s definition of AI.
computer science techniques aimed at analyzing data to
help understand and navigate real-world phenomena Applying AI to real-world problems is common practice
through: today. The technology has proven its ability to help public
and private organizations have a better understanding of
• providing better information (descriptive use cases), their context, provide better services, and improve their
operational performance.
Technology General Example Climate-Related Example
Advanced Analytics Supermarket Inventory Energy Consumption
Management. Advanced analytics Optimization. Advanced analytics
The use of advanced mathematical and
can identify best sellers and demand can optimize a building’s carbon
statistical techniques to develop insights
dynamics, enabling more efficient footprint by adjusting heating,
from structured and unstructured data.
shelving and restocking strategies, cooling, and lighting systems in
thereby reducing waste and ensuring response to real-time data from
popular items are always in stock. sensors and weather forecasts.
Machine Learning Credit Card Fraud Detection. Predicting Wildfires. Machine
Machine learning helps banks and learning models can analyze weather
Training computers to learn and make
credit card companies detect data, satellite imagery, and terrain
predictions from data. Historical data
unusual transactions, enabling them information to predict the likelihood
constitutes the inputs, while predictions
to alert card holders and minimize of wildfires, helping authorities take
based on new or unseen data are the
fraud losses. preventive measures and optimize
outputs.
resource allocation.
Deep Learning Medical Image Evaluation. Extreme Weather Prediction.
Applied to the analysis of medical Deep learning can analyze vast
A specialized form of machine learning
images such as X-rays and MRIs, amounts of historical and real-time
that uses artificial neural networks to
deep learning helps doctors diagnose meteorological and satellite data,
generate hierarchical insights from
diseases and other abnormalities leading to more accurate forecasts
diverse data sets, such as images, text, or
more accurately, enabling more for hurricanes, tornadoes, and
audio. These models are able to recognize
timely and effective treatments. typhoons.
patterns or features within the data, for
example, by identifying objects in images.
Large Language Models Customer Service Chatbots. Green Technology Innovation.
Large language models enable Large language models can
Advanced AI models trained on vast
companies to automate the process accelerate innovation by digesting
amounts of text data—and able to
of answering customer questions research papers and patent
generate human-like text as output, such
and helping troubleshoot issues, applications and rapidly surfacing
as for Generative AI use cases.
enhancing the efficiency of, and ideas and identifying knowledge gaps.
satisfaction with, customer service
operations.
BOSTON CONSULTING GROUP GOOGLE 4
The Climate Action Imperative
and the Promise of AI
D
espite significant progress over the last several years Even if the world succeeds in limiting warming to 1.5°C,
in mobilizing the global community to intensify its there will still be adverse impacts. Already today at 1.1°C,
climate actions and ambitions, the world is not on the IPCC reports that over 3 billion people live in areas
track to meet the Paris Agreement’s target to limit tem- highly vulnerable to climate impacts. We are already
perature rise to 1.5°C. This target was selected because seeing the impact on weather, agriculture, water security,
scientists believe that above that level, the effects would and migration. If we overshoot the target, the picture
be catastrophic and potentially irreversible. At present— becomes increasingly dire: seas will rise further, droughts
based on updated national pledges since COP26 in 2021— will be worse, and extreme weather events will be more
the United Nations Environment Programme currently common.
estimates that we are on a path to warming by 2.8°C.2
5 ACCELERATING CLIMATE ACTION WITH AI
In a 1.5°C world, the IPCC forecasts that 48% of the world’s Climate Action Has an Analytical Challenge—
population will be exposed to deadly heat levels for more and AI Can Help
than 20 days a year. In a 3° to 4°C world, that number
increases to 74%. If we stay on our current trajectory, the Leaders increasingly understand the urgency. So far, 194
World Bank estimates an additional 143 million people— parties to the Paris Agreement have developed Nationally
more than the combined populations of the United King- Determined Contributions (NDCs)—each representing
dom, Morocco, and Malaysia—could be displaced.3 And, detailed commitments for how their country will help the
absent significant investments in resilience, major global world meet the Paris Agreement’s 1.5°C goal—up from 75
cities—for example, Tokyo, Osaka, Mumbai, Bangkok, parties in February 2021.4
New York, London, and Lagos—will find themselves partly
under water. But avoiding the most catastrophic impacts of warming
requires more than political will. To achieve real progress,
We urgently need new tools to accelerate the reduction we need to develop a much richer analytical understanding
and removal of GHG emissions—and to help citizens, of a complex system comprising many variables and feed-
cities, regions, countries, and businesses make plans to back loops. (See Exhibit 1.)
adapt to the inevitable impacts of warming. AI offers much
promise.
Exhibit 1 - Climate is an interlinked, multi-parameter system
Core climate characteristics
Water Emissions have varying
Changes in tempera- impacts on core
precipitation ture
climate characteristics,
Salinity
and changes in these
Ice cap
Human activities, such as melting Ocean processes can worsen
fossil fuel burning and land Clouds circulation the greenhouse
use changes, create significant upheaval gas effect.
volumes of greenhouse gas.
Climate change processes
Human activities
Carbon Average Gulf Stream
cycle temperature modification
disturbance rise
Global
imIn pc er re mas ee a i bn l e warming Abrupt
surfaces (enhanced) climate Europe
Greenhouse change cooling
effect
Urbanization
Land use Sea level
changes CO rise
2 NO Fluctuations in
2
Deforestation CH climate characteristics
4 Cyclones drive major
impacts—natural,
Food physical, and
Greenhouse Heat Loss of socioeconomic—at
traditional
Transport gas emissions waves lifestyles both local and global
Droughts scales.
Disease
spread
Fossil fuel
burning Disasters
Heating Biodiversity
losses
Agriculture
Casualties
Industry
Economic
losses
Famines
Major impacts
Source: Philippe Rekacewicz, Emmanuelle Bournay, UNEP/GRID-Arendal; BCG analysis.
BOSTON CONSULTING GROUP GOOGLE 6
Developing models is essential to understanding the rela- Estimating AI’s Potential Contribution
tionships among variables—and to anticipating the likely
impact of different strategies and choices. But modeling Based on our research and experience, the three broad
these complex interconnections on a local and global scale areas in which AI can accelerate climate progress are the
is a huge challenge. It requires assembling massive, longi- following:
tudinal, and real-time global data sets. Information is need-
ed on climate (for example, temperatures, ocean process- • Mitigation. Helping with both the reduction and remov-
es, and meteorological phenomena) and on human al of emissions—and with the underlying measurement
activities (for example, emissions, and land use changes). needed to size the challenge and track progress
And not all the necessary data is even available.
• Adaptation and Resilience. Aiding citizens, countries,
But understanding the complex systems that drive regions, cities, and businesses to prepare for and
climate-relevant outcomes is exactly the kind of challenge respond to the inevitable impacts of a warming planet
at which AI excels. By amalgamating and processing
massive data sets, AI can reveal elusive patterns and • Foundational Capabilities. Enabling climate action
valuable insights, facilitate scenario development and via improvements in climate modeling, climate eco-
prediction, accelerate the evaluation of multiple courses nomics, and climate education, as well as accelerating
of action, enable operational optimizations, and help breakthrough innovations that will open new horizons
monitor progress toward predefined goals. for climate action
Business leaders agree. In a 2022 BCG survey of senior
executives with leadership roles related to climate or AI
(see AI is Essential for Solving the Climate Crisis), 87%
viewed AI as a helpful unlock for climate issues. They saw
supporting emissions reduction as the top climate use
case for AI in their organizations, but expressed interest in
other applications as well. (See Exhibit 2.)
Exhibit 2 - Leaders believe AI can play a role in climate action, especially
in helping to reduce emissions
In which areas of climate-related advanced analytics and AI do you see the
greatest business value for your organization? (%)
Reducing emissions 61%
Measuring emissions 57%
Predicting hazards 44%
87%
of respondents say that Managing vulnerabilities 42%
AI is a helpful tool to
address climate change
Removing emissions 37%
Facilitating climate research,
28%
climate economics, and education
Mitigation Adaptation & Resilience Foundational Capabilities
Source: BCG Climate AI survey 2022. All respondents have decision-making authority over climate or AI topics at their organizations. Respondents
were permitted to give more than one answer.
7 ACCELERATING CLIMATE ACTION WITH AI
Regarding emissions reduction potential, a 2021 BCG And AI offers many foundational capabilities that sup-
study (see Reduce Carbon and Costs with the Power of AI) port both short-term and long-term climate action. For
estimates that currently proven AI-enabled use cases could example, it can support today’s climate research with
reduce emissions by 5% to 10% by 2030. If that potential is higher-fidelity climate change simulations. But it also has
fully realized, AI-driven applications would be responsible the potential to accelerate breakthrough innovations in
for achieving roughly between 10% and 20% of the IPCC’s domains such as physics, chemistry, biology, and material
2030 interim emissions-reduction target for the world to science that could “bend the curve” on climate progress.
achieve net zero by 2050.5 Similarly, a Microsoft/PwC study
looking at four sectors (agriculture, energy, transport, and All of our estimates are based on the current state of AI
water) estimates that AI has the potential to reduce global technology—and thus speak primarily to AI’s potential in
GHG emissions by 4%.6 Further, respondents in a Capgemi- currently proven applications. Today, we are in the early
ni survey of companies that had leveraged AI for climate stages of the adoption curve. Transforming potential to
action reported that their efforts to date had achieved GHG achievement will require that all organizations fully em-
reductions of between 11.3% and 14.3% depending on the brace AI as an essential enabler of their climate actions.
sector—and these executives believe that AI could reduce And it is important to note that our assessment does not
overall GHG emissions by 15.9% in the next three to five encompass major AI-driven disruptions and break-
years.7 throughs—for example, new materials for batteries, new
drought-resistant crops, novel carbon removal technolo-
On adaptation and resilience, AI can help cities forecast gies, and scalable approaches to nuclear fusion—that
their climate vulnerability, develop estimates of the cost of could unlock massive positive impact.
inaction, and model the impact of different climate inter-
ventions. These insights can aid them in identifying the The promise of AI is real. While we are already seeing
actions with the greatest benefit, generating private-sector benefits, we need to accelerate its contribution to
enthusiasm for funding investable projects, and securing planet-saving climate impact. The next chapter offers a
public and philanthropic support for essential, but deeper dive into the primary known climate-related use
non-bankable, adaptations. It also can help guide real-time cases for AI—and highlights some examples of how and
decision-making in agriculture—for example, increasing where AI is already making a positive difference.
crop production through intelligent irrigation systems—or
in fast-moving crises such as wildfires.
BOSTON CONSULTING GROUP GOOGLE 8
How AI Can Help Accelerate
Climate Action
A I has demonstrated the potential to enable and AI’s Role in Emissions Mitigation
catalyze climate progress in three broad areas:
taking emissions mitigation to the next level, shap- Getting smarter on reducing and removing emissions is
ing strategies for adaptation and resilience, and supporting essential. And AI is already delivering significant wins that
both climate research and reinforcing technologies. Some need to be scaled. Its contributions fall into two broad
AI applications are in early stages, some are being tested, areas: measurement and monitoring, and reduction and
and others are already being scaled. But all will need to be removal.8
embraced more broadly if we are to fulfill the promise of AI
to limit warming to less than 1.5°C. Measurement and Monitoring
Without reliable, clean, and independently verifiable data,
Exhibit 3 summarizes the most promising of the currently effective climate action is difficult. Countries and compa-
known AI use cases for climate. The rest of this chapter will nies need to know their baselines and track their progress,
offer more detail on each, along the way highlighting inspir- both at the macro level (“What are our total GHG emis-
ing examples of how AI is helping unlock and accelerate sions?”) and the micro level (“Which aspects of our opera-
climate progress. tions and broader supply chain are the big drivers? Are our
efforts at reduction or removal delivering the expected
results?”).
9 ACCELERATING CLIMATE ACTION WITH AI
Exhibit 3 - Key AI applications to accelerate climate progress
Mitigation Adaptation and Resilience
Measurement Reduction Hazard Vulnerability
& Monitoring & Removal Prediction Management
Macro-level measurement Enabling emissions reduction Building early warning systems Responding to crises
e.g., calculating carbon footprint e.g., integrating renewable energy e.g., predicting near-term e.g., monitoring drought and
at the country level into smart grids, optimizing extreme events such as flooding, wildfire spread
transportation of goods drought, and cyclones
Micro-level measurement Supporting nature-based & Projecting long-term trends Building resilient infrastructure
e.g., calculating carbon technological removal e.g., modeling localized sea-level & protecting biodiversity
footprints of individual products e.g., assessing natural carbon rise and drought frequency e.g., intelligent irrigation,
stocks monitoring of endangered species
Foundational Capabilities
Climate modeling
e.g., monitoring drought and wildfire spread
Climate economics
e.g., developing cost of inaction assessments
Education & behavioral change
e.g., developing recommendations for climate-friendly consumption
Innovation & breakthroughs
e.g., supporting research on fusion
Source: BCG, AI for the Planet Alliance.
Effective measurement and monitoring solutions leverage Solutions are emerging for micro-level measurement as
AI to process and analyze data from multiple sources such well. Google’s Environmental Insights Explorer (EIE) uses
as satellite data, weather data, sensors, and other heavy machine learning to offer city planners annual estimates
data sets—which can, for example, help an organization of emissions from buildings and transportation, tree
develop a baseline for its Scope 1, 2, and 3 emissions. AI canopy status, and emissions reduction opportunities
can also deliver insights, revealing patterns in emissions such as the potential for expanded rooftop solar. Houston,
and suggesting the best ways to prioritize abatement Texas, used EIE to perform a solar assessment and inform
efforts. the development of its 5 million MWh Solar Energy Target
Plan. Similarly, CO2 AI, a novel SaaS platform, enables
In the domain of macro-level measurement, Climate business leaders—together with their value chain
TRACE has been an early mover. This nonprofit offers free partners—to develop an accurate estimate of their
emissions data for more than 80,000 individual sources organizations’ Scope 1, 2, and 3 emissions down to the
and facilities around the globe, providing a data foundation product level. It also helps them to model and evaluate
to help organizations get started with mitigation plans. Its emissions reduction opportunities. (See the sidebar
data could, for example, assist countries seeking to transi- CO2 AI: Helping Business Ecosystems Reduce their
tion away from coal and other fossil-fuel based electricity Carbon Footprints.)
generation by pinpointing the largest emitters and reveal-
ing the mix of power sources by region. (See the sidebar
Climate TRACE: Providing Timely, Independent Emissions
Data—for Free.)
BOSTON CONSULTING GROUP GOOGLE 10
Climate TRACE: Providing Timely,
Independent Emissions Data—for Free
Making real progress on climate requires timely and accu- Supported by Google.org, among others, Climate TRACE
rate data on emissions to inform government policy and uses AI and machine learning to calculate GHG emissions
business action. But historically, emissions data has been on a global scale, with the goal of moving toward real-time
based on self-reporting, calculated using varying algo- precision. To achieve this, its model analyzes more than 59
rithms, and submitted years after the fact. Climate terabytes of data from over 300 satellites and more than
TRACE—a global coalition of nonprofits, tech startups, and 11,000 sensors to create highly granular emissions data for
researchers—offers a powerful, free, and independent over 80,000 sources globally. That number is expected to
alternative: the first comprehensive source-level global grow to more than 70 million sources by the end of 2023.
inventory of GHG emissions.
Application area: Macro-Level Measurement
Climate TRACE tracks global emissions
Source: Climate TRACE. Used with permission.
11 ACCELERATING CLIMATE ACTION WITH AI
CO2 AI: Helping Business Ecosystems
Reduce their Carbon Footprints
In order to make real progress on decarbonization, organi- In one example, a global health care company seeking to
zations need a more granular and actionable view of their reduce its Scope 3 emissions by 20% by 2030 embraced
carbon footprints, both across their Scope 1, 2, and 3 emis- CO2 AI. The platform enabled it to incorporate 50 times
sions and at the level of individual product areas. Until more factors into its calculations and to develop an
now, that kind of single source of truth has not been avail- activity based emissions baseline that was 20% more
able to help operations leaders understand emissions hot precise. And CO2 AI’s simulation and roadmapping tools
spots and explore potential solutions. enabled it to identify decarbonization opportunities that
would deliver 120% of its emission reduction target.
CO2 AI, an innovative SaaS platform, helps organizations
seamlessly map emissions across their value chains and Application area: Micro-Level Measurement
leverage those insights to drive climate action. AI plays a
central role in both assembling emissions data and match-
ing it to activities and products—and in simulating solu-
tions and building decarbonization roadmaps.
Measuring and managing emissions with CO2 AI
Source: CO2 AI. Used with permission.
BOSTON CONSULTING GROUP GOOGLE 12
Reduction and Removal In the realm of agriculture, the integration of AI tools with
AI has the potential to aid organizations in reducing and technologies such as drones can help farmers monitor
removing emissions in two ways: enabling emissions their crops in real time for better field management, thus
reduction and supporting nature-based and technology- enhancing agricultural productivity while minimizing GHG
based carbon removal. emissions. Moreover, AI-driven precision farming helps
empower farmers to make well-informed, data-driven
Enabling Emissions Reduction. AI can contribute to the decisions regarding farming practices, crop selection,
creation of more efficient and cleaner energy systems in irrigation, fertilizing, pest management, and harvesting.
multiple ways. It can, by consolidating information from This approach streamlines resource utilization and, if done
dozens of different organizations and grid components, purposefully, can minimize the environmental impact
provide insights on how to optimize electric grid opera- associated with farming practices. For example, Alphabet’s
tions—and support informed decision-making on grid project Mineral is using robotics, AI, and computer vision
planning. It can also help speed transition from fossil fuels to create a more sustainable food production system. It is
to alternative energy sources through better supply and developing perception-powered solutions with partners
demand forecasts that reduce the need for battery storage across the agriculture value chain—from grocery retailers
and standby power and enable more efficient real-time and enterprise farms to equipment manufacturers and
balancing of electric grids. crop protection companies—to develop a better under-
standing of the complex interactions of plants, their grow-
For example, Tapestry, an Alphabet project, is creating a ing environments, and farm management practices.12
single virtualized view of the electricity system with the
goal of lowering emissions, minimizing outages, shortening Another interesting use case involves using AI to reduce
interconnection queues, and integrating more renewables contrails. Contrails, the white clouds that sometimes form
into the grid. AI is at the heart of its computational and behind airplanes when they fly, are responsible for about
simulation tools. Relatedly, on the subject of renewables, 35% of the aviation sectors’ global warming impact. AI
France’s Engie has partnered with Google Cloud to develop solutions developed by Google Research in partnership
and pilot an AI-powered tool that can provide grid opera- with Breakthrough Energy have enabled airline pilots in
tors with more accurate real-time forecasts of wind energy trial studies to reduce contrails by up to 54%.13 (See the
supplies.9 sidebar The Contrails Impact Task Force: Addressing Avia-
tion’s Other Contribution to Warming.)
In Africa and India, Husk Power Systems provides “pay-as-
you-go” 100% renewable power to off-grid and weak-grid Supporting Nature-Based and Technology-Based
communities that is 30% cheaper than the alternative: Removal. According to the IPCC, limiting warming to
diesel generation. Husk estimates that its AI model en- 1.5°C by 2100 will require an extensive deployment of CO2
ables it to predict user demand with 80% accuracy across removal measures, of which there are two broad types:
its microgrids, thereby improving capacity utilization, re- nature-based removal in which carbon is removed by and
ducing costs, delivering lower prices, and guiding capital stored in natural sinks such as forests, algae, and wetlands,
investment in additional capacity. and technology-based removal via approaches such as
direct air capture (DAC).14 AI can play a supporting role in
Moreover, AI-driven insights can also enable people and both types of removal.
organizations to make smarter decisions that decrease
emissions. For instance, as a result of using AI to improve In nature-based removal, AI-based solutions can help
demand forecasting, manufacturers can avoid both over- quantify and verify the level of carbon sequestration
production and the emissions those unsold goods would achieved in an ecosystem, enabling public and private
produce. Similarly, AI-optimized transportation can reduce sector leaders to make informed decisions regarding the
emissions by identifying and directing drivers to the most deployment of natural solutions, including land manage-
efficient routes. As of September 2023, Google Maps’ ment and reforestation efforts. One actor in this space is
|
282 | bcg | ai-in-india-a-strategic-necessity.pdf | AI in India -
A Strategic Necessity
A pragmatic playbook for Indian
organization to leapfrog on AI maturity
July 2023
Boston Consulting Group partners with leaders The Brij Disa Centre for Data Science and Artificial
in business and society to tackle their most Intelligence (CDSA) is a research centre at the
important challenges and capture their greatest Indian Institute of Management Ahmedabad
opportunities. BCG was the pioneer in business (IIMA). It offers a platform for faculty, scholars
strategy when it was founded in 1963. Today, and practitioners to conduct cutting-edge research
we work closely with clients to embrace a on data analytics and artificial intelligence,
transformational approach aimed at benefiting all providing solutions for businesses, governments
stakeholders—empowering organizations to grow, and policymakers. Besides generating action-
build sustainable competitive advantage, and oriented insights, CDSA conducts seminars,
drive positive societal impact. workshops, and conferences to disseminate
knowledge on artificial intelligence and analytics
Our diverse, global teams bring deep industry and
to a wider audience across the world.
functional expertise and a range of perspectives
that question the status quo and spark change. The Indian Institute of Management Ahmedabad
BCG delivers solutions through leading-edge (IIMA) is a premier, global management Institute
management consulting, technology and design, that is at the forefront of promoting excellence in
and corporate and digital ventures. We work in a the field of management education. Over the 60
uniquely collaborative model across the firm and years of its existence, it has been acknowledged
throughout all levels of the client organization, for its exemplary contributions to scholarship,
fueled by the goal of helping our clients thrive and practice and policy through its distinctive teaching,
enabling them to make the world a better place. high-quality research, nurturing future leaders,
supporting industry, government, social enterprise
and creating a progressive impact on society.
Contents
04 | Foreward 39 | Strategic Planning
in the Era of AI
06 | Executive 42 | The Road Ahead
Summary Adoption Process
of Analytics in
Organizations
09 | The Evolving Global 45 | Responsible AI: A
Perspectives on AI foundational pillar
in India’s growth
16 | The AI Maturity 48 | India’s AI Policy:
Survey The Current
Position and the
Way Forward
Maturity1, because their AI-derived benefits are marginal high AI Maturity, at par with global benchmarks. However,
at best. They have not fully harnessed AI to redesign their the survey also finds that around three out of four compa-
offerings or processes to achieve a sustainable competitive nies in CG and IG are classified as ‘Laggards’ in AI Matu-
advantage or higher margins. Indeed, our findings rity. Given that AI adoption will be a driver of competitive-
indicate that the margins of AI ‘Maturity Leaders’ ness, the low AI maturity of majority of CG and IG
are 3-5 percentage points above their Laggard peers. companies may constrain their global ambitions
if the issue is not addressed soon.
We find that AI delivers its best results when AI-driven
transformation is a strategic priority. Therefore, this study The report discusses the roadmap for improving AI
is designed to inform senior decision-makers in Indian maturity focusing on the organisation’s current maturity
organizations on the state of AI in India, and the way level and industry. Specifically, it defines a roadmap for
forward. It is based on a structured survey and discussion laggards in AI adoption to kickstart their AI transforma-
with CXO-level leaders in Technology, Data analytics, tion, build up their AI Maturity and ultimately achieve
Digital Transformation and Business Heads from 130 success in AI adoption. Further along the spectrum,
organizations across Banking, Financial Services and it also offers roadmaps for players with mid-level AI
Insurance (BFSI), Consumer Goods (CG) and Industrial Maturity (the so-called ‘Steady Followers’ and
Goods (IG). ‘Leapfroggers’) to graduate to global best-in-class AI
Maturity levels. For ‘Leaders’ the report focusses on
The study brings the latest research on impact of AI on the next frontiers of AI excellence to be conquered.
organisations along with the best on-ground AI-led trans-
formation experience. It unearths several encouraging This report is a joint initiative of IIM Ahmedabad’s Brij
findings—for instance, a significant number of Banking Disa Centre for Data Science and Artificial Intelligence
sector participants, and a smaller number of corporates and BCG X, the AI and Digital Transformation unit of BCG.
in Consumer Goods (CG) and Industrial Goods (IG) have
Foreword
A
rtificial Intelligence (AI) has evolved remarkably zational productivity and efficiency, changing the competi-
since its genesis in the 1950s. Today, it permeates tive landscape. The success of a country’s businesses in
every aspect of our daily lives—from the phones adopting AI will be an increasingly crucial determinant
in our hands, to the products on our supermarket shelves; of its competitiveness.
from selecting the route for our commute, to suggesting
the next movie on our entertainment platforms. It is In light of the high stakes involved, this study aims to
equally pervasive at the macro level, assisting in tasks gauge the status of AI adoption in Indian organizations,
as varied as studying the impact of weather on crops, and their success in translating it into business perfor-
optimizing supply chain risk and determining the best mance. To this end, it examines the AI Maturity of these
drug molecule for diseases. organizations. Many have already dipped their toes in
AI—for instance, the larger players in most sectors apply
Sentient AI robots may be a while away, but AI today has machine learning algorithms to make predictions on select
the potential to transform entire industries, by redefining business metrics. However, such behavior by itself does not
products, services and reshaping supply chains. Successful imply high AI maturity for such organisations. In fact, some
AI adoption is already having a profound impact on organi- of these organisations may still end up as Laggards in AI
1. We have classified companies into four groups based on their AI Maturity – Leaders (most mature), Steady Followers (less mature but steadily
catching up), Leapfroggers (less mature but recently made rapid strides) and Laggards (least mature). See pg 16 for more.
4 AI IN INDIA - A STRATEGY NECESSITY BOSTON CONSULTING GROUP + INDIAN INSTITUTE OF MANAGEMENT AHMEDABAD 5
data maturity. It draws on the concepts of the BCG AI For legacy companies, size and scale alone offers little
Iceberg and academic literature, which assert that a protection against deft small competitors who are master-
successful implementation of AI is one that impacts ing AI usage. We are seeing a rapid rise in mid- and small-
the revenues, margins and sustainability of the business. size players (as many as 16% of the companies studied)
Among the key contributors of organizational success which are well-positioned to capture greater market share.
from AI adoption, algorithms drive approximately 10% Unencumbered by legacy issues, they have thrown down
of the success, while data and technology infrastructure the gauntlet, not only to larger more established players
adds a further 20%. The remaining 70% hinges on people, in their industry but also to AI maturity Leaders.
processes and business transformation.
Leaders and Leapfroggers tend to adopt a ‘use case-first’
This study particularly draws on the views of Chief Data approach to AI adoption. They take time to identify use
Officers (CDO), Chief Analytics Officers (CAO), Chief Tech- cases which will have a palpable impact on the balance
nology Officers (CTO) and Chief Digital Transformation sheet. They then deploy technology, people and processes
Officers from leading organizations across the BFSI, CG to support those use cases. Laggards, on the other hand,
and IG sectors. It also draws on interviews with Business take a technology-first approach. They often end up with
Unit heads to gauge their views on the impact of AI on white elephant technologies which have limited impact on
business outcomes. The result is a detailed, calibrated business outcomes. Steady Followers lie in between these
understanding of a) these organizations’ plans with respect two groups. They tend to choose use cases that are tentative
to AI; b) the investments and measures taken to operation- and small-scale, and thus rarely transform the organization
alize those plans; c) the changes underway across technol- to the extent required to let AI play out at scale.
ogy, organization, people and procedures; and d) the
observed outcome. The country’s ecosystem plays a vital role in this endeavor,
as both a supplier and enabler of essential talent. If the
The study reveals that select Indian BFSI companies top 500 listed companies in India made AI a strategic prior-
(particularly banks and new-age NBFCs) have very high AI ity, they would need at least 25,000 to 30,000 advanced
Maturity, on par with global frontrunners. We have divided practitioners of AI-ML in the next 3-5 years. This covers
companies into four groups based on their maturity level— the entire gamut of AI professionals, from data scientists
Leaders, Steady Followers, Leapfroggers and Laggards. 11% and data engineers to enterprise architects. But it does
of companies in the set were adjudged Leaders. Their lead- not include managerial and leadership talent, nor the
ership position is facing a stiff challenge from the Leapfrog- workforce in AI vendor ecosystems and support infrastruc-
gers, who make up 9% of the companies. Leapfroggers ture which must enable these AI initiatives. Even with
started their AI-driven transformation journey late but India’s engineering and science talent, the quest for higher
have improved sharply in AI Maturity in the last three years, AI Maturity requires significant training and upskilling
converging with the Leaders on most aspects of AI Maturity. across data engineering, enterprise architecture, product
Executive Summary management, design thinking, domain knowledge, Agile
However, the concern is that 2/3rds of the companies in working and management of digital organizations. Finding
the set remain Laggards. These are companies with some and training talent in requisite numbers will be a critical
exposure and investment in AI in their Technology, Data determinant of whether India gains competitiveness in AI.
and Analytical capabilities. But AI is not a strategic priority
for them. Three out of four companies in Consumer Goods Research shows that AI investments augmenting end-user
and Industrial Goods are Laggards by this assessment. value and topline growth could drive significant economic
Just 5% of IG and CG organizations surveyed are AI Maturi- and wage expansion. The opportunity is India’s for the
T
he age of AI is upon us. As with previous General must invest in significant upskilling of mid- and senior-level ty leaders. The AI laggardness could have severe implica- taking—the challenge is now to turn the enormous poten-
Purpose Technologies like the steam engine and management on the business aspects of AI, digital trans- tions for the competitiveness of Indian manufacturing if it tial of AI into reality.
the internal combustion engine, or more recently, formation, ‘Agile’ ways of working and more. This study remains unaddressed.
computers and the internet, AI will have a transformative estimates that just the top 500 Indian companies would
impact on economies, societies and civilization at large. require at least one million hours of training.
In India alone, successful adoption of AI could add up to
1.4 percentage points annually to real GDP growth. From Companies cannot assume that benefits of AI will accrue
the perspective of corporates, successful adoption of AI is to them in due course. Companies have a choice to priori-
expected to add over a five year period, INR 1.5-2.5 trillion tise AI and adopt it or perish—and the nature of this tech-
in incremental pre-tax profit for the top 500 Indian compa- nology is such that either scenario would come about very
nies alone. quickly. The key to success in AI is achieving an advanced
level of AI maturity—the core theme of this report.
Investments into AI could deliver extraordinary returns
but success hinges on deploying AI at scale, as opposed AI maturity captures the overall ability of a company to
to restrictive incrementalism. Senior leaders must develop leverage AI to drive its strategic objectives and enhance
a more granular and precise understanding of the implica- its financial and operational performance. AI maturity
tions of AI for their business. For starters, organizations goes well beyond the existing measures of analytical or
6 AI IN INDIA - A STRATEGY NECESSITY BOSTON CONSULTING GROUP + INDIAN INSTITUTE OF MANAGEMENT AHMEDABAD 7
Key Highlights of the Report
AI benefit to Indian Economy AI benefit to Indian companies
Successful AI adoption by Indian businesses Successful AI adoption can add INR 1.5-2.5 trillion
could consistently add ~1.4 percentage point to real in incremental pre-tax profit, for the top 500 Indian
GDP growth. companies, over following 5 years.
AI Maturity: key to successful AI adoption India has exemplars in AI Maturity
Measures the ability of a company to leverage AI to Select Indian BFSI companies (particularly banks and
drive its strategic objectives and enhance its financial new-age NBFCs) have very high AI Maturity, on par with
and operational performance. global frontrunners.
AI Maturity level worrisome for most Future competitiveness may be impacted
The study finds overall 2 out of 3 Indian companies are In consumer goods and industrial goods sectors, 3 out
laggards in terms of AI adoption and maturity. of 4 companies are AI maturity laggards. Companies
The Evolving Global Perspective on AI
who strive for global competitiveness need to address
their low AI maturity quickly.
A
I may have started out as a research concept eight expensive computational setups. This period may be
decades ago, but it has since grown profoundly in its called the Pre-Democratization era of AI (PD-AI).
Use-case first vs technology first Other differentiators between leaders and laggards
scope and power, moving out of the laboratory into
Even laggards invest in data and technology. However Leaders tend to be aware that for AI adoption success:
everyday life. The last three decades have seen specialized Since then, three factors have lowered the entry barriers
laggards take a technology first approach and oen the algorithms drive 10% of the success, technology and
usage of increasing intensity. Today, its potential applications in AI adoption. Firstly, rapid fall in cost of data storage and
use-cases are not detailed out. Leaders first prioritise data infrastructure drive another 20%. 70% of the
cover every area of human activity, and no company can computational power. Secondly, cost effective cloud-based
the use-cases and then decide the optimal choice of success is driven by people, organisations and processes.
afford to ignore it. We have identified five factors that explain data and computational architecture which converts high
technology, algorithms, people and processes to make
the increasing pervasiveness of AI in recent decades. upfront technology CapEx to more manageable and
the use-case successful.
scalable OpEx. Lastly, coding platforms with low code or
The democratization of AI: Originally funded for military no code environment allowing companies to get started
purposes, AI was subsequently nurtured in research labs on basic AI use-cases. The last 5-7 years have thus been
and universities. Since the early 2000s, it has increasingly a period of Democratization of AI (D-AI). It is likely that
been deployed in industrial and real-world applications. this democratization is in its early phase and
But till as recently as the last decade, AI implementation improvements in technology will further reduce costs and
and adoption was limited to organizations with advanced increase AI deployment—but this alone will not ensure
Requirement of AI specialist Massive managerial upskilling required resources, high investment in data infrastructure and success in AI.
Just the top 500 Indian companies they would need at Just the top 500 Indian companies would require at
least 25,000 to 30,000 advanced practitioners of AIML least 1 Million hours of training in upskilling mid and
in the next 3-5 years. senior level management on the business aspects of AI,
digital transformation, Agile ways of working and more.
8 AI IN INDIA - A STRATEGY NECESSITY BOSTON CONSULTING GROUP + INDIAN INSTITUTE OF MANAGEMENT AHMEDABAD 9
Data Source and Analytics in the Pre-Democratization era of AI
Generative AI- A brief
PRE-DEMOCRATIZATION AI
B2C B2B
What is Generative AI
from patterns and structures present in large datasets,
Auto, Consumer Durable:
Generative AI refers to a subset of artificial intelligence enabling it to generate novel outputs that closely resemble
Data source: Sales & Distribution data; Industrial & Engineering Goods:
that focuses on creating new content rather than solely the original data. This technology draws inspiration from
Production Data from ERP Data source: Order details & production
Goods analyzing or predicting existing data. This new content fields such as deep learning, neural networks, and
& Supply Chain data from ERP & Supply Chain
ranges from music to art to text to software code. At its probabilistic modeling to mimic human creativity in
Analytics: MIS, Trend Analysis, Root Cause Basic descriptive analytics, MIS; Trend core, generative AI operates on the principle of learning a machine-driven manner.
Analysis, Linear Extraploation, Regres- Analytics of production data, ERP
sions for prediction
BFSI, Healthcare: Data Source: customer Engineering & Construction Services;
A View of Gen AI Apps
level data Data Source: Logistics, Cargo Shipping,
Services
Order data & service delivery level data
Analysis: Descriptive and trend analysis;
Customer specific and transaction specific Basic descriptive analytics, MIS; Trend
predictive models Analytics of production data, ERP
The data deluge: Data analysis has always facilitated AI deployable at scale: AI is transforming the way
decision-making at the transactional, operational and business is conducted across industries. Companies are CHATGPT Lexica Mage.space Jasper
strategic level. In the pre-AI era, companies with rigor investing heavily in AI solutions in the hope of substantial Natural text generation Browse AI generated Prompt-based Creative writing
on decision quality invested heavily in extensive setups returns. The growing prevalence of industrial robots, images and the prompts image generation (ads/ blog articles,
for data analysis. These efforts focused on descriptive computerized production equipment, marketing chatbots that have been used product descriptions)
analysis to understand the drivers and causality of past and machine learning investment algorithms is constantly
performance, with limited predictive analysis to gauge expanding the range of tasks that machines can perform.
future trends and transactional events.
AI’s inflection point- Generative AI: The sheer public
But the D-AI period is driven by cheap and plentiful excitement generated by a Generative AI app- ChatGPT
computing power, enabling easier execution of AI at its launch and afterwards is unprecedented. ChatGPT
algorithms. The increasing ability to capture and store reached 1 million users in just 5 days after its launch.
large amounts of data from communication devices, ERP In comparion, Instagram and Spotify took an estimated
Synthesia.io Midjourney Runway KAEDIM
systems or satellite data has provided the ideal setup for 75 days and 150 days respectively. If one goes by Google
Convert text to Prompt-based Image/video editing Convert 2D images to
such algorithms to run. This has opened up new use cases search count as a measure of interest, interest in ChatGPT
speaking avatar image generation and enhancement 3D objects
and created an enabling environment for new business is 7 to 8 times higher than the peak interest in Metaverse.
models on the lines of ‘X-as-a-service’, where X assumes The hype around currently available generative AI
various forms—banking, payment, logistics, manufacturing applications could be due to their ease of access and
and infrastructure management—limited only by market simple yet intuitive user interface. By formulating the
size and cost-effective execution. appropriate question in English, one can effortlessly and
quickly access information derived from a vast dataset,
Search for sustainability during uncertain times: in a user-friendly format.
Stakeholders are increasingly demanding that businesses
deliver sustainable profitability with social responsibility— Organisations in industries ranging from BFSI, Healthcare,
even in a period of economic volatility and uncertainty. Consumer Goods and Services, Technology to name a few,
This has added to the challenge of operational planning are finding powerful use-cases based on Gen AI.
and strategic decision-making. AI is well-placed to help
businesses balance these difficult imperatives.
10 AI IN INDIA - A STRATEGY NECESSITY BOSTON CONSULTING GROUP + INDIAN INSTITUTE OF MANAGEMENT AHMEDABAD 11
India’s current AI strides
The global AI market is estimated to reach US$450 billion Management services to global clients. Today, the IT-BPM
in 2022, growing at a rate of more than 20%.3 In India, AI industry is India’s largest private sector employer, account-
expenditure reached US$665 million in 2018 and is expect- ing for ~11% of the urban workforce.
ed to reach US$11.78 billion by 2025, with a CAGR of 39%
from 2019-2025.4 With AI, a growing body of evidence suggests that the
automation of repetitive tasks has led to the disappear-
As with most new technologies, there has been concern ance of middle-skilled jobs and increased wage inequality.
about the impact of AI on labor markets. While these On the other hand, there is also growing demand for labor
concerns are understandable, large scale job losses due trained in advanced technology and adept in socio-behav-
to technological innovation can be averted. A case in point ioral skills. Experts suggest that emerging technologies
is India’s thriving Information Technology sector and the may increase the productivity of existing jobs as well as
opportunities it has created. As recently as the 1990s, there create new roles which are difficult to envisage today.
were fears of computers replacing humans – yet the sector These new roles may require a combination of skills such
ended up creating large numbers of new jobs. India has as higher technological acumen, better empathy, people
become a major offshoring hub for the global software connect and critical thinking.
industry providing Business Process Outsourcing and
Exhibit 2 - Global Research and Patenting in AI by Country
1,000,000
100,000
10,000
1,000
100
10
1
3. IDC: Worldwide Semiannual Artificial Intelligence Tracker
4. https://www.ibef.org/download/AI-Revolution.pdf
Source: Exhibit 2 - Emerging Technology Observatory’s Country Activity Tracker: Artificial Intelligence: https://cat.eto.tech/
12 AI IN INDIA - A STRATEGY NECESSITY BOSTON CONSULTING GROUP + INDIAN INSTITUTE OF MANAGEMENT AHMEDABAD 13
selcitrA
latoT
)elacs
laitnenopxe(
snoitacilppA
tnetaP
Exhibit X
Exhibit 1 - How are companies leveraging GEN AI
Model topology Pre-trained model Organisation Specific
!
(Outcome of academic (E.g., GPT-3 Davinci) use cases supported
& industrial research)
Question answering
Sentiment analysis
Training data
! Training Fine-tuning Information extraction
(Text, images,
Publicly Little amounts
code & other data)
available (MBs) of specific
information from domain data or Fine-tuning Image labeling
huge dataset. inhouse data-low data
High computa- resource and time (Text, images, Object recognition
tional power, requirement code & other data)
time
Content creation
On average, 1 in 3 respondents in various roles predict Ethical Considerations and Future Prospects
a 25-50% gain in productivity from using these tools.2
The tasks which are currently getting revolutionised by As generative AI continues to evolve and permeate various
GEN AI are: aspects of society, it is crucial to address the ethical consid-
erations surrounding its use. Questions arise regarding the 1,400,000
279,962
• Writing blogs/ posts/ mails ownership and authenticity of generated content, the
potential for misuse or manipulation, and the impact 1,200,000 77,635
• Marketing material on employment and creativity. Striking a balance between
16,499
innovation and responsible implementation is essential
1,000,000
• Structured information extraction for harnessing the full potential of generative AI.
2,184 4,160 1,113 3,282 3,082
• Client outreach 800,000 983
• Writing code 600,000 176
• Answering customer queries
400,000
• Project management
200,000
0
China US India UK Germany Japan France Australia Canada Italy
Articles PatentApplications
2. https://www.sortlist.com/datahub/reports/chat-gpt-statistics/
Source: Exhibit 1 - “On the Opportunities and Risks of Foundation Models”, Center for Research on Foundation Models, arXiv, 2021; BCG analysis
Exhibit 4 - Private investment in AI and R&D
AI Investments: Companies and Amount
• Government Intervention: Arguably the biggest and government could nudge companies through suitable
most successful digital pioneer in India is the Govern- tax incentives for AI research and innovation. It should
ment of India. Its innovations at scale include Aadhaar strive to upgrade the curriculum and boost the resource
(universal biometric ID), the Unified Payments Inter- of India’s top institutions to focus on advanced technol-
face and the Open Network for Digital Commerce. The ogy education, especially in AI.
Source: “Emerging Technology Observatory’s Country Activity Tracker: Artificial Intelligence: https://cat.eto.tech/
14 AI IN INDIA - A STRATEGY NECESSITY BOSTON CONSULTING GROUP + INDIAN INSTITUTE OF MANAGEMENT AHMEDABAD 15
DSU
snoillim
ni
stnemtsevni
IA
tnemtsevnI
IA
htiw
seinapmoC
fo
rebmuN
Ultimately, India’s success in leveraging AI will be shaped • R&D and Intellectual Property: While India ranks
by four key factors: in the global top 10 for AI research and patents, the
associated value being captured is relatively miniscule.
• Nurturing AI talent: India only has around 4.5%5 of the The pay-off in terms of patents, products and profits
world’s AI professionals, and the talent crunch will get remains low relative to the volume of research conduct- 600,000 8000
more acute. 76% of India’s data talent is currently hired ed. The reasons for this are manifold. Research in latest 6,773
by the IT Services industry. However, companies are technologies is often limited to incrementalism in most 7000
500,000
struggling to find AI talent with the requisite business firms. Additionally, a fledgling collaboration between
and sector understanding. As a result, direct hire of AI industry-academia limits monetizable research and IP 6000
talent remains low despite high demand. A NASSCOM creation. This scenario further discourages advanced tal- 400,000
5000
report6 also projects that the demand-supply gap for dig- ent development. Currently, less than 3% of graduates
ital technology talent will grow 3.5x+ by 2026 to 1.4-1.8 pursue a PhD in the field.
300,000 4000 million. The current study estimates that core AI talent-
data scientist, data engineer, enterprise architect would • Investments in AI: The last decade has seen a rapid
3000
be 15% to 20% of the headline number. Further just the uptick in the number of Indian startups. Indian orga- 200,000
top 500 corporates (listed corporates by revenue) would nizations have also accelerated their adoption of da- 1,422 2000
1,198
need at least 25,000 to 30,000 advanced practitioners of ta-driven use cases. However, when it comes to private
100,000 711 642 AIML in the next 3-5 years. To handle AI driven transfor- investment in AI, the US and China lead the rest of the 557 418 294 186 127 1000
mations, the existing senior and middle managements world by a huge margin.
of these 500 companies would require a minimum a 0 0
million hours of training! US China UK Canada India Germany France Japan Australia Italy
Disclosed Investments Estimated total investments Companies
Exhibit 3 - Published AI articles by category and application (India)
India: Published AI articles by Category
Computer 1395
SpeechRecognition 4698
Simulation 656
RealTimeComputing 3540
PatternRecognition 39446
OtherAI
NaturalLanguageProcessing
MathematicsEducation 256
MathematicalOptimization
MachineLearning 15281
Lingusitics 575
Informationretreival 4690
Human-ComputerInteraction
DataScience
DataMining
ControlTheory
ControlEngineering 1118
ComputerVision
CognitivePsychology 542
Algorithm 6045
0 5000 10000 15000 20000 25000 30000 35000 40000 45000
Source: Emerging Technology Observatory’s Country Activity Tracker: Artificial Intelligence: https://cat.eto.tech/
Note: Chart shows the number of AI articles published by authors from the country. Author countries are inferred from where their institutions
are located.
5. OECD.AI (2023), visualisations powered by JSI using data from LinkedIn, accessed on 06/2/2023
6. https://community.nasscom.in/communities/emerging-tech/indias-tech-industry-talent-demand-supply-analysis
seirogetaC
elcitrA
33351
7692
2551
3739
3681
8003
2942
14024
AI Maturity goes beyond the existing measures of benefits will accrue only through fundamental changes
analytical and data maturity. Instead, it draws on the to the policy, operating processes, and behavior of the
concept of the BCG AI Iceberg and academic literature. entire loan origination setup to enable digital workflow
The Iceberg asserts that a successful implementation and data driven decision making. In the absence of such
of AI is one that impacts the revenues and margins of transformations, even organizations with a bespoke
the business. Approximately 10% of this success can AI model will be deemed to have low AI Maturity.
be attributed to algorithms, and another 20% may be
attributed to enabling data and technology capability. On the other hand, consider a Consumer Goods company
But the bulk of the success, that is 70%, hinges on that has begun to embrace AI by adopting a reasonable
people, processes and business transformation. (though not cutting-edge) demand forecasting model for
its products. The organization has expedited the adoption
This means that simply acquiring technology or using the of this model at the district level to ensure consistency
latest Machine Learning tools would not confer a strategic in supplies. It closely tracks any divergence of the actual
benefit by itself. For instance, a bank may have a well-built demand from predicted demand. These gaps are then
ML-driven risk score. But this will not transform the loan fed back into the system to improve the predictions.
acquisition and underwriting process if the model scores The front-end decision makers are trained to identify
are regularly overwritten by subjective judgment and the 5-10% instances where they will overwrite the model
ineffective credit policies. results, but these human interventions are also tracked
for quality of decision making. An organization like this is
Such an implementation of the risk model may offer exhibiting higher AI Maturity than the financial institution
limited benefits to the organization. However, large-scale in the previous example.
Exhibit 5 - How we measure AI Maturity—the seven components
VISION
STRATEGY
The AI Maturity Survey
ANALYTICS
Industries and companies across the board are Understanding AI Maturity
increasingly looking at AI to deliver a long-term
competitive advantage. AI is seen as critical not just AI Maturity measures the overall capability of a company
ETHICS & GOVERNANCE
to their growth, but to their very survival in the medium to leverage AI to drive the strategic objectives and
to long run. With the stakes this high, it is vital for the ensuing operational performance of the company.
companies to assess their AI capabilities and build These objectives may include (inter alia) sustained
a robust plan to harness the value of AI. growth in revenues and consistent margins to enhance
shareholder value. In other words, AI Mat |
283 | bcg | gen-ai-increases-productivity-and-expands-capabilities.pdf | GenAI Doesn’t Just Increase
Productivity. It Expands
Capabilities.
SEPTEMBER 05, 2024
By Daniel Sack, Lisa Krayer, Emma Wiles, Mohamed Abbadi, Urvi
Awasthi, Ryan Kennedy, Cristián Arnolds, and François Candelon
READING TIME: 12 MIN
This is the second major field experiment led by the BCG Henderson Institute designed to help business
leaders understand how humans and GenAI should collaborate in the workplace. Our previous study
assessed the value created—and destroyed—by GenAI when used by workers for tasks they had the
© 2024 Boston Consulting Group 1
capabilities to complete on their own. Our latest experiment tests how workers can use GenAI to complete
tasks that are beyond their current capabilities.
A new type of knowledge worker is entering the global talent pool. This employee, augmented with
generative AI, can write code faster, create personalized marketing content with a single prompt, and
summarize hundreds of documents in seconds.
These are impressive productivity gains. But as the nature of many jobs and the skills required to do
them evolve, workers will need to expand their current capabilities. Can GenAI be a solution there as
well?
Based on the results of a new experiment conducted by the BCG Henderson Institute and scholars
from Boston University and OpenAI’s Economic Impacts research team, the answer is an
unequivocal yes. We’ve now found that it’s possible for employees who didn’t have the full know-how
to perform a particular task yesterday to use GenAI to complete the same task today.
METHODOLOGY
Our research involved a carefully structured experimental design to evaluate the
impact of generative AI on the ability of nontechnical knowledge workers to perform
technical, data-science tasks.
In total, 480 BCG consultants and 44 BCG data-scientist volunteers completed this
controlled study. The study participants were general consultants, for whom data-
science expertise is not typically required. This expertise exists among the data
scientists of BCG X who also volunteered to support the study by establishing
benchmarks. The performance of general consultants was evaluated by comparing
their output to that of BCG data scientists who completed the same tasks.
General consultants were randomly assigned to either a GenAI-augmented group,
which received interactive training on using Enterprise ChatGPT-4 with the Advanced
Data Analysis Feature for data science tasks, or a control group, which was asked not
to use GenAI and received interactive training on traditional resources like Stack
Overflow. The tasks assigned to participants included coding, statistical
understanding, and predictive modeling, all of which required skills that are typically
outside the expertise of nontechnical workers but within the day-to-day expertise of
BCG’s data scientists.
© 2024 Boston Consulting Group 2
Data collection was carried out in four phases: a pre-experiment survey to assess
baseline skills and attitudes, a tailored training session for each group, the
completion of two out of three randomly assigned data-science tasks, and a post-
experiment survey to measure knowledge retention without the use of AI tools.
The tasks were designed by BCG data scientists to ensure that they were challenging
enough that AI could not solve them independently. Analysis focused on comparing
the performance of the treatment and control groups against the benchmarks set by
BCG data scientists, examining the completion rates, time taken, and correctness of
the responses.
Similar to our first GenAI study, we “put our feet to the fire”—with a goal of deeply
understanding GenAI’s impact on ways of working.
With that in mind, leaders should embrace GenAI not only as a tool for increasing productivity, but as
a technology that equips the workforce to meet the changing job demands of today, tomorrow, and
beyond. They should consider generative AI an exoskeleton: a tool that empowers workers to
perform better, and do more, than either the human or GenAI can on their own.
Of course, there are important caveats—for example, employees may not have the requisite
knowledge to check their work, and therefore may not know when the tool has gotten it wrong. Or
they may become less attentive in situations where they should be more discriminating.
But leaders who effectively manage the risks can reap significant rewards. The ability to rapidly take
on new types of work with GenAI—particularly tasks that traditionally require niche skills that are
harder to find, such as data science—can be a game-changer for individuals and companies alike.
How GenAI Can Equip Knowledge Workers
In the previous experiment, we measured performance on tasks that were within the realm of the
1
participants’ capabilities. (See top row of Exhibit 1.) For tasks where GenAI is highly capable, we
found that augmented workers perform significantly better than humans working without the
technology. However, when the technology is not capable of performing the task at expert level,
humans tend to over-rely on GenAI and perform worse than if they had completed the task on their
own.
© 2024 Boston Consulting Group 3
But what happens when, instead of using GenAI to improve performance within their current skillset,
people use GenAI to complete tasks that are outside their own capabilities? Does being augmented
with GenAI expand the breadth of tasks people can perform?
For our latest experiment, more than 480 BCG consultants performed three short tasks that mimic a
common data-science pipeline: writing Python code to merge and clean two data sets; building a
predictive model for sports investing using analytics best practices (e.g. machine learning); and
validating and correcting statistical analysis outputs generated by ChatGPT and applying statistical
2
metrics to determine if reported findings were meaningful.
While these tasks don’t capture the entirety of advanced data scientists’ workload, they are
sufficiently representative. They were designed to present a significant challenge for any consultant
3
and could not be fully automated by the GenAI tool.
To help evaluate the performance impact of GenAI, only half of the participants were given access to
the GenAI tool, and we compared their results to those of 44 data scientists who worked without the
assistance of GenAI. When we dive deeper into the results, three critical findings emerge.
The Immediate Aptitude-Expansion Effect
When using GenAI, the consultants in our study were able to instantly expand their aptitude for new
tasks. Even when they had no experience in coding or statistics, consultants with access to GenAI
were able to write code, appropriately apply machine learning models, and correct erroneous
4
statistical processes. (See Exhibit 2.)
© 2024 Boston Consulting Group 4
We observed the biggest aptitude-expansion effect for coding, a task at which GenAI is highly adept.
Participants were asked to write code that would clean two sales data sets by correcting missing or
invalid data points, merging the data sets, and filtering to identify the top five customers in a
specified month.
Participants who used GenAI achieved an average score equivalent to 86% of the benchmark set by
data scientists. This is a 49-percentage-point improvement over participants not using GenAI. The
GenAI-augmented group also finished the task roughly 10% faster than the data scientists.
Even those consultants who had never written code before reached 84% of the data scientists’
benchmark when using GenAI. One participant who had no coding experience told us: “I feel that I’ve
become a coder now and I don’t know how to code! Yet, I can reach an outcome that I wouldn’t have
been able to otherwise.” Those working without GenAI, on the other hand, oen did not get much
further than opening the files and cleaning up the first “messy” data fields; they achieved just 29% of
the data-scientist benchmark.
It’s important to note that most consultants are expected to know the basics of data cleaning and
oen perform data-cleaning tasks using no-code tools such as Alteryx. Therefore, while they did not
have experience doing the coding task in Python, they knew what to expect from a correct output.
This is critical for any GenAI-augmented worker—if they don’t have enough knowledge to supervise
the output of the tool, they will not know when it is making obvious errors.
A Powerful Brainstorming Partner
For the task that involved predictive analytics, our participants faced a challenging scenario: neither
they nor the GenAI tool were highly adept at that task. Here, the technology was still valuable as a
brainstorming partner.
© 2024 Boston Consulting Group 5
While all the tasks in our experiment were designed such that the GenAI could not independently
solve them, the predictive-analytics task required the most engagement from participants. They were
asked to create a predictive model, using historical data on international soccer matches, to develop
an investment strategy. Their ultimate goal was to assess how predictable, or reliable, their model
would be for making investment decisions.
Many participants used GenAI to brainstorm, combining their knowledge
with the tool’s knowledge to discover new modeling and problem-
solving techniques.
As shown in Exhibit 2, this was the task on which the GenAI-augmented consultant was least likely to
perform on par with a data scientist, regardless of previous experience in coding or statistics. This is
because the GenAI tool is likely to misunderstand the ultimate goal of the prompt if the entire task is
copied and pasted directly into the tool without breaking the question into parts or clarifying the
goals. As a result, participants with access to GenAI were more likely to be led astray than their
nonaugmented counterparts.
Even so, we found that, with the support of GenAI, many participants were able to step outside their
comfort zone. They brainstormed with the tool, combining their knowledge with GenAI’s knowledge
to discover new modeling techniques and identify the correct steps to solve the problem successfully.
The GenAI-augmented participants were 15 percentage points more likely to select and appropriately
apply machine-learning methods than their counterparts who did not have access to GenAI.
Reskilled, but Only When Augmented
Participants’ aptitude for completing new and challenging tasks was immediately boosted when
using GenAI, but were they reskilled? Reskilling is defined as an individual gaining new capabilities or
knowledge that enables him or her to move into a new job or industry. We found in our study that
GenAI-augmented workers were in a sense “reskilled,” in that they gained new capabilities that were
beyond what either the human or GenAI could do on their own. But GenAI was only an exoskeleton;
the participants were not intrinsically reskilled, because “doing” with GenAI does not immediately
nor inherently mean “learning to do.”
While each participant was assigned just two of the three tasks in the experiment, we gave everyone
a final assessment with questions related to all three tasks to test how much they actually learned.
For example, we asked a coding syntax question even though not everyone did the coding task—and
therefore not everyone would have had a chance to “learn” syntax. Yet the people who participated
© 2024 Boston Consulting Group 6
in the coding task scored the same on the assessment as people who didn’t do the coding task.
Performing the data-science tasks in our experiment thus did not increase participants’ knowledge.
Of course, participants only had 90 minutes to complete the task. With repetition, more learning
might have occurred. We also didn’t inform participants that they would be tested at the end, so
incentivizing learning might also have helped. This is important, because we found that having at
least some background knowledge of a given subject matters.
We found that coding experience is a key success factor for workers who use
GenAI—even for tasks that don’t involve coding.
GenAI-augmented participants with moderate coding experience performed 10 to 20 percentage
points better on all three tasks than their peers who self-identified as novices, even when coding was
5
not involved. In fact, those with moderate coding experience were fully on par with data scientists
for two of the three tasks—one of which had zero coding involved.
Based on this, we posit that it is the engineering mindset that coding helps develop—for example,
having the ability to break a problem down into subcomponents that can be effectively checked and
corrected—that ultimately matters, more so than the coding experience itself.
The risk of fully automating code, then, is that people don’t form this mindset—because how do you
maintain this skill when the source of its development is no longer needed? This is part of a larger
discussion: What other seemingly automatable skills have such importance? Will these skills become
the new Latin, taught mostly to cultivate a particular mindset?
Managing the Transition
While we have used data science as a case study, we believe that our finding—that augmented
workers can skillfully perform new tasks—can be applied to any field that is within the tool’s
capabilities. We’ve identified five core implications for company leaders. (See Exhibit 3.)
© 2024 Boston Consulting Group 7
Talent Acquisition and Internal Mobility. The results across our workforce experiments have
shown that what an individual can perform on his or her own by no means approaches what can be
accomplished when augmented by GenAI. This suggests that the talent pool for skilled knowledge
work is expanding.
Recruiters should therefore incorporate GenAI into the interview process to get a more complete
picture of what a prospective employee might be capable of when augmented by the technology.
Leaders may also find that an unlikely person inside their organization can fill an open role. We’re
not suggesting that nontechnical generalists can immediately become data scientists. But a
generalist marketer could, for example, take on marketing analyst tasks or roles.
Learning and Development. What does this mean for employees seeking paths to senior roles
and/or leadership? How should members of the GenAI-augmented workforce, who can flexibly take
on various roles, cultivate the right skills for career advancement—and what are the most important
skills for them to retain long term?
While GenAI has an immediate aptitude-expansion effect, learning and development remain the
most import lever for cultivating advanced skills and supporting each employee’s professional
trajectory. Leaders therefore must ensure that employees have incentivized and protected time to
learn. Other research has shown that when specifically used for learning (and, unlike our
participants, people are generally incentivized to learn in their jobs), GenAI is an effective
personalized training tool.
© 2024 Boston Consulting Group 8
Leaders should ensure that future implementations of GenAI tools include
the functionality to inform the user if a task is outside the technology’s
capability set.
Our analysis also suggests that developing some technical skills leads to greater performance, even
for nontechnical workers. Regardless of the training employees receive, company leaders should
ensure their future implementations of GenAI tools include the functionality to inform the user if a
task is outside the technology’s capability set—information that should be compiled from regular
benchmarking.
Companies are likely to find competitive advantage from developing tools and processes that
precisely assess the capabilities of GenAI models for their use cases. As shown in Exhibit 1, how a
worker should use GenAI greatly depends on understanding where a task lies within their own skill
set and within the capabilities of the technology.
Teaming and Performance Management. Although our results show it is possible for a generalist
to take on more complex knowledge work, it will be crucial to manage their performance and ensure
the quality of their output. This could mean designing cross-functional teams to provide generalists
with easy access to an expert when they need help and establishing regular output-review
checkpoints—because an overconfident generalist may not always know when to ask for support.
Leaders will need to run pilots to ensure their teaming configurations lead to the best outcomes. This
may be an opportunity to break silos and integrate teams of generalists with experts from various
centers of excellence.
Strategic Workforce Planning. Given the implications for talent and teaming, how should
organizations think about specialized expert tracks and the structure of their workforce? What does
strategic workforce planning for knowledge work mean in a world of constant job transformation
and technological advancement? We don’t have all the answers. But we do see that the skills needed
for a given role are blurring, and workforce planning will no longer be solely focused on finding a
certain number of people with a specific knowledge skill, such as coding.
Instead, planning should include a focus on behavioral skills and enablers that will support a more
flexible workforce. While knowledge workers may be technically capable of taking on new roles with
the help of GenAI, not everyone is equally adept at embracing change.
Professional Identity. The impact of GenAI on professional identity is an important and
contentious topic. But a recent survey suggests that negative impacts can be mitigated when
employees feel supported by their employers.
© 2024 Boston Consulting Group 9
In fact, in our study, we found that 82% of consultants who regularly use GenAI for work agree with
the statements “Generative AI helps me feel confident in my role” and “I think my coworkers enjoy
using GenAI for their work,” compared to 67% of workers who don’t use it on a weekly basis. More
than 80% of participants agreed that GenAI enhances their problem-solving skills and helps them
achieve faster outputs.
This suggests that highly skilled knowledge workers genuinely enjoy using the tool when it allows
them to feel more confident in their role—which aligns with our previous findings that mandating
the use of AI can actually improve employee perception of AI. However, this is only true if employees
believe that AI is being deployed to their benefit.
We are only at the beginning of the GenAI transformation journey, and the technology’s capabilities
will continue to expand. Executives need to be thinking critically about how to plan for this future,
including how to redefine expertise and what skills to retain in the long term.
But they are not alone: Skill development is a collaborative effort that includes education systems,
corporate efforts, and enablement platforms such as Udemy and Coursera. Even the providers of
GenAI models should be thinking about how their tools can further enable learning and
development. Preparing for the GenAI-augmented workforce must be a collective endeavor—
because our collective future depends on it.
bhi-logo-image-gallery-2-tcm9-239323.jpg
The BCG Henderson Institute is Boston Consulting Group’s strategy think tank, dedicated to
exploring and developing valuable new insights from business, technology, and science by embracing
the powerful technology of ideas. The Institute engages leaders in provocative discussion and
experimentation to expand the boundaries of business theory and practice and to translate
innovative ideas from within and beyond business. For more ideas and inspiration from the Institute,
please visit our website and follow us on LinkedIn and X (formerly Twitter).
© 2024 Boston Consulting Group 10
Authors
Daniel Sack
MANAGING DIRECTOR & PARTNER
Stockholm
Lisa Krayer
PRINCIPAL
Washington, DC
Emma Wiles
ASSISTANT PROFESSOR OF INFORMATION SYSTEMS,
BOSTON UNIVERSITY’S QUESTROM SCHOOL OF BUSINESS
Mohamed Abbadi
CONSULTANT
Washington, DC
Urvi Awasthi
DATA SCIENTIST
New York
Ryan Kennedy
AI ENGINEER
Boston
Cristián Arnolds
CONSULTANT
New York
François Candelon
ALUMNUS
© 2024 Boston Consulting Group 11
1 That experiment was conducted using the first version of GPT-4.
2 Of the consultants who originally signed up to participate, 480
completed the experiment. Participants were randomly split into a
control group that was not allowed to use GenAI for the tasks and a
“treatment” group that was asked to use GenAI. Each participant was
randomly assigned two of the three tasks; each task was timeboxed for
90 minutes. It is important to note that we did not test a full end-to-end
data science workflow from ideation to delivery or advanced topics such
as deep learning.
3 We used Enterprise ChatGPT with GPT-4 and its Advanced Data
Analysis feature.
4 The findings for the statistical understanding task were consistent with
those of the coding and predictive analytics tasks; as a result, we
focused on the first two tasks in this article.
5 Coding experience was based on a self-assessment. We define moderate
experience by those who selected “I know how to code but am not an
expert” and novices as those who selected “I only know the basics of
coding” and “I don’t know how to code.”
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 2024. All rights reserved.
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).
© 2024 Boston Consulting Group 12 |
284 | bcg | bcg-how-digital-and-ai-solutions-will-reshape-health-care-in-2025.pdf | How Digital and AI Will
Reshape Health Care
in 2025
January 2025
Boston Consulting Group BCG X
Boston Consulting Group partners with leaders BCG X is the tech build & design unit of BCG.
in business and society to tackle their most
Turbocharging BCG’s deep industry and
important challenges and capture their greatest
functional expertise, BCG X brings together
opportunities. BCG was the pioneer in business
advanced tech knowledge and ambitious
strategy when it was founded in 1963. Today,
entrepreneurship to help organizations enable
we work closely with clients to embrace a
innovation at scale.
transformational approach aimed at benefiting all
stakeholders—empowering organizations to grow,
With nearly 3,000 technologists, scientists,
build sustainable competitive advantage, and
programmers, engineers, and human-centered
drive positive societal impact.
designers located across 80+ cities, BCG X builds
and designs platforms and software to address
Our diverse, global teams bring deep industry and
the world’s most important challenges and
functional expertise and a range of perspectives
opportunities.
that question the status quo and spark change.
BCG delivers solutions through leading-edge
Teaming across our practices, and in close
management consulting, technology and design,
collaboration with our clients, our end-to-end
and corporate and digital ventures. We work in a
global team unlocks new possibilities. Together
uniquely collaborative model across the firm and
we’re creating the bold and disruptive products,
throughout all levels of the client organization,
services, and businesses of tomorrow.
fueled by the goal of helping our clients thrive and
enabling them to make the world a better place.
How Digital and AI Will Reshape
Health Care in 2025
T
he definition of digital health is evolving. The era A growing number of individually tailored apps and digital
spurred on by the Covid-19 pandemic—think tele- platforms will give patients more control over their medical
medicine and digital therapeutics, which have strug- conditions, predict flare-ups, and suggest real-time interven-
gled to scale—is giving way to one defined by artificial tions. We expect consumers to increasingly rely on AI chat-
intelligence (AI) and solutions that strengthen the bond bots and virtual assistants for answers to health questions.
between health care professionals and patients in an
integrated manner, with appropriate economics to support Digital health will continue to offer solutions to address
them. gaps in women’s health care, including femtech innova-
tions to redesign traditional “hardware” used for women’s
We see this shift reflected in trends that experts across health (such as the speculum), with the female experience
BCG and BCG X anticipate will shape digital health in at the center. It’s a needed shift: A recent BCG X survey
2025. As AI matures, it is rapidly expanding possibilities for found that fewer than half of women respondents across
patients, providers, and health care organizations alike. the globe (41%) agreed that there are sufficient services to
New digital solutions are being leveraged to address gaps address their specific health concerns.
in care for chronic conditions such as heart failure, diabe-
tes, and mental health. And the growing influence of gener- We are also beginning to see a maturing of partnerships
ative AI (GenAI) on every aspect of health care—from between femtech health and wellness brands that can lead
personalized care to automated workflows—is a key theme to interoperable ecosystems that pool women’s health
for the upcoming year, as it was in 2024. data and ultimately drive improved health outcomes.
Let’s dive deeper into how we expect digital and AI solu- Provider Empowerment
tions to reshape health care in 2025. Providers will be empowered and enabled by digital tech-
nology as well. AI can provide the analytical muscle to
Patient Support process vast quantities of personal patient data, powering
This year, digital health tools will continue to transform highly personalized medical treatment tailored to individu-
patient care, improving their support and access. Smart als based on their unique health data from continuous
implants and wearable devices that allow providers to monitoring devices, lifestyle inputs, and individual genetics.
monitor patients’ cardiac activity, blood sugar levels, and This enables providers to adjust treatment dynamically
other biological functions in real time from remote loca- based on feedback in real time.
tions will enable better chronic disease management and
improve patients’ quality of life. As sleep continues to gain Artificial intelligence decision-making tools will become
attention as a crucial biomarker for overall well-being, mainstream in 2025, giving doctors immediate access to
health tech companies are creating more advanced, accu- evidence-based research and treatment guidelines. GenAI
rate sleep-tracking tools. applications will accelerate diagnoses and minimize diag-
nostic errors, while speeding the delivery of patient care
and more accurately predicting patient outcomes.
BOSTON CONSULTING GROUP 1
Emergence of Ecosystems While GenAI continues to generate tremendous excite-
At the organizational level, our experts anticipate that the ment in the digital health care space, it’s not a panacea.
coming year will see an expansion of the use of AI to orga- Our experts recognize that some of these programs won’t
nize and automate entire workflows instead of just specific deliver anticipated results in 2025. When that happens, we
tasks. For example, rather than an AI tool that facilitates emphasize the importance of going back to the basics:
physician note-taking or scheduling, intelligent agents will focusing on business outcomes and tracking key perfor-
automate an entire patient episode of care, from intake mance indicators. In this way, AI failures can drive more
through treatment plan. Working across departments, AI focused, sustainable transformation in the long term.
programs will learn as they go, improving efficiency and
outcomes at both the patient and system level. Health Clearly, 2025 promises to be a transformative year.
systems will benefit, but so will other types of health care We’re excited to see how AI and more digital solutions
organizations such as pharmaceutical companies, where reshape health care.
GenAI can transform key activities such as clinical trials
and regulatory submissions.
AI-driven data processing will also allow access to data that
has until now been considered too disorganized to be useful,
such as medical records, clinical notes, and physician/
patient interaction information. Clinicians, payers, and
drug companies alike will be able to draw out actionable
insights from these data sets to improve patient care and
outcomes. At the same time, expanded access will en-
hance different systems’ ability to interact with one anoth-
er, facilitating more seamless collaboration.
2 HOW DIGITAL AND AI WILL RESHAPE HEALTH CARE IN 2025
“With AI-driven solutions, wearable devices, and
digital triaging tools, patients are no longer passive
participants in their care but active managers of their
health journey.”
Ashkan Afkhami
Managing Director and
Senior Partner
Every patient will soon have the tools to find the right By capturing, analyzing, and applying data to drive better
care, support, and treatment tailored to their unique treatments, therapies, and operational efficiency, health
needs. We’re at a tipping point in patient care. With AI-driven care organizations will realize their true potential.
platforms, wearable devices, and digital triaging tools, patients By leveraging real-time insights and advanced analytics,
are no longer passive participants in their care but active man- organizations can detect diseases earlier, streamline care
agers of their health journey. Technology is closing critical gaps pathways, and optimize operations. Interoperable systems
in care, particularly in underserved communities, by enabling and secure data-sharing frameworks are critical for achieving
timely guidance, remote consultations, and personalized care these outcomes, ensuring data can flow seamlessly across
plans. These tools empower patients to take charge of their stakeholders. As we move forward, ethical AI frameworks and
health and promote a seamless, personalized experience that integrated data strategies will be the catalysts for change,
meets them wherever they are—whether at home, in the clinic, transforming health care into a precision-driven, efficient,
or on the go. and impactful ecosystem.
Digital health solutions that simplify workflows,
optimize resources, and improve patient monitoring
will enable clinicians to deliver continuous,
high-quality care. AI-assisted technologies are helping to
address capacity challenges, reduce diagnostic turnaround
times, and improve treatment accuracy. Similarly, deci-
sion-support tools and real-time analytics are enabling
smarter, safer care delivery. By integrating remote patient
monitoring, automation, and predictive analytics, health
care professionals can focus on what matters most: treat-
ing and supporting patients beyond the clinic walls. This
shift will improve efficiency and empower clinicians to
deliver proactive, patient-centered care.
BOSTON CONSULTING GROUP 3
“Digital tools can help bridge R&D and access gaps,
driving equity in health care for women.”
Johanna Benesty
Managing Director and
Senior Partner
Women continue to face barriers to accessing health To fully harness digital health’s potential in low- and
care. Many factors contribute to lack of access, even in middle-income countries (LMICs), the health care
high-income countries—including economic disparities, ecosystem must overcome several challenges. Scaling
limited R&D on women-specific health, and systemic bias- digital health initiatives effectively remains a primary hurdle
es. For instance, 26% of US women delay care due to cost. because many digital solutions that show promise in pilot
For many low-income women, this limits access to essen- stages struggle with long-term sustainability across diverse
tial reproductive health services. Additionally, lack of R&D regions. Investment in digital infrastructure, talent develop-
on women’s conditions like endometriosis or menopause ment, and skills training is essential, as health care workers in
leads to delayed diagnoses and inadequate treatment. LMICs often lack the technical training needed to operate and
Digital health offers solutions to address these gaps. support digital health tools. Moreover, ethical considerations,
Telehealth has expanded access to underserved areas, and especially regarding AI use, are paramount. Without clear
virtual consultations are proven to support women in rural guidelines and regulatory frameworks, AI risks exacerbating
areas who need mental health services. Health apps such health inequities rather than reducing them. Cultural adapta-
as Eve and Flo collect critical health data about women’s tion and community trust in these technologies are also critical,
menstrual cycles and ovulation, advancing research on requiring a user-centered approach that aligns digital health
women’s health and creating awareness. And AI can help solutions with local values and health care practices. Address-
reduce the cost of R&D on target groups. These digital ing these challenges will be key to advancing equitable access
tools can help bridge R&D and access gaps, driving equity to health care in LMICs through digital innovation.
in health care for women.
4 HOW DIGITAL AND AI WILL RESHAPE HEALTH CARE IN 2025
“As smart implants become more sophisticated,
they will enable more proactive and tailored
health care.”
Diego Bernardo
Principal
Smart implants will play an increasingly significant AI-powered N-of-one studies will expand, providing
role in patient-centered care. Smart implants are ad- the computational power and advanced analytics
vanced medical devices that integrate with the body’s needed to make personalized medicine feasible. AI
biological systems to monitor, diagnose, or treat various and machine learning are key enablers of N-of-one research.
conditions in real time. These implants are equipped with In N-of-one studies, a single patient’s unique data—such as
sensors, microprocessors, and wireless communication genetic information, lifestyle habits, and continuous health
technologies, enabling them to gather critical health data monitoring—is collected and analyzed in real time. AI and
and adjust their function based on patient needs. For machine learning algorithms can process this vast array of
example, smart cardiac implants can regulate heart individual data, identifying patterns, predicting health out-
rhythms or detect arrhythmias, while glucose-monitoring comes, and optimizing treatments specific to the patient.
implants continuously track blood sugar levels for diabet- These tools allow continuous learning from a patient’s
ics. Neuro-prosthetics and brain-computer interfaces evolving responses to interventions, enabling dynamic
(BCIs) are also part of this revolution, allowing patients to adjustments to therapies based on real-time feedback.
control prosthetic limbs with their minds or even restore This personalized, data-driven approach can lead to more
motor functions in cases of paralysis. These devices offer effective treatments and improved patient outcomes,
continuous, real-time monitoring and treatment, reducing making AI and machine learning critical to the future of
the need for frequent medical interventions and improving precision medicine.
the quality of life for patients with chronic conditions. As
smart implants become more sophisticated, they will
enable more proactive and tailored health care.
BOSTON CONSULTING GROUP 5
Vocal biomarkers that detect early signs of disease
have the potential to speed interventions and im-
prove patient outcomes. Vocal biomarkers represent a
cutting-edge trend in digital health, where subtle changes in
voice patterns are analyzed to diagnose and monitor vari-
ous health conditions. By using artificial intelligence and
machine learning algorithms, vocal biomarkers can detect
early signs of diseases such as Parkinson’s, Alzheimer’s,
respiratory infections, and even mental health disorders like
depression and anxiety. These tools analyze factors such as
tone, pitch, cadence, and even micro-tremors in the voice,
offering a non-invasive, scalable method for continuous
health monitoring. Vocal biomarkers hold particular prom-
ise in telemedicine, where remote assessment is critical.
As the technology advances, voice analysis could become a
routine tool in both preventive care and chronic disease
management, offering early intervention possibilities and
improving patient outcomes through real-time data collec-
tion. This approach also aligns with the growing trend of
passive health monitoring using everyday interactions.
As AI matures, it is rapidly
expanding possibilities for
patients, providers, and health
care organizations.
6 HOW DIGITAL AND AI WILL RESHAPE HEALTH CARE IN 2025
“Resilient manufacturing processes will look to AI as a
strategic enabler, helping pharma manufacturers meet
growing demand with improved accuracy.”
Satty
Chandrashekhar
Managing Director and Partner
Drug discovery and development continue to accelerate. Quality control comes of age in pharma manufacturing.
Forward-thinking pharma and biotech organizations will Generative AI (GenAI) has started to play a transformative
continue to reshape their R&D agendas, leveraging cus- role in ensuring higher standards of quality control in the
tomized language models to improve understanding of manufacturing of pharmaceutical and medical device prod-
disease biology and accelerating processes to identify ucts. Leveraging this capability to enhance the detection and
promising compounds. Models, both commercial and mitigation of deviations in manufacturing processes, AI will
open, already present the potential to analyze vast bio- help organizations address quality issues in more standard
medical data sets to suggest novel molecular structures or ways across manufacturing facilities, many of them global,
predict drug interactions. Combined with causal modeling by analyzing vast operational data streams from production
approaches, the ability to identify clues previously undis- environments. This approach to quality control will allow
covered or underrepresented in clinical data will continue manufacturers to adjust processes, reduce waste, improve
to evolve. And in 2025, this trend will further shorten dis- yield, and increase product quality. An issue-resolution
covery cycles and reveal more promising candidates to test GenAI solution trained with historical data, for example,
in clinical settings. Clinical development will also continue has the potential to help organizations assess the effects of
to accelerate. By using AI to improve data quality, better minor changes on product outcomes, enabling companies
understand data lineage, and enable evolved uses of oper- to reimagine processes without extensive and often manual
ational and patient data to find the right sites and more trial-and-error tests. Enhancing safety while staying compli-
precise populations for clinical studies, the industry will ant with regulations is critical to this effort—and at-scale
force a reckoning with operational data readiness. can accelerate the speed at which new treatments reach the
market. Resilient manufacturing processes will look to AI as
a strategic enabler, helping pharma manufacturers meet
growing demand with improved accuracy and lower produc-
tion costs.
BOSTON CONSULTING GROUP 7
Pharma commercial enterprises will reimagine how
they make data-driven decisions. Many pharma compa-
nies continue to transform how they generate insights and
make strategic decisions. With the proliferation of real-world
data from sources like electronic health records, patient
registries, data related to social determinants of health, and
other non-traditional sources, companies are harnessing AI
to derive actionable insights at unprecedented speed and
scale. AI platforms will synthesize complex data sets into
clear insights that inform everything from market access
strategies to patient engagement and salesforce optimiza-
tion. Commercial teams will increasingly rely on predictive
models to forecast trends, identify emerging therapeutic
needs, and optimize pricing strategies in business-time.
These models will allow mature companies to pivot quickly
based on evolving global dynamics, competitor actions, and
regulatory shifts. This will lead to new operating models and
new ways of working to harness this data across the “extended
enterprise” with speed and precision—to move beyond
traditional silos and integrate information to drive more
cohesive decisions in the organization and anticipate and
respond to trends more effectively.
The growing influence of GenAI on
every aspect of health care—from
personalized care to automated
workflows—is a key theme.
8 HOW DIGITAL AND AI WILL RESHAPE HEALTH CARE IN 2025
“Consumer behavior patterns will likely force
established online health information gateways to offer
their own bespoke AI tools.”
Nick Cristea
Vice President,
Experience Design
Consumers will increasingly use AI chatbots for At-home diagnostic solutions will become a more
health questions, and a growing number of people ingrained part of how patients and providers interact.
will see them as trusted resources. A KFF poll from These tools should help simplify logistics for patients, espe-
August 2024 found that one in six adults say they use AI cially those in more rural settings. One example is TytoCare,
chatbots at least once a month to find health information a portable medical exam kit with an app that sends informa-
or advice, rising to 25% for adults under 30 years old. As tion to providers. The company’s recent announcement of a
the technology improves, these consumer behavior pat- new integration with Epic’s MyChart, in collaboration with
terns will likely force established online health information the University of Miami Health System, will leverage Tyto-
gateways to offer their own bespoke AI tools or risk losing Care for medical exams and enable asynchronous workflows
web traffic. Once providers move past their risk-averse for remote patient monitoring and primary care. With big
strategies, they will be able to start realizing significant hardware companies continuing to invest in consumer
operational efficiencies and competitive advantage by health care solutions, there will be an increasing expectation
leveraging their “clinical expert” brands to attract patients of preventive care benefits rather than simply vitals monitor-
to their AI services, while also reducing the burden on ing, and more formalized partnerships will emerge. As
humans who staff the 24/7 triaging capabilities that they clinical researchers discover new ways of detecting the early
offer. Deploying these tools first as co-pilots for their “nurse onset of disease through measurable biomarkers, patients
line” staff provides an early stepping stone for building and with a history of chronic conditions will be able to sign up
testing their capabilities and ensuring that they do not get with remote monitoring programs, which in turn will feed
bypassed in the future as patients seek greater access to more data back to the research teams. And, with the colla-
immediate answers and strategies to relieve their symp- tion of richer personalized health data, providers will gain a
toms. The providers who architect these solutions most deeper understanding of how best to automate solutions for
effectively will be able to realize a host of downstream patients or escalate to the right people at the right time,
opportunities in attracting patients, collecting data, and improving their operational efficiency while also decreasing
increasing effectiveness in triaging and routing patients to time to treatment.
the most appropriate sources of care.
BOSTON CONSULTING GROUP 9
“By interpreting and synthesizing unstructured
clinical data into actionable insights, GenAI will
streamline workflows and improve efficiency.”
Andre Heeg
Managing Director and Partner
The combination of AI, genomics, and wearable tech GenAI will revolutionize health administration by
is paving the way for highly personalized treatments. automating the creation and updating of medical
Digital therapeutics will evolve to offer precision care based records, reducing physicians’ time spent doing
on general population data and individual genetics, lifestyle, paperwork. By interpreting and synthesizing unstructured
and real-time health data. For instance, we’ll see more apps clinical data into actionable insights, GenAI will streamline
and platforms tailored to individuals managing chronic workflows and improve efficiency. This will free up more
diseases that can predict flare-ups and suggest real-time time for health care providers to focus on patient care
interventions based on continuous health monitoring. while ensuring that health records are more accurate
and comprehensive.
AI-powered decision-making tools will become
mainstream, improving diagnostics, treatment plans,
and patient outcomes. In particular, GenAI will give physi-
cians near-instant access to research insights, treatment
guidelines, and real-world evidence, allowing for more in-
formed decisions. This will significantly reduce diagnostic
errors and speed up patient care delivery.
10 HOW DIGITAL AND AI WILL RESHAPE HEALTH CARE IN 2025
“Medicine 3.0 prioritizes healthspan driven by improved
prevention, personalization, and participation.”
Iana Kouris
Managing Director
The distinction between longevity and healthspan is The concept of Medicine 3.0 and shifting to proactive
becoming more and more important. According to The approaches in health care is gaining momentum.
Longevity Imperative, by Andrew Scott of the London Business While Medicine 2.0 focuses mainly on extending lifespan
School, “In the UK in 1965, the most common age of death and disease treatment, Medicine 3.0 expands this and
was in the first year of life. Today the most common age to prioritizes healthspan driven by improved prevention, per-
die is 87 years old.” This sounds like great news. However, a sonalization, and participation, as noted in Outlive by Peter
longer life doesn’t always mean a healthy life. While the Attia. This drives demand for regular heath checks, popular-
global life expectancy has continued to rise (from 66.8 years izes prevention approaches, and leads to the emergence of
in 2000 to 72.5 years in 2020) and the healthy life expectan- new businesses in these areas.
cy (HALE or healthspan) has also increased (from 58.1 in
2000 to 62.8 in 2020), the gap between life span and Generative AI in health care is driving better access,
healthspan has actually increased by 1 year during that time personalization, and quality. In particular, GenAI can
(from 8.7 to 9.7 years). Hence, on average, we spend one support health care in the area of Medicine 3.0. For instance,
more year with disability/in poor health than we used to, it can drive personalization through chatbots, virtual assis-
according to the World Health Organization. This results in tance, and more precise analysis of health-related data. New
additional medical costs, challenges for our insurance sys- GenAI models focused on medical applications will be devel-
tems, and increased years spent with suffering/discomfort. oped that can, for example, interpret data from medical
imaging, lab results, and electronic health records to produce
written or spoken recommendations.
BOSTON CONSULTING GROUP 11
“If 2023 was about GenAI experimentation and 2024
was about point solutions, 2025 will be about value
delivery through end-to-end transformation.”
Julius Neiser
Managing Director and Partner
More than a third of ongoing GenAI programs will physician note taking or scheduling, we will witness inte-
fail to deliver value in 2025, and some health care grated systems that automate entire workflows, for exam-
players will draw the wrong conclusions from that. ple, from patient intake to treatment plans. These intelli-
Many GenAI solutions are delivering true impact by, for gent agents will coordinate across departments, learning
example, reducing medical/regulatory writing effort by 50% from each interaction to improve efficiency and outcomes.
and shaving valuable months off drug launch timelines. For example, in pharma, key processes that will be trans-
But more than a third of programs fail. The takeaway from formed with GenAI include clinical trials, regulatory sub-
these failures should not be to reduce funding, but rather missions, medical legal regulatory review, and omnichan-
to build on the lessons learned: obsess about business nel engagement.
outcomes, rigorously track key performance indicators, and
concentrate on the important people aspect of GenAI Unstructured health care data will become the new
transformations. The failure of some programs will ulti- structured health care data. Advances in AI-driven data
mately pave the way for more sustainable and impactful processing will allow systems to analyze and organize vast
transformations, driving a sharper focus on integrating amounts of medical records, clinical notes, and physician/
GenAI into existing health care workflows. patient interactions previously considered too disordered to
leverage. This shift will enable health care professionals,
We will see an evolution from health care GenAI pharma companies, payers, and providers to extract action-
point solutions to agentic end-to-end process able insights from a much larger data set, improving patient
transformations. If 2023 was about GenAI experimenta- care and outcomes. This transformation will enhance in-
tion and 2024 was about point solutions, 2025 will be teroperability across different systems, facilitating seamless
about value delivery through end-to-end transformation. collaboration between providers and empowering more
Instead of isolated AI tools focused on specific tasks like personalized and precise medical treatments for patients.
12 HOW DIGITAL AND AI WILL RESHAPE HEALTH CARE IN 2025
“GenAI chatbots will reshape the cost profile of contact
centers and drive significant efficiencies.”
Etugo Nwokah
Managing Director and Partner
GenAI chatbots will define a new table-stakes for Further innovation around clinical specialties will
payers and providers to deliver an omni-channel drive increased value-based payment arrangements
customer experience that patients and members will and end-to-end management for specific diseases.
expect. These capabilities will reshape the cost profile of Specialties such as oncology, orthopedics, and behavioral
contact centers and drive significant efficiencies. Health health will continue to see higher spending by payers,
plans will see a significant improvement in the adoption whose ability to influence and measure outcomes, well-
of self-service capabilities by their members that will drive defined populations, and episodes of care will be crucial to
major cost efficiencies, which will be reinvested to support leveraging new technology enablers such as large language
ambitious growth agendas and M&A. models (LLMs).
Organizational models that focus on digital products
and services will further evolve in hospital systems
and payer organizations of all sizes. As Chief Digital
Officer or Chief Product Officer roles become more preva-
lent in these organizations, principles around building
technology with measurable outcomes and customer-
centric experiences will overcome historical challenges of
IT functions becoming “digital feature factories” building a
lot of capabilities that no one uses.
BOSTON CONSULTING GROUP 13
“Forward-thinking health care companies will focus on
driving AI adoption internally to accelerate scaling.”
Sid Thekkepat
Managing Director and Partner
Many health care AI use cases will be slow to scale Sleep monitoring is emerging as a mainstream bio-
and drive value, leading to increased pressure on IT marker that consumers are increasingly prioritizing
and change management teams. Winning health care in their health-tracking routines; providers and life
companies will experiment with new approaches to sciences will pay more attention. As awareness grows
drive adoption. By fostering a culture of experimentation, around the critical role sleep plays in overall well-being—
collaborating with frontline staff, and prioritizing impacting everything from mental health to chronic disease
user-friendly AI solutions, these organizations can enhance prevention—health tech companies are responding by
adoption rates and realize AI’s full potential. Health care focusing on developing more advanced, accurate
companies that thrive will be those that treat adoption as sleep-tracking tools. This trend is driving innovations in
a critical aspect of AI implementation, leveraging iterative wearables, apps, and even non-invasive monitoring devices
learning and adaptive frameworks to drive sustainable value. designed to provide deeper insights into sleep patterns,
quality, and its correlation with other health metrics. Expect
As valuations drop, consolidation and M&A activity sleep tracking to become a cornerstone of personalized
will intensify in the health tech space, leading to the health solutions as tech continues to refine its capabilities.
emergence of more scaled, sustainable platforms.
Larger strategic players will seize this opportunity to
strengthen their positions by acquiring or merging with
innovative but financially constrained startups. Private
equity and venture capital firms, facing market uncertainty,
may hold back for now, allowing established companies to
lead the charge with bold and creative deals.
14 HOW DIGITAL AND AI WILL RESHAPE HEALTH CARE IN 2025
“The focus on tech-enabled mental health will continue
to grow, notably by further integrating mental health
services into primary care.”
Gunnar Trommer
Managing Director and Partner
More providers/hospitals will develop their own Products and solutions that are enabled by machine
workflow efficiency solutions leveraging GenAI, learning—and soon, GenAI—will see accelerated
mainly aiming at reducing the administrative burden adoption in diagnosing diseases, analyzing medical
on clinicians and improving workflows. As long as imaging, and predicting patient outcomes. AI-powered
providers develop tech-enabled solutions for their own |
285 | bcg | digital-government-in-the-age-of-ai-championing-gcc-next-gen-citizen-services.pdf | Digital Government in the
Age of AI: Championing GCC
Next-Gen Citizen Services
November 2024
By Rami Mourtada, Dr. Lars Littig, Miguel Carrasco, Semyon Schetinin, Akshara Baru
Contents
01 GCC digital government services 03 Citizen sentiment remains
lead globally while citizen the key to digital government
expectations are heightened service adoption
• What usage and satisfaction
levels reveal about GCC digital
government services
04 Government AI: A clear
path forward
02 AI and GenAI present • From promise to leadership: How GCC
governments can capitalize on progress
new opportunities for
GCC governments
• GCC public investment as a launchpad
for AI leadership
• Government investment in AI
and GenAI aligns with widespread
citizen adoption
• Seizing the GenAI opportunity while
addressing potential challenges
About the Digital Government
Citizen Survey in GCC
Conducted every other year, the Digital with The Kingdom of Saudi Arabia (KSA)
Government Citizen Survey (DGCS) and the United Arab Emirates (UAE)
is the most comprehensive and long- participating in the survey since its
running survey of global citizens on inception and Qatar joining in 2020.
digital government, with data spanning
Digital government services include 27
a decade from 2014 to 2024. The 2024
priority high-touch and citizen-facing
study surveyed 41,600 regular internet
online services across social services,
users (respondents) across 48 global
taxation, housing, health, education,
countries, representing 73% of the
transport, and immigration.
world’s population.
This report focuses on digital
government services in the GCC region,
Introduction
T
he significance of digital government services in
addressing citizens’ needs globally cannot be
overstated. These services are vital to support
individuals through key life events, from registering a
birth to healthcare and education, to public safety and
job support, social assistance and pensions, and more.
Digital government services also streamline critical
business processes like registering a company, filing
taxes, and ensuring regulatory compliance. They are
integral to the broader functioning of a nation and its
economy, acting as the backbone that ensures
societal and economic systems operate smoothly,
quickly, and efficiently.
Citizens’ satisfaction with their government services
and their experiences accessing them translate quickly
into positive overall perceptions of government
effectiveness and support. As such, building and
operating seamless digital offerings to better serve
citizens’ evolving needs should be a key priority for
governments and recognized as a core driver of
socioeconomic development. Innovation and
technological advancements, in this regard, provide
governments with vital opportunities to do even better.
This report showcases citizens’ experiences with digital
government services within the Gulf Cooperation Council
(GCC) region, based on data from BCG’s flagship 2024
Global Digital Government Citizen Survey (DGCS). A
particular focus in this edition was on citizens’ attitudes
towards government use of Artificial Intelligence (AI), and
especially Generative AI (GenAI).
1. “Citizen/s” refers to respondents who are nationals or international residents living within each country
1 DIGITALGOVERNMENTINTHEAGEOFAI:CHAMPIONINGNEXT-GENCITIZENSERVICESINTHEGCC
01. GCC digital government
services lead globally while citizen
expectations are heightened
P
revious surveys have shown that GCC governments technologies like AI and GenAI (discussed further
receive strong citizen approval for their digital below) are increasing citizens’ expectations in terms
services. This positive trend continued in 2024, of personalization and improved user experience.
as GCC countries maintained their global lead in This year’s data also highlights a strong link between
satisfaction with a net approval score of 81%2 [Exhibit 1], satisfaction levels and the perception that government
significantly higher than the global net average of 65%. digital services are on par with those of the private sector,
which are typically best-in-class. This underscores the
While this year’s survey has reconfirmed cumulative trends need for government services to match the efficiency
from past editions, it has also shown an evolution in citizen and effectiveness of the private sector while fostering
expectations. Accelerated digitalization and newer continued innovation in the future.
2. “Netexperience”ispositiveexperienceminusnegativeexperience.BCGAIRadar,FromPotentialtoProfitwithGenAI,2024
2 DIGITALGOVERNMENTINTHEAGEOFAI:CHAMPIONINGNEXT-GENCITIZENSERVICESINTHEGCC
Exhibit 1 | GCC countries lead globally in citizen satisfaction with
government digital services, especially when they match or exceed
private sector standards
Q. How satisfied are you with the use of the internet in delivering various types
of government services?
Indonesia
UAE
75 Thailand New Zealand Estonia Egypt Singapore
Australia Netherlands Italy Philippines Hong Kong-China Vietnam
Kenya Norway Canada Greece
Brazil Sri Lanka
Ukraine Spain Nigeria Mexico Bangladesh
Chile Turkey
50 Argentina
Morocco Switzerland
Malaysia
Laggards Emerging Leaders
30
0 35 70
Perception of government digital services compared to private-sector services2 (%)
Q. Are government online services better
than those offered by the private sector?
1. Net satisfaction is the percentage of satisfied respondents minus the percentage of dissatisfied respondents.
2. Respondents that agree government online services are better compared to those offered by private sector.
Source: 2024 BCG Digital Government Citizen Survey.
3 DIGITALGOVERNMENTINTHEAGEOFAI:CHAMPIONINGNEXT-GENCITIZENSERVICESINTHEGCC
latigid
htiw
1noitcafsitas
teN
)%(
secivres
tnemnrevog
KSA Qatar
Cambodia
Kazakhstan
This year’s survey shows citizens’ increased interest and What usage and satisfaction levels reveal
comparison to leading players in both private and public about GCC digital government services
sectors. For instance, while 42% of GCC respondents (a
steady proportion since 2022) expect quality standards It is important to note that while citizens expect
similar to those of global digital leaders, including top increasingly higher standards for digital government
high-tech private-sector companies. Addiotionally, 23% of services, the consistent growth in satisfaction levels in the
GCC respondents (4% higher than 2022) expect quality GCC is reflected in continued widespread usage. Trending
standards to match those of the best online government upward since 2022, the GCC records the highest usage
services globally, indicating citizens are rates of digital government services globally in 2024
more aware of global best practices across both private [Exhibit 2]. This overall usage level is a remarkably positive
and public sectors. indicator for governments in the GCC. However, examining
how both the frequently and less frequently used services
perform reveals what is going well and what still
needs improvement.
Exhibit 2 | Government service usage in GCC countries is well above
global average (+22%) with an upward uptake trend since 2022
Q. How often do you access government services online?
% of respondents across regions (2022 to 2024)
+4%
+2%
-2%
+22%
+5%
45%
Global
63 67 Average
54 56
46 44
33 38
Asia-Pacific Europe NorthAmerica GCC
2022 2024 Global Average
1. Respondents using government services at least once a week or more.
Source: 2024 BCG Digital Government Citizen Survey.
Overall, satisfaction scores of high-usage digital services frequently used services highlights ongoing or new user
(see service ranking in exhibit slides) are higher than the experience challenges that require attention. Addressing
average satisfaction across all services in the GCC. This these gaps, especially in less-used services, will reduce the
indicates that governments are focusing on improving risk of eroding satisfaction and trust (more on this year’s
these more frequently used services with higher repeat UI/UX challenges below).
value. However, the below-average satisfaction with less
4 DIGITALGOVERNMENTINTHEAGEOFAI:CHAMPIONINGNEXT-GENCITIZENSERVICESINTHEGCC
02. AI and GenAI present new
opportunities for GCC governments
T
he next generation of digital services for governments AI has rapidly become a global game changer, rapidly
will be driven by the newest emerging technologies, expanding the scale and scope of relevant use cases,
including AI and GenAI. These technologies will and accelerating adoption across organizations. Most
enable a whole new range of delivery methods and a interestingly, GenAI has emerged as an important
variety of high-value service use cases, from personalized transformational AI technology for direct citizen services.
recommendations and proactive nudges to more advanced Intuitive GenAI tools like OpenAI’s ChatGPT and Google’s
chatbots that will make usage easier and more productive, Gemini have seen rapid and widespread adoption around
reducing the burden on the citizen. the globe and across industries and users. And just as
GenAI is driving productivity and competitive advantage
• AI: The use of computer systems to perform tasks traditionally in private-sector customer service, it is also starting to
requiring human intelligence, such as learning, reasoning, transform digital government service quality and
problem-solving, and language. citizens’ experiences.
• GenAI: A type of AI that uses foundational, multi-modal
models to generate novel content, including text, images,
and audio/video, and supports interaction using natural
language prompts.
5 DIGITALGOVERNMENTINTHEAGEOFAI:CHAMPIONINGNEXT-GENCITIZENSERVICESINTHEGCC
GCC public investment as a launchpad development. Governments are integrating advanced
for AI leadership technologies like AI and IoT into public services while
also investing in AI-driven economic development. Saudi
A recent global BCG report estimates that GenAI has Arabia’s National Strategy for Data and AI, which aims to
the potential to automate 10–20%3 of an employee’s contribute SAR 500 billion ($133.3 billion) to GDP by 2030,
everyday tasks, freeing up time for more strategic work. is focused on developing AI capabilities, data governance,
In this context, it can also help make government services and analytics5,6. Recent developments include Saudi
simpler, more accessible, and more personalized. In the Arabia’s early adoption of UNESCO’s recommendations on
public sector, BCG estimates a $1.75 trillion4 annual AI ethics and the establishment of the International Center
productivity opportunity over the next decade. By delivering for Artificial Intelligence Research & Ethics (ICAIRE), which
services more efficiently and effectively, GenAI allows aims to foster ethical AI practices, support policy
governments to address backlogs, allocate further resources development, and ensure responsible AI implementation
to human-centric support, and maximize the value and across sectors. Qatar is driving its digital growth through
impact of public funds. However, appropriate AI safeguards collaboration with institutions like Qatar University and
are still needed to safely achieve these benefits, reduce tech providers to upskill ICT professionals in AI, 5G, and
potential risks (discussed below), and retain citizen trust. cloud computing7. With the Falcon LLM’s open-source
nature and cutting-edge performance, the United Arab
GCC countries have undertaken extensive investment, Emirates is positioning itself as a global AI leader; further
partnership, and upskilling efforts to leverage this bolstering its position by forming strategic alliances with
opportunity for government advancement, citizen tech giants to establish secure data centers and leverage
satisfaction, private-sector growth, and national the immense potential of data8.
Exhibit 3 | GCC countries have highest overall usage of AI and GenAI tools
compared to other regions globally
Q. How frequently1 do you use AI/GenAI2 tools ?
% of respondents by regions (2024)
79%
75%
64%
50% 50%
36%
37%
33%
27% 30%
24%
20%
Asia-Pacific Europe NorthAmerica GCC
Regional avg. AI/GenAI usage range within the region
1. Accessing digital government service and GenAI tools at least once a week or more.
Source: 2024 BCG Digital Government Citizen Survey.
3. BCGAIRadar,FromPotentialtoProfitwithGenAI,2024
4. BCG Article, Generative AI for the Public Sector: From Opportunities to Value, 2023
5. https://oxfordbusinessgroup.com/reports/saudi-arabia/2023-report/ict/digital-drive-strong-government-support-and-foreign-investment-are-
6. https://www.idc.com/getdoc.jsp?containerId=prMETA51181123
7. https://oxfordbusinessgroup.com/reports/qatar/2022-report/economy/digital-drive-both-the-public-and-private-sectors-turn-to-online-solutions-
while-the-country-taps-emerging-segments-such-as-e-sports
8. https://www.reuters.com/technology/uae-us-see-more-ai-partnerships-uae-minister-says-2024-05-21/#:~:text=The%20UAE%2C%20led%20by%20
government,activity%20outside%20the%20United%20States.
6 DIGITALGOVERNMENTINTHEAGEOFAI:CHAMPIONINGNEXT-GENCITIZENSERVICESINTHEGCC
Government investment in AI and GenAI government services, overall citizen satisfaction, and
aligns with widespread citizen adoption citizens’ trust in their government to use AI responsibly
(see slide 5 in exhibit slides).
GCC citizens are significantly more familiar with GenAI
than those in other regions, as reflected in their higher In GCC countries, which have high citizen satisfaction
general usage of GenAI tools [Exhibit 3]. This creates a scores for digital government services, citizens trust their
strong opportunity for GCC governments to accelerate governments even more than private sector entities to
the integration of AI and GenAI into their digital services. use AI responsibly. The net average trust in governments’
responsible use of AI is 71% across GCC countries – 49%
Citizen trust is imperative for governments to successfully higher than the global average – compared to 52% for
leverage AI and GenAI in digital services, especially given the private sector [Exhibit 4]. This trust gives GCC
GCC citizens’ heightened familiarity with GenAI tools. governments the opportunity to rapidly yet safely
Without trust, people are less likely to engage with AI- deploy GenAI to further enhance service efficiency,
driven solutions or features, squandering many related accessibility, and personalization, thereby fostering
service improvement opportunities. Interestingly, the data even greater trust and satisfaction.
shows a positive link between the quality of digital
Exhibit 4 | Respondents trust governments more to use AI/GenAI
responsibly than private sector, with trust in GCC countries exceeding
the global average
Q. To what extent do you trust organizations to use AI/GenAI responsibly?
Distrust Trust Net Difference to
trust1 global average
Overall
Regional 4 9 49 35 71 % +49 %pts
Government
National 4 8 32 54 74 % +51 %pts
State 5 10 37 44 +19%pts 66 % +15 %pts
Local 3 9 37 49 74 % +12 %pts
Regional 5 17 45 29 52 % +22 %pts
Private Sector
Highly distrust Somewhat distrust Somewhat trust Highly trust
1. Net trust is the percentage of positive trust minus the percentage of negative trust among respondents regarding the government's use of AI/GenAI.
Note: Responses with option "I don't know" are not represented.
Source: 2024 BCG Digital Government Citizen Survey
7 DIGITALGOVERNMENTINTHEAGEOFAI:CHAMPIONINGNEXT-GENCITIZENSERVICESINTHEGCC
Seizing the GenAI opportunity while GenAI for innovation, most are still in the early stages
addressing potential challenges of their journey, with only 8% applying it at scale.
The lightning speed of GenAI’s technological advancement Second, governments face citizen concerns as they assess
and its widespread global adoption come with several their pilot AI and GenAI use cases. In this year’s survey,
challenges. First, governments need to move fast enough citizens shared a number of these concerns [Exhibit 5],
to keep up in an effective and responsible way, despite including potential economic and legal issues. Citizens
rarely innovating this quickly. For example, BCG’s latest also cited the need to address technology adoption-related
Most Innovative Companies report9 ,which also assessed limitations like user capabilities and service accuracy
the public sector’s readiness to leverage GenAI, reveals to ensure high-quality outcomes. A further area of
it to score among the lowest on innovation readiness notable concern relates to ethical and social issues,
compared to other industries, and overall readiness has including the potential for service bias or discrimination
declined noticeably over the past two years. Globally, while and lack of transparency.
83% of public sector organizations are beginning to use
Exhibit 5 | While GCC respondents have high AI/GenAI usage, they have
material concerns across multiple important areas related to
Responsible AI
Q. What concerns you the most about the use of AI/GenAI?
Category Key concerns % of Respondents
Economic The potential loss of jobs and impact to the economy
andlegal
concerns The potential intellectual property risks
The capability of individuals to use AI
Technical
The accuracy of the results and analysis
concerns
The large volumes of data needed
The moral or ethical issues have not been resolved
Ethicaland
social The potential for bias or discrimination
concerns
The lack of transparency
Noconcerns I don't have any concerns
00 10 20 30 40
GCC Global range and average
Note: The data represents the top two concerns selected by respondents.
Source: 2024 BCG Digital Government Citizen Survey
It is therefore important for governments to be discerning capabilities across government organizations. Safeguards,
in how they use GenAI for productivity and efficiency including increased regulation and transparency and
gains, applying careful consideration and a robust, focused skill and talent development are especially
responsible AI framework10. They need to address the relevant in GCC countries, where governments are often
risks when integrating GenAI into their services or when major employers at the national level.
setting overall national AI strategies and building GenAI
9. BCG Research, Innovation Systems Need a Reboot, 2024
10. BCG Experience, Responsible AI | Strategic RAI Implementation | BCG
8 DIGITALGOVERNMENTINTHEAGEOFAI:CHAMPIONINGNEXT-GENCITIZENSERVICESINTHEGCC
03. Citizen sentiment remains
the key to digital government
service adoption
O
ver the past decade, GCC governments have First, governments must focus on maintaining high levels
successfully delivered impactful digital services of citizen satisfaction and trust, finetuning their service
to serve citizen needs, as evidenced by high levels features, functionality, and ease of use to enhance citizens’
of adoption and satisfaction. Meanwhile, citizens’ digital experiences. A comparison with 2022 data reveals
expectations continue to increase as the global technology a 5% improvement in the proportion of respondents who
landscape rapidly advances. As with any successful digital experienced no problems while using digital government
product, governments must continue to evolve their services. However, 72% of GCC survey respondents still
services to keep pace with demands for customer- encounter issues—for example, 26% experienced technical
centricity, personalization, and other enhancements difficulties completing requests, and 24% deemed the
enabled by emerging technologies. This presents overall process too long or difficult [Exhibit 6].
governments with a twofold challenge.
9 DIGITALGOVERNMENTINTHEAGEOFAI:CHAMPIONINGNEXT-GENCITIZENSERVICESINTHEGCC
Exhibit 6 | Percentage of survey respondents experiencing no issues
increased by %5 since 2022, albeit challenges remain across digital
government services
Q. Which of the following problems have you encountered while using digital government services?
Problems Encountered by % of Respondents (2022-2024)
I experienced technical difficulties or issues completing my request 3
+ %
The overall process took too long or was too difficult
share of
I couldn’t remember my username or password
respondents
who found their
I needed help but none was available at the time
service online
I didn’t have all the information or paperwork
15% 5
The service I needed is not available online + %
18%
I didn’t understand the instructions or didn’t know what to do share of
respondents
I could not find what I needed who experienced
no issues
28%
I have experienced no problem or issue 23%
0 10 20 30
2022 2024
Source: 2024 BCG Digital Government Citizen Survey.
These concerns can be addressed through strategic GenAI solutions. A crucial enabler for GCC governments
interventions. For example, further expanding Digital becoming global leaders in the AI-enabled digital service
IDentity (DID) efforts simplifies access, eliminates issues space is that citizens have shown a globally leading
like forgotten passwords, and eases navigation between (by a wide margin) level of trust, and they are comfortable
services. GenAI solutions, such as proactive service with a range of AI use cases in digital government
suggestions, can enhance overall service adoption, services [Exhibit 7].
while cutting-edge virtual chatbots and dynamic
assistance allow for personalized support and effective For example, GCC respondents are most comfortable
troubleshooting with 24/7 citizen accessibility. with customer support and engagement use cases, with
an average comfort level of 83%, which is 16% higher than
Distinguishing services based on frequency of use can help the global average of 67%. GCC respondents are similarly
governments make informed decisions about where to comfortable with government use of GenAI for tech
focus resources for improvement. This strategic approach development and operational efficiency, reporting a 19%
ensures that governments prioritize areas with the highest higher comfort level than the global average, at 80%
impact on citizen satisfaction and service efficiency while compared to 61%. For public relations and
ensuring improvement across their entire service portfolio. communications, the comfort level is 79%, also 19% higher
than the global average of 60%. These findings consistently
Perhaps most importantly going forward, governments express GCC citizens’ greater comfort with AI and GenAI
should identify top-priority use cases for deploying AI and use across multiple areas than their global counterparts.
10 DIGITALGOVERNMENTINTHEAGEOFAI:CHAMPIONINGNEXT-GENCITIZENSERVICESINTHEGCC
Exhibit 7 | GCC respondents are comfortable with a wide range of
AI/GenAI use cases, at a level significantly higher than the global average
Q. How comfortable you are with the government using AI/GenAI across use cases?
Categories AI/GenAI use cases Net comfort1 level % (GCC) Comfort GCC Vs Global
Difference in avg com level %
Multilingual citizen communication -15 82 67%
Customer Personalization of citizen services -16 81 65%
service and 16
24/7 chatbot access -13 84 71% + %pts
support
Virtual agent support -13 84 71% 67% 83% Avg Difference
Difference in avg com level %
Tech Software development -16 81 65%
development
Administrative tasks -16 81 65% 19
and operational + %pts
efficiency Service decision support -20 77 57% 80% Avg Difference
61%
Difference in avg com level %
Public
engagement Public information campaigns -17 80 63%
19
and + %pts
Social sentiment analysis -20 77 57%
communications 79% Avg Difference
60%
- Not Comfortable - GCC Average
- Comfortable - Global Average
1. Net comfort is the percentage of respondents who feel comfortable minus the respondents who feel uncomfortable with the government's use of AI/GenAI across the use cases.
Note: Responses with option "I don't know" are not represented.
Source: 2024 BCG Digital Government Citizen Survey
Governments should also assess pilot use cases for the governments optimize resources through intelligent,
greatest benefit to citizens’ experience and thus the focused staffing, where reducing manual intervention in
highest return on investment. Behind the scenes, GenAI repetitive tasks, for example, can facilitate the swift and
drives faster code development for technology solutions, effective reassignment of roles. BCG estimates that by
enabling quicker updates to meet changing user needs 2033, GenAI in the public sector workforce could yield
and resulting in more efficient service delivery and more than $65 billion in annual productivity gains
reduced administrative wait times. GenAI will also help across the GCC11.
11. BCG Article, Generative AI for the Public Sector: From Opportunities to Value, 2023
11 DIGITALGOVERNMENTINTHEAGEOFAI:CHAMPIONINGNEXT-GENCITIZENSERVICESINTHEGCC
04. Government AI:
A clear path forward
W
hile GCC citizens already report high levels of The “black box” nature of many GenAI models
trust in their governments’ use of AI, there are makes it difficult to understand how decisions are
still risks that need to be addressed. Ongoing made, reducing transparency and accountability. In
efforts to maintain and enhance safeguards and providing government services, current GenAI models
transparency should further bolster citizens’ confidence can struggle with language nuances and context,
in AI applications. The survey suggests that there is no potentially leading to misunderstandings.
single key factor that will drive a significant increase in
citizens’ trust in government use of AI, but rather a holistic To address these, specific laws and regulations
and studied set of actions, regulations, and initiatives. governing AI use can be implemented to provide
a clear legal framework that assures citizens of the
As a fairly recent technology, GenAI brings several ethical deployment of AI technologies. When asked
adoption- and design-related risks. Bias in AI algorithms which AI regulations and policy considerations would
can perpetuate and even amplify existing inequalities, increase their trust, GCC respondents were generally
leading to unfair outcomes for citizens. AI systems can on par with the overall global sentiment.
also produce “hallucinations,” generating incorrect or
nonsensical information, which can mislead users.
12 DIGITALGOVERNMENTINTHEAGEOFAI:CHAMPIONINGNEXT-GENCITIZENSERVICESINTHEGCC
It is interesting to note that introducing specific laws and communication on the benefits and risks of AI (30%),
regulations on government use of AI, and establishing mandatory reporting of adverse AI-related incidents (27%),
clear rules on how Personally Identifiable Information and disclosure of AI use in government process (27%), were
(PII) is used and protected, were slightly higher on the also important to citizens in the GCC.
priority agenda for citizens globally, while gaining 37%
and 32% average support respectively from citizens On a closing note, governments today enjoy strong
in the GCC [Exhibit 8]. support for responsible AI adoption. Ninety-four percent
of respondents in the GCC and 90% of respondents
Nonetheless, 30% of GCC citizens believe that applying globally believe that implementing at least one of the
fairness and safety measures through rigorous testing, and proposed safety and regulatory measures could increase
regular AI systems audits, are critical to build trust. Other their trust in government AI adoption.
initiatives to increase transparency, such as open
Exhibit 8 | 94% of GCC respondents say government can build trust in
AI/GenAI with key actions related to regulation, communication,
and transparency
Q. Which of the following would increase your trust in the use of AI/GenAI
by governments?
% of respondents in GCC countries (2024)
48% 47%
42% 40% 20%
34% 33%
% of GCC
37%
respondents
32% 32%
30% 29%
27%
26%
20%
28% 18% 20% 10%
15%
6%
3%
Factors that Specific laws and Rules on how Communication The application Mandatory reporting Disclosures if AI has Nothing would
increase trust regulations on how personally identifiable about the of fairness and of AI-related adverse been used in a influence my trust
AI can be used by information must be potential benefits safety measures events, breaches or government process in government to
the government safeguarded and and risks of AI incidents or decision making use AI responsibly
protected
Global average Global range Regional average
Source: 2024 BCG Digital Government Citizen Survey.
From promise to leadership: How GCC in the GCC. However, with accelerated global
governments can capitalize on progress advancements in digital services, especially in AI and
GenAI, GCC citizens’ expectations are evolving rapidly.
GCC governments have already made remarkable progress While they continue to benchmark their government
in delivering digital services with world-leading levels of services against global private-sector leaders, a notable
adoption and citizen satisfaction. They are well prepared— shift indicates that 98% now expect government digital
through supportive national strategies, matching services to rival the best private and public sector
investments, and citizen trust in their ability to do so platforms worldwide. This points to a new era of
responsibly—to expand their adoption of AI and GenAI. heightened expectations, which governments must
adapt to quickly.
Over the years, effective digital service delivery strategies
have achieved an 81% net approval score of from citizens
13 DIGITALGOVERNMENTINTHEAGEOFAI:CHAMPIONINGNEXT-GENCITIZENSERVICESINTHEGCC
Moreover, the trust GCC citizens place in their 3 Accelerate: the identification and adoption of
governments regarding AI remains an unparalleled winning AI and GenAI use cases where citizens
opportunity. Their 71% trust level in governments’ are most comfortable and where adoption can
responsible use of AI far exceeds global averages. This generate the highest impact on citizen experience
paves the way for GCC governments to further integrate and satisfaction.
AI into service delivery, leveraging this trust to introduce
transformative next-generation solutions. The challenge 4 Focus: on the trust-building actions citizens favor by
now is for governments to act swiftly and responsibly, setting a balanced strategic course, collaborating with
capitalizing on citizens’ expectations while ensuring the all relevant stakeholders to implement responsible AI
necessary safeguards to maintain and build upon their frameworks, and quickly introducing comprehensive
trust. By doing so, GCC governments can secure their guardrails to advance AI adoption.
leadership position in the digital public sector landscape
in a new responsible AI-enabled era. The time to take action is now. Tremendous opportunities
await the GCC governments that can capitalize on their
We see four steps to guide governments’ path forward: current digital service momentum, keep pace with AI
and other fast-moving technology, to exceed citizen
1 Innovate: continuously to keep pace with heightened expectations and maintain their trust and confidence.
citizen expectations and a rapidly advancing global
technology landscape.
2 Prioritize: addressing lingering service usability
and user experience issues to maintain high levels
of satisfaction. This includes diversifying investment
and improvement efforts across their entire portfolio
of digital services.
14 DIGITALGOVERNMENTINTHEAGEOFAI:CHAMPIONINGNEXT-GENCITIZENSERVICESINTHEGCC
About the Authors
Rami Mourtada is a Partner and Director in the firm’s Dr. Lars Littig is a Managing Director & Partner at the
Middle East offices. He leads BCG’s digital transformation firm’s Middle East offices. He is the EMESA Leader of
topic in the Middle East and is part of BCG’s Center for BCG’s Center for Digital Government. You may contact
Digital Government. You may contact him by email at him by email at [email protected].
[email protected].
Semyon Schetinin is a Managing Director and Partner in
Miguel Carrasco is a Managing Director and Senior BCG’s Middle East offices. He has vast experience in
Partner in BCG’s Sydney office. He is the global leader and technology & digital topics in Public Sector, TMT and other
founder of BCG’s Center for Digital Government and can industries. You may contact him by email at
be reached by email at [email protected]. [email protected].
Akshara Baru is a Senior Knowledge Analyst in BCG’s
London office and provides worldwide support for the
Digital Government practice.
For Further Contact Acknowledgments
If you would like to discuss this report, please contact The authors are grateful to their BCG colleagues, whose
the authors. insights and experience contributed to this report. In
|
286 | bcg | BCG_Most-Innovative-Companies-2023_Reaching-New-Heights-in-Uncertain-Times_May-2023.pdf | Most Innovative Companies 2023
Reaching New Heights
in Uncertain Times
May 2023
Contents
01 T he Formula for Innovation from 14 How Early Winners Are
Leading Companies Unlocking AI’s Potential
• What Winners Are Doing • From Implementation to Impact
• Bosch: A Culture of Innovation • H&M: Leveraging AI and Human Input
for Amplified Intelligence
• The 50 Most Innovative Companies
of 2023 • Four Impact Success Factors
• Innovation and Performance • Moderna: Pioneering AI-Driven
Innovation in the Fight Against Cancer
• Samsung: Leading the
Commercialization of Consumer Tech
19 Methodology
07 A Downturn Ups the Stakes in
Innovation
24 About the Authors
• A New Outlook?
• Investing with Focus
• McDonald’s: Driving Growth with 25 Acknowledgments
Digital Innovation
• Good Advice (Updated) Stands
the Test of Time
The Formula for Innovation from
Leading Companies
By Justin Manly, Michael Ringel, Amy MacDougall, Will Cornock, Johann Harnoss, Konstantinos Apostolatos,
Ramón Baeza, Ryoji Kimura, Michael Ward, Beth Viner, and Jean-Manuel Izaret
F
or the third straight year, the evidence is mounting: In this year’s Most Innovative Companies report, we
companies that both prioritize innovation and make examine what innovation-ready leaders (those that are ready
sure that they are ready to act are widening the gap to develop product, process, and business model innovations
over less capable competitors. The leaders at these firms that can deliver sustainable impact) are doing to pull ahead
are consistently delivering new products, entering new and how innovation is building their resilience to economic
markets, and establishing new revenue streams. The uncertainty and fueling their pursuit of lower emissions. In
laggards struggle to make headway beyond incremental “A Downturn Ups the Stakes in Innovation,” we explore how a
improvements. potential downturn in 2023 is evoking a much different
response than did the 2009 financial crisis, especially among
This year, the findings from our global innovation survey leading firms. In “How Early Winners Are Unlocking AI’s
dovetail with other new BCG research showing that compa- Potential,” we dig into the critical role of artificial intelligence
nies built for the future share a common set of attributes (AI) in innovation as in many other areas of business today.
that enable them to exhibit superior performance, be more
resilient to shocks and disruptions, and exploit innovation
faster for value-creating growth. In addition to people and
technology capabilities (including, importantly, AI), one of
these attributes is an innovation-driven culture.
BOSTON CONSULTING GROUP 1
What Winners Are Doing ambitions. Last year, we examined companies’ readiness in the
context of climate and sustainability (C&S), which two-thirds of
Despite global economic uncertainty, innovation rose as a top companies ranked as a top corporate priority. Only about one
corporate priority in 2023, with 79% of companies ranking it in five companies was ready to take effective action.
among their top three goals. (See Exhibit 1.) This is up from
75% in 2022 and close to 2019’s high of 82%. The top areas of This year, two out of three ready companies rank innovation as
innovation emphasis are new products and exploring adja- their top priority, and 90% expect to increase spending—
cent business models. Cost is a key driver for 62% of respon- almost all by more than 10%. (See Exhibit 2.) Moreover, while
dents and a top reason for innovation. Companies remain all companies on average expect to allocate more money
bullish on their innovation prospects: 42% expect to signifi- toward incremental innovations close to the core (an under-
cantly increase spending this year, a jump of 16 percentage standably conservative approach in uncertain times), ready
points over the last economic downturn in 2009. companies are allocating fully one-third of spending toward
developing breakthrough innovations. Expanding into adjacent
These are impressive figures, especially in the current macroeco- business models is also a priority in 2023. And 89% of ready
nomic and geopolitical environment. But there is also an emerg- companies prioritize C&S compared with 58% of all companies.
ing group of companies that is going much further and putting
innovation front and center in their future growth strategies. Ready companies use a wide array of strategic tools to
strengthen their innovation platforms and practices. They
Two years ago, as the world began to emerge from the access capabilities and expertise from outside their own walls,
pandemic, we observed that successful innovation takes three and they have systems in place to leverage these tools for
things: making innovation a priority, committing investment value. These companies are much more aggressive in their
and talent to it, and being ready to transform investment into use of M&A, for example, targeting innovative technologies or
results. We found that only about one company in four was processes or acquiring leaders and employees with a demon-
“innovation ready”—that is, it met all three criteria, particu- strated ability to innovate. They are also more likely to involve
larly possessing the elements of leadership and teaming that innovation experts in target analysis and selection.
enable effective execution of a company’s innovation
Exhibit 1 - Nearly 80% of Respondents Named Innovation as a the Top-Three
Priority, While Two-Thirds of Ready Companies Ranked It as Top Priority
Where does innovation, R&D, and product development rank among your Respondents who cite innovation,
company’s priorities? (%) R&D, and product development as
their company’s top priority (%)
35
82
76 77 75 79 76 77 75 75 79
72 71
66 66 66 64 65
47
32 47 42 42 46
45 52 53 53 57 53
47 43 43 39 42 65
19 40 23 23 25 26 24 24 22 22 23 30 35 23 33 33 33 30
2005 2006 2007 2008 2009 2010 2012 2013 2014 2015 2016 2018 2019 2020 2021 2022 2023 Unready Ready
Top 3 Priority Top Priority
Sources: BCG Global Innovation Survey 2023; BCG analysis.
Note: n = 1,023 for global respondents. No data for 2011 and 2017 available. Totals may not sum due to rounding. “Ready” companies are those that
are ready to develop product, process, and business model innovations that can deliver sustainable impact.
2 MOST INNOVATIVE COMPANIES 2023
Exhibit 2 - Nearly 90% of Ready Companies Plan to Increase Innovation
Spending
How will your company's innovation, R&D, and product development spending change this year in response to
macroeconomic factors? (%)
32
30
26
23
22
16
15
11
10
7
6
3
Increase Increase by Increase by Increase slightly No change Decrease
by >30% 21% to 30% 11% to 20% (5% to 10%) (<5% change)
Increase No Change Decrease
Ready Unready
Sources: BCG Global Innovation Survey 2023; BCG analysis.
Note: n = 1,023 for global respondents. Totals may not sum due to rounding. “Ready” companies are those that are ready to develop product,
process, and business model innovations that can deliver sustainable impact.
For similar reasons, they also are more likely to orchestrate or strategy is its centralized Bosch Research unit. With 1,800
participate in ecosystems, engaging with external partners, highly specialized employees, this unit generates about a
even competitors, on innovations. They determine what they quarter of all Bosch patents. Bosch Research focuses on
need, whether it’s technology, data, or something else, and enabling technologies that can be applied across The Bosch
then work out the most effective way to access it. Group, such as AIoT, which combines AI and the Internet of
Things, to move from fundamental research to actual
They drive digital innovation with a clear bias towards new product innovation and large-scale commercialization.
digital products, agile teaming, and improving customer and Bosch builds on a broad ecosystem of internal business
marketing insights. They regularly review the performance of units and external partners to generate innovation ideas.
innovation units or vehicles (such as venture capital funds,
accelerators, incubators, and R&D) and shift resources toward While three-quarters of R&D spending has been devoted to
centers of success. And they understand that effective portfolio the company’s Mobility Solutions business and topics such
governance and management, especially with respect to data as electrification, driver assistance systems, semiconductors,
transparency, are key to driving impact. and sensors, Bosch supplements internal R&D investments
with targeted acquisitions to support high-priority areas, such
as its automated driving product portfolio.
Bosch: A Culture of Innovation
In 2022 alone, the company made three investments to
The Bosch Group (number 37 on the 2023 Most Innovative acquire IP for the next generation of mobility, consistent
Companies list) states in its annual report that “the basis with its goal of making Bosch a one-stop shop for “all the
for the company’s future growth is its innovative strength.” necessary building blocks of automated driving—from
While Bosch has a special ownership structure that actuators and sensors to software and maps,” according to
facilitates long-term planning and up-front investments, Mathias Pillin, president of the Cross-Domain Computing
it is a strong culture of innovation that underpins. Solutions division.
Bosch has a global R&D organization of about 84,800 For example, Bosch’s Semiconductor Ideas to the Market
employees, 44,000 of whom are software developers, in 130 team specializes in high-frequency-processing “System-on-
locations. From 2018 through 2021, the company has Chips” used in control units for the automotive industry. Its
maintained steady R&D spending as share of sales at FiveAI unit provides a modular cloud platform designed for
between 7.6% and 8.2%. A core pillar of Bosch’s innovation building software components and development platforms
BOSTON CONSULTING GROUP 3
for safe automated driving systems, particularly supporting These new bases of advantage are rooted in superior capa-
solutions used in complex urban environments. “We want bilities, especially those related to digital, AI, and innovation.
Five to give an extra boost to our work in software develop- These capabilities are more difficult to establish but much
ment for safe automated driving,” said Markus Heyn, more enduring for two reasons. First, technology is evolving
member of the Bosch board of management and chair- rapidly, and proficiency in a technology today, such as
man of the Mobility Solutions business. Bosch’s Atlatec AI, means that as the technology grows more powerful,
team, meanwhile, creates high-resolution digital maps that a company can be faster at deploying it. Second, compa-
are critical to automated driving functionality. nies that have these capabilities benefit from a flywheel
effect: they can invent, deploy, adapt, and reinvent more
The 50 Most Innovative Companies of 2023 quickly and with greater impact than their competitors
can. They also get better at co-creating with customers
The 50 most innovative companies for 2023 are a and ecosystem partners and at democratizing the use of
geographically diverse group, roughly evenly split between data throughout their organization.
North America and the rest of the world. Europe and Asia
are well represented, and the Middle East joins the list for Samsung: Leading the Commercialization of
the first time with Saudi Aramco at number 41. (See Exhibit 3.) Consumer Tech
Auto companies held multiple positions in the 2022 list;
international energy companies hold five spots this year. Consumer electronics giant Samsung is an example of a
This may be a sign of respondents’ concerns over climate company that uses all the tools available to drive perfor-
change and the fact that they are looking to the energy indus- mance by innovating at multiple stages of the value chain.
try to be a large and creative part of the solution. In spite of the Samsung regularly brings new technology to the mass
market headwinds that they experienced in 2022, tech compa- market through a focus on component-level technology
nies continue to dominate the top 50, including the top ten. innovations and advances in scaled manufacturing. Over the
years, as its core products and markets (such as smart-
The big story, though—this year and for most of the past phones and TVs) have matured, Samsung, known for its
decade—is the ability of innovation to drive performance. dizzying array of products, has proved adept at pushing into
Since 2005, our portfolio of the 50 most innovative adjacent markets and developing new business models.
companies has outpaced the broader market in sharehold-
er returns by a significant margin—an average of 3.3 Samsung innovates along two dimensions: component-
percentage points per year. (See Exhibit 4.) level advances (improving existing technologies with inno-
vations, such as foldable phones), and adoption (increasing
Innovation and Performance accessibility to products through mass production, lower
costs, and technological advancements). The company is a
The consistent outperformance of BCG’s most innovative global innovation leader across R&D, patents, and innova-
companies compared with the broader market is one tion vehicles such as labs and incubators. It invests heavily
indication of the correlation between superior innovation in R&D, spending more than $17 billion (9% of sales) in
capabilities and performance. There are others. 2021 alone, making it the largest non-US R&D spender.
Boasting about 10,000 researchers and developers dedicat-
New BCG research into the shifting drivers of performance ed to the development of future tech, the company has
and sustainable competitive advantage show that the developed a robust patent portfolio: it was granted 6,300
ability to innovate consistently over time is fast rising in patents in 2022, the most in the US.
importance. Traditional markets have plateaued, and
growth, which accounts for 60% to 70% of shareholder As Samsung has developed new products and sought out
returns in the medium term, is found primarily in new new markets, it has moved from displays and electronic
markets, including those created by technology disruption, components into robotics, smart home products, connected
such as e-commerce, streaming media, cloud-based inter- cars, medical equipment, virtual assistants, and 5G
actions, mobility solutions, and smart energy solutions. A connectivity. The company has captured significant shares
small number of companies that have embedded the of the global market for smartphones, QLED TVs, and
capabilities that enable them to exploit innovation for IoT products.
value-creating growth are widening the performance gap
with their competitors and generating shareholder returns The connection between innovation and both growth
almost three times greater than those of the S&P 1200. and advantage is becoming stronger than ever. Winners
understand this and invest in their innovation engines
accordingly. They will continue to widen their lead over
others until the laggards reset their own priorities and
investments for the future.
4 MOST INNOVATIVE COMPANIES 2023
Exhibit 3
The 50 Most Innovative Companies of 2023
Ranking
1–10 1 Apple 2 Tesla (+3) 3 Amazon 4 Alphabet 5 Microsoft 6 Moderna 7 Samsung 8 Huawei 9 BYD 10 Siemens
(-3) (+1) (-1) Company (+10)
11–20 Pfizer (+7) J&J (+15) SpaceX Nvidia (+1) ExxonMobil Meta (-5) Nike (-5) IBM (-8) 3M (+18) Tata Group
21–30 Roche Oracle (-3) BioNTech Shell Schneider P&G (+8) Nestlé (+22) General Xiaomi (+2) Honeywell
Electric Electric (+1)
31–40 Sony (-22) Sinopec Hitachi (+6) McDonald's Merck ByteDance Bosch (-11) Dell (-24) Glencore Stripe
41–50 Saudi Coca-Cola Mercedes- Alibaba Walmart PetroChina NTT Lenovo (-24) BMW Unilever
Aramco (-6) Benz Group1 (-22) (-32)
xxx - Returnee xxx - New entrant
Sources: BCG Global Innovation Survey 2023; BCG analysis.
Note: +/- indicates change from 2022 MIC ranking.
1Mercedes-Benz Group was previously identified as Daimler.
BOSTON CONSULTING GROUP 5
Exhibit 4
Innovators Create Value
Total shareholder return
2005 = index 100
482
+3.3
percentage
points
Annual TSR
287
100
2004 2007 2010 2013 2016 2019 2022
BCG MIC 50 MSCI World Index
Sources: BCG Innovation Journey Analytics Database; CapitalIQ.
Note: Total shareholder return (TSR) performance of publicly listed MIC top 50 companies. Chart compares their one-year TSR performance for
that year against global performance index (MSCI World). We reweight the MIC 50 basket annually to reflect changes to the list. BCG MIC 50 has
outperformed the index in 12 of 17 years.
6 MOST INNOVATIVE COMPANIES 2023
A Downturn Ups the Stakes in
Innovation
By Justin Manly, Michael Ringel, Amy MacDougall, Will Cornock, Johann Harnoss, Konstantinos Apostolatos,
Ramón Baeza, Ryoji Kimura, Michael Ward, Beth Viner, and Jean-Manuel Izaret
F rom an innovation perspective, this downturn (if it A New Outlook?
occurs) looks to be very different from others. During
the last recession, in 2009, companies reined in What’s going on in 2023? For one thing, as growth has
spending. The innovation plans we reported on in that slowed in core markets, the importance of being able to
year’s Most Innovative Companies report reflected that innovate new products and services that carry companies
belt tightening. Less than two-thirds of companies ranked into new markets with new business models has increased.
innovation as a top-three priority that year, and only 58% We reported in 2021 that most companies are at least
planned to increase spending. Nearly 15% expected to cut gradually or partially altering their core business models,
innovation investment. This year, by contrast, 79% of com- with digital opportunities and the sustainability imperative
panies see innovation as a top-three priority—15 points as two key forces driving the change. This dynamic
more than in 2009—and 66% plan to increase spending, continues today.
42% by more than 10%. (See Exhibit 5.)
More companies are building adjacent and new-
frontier business models to serve as growth engines. Many
firms are also planning to increase their spending on such
tools as M&A, innovation labs, and open innovation ecosys-
tems, despite the possible downturn. (See Exhibit 6.)
BOSTON CONSULTING GROUP 7
Exhibit 5 - Innovation Priority and Spending Plans Are Much Stronger in
2023 Than They Were in 2009
Where does innovation, R&D, and product development rank How will your company's innovation, R&D, and product
among your company’s priorities? (%) development spending change this year in response to
macroeconomic factors? (%)
100 10 3 100 5 6
18 9 7
80 80
26 20
28
60 +15pp 60
46 24
39
40 40 32
20 +8pp 20 +16pp 42
33
25 26
0 0
2009 2023 2009 2023
Increase significantly (>10%) Decrease slightly (-1% to -10%)
Top priority Top 3 Top 10 Not on the list
Increase slightly (1% to 10%) Decrease significantly (> -10%)
Stay roughly the same
Sources: BCG Global Innovation Survey 2023; BCG analysis.
Note: n = 1,023 for global respondents. pp = percentage points.
Companies are raising their commitment to innovation, In addition, the old adage about downturns separating
although many are not improving their capabilities as fast as winners and losers may be gaining real traction this time
they would like. Last year, for example, nearly 80% (39) of around. Top performers realize that playing a game of wait
BCG’s 50 most innovative companies ranked among the top and see can easily backfire in unstable times, as it gives
climate and sustainability (C&S) innovators, according to forward-looking competitors more time to position them-
global peer votes. Fully 60% of high-emitting companies were selves to win. For example, leaders know that building
targeting deep-tech innovation, and deep tech was the medium- to long-term resilience requires ongoing focus on
number one or number two innovation focus for those firms. C&S: 89% of innovation leaders cite it as a top-three priori-
In addition, many more committed C&S innovators are ty, and 49% of all companies have confidence in their C&S
leveraging external innovation vehicles that are typically used investment decisions (up from just 23% last year). Among
for longer term or more technologically advanced solutions. those prioritizing C&S, average “C&S readiness” increased
to 37% in 2023 from 28% in 2022, based on BCG’s innova-
Separate BCG research has shown that a group of companies tion to impact (i2i) parameters.
representing about 25% of the S&P 1200 have put in place
the capabilities that enable them to pivot from shoring up the
digital basics of their value chains to focusing on growth from
innovation. Some are on the leading edge of disruption in
their sectors and demonstrating considerable resilience in the
face of uncertainty. These companies are delivering impres-
sive results, far outpacing their peers on such key metrics as
shareholder returns and revenue and earnings growth.
8 MOST INNOVATIVE COMPANIES 2023
Exhibit 6
Companies Expect to Increase Spending On Key
Innovation Enablers, Even in the Face of a Downturn
How do you anticipate the use of innovation vehicles to change in response to macroeconomic factors such as a potential
downturn, inflation, or uncertainty? (%)
19 14 13 15 15 22 18
46 53 55 52
58
55 60
35 33 32 33
26 23 22
M&A Digital and Open innovation R&D Accelerator CVC fund Incubator
innovation lab ecosystem labs
Increase spending Maintain spending Decrease spending
Sources: BCG Global Innovation Survey 2023; BCG analysis.
Note: n = 1,023 for global respondents.
BOSTON CONSULTING GROUP 9
Investing with Focus Microsoft (which has ranked among the top five innovative
companies since the first edition of this report in 2005) is famous
This year’s survey found a significant percentage of companies for using M&A as well as partnerships and alliances to fill strate-
that are not only prioritizing innovation but investing in it. They gic needs that further its innovation agenda. The most recent
are also focusing their investments for greater impact and example is the company’s investments in OpenAI and the
managing their portfolios for improved results, perhaps directly integration of ChatGPT into multiple Microsoft product offerings.
in response to economic uncertainty or the prospect of turmoil.
They understand that innovation leads to advantage, or as Eric Another example is Siemens (number 10 this year and in the top
Schmidt, former chairman and CEO of Google, recently 50 for 13 of the previous 16 years), which uses M&A in various
observed in Foreign Affairs, “The main reason innovation now ways. In 2018, it spun off 25% of its Siemens Healthineers medi-
lends such a massive advantage is that it begets more innovation.” cal device business to spur entrepreneurial independence.
Siemens Healthineers has taken advantage of the flexibility to
Innovating through uncertainty requires tough prioritization pursue big bets in health care, continuing through the pandemic.
around portfolio management, rigorous governance, investment In August 2020, it announced the acquisition of longtime partner
in M&A opportunities, and the continuous building of talent and Varian Medical Systems for $16.4 billion. The Varian acquisition
internal capabilities. In practice, this means companies should positioned Siemens Healthineers as the player with the most
focus on five things. comprehensive integrated cancer care portfolio, across screening,
diagnostics, and treatment. It has allowed the company to
M&A. Leaders look to acquire missing technologies, capabilities, realize innovation synergies, combining Siemens’ imaging tech-
and talent. Our research found that innovation-ready companies nology with Varian’s therapeutic technology and AI to enhance
(those that are ready to develop product, process, and business existing products and create new ones. For example, Siemens
model innovations that can deliver sustainable impact) use a Healthineers’ new radiotherapy product combines imaging
wide array of strategic tools to strengthen their platforms and capability with AI to do rapid assessments and real-time optimi-
practices. They are much more aggressive in their use of M&A to zation of treatment while patients are receiving therapy.
further objectives by accessing new technologies and processes
or acquiring leaders and employees with a demonstrated ability
to innovate. (See Exhibit 7.) They are also more likely to involve
innovation experts in target analysis and selection.
Exhibit 7 - Companies Use M&A to Access Innovative Technologies and
Processes, and They Involve Innovation Experts in Target Selection
What role do M&As play in your company's innovation strategy during times of downturn,
inflation, or uncertainty (select one)? (%)
+3
+23
-20
-14
+7
36
33 32
26
22
17
9 10 8
6
Acquire innovative Include innovation Acquire leaders M&A does not Access to new markets
technologies or experts in our target and employees with play a significant role
processes screening and due demonstrated ability
diligence process to innovate
All companies Ready companies
Sources: BCG Global Innovation Survey 2023; BCG analysis.
Note: n = 1,023 for global respondents. ”Ready” companies are those that are ready to develop product, process, and business model innovations
that can deliver sustainable impact.
10 MOST INNOVATIVE COMPANIES 2023
Exhibit 8 - Ready Companies Lean Away from the Core Toward
Breakthrough Innovations
All companies Ready companies
100%
47 41 49 45 38 33 35 33
32 32 32 31 31
30 29 29
23 28 22 26 30 35 34 36
Past year (2022) Expected allocation in Past year (2022) Expected allocation in
uncertainty/downturn uncertainty/downturn
Distance from core: Near-in/extension New to company (adjacencies) New to world (breakthrough)
Level of advantage: Near-in/sustaining Incremental Distruptive
Source: BCG Global Innovation Survey 2023; BCG analysis.
Note: n = 1,023 for global respondents. Totals may not sum due to rounding. “Ready” companies are those that are ready to develop product,
process, and business model innovations that can deliver sustainable impact.
Portfolio Prioritization. Innovation-ready companies Portfolio Management. Embedding effective portfolio
emphasize breakthrough or disruptive innovations, while less- management and governance helps ensure ROI. The availabili-
sophisticated innovators allocate resources more heavily to ty of end-to-end tracking tools for innovation portfolios actually
“near-in” or sustaining innovation. Significant percentages of declined from 2022 to 2023, and only 38% of companies
all the companies in our 2023 survey are increasing their report a strong reliance on metrics to inform decision making
focus on digital product innovation (34%), adjacent new and governance. Less than a quarter of survey respondents
business models (30%), lowering costs (23%), and new ways of said they have successfully implemented clear KPI and deci-
working (30%). Innovation-ready companies are shifting their sion-making criteria to make portfolio decisions, and only 23%
allocation of resources away from incremental innovations use stage-gate processes with clearly defined decision criteria,
that sustain current positions or advantages toward break- reporting requirements, and performance metrics. By
through or disruptive innovations that create new markets or contrast, almost all innovation-ready companies employ
revenue streams. (See Exhibit 8.) In the event of a downturn, end-to-end tracking to assess progress and make informed
ready companies expect to allocate 50% more investment decisions about an initiative’s value. (See Exhibit 9.)
toward breakthrough innovations (34% versus 22% for other
companies) and almost that much more toward disruptive Overall revenue growth and customer satisfaction remain
innovations (36% versus 26%). Ready companies are planning the top metrics for innovation success, used by 41% and
to increase downturn spending in both areas while others 35% of companies, respectively. These are the same met-
tread water or retrench. rics, used by roughly similar percentage of companies, as
we recorded during the last recession in 2009. Impact on
environmental, social, and governance goals jumped into
the top-five success metrics tracked this year and is the
most common among ready companies.
BOSTON CONSULTING GROUP 11
Exhibit 9 - Nearly All Ready Companies Have Implemented End-to-End
Portfolio Tracking
Have you implemented a holistic view of the portfolio What metrics does your company use in innovation, R&D,
with end-to-end tracking available centrally as a source and product development (select up to three)? (%)
of truth to guide portfolio decisions? (%)
41
Overall revenue growth
33
35
Customer satisfaction
38
94
26
Impact on ESG goals
43
42 43
26
14 Return on innovation spending
23
Implemented Implemented 22
with impact1 Margin accretion
25
All companies Ready companies
Sources: BCG Global Innovation Survey 2023; BCG analysis.
Note: n = 1,023 for global respondents. “Ready” companies are those that are ready to develop product, process, and business model innovations
that can deliver sustainable impact.
1 “Implemented with impact” includes the subset of companies implementing portfolio transparency that also saw impact from those efforts.
Data, Targets, and Collaboration. Ready companies The war for talent is a perennial issue. The job market is
emphasize use of fundamental tools and ensure greater still strong, but it has softened in many parts of the world,
data transparency, clearer portfolio targets, and more meaning it may be easier now to build or strengthen
collaboration. Three-quarters have full data transparency internal teams. The same conditions that provide
to support decision making, compared with only 35% of opportunities for M&A also offer the opportunity to
not-ready companies. Almost 60% have clear portfolio acquire qualified, innovation-focused talent.
targets, and more than half use regular portfolio meetings
to assess process. (See Exhibit 10.)
McDonald’s: Driving Growth with Digital
Innovation-Focused Talent and Culture. For many Innovation
“almost ready” companies, talent and culture is the dimen-
sion holding them back from realizing the full potential of Consider the case of McDonald’s (number 34 on the 2023
their innovation function. In fact, companies that are top 50 list), a restaurant industry frontrunner in technology
almost ready (according to BCG’s i2i assessment) lag ready innovation and investment, which has combined many of
companies more on talent and culture than on any other these practices to bolster its leading position in the indus-
dimension. Ready companies focus on an innovation- try. McDonald’s spends heavily on innovation through
focused culture and talent pipeline. They are three to four partnerships, labs, digital tools, and acquisitions. It was
times as likely as their almost-ready counterparts to have investing in AI (through M&A) as early as 2019 when it
successfully implemented a strong innovation-focused acquired Apprente, which develops voice-based, conversa-
recruiting and talent acquisition foundation across all tional technology, and personalization startup Dynamic
stages of the talent pipeline. And, as we will see in “How Yield to better customize the drive-through experience.
Early Winners Are Unlocking AI’s Potential,” the third
article in this year’s report, companies that realize impact
from AI have more than three times as many people
dedicated to innovation as those who don’t.
12 MOST INNOVATIVE COMPANIES 2023
Exhibit 10 - Ready Companies Rely on Full Data Transparency, Clear
Portfolio Targets, and Regular Portfolio Meetings
How do you support effective decision making on your innovation portfolio? (%)
+37 +20 +13
72
58 51
35 38 38
Full data transparency Clear portfolio targets Regular portfolio meetings
Ready companies Unready companies
Sources: BCG Global Innovation Survey 2023; BCG analysis.
Note: n = 1,023 for global respondents. “Ready” companies are those that are ready to develop product, process, and business model innovations
that can deliver sustainable impact.
When the onset of the pandemic threated its in-restaurant • Go bargain hunting. Find attractive acquisitions and
business, McDonald’s launched the “Accelerating the build innovation into the M&A process by including
Arches” growth strategy, doubling down on digital, drive- innovation expertise on M&A teams and targeting new
through, and delivery. Digital innovations cut 30 seconds technologies and talent.
off drive-through ordering times during the pandemic,
enabling the restaurant chain to move more cars through • Acquire intellectual property (I |
287 | bcg | ceo-guide-to-ai-revolution.pdf | The CEO’s Guide to the Generative AI
Revolution
MARCH 07, 2023
By François Candelon, Abhishek Gupta, Lisa Krayer, and Leonid Zhukov
READING TIME: 15 MIN
The release of ChatGPT in late 2022 created a groundswell of interest in generative AI. Within hours,
users experimenting with this new technology had discovered and shared myriad productivity hacks.
In the weeks and months since, organizations have scrambled to keep pace—and to defend against
unforeseen complications. Some organizations have already adopted a more formal approach,
creating dedicated teams to explore how generative AI can unlock hidden value and improve
efficiency.
© 2025 Boston Consulting Group 1
For CEOs, however, generative AI poses a much bigger challenge. Today’s focus might be on
productivity gains and technical limitations, but a revolution in business-model innovation is coming.
Much as Mosaic, the world’s first free web browser, ushered in the internet era and upended the way
we work and live, generative AI has the potential to disrupt nearly every industry—promising both
competitive advantage and creative destruction. The implication for leaders is clear: today’s
breathless activity needs to evolve into a generative AI strategy owned by the C-suite.
This is no small task, and CEOs—who are likely several steps removed from the technology itself—
may feel uncertain about their next move. But from our perspective, the priority for CEOs isn’t to fully
immerse themselves in the technology; instead, they should focus on how generative AI will impact
their organizations and their industries, and what strategic choices will enable them to exploit
opportunities and manage challenges. These choices are centered on three key pillars:
Each pillar raises an urgent question for CEOs. What innovations become possible when every
employee has access to the seemingly infinite memory generative AI offers? How will this technology
change how employees’ roles are defined and how they are managed? How do leaders contend with
the fact that generative AI models may produce false or biased output?
Clearly, generative AI is a rapidly evolving space, and each of the pillars above involves short- and
long-term considerations—and many other unanswered questions. But CEOs need to prepare for
the moment when their current business models become obsolete. Here’s how to strategize for that
future.
Potential: Discover Your Strategic Advantage
AI has never been so accessible. Tools such as ChatGPT, DALL-E 2, Midjourney, and Stable Diffusion
allow anyone to create websites, generate advertising strategies, and produce videos—the
possibilities are limitless. This “low-code, no-code” quality will also make it easier for organizations
to adopt AI capabilities at scale. (See “The Functional Characteristics of Generative AI.”)
THE FUNCTIONAL CHARACTERISTICS OF GENERATIVE AI
The transformative potential of generative AI can be summed up by three key
functional characteristics.
© 2025 Boston Consulting Group 2
Seemingly “Infinite” Memory and Pattern Recognition. Because generative AI is
trained on huge amounts of data, its memory can appear infinite. For example,
ChatGPT has been trained on a massive portion of publicly available information on
the internet. To put this in context, as of 2018 the internet generated 2.5 quintillion
bytes of new data daily, according to Domo—the equivalent of 1.2 quintillion words.
That number is likely much higher today. Generative AI can also create connections
(or recognize patterns) between distant concepts in an almost human-like manner.
Low-Code, No-Code Properties. When describing the impact of ChatGPT, Andrej
Karpathy, a founding member of OpenAI, said “the hottest new programming
language is English.” That’s because generative AI’s natural-language-processing
interface allows nonexperts to create applications with little or no coding required.
By contrast, coding assistant systems such as Github Copilot still require competent
programmers to operate them.
Lack of a Credible Truth Function. Generative AI’s “infinite” memory can become
an infinite hallucination. In reality, the level of error in today’s generative AI systems
is an expected characteristic that makes it useful for generating new ideas and
content. But because generative AI does not use logic or intelligent thought, instead
predicting the most probable next word based on its training data, it should only be
used to generate first dras of content.
Companies are working to make generative AI’s output significantly more reliable by
using an approach known as reinforcement learning with human feedback; other
approaches that combine generative AI with traditional AI and machine learning
have also been considered. Improvements to generative AI are expected soon, with
some predicting that it will be able to produce final-dra content by 2030.
The immediate productivity gains can greatly reduce costs. Generative AI can summarize documents
in a matter of seconds with impressive accuracy, for example, whereas a researcher might spend
hours on the task (at an estimated $30 to $50 per hour).
But generative AI’s democratizing power also means, by definition, that a company’s competitors will
have the same access and capabilities. Many use cases that rely on existing large language model
1
(LLM) applications—such as productivity improvements for programmers who use Github Copilot
© 2025 Boston Consulting Group 3
and for marketing content developers who use Jasper.ai—will be needed just to keep pace with other
organizations. But they won’t offer differentiation, because the only variability created will result
from users’ ability to prompt the system.
Identify the Right Use Cases
For the CEO, the key is to identify the company’s “golden” use cases—those that bring true
competitive advantage and create the largest impact relative to existing, best-in-class solutions.
Such use cases can come from any point along the value chain. Some companies will be able to drive
growth through improved offerings; Intercom, a provider of customer-service solutions, is running
pilots that integrate generative AI into its customer-engagement tool in a move toward automation-
first service. Growth can also be found in reduced time-to-market and cost savings—as well as in the
ability to stimulate the imagination and create new ideas. In biopharma, for example, much of
today’s 20-year patent time is consumed by R&D; accelerating this process can significantly increase
a patent’s value. In February 2021, biotech company Insilico Medicine announced that its AI-
generated antifibrotic drug had moved from conceptualization to Phase 1 clinical trials in less than
30 months, for around $2.6 million—several orders of magnitude faster and cheaper than traditional
drug discovery.
Once leaders identify their golden use cases, they will need to work with their technology teams to
make strategic choices about whether to fine-tune existing LLMs or to train a custom model. (See
Exhibit 1.)
© 2025 Boston Consulting Group 4
Fine-Tuning an Existing Model. Adapting existing open-source or paid models is cost effective—in
a 2022 experiment, Snorkel AI found that it cost between $1,915 and $7,418 to fine-tune a LLM
model to complete a complex legal classification. Such an application could save hours of a lawyer’s
time, which can cost up to $500 per hour.
Fine-tuning can also jumpstart experimentation, whereas using in-house capabilities will siphon off
time, talent, and investment. And it will prepare companies for the future, when generative AI is likely
to evolve into a model like cloud services: a company purchases the solution with the expectation of
achieving quality at scale from the cloud provider’s standardization and reliability.
But there are downsides to this approach. Such models are completely dependent on the
functionality and domain knowledge of the core model’s training data; they are also restricted to
available modalities, which today are comprised mostly of language models. And they offer limited
options for protecting proprietary data—for example, fine-tuning LLMs that are stored fully on
premises.
Training a New or Existing Model. Training a custom LLM will offer greater flexibility, but it comes
with high costs and capability requirements: an estimated $1.6 million to train a 1.5-billion-
parameter model with two configurations and 10 runs per configuration, according to AI21 Labs. To
put this investment in context, AI21 Labs estimated that Google spent approximately $10 million for
© 2025 Boston Consulting Group 5
2
training BERT and OpenAI spent $12 million on a single training run for GPT-3. (Note that it takes
multiple rounds of training for a successful LLM.)
These costs—as well as data center, computing, and talent requirements—are significantly higher
than those associated with other AI models, even when managed through a partnership. The bar to
justify this investment is high, but for a truly differentiated use case, the value generated from the
model could offset the cost.
Plan Your Investment
Leaders will need to carefully assess the timing of such an investment, weighing the potential costs of
moving too soon on a complex project for which the talent and technology aren’t yet ready against
the risks of falling behind. Today’s generative AI is still limited by its propensity for error and should
primarily be implemented for use cases with a high tolerance for variability. CEOs will also need to
consider new funding mechanisms for data and infrastructure—whether, for example, the budget
should come from IT, R&D, or another source—if they determine that custom development is a
critical and time-sensitive need.
The “fine-tune versus train” debate has other implications when it comes to long-term competitive
advantage. Previously, most research on generative AI was public and models were provided through
open-source channels. Because this research is now moving behind closed doors, open-source
models are already falling far behind state-of-the-art solutions. In other words, we’re on the brink of
a generative AI arms race. (See “The Future State of the LLM Market.”)
THE FUTURE OF THE LLM MARKET
Until recently, most generative AI research has been publicly accessible. But many
companies are choosing to stop or delay publishing their research findings and are
keeping model architectures as proprietary knowledge. (For example, GPT-2 was
open-source but GPT-3 is proprietary.)
The next improvements to generative models with vast number of users will likely
come from logs of their user interaction, giving these models a significant
competitive advantage over new entrants. This reality, combined with the heavy data,
infrastructure, and talent costs required to train LLMs, means that the LLM market
has both economy and quality of scale. Advances in generative AI therefore might be
limited to large companies, while the democratization of AI development for small
and medium-sized enterprises could be limited to nondifferentiated use cases.
© 2025 Boston Consulting Group 6
The jury is still out, but this dynamic appears comparable to the “search-engine
wars.” Several large companies invested heavily in search solutions, but Google’s
user-friendliness and accuracy helped set it apart from competitors. Once users
preferred Google, other engines could not keep up—because every search request
Google received made it better and smarter. Soon, all other B2C solutions faded
away. A similar winner-take-all situation could play out in the LLM market, with the
big, early entrants eventually owning the models and having full control over access.
A winner-take-all situation could play out in the LLM market.
It’s worth noting, however, that Google did not achieve the same level of success in
the enterprise search market, which has unique requirements and challenges
compared to B2C. At the enterprise level, search-engines lack the scale to build
domain expertise and lack the volume of user data to build that capability. Similarly,
businesses will get the most value out of LLMs that are trained on their proprietary
data and that have modalities that drive unique use cases. This could make it
difficult for any single player to dominate the B2B market.
There is also the potential for companies and governments to fund open-source
models to keep them state of the art—similar to how IBM funded Linux.
These market dynamics have key implications for CEOs as they make customization
and implementation decisions:
• It is unlikely that any single LLM provider will dominate the B2B market; the key
for companies is to find large models with the modality and functionality that
match their golden use cases or use cases that require sensitive data.
• While training LLMs is an option for large businesses, the quality of scale could
make purchasing solutions more reliable (similar to cloud).
© 2025 Boston Consulting Group 7
• If choosing to train in-house, be wary of relying too much on individual
researchers. If only a small number of people have the expertise to advance and
maintain the model, this will cause a single point of failure if those researchers
choose to leave.
As research accelerates and becomes more and more proprietary, and as the algorithms become
increasingly complex, it will be challenging to keep up with state-of-the-art models. Data scientists will
need special training, advanced skills, and deep expertise to understand how the models work—their
capabilities, limitations, and utility for new business use cases. Large players that want to remain
independent while using the latest AI technology will need to build strong internal tech teams.
People: Prepare Your Workforce
Like existing forms of artificial intelligence, generative AI is a disruptive force for humans. In the near
term, CEOs need to work with their leadership teams as well as HR leaders to determine how this
transformation should unfold within their organizations—redefining employees’ roles and
responsibilities and adjusting operating models accordingly.
Redefine Roles and Responsibilities
Some AI-related shis have already occurred. Traditional AI and machine-learning algorithms
(sometimes incorrectly referred to as analytical AI), which use powerful logic or statistics to analyze
data and automate or augment decision making, have enabled people to work more autonomously
and managers to increasingly focus on team dynamics and goal setting.
Now generative AI, in its capacity as a first-dra content generator, will augment many roles by
increasing productivity, performance, and creativity. Employees in more clerical roles, such as
paralegals and marketers, can use generative AI to create first dras, allowing them to spend more
of their time refining content and identifying new solutions. Coders will be able to focus on activities
such as improving code quality on tight timelines and ensuring compliance with security
requirements.
Of course, these changes cannot (and should not) happen in a vacuum. CEOs need to be aware of
the effect that AI has on employees’ emotional well-being and professional identity. Productivity
improvements are oen conflated with reduction in overall staff, and AI has already stoked concern
among employees; many college graduates believe AI will make their job irrelevant in a few years.
But it’s also possible that AI will create as many jobs as it will displace.
The impact of AI is thus a critical culture and workforce issue, and CEOs should work with HR to
understand how roles will evolve. As AI initiatives roll out, regular pulse checks should be conducted
© 2025 Boston Consulting Group 8
to track employee sentiment; CEOs will also need to develop a transparent change-management
initiative that will both help employees embrace their new AI coworkers and ensure employees retain
autonomy. The message should be that humans aren’t going anywhere—and in fact are needed to
deploy AI effectively and ethically. (See Exhibit 2.)
As AI adoption accelerates, CEOs need to learn as they go and use those lessons to develop a
strategic workforce plan—in fact, they should start creating this plan now and adapt it as the
technology evolves. This is about more than determining how certain job descriptions will change—
it’s about ensuring that the company has the right people and management in place to stay
competitive and make the most out of their AI investments. Among the questions CEOs should ask
as they assess their company’s strengths, weaknesses, and priorities are:
• What competencies will project leaders need to ensure that individual contributors’ work is of
sufficient quality?
• How can CEOs create the optimal experience curve to produce the right future talent pipeline—
ensuring, for example, that employees at a more junior level are upskilled in AI augmentation
and that supervisors are prepared to lead an AI-augmented workforce?
• How should training and recruiting be adjusted to build a high-performing workforce now and in
the future?
Adjust Your Operating Model
© 2025 Boston Consulting Group 9
We expect that agile (or bionic) models will remain the most effective and scalable in the long term,
but with centralized IT and R&D departments staffed with experts who can train and customize
LLMs. This centralization should ensure that employees who work with similar types of data have
access to the same data sets. When data is siloed within individual departments—an all-too-
common occurrence—companies will struggle to realize generative AI’s true potential. But under the
right conditions, generative AI has the power to eliminate the compromise between agility and scale.
Because of the increased importance of data science and engineering, many companies will benefit
from having a senior executive role (for example, a chief AI officer) oversee the business and
technical requirements for AI initiatives. This executive should place small data-science or
engineering teams within each business unit to adapt models for specific tasks or applications.
Technical teams will thus have the domain expertise and direct contact to support individual
contributors, ideally limiting the distance between the platform or tech leaders and individual
contributors to one layer.
Structurally, this could involve department-focused teams with cross-functional members (for
example, sales teams with sales reps and dedicated technical support) or, preferably, cross-
departmental and cross-functional teams aligned to the business and technical platforms.
Policies: Protect Your Business
Generative AI lacks a credible truth function, meaning that it doesn’t know when information is
factually incorrect. The implications of this characteristic, also referred to as “hallucination,” can
range from humorous foibles to damaging or dangerous errors. But generative AI also presents
other critical risks for companies, including copyright infringement; leaks of proprietary data; and
unplanned functionality that is discovered aer a product release, also known as capability
overhang. (See Exhibit 3.) For example, Riffusion used a text-to-image model, Stable Diffusion, to
create new music by converting music data into spectrograms.
© 2025 Boston Consulting Group 10
Prepare for Risk
Companies need policies that help employees use generative AI safely and that limit its use to cases
for which its performance is within well-established guardrails. Experimentation should be
encouraged; however, it is important to track all experiments across the organization and avoid
“shadow experiments” that risk exposing sensitive information. These policies should also guarantee
clear data ownership, establish review processes to prevent incorrect or harmful content from being
published, and protect the proprietary data of the company and its clients.
Another near-term imperative is to train employees how to use generative AI within the scope of their
expertise. Generative AI’s low-code, no-code properties may make employees feel overconfident in
their ability to complete a task for which they lack the requisite background or skills; marketing staff,
for example, might be tempted to bypass corporate IT rules and write code to build a new marketing
tool. About 40% of code generated by AI is insecure, according to NYU’s Center for Cybersecurity—
and because most employees are not qualified to assess code vulnerabilities, this creates a
significant security risk. AI assistance in writing code also creates a quality risk, according to a
Stanford University study, because coders can become overconfident in AI’s ability to avoid
vulnerabilities.
Leaders therefore need to encourage all employees, especially coders, to retain a healthy skepticism
of AI-generated content. Company policy should dictate that employees only use data they fully
understand and that all content generated by AI is thoroughly reviewed by data owners. Generative
AI applications (such as Bing Chat) have already started implementing the ability to reference source
data, and this function can be expanded to identify data owners.
© 2025 Boston Consulting Group 11
Ensure Quality and Security
Leaders can adapt existing recommendations regarding responsible publication to guide releases of
generative AI content and code. They should mandate robust documentation and set up an
institutional review board to review a priori considerations of impact, akin to the processes for
publishing scientific research. Licensing for downstream uses, such as the Responsible AI License
(RAIL), presents another mechanism for managing the generative AI’s lack of a truth function.
Finally, leaders should caution employees against using public chatbots for sensitive information. All
information typed into generative AI tools will be stored and used to continue training the model;
even Microso, which has made significant investments in generative AI, has warned its employees
not to share sensitive data with ChatGPT.
Today, companies have few ways to leverage LLMs without disclosing data. One option for data
privacy is to store the full model on premises or on a dedicated server. (BLOOM, an open-source
model from Hugging Face’s BigScience group, is the size of GPT-3 but only requires roughly 512
gigabytes of storage.) This may limit the ability to use state-of-the-art solutions, however. Beyond
sharing proprietary data, there are other data concerns when using LLMs, including protecting
personally identifiable information. Leaders should consider leveraging cleaning techniques such as
named entity recognition to remove person, place, and organization names. As LLMs mature,
solutions to protect sensitive information will also gain sophistication—and CEOs should regularly
update their security protocols and policies.
Generative AI presents unprecedented opportunities. But it also forces CEOs to grapple with
towering unknowns, and to do so in a space that may feel unfamiliar or uncomfortable. Craing an
effective strategic approach to generative AI can help distinguish the signal from the noise. Leaders
who are prepared to reimagine their business models—identifying the right opportunities, organizing
their workforce and operating models to support generative AI innovation, and ensuring that
experimentation doesn’t come at the expense of security and ethics—can create long-term
competitive advantage.
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
© 2025 Boston Consulting Group 12
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 2025. All rights reserved.
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).
1 Large language models, also known as foundation models, are deep- learning algorithms that can
recognize, summarize, translate, predict, and generate content based on its training data. Today these
models are mostly trained on text, images, and audio, but they can also go beyond language and
images into signals, biological data, and more. Models trained on data beyond language are called
multimodal models.
2 “How Generative AI Is Changing Creative Work,” Harvard Business Review, November 14, 2022.
Authors
François Abhishek Gupta
Candelon
SENIOR SOLUTION
MANAGING DELIVERY
DIRECTOR & MANAGER,
SENIOR PARTNER; RESPONSIBLE AI
GLOBAL
Montreal
DIRECTOR, BCG
HENDERSON
INSTITUTE
Lisa Krayer Leonid Zhukov
PRINCIPAL VICE PRESIDENT,
DATA SCIENCE
Washington, DC
New York
© 2025 Boston Consulting Group 13 |
288 | bcg | leading-with-genaiai-how-vietnamese-companies-can-stay-ahead-pdf.pdf | Leading with GenAI/AI:
How Vietnamese Companies
Can Stay Ahead
November 2024
By Il-Dong Kwon, Arnaud Ginolin, Hanno Stegmann
Boston Consulting Group partners with leaders BCG X is the tech build & design unit of BCG.
in business and society to tackle their most Turbocharging BCG’s deep industry and
important challenges and capture their greatest functional expertise, BCG X brings together
opportunities. BCG was the pioneer in business advanced tech knowledge and ambitious
strategy when it was founded in 1963. Today, entrepreneurship to help organizations
we work closely with clients to embrace a enable innovation at scale. With nearly 3,000
transformational approach aimed at benefiting all technologists, scientists, programmers, engineers,
stakeholders—empowering organizations to grow, and human-centered designers located across 80+
build sustainable competitive advantage, and cities, BCG X builds and designs platforms and
drive positive societal impact. software to address the world’s most important
challenges and opportunities. Teaming across
Our diverse, global teams bring deep industry and
our practices, and in close collaboration with our
functional expertise and a range of perspectives
clients, our end-to-end global team unlocks new
that question the status quo and spark change.
possibilities. Together we’re creating the bold and
BCG delivers solutions through leading-edge
disruptive products, services, and businesses of
management consulting, technology and design,
tomorrow.
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.
Table of contents
Case Study: Transforming a
01 16
Executive Summary Leading Southeast
Page Page
Asian Bank with GenAI
Build for the Future -Establishing
03 17
Advice for Business Leaders Solid Enterprise Foundations for
Page Page
Long-term Success
04 GenAI/AI is Already 20
Where Next for AI in Vietnam?
Page Transforming Businesses Page
GenAI/AI is a Top Strategic
07 21
Priority, yet Many Companies Appendix
Page Page
are Falling Behind
GenAI/AI Landscape in
09 22
Vietnam -Challenges and About the Authors
Page Page
Opportunities Await
From Hesitation to Execution –
14
Practical Recommendations for
Page
Beginning Your GenAI/AI Journey
Executive Summary
Artificial Intelligence (AI) is unlocking new opportunities AI and GenAI are rapidly reshaping industries worldwide,
for businesses through the complementary capabilities of and have become a top priority for executives in Vietnam
traditional predictive AI and emerging generative AI and beyond. [Exhibit 1.] GenAI/AI’s potential to reshape
(GenAI). Predictive AI focuses on decision-making and business processes, enhance productivity, and create new
optimization within defined problem areas, while GenAI business models offers immense opportunities for
generates new, context-aware content and solutions, companies aiming to stay competitive. However, while the
offering a more flexible and expansive range of technology is powerful, it must ultimately be treated the
applications. same as any other tool—requiring a strategic and well-
founded approach to leverage its benefits fully.
Exhibit 1 - Generative AI will revolutionize the world – and executives want to capitalize
Source: BCG AI Radar (2024); n = 1,406 in 50 markets.
BOSTON CONSULTING GROUP 1
Despite the promise of GenAI/AI, most Vietnamese Activating this opportunity involves securing strong
companies are still in the early stages of adoption, alignment and commitment from leadership, focusing on
grappling with significant internal challenges. These high-impact use cases, setting up robust governance,
hurdles range from strategic issues like insufficient GenAI/ upskilling the workforce, ensuring robust data and security
AI expertise among leadership and unclear strategies, to practices, and leveraging external expertise and technology
implementation barriers such as data limitations and partnerships to accelerate progress and deliver immediate
talent gaps. These challenges are further compounded by impact.
external factors like inadequate computing resources and
regulatory uncertainty. Nevertheless, Vietnam’s growing However, to truly capitalize on GenAI/AI, organizations
support ecosystem—bolstered by enhanced GenAI/AI must move beyond these initial steps and embark on a
education, localized models, and a burgeoning GenAI comprehensive transformation journey. This
startup scene—provides a strong foundation for transformation requires prioritizing a mix of strategic value
overcoming these obstacles and accelerating GenAI/AI plays—Deploy, Reshape, and Invent—to maximize GenAI/
adoption in the near future. Given the potential of GenAI/ AI’s benefit. [Exhibit 2.] Additionally, building strong
AI, and the rapid pace of advancements, companies must enterprise foundations by transforming and enhancing
act now to embark on their GenAI/AI transformation technology, people, operating models, processes and risk
journey. management is crucial to fully capture the benefits of
GenAI/AI, ensure responsible and compliant AI use and
enable scalable deployment across the enterprise.
Exhibit 2 - Three strategic plays for AI integration
RESHAPE
critical functions end to INVENT
DEPLOY
end for radical efficiency new experiences,
GenAIin everyday tasks
and effectiveness offerings and business
for broad enterprise
models powered by
productivity
GenAI
In summary, GenAI/AI presents a transformative [Exhibit 3.] The time to act is now—those companies who
opportunity for Vietnamese companies, but success will embrace GenAI/AI will secure a lasting competitive edge in
hinge on a thoughtful, strategic approach that integrates the market.
technology with business goals, people and processes.
Exhibit 3 - A holistic approach integrates strategy, pilots, and multiple functional
transformation while continuously building foundations
GENAI/AI SETUP DEPLOY GenAI/AI in everyday tasks across the enterprise
Define value pools
In every business function:
Articulate visionand strategy RESHAPE • Build MVPs and re-design workflows Transformation of function 1
critical to prove value Transformation of function 2
Select priority opportunities
functions • Design future state & set targets
across Deploy, Reshape, Invent
• Cascade changes & scale E2E
Transformation of function 3
PLAYS
Build the business case INVENT New business models and products
E2E CHANGE MANAGEMENT & DELIVERY
Enable leaders and upskill
Drive adoption, engagement & culture change, leveraging behavioral science
Steer through central governance and measure impact via AI delivery office
ENTERPRISE FOUNDATIONS
Build capabilities to drive & sustain transformation
Make coordinated investments in core tech & data, people,and responsible AI
2 LEADING WITH GENAI/AI: HOW VIETNAMESE COMPANIES CAN STAY AHEAD
Advice for Business Leaders
The potential of AI and GenAI in Vietnam’s business While the journey may seem daunting, business
landscape is immense—these solutions can leaders can initiate action today by leveraging
revolutionize operations, drive innovation, and create internal and external expertise—starting with high-
substantial new avenues for growth. However, impact, quick-win initiatives and laying the key
Vietnamese companies are still in the early stages of foundations of governance, data and people.
adoption, with many GenAI/AI initiatives still at pilot Alongside these efforts, companies need a strategic
stages and primarily focused on internal use cases to approach that integrates GenAI/AI with clear
tactically improve efficiency. business objectives, robust governance, and strong
data, technology, and security frameworks. These
Realizing the full potential of GenAI/AI requires more efforts should include a special focus on talent
than just technology upgrades; it demands a holistic, development, process optimization, and
end-to-end transformation of the business and organizational redesign to achieve long-term,
organization. sustainable competitive advantage.
BOSTON CONSULTING GROUP 3
GenAI/AI is Already Transforming Businesses
GenAI represents a cutting-edge advancement in artificial algorithms are built on foundation models, trained with
intelligence, designed to create seemingly new and realistic self-supervised methods to uncover underlying patterns
content—whether it’s text, images or audio—based on across a wide array of tasks. GenAI’s capabilities can be
vast amounts of training data. [Exhibit 4.] These powerful broadly categorized into three areas:
1 Generating content and ideas. GenAIcan produce unique outputs across a variety of formats,
from crafting compelling video advertisements to discovering new proteins with antimicrobial properties.
2 Improving efficiency. GenAIstreamlines manual and repetitive tasks, such as writing emails, coding or
summarizing extensive documents
Personalizing experiences. GenAItailorscontent and information to meet specific audience needs,
3
enabling personalized customer interactions through chatbots and delivering targeted advertisements
based on individual behaviors.
Exhibit 4 - Main capabilities of GenAI that drive value
Content
Tech
generation/ Knowledge Problem solving
capabilities
Conversation transcription Summarization Ideation extraction & Insights AI agents
Interactive The creation Summarization Generation Extraction of Logical & The solving of
& dynamic of specific of large of new & structured reasoning complex tasks
engagement types of amounts of Innovation knowledge process to by planning
of information, content (e.g., information ideas, concepts from make and executing a
ideas, or text, images, or text into or designs unstructured or inferences, set of actions
questions videos, audio, shorter, more (e.g., unique semi-structured draw using a suite of
between humans codea) concise versions, product solutions, data sources conclusions, tools
Description
& AI systems, that capture the exploration of make informed
responding to key points of uncharted judgements,
questions and the content territories in and derive new
generating scientific fields) insights based on
appropriate available
responses information, data,
or knowledge
4 LEADING WITH GENAI/AI: HOW VIETNAMESE COMPANIES CAN STAY AHEAD
GenAI complements traditional Predictive AI, which excels experiences. When integrated, GenAI and Predictive AI
in decision-making tasks like descriptive analytics, unlock unprecedented opportunities for businesses,
recommendation engines, and fraud detection. While allowing them to rethink operational processes and
Predictive AI provides the analytical backbone for informed revolutionize their business models, ultimately driving
decisions, GenAI takes creativity and engagement to new greater value and innovation.
heights, automating content creation and enhancing user
Exhibit 5 - GenAI complements and co-exists with previous efforts on predictive AI,
enabling new and broader applications
Predictive AI GenAI
“Left brain” “Right brain”
• Decision making and • Content creation, qualitative
optimization reasoning, orchestration of
other systems
• Each algo constrained to
specific problem space • Multi-applications
• Limited range of • Unlimited range of
possible outputs possible outputs
Impact: 5% of employee tasks and workflows; Impact: 50% of employee tasks and workflows;
focuses on key decisions complemented with impact by Predictive AI
Technology for business has been advancing for decades, • High-value applications. Over 50% of executives
and many companies have worked to digitize their identify marketing and sales, customer operations,
processes and functions— but GenAI/AI is different. In our R&D and IT/software engineering as key areas where
research, roughly two-thirds (65%) of senior executives say GenAI/AI provides the most value. Additionally, sector-
it has the most significant disruptive potential of any specific applications offer further opportunities.
technology over the next five years. One-third of
respondents have already increased investments despite • New revenue streams. Initial GenAI/AI deployments
the challenging cost environment. Why are executives so typically focus on boosting productivity or enhancing
bullish on GenAI/AI? customer service. However, GenAI/AI also opens up
new revenue streams, with approximately 60% of
• Rapid time to value. GenAI/AI solutions can be initiatives aimed at cost reduction and 40% focused on
implemented quickly, delivering benefits within as revenue growth through increased engagement and
little as three months, particularly when using plug- customer satisfaction.
and-play applications. Basic tasks can see productivity
increases of 10% to 20%.
• Larger impact from ambitious applications. While
more complex applications—such as reshaping
business functions or inventing new models—may
take one to three years to implement, they can deliver
significantly larger benefits.
BOSTON CONSULTING GROUP 5
Several leading companies have already demonstrated the solutions. This resulted in a 5 to10% productivity
transformative power of GenAI/AI in complex scenarios. increase, a 15% to 20% reduction in job duration and
rework, and a 0.5% to 1% decrease in attrition,
• A global payment technology company enhanced fraud enhancing asset uptime and revenue.
detection by integrating GenAI/AI, improving real-time
fraud identification and reducing false positives. This • A financial information company transformed its core
allowed it to strengthen customer security and service by integrating GenAI, turning data and analysis
reinforce its reputation for innovation. into a conversational insight platform. It is projected to
generate up to US$100 million in additional revenue
• A leading global consumer goods organization used and significantly boost its financial profile.
GenAI to gain a competitive edge in consumer
engagement. The AI-driven marketing engine • An industrial goods client diversified revenue by
identified trends, automated actions and optimized commercializing its AI-powered image analytics model.
campaigns, leading to a 40% to 60% increase in Integrating AI and GenAI into inspections increased
engagement, up to 80% in cost savings and a three- defect detection efficiency 10-fold, improved data
month acceleration in campaign creation. ingestion by 160 times and boosted workflow efficiency
by 62%. Tasks are now completed 50% faster and
• A renewable energy developer improved the employee reporting is 67% quicker.
value proposition and boosted productivity by
reimagining maintenance processes with AI and GenAI
6 LEADING WITH GENAI/AI: HOW VIETNAMESE COMPANIES CAN STAY AHEAD
GenAI/AI is a Top Strategic Priority, yet Many
Companies are Falling Behind
AI has rapidly ascended to the top of the executive • 89% of executives rank AI and GenAI in their top three
agenda, with GenAI opening up a new world of business technology priorities for 2024.
opportunities that leaders are eager to capitalize on.
[Exhibit 6.] According to BCG’s survey of over 1,400 C-suite • 54% of leaders expect AI and GenAI to deliver cost
executives: savings in 2024, with roughly half of them anticipating
savings of more than 10%—primarily driven by
• 71% of leaders plan to increase their company’s tech productivity gains in operations, customer service and
investments in 2024, up from 60% in 2023. An even IT.
higher percentage (85%) plan to boost spending
specifically on AI and GenAI.
Exhibit 6 - A global wave of rising tech and GenAI/AI Investment
Executivesplanningtoincrease their 71% Executives planning to increase their 85%
techinvestmentin2024 overall AI/GenAIinvestment in 2024 overall
MiddleEast 85% MiddleEast 93%
Asia-Pacific 80% Europe 86%
Africa 77% Asia-Pacific 85%
Europe 68% North America 85%
NorthAmerica 65% Africa 82%
SouthAmerica 63% South America 75%
Source: : BCG AI Radar (2024); n = 1,406 in 50 markets.
Note: In Asia-Pacific, n = 308; in North America, n = 303; in Europe, n = 647; in the Middle East, n = 28; in South America, n = 51; in Africa, n = 69.
Despite the enthusiasm, many organizations still investment priorities (47%), and the absence of a
hesitate to embrace GenAI/AI fully. [Exhibit 7.] BCG’s strategy for responsible AI (42%).
findings highlight a significant gap between ambition and
action: • Only 6% of companies have managed to train
more than 25% of their people on GenAI tools
• 90% of leaders are either cautiously waiting for AI and so far.
GenAI to prove itself beyond the hype or are limiting
their efforts to small-scale experiments. • 45% of leaders say that they don’t yet have
guidance or restrictions on AI and GenAI use
• 66% of executives are ambivalent or outright at work.
dissatisfied with their organization’s progress on AI
and GenAI. They highlighted several challenges,
including a shortage of talent and skills (62%), unclear
Exhibit 7 - Executives worldwide must boost upskilling
Executives who report that more than 25%
of their workers have trained on GenAItools 6%
Midle East 11
North America 8
Asia-Pacific 7
Europe 5
Africa 3
South America 2
Source: BCG AI Radar (2024); n = 1,406 in 50 markets.
Note: In Asia-Pacific, n = 308; in North America, n = 303; in Europe, n = 647; in the Middle East, n = 28; in South America, n = 51; in Africa, n = 69.
BOSTON CONSULTING GROUP 7
It’s clear that the gap between AI leaders and • 10% of companies are actively scaling GenAI, with 61%
laggards is widening. Companies with the highest AI of these having already scaled multiple predictive AI
maturity are further extending their lead by scaling GenAI use cases and realizing substantial benefits.
applications. [Exhibit 8.] According to BCG’s Build for the
Future C-level GenAI Survey in 2023/2024: • Meanwhile, 50% are still in the phase of limited
experimentation and small-scale pilots, while 40%
have yet to take significant action.
Exhibit 8 - 10% of companies already scaling GenAI; building on their higher predictive
AI maturity and widening the gaps
~40% ~50% ~10%
SCALING
companies companies companies
Are scaling
1 or more GenAI
PILOTING applications
across functions/
enterprise
Developing few focused MVPs
NO-ACTION YET to test value from GenAI
Taking no action
on GenAIyet Majority have
historically
scaled several
Majority have historically Majority have historically piloted a number of predictive AI
lacked predictive AI project predictive AI projects, but few have initiatives in a
execution capability successfully scaled few functions
% high AI maturity 25% 46% 61%
% who have
realized significant
7% 10% 44%
value from scaled
predictive AI cases
1. % who have scaled numerous predictive AI cases
8 LEADING WITH GENAI/AI: HOW VIETNAMESE COMPANIES CAN STAY AHEAD
GenAI/AI Landscape in Vietnam - Challenges and
Opportunities Await
AI has been making waves in Vietnam for the past five recognition, electronic know-your-customer (eKYC), optical
years, with both government and private sectors character recognition (OCR), and call or voice bots
increasingly recognizing its potential to drive innovation deployed in contact centers, collections and credit card
and digital transformation. activation. In other industries, some advanced applications
are emerging, such as image diagnostics in healthcare and
In 2021, the Government of Vietnam took a significant step AI/ML-driven analytics in some global CPG/FMCG
forward by issuing Decision 127/QD-TTg, launching the companies.
National Strategy on Research, Development and
Application of AI until 2030. This strategy aims to position Late 2023 and early 2024 marked a turning point for AI in
Vietnam as a leading AI hub in Southeast Asia by focusing Vietnam, driven by the rapid rise of the ChatGPT
on five key areas: (1) establishing a robust legal framework, phenomenon. As a result, Vietnamese companies began
(2) building data and computational infrastructure, (3) increasing their focus on AI, with a growing interest in
developing a thriving AI ecosystem, (4) promoting AI GenAI. However, as GenAI is still a relatively new
applications, and (5) enhancing international cooperation. technology, its adoption remains in the early stages. Some
This strategic direction has spurred AI research and large companies have started piloting GenAI, mainly for
application at both central and local government levels, internal purposes such as document and knowledge
with a focus on enhancing public services and addressing navigation, HR tasks like automated CV scanning and
social challenges. Vietnam currently ranks 59th out of 193 profiling, staff training, and IT support through code co-
countries in the Government AI Readiness Index report by pilots. External-facing use cases are still largely focused on
Oxford Insights (UK), surpassing the global average and GenAI-powered chatbots and voice bots.
improving by one position regionally to rank 5th out of 10
countries in Southeast Asia. GenAI/AI is still in its early stages in Vietnam, but this
presents significant opportunities for companies to fully
In the private sector, AI has gradually expanded across harness its potential. While many Vietnamese companies
industries in Vietnam since 2019, including banking, are initially focusing on the productivity and efficiency
insurance, consumer finance, retail, healthcare, consumer benefits of GenAI/AI—the most tangible and easiest to
packaged goods (CPG) and manufacturing. However, the AI implement—the real power of GenAI/AI is to
use cases currently being implemented remain relatively fundamentally reshape how entire functions operate,
nascent compared to those in other countries. For driving deeper productivity gains and reducing costs, and
example, in the banking industry—one of the leading even inventing new business models to create new sources
sectors in AI adoption—AI is primarily used for facial of competitive advantage to improve top-line growth.
BOSTON CONSULTING GROUP 9
Taking a broader view, there are three value plays to maximize GenAI/AI’s potential:
Deploy GenAI/AI in everyday tasks to realize 10% to 20% productivity potential. Select and test
GenAI/AI tools, deliver massive upskilling, roll out solutions to support workers in day-to-day tasks and
carefully evaluate the costs of deployment.
Reshape critical functions for 30% to 50% enhancement in efficiency and effectiveness.
Anticipate the impact of GenAI/AI on your workforce and core functions, create new roles, reallocate
budgets and guide a series of pilots that can reliably scale up.
Invent new GenAI/AI business models to build a long-term competitive advantage. Develop a
strong customer-centric approach, and leverage first-party data and intellectual property to create
interactions that customers can't find anywhere else.
No single approach can capture the full scope of what This is particularly crucial for companies in Vietnam, where
GenAI/AI offers. Instead, companies should evaluate their lower labor costs can make it challenging to justify the
opportunities across three broad value plays. Ideally, these value of GenAI/AI if efforts are limited to isolated
aren’t just small-scale pilots to appease investors or productivity improvements. For GenAI/AI to truly deliver its
boards—they require a deep, organization-wide potential, it must be embedded in a way that drives
commitment to embedding GenAI/AI into every aspect of comprehensive transformation rather than just
the business from budgets and processes to roles and incremental automation. [Exhibit 9.]
culture, all while adhering to responsible AI principles.
Exhibit 9 - Three strategic plays to leverage when considering adopting and
implementing GenAI/AI
DEPLOY RESHAPE INVENT
Deploy GenAIsolutions to augment the GenAIserves as an enabler to GenAIsolutions reimagine business
productivity of everyday tasks transform an entire function(s) and models and/or create new sources of
enhance its productivity competitive advantage
Improve effectiveness of Re-engineer one or multiple Reimagine the org. business
existing tasks within a role e.g., critical functions e.g., Marketing models e.g., reinventing CX or
meeting summary or code or Customer Service insights and innovation platforms
development
Difficult to directly measure Direct capturable impact on bottom Potential for significant long-term
bottom-line impact but has great line, therebyenhancing company impact on bottom line, offering a
potential to familiarize the org. productivity competitive advantage
with GenAI
Change focus on driving adoption Reshapes how employees work Reimagines what employees do in
and improving effectiveness in a specific function(s), the organization, necessitating work
within existing setup necessitating work with change, with change, new skillsets, WoW
new skillsets, WoW and and potentially org setup
potentially org setup
Increasing business value
Note: The change effort will depend on the organization’s starting point in data, people, and process readiness
10 LEADING WITH GENAI/AI: HOW VIETNAMESE COMPANIES CAN STAY AHEAD
Implementation challenges
• Data challenges. Fine-tuning models is difficult due
to the lack of structured, high-quality and diverse data.
Additionally, the Personal Data Protection Law
imposes strict limitations on how companies can store
and process data, further complicating efforts in
certain use cases.
• Talent gaps. There is a shortage of skilled GenAI/AI
professionals at both leadership and operational
levels, hampering effective implementation.
• Security and privacy concerns. There are significant
concerns about data leakage when sharing
information with third parties, as well as concerns
about potential misuse of data for training models
• Change management. The lack of a robust
governance structure for GenAI/AI projects, along with
resistance to new processes, cultural shifts, and
changes in mindsets and ways of working, significantly
impacts implementation.
Despite the potential, Vietnamese companies starting to Beyond that, there are two fundamental external
adopt and pilot GenAI/AI typically face several common challenges that affects the ability of Vietnam companies
internal challenges: to scale and capture the full potential of GenAI/AI:
Strategic challenges • Insufficient computing resources and
infrastructure. Securing the necessary computing
• Insufficient GenAI/AI proficiency among the resources, particularly graphics processing units
executive team. This expertise gap hinders the (GPUs), and setting up the infrastructure for GenAI/AI
ability to drive and adapt to change, limits the inference is both challenging and costly. This issue is
recognition of GenAI/AI’s transformative potential for further compounded for companies seeking to deploy
the business and results in a lack of top-down these solutions on private infrastructure.
strategic direction for GenAI/AI adoption. [Exhibit 10.]
• Regulatory uncertainty. The absence of legally
• Lack of clear AI strategy: Many GenAI/AI initiatives binding regulations for GenAI/AI, along with limited
are currently embedded within broader guidance on ethics and responsible AI (RAI), poses
transformation efforts or are focused on solving significant challenges to broader adoption.
specific, discrete problems, leading to a fragmented
approach.
Exhibit 10 - The need to upskill extends to the C-suite
CoCnofindfiednecneceininthteheexeexceuctuivtievetetaemam’s’GseGneAnIApIrporfiocfiiceinecnycy
CoCmomplpetleetlye lcyo cnofindfiednetnt 1%1%
5599%%
VeVreyr cyo cnofindfiednetnt 111%1%
CoCnofindfiednetnt 292%9%
ofo lfe laedaedresr ssu sruvrevyeeyde dsa sya tyh tehyey
hahvaev lei mlimiteitde do ro nr on o
LimLimiteitde dco cnofindfiednecnece 404%0%
cocnofindfiednecneceini nth tehire ier xeexceuctuivteiv e
tetaemam’s ’ps rporfiocfiiecniecnyc iyn i nG eGneAnIA.I.
NoN coo cnofindfiednecnece 191%9%
Source: BCG AI Radar (2024); n = 1,406 in 50 markets.
BOSTON CONSULTING GROUP 11
While these challenges exist, Vietnam also has unique improving university training, specialized programs
strengths and a rapidly evolving support ecosystem that is with global experts and corporations are expected to
poised to accelerate GenAI adoption in large enterprises: help build a skilled AI workforce. Recently, a global
tech giant partnered with the Ministry of Planning and
• Building talent foundations. The number of Investment to offer 40,000 AI scholarships, while
educational programs specifically focused on AI and another leading tech company launched AI training for
GenAI is increasing, especially in leading institutions. university lecturers to strengthen Vietnam’s AI
For example, Hanoi University of Science and capabilities. Combined with the unique strength of
Technology has launched Vietnam’s first specialized Vietnam’s young, tech-savvy, low-cost workforce, these
engineering program in GenAI. Vietnam National efforts are expected to produce a high-quality AI talent
University, Ho Chi Minh City is also focusing on AI, pool in the coming years. [Exhibit 11.]
and, with 6,000 undergraduates, it aims to provide the
nation with highly qualified AI talent. Beyond
Today’s students are remarkably advanced in AI and GenAI. Through our
engagement with these students in GenAI competitions, we’ve been impressed
by how many undergraduate students can solve complex problems and develop
prototypes in a short period of time.
AI Expert at a Leading Commercial Bank
Exhibit 11 - Vietnam’s affordable talent advantages
Average monthly IT Engineer salary in 2024, in dollars
Singapore 5,627
Taiwan 3,782
South Korea 2,826
Thailand 2,174
Malaysia 1,313
Vietnam 665
0 2,000 4,000 6,000
Source: Salary Explorer, Nikkei Asia, BCG Estimation
12 LEADING WITH GENAI/AI: HOW VIETNAMESE COMPANIES CAN STAY AHEAD
• Localized AI models. More than 20 large language
models (LLMs) developed or fine-tuned for the
Vietnamese language and context have been
introduced recently, offering tailored options beyond
models from abroad. This development has attracted
a diverse range of contributors, from large technology
companies to startups, independent research groups,
and universities. While these models are still in
development, and face challenges such as limited
computing power and data resources, they are
expected to meet the needs of Vietnamese companies
within the next one to two years.
• Growing GenAI startup ecosystem. The rise of new
GenAI startups in Vietnam is accelerating adoption by
creating competitive pressure and providing proven
use cases for large enterprises. These startups also
offer GenAI/AI applications and services that can be
integrated into larger enterprise GenAI/AI systems.
We’re witnessing a surge in GenAI
startups in Vietnam, which now
ranks #2 in the region for number
of GenAI startups. This marks a
significant milestone, as Vietnam
is emerging as a strong contender
in this field, closing the gap with
Singapore for the first time.
Business Development
Manager for Startups at
Global Hyperscaler
Given the rapid advancements in GenAI/AI and the
significant opportunities it presents, companies need to
act now and prepare for the future:
Start now with small but high-
1 impact use cases to quickly adopt
and realize GenAI/AI benefits and
integrate GenAI/AI into daily operation.
In parallel, prepare for business and
2 organizational transformation to
scale and fully capture the potential of
GenAI/AI in the long term.
BOSTON CONSULTING GROUP 13
From Hesitation to Execution - Practical
Recommendations for Beginning Your
GenAI/AI Journey
For companies that have yet to embark on their GenAI/AI
journey or still at the early stage, moving forward can seem
daunting. However, the rewards of successfully adopting
and scaling GenAI/AI are immense. Here’s how to take
those critical first steps and move from hesitation to
execution.
14 LEADING WITH GENAI/AI: HOW VIETNAMESE COMPANIES CAN STAY AHEAD
Turn Leadership into Informed Upskill Your Workforce
1 4
Advocates
Empowering your team to work effectively
GenAI/AI adoption starts at the top. Leaders with GenAI/AI is crucial for success:
must become well-informed advocates • Develop training programs.
for GenAI/AI, understanding its potential Implement training initiatives and
value and pushing the organization to communication strategies that shift
embrace the technology. This is particularly mindsets and equip users with the
crucial because piloting GenAI/AI use skills to maximize GenAI/AI’s value.
cases can sometimes be challenging, with • Provide hands-on support. During
results that may fall short of expectations. implementation, offer necessary
However, GenAI/AI is a long-term journey, handholding to hel |
289 | bcg | mitsmr-bcg-ai-report-november-2024.pdf | In collaboration with
November 2024
Learning
to Manage
Uncertainty,
With AI
by Sam Ransbotham, David Kiron,
Shervin Khodabandeh, Michael Chu,
and Leonid Zhukhov
AUTHORS
Sam Ransbotham is a professor of analytics at Michael Chu is a vice president of data science
the Carroll School of Management at Boston at BCG, where he focuses on applying AI and
College, as well as guest editor for MIT Sloan machine learning to business problems in
Management Review’s Artificial Intelligence commercial functions, including optimizing
and Business Strategy Big Ideas initiative. pricing, promotions, sales, and marketing. He
can be reached at [email protected].
David Kiron is the editorial director, research,
of MIT Sloan Management Review and program Leonid Zhukhov is a vice president of data
lead for its Big Ideas research initiatives. science at BCG and leads the Tech & Biz
Lab at the BCG Henderson Institute. He
Shervin Khodabandeh is a senior partner and
leads the design and build of AI and machine
managing director at Boston Consulting Group
learning solutions for BCG clients across
(BCG) and the coleader of its AI business
a range of sectors. He can be reached at
in North America. He is a leader in BCG X
[email protected].
and has over 20 years of experience driving
business impact from AI and digital. He can be
contacted at [email protected].
CONTRIBUTORS
François Candelon, Todd Fitz, Kevin Foley, Sarah Johnson, Michele Lee DeFilippo, Meenal Pore,
Namrata Rajagopal, Allison Ryder, Barbara Spindel, and David Zuluaga Martínez
The research and analysis for this report was conducted under the direction of the authors as part of
an MIT Sloan Management Review research initiative in collaboration with and sponsored by Boston
Consulting Group.
To cite this report, please use:
S. Ransbotham, D. Kiron, S. Khodabandeh, M. Chu, and L. Zhukov, “Learning to Manage Uncertainty,
With AI,” MIT Sloan Management Review and Boston Consulting Group, November 2024.
SUPPORTING SPONSORS
Copyright © Massachusetts Institute of Technology, 2024. All rights reserved. REPRINT #: 66262
CONTENTS
1 Uncertainty Abounds
2 Combining Organizational Learning and AI-specific
Learning Leads to Augmented Learning
4 Augmented Learners Are Better Prepared for
Many Types of Uncertainty
8 Three Ways to Enhance Organizational
Learning With AI
11 Developing Augmented Learning Capabilities
13 Learning With AI Is Key to Navigating Uncertainty
14 Appendix: The State of AI in Business
Uncertainty Abounds
Uncertainty is all about the unknown. The less an organi- ELC is not alone: The company is among the 15% of orga-
zation knows, the greater its uncertainty and the less able nizations that integrate AI into their learning capabilities.
it is to manage resources effectively. Managing uncertainty, These organizations — what we refer to as Augmented
therefore, requires learning. Companies need to learn Learners — are 1.6 times more likely than those with lim-
more, and more quickly, to manage uncertainty. ited learning capabilities to manage various environmental
and company-specific uncertainties, including unexpected
Addressing uncertainty constitutes a pressing challenge
technological, regulatory, and workforce changes. These
for leadership, especially today, when geopolitical tensions,
companies are twice as likely to be prepared to manage
fast-moving consumer preferences, talent disruptions,
talent-related disruptions compared with organizations
shifting regulations, and rapidly evolving technologies
that have limited learning capabilities. What’s more, these
complicate the business environment. Companies need
organizations are 60%-80% more likely to be effective at
better tools and perspectives for learning to manage
managing uncertainties in their external environments
uncertainty arising from these and other business disrup-
than Limited Learners — companies with limited learning
tions. Our research finds that a major source of uncertainty,
capabilities. By doing so, they reap advantages with AI well
artificial intelligence, is also critical to meeting this chal-
beyond direct financial benefits.
lenge. Specifically:
Based on a global survey of 3,467 respondents and inter-
Companies that boost their learning capabilities with AI views with nine executives, our research quantitatively and
are significantly better equipped to handle uncertainty qualitatively establishes a relationship between organiza-
from technological, regulatory, and talent-related dis- tional learning, learning with AI, and the ability to manage
ruptions compared with companies that have limited rapidly changing business environments. Organizational
learning capabilities. learning itself has long been associated with improved per-
formance. Integrating AI with an organization’s learning
The Estée Lauder Companies (ELC) offers a case in point.
capabilities significantly improves corporate responses to
The cosmetics company has a strategic need to anticipate
uncertainties from talent mobility, new technology, and
consumer trends ahead of its competitors. In earlier times,
related regulations. This report defines an AI-enhanced
consumer preferences might have shifted seasonally. Now,
organizational learning capability (augmented learning),
preferences are less certain; shifts happen more quickly due
explains its use in reducing the considerable uncertainty
to social media and digital influencers. Fashion trends can
managers face today, and offers key takeaways for exploit-
change by the week. If the color peach suddenly captures
ing these new abilities.
the public’s interest, the company needs to discern that
trend as quickly as possible. It uses AI to detect and rap-
idly respond to consumer trends. Sowmya Gottipati, vice
president of global supply chain technology at ELC, reports
that the company, which carries products across more than
20 brands and “hundreds of different shades,” uses fuzzy
matching to figure out which products can meet the demand
and delight consumers. “We are looking to AI to discover
Companies need to learn
consumer trends and then match up our existing products to
the trends so that we can repackage them and position them
more, and more quickly,
in the market for that trend,” Gottipati explains. ELC uses
AI to detect sudden changes and have a market response to manage uncertainty.
ready so it can redeploy inventory and supply chain pro-
cesses to meet demand efficiently. Companies can’t control
the changes but can use AI to manage their responses.
Learning to Manage Uncertainty, With AI 1
ABOUT THE RESEARCH
This report presents findings from the eighth annual Combining Organizational Learning
global research study on artificial intelligence and busi-
and AI-specific Learning Leads to
ness strategy by MIT Sloan Management Review and
Boston Consulting Group. In spring 2024, we fielded a Augmented Learning
global survey and subsequently analyzed records from
3,467 respondents representing more than 21 industries Organizational learning is an organization’s capability to
and 136 countries. We also interviewed nine executives change its knowledge through experience.1 Organizations
leading AI initiatives in a broad range of companies and that learn from mistakes, tolerate failure, capture best
industries, including financial services, technology, retail, practices, and support new ideas have an advantage over
travel and transportation, and health care. organizations that don’t: They learn to get better. Those
Our research connects organizational learning, learning that struggle to learn will struggle to navigate increasing
with AI, and the ability to manage rapidly changing environ- uncertainties. Extensive past research demonstrates the
ments. This report defines an AI-enhanced organizational benefits of general organizational learning.
learning capability, explains its use in reducing several types
General organizational learning capabilities don’t neces-
of uncertainty managers face today, and offers key leader-
sarily depend on AI; organizations can have strong organi-
ship takeaways for exploiting these new abilities.
zational learning capabilities without using the technology.
To assess whether organizations have “high” or “low”
Conversely, organizations can use AI to learn even if they
organizational and AI-specific learning capabilities, we
don’t otherwise have strong organizational learning capa-
analyzed survey responses to these statements using an
bilities. Managers can learn from generative AI tools, use
agree-disagree Likert scale:
AI to deepen their understanding of performance, and iter-
ate with AI to develop new insights and processes. These
›
My organization learns through experiments. (organi-
individual learning experiences create value from AI but
zational learning)
may not constitute an organizational learning capability.
›
My organization tolerates failures in experiments.
Our research finds that organizations that combine organi-
(organizational learning)
zational learning with AI-specific learning — Augmented
›
My organization learns from postmortems on both Learners — outperform organizations that employ either
successful and failed projects. (organizational learning) approach in isolation. As businesses adopt AI and embrace
› successively more powerful AI tools in various contexts, they
My organization codifies its learning from initiatives.
have new opportunities to strengthen their learning capa-
(organizational learning)
bilities — for both human workers and their machines. Our
›
My organization gathers and shares information that prior research, “Expanding AI’s Impact With Organizational
employees learn. (organizational learning) Learning,” found that organizations with superior learning
› capabilities are more likely to obtain significant financial ben-
My organization’s use of AI leads to new learning.
efits from their AI use.2 In our latest research, we find that
(AI-specific learning)
the reverse is also true: Using AI can improve organizational
›
My organization uses AI to learn from performance. learning capabilities, and these learning improvements are
(AI-specific learning) tied to not only enhanced financial results but also the ability
› to manage strategy-related uncertainties.
My organization builds AI solutions with human feed-
back loops. (AI-specific learning)
Assessing Learning Capabilities
›
Employees in my organization learn from AI solutions.
Our survey instrument measured each enterprise’s orga-
(AI-specific learning)
nizational learning capability using five questions. We
also assessed how individuals and systems learn with AI
We then grouped respondents into four categories:
through a different set of four questions. Together, these
Limited Learners, Organizational Learners, AI-specific
questions probe several aspects of organizational learning
Learners, and Augmented Learners. (See Figure 2, page 3
and AI-specific learning: knowledge capture, synthesis,
For theSe breakdownS.)
and dissemination. (see figure 1, page 3.)
2 MIT SLOAN MANAGEMENT REVIEW • BCG
Becoming adept at these learning activities — which rep-
Organizational learning AI-specific learning
resent only a slice of an organization’s overall learning
capability — significantly improves a company’s ability to
• Learns through experiments and • Uses AI to lead to new learning
manage uncertainty. tolerates failure
• Uses AI to learn from
• Supports employees presenting performance
Most Companies Have Limited new ideas
• Builds AI solutions with human
Learning Capabilities • Learns from postmortems on feedback loops
successful and failed projects
Given the uncertainties facing many companies, it’s strik- • Enables employees to learn
• Codifies learning from initiatives from AI solutions
ing that most organizations have limited learning capabili-
• Gathers and shares information
ties; 59% of all companies represented in our sample report
that employees learn
low levels of both organizational learning and AI-specific
learning. Only 29% of respondents agree or strongly agree
that their enterprise has organizational learning capabili-
ties. While 27% of organizations report learning with AI, FIGURE 1
only 15% combine AI-specific learning with organizational Characteristics of Organizational
Learning and AI-specific Learning
learning capabilities. These Augmented Learners are the
focus of this report. We outline characteristics of organizational and
AI-specific learning based on nine survey questions.
12 15
% %
59 14
% %
In our global survey, we assessed an organization as having “high”
or “low” organizational and AI learning capabilities. For more detail,
see “About the Research,” page 2.
gninraeL
cfiiceps-IA
AI-specific Learners Augmented Learners
Limited Learners Organizational Learners
Low High
Organizational Learning
hgiH
woL
12 15
% % OrganizatiOnal learning
An organization’s capability to change its
knowledge through experience.
ai-specific learning
The measure of organizations’ use of
AI for learning.
59 14
augmented lear%ners %
Organizations that score high on organizational
learning and AI-specific learning.
limited learners
Organizations that score low on organizational
learning and AI-specific learning.
In our global survey, we assessed an organization as having “high”
or “low” organizational and AI learning capabilities. For more detail,
see “About the Research,” page 2.
FIGURE 2
Learning Capabilities Vary
Only 15% of organizations are Augmented Learners —
organizations that enhance organizational learning with AI.
gninraeL
cfiiceps-IA
AI-specific Learners Augmented Learners
Limited Learners Organizational Learners
Low High
Organizational Learning
hgiH
woL
Fig. 2
Title: Learning Capabilities Vary
Caption: Only 15% of organizations are Augmented Learners —
organizations that enhance organizational learning with AI.
Learning to Manage Uncertainty, With AI 3
Limited learning capabilities constrain opportunities and disruptions from talent mobility, changing technology,
undermine organizations’ ability to manage uncertainty. and evolving regulatory and legal requirements. (see fig-
ure 5, page 6.)
Augmented Learners Are Better
at Managing Uncertainty Disruptions From Talent Mobility
Among our sample, 15% of organizations report high lev- Elevated rates of workers quitting, retiring, being laid off,
els of both organizational learning and AI-specific learning. or even ghosting employers create risks and ambiguities
These Augmented Learners display abilities and advantages for organizations striving to compete. Shilpa Prasad is head
that lead to better outcomes than organizations with limited of incubation, AI Ventures at LG Nova, the subsidiary of
LG Electronics that works with startups to fuel innova-
capabilities. They are more likely to improve financial out-
tion for the company. She observes that “60% of the work-
comes with AI than Limited Learners: 99% of Augmented
force will likely hit the age of 65 by the year 2028 or 2030,
Learners report annualized revenue benefits from AI. (see
which means that a lot of knowledge will go out from the
sidebar, “enhancing OrganizatiOnal learning With
workforce because they’ll retire, not because they’re going
ai imprOves financial OutcOmes,” page 5.) What’s more,
somewhere else to work.” When employees leave organi-
they are much more likely to be prepared to deal with uncer-
zations, their knowledge can leave with them unless the
tainty from talent, technology, and legal disruptions.
company has effective organizational learning capabilities.
Figure 3 shows that organizational learning alone or
These problems are not new for organizations. In indus-
AI-specific learning alone offers some benefits, but their
tries like chemicals, aerospace, and oil and gas, retirement
combination represents the most powerful hedge against
rates have been an increasing cause for alarm for years.
multiple types of uncertainty. Organizational learning with
However, companies have new resources to address these
AI may well prove to be a source of resilience against other
challenges. Augmented learning is a valuable resource for
forms of disruptions or uncertainty.
addressing disruptions from talent mobility. Only 39% of
organizations with limited learning feel prepared to han-
dle the disruption in knowledge from departing employ-
Augmented Learners Are
ees, but this readiness increases to 64% if the companies
Better Prepared for Many have organizational learning capabilities. Using AI can
Types of Uncertainty further contribute to this readiness: Eighty-three percent
of Augmented Learners are prepared to deal with the
Combining organizational learning and AI-specific learn- uncertainty of knowledge disruption from talent mobil-
ing capabilities helps enterprises manage uncertainty and ity — twice as much as Limited Learners.
FIGURE 3
Learning With AI Helps
AI will allow us to manage uncertainty Organizations Manage
in our industry. 1.6× Uncertainty
Organizations that combine
organizational and AI-specific
Limited Learners 53% learning (Augmented Learners)
are 1.6 times more likely to feel
Organizational Learners 58%
prepared to manage uncertainty
AI-specific Learners 76% than organizations with limited
learning capabilities.
Augmented Learners 82%
Percentage of respondents in each learning category who strongly agree or agree with the above statement.
4 MIT SLOAN MANAGEMENT REVIEW • BCG
SIDEBAR
ENHANCING ORGANIZATIONAL LEARNING
WITH AI IMPROVES FINANCIAL OUTCOMES
Numerous studies now show the direct financial benefits of to recognize some revenue benefits from AI compared with
AI adoption. Clearly, organizations are finding ways to extract organizations with limited learning capabilities. Indeed, virtually
financial benefits through AI, even if many such efforts fail all of these organizations (99%) recognize or observe some
or their costs exceed revenues. Extensive past research also revenue benefits from AI. What’s more, organizations that
surfaces the general benefits of organizational learning for combine AI and organizational learning are significantly more
companies. In prior research, we found that organizations with likely to have realized revenue benefits from AI compared with
superior learning capabilities are more likely to obtain signifi- companies that excel at organizational learning but not learning
cant financial benefits from their AI use than organizations with with AI, and with companies that excel at AI-specific learning
lesser learning capabilities. but not organizational learning. That is, combining organiza-
tional learning and AI-specific learning enables organizations to
In this study, we find that using AI can improve organizational
cross a revenue benefit threshold that neither type of learning
learning capabilities and that these learning improvements are
alone can generate.
similarly tied to improved financial results. Organizations using
AI to improve organizational learning are 1.4 times more likely
Over the past three years, AI has
1.4×
created additional business value.
Limited Learners 66%
Organizational Learners 76%
AI-specific Learners 89%
Augmented Learners 95%
Percentage of respondents who strongly agree or agree that AI has created additional business value over
the past three years.
Our organization has realized revenue
1.4× FIGURE 4
benefits from AI on an annualized basis. Enhancing Organizational
Learning With AI Leads to
Financial Benefits
Limited Learners 71%
Organizations that combine
Organizational Learners 72% organizational learning and
AI-specific learning (Augmented
AI-specific Learners 79%
Learners) are 1.4 times as likely
Augmented Learners 99% to realize additional business
value and annualized revenue
Percentage of respondents who report revenue benefits from AI. benefits from AI.
Learning to Manage Uncertainty, With AI 5
As more and more workplace communications occur Generative AI tools can help synthesize and disseminate
via digital channels, emerging AI capabilities can make personalized knowledge. “GenAI helps you get more value
this raw data sensible, and tacit knowledge accessible, out of this knowledge so that you can find what you’re
on demand. Jackie Rocca, former vice president of prod- looking for and be more effective in using all that data that
uct at Slack, describes how AI can surface and distill the has been available to you but hasn’t been very easy for you
trove of information from past conversations in a platform to access and use,” Rocca says. While tools like wikis make
like Slack when people need it. “People can get context it easier for people to record knowledge, AI capabilities can
from coworkers who left the company months or years bolster organizational learning about what workers know.
ago and still learn from that knowledge,” she points out. That enables organizations to better handle knowledge
My organization is prepared to deal with uncertainty from …
Talent disruptions 2.2×
Limited Learners 39%
Organizational Learners 64%
AI-specific Learners 58%
Augmented Learners 83%
Technology disruptions 1.8×
Limited Learners 49%
Organizational Learners 71%
AI-specific Learners 68%
Augmented Learners 86%
Legal disruptions 1.6×
Limited Learners 48%
FIGURE 5
Organizational Learners 61%
Combining Organizational
AI-specific Learners 68% Learning With AI Learning Helps
With Many Types of Uncertainty
Augmented Learners 79%
Organizations that combine
Percentage of respondents in each learning category who strongly agree or agree that their organization is organizational and AI-specific learning
prepared to deal with each type of uncertainty. Some values calculated with rounding. (Augmented Learners) are more likely
to manage talent, technology, and
legal disruptions.
6 MIT SLOAN MANAGEMENT REVIEW • BCG
loss from talent mobility, reducing uncertainty around of Excellence at Novo Nordisk, notes that “technology is
how and when to capture tacit knowledge. evolving faster than organizations can address. Combining
that with the hype around technology’s possible effects pulls
One cloud services provider wasn’t preparing for a poten-
the organization to do something.” Emerging technologies
tial pandemic when it developed its learning tool, but
become “a propeller for the organization,” she observes,
when in-person meetings were no longer possible due
even if it’s initially unclear what the business case is or where
to COVID-19, its platform and micro-learning content
investments should go. Reassessing technology investments
enabled the company to sustain and even enhance mean-
can be beneficial, even if organizations don’t end up adjusting
ingful educational experiences. The company’s responsible
their strategies but, rather, reinforce them to work within the
AI lead explains how an innovative learning tool turned into
new technological landscape.
a powerful tool for managing uncertainties wrought by the
pandemic. Before the pandemic, the company had begun What’s more, technology adoption can lead to more, and
shifting its learning modules to shorter, AI-supported more complex, regulatory scrutiny and compliance issues,
“micro-adaptive” approaches suitable for a “TikTok world.” raising difficult questions about how to navigate increas-
The pandemic necessitated a remote work environment ingly uncertain legal environments. Surprisingly, using AI
that changed what employees needed to know and, further- to amplify organizational learning dramatically improves a
more, made it more difficult for the company’s educational company’s ability to manage uncertainty from both tech-
content providers to determine what employees knew and nology and regulatory disruptions. Compared with orga-
didn’t know on an ongoing basis. nizations with limited learning capabilities, Augmented
Learners are significantly more likely to be prepared to deal
The adaptive modules tailored content recommendations
with uncertainty from technology disruptions (86% versus
to each individual as the system assessed individual users’
49%) and regulatory disruptions (79% versus 48%). (see
learning capabilities. “AI became a huge part of that,” this
figure 5, page 6.)
executive says. “We monitored users’ self-reporting and
skills self-assessments in their profiles and from the learning Learning to manage uncertainty that comes from a depen-
platform.” By analyzing skills and competency proficiency dence on older technology and from future waves of tech-
across systems throughout the organization, the company nology is a growing opportunity for Augmented Learners.
identified what its employees were learning and needed Shelia Anderson, CIO of Aflac U.S., shares how the insurer
to learn. She adds, “The AI-enabled modules did not just uses generative AI to reverse-engineer code in certain leg-
enable a different delivery of content; the platform helped acy systems. This approach is projected to boost current
people better understand what they knew and how that levels of system productivity by five to 10 times by revealing
intersected with what they needed to know.” Drawing on hidden complexities. “We have built in approaches to learn-
ing that leverage AI and actually help to inform our organi-
the learning approaches and habits of many of the company’s
zation on how AI can be used as well,” Anderson says. She
workers, the learning modules made tailored recommenda-
notes that Aflac also has a technology incubator that uses
tions based on individual needs that reduced uncertainty
AI to evaluate new technologies and rapidly prototype lead-
about what an individual needed to learn next. Enhancing
ing candidates to prove out concepts for the business. If a
organizational learning with AI provided flexibility to man-
prototype appeared to be viable for the business, Anderson
age necessary changes during an unanticipated crisis.
says, “we would use AI to build a full business model with
Technological and Regulatory Uncertainty the return on investment or productivity savings or what-
ever business value metric we’re looking to achieve.”
Increasingly frequent technology innovations lead to signifi-
cant strategic and operational uncertainty. Adapting systems On the regulatory front, large organizations with global
over and over again can be exhausting and disruptive to tech- operations can use AI to navigate complex, uncertain reg-
nologists and business users alike. Just when companies had ulatory frameworks that vary from one country to the next.
begun to understand how to incorporate AI use into their For example, ELC’s Gottipati observes, “From a company
business strategies, generative tools introduced changes that point of view, you make one product and distribute it. But
required a reassessment. (see “the state Of ai in business,” then, if the requirements are different for different countries,
page 14.) Tonia Sideri, director of the AI and Analytics Center and also certain ingredients are limited in certain countries,
Learning to Manage Uncertainty, With AI 7
it becomes time intensive to keep track of all these changing Surman also sees open source as critical for managing that
regulations and at the scale at which we operate.” The com- uncertainty. He expects that “open source will play a huge
binatorial explosion of products in a large number of mar- role in the organizational learning and corporate AI space
kets is difficult to keep up with. But Gottipati sees potential because it lends itself to on-premises privacy, respecting
in using AI to help manage the myriad combinations. “That’s customized models. It creates market demand for open-
where I think AI can play a huge role: to identify the right source models that you can fine-tune on your own data.”
combination of products before we ship anything, or send- However, taking full advantage of open source requires
ing us alerts and assisting with compliance,” she notes. Using organizational learning.
AI can offset growing legal complexity.
Openness isn’t restricted to the models themselves. For
Technology and regulatory uncertainty are inherently example, federated learning allows multiple organizations
intertwined. As difficult as technological disruptions can to train models collaboratively while keeping their data pri-
be, legal disruptions can exacerbate technological uncer- vate. Surman believes that “private AI — using open-source
tainty in addition to creating uncertainty on their own. models — becomes a hedge against regulatory uncertainty.
Mark Surman, president of the Mozilla Foundation, notes Federated learning, in which I benefit from my data in the
that the software company is still very early in the process partnership, you benefit from your data in the partnership,
of figuring out the legal implications of AI. He says, “The and there’s some area where we collectively benefit or at
core piece is, there’s just so many questions about copyright
least we can operationalize on each other’s data, is super
and what it means to own knowledge. Maybe the copyright
juicy.” While standards for federated learning are still being
law we have just needs to be interpreted for the AI era. Or
worked out, Augmented Learners are better able to manage
maybe we need new copyright law.” Within and beyond the
regulatory and legal uncertainties like these.
boundaries of organizations, AI has turned the question of
knowledge ownership upside down because, as Surman
points out, “many of the main large language models are Three Ways to Enhance
built on stuff that, arguably, doesn’t belong to them.”
Organizational Learning With AI
Augmented Learners have an advantage here because they
have abilities that those unable to learn with AI lack. For While it may be tempting to identify organizational learn-
example, knowing how to build your own AI tools could ing as — or, more pointedly, reduce it to — knowledge
hedge against uncertainty from third-party solutions sub- management or learning and development, organizational
ject to upcoming copyright regulation. Surman explains learning involves far more than these important activities.
that this knowledge can help an organization navigate It encompasses whether organizations view unsuccess-
current legal uncertainties: “The one thing that is known ful experiments as failures or as sources of learning; how
and safe is [what’s] inside the organization … to the degree organizations develop, not just manage, knowledge; and
that you have good practices that information is clean and how organizations anticipate the unknown rather than
belongs to you. So if you train a large language model on merely capture what is known. It occasionally requires
your company’s information, it’s yours.” setting aside old ways of working to make learning new
Combining organizational learning with
AI-specific learning yields more benefits
than taking either approach alone.
8 MIT SLOAN MANAGEMENT REVIEW • BCG
capabilities possible.3 What’s more, organizational learn- AI technologies represent new capabilities for capturing
ing encompasses synthesizing and analyzing information existing tacit knowledge. In a more down-to-earth context,
to glean what is and is not working in the enterprise. It LG Nova’s Prasad observes that AI-based augmented reality
also involves optimizing metrics, not merely maximizing (AR) glasses have the potential to capture the tacit knowl-
performance on existing metrics. Finally, organizational edge of factory workers on the shop floor who have mastered
learning addresses the communication, dissemination, and a certain way of working with machines. “If they’re doing a
accessibility of knowledge. technique on the shop floor that only they know, AR glasses
can allow real-time content creation,” she says. While AR
Combining organizational learning with AI-specific learn-
use is not common today, Prasad states this use case has the
ing yields more benefits than taking either approach alone.
potential to become a more significant approach to capturing
AI-specific learning can significantly enhance (at least)
tacit knowledge as the technology/hardware matures.
three areas of organizational learning: knowledge capture,
knowledge synthesis, and knowledge dissemination. These Using AI to distill information at scale enables the capture
are not incremental additions; Augmented Learners multi- of salient information that would otherwise be impossible
ply their abilities in these areas. for humans to discern. Since 2021, LG Nova’s mandate has
been to work, develop, and collaborate with startups to build
Knowledge Capture
new business ventures — a typically daunting task, given the
Adopting generative AI and embracing developments in tra- sheer number of potential targets worldwide. Prasad summa-
ditional AI can expand an organization’s ability to capture rizes the question |
290 | bcg | June 2023 - BCG Perspectives on GenAI_s Impact on Private Equity - Summary.pdf | Generative AI:
Implications for PE Investors
Operational and Portfolio Overview
June 2023
1
2
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
3202
©
thgirypoC
Table of contents | GenAI perspectives for private equity
Scope of the assessment
Overview of foundation models, the tech stack, why now, what the
1 What is GenAI, why now, and why it matters
current capabilities are, and what the expected impact is
2 GenAI's impact on Private Equity
Framework for how to evaluate GenAI's impact on portfolio
2a Impact on Portfolio and new deals
industries, functions and targets
GenAI use cases in fund mgmt. from fundraising to investing to
2b Impact on PE operations
portfolio mgmt.
Immediate next steps; Introduction to GenAI control tower to
3 How to proceed with Gen AI deployment
coordinate fund-wide efforts
4 Why BCG for the GenAI journey Why BCG is a thought leader in GenAI
5 How to partner going forward Areas where BCG can support, our approach and commercial offer
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
3202
©
thgirypoC
GenAI is already proving to be a game-changer: Significant productivity increases
1 have already been reported and companies' value propositions are being challenged
with significant impact on enterprise value
GenAI's impact on an industry basis should be analyzed from two complementary
2 lenses: Productivity gain potential (cost and/or effectiveness improvements) and its
Value proposition change (new product & business models) distinctly impact businesses
Actions required depend on type of impact: Funds should scan portfolios for horizontal
Key Messages
3 productivity gain potential and PortCo-specific value proposition changes. Productivity
gain potential implies org changes, value prop changes require strategic reviews
People and process change are the most critical factors to succeed: Whilst data and
4
technology are important, organizational transformation through GenAI is 70% about
People, 20% about Processes, and only 10% about Tech
As an immediate step, funds need to start assessing impact across its portfolio, in
5 fund operations and its investments strategy; as efforts develop, a GenAI control
tower setup can serve as a focal point of efforts and drive coordination
In addition to risks, GenAI presents meaningful investments opportunities across industries. These
requires assessments on a sub-sector basis, and will be the focus of upcoming perspectives
3
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
3202
©
thgirypoC
What is GenAI, Why now, and Why it matters
GenAI's impact on Private Equity
Impact on portfolio and new deals
Impact on PE operations
Agenda
How to proceed with GenAI deployment
Why BCG for the GenAI journey
How to partner going forward
4
5
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
3202
©
thgirypoC
Generative AI (Gen AI) refers to the application of foundation models in order
to create original content across various modalities
Input Activity Outputs
Pretraining Finetuning
• Large datasets (TBs) • Little amounts data
Data needed (MBs) of specific
• No labels needed domain data needed
• Labels may be needed
Question
Answering
Text
Sentiment
Analysis
Images
Information
Extraction
Speech
Foundation
Content
model
generation
Structured Data
Object
Recognition
Signals
Instruction
Following
Generation Task
Traditional AI Task
Source: “On the Opportunities and Risks of Foundation Models”, Centerfor Research on Foundation Models, arXiv, 2021; BCG Analysis
6
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
3202
©
thgirypoC
The GenAI tech stack builds from data, foundation models, & infrastructure
layers to support end-applications
Data modality
Text Code Image Video Speech Other
Vertical AI applications
Vertical applications
Generative AI applications that are tailored to solve vertical-
End- specific use cases
Horizontal applications
applications
Horizontal AI applications
General productivity applications Generative AI applications that solve cross-cutting functional
department use cases such as sales & marketing
MLOPs
General productivity applications
Model Generative AI applications that solve cross-cutting non-functional
Model distribution and non-vertical specific use cases that improve productivity,
infrastructure
such as software development
Model tools & updating
Model infrastructure
Generative AI platforms and tools that aid deployment and
Domain adapted models
integration, improve model performance, expand model
distribution, and improve model training and experimentation
Foundation models
Model APIs
Model APIs
Proprietary foundational models
Model providers who develop foundation models to support
generative AI end-applications and use cases (going forward,
Open-source foundation models domain-adapted models could also emerge, building from
foundation models)
Horizontal, vertical & synthetic
data Data
Enablers
Cloud platforms and specialized Providers of input data used to train and fine-tune foundational
hardware models
Source: Expert interviews; BCG analysis 6
7
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
3202
©
thgirypoC
GenAI is already proving to be a game-changer
Productivity gains are Companies' value prop Barriers to AI are Widely expected to
real and proven are being challenged lower than ever create outsized value
55% -49% Conversational UX eases ~$20B+
human adoption
Same model can handle
faster completion of
drop in Chegg's stock share
multiple downstream of committed VC funding for
coding tasks with higher
price after CEO attributed
tasks Generative AI in the last
success rate using GitHub
the slowdown in three years alone5
CoPilot1 Robust against
subscriptions to Chat GPT
unstructured, unlabeled
messy data
37%
2 months
faster completion of
to 100M users for OpenAI
knowledge work with
ChatGPT,
comparable quality results
the fastest product on
using OpenAI ChatGPT2
record4
1. https://github.blog/2022-09-07-research-quantifying-github-copilots-impact-on-developer-productivity-and-happiness/
2. https://joshbersin.com/2023/03/new-mit-research-shows-spectacular-increase-in-white-collar-productivity-from-chatgpt
3. https://www.cnbc.com/2023/05/02/chegg-drops-more-than-40percent-after-saying-chatgpt-is-killing-its-business.html
77
4. Reuters, Yahoo! Finance, OpenAI
5. Crunchbase, Pitchbook, BCG Analysis
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
3202
©
thgirypoC
What is GenAI, Why now, and Why it matters
GenAI's impact on Private Equity
Impact on portfolio and new deals
Impact on PE operations
Agenda
How to proceed with GenAI deployment
Why BCG for the GenAI journey
How to partner going forward
8
9
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
3202
©
thgirypoC
Early adopters and uses cases likely to emerge in more error-tolerant or easily
audited use cases
Timing of impact depends on creative requirements and error
tolerance… …with uses cases falling into 3 key categories
Error tolerance Nature of use case / task Near-term | Creative, repetitive, highly error-tolerant or easily-
1
Requirement that GenAI output be Degree of flexibility in type of work audited use cases
factually accurate and precise (e.g., creative vs. execution-based)
• Use cases for frequently generating new material, and that do not vary widely
between verticals (e.g., draft marketing copy vs. research report)
• Use cases involving time-consuming, repetitive workflows, and that would
Adoption Illustrative
create competitive or cost advantages for mature companies
• Examples: draft marketing copy, logo creation, ad text/image generation, art
media, grading for short-answer questions, audio/video/image correction,
product review summarization / aggregation
2 Mid-term | Explorative, error-tolerant, moderately auditable use cases
Creative,
Repetitive, • Use cases for more complex, vertical-specific creative tasks, targeting
high-/lower identification of potential new solutions to complex problems
Explorative, error tolerance Long runway for • Examples: Biopharma drug discovery; chemical synthesis pathways; CO2 generative AI adoption
high error (0-3 yrs.) alternative material discovery
Operational, tolerance
no error (3-5 yrs.)
tolerance
(5+ yrs.) 3 Long-term | Operational, no error tolerance, limited auditability
• Use cases for generating, execution, or facilitating complex manual processes
1
with highly specific requirements
3 2
• Examples: Industrial facility design; Machinery part design and tooling / 3D
Industry maturity
printing
Source: Industry interviews; BCG analysis Legend: Adoption range due to variety of underlying use cases within category 9
10
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
3202
©
thgirypoC
GenAI has implications on three key dimensions for PEs
Focus of following sections Focus of upcoming perspectives
Impact on portfolio Impact on fund Impact on
and new investments operations Investment Strategy
GenAI impact on key portfolio Use cases in fund mgmt. from Investment themes emerging from
industries, impact time horizon fundraising to investing to GenAI, Competitive landscape and
and how to prepare portfolio mgmt.; shortlist dynamics
'transformative' use-cases
10
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
3202
©
thgirypoC
What is GenAI, Why now, and Why it matters
GenAI's impact on Private Equity
Impact on portfolio and new deals
Impact on PE operations
Agenda
How to proceed with GenAI deployment
Why BCG for the GenAI journey
How to partner going forward
11
12
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
3202
©
thgirypoC
GenAI will impact industries through both improvements in productivity, and
fundamental changes to customer value proposition
1 Productivity 2 Value proposition
gains Benefits of GenAI change
are expected to be
By automating or augmenting By synthesizing large volumes of
two-fold
repetitive tasks, GenAI models can complex data, GenAI models can
unlock cost benefits in back office enable new offerings; while some
processes, and/or revenue growth industries likely to remain
via improvements in service unaffected, many will benefit from
delivery, personalization, or complementary offerings and others
customer experience may be exposed to new competition
Need to urge portfolio companies to Need to help portfolio companies
understand and adopt available GenAI tools to understand potential
and evolve their org & op model to realize opportunities and risks and
productivity gains navigate strategic changes
13
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
3202
©
thgirypoC
Initial view – list is dynamic as new use-cases emerge
Productivity gains | Most impacted corporate roles expected to be within
1
marketing, customer service, legal, SW and knowledge management
Non-exhaustive
Degree of generative AI impact on roles
High impact Medium impact Low impact
Functions that will likely see extensive or Functions likely to be benefited by Functions with less direct impact from GenAI,
eventually full automation, resulting in one or automation of certain tasks, with potential but likely some benefit from general
more of cost reductions, demand generation cost reduction or per-head productivity productivity applications
due to via higher quality service, or ability to benefits • E.g. email generation
focus resources on higher-value tasks • E.g. generation of finance reports
• E.g., GenAI customer service chatbots to
support call center efficiency
Marketing & advertising Finance & Administration Operations
Human resources Management
Customer service
Product development Community & social services
Legal
Engineering Manufacturing workers
Software development
Business development & Sales Construction workers
Research services & knowledge management IT Support Transportation
Research Mining & extraction labor
Healthcare Services Maintenance and repair
Key use-cases per role covered ahead 13
Source: Expert interviews, LinkedIn Sales Insight, BCG analysis, GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models ( arXiv:2303.10130)
14
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
3202
©
thgirypoC
Initial view – list is dynamic as new use-cases emerge
Value proposition impact | Most industries will see either competitive shifts
2
from new services, or little change; few expected to be completely automated
Non-exhaustive
Degree of generative AI impact on value proposition
Significant impact Industry disruption Limited impact
Businesses whose offerings can be entirely Industries where players can gain competitive Businesses predominantly revolving around
replaced with GenAI services, being both advantage / customer benefits by automating providing or tracking physical goods and
undifferentiated / commoditized and not workflows, but differentiate on existing services, where the core value proposition
dependent on making or tracking a change in a product, services and non-public data. Fast- relates to a change in the customer's physical
customer's physical environment adapting incumbents can retain share against environment
• e.g., translation services, copywriting new entrants, but could still lose revenue • e.g., dentists, food delivery, mining
due to GenAI impact on profit pool.
• e.g., legal services, medical diagnostics
Translation services Capital markets and institutions Consumer and business products
Media Retail & apparel
Personal assistants
Software Restaurants, hotels, leisure
Chatbots
Biopharma Tech hardware and networking
Public information aggregators
Healthcare diagnostics Transportation
Fact-based reporting (e.g., sports results) White-collar services Insurance
Education Healthcare delivery
Proprietary data vendors / aggregators Manufacturing
14
15
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
3202
©
thgirypoC
In aggregate, GenAI's impact on value proposition and productivity will most
disrupt knowledge & content industries; less-digital sectors less impacted (1/2)
Greater productivity
per head, driving
reduced costs or
improved quality of
outputs & delivery
Require adoption of
available GenAI tools
and redesigning org &
op model accordingly
to unlock productivity
Limited Value proposition impact Significant
Significant value prop impact requires deep understanding of
sector/portco-specific dynamics to make strategic choices
Source: BCG Analysis, Expert Interviews, GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models (arXiv:2303.10130)
hgiH
laitnetop
niag
ytivitcudorP
woL
Highly-impacted industries focused on knowledge or content
generation, where GenAI can replace / automate large portions
of operations and service
Industries with significant
changes to value proposition,
Moderately-impacted industries, where many tasks can be leading to turbulence
automated, but value is provided in the form of a product or manifested through new
face-to-face service products / services emerging,
changes in customer demand,
changes in competitive
Low / no impact due to high dynamics, pricing evolution,
dependence on physical and changes in employee bases
products, infrastructure, or
face-to-face service delivery
16
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
3202
©
thgirypoC
In aggregate, GenAI's impact on value proposition and productivity will most
disrupt knowledge & content industries; less-digital sectors less impacted (2/2)
Highly-impacted Potentially disrupted
Accounting Newspaper
Legal services Proofreading & copying
& payroll publishing
services
Translation services
Moderately-impacted
Hospitals & Museums &
Travel & reservation
Utilities
clinics historic sites services
Scientific Clothing
research retailers Graphic design
Digital health
Low / no impact Tax preparation services
Financial
Food & Oil & gas
vehicles Motion picture / sound
beverage extraction
recording
Libraries &
Telecom. Real estate Web search / information
archives
services
Limited Value proposition impact Significant
Non-exhaustive
Low / no impact due to high Moderately-impacted due to task Highly-impacted as Gen AI can Fundamentally impacted ted as the
dependence on physical products automation, but value provided in replace / automate knowledge & value proposition has changed & new
/ services product / personal service content generation products / services have emerged
Source: BCG Analysis, Expert Interviews, GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models (arXiv:2303.10130) 16
hgiH
laitnetop
niag
ytivitcudorP
woL
GenAIimpact on PE Portfolio and new deals Impacted industries
Directional
17
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
3202
©
thgirypoC
A set of consideration factors requires assessment to gauge the productivity
gain potential and value proposition impact for portfolio and new investments
GenAI impact consideration factors
Illustrative
GenAI impact Consideration factors Questions to determine GenAI impact
Bespoke assessments of
1 consideration factors
How much of companies' total cost basis can be reduced by adopting GenAI-enabled
conducted to determine
Efficiency gains
software solutions?
1. Productivity gain potential
and 2. Value prop. impact for
a given sector or company
Effectiveness gains How can GenAI impact the quality of outputs?
driven by GenAI technology
Productivity
To what degree will companies' operating models (e.g., workforce composition,
Operating model shifts
gain potential
workflow, job responsibilities) change due to AI?
To what degree do sector-specific barriers exist (e.g., regulation, legal threats) and
Adoption barriers
what is their impact (e.g., require legislation, customer preference shift)?
2
To what degree does GenAI enable new product offerings, which may either reset
Competitive impact
competitive dynamics through substitution or induce greater competitive intensity?
To what degree will GenAI erode or increase demand? Is the mix shift favorable or
Demand impact
Value unfavorable?
proposition
Impact Price and margin
To what degree will GenAI drive a commoditization of prices or margin?
impact
Alternatively, would companies be able to improve margin given productivity gain?
Assessment of each criteria includes evaluation of current GenAI technology, and how it is expected to evolve
Source: BCG analysis 17
18
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
3202
©
thgirypoC
Illustrative &
Not exhaustive
GenAI capabilities are continuously developing; as such, assessing GenAI's
impact requires considering current as well as potential future capabilities
High degree of feasibility Unclear degree of feasibility
GenAI capabilities with path to GenAI capabilities requiring
Current GenAI capabilities
development in near future substantial incremental R&D
• Generate 'second drafts' of long-form responses • Use of e.g. LoRa for organization specific fine tuning • Ability to reliably assess factual accuracy of output
to general prompts based on learned patterns to increase accuracy of output (e.g., scientific • Move beyond token window of LLMs to work on large
Text from training data papers) bodies of text or extended conversations
• Enhance search capabilities through • Agent-based capabilities to generate and perform • Architecture to hold long conversations without
information aggregation and extraction lists of tasks "drifting" from initial parameters
• Generate basic code from natural language • Expansion of support to more programming • Interpret intended use case and generate context-
Code/ prompts and autocomplete existing base code languages and frameworks driven code
Data • Code translation from one language to another • Integration with other platforms for specific tasks • Ability to train on self-generated data via
(e.g., Java to Python) (e.g., Wolfram Alpha for mathematical calculations) unsupervised learning
• Generate unique art, logos and photographs • Ability to tweak generated output via iterative • Ability to create 3d models requiring internal
Images
from natural language feedback from user in natural language interface consistency (e.g., architectural models, CAD renders)
• Short clips from natural language prompts; first • Generate video 'first drafts' from natural language • Using natural language to create complex video
attempts at 3D renders and video models • Generate high quality 3D scenes with consistency scenes/models that are fully 3D modeled, resulting
Video/
• Advanced speech recognition, transcribing, between frames in highly personalized media
Speech
highlighting and outlining meeting notes • Ability to tweak generated output via iterative
feedback from user in natural language interface
Current-state Future-state
Source: BCG analysis, Generative AI: A Creative New World | Sequoia Capital US/Europe
19
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
3202
©
thgirypoC
GenAIimpact on PE Portfolio and new deals Impacted industries
40-50% of avg PE portfolio within industries expected to have high impact from
GenAI – though sub-sector & company specific analyses are warranted
Example of Industries1 - Average global top 20 PE funds portfolio (% of total)
Professional Services
Software
Biotech
30% Media
Retail & Apparel
IT Services
Capital Markets
Healthcare Services
Insurance
Consumer Products
Chemical & Gases
Commercial Products
Energy Equipment
Source: BCG analysis; Notes: (1) example set and not exhaustive of all industries; penetration is estimation 19
hgiH
laitnetop
niag
ytivitcudorP
woL
Directional
5% 10%
3%
Whilst general patterns
3% exists, GenAI's impact
5%
will be vastly different
4%
5% for sub-sectors and
3%
companies within
industries
5%
4%
Size of bubble = Approx.
Limited Value proposition impact Significant
portfolio share of PE funds
20
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
3202
©
thgirypoC
GenAI impact on PE Portfolio and new deals Implications for PE
What does this
mean for PE?
21
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
3202
©
thgirypoC
GenAIimpact on PE Portfolio and new deals
PortCo implications | Org. changes required in PortCos exposed to high
productivity gain potential, whilst value prop changes require strategic review
Highly-impacted industries | "Organizational Re-haul" Fundamentally impacted
industries| "Moments of Truth"
Significant organizational implications: Significant workforce changes
required to adapt to GenAI – e.g. re-skilling, hiring, and operating model Strategic shifts: Radical change
(positive or negative) in
Limited strategic implications: Limited effects on strategic positioning and
competitive differentiation strategic positioning of
competitors
Potential market share capture: Players with strong data governance
Demand impact: Change in
frameworks and tech org hold an initial speed-to-adoption advantage
demand for core
Low/Moderately-impacted industries | "Fast following" products/services
Negligible short-term impact: Limited (short-term) effects on productivity Intensified competition:
and value proposition Heightened competition with
new players and improved
"Second-mover" advantage: Oppt'y to adopt GenAI at lower cost/risk later in
offerings
time through monitoring and gaining familiarity with GenAI developments
21
hgiH
laitnetop
niag
ytivitcudorP
woL
Implications for PE
Directional
Limited Value proposition impact Significant
22
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
3202
©
thgirypoC
GenAIimpact on PE Portfolio and new deals Implications for PE
Fund actions | Identify cross-portfolio productivity gain potential and assess
strategic implications of value proposition changes for impacted PortCos
Productivity gain
Stand up Prepare for
potential:
Set objectives Size the prize
functional CoEs1 implementation
Identify and • Determine sub-set of • Aggregate headcount by • Identify & drive best • Assess implications on
implement PortCos and functions to (sub-) function across practices across PortCo people, processes & tech
be evaluated portfolio
operational best • Set up GenAI focused • Consider extent which
practices across • Consider the end-state • Estimate the productivity teams across key productivity translates to
portfolio, function goal of the evaluation improvement potential impacted functions (e.g. cost take out, workstream
(e.g, cost take out vs. by (sub-) function call centers) reinvention or op model
by function
quality improvement) enhancement
Value proposition
Identify high Assess scenarios Develop option Set up war room
impact:
impact-sectors sets
• Screen the portfolio for • Initiate deep-dive • Evaluate options: e.g. • Assemble war room
Screen for highly
high impact industries analysis for prioritied product dev., M&A, involving mgmt & board
impacted industries
PortCos to estimate size partnerships
• Look for anticipated • Develop action plan and
and assess strategic
and scope of impact
changes in core • Estimate execute with high
implications PortCo-
offerings, customer • Assess PortCo's positions costs/investments urgency
by-PortCo
demand, competitive vs. key competitors required and potential
dynamics outcome
1. Center of excellence 22
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
3202
©
thgirypoC
What is GenAI, Why now, and Why it matters
GenAI's impact on Private Equity
Impact on portfolio and new deals
Impact on PE operations
Agenda
How to proceed with GenAI deployment
Why BCG for the GenAI journey
How to partner going forward
23
24
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
3202
©
thgirypoC
Non-exhaustive
Four main GenAI capabilities that can be leveraged across PE operations
Portfolio
Fund operations Deal activity
operations
Fund support Assessment & inv. Exit & re- Portfolio mgmt. &
Fund strategy Deal sourcing
functions decision investment monitoring
Automating/Assisting analysis: e.g, Sentiment analysis, data cleaning, market dynamics and trends, areas of concern / uncertainty
Content summarization & synthesis: e.g., summarization of expert interviews, long documents, publications,. (IR updates/ portfolio developments)
Knowledge mgmt. & access: Increased self-service / access to internal data, policies, training and advanced applications
Content creation: E.g., presentations (pitch decks etc.) and documents (emails, marketing material, legal contract drafting, etc.)
24
Source: Expert interviews; BCG analysis
seitilibapaC
View as of April 2023
25
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
3202
©
thgirypoC
View aBs aocfk Mupay 2023
Non-exhaustive
Indicative
Potential products for private equity operations enabled by GenAI (1/2)
quantification
Adoption Value &
Product Description timeline impact
Investment themes Early identification of emerging investment themes based on e.g. social media, news
identification discussions, market publications and developments, investments and/or expert interviews
Investment success factor Identifying and detailing investment criteria for target screening based on past successful
identification investments, market developments and desired investment themes to invest behind
Investor relations (Fundraising Identifying new potential investors across data sources and internet outside of traditional
analytics, relationships) databases, and based on their lifecycle tailor products and timing to their emerging needs
Product development assistant General product development, incl. e.g., code writing assistance for IT department to
(e.g. coding) accelerate IT developments and data science efforts across the company
Risk monitoring (LP financing, Risk and compliance monitoring across support functions, incl. KYC/AML, invoicing and
and compliance) expense compliance, LP agreement and regulatory / legal compliance monitoring
Talent mgmt. (marketing, Create job descriptions and personalized outreach, and by analyzing missing team
interview assessment, capabilities using employee data, such as performance reviews and job history, it can help
recruiting) suggest applicants to target or review, as well as to help interview process
Landscape assessment
Analyze large volumes of data from various sources and generate insights that aid
(dynamics in particular
landscape assessment ahead of specific target identification
segments)
Target identification, filtering Identify potential targets based on investment criteria, competitive environment and
and comparison investment thesis
Note: Adoption and impact quantification is to be considered relative to one another (what will be the first, middle and last group of applications emerging)
Source: Desktop research; BCG analysis 25
Short- Long-
Degree of impact Wide adoption in: Short-term <3yrs Long-term >5yrs
term term
26
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
3202
©
thgirypoC
View aBs aocfk Mupay 2023
Non-exhaustive
Indicative
Potential products for private equity operations enabled by GenAI (2/2)
quantification
Adoption Value &
Product Description timeline impact
Generative AI creates research and outline content of (pre-) due diligence (incl. market
Due diligence assistance
dynamics, landscaping etc.) in addition to identifying key analysis questions
Tool to draft deal structure and align incentives efficiently across stakeholders to create
Deal structuring tool
customized offer to decision makers
Market survey creation and Creation of market survey questions in research process, e.g., customer sentiment or
analysis customer KSC, and preliminary analysis of survey results
VDD & Sales material Draft first version of VDD and sales prospectus including company profile, historical
assistance financial information etc., and facilitate data collection, research as part of prospectus
Automated exit assistant Generative AI powered assistant to manage and co-ordinate process stakeholders with
(process mgmt.) scheduling, automated responses and transaction marketing to potential buyers
Continuous portfolio monitoring of emerging risks and performance based on updating
Portfolio monitoring (risks and
market conditions and developments with potential to synthesize data for outlook and
performance etc.)
sensitivity assessments
Value creation plans and Creating creative value creation plans and ideas through analysis of current events and
Variable
assistance market dynamics, to best help portfolio companies grow their business
In addition, range of horizontal Range of horizontal GenAI applications for HR processes (recruiting, interviewing, etc),
Variable Variable
applications non-PE specific legal (contracting, compliance reviewing, etc.), finance (invoicing, expensing), marketing (
Note: Adoption and impact quantification is to be considered relative to one another (what will be the first, middle and last group of applications emerging)
Source: Desktop research; BCG analysis 26
Short- Long-
Degree of impact Wide adoption in: Short-term <3yrs Long-term >5yrs
term term
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
3202
©
thgirypoC
What is GenAI, Why now, and Why it matters
GenAI's impact on Private Equity
Impact on portfolio and new deals
Impact on PE operations
Agenda
How to proceed with GenAI deployment
Why BCG for the GenAI journey
How to partner going forward
27
28
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
3202
©
thgirypoC
Portfolio Actions
• Assess entire portfolio and identify opportunities,
efficiencies and risks arising from GenAI, analyzing its
impact on productivity and value proposition
• Start identifying 'transformative use cases' for
internal operations, partner up to experiment with
Several
technology
immediate steps
• Start evaluating 'Responsible AI' policies for the
for your GenAI organization to manage potential risks
roadmap
Investment Actions
• Review investment strategy, including identifying
investment themes in GenAI
Setup GenAI control tower to coordinate efforts across
the organization and within the portfolio
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
3202
©
thgirypoC
Generative AI does still
come with risks...
… but many of
Intellectual property and these can
copyright infringement
eventually be
mitigated with
Biased outputs
the right
Responsible AI
Cybersecurity and
data privacy
approach
Hallucination / confidently
wrong answers
Midjourney | "leaping between two boulders"
29
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
3202
©
thgirypoC
What is GenAI, Why now, and Why it matters
GenAI's impact on Private Equity
Impact on portfolio and new deals
Impact on PE operations
Agenda
How to proceed with GenAI deployment
Why BCG for the GenAI journey
How to partner going forward
30
31
.devreser
sthgir
llA
.puorG
gnitlusnoC
notsoB
yb
3202
©
thgirypoC
Cross-functional teams with specialized
1 skillsets across the AI/ML stack with proven
3,000+
track record of enabling applications in GenAI
tech build & design team
Advisor to
Deep understanding of how investors are
2
world leading PE Funds
assessing the GenAI landscape globally
What makes and Software Investors
BCG unique
in this space?
Prioritized ac |
291 | bcg | harnessing-the-power-of-genai-in-indonesian-financial-services.pdf | + +
Harnessing
the Power of (Gen)AI
in Indonesian
Financial Services
August 2024
Table of Contents
02 Foreword: Unlocking New Ventures & Financial
Empowerment in a Booming Economy
03
Executive Summary
06
The Role of AI & GenAI in Transforming
Indonesian Financial Services
11
3 Strategic Plays for GenAI Integration
13
Themes Defining the Opportunity for AI & GenAI
23
Inherent Risks & New Challenges in the Age of GenAI
27
Enterprise Foundations Framework for (Gen)AI
Success
32
Authors
Harnessing the Power of (Gen)AI in Indonesian Financial Services
Foreword
Unlocking New Ventures &
Financial Empowerment
in a Booming Economy
When we set out to build this report, we is leading to fast-paced investment
sought to explore the strategic impact of discussions about how to multiply and
Generative AI (GenAI) on the financial bolster local data centers for domestic
services sector in Indonesia. We wanted to GenAI development. Because data centers
understand its potential to drive significant consume so much power, public and
sector-wide transformation and deliver a private sector stakeholders are having
new generation of investment and venture serious discussions related to Indonesia’s
opportunities in an economy growing at 5% energy infrastructure, the clear role that
annually and projected to reach US$1.47 renewables must play, and how to finance it
trillion this year.
all.
The report was motivated by a global shift This report provides an in-depth analysis of
toward GenAI, which is rapidly altering the Predictive AI and GenAI adoption within
competitive landscape across many Indonesian financial services. It outlines
industries. In local financial services, this emerging opportunities, highlights
technology offers not just incremental fascinating case studies, and discusses
improvements, but foundational shifts that ways to address key challenges. It also
enable new business models and enhance offers actionable insights for companies
operational efficiencies.
seeking to enter the market as well as
established entities aiming to integrate new
AI has the potential to increase access to technologies.
financial services across broader
demographics and underserved markets. In As we release this report, we urge
particular, it can help rural and remote policymakers, business leaders, and
communities access services such as investors to consider its contents and
credit, insurance, and savings, thus insights. The integration of GenAI presents a
stimulating economic engagement at unique opportunity to redefine financial
multiple levels.
services in Indonesia, enhancing both
economic efficiency and empowerment.
For global institutional investors, there is a
robust and multifaceted opportunity in
Pandu Sjahrir
Indonesia as we speak. AI and GenAI
promise great applicability in sectors such Founding Partner, AC Ventures
as healthcare and education, two spaces Head of Technology & Digital
that tie directly into financial services via Finance, Kadin Indonesia
insurance and credit.
Meanwhile, as the government seeks to
establish a regulatory framework for this Andy Lees
new tech, the concept of “Sovereign AI” has Managing Director & Partner,
become top-of-mind for the incoming BCG X
administration. In a cascading fashion, this
Harnessing the Power of (Gen)AI in Indonesian Financial Services 2
Executive Summary
AI and GenAI are transforming the financial sector, reducing barriers to entry and enabling new
players to compete in the market against established institutions. In Indonesia, these technologies
are enhancing financial education, credit scoring, customer service, and more.
How Financial Institutions View GenAI:
85% of global financial institutions view GenAI as highly disruptive or
transformational, yet only 18% have a clear strategy and are implementing it.
58% of GenAI users globally are saving at least five hours a week, shifting focus to
using this extra time for value creation and customer joy.
Global worker confidence in GenAI has increased by 16 percentage points to 42%
since 2023, but fear of job loss has also risen by 5 percentage points to 17%.
51% of Indonesian financial institutions are focusing on deploying GenAI for
everyday tasks, and an additional 27% see great opportunity in inventing new
products and services.
49% of business leaders in Indonesia's financial sector prioritize GenAI to enhance
customer service, with 34% already seeing tangible benefits from its deployment.
61% of Indonesian financial institutions are confident they have strong
technological readiness with established data and tech stacks for GenAI.
3 Strategic Plays for AI Integration
Deploy Reshape Invent
GenAI in everyday critical functions end-to- new experiences, offerings,
tasks for broad end for radical efficiency and business models
enterprise productivity and effectiveness powered by GenAI
Financial institutions in Indonesia can GenAI pilots are in progress across all major
successfully integrate AI and GenAI into their Indonesian financial institutions, with many
operations using the proven strategic plays of already transitioning these initiatives into
Deploy, Reshape, and Invent.
scalable projects. These efforts aim to
democratize financial access and inclusion,
This involves deploying enhancements to especially for underserved communities,
everyday use cases, both internally and on the aligning with the stringent compliance
customer-facing side; reshaping in-house requirements of Indonesia’s Personal Data
processes and skills for greater efficiency; and Protection (PDP) Law. The government is also
inventing completely new products to achieve enhancing AI and GenAI regulatory frameworks.
greater savings and create new revenue
streams.
Harnessing the Power of (Gen)AI in Indonesian Financial Services 3
Advice for Business Leaders
The potential of GenAI in
Indonesia's financial sector is
evident—it can expand financial
access, enhance customer
experiences, enable rapid
service scaling, and more.
Our research indicates that both
major financial institutions and
fintech startups have quickly
adopted this technology. But
many of these initiatives are still
in the pilot stage and have not
yet delivered substantial
business value at scale.
Financial institutions need a strategic
framework for AI that includes governance,
operations, and talent management to align
with business objectives. This report reveals
that neglecting the business perspective can
cause AI implementations to fail. It is crucial
to ensure AI projects match actual business
needs, emphasizing data security and
regulatory compliance, while also focusing
on goal setting, responsible AI development,
and staff training. This approach balances
technological readiness with business and
ethical considerations, integrating AI
effectively within organizational practices.
Harnessing the Power of (Gen)AI in Indonesian Financial Services 4
Kadin Indonesia’s Advice
to Policymakers
AI and GenAI have clear potential to boost To realize this, Indonesia may strategically invest in
Indonesia's economic growth and global robust, sustainable data center infrastructure
competitiveness by transforming the way state- enhanced by renewable energy sources, and
owned enterprises and government agencies underpinned by stringent legal frameworks for data
operate. With this in mind, the local administration privacy and autonomy. Emphasizing public-private
could prioritize the creation of trusted AI within its partnerships and collaborations could accelerate
overarching strategy.
the process, while strict cybersecurity measures can
safeguard the country’s critical data assets.
The strategic development of sovereign AI offers
promise across various sectors, from financial While this report highlights private-sector strategies
services to clean energy, commerce, and beyond. for AI and GenAI implementation, government
Developing this technology under the oversight of leaders may want to observe, adapt, and modify
the incoming administration has the potential to these approaches for their own initiatives while also
bolster national security and ensure Indonesia's considering three core strategies:
technological independence for future generations.
1
Establishing a National AI Research & Development Fund
Central to a potential sovereign AI initiative could be Cost-Benefit Outlook: The primary expenditures
the creation of a National AI Research and would involve initial funding for research and key
Development Fund. This fund would focus on infrastructure, such as data centers. The future
spurring AI innovation in critical economic sectors benefits may include but are not necessarily limited to
such as healthcare, agriculture, and manufacturing. enhanced sector efficiencies, job creation in AI and
Government agencies would need to allocate GenAI development, and improved international
appropriate resources to this fund, sourced or business competitiveness, all of which are relevant to
redirected from Indonesia's existing technology- annual GDP growth due to productivity improvements
related investment funds and programs. and new business opportunities.
2
Implementing AI Education & Training Programs
Integrating AI and GenAI into Indonesia’s state- Cost-Benefit Outlook: Costs on this frontier involve
owned digital ecosystem—and society at large— the operational expenses of education programs.
would require a strong emphasis on education and The anticipated benefits are a skilled workforce,
training. These initiatives could span all reduced unemployment, and increased average
educational levels, incorporating academic and income levels. Such educational initiatives may
vocational training to prepare a workforce skilled in reduce skill gaps relative to Indonesia’s global
AI technologies. Significant annual investments peers, enhance employment rates in the
may need to be made to develop curricula, train technology sector, and boost median salaries due
educators, and establish AI-equipped learning to an influx of skilled labor.
centers for government agencies, as well as
Indonesia’s secondary and tertiary schools.
3
Incentivizing AI Startups & Foreign Investment
For a thriving sovereign AI ecosystem, the Cost-Benefit Outlook: While administrative costs of
Indonesian government could also support AI managing these drives may be clear and present,
startups and attract direct foreign investment with such incentives could spur more domestic and
favorable policies and incentives. A specific portion international investments in Indonesia, driving
of the annual budget could be allocated for tax technological advancements and further economic
incentives, direct subsidies, and support services expansion. A few key potential impacts may include
for AI companies licensed to operate in Indonesia.
an increase in foreign direct investment, the creation
of high-tech jobs, and a more competitive market
environment driven by innovation.
Harnessing the Power of (Gen)AI in Indonesian Financial Services 5
The Role of
AI & GenAI
in Transforming
Indonesian
Financial
Services
Harnessing the Power of (Gen)AI in Indonesian Financial Services 6
The Basics:
Predictive AI vs GenAI
In the context of financial services, the use of AI is compliance checks, but also at making consistent
not a particularly new trend. However, it is crucial to and accurate inferences—often surpassing human
differentiate between Predictive AI and GenAI. capability—from vast data sets, particularly in
Predictive AI excels not just at automating routine areas like risk and fraud detection. [Exhibit 1]
tasks such as transaction processing and
GenAI complements and co-exists with previous
Predictive AI efforts, enabling new and broader
applications
[Exhibit 1]
Predictive AI Generative AI
“Left brain” “Right brain”
Decision-making and Content creation,
optimization
qualitative reasoning,
orchestration of other
Each algorithm systems
constrained to specific
problem space
Multi-applications
Limited range of possible Unlimited range of
outputs possible outputs
Impact: 5% of employee Impact: 50% of employee
tasks and workflows; tasks and workflows;
focuses on key decisions complemented with impact
by Predictive AI
Exhibit 1
In contrast, GenAI, as a newer technology, takes a This capability not only saves
significant leap forward by not only understanding
the user's context—including who they are and why time and energy but also
they are requesting information—but also by
broadens the use cases from
communicating in a more flexible, fluid, and natural
manner compared to the more algorithmic, rules- very specific applications with
based interactions typical of Predictive AI. This
capability allows GenAI to generate new, context- Predictive AI to a wider range of
aware content on demand, rather than simply
applications with GenAI.
retrieving existing information.
Harnessing the Power of (Gen)AI in Indonesian Financial Services 7
GenAI's conversation, summarization, and knowledge extraction capabilities may help create broader
access to loans and working capital for Indonesians. This could noticeably influence the nation's economic
landscape and potentially support financial inclusion. [Exhibit 2]
Main capabilities of GenAI that drive value
[Exhibit 2]
Content Problem
Tech Knowledge
Conversation generation/ Summarization Ideation solving
AI agents
capabilities extraction
transcription & Insights
Interactive The creation of Summarization G eneration of Extraction of Logical & The solving of
& dynamic specific types of of large new
structured reasoning complex tasks
engagement of content (e.g., text, amounts of & innovative knowledge from process to by planning and
information, ideas, images, videos, information or ideas, concepts unstructured or make executing a set
or questions audio, code) text into shorter, or designs (e.g., semi-structured
inferences, draw of actions using
Description between humans more concise unique product data sources conclusions, a suite of tools
& AI systems, versions, that solutions, make informed
responding to capture the key exploration of judgments, and
questions and points of the uncharted derive new
generating content territories in insights based
appropriate scientific fields) on available
responses information,
data, or
knowledge
Exhibit 2
BCG and ACV conducted a survey of 41 business
leaders from traditional financial institutions in Organizational perspectives
Exhibit 3
on Predictive vs GenAI
Indonesia for this report. The majority of leaders
view Predictive AI and GenAI as complementary
technologies. However, nearly a quarter of the
leaders consider Predictive AI more crucial,
particularly due to its role in core financial products
like credit scoring and credit limit calculations. 66%
Despite this, it is evident that GenAI has quickly
established itself as a legitimate and valuable tool
for the financial sector within just two years.
[Exhibit 3]
A majority of business 22%
12%
leaders see Predictive AI
and GenAI as
They are Predictive AI is GenAI is more
complementary more important important
complementary
[Exhibit 3]
Harnessing the Power of (Gen)AI in Indonesian Financial Services 8
Economic Impact & Growth
Projections
The economic impact of AI and GenAI in Southeast Asia is poised to be significant, with projections
indicating a substantial US$1 trillion GDP uplift by 2030.
In our survey, we found that business leaders' top three expectations from GenAI include improving
operational efficiency within their organizations, enhancing customer experience, and spurring innovation in
financial products and services. [Exhibit 4].
The most significant
Anticipated benefits of Integrating GenAI Exhibit 4
anticipated benefit
85%
of GenAI is the
76%
improvement of 66%
56%
operational
41%
efficiency, such as 32%
the automation of
basic tasks.
[Exhibit 4]
Improved Enhanced Innovation in Improved Cost Advanced
operational customer product/ employee reduction risk
efficiency experience services productivity management
Harnessing the Power of (Gen)AI in Indonesian Financial Services 9
Impact on Traditional Financial
Institutions & Fintech Startups
Traditional Financial Institutions
GenAI adoption by banks has been surprisingly rapid. All mid to All mid to large-sized
large-sized financial institutions surveyed in Indonesia are at financial institutions
the minimum running pilots and proof of concepts with GenAI, surveyed are at least
with a majority having implemented a few use cases at scale
piloting GenAI, with a
for customers or employees. Considering GenAI has only been
majority having
on the market for two years, its rapid adoption by traditional
financial institutions, even within a highly regulated industry, implemented a few use
underscores how seriously these businesses are taking the cases at scale for customers
technology. [Exhibit 5] or employees.
[Exhibit 5]
Exhibit 5
GenAI use in traditional financial institutions %
Piloting the technology and running 41%
proof of concepts
Some instances of implementation at 39%
scale for employees or customers
Several instances of implementation at
20%
scale for employees or customers
Fintech Startups
The AI and GenAI boom is lowering barriers to entry The rise of fintech companies, empowered by AI
in Indonesia's financial services sector, enabling and GenAI, is transforming Indonesia's credit
fintech startups to challenge traditional banks services sector by offering personalized, accessible,
more effectively. This shift also encourages and efficient credit solutions—tasks that previously
strategic partnerships that merge fintech required large departments and specialized
innovation with the robust infrastructure and expertise. These AI-driven innovations enable
operations of conventional banks, enhancing fintech firms to gain significant market share,
operational efficiencies and expanding service challenging the dominance of traditional financial
offerings.
institutions and forcing them to innovate.
Many of Indonesia's large banks have introduced
Innovative Dynamics in the signature digital banking apps, allowing users to
Financial Landscape
manage their financial lives fully online—from
opening accounts to making investments—without
AI and GenAI are blurring the lines within the fintech ever visiting a bank branch.
ecosystem, enabling SaaS companies to move into
fintech, fostering partnerships between fintechs Mid to large-sized banks are increasingly exploring
and banks, and pushing banks to adopt fintech-like microfinance. Our data shows that 51% of surveyed
strategies.
entities are testing or have implemented GenAI to
simplify lending and loan disbursement in this area,
For example, ESB has transitioned from a restaurant but long-term benefits have yet to be realized
SaaS platform to facilitating loans by leveraging (Exhibit 9). 12% of respondents see significant
transaction data to assess SME credit risks more potential in GenAI for microfinance, as it enhances
effectively.
credit assessment with unstructured data sources.
Harnessing the Power of (Gen)AI in Indonesian Financial Services 10
3 Strategic
Plays for GenAI
Integration
Harnessing the Power of (Gen)AI in Indonesian Financial Services 11
Both established institutions and early-stage Through our survey results, we found that financial
fintech startups recognize the importance of a institutions are prioritizing the integration of GenAI
diversified set of strategic plays when integrating AI into everyday tasks (Deploy) to enhance specific
and GenAI into their existing technology stacks and operations, such as customer service. This is
operational procedures. Based on our findings, followed by a focus on the creation of new products
companies successfully capturing outsized upside and services (Invent), indicating that companies
engage in one or more of the following strategic are addressing immediate needs while also aiming
plays. We’ve broken them down into three key to leverage GenAI for the development of entirely
categories: Deploy, Reshape, and Invent. new products and services. [Exhibit 7]
There are 3 strategic plays to leverage when
considering integrating GenAI into products, services,
and systems
[Exhibit 6]
Deploy Reshape Invent
GenAI in everyday critical functions end-to- new experiences,
tasks for broad end for radical efficiency offerings, and business
enterprise productivity and effectiveness models powered by GenAI
This refers to the use of AI and Reshape involves the Invent is the practice of
GenAI for everyday tasks, reimagination of specific in-house leveraging GenAI to pioneer new
typically involving an existing processes, for radical operational experiences, offerings, and
product. Examples include efficiency changes. Examples business models. Firms create
employees using GenAI to could include GenAI-powered hyper-personalized financial
summarize meetings and calls chatbots that handle customer products and services that meet
in seconds, highlight key inquiries with speed and precision, the unique needs of Indonesia's
takeaways from lengthy reports, tools that detect fraudulent diverse population. GenAI also
and quickly create first drafts of activities by analyzing patterns in facilitates the rapid testing and
communications. Additionally, real time. This requires a holistic, deployment of innovative
GenAI can assist with more centrally coordinated effort to business models, such as
technical and specific aspects transform work and workforce mobile-first banking solutions
of workflows, such as data dynamics with responsible AI and micro-financing tailored to
labeling and analysis. This enablers where possible, rather rural communities. New products
approach dramatically saves than a set of isolated use cases enhance customer engagement
time and also increases the with limited scalability and but also expand financial
quality of outputs.
decentralized rollout.
inclusion across the archipelago.
Traditional financial Exhibit 7
GenAI strategic priorities for
institutions are focusing on
financial institutions in Indonesia
deploying GenAI for
51%
everyday tasks, and see
great opportunity in
27%
22%
inventing entirely new
products and services.
[Exhibit 7]
Deploy Reshape Invent
Harnessing the Power of (Gen)AI in Indonesian Financial Services 12
Themes
Defining the
Opportunity for
AI & GenAI
Harnessing the Power of (Gen)AI in Indonesian Financial Services 13
In addition to surveying 41 business leaders from relationship managers (RM) to better engage
traditional financial institutions in Indonesia, we with customers. About one-fifth (22%) of
also conducted interviews with five fintech startups companies have introduced such RM tools and
in the country’s financial services sector. These are now seeing positive outcomes.
executives are integrating AI and GenAI into their
operations, and we’ve identified six recurring Hyper-personalization and fraud management
themes of opportunities across a range of are identified as areas with significant potential,
company types, from emerging micro-fintech firms though they are still in the nascent stages of
to well-established financial institutions.
implementation, with only 7% of companies
having launched related products and seen
Per our survey, nearly half (49%) of business leaders gains. There is a notable opportunity for
see customer service as the primary application for organizations to explore hyper-personalization
GenAI, with a third (34%) of companies already further, which, if effectively implemented, can
reaping benefits from its implementation. The next lead to successful upselling and cross-selling on
promising area for is enhancing tools for financial platforms. [Exhibit 9]
GenAI leads in customer service among business leaders, with 49%
identifying it as the primary application for the technology. 34% are
already seeing benefits, while relationship management, hyper-
personalization, and fraud prevention show potential.
[Exhibit 9]
Where financial institutions see the most future potential for GenAI Exhibit 9
5% 7%
7% 15% 12%
10% 15% 29%
7% 15% 22% 37%
12%
22% 15% 20% 22%
24% 17%
15%
37% 37% 15% 20%
49% 15%
20% 12%
7% 12% 2%
15%
7% 7% 7% 12%
Customer service Tools for Relationship Fraud management Hyper-personalization Front and back office Easier lending and loan
agents to improve Managers to engage with Generative AI to engage and upsell employee productivity disbursement in
customer experience customers detection and customers leveraging Generative microfinance
management AI tools
Most potential High Potential Moderate Potential Low Potential Least potential Not Ranked
Progress of GenAI implementation & realization of benefits Exhibit 9
0%
5% 7% 9% 10%
22% 17% 20%
20%
37% 39%
29%
34%
49%
51%
10% 24%
37% 39%
7%
34% 15% 10% 5% 22%
22%
7% 7% 7% 5%
Customer service Tools for Relationship Fraud management Hyper-personalization Front and back office Easier lending and loan
agents to improve Managers to engage with Generative AI to engage and upsell employee productivity disbursement in
customer experience customers detection and customers leveraging Generative microfinance
management AI tools
GenAI rolled out and benefits are being realized GenAI rolled out, but benefits are not yet being realized
Piloting GenAI Planning to implement GenAI, but not yet in pilot phase No plans for GenAI
Harnessing the Power of (Gen)AI in Indonesian Financial Services 14
1
Customer
AI-driven customer service improves
Success
customer experiences by efficiently handling
inquiries through natural language
of the Future processing, enhancing both customer
satisfaction and operational efficiency.
Fintech Startups Traditional Financial Institutions
Customer service is pivotal in helping fintech Banks are not just focusing on enhancing customer
startups scale by reducing costs and maintaining service with GenAI but are aiming for a
the necessary 24/7 support for sensitive products comprehensive service transformation across
like credit cards. With government caps on credit multiple touchpoints. They are employing GenAI-
lending profits, minimizing non-core operating enabled tools and processes to quickly respond to
expenses is essential. Utilizing GenAI as the primary customer inquiries, build stronger relationships
interface in customer service ensures consistent between clients and relationship managers, and
support at significantly reduced staffing costs. This shift conversations from service to sales.
strategy not only enhances customer experience
but also lowers non-core expenses.
Fintech Case Study
SkorLife, an Indonesian credit builder,
aimed to reduce customer service
costs by 50%.
As customer service constitutes 40% of its operating expenses, the
team implemented GenAI assistance to provide consistent, 24/7
support. Recognizing that such use cases are applicable across
industries, SkorLife chose to partner with a vendor to develop its
GenAI customer service capabilities instead of building them in- Ongki Kurniawan
house. Co-founder & CEO
44% of business leaders stated that their primary
Exhibit 10
Largest benefits of using goal for implementing GenAI in customer service
GenAI customer service is to establish 24/7 support, crucial for addressing
urgent issues like credit card fraud and loss that
demand immediate action. Additionally, 37% aim
to use GenAI to enhance the quality of customer
44%
service. Only 5% of traditional financial institution
37% respondents consider cost reduction as a goal for
adopting GenAI, highlighting a significant
difference from the priorities of smaller fintech
companies. [Exhibit 10]
Prioritizing 24/7 customer
12%
support with GenAI in
5% traditional financial
2%
institutions: A key strategy
24/7 Higher quality Reduced Lower cost to Improved
customer customer waiting serve rates of debt
support service period for customers collection beyond cost reduction.
customers [Exhibit 10]
Harnessing the Power of (Gen)AI in Indonesian Financial Services 15
2
Productivity GenAI is transforming employee productivity. A
recent BCG survey on AI usage in the workplace
Co-Pilot
identified Asia Pacific (APAC) as one of only two
regions globally where a majority of frontline
workers utilize GenAI tools. Furthermore, nearly a
third of APAC employees have received training
on how GenAI will impact their roles, a figure that
surpasses the global average.
Fintech Startups Traditional Financial Institutions
Enterprise-grade instances of GenAI The potential productivity gains for banks in Indonesia are
remain too costly to deploy across entire immense at scale. With thousands of employees spread
companies. Instead of adopting a broad across various branches, the implementation of GenAI
use case approach, leaders are first could revolutionize operations, streamlining tasks that
assessing the costs and benefits to ensure previously took hours into mere minutes. This results in
that their employees can effectively utilize significant time savings across the board. GenAI users
the technology. This strategy focuses on report saving at least five hours a week using these tools.
addressing time-intensive, low-value tasks Additionally, there is a significant opportunity to invest in
within specific teams and processes. upskilling employees to effectively interact with GenAI tools.
Fintech Case Study
JULO, a credit lending app, has its data
analytics team utilizing GitHub Co-Pilot to
boost coding productivity, resulting in a 2x
increase in engineer efficiency.
For data warehouse queries in BigQuery, the team employs Gemini, which
accelerates query responses by 2x to 3x, significantly reducing the time to
gain insights. Additionally, mundane tasks like data classification are
managed by GenAI, which automates the tagging and labeling of data, Martijn Wieriks
minimizing the need for manual intervention. Chief Data Officer
Exhibit 11
37% of business leaders are leveraging GenAI to
Largest benefits of using
improve the quality of their employees' work,
while 27% are prioritizing its use to increase GenAI for employee productivity
employee productivity through capabilities in
content generation, summarization, and flexible
task completion. Notably, leaders aim for
37%
employees to enhance the quality and speed of
their core tasks rather than freeing up time for
more strategic work. [Exhibit 11] 27%
24%
Business leaders are looking
to improve the quality and
10%
speed of employees' core work
rather than to free up time for
more strategic work. Improve quality Increase output Reduce time Free up time for
[Exhibit 11]
of employee and speed of spent on more strategic
work employees administrative work
tasks
Harnessing the Power of (Gen)AI in Indonesian Financial Services 16
3
Rapid Lending Predictive AI and machine learning have been crucial
in analyzing borrower profiles, accelerating decision-
with Reduced making, and reducing non-performing loans. Among
other things, Predictive AI can deliver deterministic
outcomes with explainability for the underlying
Risk
rationale—a capability that GenAI has yet to fully
develop. However, GenAI excels at utilizing
unstructured data sources that contain quality data,
providing additional data points to enhance models.
Fintech Startups Traditional Financial Institutions
While GenAI shows promise, leaders are not With the abundance of unstructured data in large
yet prepared to replace Predictive AI for companies, it can be easy to overlook valuable data
mission-critical tasks. Instead, they are using sources that have not yet been integrated into models.
it to enhance supporting functions with GenAI has the ability to enhance Predictive AI credit
capabilities where it excels. models by incorporating supplementary information
from unstructured data sources, such as rejected
applications, to improve the models' understanding.
Fintech Case Study
Broom is a company that provides an
end-to-end financial solution for auto
dealer inventories in Indonesia. It
leverages AI to expedite credit
approvals.
Users may submit their applications before mid-day and receive
funds within a few hours. AI reduces underwriting time from an
hour to just 10 minutes. An employee then double-checks to
ensure 100% accuracy, serving as the final checkpoint and
demonstrating the effectiveness of a bionic process. Pandu Adi Laras
Co-founder & CEO, Broom
Finku, a personal finance
application that helps Indonesian
consumers manage their finances
and provides them loans, initially
underwrote loans using credit
bureau data.
Over time, the firm incorporated deep learning and machine
learning into its underwriting process as the basis of its credit
Reinaldo Tendean
lending model, significantly cutting down non-performing
Co-Founder, Finku
l |
292 | bcg | The-GenAI-Imperative-for-Telco-B2B-Sales-Teams.pdf | WHITE PAPER
The GenAI Imperative for
Telco B2B Sales Teams
February 2024
By Bryan Gauch, Alexa Vignone, Adolfo Magan, Jean-Marie Pierron, Johannes Goltsche, Basir Mustaghni,
Phillip Andersen, Alfonso Abella, and Ignacio Hafner
Boston Consulting Group partners with leaders Salesforce is the #1 AI CRM for Communications.
in business and society to tackle their most Salesforce for Communications enables service
important challenges and capture their greatest providers to find more prospects, close more
opportunities. BCG was the pioneer in business deals, deliver services more rapidly, and serve
strategy when it was founded in 1963. Today, customers more efficiently by connecting with
we work closely with clients to embrace a customers in a whole new way.
transformational approach aimed at benefiting all
Salesforce brings together all your data, from any
stakeholders—empowering organizations to grow,
source. Salesforce for Communications, powered
build sustainable competitive advantage, and
by Einstein 1, unites your marketing, sales,
drive positive societal impact.
commerce, delivery, service, and IT teams with a
Our diverse, global teams bring deep industry and single, shared view of customer information - at
functional expertise and a range of perspectives scale. With Communications Cloud, an asset-
that question the status quo and spark change. based, catalog-driven, modular solution built on
BCG delivers solutions through leading-edge industry standards, you can further spur growth
management consulting, technology and design, and reduce costs with industry specific functions
and corporate and digital ventures. We work in a like churn predictions, order management and
uniquely collaborative model across the firm and pricing and product designer. With artificial
throughout all levels of the client organization, intelligence embedded within our platform and
fueled by the goal of helping our clients thrive and apps, Salesforce helps augment everyone in your
enabling them to make the world a better place. company to work more productively and better
deliver the personalized experiences customers
love.
The GenAI Imperative for Telco B2B
Sales Teams
T
he future—in the form of both predictive and generative artificial intelligence (AI)—is
calling communications service providers. In every conversation with customers,
partners, and industry analysts, we have heard how excited telco B2B sales teams are
to embrace the new era of data, predictive AI, and now generative AI. They’re so eager that
sales ops teams are even reinventing their role as AI ops. It’s clear that AI could energize
service providers and help them deliver the true potential for their customers and experience
accelerated growth themselves. Harnessing the technology’s power, however, is a journey
that requires a progressive approach that generates value at each incremental step. Many
have started that journey but have yet to harness its full potential.
Transforming an organization to take advantage of AI/GenAI will leverage data, advanced
technology, infrastructure, and human interaction to create powerful end-to-end sales
processes. Harnessing this, we see a total value potential for telco operators of 40 percent to
70 percent uplift in EBITDA from driving both top-line growth and bottom-line efficiencies.
The telco industry, like utilities and other mature industries, faces a fundamental challenge:
growing their business while accelerating cost optimization and initiatives to find economies
of scale. This is particularly true in the B2B sector, where administrative tasks take up the
bulk of sales representatives’ time, the catalog of products is increasingly complex, and
legacy applications are clogging the system—frustrating reps and customers alike. In BCG’s
view, transforming the organization and its ways of working, supported strongly by developing
and deploying AI and GenAI, presents the best, perhaps the only, viable path for
organizations that want to break out of their stasis and generate positive momentum.
Even if organizations are beginning to employ AI/GenAI in some of their operations, its
deployment among sales professionals is often very much suboptimal (see Exhibit 1).
Exhibit 1 - How sales reps spend their time
9.2% Prioritizing leads/ opportunities 10.4% Meeting in person with customers
9.3% Researching prospects
9.4% Connecting virtually with customers
28%
9.0% Preperation and planning 8.7% Prospecting
Selling
72%
9.4% Generating quotes/
proposals and gaining approvals
Non-selling 8.8% Internal meetings and trainings
8.8% Manually entering customer
and sales information
8.3% Downtime
8.8% Administrative tasks
Source: State of Sales, 5th ed. report, Salesforce.com 2022
BOSTON CONSULTING GROUP + SALESFORCE 1
Sales reps spend 72 percent of their time on administrative and non-selling tasks, including
prioritizing leads, researching prospects, and planning. BCG and Salesforce have designed
and deployed several scenarios demonstrating how combining different types of AI can
transform B2B sales for telcos.
Leveraging these technologies and properly integrating them into sales processes will drive
productivity by automating or accelerating many steps. Predictive AI analyzes and evaluates
information, and GenAI synthesizes information and relays original output in natural
language—creating, for instance, an interactive engine to understand solution options for
specific customer needs (see Exhibit 2).
The unique characteristic of GenAI, which is based on powerful large language models
(LLM), is its ability to synthesize data that was input or extracted from unstructured data -i.e.
data that is typically categorized as qualitative data, does not have a predefined data model,
and cannot be processed and analyzed via conventional data tools and methods. GenAI can
then generate original data in different formats—text, images, sound, etc. GenAI platforms
trained in text, such as ChatGPT, Cohere, Anthropic, etc. can “understand” conversational
prompts and create original text, a complementary genre of output than, say, predictive
forecasting or recommendations based on crunching years of behavioral data (see Exhibit 3).
Our vision for AI-assisted sales
How well organizations integrate such AI models in the day-to-day of their employees and
their customer engagements is the key to success. Most clients we have worked with over-
index on the wrong elements, severely underestimating the change effort required. Based on
numerous deployments in different industries, we are convinced that three factors are
critical to success:
Exhibit 2 - PredAI and GenAI need to be combined to drive new frontiers
and accelerate existing applications
Not exhaustive
PredAI/ML GenAI
Use Predictive AI for decision-making Use Generative AI for content generation
Unstructured data
Dynamic pricing engines
ingesting & interpretation
Lead scoring and Synthesize findings in
prioritization large datasets
Demand
forecasting
Cross-sell / upsell Write and debug code
Protein
Churn prevention design & Creative content generation
selection
Other PredAI/ML applications Other GenAI applications
Use the combination of PredAI &
GenAI to maximize impact
generation
Source: BCG Analysis
2 THE GenAI IMPERATIVE FOR TELCO B2B SALES TEAMS
• 70 percent of the success revolves around processes and people: business process
reinvention, adoption at scale within the organization (which increases dramatically if the
“why” is clear behind the prediction), change (which requires the sponsorship and buy-in
of leaders), and rewiring the operating model. Equally important are well-defined business
objectives. Introducing new platforms without prioritizing the desired high-value, scalable
business outcomes, or simply using them for isolated applications, would vastly under-
exploit their potential for business transformation.
• A much smaller amount—20 percent—will be directly related to the technology stack and
foundations to make it work: model infrastructure, machine learning operations (MLOps),
data quality assurance, architecture design, app integration, and leveraging digital
platforms in the cloud, especially for their business support system (BSS) stack.
• Finally, 10 percent can be linked to the most disruptive technological advancements:
GenAI and AI/ML models (see Exhibit 4).
How to realize the value of AI
To make this tech-enabled organizational transformation a reality, organizations need to
actively consider their readiness and willingness to embark on this journey.
We understand that companies have different horizons in terms of their readiness to deploy
AI tools. We see three distinct horizons to choose from, advancing in maturity while implying
a greater need for transformational change and shift in the go-to-market strategy and
customer engagement.
1. Task automation and augmentation. Easier to realize, enhancing the business as usual
with tools, often out of the box from vendors, that speed and improve the process, such as
automating call summaries, generating targeted customer insights and emails, updating
CRM records, etc. Leveraging just these capabilities would allow sales teams to gain
substantial productivity already, addressing mostly their non-selling time.
Exhibit 3 - GenAI will not replace AI, but rather seamlessly incorporate it
to improve enterprise capabilities across the value chain…
Generative AI is complementary to AI & ML offerings Implications
Generative AI
• Step-change in ability of models to summarize, Gen AI extends AI capabilities…
categorize and generate language
• …by simplifying user interfaces, embedding in
• Generation of novel examples by learning workflows and enabling efficiency and
patterns in the data it is trained on and ability effectiveness across industries
to work with other media than just text/language
• Creation of true conversational user interfaces,
giving rise to new class of applications
AI AI use cases will persist….
• ...such as attribution modelling, budget
• Human-led programming sometimes constrained by allocation, personalization, and forecasting.
predefined rules
• Performance restricted to tasks within the programming
scope
ML
• Algorithms designed to analyze vast amounts of data and infer
ML allows AI to learn and improve accuracy…
correlations and causations
• …by allowing AI to find and learn patterns in data
• Provides the fuel to enable AI to autonomously improve outcomes
without being explicitly programmed
• …by helping AI make informed decisions based on
data with high degrees of accuracy
Source: Forbes; BCG analysis
BOSTON CONSULTING GROUP + SALESFORCE 3
2. Reimagined individual workflows. Reshape end-to-end solutions for sales agents.
Typical outcomes include lead outreach and qualification, quote updates, approval
processes, or the automated creation of proposals, hence unlocking revenue upsides and
competitive advantages from time to market.
3. Transformational change. This represents a fundamental shift in sales motions and
how the telco interacts with its customers. It requires the organization to adopt new
ways of working, undertake complete end-to-end redesign of processes crossing different
departments, and orchestrating across solutions that span many systems in their
architectural landscape, leveraging both structured and unstructured data (see Exhibit 5).
The outcome will enable processes to be executed at hyper speed and will create a new type
of empowered sales agent. An end-state vision is a no-touch sales process with minimal
human oversight, digitized end-to-end and driving quality engagements across the funnel:
the GenAI seller.
Managing the change effort across all of these horizons unlocks the true value of AI/GenAI.
To guide this effort, the following key principles have helped organizations embark on this
journey and make it a success:
• Set the top-down vision and ambition
• Simplification. The goal is to create simple, consistent, and seamless experiences for the
clients, the employees, and the partner ecosystem
• Choose a “lighthouse” use case. Stronger focus on value creation from day one
• Drive rapid organizational change while nurturing a sense of opportunity
• Embed analytics into the operating model and incentivize adoption
• Human-Centered AI (creating AI systems that amplify and augment rather than displace
human abilities)
Exhibit 4 - We can help you to deliver…
10% Disruptive 20% Technology 70% Business
Models Stack Transformation
Define the most strategic models Collaborate to define necessary Deliver lasting capabilities and
based on what is already available tech capabilities, developing and assets, reimagining business
and what we can build together deploying on the existing processes while ensuring fully
infrastructure enabaled teams
Guiding • What platform investments have • What capabilities does the tech • What is the vision for an AI-powered
questions been made? stack already have? sales organization?
• What models are already • What additional capabilities will • What traditional AI and Gen AI use
available? it need to support each model? cases support the desired business
• What combination of traditional • How can we equalize the risks of outcomes?
AI and Generative AI models is existing applications? • How do we adapt our ways of
best suited to the strategic goals? working to make change sustainable?
• How do we bring our teams along the
journey?
• How should we approach training to
ensure independence?
Source: BCG Analysis
4 THE GenAI IMPERATIVE FOR TELCO B2B SALES TEAMS
Salesforce and BCG customer stories
BCG and Salesforce have launched numerous AI/GenAI initiatives and projects for customers
in every industry and in all domains, with a goal to experiment, bring value, and augment
sales and customer operations. We are starting to collect great stories and lessons learned,
as illustrated in these four B2B comms, media, and high-tech examples.
Salesforce reference clients AI transformation project Metrics
Tier-1 Telco operator in EMEA; Einstein Copilot in local language of FAQ for • Deployed in few weeks
comms industry, B2C & B2B employees, including 1,800 articles • Time-saving
• Value enhancement
• Quality enhancement
Business information services Main objective: to mine sales interactions to Improved monetization of
leader; media industry, B2Ba increase sales efficiency and revenue with Einstein, existing customer base
including the following business capabilities: using predictive cross-sell,
up-sell models, pricing
• Summarize insights from customer calls to train
signals, and next best
reps and share feedback with the product team
action
• Mine all sales rep interactions with the
customer & summarize into key insights
• Build a recommendation engine for the print
business to identify customers ready for
upgrades
World leader in artificial This company relies on MuleSoft and Salesforce to • 40% reuse rate
intelligence computing; high- combine the power of APIs and AI to drive employee • Developer time + asset
tech industry, B2B productivity. It can connect back-office systems and reuse, multimillions
AI to build an intuitive chatbot to allow employees savings
to self-serve customer information for faster support.
By leveraging MuleSoft for AI-related projects like
the chatbot, employees can focus their energy on
tasks that require more hands-on attention.
BCG reference clients AI transformation project Metrics
US-based provider for • 3-year bot program, targeting comprehensive • >6% of net recurring
collaboration and capability uplift across the organization; revenue uplift, across
communication tools; comms/ leveraging the best of BCG across while aligning all customer segments
hi-tech, B2B incentives and putting skin in the game from SMEs to large B2B
• Inserting data-centricity into every customer enterprises
interaction, enhancing the technology stack e2e • Uplift x-sell and upsell by
while upskilling >500 sales agents +95%
• Churn -40%
• Price realization by +10%
Integrated US-based provider Sales acceleration through operating model • 2x lead conversion rates
for telecommunication transformation and AI-driven models’ deployment • Reduced Priority 1 stalled
services to drive execution velocity and sales engagement deals by 70%
in high-value deals
Integrated European-based Pipeline push focused on cross-selling enabled >4% of revenue uplift
provider for through win-rooms and AI-driven models’ to drive
telecommunication services opportunity identification and prioritization
BOSTON CONSULTING GROUP + SALESFORCE 5
“
“We are seeing different levels of AI
maturity in telco and recognize the
importance of trust & security and real
ROI in this cost conscious market.
Our customers are excited about the
tangible business results they are
seeing with our unified & composable
architecture that gets enterprise AI
solutions into the hands of the people
who need it, right in the flow of work.”
Alexa Vignone,
Executive Vice President, Salesforce
BCG’s sales transformation client story
The BCG client story began when we were approached by a B2B SaaS provider with annual
revenue between $1-5B. The company’s desire was to become the market leader in their
industry. It was suffering high monthly loses from churn and down-selling, so one goal was
clearly to do better at retaining customers. We undertook an initial diagnostic and noted that
the approach had room for improvement. The sales agents had only adopted existing tools in
a limited way and they lacked a clear understanding of how these could deliver value for
them.
The initial focus was on building the required capabilities—enhancing the platform and
building AI into it, while accelerating the launch of initiatives. To implement, BCG integrated
selected assets with the Salesforce.com platform’s capabilities in a modular approach.
Specifically, the BCG team:
• Developed the base data and infrastructure layer to support personalized account
management and unlocked access to additional data sources and facilitated ingestion.
• Managed campaign and experimentation enabled by a campaign manager-optimizer—
responsible for AI-based action codification, audience selection, and monitoring and
measurement. In addition, this campaign manager allowed us to launch targeted
experiments with new actions and enable an efficient test-learn cycle for the AI models.
• Gathered account intelligence, by building more than 15 different use cases on top of this
enhanced platform, leveraging AI from cross-sell to churn prevention.
• Developed more than 10 AI targeting models whose adoption was facilitated by translating
the models into natural language to build trust and improve decision-making.
Exhibit 5 - We help you to create value in the short- and long-term by iden-
tifying quick wins and by creating a roadmap to re-define workflows and
business models
Source: Forbes; BCG analysis
BOSTON CONSULTING GROUP + SALESFORCE 7
ytivitcudorP
Augment & automate tasks Reimagine individual workflows Drive transformation change
Do what is done today, but faster Change the way work gets done with with a fundamental shift in sales
and better new and redefined workflows motions & customer engagement
• Data curation to help sellers sell more • Advanced automation to take on significant • Execute back-office processes at hyperspeed
effectively (e.g., customer insights) work from sales (e.g., RFP response creation, (e.g., sales ops, deal desk)
• Task level automation to help sellers get quote updates, approval triggers) • Enable direct engagement with customers
more done faster (e.g., CRM updates, call • Precise prioritization (e.g., rethink demand and partners with GenAI 'sellers' and
summaries, tailored emails) gen with GenAI powered lead qualification) QA teams (e.g., RFP response teams
• Advanced automation (e.g., quote updates, to autonomously bid on work)
trigger approvals)
Deploy out of the Configure, build and integrate Invent new, cross-
box capabilities solutions within platform architecture
solutions
Time
• Activated channels by integrating target audiences and activating them in online channels
with Salesforce Sales Cloud for Contact Centers, further enhanced by customization with
BCG’s Agent desktop solution.
BGC teams provided support in all domains, combining classic strategic capabilities and
tech, including consultants, data scientists, engineers, and developers, among others.
From the beginning, we set a clear ambition to enable sustainable value delivery.
Consequently, towards the end of our program, we shifted the joint efforts to ensure
capabilities and ownership could be transferred effectively to the client’s teams across its AI,
tech, campaign management, and execution functions. Overall, the transformation delivered
significant impact, uplifting revenue by 5-8% annually and the effective transfer allowed the
client teams to maintain the performance and continue to iteratively enhance its
capabilities.
Talking about the tech stack
The common thread across all discussions we have with clients, partners, and industry
analysts about the new era of data and AI is concern about tech stack consolidation—not
just about trimming costs, but about accelerating productivity and unleashing growth.
As the case studies above show, the list of ingredients necessary for successful
transformation starts with data. We know that AI is only as knowledgeable as the data it’s
grounded on. Many customers tell us that to fully leverage AI, they need and want a single
platform. Point solutions atop the CRM create siloed data pockets that increase risk,
duplicate capabilities, reduce seller productivity, and increase costs.
AI is the next ingredient. A solid data foundation built in one CRM ensures that predictive
and generative AI bring real productivity gains. AI can automate emails, take actions, and
create account summaries based on CRM context—or tell your sales agents which products
are ripe for cross-selling opportunities. The possibilities are just beginning to be understood.
The number one goal is securing the AI architecture. That requires a trust layer, natively built
into the platform, with strong components to support data residency and compliance. As an
example, the Salesforce Einstein Trust Layer is equipped with security guardrails (see Exhibit 6).
Exhibit 6 - Salesforce’s Trust layer incorporates guardrails
Models
Customer, company, and outcome data
Promt Secure Data Dynamic Data Prompt Hosted Models
Retrieval Grounding Masking Defense
in Salesforce
Zero Trust Boundary
Retention
Audit Trail Data Toxicity Response Bring your own
Demasking Detection
CRM Einstein models, your own
Infrastructure
apps Copilot
Secure Gateway
Einstein Trust Layer External models
with Shared
Trust Boundary
Source: Einstein Trust Layer, Salesforce.com 2023
8 THE GenAI IMPERATIVE FOR TELCO B2B SALES TEAMS
As Exhibit 6 illustrates, when you generate a prompt that is built in Prompt Builder, the prompt is
sent to the Einstein Trust Layer, which masks any sensitive data before sending it to the LLM.
When prompts are sent to external models through the shared trust boundary, your data is
encrypted to ensure its security in transit. Additionally, any sensitive information within the
prompts is masked.
Another great concern is risk mitigation, both tactical and strategic. In a recent survey,1 73
percent of employees believe GenAI introduces new security risks, underscoring the need for
organizations to leverage GenAI technologies built with trust first. In this regard, the zero
retention of the Salesforce Trust Layer ensures that no information is stored or remembered by
large language models, prioritizing user privacy and data security.
There are indeed many types of risks related to AI, including trust, data property, data quality,
biases, hallucinations, costs of AI, value for money, human-centric vs. Full-bot approaches, business
deployment, and so forth. These risks are recognized by dynamic, new regulations and frameworks
that companies will have to carefully investigate. Our vision at BCG and Salesforce is to join forces
to propose a strategic enterprise approach for AI implementation, taking into consideration all
these dimensions, and not simply proposing to launch Proof of Concepts or a pilot.
To set the stage, Salesforce and BCG have developed a framework of AI-driven Sales-related use
cases matching the needs of companies in the Telco sector as illustrated in Exhibit 7.
We observed from most Salesforce customers that use cases relied on a blend of predictive and
generative AI, as well as analytics and automation.
Customer use cases also relied on knowledge extraction across a mix of unstructured and
structured data from multiple applications. This could include text documents, chats, audio, or
video. The importance of unstructured data should not be underestimated.
These varied data sources and data types come with different data latency requirements that must
be considered and coordinated for different use cases (e.g. real-time data streams or near real-time
data latency requirements in addition to batch data from different systems and schedules).
1. FY23 Salesforce Customer Success Metrics
Exhibit 7 - Trusted AI, Built Into the Flow of Work
Accelerate pipeline + Supercharge productivity + Unlock revenue
SALES MARKETING SERVICE
Close better Create more Elevate Customer
deals faster resonant content Engagement
Product Finder / Proactive retention
Sales Copilot Campaign Copilot Bahavior scoring Service Copilot
FAQ Search Engine campaign
Lead / Opportunity "Where is my order" Order Management
Call Summaries Content creation Engagement scoring
scoring Assistant Insights
Personalized Attrition Reduction Customer & Network
Activity 360 Content tagging Engagement frequency
conversations Assistant Serviceability Insights
Account /
Next Best Products Segment creation Spend Time optimization Renewal Assistant Next Best Actions
Opportunity overview
CPQ Assistant Pipeline inspection Smart promotion creation Smart return analysis Conversations Catch Up Next Best Products
Proposal / Contract Product Product Finder /
Forecasting Simulation Demo creation
Generation recommendations FAQ Search Engine
Source: Salesforce Industry Advisory 2024
BOSTON CONSULTING GROUP + SALESFORCE 9
“
“The building narrative we are
hearing across our Telco clients
is clear – GenAI’s imperative for
sales is to create smarter and
more impactful customer
interactions.”
Bryan Gauch,
Managing Director and Partner, Global Salesforce Offering,
Boston Consulting Group
The majority of these use cases and features are already available. They can be mapped to
service provider needs and help create a value-based transformation journey.
The business outcomes are already impressive. While security risks are a continuing concern
for employees we have surveyed, 68 percent say that GenAI will help them better serve
customers and save them an estimated five hours on average each week. Recent benchmarks
on ongoing Sales AI pilots have shown a potential of 29 percent increase in productivity.
How to move from strategy to execution at scale
BCG and Salesforce have teamed up and designed two engagement archetypes that can be
leveraged to kickstart the AI/GenAI journey for B2B sales organizations in the telco sector
and create an aligned starting point and vision:
1. Value Assurance. This is directed to customers that want to maximize the value from
their technology investment. To achieve this, Salesforce and BCG undertake a tailored
diagnostic to pinpoint concrete opportunities for value enhancement, delivered by a joint
team from BCG, Salesforce, and the client.
2. AI/GenAI Exploration. Also delivered jointly, in the format of a workshop, it will leverage
best-in-class GenAI strategy and capabilities to drive sustained success. The outcome will
be an industry point of view on the relevant use cases and will enable our clients to pilot
and scale their AI/GenAI practice, providing the basis for a smooth implementation and
transformation—faster and at a lower cost.
Following the initial vision and design, different approaches are available to realize the
potential, depending on the value at stake and the readiness—from further opportunity
exploration and detailing to full-scale transformation and support.
BCG and Salesforce teams have seen the potential of a predictive AI/GenAI future for service
providers. Now is the time to make that future happen and embark jointly on this exciting
journey.
Exhibit 8 - BCG x Salesforce AI/GenAI combination enables accelerated
value generation of GenAI while de-risking the implementation
• Value-driven strategies & transformations paired with • Integrated CRM platform with industry and functional
strong CRM transformation management capabilities specific clouds
• Deep industry & functional expertise, best-in-class Go-to- • Strong GenAI solutions and capabilities as well as an
Market frameworks and toolkits, optimized for CRM extensive marketplace with third party apps
• Sales, marketing and service persona-specific change • Product and platform expertise to support deployment
management approach of high-value CRM capabilities
• Trusted GenAI leader with purpose-built AI assets • Platform and product roadmap insights
(e.g., Deep.AI) integrated into Salesforce
• Best in class GenAI strategy & capabilities to drive sustained success
• Deliver platform and GenAI implementations faster and at lower cost
Source: BCG
BOSTON CONSULTING GROUP + SALESFORCE 11
12 THE GENAI IMPERATIVE FOR TELCO B2B SALES TEAMS
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.
© Boston Consulting Group 2024. All rights reserved.
2/24 |
293 | bcg | unlocking-the-genai-opportunity-in-latin-america.pdf | Unlocking the Gen AI
Opportunity in Latin America
Insights Generated from the 2024 Latam Tech Forum (LTF), an Invitation-Only Private Gathering of
the CEOs and Founders of the Largest and Leading Tech Companies from Across Latin America
July 2024
By Lucas Frenay, Julian Herman, David Marín, Federico Muxi and Joan Viñals
Since its inception in 2011, The Latin America Tech Forum
(LTF) has become a prestigious private gathering for
CEOs and founders of Latin America’s largest and leading
technology companies. The forum brings together founders
and c-suite executives from Latin America, alongside Latin
America heads of global technology companies, global
luminaries, and a small sub-set of investors and technology
advisors. The mission of LTF is to provide a platform for
leaders to collaborate, build trust and develop long standing
relationships across the technology ecosystem in Latin
America, which helps further economic development and
prosperity across the region.
Private, off-the-record, and by invitation-only, the forum
is held once annually and includes thought-provoking
interactive Executive Sessions, Fireside Chats with global
business leaders and renowned personalities, and other
activities relevant to this peer group across several days.
Attendance is limited to ensure the right environment for
developing new and meaningful connections.
Organized by Riverwood Capital, LTF is an industry initiative
supported by several leading institutions with the objective
of expanding the Latin America technology ecosystem.
TABLE OF CONTENT
Key Survey Insights .................................................................... 3
Introduction ............................................................................... 4
Chapter 1 | The Case for Gen AI at Scale .................................. 5
Chapter 2 | Use Case Deployment and Success Cases ............ 8
Chapter 3 | Key Challenges and How to Overcome Them ....... 11
Acknowledgements .................................................................... 14
About the Authors ...................................................................... 14
LTF 2024 17
Key Survey Insights
170
~100 ~70
Respondents
Tech company Company executives, tech investors,
Founders/CEOs and other industry leaders from
across Latin America
95 20 50
% % %
of top leaders say Gen AI consider they have a have passed the
will have a great impact well-defined Gen AI “Quick Win stage”and are
or be transformative for vision and ambition. now reshaping their core
their organizations. functions with the
use of Gen AI.
3 most relevant use cases stated: Main challenges perceived:
• Product enhancement (42%) • Talent (70%)
• New product development (29%) • Responsible Gen AI and data privacy (50%)
• Code generation assistance (26%) • Data and tech readiness (42%)
Note: The data set from Latam tech executives is displayed alongside responses from a BCG survey of global large traditional companies for context
in some areas of the report.
LTF 2024 3
Introduction
The Latam Tech Forum (LTF) 2024 yielded significant insights on
companies at the forefront of innovation in the region.
During the Forum, and to provide an in-depth reduction in customer inquiry costs, a 25% reduction in
understanding of how tech leaders perceive one of this marketing content creation time, and a 30% boost in
decade’s most pivotal technologies, BCG conducted a content production efficiency—all contributing to improved
survey on Gen AI with the participation of more than customer satisfaction and faster issue resolution. These
170 CEOs and C-level executives of the most prominent successes illustrate the substantial impacts achievable
tech companies in Latam. when Gen AI is deployed effectively.
BCG worked together with Riverwood Capital to plan and The report also serves as a strategic tool for decision-
execute Executive Sessions on Gen AI. This report makers to evaluate their Gen AI progress and understand
synthesizes key learnings from these discussions, together the evolving landscape, informing their strategic
with the insight of BCG’s experience across over 200 global engagements moving forward.
client cases.
It sheds light on the ongoing Gen AI adoption level within
the Latam tech sector, highlighting both achievements and
areas for improvement, and reveals how Gen AI has
demonstrated significant benefits, such as a tenfold
4 UNLOCKING THE GEN AI OPPORTUNITY FOR TECH PLAYERS IN LATIN AMERICA
Chapter 1 | The Case for Gen AI at Scale
GEN AI: TRANSFORMATIVE POTENTIAL FOR COMPANIES AND INDUSTRIES
One of the most spread beliefs gathered during the Forum is the recognition of Gen AI’s exponential growth and impact.
Over 90% of the participating C-level executives see Gen AI as a high-impact, transformative force within their sectors and
the competitive landscape (Exhibit 1). Gen AI is poised to revolutionize competitive dynamics and operational strategies in
the business world.
.
Selected examples of impact delivered:
30-40 95 10-15 20
% % x %
increase in reduction in fraud gains in marketing increase in
service desk prevention content generation developer
productivity manual tasks productivity productivity
Exhibit 1 | Almost every company perceives high value in Gen AI
What impact do you believe Gen AI will have in your business/industry?
% of total responses
85
63
32
13
5
2
Moderate impact High impact Completely transform
Global large traditional companies (BCG Build for the Future Survey)1 Latam tech companies (LTF Survey)
1. Traditional companies are based on a 2023 survey of 159 large companies (>USD 25 billion value) in North America, Europe and Asia-pacific –
covering various sectors.
Source: LTF 2024, BCG Survey; BCG Build for the Future C-level Gen AI survey
LTF 2024 5
TRADITIONAL COMPANIES’ GEN AI APPROACH TECH COMPANIES’ GEN AI APPROACH
Recognizing the impact and potential of Gen AI, traditional Top Latam tech companies are taking a different approach
companies are strategically harnessing its power. More to harnessing Gen AI. They have adopted Gen AI quickly
than 50% of these companies have adopted a well-defined and efforts are proliferating across functions and use cases,
strategic approach to Gen AI, emphasizing clear ambitions not only in the form of quick wins but also by reshaping
and prioritizing scalable use cases based on impact and their core businesses. Yet, just 21% of these companies
feasibility (Exhibit 2). declare having set an ambition (North Star), a Gen AI
strategy, and a clear implementation plan (Exhibit 2).
Moreover, more than 75% of companies scaling Gen AI are
actively engaged in shaping strategies, defining roadmaps, Only companies whose business models are
and discussing proactive investment plans of Gen AI. fundamentally based on Gen AI largely state to have a
well-defined ambition and a strategic approach to Gen AI.
Traditional companies are focusing on envisioning a
sustainable end-state operating model, despite the nascent Despite lesser planning, tech companies have adopted
stage of this technology and the lack of consensus on the Gen AI more rapidly than traditional businesses. At
optimal approach. Current models vary from decentralized LTF 2024, examples of quick wins through operational
structures, where AI engineers or specialists are distributed efficiencies, such as automating customer service with
across teams, to centralized or federated models. chatbots and optimizing data management, were
prevalent, as well as examples of product enhancements
Highlighting the importance of responsible AI together to reduce costs and drive top-line growth.
with creativity, an increasing number of organizations
see the federated model as the right landing point. This This is also reflected in the limited progress in defining and
model not only helps set and enforce clear policies but implementing a Gen AI operating model, with only 5% of
also ensures rigorous prioritization based on return and companies stating to have defined and implemented
the widespread dissemination of Gen AI resources delivery Gen AI teams with clear structure and KPIs.
throughout the company, promoting broad accessibility
and integration.
However, establishing specialized teams with clear roles
Start the journey collecting data and
and governance across the company remains a significant
challenge, with only 28% of traditional companies defining your strategy, and only then
successfully implementing such teams so far (Exhibit 3).
connect to specific capabilities and
use cases to continue evolving.
Leading Tech Executive
and LTF 2024 Participant
Exhibit 2 | Tech companies are far behind in defining strategy
Do you have an ambition (North Star), strategy and implementation plan underway?
% of total responses
61
51
36
21
17
13
Not defined Partially defined Well-defined
Latam tech companies (LTF Survey) Global large traditional companies (BCG Build for the Future Survey)1
1. Traditional companies are based on a 2023 survey of 159 large companies (>USD 25 billion value) in North America, Europe and Asia-pacific –
covering various sectors.
Source: LTF 2024, BCG Survey; BCG Build for the Future C-level Gen AI survey
6 UNLOCKING THE GEN AI OPPORTUNITY FOR TECH PLAYERS IN LATIN AMERICA
Exhibit 3 | Only 28% of traditional companies have AI dedicated teams
at scale
Have you defined and implemented Gen AI delivery teams/squads with clear structure and KPIs?1
% of total responses
43
33 33
32
28
19
6 5
No delivery teams Partially in some areas1 At scale in some areas1 At scale company wide
Latam tech companies (LTF Survey) Global large traditional companies (BCG Build for the Future Survey)2
1. These options were consolidated into one in the BFF survey (answered by Global large traditional companies): “Scale at pockets, traditional
governance.”
2. Traditional companies are based on a 2023 survey of 159 large companies (>USD 25 billion value) in North America, Europe and Asia-pacific –
covering various sectors.
Source: LTF 2024, BCG Survey; BCG Build for the Future C-level Gen AI survey
Key Takeaways
In the early stages of adoption, business leaders widely recognize the potential of Gen AI. Traditional companies are
proactively setting ambitions and developing new operating models for the Gen AI era.
Meanwhile, the Latam tech sector, eager to rapidly advance in experimentation across functions and use case
escalation, shows room for more defined strategy and governance.
Fast and nimble deployment and innovation is inherent to tech companies. Nevertheless, a scattergun approach could
lead to operational inefficiencies, and without a clear strategy across the organization, businesses face a significant
challenge in identifying and measuring potential value. This in turn could impact the correct allocation of efforts and
resources, delaying investment decision and stalling execution capabilities altogether.
In our opinion, developing a structured Gen AI strategy and establishing clear ambitions is essential, a stance
supported by numerous tech leaders at LTF 2024.
LTF 2024 7
Chapter 2 | Use Case Deployment and Success Cases
Latam’s dynamic tech ecosystem is on the cusp of a 3. Inventing new Gen AI-driven business models:
major shift, driven by the integration of Gen AI into the New value proposition and revenue streams
business landscape.
To navigate this transformative era, companies can deploy The boldest players in the Latam tech scene are exploring
Gen AI to capture quick wins, reshape critical functions new horizons by leveraging Gen AI to create innovative
through Gen AI or invent new Gen AI driven business business models and long-term competitive advantages.
models (Exhibit 4). These examples were rarer in the discussions at LTF 2024,
since only 24% of companies surveyed were inventing new
1. Deploying Gen AI in everyday tasks: Broad Gen AI-driven business models. However, the ones that do,
enterprise-wide productivity enhancement have the potential to disrupt their industries and establish
and quick wins entirely new market spaces.
A common starting point is to capture quick wins. Several companies have not only embraced Gen AI but
According to the survey, 55% of the companies and CEOs also made significant strides in their application. Some
surveyed reported that they are already doing so. success cases are illustrated on the next two pages.
2. Reshaping critical functions: Radical productivity,
speed and quality improvements
The natural next stage is to integrate Gen AI deeply within
core functions.
In comparison with most traditional companies, that
are still at the Gen AI proof-of-concept stage, over half
of tech businesses in Latam are reshaping their core
functions like marketing, sales, and HR, with Gen AI to
increase both efficiency and effectiveness (with impacts
in the +50% zone).
Currently, tech companies’ main area of focus in Latam
is improving the product portfolio, 42% mostly enhancing
core products, while new products or services positions
as the second use case with ~30% of respondents
(Exhibit 5).
Exhibit 4 | Focus and Gen AI adoption in Tech companies in Latam
Q: Which of following statements best describes the focus and degree of Gen AI adoption in
your company?
55%
50%
24%
4%
No action on Deploy: Capturing Reshape: Embedding Invent: Innovating
Gen AI yet quick wins Gen AI in core functions new business models
1. Requested answer: select all that apply
Source: LTF 2024, Participants Survey
8 UNLOCKING THE GEN AI OPPORTUNITY FOR TECH PLAYERS IN LATIN AMERICA
Exhibit 5 | Most relevant use cases in tech companies in Latam
Q: What are the 3-5 most relevant use cases you are currently implementing/discussing
in your company?
Standard use cases
Product Core product enhacement 42%
New products or services 29%
Code generation 26%
Tech Code review/auditing 14%
Code documentation 9%
Chatbot for sales/support 15%
Copilot for customer support 8%
Sales and CX Personalized add creation/marketing 7%
Copilot for salesforce 3%
Claims/complaint evaluation 2%
Finance 9%
Procurement 9%
Support functions
HR 9%
Legal 7%
Knowledge management 9%
General productivity Document/meeting summaries 9%
Copilot for office tools (ppt, e-mail, ...) 5%
1. Requested answer: select 3-5 options
Source: LTF 2024, Participants Survey
INVGATE
InvGate is an IT Management software company with a in proprietary models and talent. They have implemented
focus on AI-enabled Enterprise Service Management and a strategic investment plan focusing on an internal AI
large-scale IT device inventory and configuration Service alongside internal models and engineering
management. It enables drastically lowered time-to-value capabilities, asserting that flexibility will be crucial in
in all of the categories it focuses on, by leveraging no-code navigating this dynamic landscape.
implementations that are an order of magnitude shorter
than competing offerings. InvGate has 1000+ customers in SENSEDIA
over 50 countries, including NASA, Arcos Dorados, Telekom
Malasia, and Collins Aerospace among others. A leading Brazilian company specializing in API
management and integration strategy, enhancing digital
Key use cases implemented | InvGate has embedded connectivity and open technology ecosystems. The
GenAI capabilities across its solutions for Service company offers solutions for integrating diverse digital
Management and Asset Management. These include: channels and adopting more modern architectures, like
microservices, APIs, events and service mesh.
• Resolution Recommendation: This feature automatically
suggests a possible solution to a ticket based on not Key use cases implemented | Three innovative Gen AI use
only knowledge-base articles but also previously cases enhancing API design and client services:
resolved tickets.
• API Copilot: boosts API design productivity by gener-
• AI-Knowledge Article Generation: Uses resolved ticket ating new documentation and improving existing ones
information to create knowledge base articles that can be based on business context.
later referenced by agents or Invgate’s AI agent.
• API Simplification: Leverages Gen Al to identify duplicat-
• Ticket Summarization: This feature reduces the time ed APIs across an organization and propose simplifica-
needed to manage IT incidents, speeds up new team tion, enabling companies to reduce costs.
members onboarding, and enhances overall support
efficiency. • API Consumption: Supports API portfolio management
by assessing individual API performance metrics and
Impact | 30%-40% increase in productivity and a noticeable defining best prioritization model for a given objective.
increase in MTTR (mean time to resolution) due to its
GenAI capabilities. Impact | Although comprehensive metrics are not yet
developed, the API Copilot shows potential to enhance API
Key learnings | The company emphasizes being vendor- design productivity by up to fivefold (5x).
agnostic when it comes to GenAI capabilities and investing
LTF 2024 9
Key learnings | Talent acquisition was the biggest hurdle, RAPPI
so the company formed a small agile squad to reduce the
learning curve and accelerate time-to-market, quickly Rappi is a leading on-demand delivery superapp. It was
sharing insights to enhance overall capabilities. founded in 2015 in Bogotá, has operations in 9 countries in
Latin America and a network of more than 300 thousand
DLOCAL businesses across multiple segments (restaurants,
groceries shops, pharmacies, and more).
Founded in 2016, the company became Uruguay’s first
unicorn and went public in 2021 with a $9.5 billion Key use cases implemented | Rappi is fully embracing
valuation. The company is known for its API-based Gen Al, being able to deliver several use cases at scale
payment solutions, serving over 330 merchants in across multiple business functions and verticals:
29 emerging markets, and supports various local
payment methods. • Customer Support Sidekick: Empowers customer
support by providing best recommended response
Key use cases implemented | Dlocal has advanced its core and retrieving relevant policy documents, and boosts
functions with productive Gen AI use cases at scale: productivity by summarizing conversations.
• Smart Router: an AI-based routing solution that selects • Account Managers Sidekick: Enables sales teams to
payment processors based on variables like payment add more value to their clients preparing them with
method, industry, country, and merchant-specific factors prioritized insights and proposed action items with
to improve conversion rates and cost. higher sales impact.
• Fraud Prevention: Gen AI automates the examination of • Developer Copilot: Supports developers to increase
merchant websites for risk assessment and formulates coding and development productivity. Supports code
questions to help prevent fraud. documentation/writing.
• Support Cloud Engineer: The GenAI Copilot has • Merchandising Content Generation: Enables
significantly transformed CI/CD pipeline management. It merchandising teams to significantly improve digital
detects errors and automatically suggests fixes, reducing storefront customization and creative content
human support from the Cloud Platform team by 90%. development. Same team is able to create 10X more
This allows the team to focus on platform development, seasonal merchandising events.
while engineering teams receive immediate answers,
accelerating their delivery speed and enhancing overall • Product search/personalization improvement: Using
satisfaction. Future plans include enabling the Copilot the same underlying Transformers technology as used
to directly fix issues in the code, further increasing the in GPT models, Rappi’s in-house recommendation
speed of error resolution. systems predict the likelihood of click or conversion for
a customer at a given time, location, product and search
Additionally, Dlocal has launched Smart Request to terms, to create a better search/browse experience.
optimize payment option selection and a company-wide
chatbot powered by OpenAI, to boost productivity by • Back Office Automation: Multiple AI-powered
assisting with various inquiries. applications, including purchase order and invoice data
extraction to drastically reduce cost, error rate and
Impact | The copilot has already reduced the need for accelerate identification of discrepancies.
human intervention in support tasks by 90%, and the fraud
prevention tool has cut manual tasks by 95%. However, Impact | Significant impact across use cases. A few
some of these metrics are still preliminary, and further examples are 10-15x productivity gains in site
operation is needed to develop a comprehensive set of merchandising quantity of events and 20% increase in
KPIs for Gen AI use cases. developer productivity.
Key learnings | Dlocal has not established a defined Gen Key Learnings | Don’t be fixated on developing the most
AI North Star per se but acknowledges the need for a sophisticated or complex AI technologies in-house.
prioritization methodology. A dedicated team reviews Prioritize tech solutions with the most direct path for
inputs from different BUs to identify and rank use cases impactful adoption and likely this involves leveraging
based on impact, acting as gatekeepers to filter, prioritize commercial/industry tech offerings. Develop a central
and allocate resources. architecture that enables whole organization to get easy
access to Gen Al and invest in upskilling - enabling teams
Dlocal recognizes the importance of adopting a new to produce solutions in a decentralized way and focusing
mindset to fully leverage Gen AI potential, and despite on delivering client value. Deliver solutions that will
progress, also acknowledges certain resistance to change. empower humans - not displace them - and pass the value
generated to customers.
10 UNLOCKING THE GEN AI OPPORTUNITY FOR TECH PLAYERS IN LATIN AMERICA
Chapter 3 | Key Challenges and How to Overcome Them
Adopting Gen AI within the dynamic Latam tech ecosystem Talent scarcity: Digital transformation has significantly
presents a unique set of challenges. This chapter increased the need for Gen AI expertise. Traditionally, tech
synthesizes the common roadblocks as expressed by companies would tackle challenges by hiring the most
industry leaders during LTF 2024. experienced talent. However, with current scarcity of
Gen AI professionals, simply hiring the top candidate no
MAIN CHALLENGES TO OVERCOME longer suffices, and leaders know it. They stated that
acquiring new skilled professionals and training existing
The three primary challenges perceived by Latam tech employees in Gen AI is a considerable task.
companies are related to Talent, including the difficulty
in recruiting new skilled personnel, training existing Talent training: Businesses now heavily rely on AI’s
employees, and ensuring leadership readiness; scalability, underscoring the urgent need to train
Responsible AI, involving data privacy, transparency, employees for AI-driven processes. Traditional problem-
policies, and regulation; and Data mastery and tech solving methods are inadequate in the face of growing
readiness, including tech integration, data governance, and complexity and the unique challenges brought by Gen AI.
model training (Exhibit 6). To effectively equip teams, an innovative training approach
incorporating external expertise and new educational
In addition, during the LTF sessions, executives highlighted methods is essential.
that infrastructure and cost management pose significant
barriers to the deployment and scaling of Gen AI. Companies are contemplating how to best allocate their
resources for talent development in AI, weighing whether
to develop and train in-house AI expertise or wait for AI
1. Talent transformation: Elevating AI competence platforms to evolve and become more user-friendly.
and empowerment
AI adoption: Leaders are aware of the talent and cultural
gap, particularly in specialized AI knowledge and usage.
There are four specific They expressed the need to set a culture of AI use and
Hiring the most topics related to Talent and understanding across all teams, not just within
experienced person to AI that need to be engineering or product development.
solve the problem will addressed: Scarcity, training,
not be the solution in adoption and leadership.
this occasion.
Leading Tech Executive
and LTF 2024 Participant
Exhibit 6 | Challenges around talent and data are the major concerns in the
implementation of AI
Which are the following dimensions that will pose the biggest challenge moving forward?
Talent/Resources 71%
Responsible AI/Data privacy 50%
Data mastery/Tech readiness 42%
Main challenges
Governance/Working model 22%
Regulation 22%
Other1 5%
0% 20% 40% 60% 80%
% of total responses: >70% 40%-30% 25%-10% >10%
1. Including: Investment, competition, mindset change, output accuracy
Source: LTF 2024, BCG Survey
LTF 2024 11
Leadership readiness: Leadership these bills will be as stringent in compliance as the EU
is crucial in the AI transformation legislation is.
journey. The challenge is to equip
leaders with an AI-ready mindset Start small, Companies need to carefully adhere to intellectual
to effectively identify and leverage conquer early wins property and copyright norms within the evolving AI
Gen AI capabilities. Focusing on to incorporate legislation to stay competitive without trespassing
leadership readiness is vital to capabilities and legal boundaries.
align the organization with funding for larger
AI capabilities and technologies. deployments! Data mastery and tech readiness
Leading Tech Executive
and LTF 2024 Participant Data mastery: When it comes to data mastery, challenges
lay in three main pillars: data capabilities, data design and
2. Responsible AI: Forging the path to accountability data governance.
and integrity in technology
First, datasets for Gen AI (foundation) models are
Gen AI will only amplify existing and new risks associated becoming ‘multimodal’, so there is a rising need to
with AI, namely litigation around copyright infringement, incorporate a much broader range of data inputs.
data and financial loss, reputational damage and
regulatory compliance, as well as new risks to consider, like Second, design considerations must be observed for
accuracy, ownership and bias or harm of outputs produced. multimodal processing, ranging from data provenance,
metadata, data lineage, output data quality and
Data privacy: In an age where data equates to currency, regulatory compliance.
privacy concerns are of utmost importance. Companies
must navigate the complexities of protecting individual Third, the incorporation of large amounts of unstructured
privacy while leveraging data for AI innovation. This data also introduces new risks (data usage outside of
balance is critical, especially for U.S.-based operations, given purpose, unauthorised data usage, output reliability,
where regulations are stringent, and the cost of non- computing cost when using unstructured data), which
compliance is high. require new approaches and governance to mitigate and
control data.
Ethical AI and transparency: Discussions from LTF 2024
reveal a trend towards establishing ethical AI frameworks Tech readiness: Tech challenges with regards to Gen AI is
and practices, addressing concerns such as models’ twofold, CIOs and CTOs will have to manage both “Gen AI
hallucination and bias, to ensure AI’s decisions are in Tech” (i.e. transform the IT organization) as well as
transparent and accountable. This commitment to ethical “Tech in Gen AI” (enable business transformation). Also,
AI extends to maintaining regulatory compliance, CIOs and CTOs will have a growing role as orchestrators to
particularly in data-sensitive areas. help businesses navigate the landscape, deliver value, and
ensure responsible use of AI while keeping the pace
The opacity of AI decision-making processes — the ‘black with Gen AI evolution.
box’ issue — requires a push for greater transparency and
understanding of AI’s internal workings. This transparency Companies, with the help of their tech leaders, will have to
is crucial for building trust among users and stakeholders. decide which model archetype to adopt. There are 4-four
main archetypes: Public, Managed Secured, Hybrid Private,
Human control: As AI technology advances, maintaining Fully Private. Each of the four archetypes has different data
human oversight is critical. Without it, AI models can and training characteristics and choosing the right model
produce harmful behaviors. Ensuring AI systems have deployment archetype will depend on business use cases.
robust human-in-the-loop mechanisms is essential to Hybrid Private is the most common and widely spread
prevent these issues and control risks associated to archetype, only to be questioned if cost of API is prohibitive
accuracy, ownership, and bias from the outputs produced or there is very high domain specific complexity.
using Gen AI models. This is even more important for
those companies in sectors with strict ethical standards Also, the platform and model partnership selection will
and regulations, making human supervision a moral and need to be carefully assessed. Partner preference,
regulatory imperative. geographical presence, hosting, portability, and security are
among the main selection criteria for platform provider
Regulatory compliance: AI regulatory compliance is selection. While performance, capabilities and complexity,
complex, requiring businesses to balance innovation with ability to fine-tune, and cost and compatibility with
legal constraints. EU is one of the few regions globally that platform are the key criteria to select the model.
has “passed” a legislation on Gen AI. Most of the Latam
countries have taken the “risk-based approach” set up in Finaly, when referring to the tech stack and architecture
the EU legislation, as the basis for their bills. If passed, new capabilities will be required, essentially in the AI Layer
12 UNLOCKING THE GEN AI OPPORTUNITY FOR TECH PLAYERS IN LATIN AMERICA
(to manage AI products and platforms), the Model Layer model and Gen AI data storage), refer to Exhibit 7 for
(to build, operate and maintain Gen AI models), the Agent more details.
Layer (to manage prompting and agents) and in the
Central Data Layer (additional storage on prompt, Gen AI
Exhibit 7 | Gen AI tech stack will require new technology capabilities
Tech stack Gen AI evolution
Modern data architecture blueprint Gen AI Simplified
Smart Business Layer Smart Business Layer
Impacted Cognitive Apps
Omnichannel App
Chat Image Video Music
builder
Cross-channel mechanisms and business components
AI Layer
AI products
Data Layer AI guardrails
New Content moderation Observability Operational
Repository & storage
Operational AI platforms AI services
data services Agent
Ingestion & distribution
Model
Data Layer
Core Transaction Layer Impacted
Data Repository Ingestion & Operational
products & storage distribution data services
ERP Other systems
Core Transaction Layer
Infrastructure/Cloud Infrastructure/Cloud
On-prem Cloud Hybrid On-prem Cloud Hybrid TPU/GPU
Key Takeaways
To overcome AI talent shortages, companies should benefiting the company overall. Identifying and
collaborate with specialized recruitment firms and categorizing the risks associated with AI systems through
create tailored AI training programs incorporating a well-defined risk taxonomy is crucial. Additionally,
external expertise. A new educational approach is companies need to create a clear governance structure
essential to upskill existing employees, while also with teams specialized in international and local AI laws,
implementing a comprehensive workstream to develop, dedicated to ensuring the continuous responsible use of
engage, anticipate, and attract skilled professionals, as AI, monitoring evolving regulations, and adapting their
well as fostering adoption and change management. policies accordingly (central small but rather senior
Additionally, investing in leadership programs focused teams). Finally, the key principles for Responsible AI,
on AI readiness and forming partnerships with peers including accountability for model outcomes,
and external entities to share insights can significantly transparency, fairness and equity promotion, safety and
enhance organizational capabilities and spur innovation. risk reduction, adverse effects avoidance and human-
Finally, they should consider allocating resources machine collaboration, must be embedded in existing
disproportionately towards the human aspects of the company policies centrally managed.
Gen AI transformation, including change management
and skill development to ensure adoption. Leaders To master Data, companies should in |
295 | bcg | unlocking-potential-strategies-driving-gccs-digital-ai-maturity.pdf | Unlocking Potential: Strategies Driving
GCC’s Digital & AI Maturity
DECEMBER 20, 2024
By Rami Mourtada, David Panhans, Lars Littig, and Hassen Benothman
READING TIME: 5 MIN
Emerging technologies are reshaping the world at an accelerating pace. As the GCC races ahead with
its ambitious economic development plans, the region has already well progressed on its technology
infrastructure and has provided many needed legislative, investment, and entrepreneurial
environments for digital-first leading organizations to emerge within the region.
To gauge this digital and AI readiness, BCG’s 2024 Build for the Future (BFF) study examined the
digital maturity of organizations in the GCC with a special focus on AI as a most transformative
© 2024 Boston Consulting Group 1
emerging technology. The study surveyed C-suite executives and senior leaders from 200+
organizations across eight sectors in Qatar, Saudi Arabia, and the United Arab Emirates.
Upon evaluating organizations across 53 core capabilities pertaining to digital maturity and AI
readiness, the BFF study methodology categorizes each surveyed organization into one of four
categories representing their stage of digital transformation, from least to most mature: Stagnating,
Emerging, Scaling, and Future-Built.
Core Capabilities: GCC Organizations to Catchup
with Global Digital and AI Maturity Levels
GCC organizations are presented with a unique opportunity and a matching challenge to build on
their capabilities to leap into global-level digital and AI maturity. In 2024, GCC organizations show
higher maturity around customer journey and digital operations capabilities, however, have yet to
fully possess many of the critical enabler capabilities that would allow them to fully deploy their
digital and (Gen)AI strategies, and will need a step-change, particularly in their data and technology
capabilities.
© 2024 Boston Consulting Group 2
Where fast-changing technology landscapes and rapid AI adoption are paramount to future success,
GCC organizations have a tangible opportunity to catch up with their global counterparts at the
overall digital maturity level. While 25% of GCC organizations fall into the top two scaling or future-
built maturity levels the global share is 31%. At a sector level, the Public Sector in the GCC exhibits
key areas at world-class digital maturity levels, while Financial Institutions and Tech companies
exhibited the highest digital and AI maturity scores across the GCC. Overall, however, digital and AI
maturity in the GCC in 2024 fell behind the global average.
Challenges & Opportunities: GCC Organizations on
the Road to AI Value Delivery
In 2024, 17% of organizations in the GCC scored into the top two AI maturity levels (AI-scaling & AI-
Future-Built). In this regard, the Financial Institutions sector had the highest share of top-level AI-
1
maturity organizations or “AI leaders” where 29% of financial institutions scored in the top two
levels in 2024, followed by Healthcare sector (23%) and the Public Sector (20%).
Additionally, our study found that the highest AI maturity organizations have three times the rate of
success extracting value from GenAI (more on this below). Yet, the GCC remains at the early stages
of (Gen)AI adoption, with only 9% of organizations surveyed at this level of value delivery. Overall half
of all organizations surveyed (53%) are either still experimenting with GenAI with no official policies
set in place, or not actively using it at all. The remaining (38%) recognize the value of adoption and
are planning to scale up with guardrails. Similar to the overall digital and AI maturity trend in the
GCC, sectors with highest share of organizations generating value with (Gen)AI are the Public Sector,
as well as the Tech and Telco sectors.
While every AI journey is tailored to each organization, we found common challenges in the region.
For instance, 6% of surveyed organizations have expressed not fully understanding GenAI, which lies
in the critical need for leadership initiatives to educate and upskill while setting an AI-and People,
Org, and Process-first strategy. For organizations further down the line of (Gen)AI adoption, several
challenges have been marked across BCG’s. “10-20-70” Algorithm- Technology- and People, Org, and
2
Processes framework.
In fact, the highest share of GCC organizations observed gaps in the people, processes, and
organizational dimension as the biggest barriers to AI maturity. This includes limited specialized
talent, a gap in overall AI literacy, and a lack of sufficient incentives for innovation and GenAI
adoption in working processes. Difficulty integrating AI within exiting IT systems and lack of access to
unified and high-quality data further hinders progress.
On the other hand, compared to stagnating and emerging organizations, AI leaders have been
successful in embedding AI for process-level productivity aimed at reshaping critical business and
© 2024 Boston Consulting Group 3
customer-facing functions as well as at integrating innovation in core corporate functions. To do so,
AI leaders in the GCC have focused on key enablers including increased investment and focused
budget allocation, as well as digital-first resource planning. High-maturity organizations allocated
2.4x more funding to AI initiatives as well as a 2.3x higher share of FTEs were dedicated to digital &
AI transformation, achieving a 1.7x higher share of (Gen)AI products scaled organization-wide and
reflecting a long-term commitment to embedding innovation.
Value Makers: Digital & AI-First Strategy for Future-
built Organizations
The 2024 BFF study highlights the need for most GCC organizations to progress beyond incremental
moves and embrace comprehensive digital and AI strategies to unlock transformative value across
sectors. While GCC organizations have made impressive progress in digital and AI capabilities, there
remain opportunities to further enhance their maturity levels in critical areas and continue building
on their strengths to lead globally in digital transformation and AI adoption.
To bridge the global maturity gap and accelerate impact, GCC organizations must embrace a bold,
digital & AI-first strategy, across 5 key recommendations:
1. Re-align organizational strategy with a digital-first vision to overcome structural barriers
like operational agility and talent development.
2. Set a bold strategic ambition for AI adoption focused on clear value pathways and
guardrails for responsible AI adoption.
3. Boost viable people and org capabilities and underlying technology platforms to support
ambition and invest in parallel to scale up.
4. Maintain a pipeline of continuing innovation pilots to rapidly and effectively adapt to
changing landscape of emerging technologies.
5. Prioritize high-profile cross-cutting lighthouse initiatives with high ROI to fund the journey
and build momentum for transformational org-wide change.
The GCC stands at a crossroads where technological advancements intersect with the region's
aspirations to lead in digital and AI innovation. By addressing these priorities, GCC organizations
can unlock transformative potential, enabling them to capitalize on emerging opportunities, catch-up
to global peers, and earn their position as future-ready pioneers in an increasingly digital world.
© 2024 Boston Consulting Group 4
Authors
Rami Mourtada
PARTNER & DIRECTOR, DIGITAL TRANSFORMATION
Dubai
David Panhans
MANAGING DIRECTOR & SENIOR PARTNER
Dubai
Lars Littig
MANAGING DIRECTOR & PARTNER
Dubai
Hassen Benothman
MANAGING DIRECTOR, BCG PLATINION
Dubai
1 Top-level AI-Maturity organizations are organizations who scored over 50
on 30 digital capabilities examined.
2 The BCG 10-20-70 digital and AI transformation model is “Focus 10% of
your efforts on algorithms, 20% on the underlying technology and data,
and 70% on people, org, and processes”
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
© 2024 Boston Consulting Group 5
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.
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).
© 2024 Boston Consulting Group 6 |
296 | gartner | gpc-genai-ocsummaryv2-content.pdf | Generative
AI Surveys
Overview Barriers, Benefits, Use Cases Open-ended Insights Peer Data & Insights Additional Insights
RESTRICTED DISTRIBUTION
1 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved.
BY THE 7 PARTICIPANTS BREAKDOWN
NUMBERS
BUSINESS FUNCTION
SPECIFIC SURVEYS
833
In a few weeks' time, we
completed 7 business
LEADERS
function specific surveys
3
with a total response of 833
North America
leaders, across 3
51%
CONTINENTS
continents, representing 21
industries about their 21 APAC
29%
impressions of generative AI
programs, and the EMEA
INDUSTRIES
19%
associated opportunities,
risks, and use cases. JOB LEVEL COMPANY SIZE INDUSTRY
30% 10,0001+ <1,001
While it remains early days employees employees Software 14%
25%
for many respondents, their 24%
21% Professional
feedback gives significant 12%
Services
insight into the potential
Finance, Banking
12%
future attitudes of about and & Insurance
applications of these tools.
5,001 – 10,000 1,001 – 5,000
C-suite VP Director Manager employees employees
RESTRICTED DISTRIBUTION
2 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved.
Generative
AI Surveys
Overview Barriers, Benefits, Use Cases Open-ended Insights Peer Data & Insights Additional Insights
RESTRICTED DISTRIBUTION
3 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved.
Top of mind risks for IT & INFOSEC LEADERS
Barriers to
Potential for vulnerabilities or
58% 57% Potential for generating
leaked secrets in AI-generated
incorrect or biased outputs
code
Generative AI
Adoption
Biggest challenges cited by SOFTWARE ENGINEERING LEADERS using Generative AI
66%
Undesirable results
43%
Lack of corporate governance policies
38%
Pushback from leadership
Reasons shared by SOFTWARE ENGINEERING respondents whose departments have not adopted Generative AI
71%
76%
Security Inaccurate or biased results
RESTRICTED DISTRIBUTION
4 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved.
Challenges of D&A LEADERS who have AI-generated synthetic data
Barriers to
51%
Generative AI
46% Not having
enough real-world
Inherited bias in source data
41%
synthetic data
Adoption
Inaccuracy caused
by statistical noise
34%
Inaccuracy
caused by
statistical noise
The top selected adoption barriers among SALES LEADERS
51%
49% 38%
Lack of widespread adoption Integrations with existing technology Availability and quality of data
RESTRICTED DISTRIBUTION
5 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved.
Generative AI adoption barriers for MARKETING LEADERS
Barriers to
55%
Skills Gaps
Generative AI
42%
Integrations with existing technology
Adoption
38%
Unforeseen security threats
Top adoption barriers submitted by SUPPLY CHAIN LEADERS
58% Integrations with existing 57%
Unforeseen security threats
technology – or a lack thereof
RESTRICTED DISTRIBUTION
6 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved.
IT & INFOSEC LEADERS expect the following for Generative AI
Identifying
66%
Tech leaders predict positive Positive impact on the bottom-line
financial performance
bottom-line impacts from
Generative AI’s large language models
(LLMs) and generative AI
apps; slightly fewer expect
59% Improve top-line financial
Benefits top-line impacts.
performance
D&A LEADERS realized benefits of synthetic data
60%
Improved model
accuracy
56%
45%
Mitigated data
Improved model
privacy concerns
efficiency
RESTRICTED DISTRIBUTION
7 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved.
SOFTWARE ENGINEERING LEADERS believe that
Identifying
Generative AI will have a
70% 23% It will have a very positive
Generative AI’s
somewhat positive impact on
impact
software engineering
Benefits
SUPPLY CHAIN LEADERS identify as expected benefits
49%
Improved agility
48%
Improved productivity
48%
Improved cybersecurity
RESTRICTED DISTRIBUTION
8 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved.
SALES LEADERS believe that Generative AI would
Identifying
Allow them to completely 37%
Generative AI’s replace a person
Benefits
MARKETING LEADERS top selected benefits
57%
43% 38%
Improved speed to market Improved productivity Improved ROI
RESTRICTED DISTRIBUTION
9 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved.
IT & INFOSEC LEADERS cite the following top use cases
Pinpointing
Use Cases
53%
30%
Marketing and advertising Research and
development
56%
Data analysis and prediction
32%
34% Fraud detection &
cybersecurity
Operations and
logistics
RESTRICTED DISTRIBUTION
10 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved.
SOFTWARE ENGINEERING LEADERS are excited about using Generative AI in
Pinpointing
Use Cases
61%
55% 48%
Code generation AI-assisted pair programming Technical document generation
SUPPLY CHAIN LEADERS are planning to put Generative AI to use for
48%
Internal knowledge base enhancement
44%
Problem resolution management
42%
Generating interactive predictive models
RESTRICTED DISTRIBUTION
11 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved.
MARKETING LEADERS selected
Pinpointing
Use Cases
62%
Content production
52%
38%
Generating ad copy
Generating product
copy
SALES LEADERS most common use cases for Generative AI
44%
48%
Create sales enablement materials Create L&D or training content
RESTRICTED DISTRIBUTION
12 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved.
Generative
AI Surveys
Overview Barriers, Benefits, Use Cases Open-ended Insights Peer Data & Insights Additional Insights
RESTRICTED DISTRIBUTION
13 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved.
IT Leaders are saying…
“
“
“We are behind on embracing generative AI for
security purposes, which is regrettable, because,
“I think there is a general nervousness about
predictably, malicious actors are not as behind.”
jumping in too soon here. I think in the next
DIRECTOR
6-12 months we will all get a better
Arts and Entertainment Industry | 10,000+ Employees
understanding of what and how we can
leverage this to our advantage as
“
businesses and as society.”
“We see the huge benefits of generative AI but
are taking baby steps with Chat GPT.”
C-SUITE
Consumer Goods Industry | 1,000 – 5,000 Employees
DIRECTOR
Professional Services Industry | 5,000 – 10,000 Employees
RESTRICTED DISTRIBUTION
14 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved.
IT Leaders are saying…
“ “
“Definitely using but cautiously and “Generative AI has to be seriously
primarily for data analysis and business considered despite its limitations and
planning and forecasting at this point. regulatory challenges, especially for
people in high-regulated industries.”
Not using clinically.”
C-SUITE C-SUITE
Healthcare Industry | 1,000 – 5,000 Employees Finance Industry | 1,000 – 5,000 Employees
RESTRICTED DISTRIBUTION
15 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved.
InfoSec Leaders are saying…
“
“
“This is a new area and all our decisions are
being questioned constantly.”
“Loss of internal IP is rising to the top of our
list as the number 1 risk for ChatGPT use
C-SUITE
within our organization with the potential for
Professional Services Industry | 1,000 – 5,000 Employees
developers to feed it source code to help
improve quality.”
“
“It's not 100% fool-proof and still benefits from
VICE PRESIDENT
Natural Resource Extraction Industry | 10,000+ Employees
human intervention.”
DIRECTOR
Healthcare Industry | < 1,000 Employees
RESTRICTED DISTRIBUTION
16 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved.
InfoSec Leaders are saying…
“ “
“We are currently assessing
“There is still no transparency about
compliance aspects [and] static
data models are training on, so the risk
analysis tool capabilities to continuously
associated with bias, and privacy is
scan AI generated code, and also
very difficult to understand and
forming guidelines for aware and ethical
estimate.”
use of generative AI tools by
engineers.”
C-SUITE
Finance Industry | <1,000 Employees C-SUITE
Finance Industry | <1,000 Employees
RESTRICTED DISTRIBUTION
17 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved.
Software Engineering Leaders are saying…
“
“
“Low-level software engineering jobs will be
replaced by AI.” “[Generative AI will] increase productivity
to a large extent, [and] create a lot of
jobs for software engineers. The
DIRECTOR
Telecommunication Services Industry | 5,000 – 10,000 Employees
department will take a more strategic
tack. More jobs will be created to
“
develop a new set of human work tasks
— many of them of higher value.”
“It will create more volume of new code than we
have resources to keep in check.”
DIRECTOR
Telecommunication Services Industry | 10,000+ Employees
DIRECTOR
Manufacturing Industry | 1,000 – 5,000 Employees
RESTRICTED DISTRIBUTION
18 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved.
Software Engineering Leaders are saying…
“ “
“[Generative AI] is going to change the
overall TAT [turnaround time] for “[Generative AI] will help speed up coding —
producing quality code. [It] may eradicate with human intervention after the main work
a lot of jobs especially at the junior is done by the AI.”
software developer level.”
DIRECTOR
VICE PRESIDENT Natural Resource Extraction Industry | 10,000+ Employees
Software Industry | 1,000 – 5,000 Employees
RESTRICTED DISTRIBUTION
19 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved.
D&A Leaders are saying…
“
“
“AI generated synthetic data is quite sensitive
and needs to be handled securely.”
“It is in [an] early stage and will be tough
to adopt across [the] entire organization
MANAGER
Finance Industry | 5,000 – 10,000 Employees and also ROI cannot be [easily]
calculated. Regulatory issues are a
“
major concern.”
“AI generated [techniques have] a high level of
myopic bias, selecting the right vendor for data
C-SUITE
remains a challenge.” Finance Industry | 10,000+ Employees
MANAGER
Finance Industry | 1,000 – 5,000 Employees
RESTRICTED DISTRIBUTION
20 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved.
D&A Leaders are saying…
“ “
“There has to be [an] integration of Human “It's difficult to reduce bias while also
Resource insights along with AI generated improving accuracy for healthcare data.
synthetic data to improve the utmost So far the only way is to tokenize real-
world data to reduce risk while
effectiveness.”
preserving data accuracy and quality.”
MANAGER DIRECTOR
Professional Services Industry | 5,000 – 10,000 Employees Finance Industry | 10,000+ Employees
RESTRICTED DISTRIBUTION
21 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved.
Supply Chain Leaders are saying…
“ “
”Generative AI can be employed to design and manage
“Ethical implications are humongous while working with
warehouse operations more effectively, optimizing space
AI/ML in supply chain industry. The AI disruptions leading to
utilization, labor allocation, and material handling
elimination of supply chain manpower from various critical
processes. By automating these tasks, logistics companies
stages of business is posing issue for businesses and
can significantly reduce their operational costs and improve
professionals globally.”
overall efficiency.”
DIRECTOR
MANAGER
Education Services | APAC | 501 – 1,000 Employees
Manufacturing | APAC | 10,001+ Employees
“ “
“[Generative AI] will be a part of the supply chain
“[Generative AI has] very bright future for accurate
technology ecosystem, and will be used to
modelling of tasks and find fastest route possible
predict outcomes, prevent issues from occurring,
and inventory replenishment.”
and prescribe actions.”
DIRECTOR VICE PRESIDENT
Consumer Goods | North America | 10,001+ Employees Consumer Goods | APAC | 51 - 200 Employees
RESTRICTED DISTRIBUTION
22 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved.
Marketing Leaders are saying…
“
“
"Once generative AI is integrated with most marketing
technology systems, I foresee prompt based images,
videos and copy being widespread. Imagine creating
"By utilizing the power of generative AI,
multivariate tests using multiple assets in multiple
languages with multiple landing pages." marketing teams can enhance customer
experience and boost sales by creating tailored
MARKETING VP
Hospitality | APAC | <1,000 Employees
content, evaluating customer feedback,
implementing precise pricing strategies,
“
launching focused marketing campaigns, and
automating customer service processes."
"[Marketing teams] should be using generative AI in
all aspects of marketing. Content, digital ad copies,
SEO suggestions, brand video and infographics."
C-SUITE
Finance Industry | 10,000+ Employees
MARKETING DIRECTOR
Finance & Banking | APAC | 1,001 – 5,000 Employees
RESTRICTED DISTRIBUTION
23 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved.
Sales Leaders are saying…
“ “
"Don't rely on it completely so that your "It can serve as a useful outline,
customers will easily find out that you however it lacks innovative thinking.
have used generative AI tool." It reports from past data."
SALES MANAGER SALES DIRECTOR
Professional Services Industry | APAC | 1,001 – 5,000 Employees Telecommunication Services | North America | 1,001 - 5,000 Employees
RESTRICTED DISTRIBUTION
24 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved.
Generative
AI Surveys
Overview Barriers, Benefits, Use Cases Open-ended Insights Peer Data & Insights Additional Insights
RESTRICTED DISTRIBUTION
25 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved.
IT –
ChatGPT policies under development, risk convos make it to the
boardroom
Currently don’t have an acceptable use policy in place for
79%
ChatGPT
In the process of
32%
developing one
69% Use ChatGPT for business purposes
21% Use paid
subscription
RESTRICTED DISTRIBUTION
26 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved.
Additional Results: IT
RESTRICTED DISTRIBUTION
27 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved.
Source: Generative AI and ChatGPT: Adoption and Usage
InfoSec –
AI working groups, data guidelines and humans in the
loop for risk mitigation
Their organization has or will establish new
44%
working groups to manage generative AI security
and risks.
61%
use or plan to usedata guidelines
55% associated with generative AI tools
orfoundational models
RESTRICTED DISTRIBUTION
28 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved. Source: Generative AI Security and Risk Management Strategies
Additional Results: InfoSec
RESTRICTED DISTRIBUTION
29 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved. Source: Generative AI Security and Risk Management Strategies
D&A –
AI-generated synthetic data can overcome real-world data
shortfalls but is not infallible
51%
57%
availability
complexity
60%
56%
adopted AI-generated
improved model
synthetic data because
of challenges with real- efficiency
60%
world data accessibility
improved model
accuracy
45%
mitigated data
privacy
concerns
Source: Generative AI for Synthetic Data
RESTRICTED DISTRIBUTION
30 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved.
Additional Results: D&A
RESTRICTED DISTRIBUTION
31 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved.
Source: Generative AI for Synthetic Data
Software Engineering
– using generative AI, but many lack
governance
More than half of respondents say generative AI
is currently used in their software engineering
department.
60% of those use it forAI-assisted pair
programming
t
78% of those respondents use
ChatGPT.
55% do not have governance policies
in place.
RESTRICTED DISTRIBUTION
32 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved. Source: Generative AI for Software Engineering Teams
Additional Results: Software Engineering
RESTRICTED DISTRIBUTION
33 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved.
Source: Generative AI for Software Engineering Teams
Supply Chain*-
leaders are looking to AI to address the corporate
brain drain and increasing unpredictability
40% of respondents are already using Generative AI
as a part of their supply chain strategy
45%
plan to deploy it soon
t
71% expect that generative AI will become a standard
in supply chain within 4 years.
Nearly half of surveyed supply chain leaders are using or plan to
use generative AI to enhance internal knowledge bases
42%
plan to use it to generate predictive models
RESTRICTED DISTRIBUTION Source: Generative AI for Supply Chain Transformation
34 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved.
*Survey still in collection phase. Results are preliminary
Marketing* –
expects generative AI to become a mainstay in the
MarTech stack, and many are already using it
100%
t
reportedthat they
believe generative AI will be a
regular aspect of marketing
team's tech stacks within6 years
76%
of marketers report their
content marketing teams are
already using generative AI,
the top choice among
respondents.
RESTRICTED DISTRIBUTION Source: Generative AI in Marketing
35 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved.
*Survey still in collection phase. Results are preliminary
Sales* -
Some believe they can completely replace a team member
with generative AI, with sales ops being most common
believe generative AI tools would allow them to
37%
completely replace a person on their team while
still producing the same results.
t
74% believe sales operations roles could be
replaced.
say they would be extremely or moderately
55%
concernedif a customer discovered their content
was AI generated.
Source: Generative AI Sales Tools
RESTRICTED DISTRIBUTION
36 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved. *Survey still in collection phase. Results are preliminary
Generative
AI Surveys
Overview Barriers, Benefits, Use Cases Open-ended Insights Peer Data & Insights Additional Insights
RESTRICTED DISTRIBUTION
37 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved.
One-Minute Insights
Individual Reports for Deeper Dives
1 2 3 4
BENCHMARK YOUR RESPONSES BENCHMARK YOUR RESPONSES BENCHMARK YOUR RESPONSES BENCHMARK YOUR RESPONSES
GET FULL REPORT GET FULL REPORT GET FULL REPORT GET FULL REPORT
RESTRICTED DISTRIBUTION
38 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved.
JOIN THE CONVERSATONS!
Join Gartner Peer Community and thousands of members to
get real-time data and insights, ask questions and get
answers from peers who have been-there, done-that.
RESTRICTED DISTRIBUTION
39 © 2023 Gartner, Inc. and/or its affiliates. All rights reserved. |
297 | ibm | ibm_2024-sustainability-readiness-report.pdf | The State of Sustainability Readiness 2024
How do we close
the gap between
ambition and action?
think report
Foreword
In the broadest sense, sustainability is New data from The State of Sustainability operational budget. At the same time,
about working to preserve continuous Readiness 2024 report shows only half of surveyed leaders feel prepared
operations over time; work that never ends organizations are well underway in to deal with increasingly disruptive
in an ever-shifting landscape of challenges grappling with this task. 9 out of 10 climate risks.
and opportunities. respondents believe in the potential
of AI to contribute to sustainability The potential and consequences of AI
Today, business leaders understand outcomes. 61% of respondents view do not stop at the lines drawn on any
AI reflects both a challenge and an investments in information technology organizational chart, and so this report is Christina Shim
opportunity. AI is accelerating the (IT) for sustainability from the for CEOs, CSOs, CIOs, COOs and more. Chief Sustainability Officer, IBM
discovery of lifesaving drugs and perspective of opportunity and growth Successfully operationalizing sustainability
sustainable materials, optimizing supply rather than cost mitigation. And almost is not about an annual report, it is about
chains and mining efforts, and supporting 90% plan to increase investments in using data and technology to tackle an
the transition to more renewable, IT for sustainability. This data signals organization’s core mission with a smart,
decentralized electric grids. Yet AI organizations understand the enormous strategic, long-term approach. The data
adoption has also driven higher energy opportunity for AI, if implemented shows that more than ever businesses
use and costs for many organizations, correctly, to drive both organizational and are starting to, and must, approach
and even forced some to reevaluate their environmental sustainability. sustainability in its broadest sense—using
sustainability targets. every tool at their disposal to mitigate
At the same time, the report shows there climate threats, support more streamlined
For business leaders, the task ahead is huge room—and need—for growth. More and cost-efficient operations, and stay
is how to maximize the business value than half of organizations (56%) are not yet competitive with their peers.
of AI—delivering the results clients need actively using AI for sustainability, and 48%
better, faster and with higher quality— say investments in IT for sustainability are
while minimizing its costs and “one-off” rather than coming from a regular
environmental impacts.
Introduction
Global industry leaders see the opportunities in The State of Sustainability Readiness 2024 The research found most organizations
report was conducted independently by understand the necessity of climate
using IT—and most prominently, AI—to elevate
Morning Consult and sponsored, analyzed risk mitigation to protect assets and
and published by IBM. Interviews were advance operations. Many acknowledge
sustainability and their companies. However, they
conducted between April and May of the importance of investment in IT,
also see gaps: outmoded policies, confidence in 2024 with 2,790 business leaders and infrastructure and human capital. Candid
decision-makers, across 15 industries, responses revealed perceptions of
tracking progress, energy consumption and lack
in 9 countries. More than 30 survey readiness and progress differ between
questions correlated to climate risk and leadership levels. But almost all the
of expertise, especially in employee skills relating
corporate responsibilities, and covered surveyed C-level executives expect AI to be
to AI and generative AI. In this report, we’ll show strategic, financial, regulatory and a change agent in furthering their business
compliance concerns. while growing climate resiliency.
you how key players are investing in sustainability
through the opportunity of technology.
3
Chapter 1 Chapter 2 Chapter 3 Chapter 4
Stronger IT, greater sustainability The urgency to think ahead The AI sustainability dilemma From risk to resilience
Chapter 5 Chapter 6 Chapter 7
Top challenges: Budget, The perception gap problem Recommendations for readiness
measurement and skills
1. Stronger IT,
greater sustainability
Investing in IT for sustainability isn’t just a matter of doing
the right thing—it can have a net positive impact on your
organization’s current and future success. Along with practical
goals, such as reducing risk to business assets and lowering
energy costs, investing in IT for sustainability can satisfy
stakeholders, attract principled employees and position their
organization for AI in the future.
88
%
of business leaders intend to
increase investments in IT to
advance sustainability efforts.
Chapter 1
Brand reputation
57% 55% 53% 70% 57% 49% 68% 64% 40% 57%
Energy cost
54% 52% 50% 63% 56% 58% 64% 55% 38% 51%
Business resilience
52% 51% 51% 51% 54% 54% 60% 61% 38% 49%
Regulator y pressure
42% 43% 42% 37% 45% 41% 48% 47% 34% 45%
Global US Canada Brazil UK Germany UAE India Japan Australia
Figure 1.
Top factors in increasing IT sustainability investments energy costs and long-term business resilience. Regulatory pressure holds less importance.
Decision-makers cite brand reputation as a key reason for investing in IT and related services for The findings suggest the importance of strategic alignment of organizational objectives and
sustainability. Globally, this reason accounts for 57% of positive responses, followed closely by desired outcomes when it comes to IT investments for sustainability.
Chapter 1
16%
61%
43%
higher rate of revenue growth
is seen by organizations that
embed sustainability.
52%
11%
7%
Global US Global US more of businesses integrating
sustainability outperform their
Cost driven Opportunity driven
peers on profitability.
Figure 2.
What drives sustainability investment? Opportunity.
The top leaders surveyed said they invest in IT sustainability initiatives based on the perceived embed sustainability are 52% more likely to outperform their peers on sustainability
opportunity rather than cost mitigation. And this opportunity mindset is paying off. The drive when compared with those that do not, according to research conducted by the
to invest in such initiatives is benefiting organizations using this approach. Organizations that IBM Institute for Business Value.1
2. The urgency
to think ahead
From Kerala to Sao Paulo, Auckland to Kyoto, global leaders
are increasingly investing more in sustainability IT—but not all
with the same urgency. Economic factors, household income,
government policy and geographic differences undoubtably
affect these investment decisions. But citizens and businesses
around the globe have one thing in common: they need their
leaders to think ahead.
Chapter 2
34%
Germany
52%
UAE
Countries are investing more
in IT for sustainability
Investing in IT for sustainability is
anticipated to grow across global markets
41%
over the next 12 months. Countries, such
Canada 41%
as India, Brazil and Australia, which have
UK
faced recent extreme climate issues, plan
24%
to invest even more than the global average
Japan
of 88%. Within this global average, 44% of
markets plan significant IT investment.
The global cost of climate change damage
is estimated to be USD 143 billion per
year.2 This cost is expected to increase
over time as the impacts of climate change
become more severe. As a result, countries
with lower incomes are at a higher risk from
the adverse economic impacts of climate
37%
change, which is why they might be 70%
US
likelier to invest in sustainability-related India
IT. Leaders in Japan, on the other hand,
answered much more conservatively across
the report’s key metrics.
58%
43%
Brazil
Australia
Figure 3.
India and Brazil plan to
invest 15% over the global
average in IT sustainability
3. The AI sustainability
dilemma
AI is a hot topic in the sustainability community, and it’s
easy to see why. It can help streamline data collection from
various sources, aid sustainability leaders in understanding
environmental risks, and assist in navigating the regulatory
compliance process and making informed decisions.
Additionally, AI can help organizations adapt their operations
to the changing climate, and better maintain their assets and
operations in response.
Chapter 3
90% 96% 95% 32%
Global Brazil United Arab Emirates United Kingdom
Very positive and positive opinions
Very positive opinions
Figure 4.
Leaders agree AI will have a positive influence in achieving sustainability goals
Almost universally, respondents had an overwhelmingly positive take on the influence of AI in responding with very positive opinions about the impact of AI on their organization’s sustainability
their organizations, especially surveyed leaders in developing markets, such as India, Brazil efforts. Whether it’s innate cultural skepticism, policy or geography, it’s tough to measure. But as
and United Arab Emirates. However, leaders in the UK were more ambivalent, with only 32% sustainability increasingly becomes a business imperative, minds may change.
“W e have a commitment to reach net zero
by 2035, and we’ve used that commitment
to influence our recently enacted
Evergreen IT strategy.”
Steve Elliott
Head of IT Services, Water Corporation
Organizations are already seeing results in applying AI to This strategy saved Water Corporation roughly 1,500 hours of
their sustainability efforts. Water Corporation, a Western manual labor annually associated with infrastructure support
Australia state-owned entity, employed AI to help migrate and cut development efforts and associated costs by 30%.
its back-office services to a responsible cloud-first approach. And all these savings helped them offset the cost of running
The team used generative AI to convert plain English into their cloud environment by more than 40%.
code recommendations for the automation functions that
Read the case study on IBM.com ↗ were targeted for the migration and upkeep of the new
environment. Additionally, they automated common, often-
repeated support tasks to be performed automatically.
Chapter 3
64%
56%
44%
40%
33%
32%
29%
28%
27%
of leaders are not using
AI for sustainability efforts
despite widespread positive
sentiments regarding AI.
Global US UK India Global US UK India
Plan to implement AI solution for Actively using AI for
sustainability soon sustainability now
Figure 5.
AI adoption for sustainability purposes shows promise percentage could improve in the near future, however, as 32% of respondents claimed they plan
The report revealed that despite the widespread positive sentiments connected with AI, to implement an AI solution for sustainability soon. India leads all surveyed countries in actively
more than half of organizations studied are not using AI for their sustainability efforts. That using AI at 64%, which is in stark contrast to the UK, where AI adoption is at just 27%.
Chapter 3
2030
AI is powerful, but at what ultimate cost?
For all the power AI can deliver,
The power demand of AI is
organizations must still account for the
expected to rise by 160%.3
energy use it demands—something leaders
are trying hard to mitigate. The good news?
This new adoption of AI is galvanizing
organizations to employ more sustainable
practices, such as utilizing foundation
models, optimizing data processing
locations, investing in energy-efficient
processors and leveraging open-source
collaborations. These strategies not only
reduce the environmental footprint of AI,
but also enhance operational efficiency and
cost-effectiveness, balancing innovation
with sustainability.
2024
Figure 6.
4. From risk to resilience
Climate resilience―anticipating, adapting to and recovering
from the impacts of climate change―is a necessary goal for any
organization. From responsibly managing natural resources to
finding skilled employees to devise solutions to environmental
challenges, business leaders are investing in climate resilience
by thinking of their IT as an engine for sustainability—
and profitability.
Climate resilience is the most
critical sustainability issue
Chapter 4
Climate risk 31%
1/2
Energy use 31%
Availability of skilled staff 30%
Measurement and analytics 26% of leaders surveyed feel fully
prepared to deal with these
aspects of climate risk.
Water use 26%
Figure 7.
A ranking of the most challenging sustainability issues that companies must solve
On average globally, organizations cited climate risk to operations and assets, energy use, and asset life. Leaders in India and Germany are particularly proactive in addressing these issues,
skilled staff availability as their top 3 sustainability challenges. By implementing a strategy that especially the urgent need to address challenges related to water use. Even so, only half of global
prioritizes repairs and replacements, they can improve asset health, predict failures and extend leaders surveyed feel confident in their readiness for climate risks.
“W e’re the largest provider of through-life
support asset management services for
passenger rollingstock in Australia.”
Adam Williams
Head of Growth, Rail and Transit Systems at Downer Group
Harnessing infrastructure data offers a great opportunity. Their AI-powered platform harnesses complex analytics and
With the correct data and proper analysis, companies can near real-time data to support predictive maintenance efforts
identify and fix early problems, extending the life of critical for more than 200 trains.
machinery and reducing maintenance and material waste.
The Downer Group turned to IBM asset management to Downer effectively doubled the number of trains it could
monitor, measure and maintain the trains in its critical maintain from one maintenance center alone—all while netting
Read the case study on IBM.com ↗ transportation infrastructure. a 20% improvement in efficiency.
5. Top challenges: Budget,
measurement and skills
The big 3 challenges facing business leaders are: how much to
allocate to sustainability efforts, how to measure sustainability
key performance indicators (KPIs) and how to stay staffed with
experienced workers amid current labor shortages.
Chapter 5
48%
Global allocations for sustainability
The first big challenge organizations face
Exceptional budget created
when it comes to investing in sustainability is
specifically for IT and aligned
financial planning. One telling statistic reveals
services sustainabilty
whether a company considers sustainability
measures as part of their operating budget or
relegates it to the lower-priority exceptional
budget—or even figures it in at all. Responses
27%
showed only 26% said IT sustainability is
part of their regular operational budget,
Dedicated sustainability budget
signaling that IT sustainability is not a priority
within the overall strategy of organizations.
What accounts for the gap between
dedicated and exceptional budgets? Quite
26%
likely, opportunity. With many sustainability
issues tied to energy consumption and
Regular operational budget
IT and the rising investment in AI, an
increased IT budget could signal that
organizations are beginning to see the
benefits and how to operationalize
sustainability through new technology.
Figure 8.
50
%
of business leaders believe their
data to measure sustainability KPIs
isn’t very mature.
Chapter 5
Total energy Renewable energy Recycling
consumption consumption
79%
72%
52%
49% 50%
48%
29% 42%
51%
35%
38%
32%
32%
27%
20%
40%
37%
14%
42%
7%
1% 1%
Plastic use GHG emissions GHG emissions Supplier metrics
Scope 2 Scope 1
Global US Brazil India Global US Brazil India Global US Brazil India
Water use Waste GHG emissions
generation Scope 3
Growing up Hitting our stride Walking confidently
Figure 9. Figure 10.
Top KPIs used to measure sustainability outcomes How mature are organizations in using data to track progress?
The second challenge is knowing where to begin. Leaders in most surveyed countries looked to Reliable data plays a significant role in improving and tracking progress on sustainability
resource efficiency, citing renewable energy consumption, total energy consumption and recycling goals. Metrics are inconsistent across markets in their ability to track sustainability KPIs.
as their top 3 KPIs for sustainability outcomes. Renewable energy consumption is cited as the top India and Brazil’s strong reliance on brand reputation, elevated expectations with AI
KPI in Brazil, UAE and the UK. Over 80% of global energy production comes from fossil fuels, which integration, relatively advanced stages of digital transformation preparedness and
are nonrenewable resources, such as coal, oil and gas.2 However, cleaner, renewable sources of commitments to skill enhancement through targeted investments may contribute to
energy—solar, wind, geothermal, hydropower, ocean energy and bioenergy—are gaining ground. this confidence level.
Chapter 5
The third pressing need for executives is
expertise in AI and generative AI, sustainable
business strategies, and renewable and clean
energy. Responses showed skills in AI were
most desired in Brazil, at 53%, and the US,
at 47%, with a strong appreciation for
sustainable business strategies in UAE at 44%,
Australia at 40% and Canada at 39%.
“T he supply chain for renewable
products was in many ways a new
kind of business, and we needed a
new foundation to build it on.”
Marko Mäki-Ullakko
Head of Integrated ERP, Neste
Along with measuring and finding efficiencies, technological As a catalyst to achieving this goal, they needed a truly global
breakthroughs also present new sustainability opportunities. supply chain strategy to effectively manage their network of
Neste, the world’s leading producer of sustainable aviation advanced renewable refineries and technologies.
fuel and renewable diesel, aims to help customers reduce
their greenhouse gas (GHG) emissions by at least 20 million IBM Consulting® provided Neste with the process design
tons annually by 2030. support it needed to optimize its enterprise resource planning
Read the case study on IBM.com ↗ (ERP) investments.
6. The perception
gap problem
There’s a real chasm in the ways organizations view the promise
of AI and the way they actually use it. For instance, many
industry leaders report a desire to bolster the resilience of their
assets, infrastructure and supply chains in the face of potential
climate hazards, but only half of respondents believe their data
is mature enough to measure sustainability KPIs. Perhaps at
the core of these disconnections is the perception gap that
exists between C-suite executives and those more likely to
implement operational decisions, with the former generally
having a rosier perception than the reality.
Different perceptions between leadership levels can signal a
divergence in direction or, perhaps worse, no direction at all.
But it presents an opportunity for dialogue. In their responses,
C-suite executives revealed a more optimistic outlook than the
vice presidents and directors who work for them—and are more
likely to implement operational decisions. The study revealed
that while reducing potential damage from climate risk is a
common goal, division can and does creep in when ambition
(what we should do) meets action (how we can achieve it).
Chapter 6
11% gap
67%
Business leaders’ perceptions differ on
approach to climate resilience efforts
Surveyed business leaders showed a
marked divergence in their perceptions
56%
of readiness and active defense against
climate risks. Top-level executives felt
more proactive when addressing and
acting on climate resilience efforts than
lower-level decision-makers.
Percentage describing their organization’s climate resilience efforts as proactive
C-suite executives
Vice presidents and directors Figure 12.
Chapter 6
13% gap 7% gap 9% gap
Confidence levels about readiness
for climate risk factors diverge
In their responses, C-suite executives
revealed more confidence than vice
55%
presidents and directors that their
52%
51%
organization was prepared to handle
different aspects of climate risks.
45%
Financial risks represented the largest
42% 42%
gap between the groups at 13%.
Financial risks Physical infrastructure risks Supply chain risks
C-suite executives
Vice presidents and directors Figure 13.
Different leadership
perceptions present an
opportunity for alignment
Chapter 6
12% gap 10% gap
51%
47%
39%
37%
Percentage claiming AI will very positively impact achieving sustainability goals Percentage claiming AI is actively used for sustainability efforts
Figure 14a. Figure 14b.
Business leaders have gaps regarding the impact of and use for AI in sustainability
As AI becomes more prevalent in operations, C-level executives tended to show a more positive
outlook than vice presidents and directors about the impact that AI could have on achieving their C-suite executives
sustainability goals and in the way their organizations put AI to use in sustainability efforts.
Vice presidents and directors
Chapter 6
32% gap 13% gap 6% gap 15% gap
57%
51%
47%
44%
42%
36%
32%
19%
Banking and financial services Telecommunications Retail Manufacturing
Figure 15.
Industries reveal perception gaps on the use of AI for sustainability efforts
When considering the impact of AI and its use for sustainability progress, leaders in the banking
and finance industry presented the strongest divergence between C-suite and lower-level
decision-makers. But responses from the retail industry showed the opposite, with 42% of
vice presidents and directors stating their organizations were actively using AI for sustainability C-suite executives
initiatives, as opposed to the C-suite executives’ response of 36%.
Vice presidents and directors
7. Recommendations
for readiness
Sustainability challenges tend to carry over from year to year,
but leaders said they feel those challenges more strongly
this year than before. Even with the optimistic attitudes cited
by C-suite executives, they accept they need to alleviate
challenges as a whole organization. Here’s how to address
notable issues head-on.
1 2
Mind the gap Invest in upskilling
Organizations should use data to obtain a To address the growing, seemingly infinite
more holistic view of their operations and need for digital skills related to climate
understand where the different perceptions risk mitigation, the workforce must make
between C-suite and other decision-makers an even stronger turn toward technology
originate. To keep an eye on changes training. One way to start is with online
and blind spots, use a data analysis and skills-based courses that offer free training
reporting tool to help maintain a state of and reskilling with purpose-designed
readiness visible to individuals across the curricula for all skill levels. Organizations
organization, so they can proactively come can also identify skills needs and bridge any
up with and implement solutions. existing and anticipated future skills gaps
by tapping their ecosystem of partners.
Chapter 7
3 4
Invest in IT for smarter assets Invest in AI
Developing predictive maintenance Trusted AI tools can help save time
practices can facilitate more efficient and money in sustainability efforts. For
resource allocation and business example, generative AI can provide insights
operations. Consider investing in an that help identify opportunities to reduce
application suite with intelligent asset carbon emissions, create scenarios
management, monitoring, predictive and algorithms for better practices,
maintenance and reliability planning in and simulate risk scenarios, including
a single platform. This investment can ephemeral details, such as weather or
enable your organization to optimize asset local disasters. An investment in AI can
performance, extend asset lifecycles and also help reduce the workforce skills talent
reduce downtime and costs. shortage many executives identified in
this report. The desire for expertise in AI
and generative AI is documented, and an
investment in that area can aid in filling the
need. Explore use cases to increase your
portfolio of ideas.
Methodology
2,790 15
This poll was conducted online by Morning
Consult among a sample of 2,790 business
leaders from each of these markets: the
US, Canada, Brazil, the UK, Germany, UAE,
India, Japan and Australia.
This report highlights global leaders overall
and by market, along with the C-suite
executives (high-level decision-makers) executives industries
and vice presidents and directors
(lowerlevel decision-makers).
30 9
To qualify as a vice president or director,
respondents must be employed at a
company with more than 1,000 employees
and work as a director or vice president.
To qualify as a C-suite executive,
respondents must be employed at a
company with more than 100 employees
and must be the owner or work in the
questions countries
C-suite. A significant proportion of the
C-suite audience includes executives
working at companies with more than
1,000 employees.
About us
About Morning Consult About IBM
Morning Consult provides global survey IBM can help you plan a profitable
research tools, data services and news We’re doing business in an path forward with open, AI-powered
to organizations in business, marketing, unpredictable world. solutions and platforms and deep industry
economics and politics. Morning Consult expertise that address your goals in
surveys thousands of people around the Success requires new levels of resilience 5 key areas: climate risk management;
world every day, pairing that exclusive, and agility rooted in responsible practices infrastructure and operations; supply chain;
forward-looking survey data with that preserve our planet for future electrification, energy, and emissions
analytical applications, to offer a distinct, generations. Sustainability is now a management; and a sustainability strategy.
competitive advantage for our users and strategic business imperative. For more information, visit
cement our leadership in the decision ibm.com/sustainability or subscribe
intelligence category. to receive sustainability updates.
1. Beyond checking the box: How to create business © Copyright IBM Corporation 2024 THE INFORMATION IN THIS DOCUMENT IS
value with embedded sustainability, IBM Institute PROVIDED “AS IS” WITHOUT ANY WARRANTY,
for Business Value, 27 February 2024. IBM, the IBM logo, IBM Consulting, and Think are EXPRESS OR IMPLIED, INCLUDING WITHOUT ANY
trademarks or registered trademarks of International WARRANTIES OF MERCHANTABILITY, FITNESS FOR
2. World energy outlook 2023, International Energy Business Machines Corporation, in the United A PARTICULAR PURPOSE AND ANY WARRANTY OR
Agency, October 2023. States and/or other countries. Other product and CONDITION OF NON-INFRINGEMENT. IBM products
service names might be trademarks of IBM or other are warranted according to the terms and conditions
3. AI is poised to drive 160% increase in data center companies. A current list of IBM trademarks is of the agreements under which they are provided.
power demand, Goldman Sachs, 14 May 2024. available on ibm.com/legal/copytrade.
The client is responsible for ensuring compliance with
This document is current as of the initial date of laws and regulations applicable to it. IBM does not
publication and may be changed by IBM at any time. provide legal advice or represent or warrant that its
services or products will ensure that the client is in
Not all offerings are available in every country in which compliance with any law or regulation.
IBM operates.
Examples presented as illustrative only. Actual results
will vary based on client configurations and conditions
and, therefore, generally expected results cannot
be provided.
38 |
298 | ibm | 6-power-moves-cfos-must-make.pdf | IBM Institute for Business Value
Global
C-suite
Series
31st Edition
CFO Study
6 power moves
CFOs must make
Tackling hard truths
in the generative AI era
Contents
2
Introduction
The imperative
of bold leadership
6
What sets leading
About the study CFOs apart?
In Q1 2024, in cooperation with Oxford Economics, the IBM Institute
for Business Value surveyed 2,000 Chief Financial Officers (CFOs). 9
Separately, a small group of executives was engaged for in-depth The six power moves Champion tech Make execution the Show me the ROI.
as core. yin to strategy’s yang.
qualitative interviews. These discussions focused on key insights from
the study and the CFOs’ on-the-ground experience leading their finance
organizations and interacting across their organizations on developing
and executing strategy in the new era of AI. Respondents span
10 18 26
26 industries and 34 locations worldwide. For more details,
Strategy Execution Investment
see “Research methodology and analysis” on page 60.
The cover concept and individual patterns in
this report were developed using generative AI.
IBM IBV designers translated each of the “power moves” into prompts, Determine your risk Make data your Ignite your talent 59
and then used these prompts within Adobe Firefly to generate tolerance, then place AI’s oxygen. revolution. Conclusion
vector-based imagery that inspired the basis and structure your big bets. Leading through
for each pattern. Similarly, the photos that appear in this report technology-driven
change and uncertainty
were identified using AI-assisted, natural-language search,
using the generated patterns as reference images.
Overall, the efficiency gained by integrating these
34 42 50 60
tools into the design process is as follows: Risk Data Talent Research
methodology
Concept—3 weeks to 1.5 days
and analysis
Patterns—2 weeks to 2 days
Photography—1 week to 2 hours
The
imperative
of bold
leadership
It’s the generative AI era. Will your Nearly two-thirds of CEOs in our 2024 CEO Study as accelerating technological modernization,
organization disrupt—or be disrupted? say they need to rewrite their organizational creating economies of scale, differentiating
playbook to remain competitive.1 You can’t products and services, and building strategic
It depends on how your organization
run the business of tomorrow with today’s skills, alliances (see Figure 1) to help them deliver this
reacts today and prepares for tomorrow.
technology, or operating models. And with 72% advantage. Yet, our study reveals some CFOs are
of top CEOs saying competitive advantage depends leaping forward while others are being cautious:
Generative AI is a once-in-a-generation
on who has the most advanced generative AI,2
opportunity to drive radical productivity – 72% of leading CFOs identify their CTO as highly
the tensions finance leaders already face are
and create new avenues for explosive important or critical to their success, more than
compounded: risk versus reward, short-term versus
“The role of the CFO has migrated any other role.
growth—but only for leaders who make long-term, optimization versus constraint, agility
the right moves. CFOs must use from a more traditional financial versus discipline, and innovation versus financial – 65% of all CFOs say their organization is under
perspective to one with a really stability. So, what’s a CFO to do? pressure to accelerate ROI across their
their influence to advance their
strong understanding of the technology portfolio.
organization in innovative ways. CFOs must be both guardians of stability and agents
business and a repository – Only 35% of finance organizations engage
of transformation.3 At the epicenter of strategy
Execution is essential. for insights.” development and execution, CFOs, working hand in early IT planning with tech leaders to set
expectations on how technology advances
Complacency is not an option. in glove with tech leaders, supercharge technology
Diana Vuong enterprise strategy.
as the transformative force in their organization.
VP, Finance & CFO, Vancouver Airport Authority
In this high-stakes environment, CFOs must make
CFOs play a crucial role in driving competitive
power moves that confront hard truths about
advantage and creating value for their organizations.
transformation, talent, innovation, and more.
They are looking to change-heavy initiatives such
2 Introduction 3
Figure 1 The CFO leaders
check once in layout
The stuff leading CFOs are made of
Competitive advantage and barriers to overcome
CFOs expect competitive advantage coming Perspective
from change-heavy initiatives and innovation
requiring tough choices. How are leading A sharp focus on a strategic future
1 Leading CFOs articulate vision and
CFOs preparing
strategy to optimize long-term
for an uncertain future? investment value. They not only
Most important enablers of competitive Greatest barriers to innovation
control and manage risk, but also
advantage over the next 3 years in the organization
finesse it with financial effectiveness.
#1 Accelerated technology #1 Management resistance
modernization to change
We’ve identified a group of leading CFOs,
representing 9% of our global data set,
#2 Scale advantages #2 Aversion to risk/ that is outperforming the competition.
(economies of scale) disruption
tie
Here are the four powerful CFO
An adeptness at strategy execution
characteristics that enable tackling 2
#2 New differentiated products #3 Limited budget/financial Leading CFOs translate decisions into
the hard truths of the generative AI era. financially astute actions—as well as
and services resources
tie tie embrace cloud-based enterprise
performance management.
#2 Strategic alliances #3 Lack of clear innovation
and partnerships strategy
tie tie
Qs. What are the most important enablers of your competitive advantage over the next three years?
Which factors present the greatest barriers to innovation in your organization?
An agile responsiveness to changing market
3 conditions and new opportunities
Leading CFOs are decisive, making decisions
with more speed and financial effectiveness than
their competitors. They’re enabling better, faster
62% decisions across their enterprises.
72%
of CEOs say they
must rewrite their
organizational playbook.4 A keen eye for technology that drives
of top CEOs say competitive 4 competitive advantage
advantage depends on who has Leading CFOs connect tech investments to
quantifiable business outcomes. They create
the most advanced generative AI.5
cross-functional teams to forecast and allocate
technology costs and manage tech budgets.
4 IntroSdtruactteiogny 55
The CFO leaders The CFO leaders
What sets leading High performance starts with financial metrics. Leading CFOs And their success transcends financials. “It is important for not only
CFOs apart? drive significantly higher annual revenue growth and operating Leading CFOs say their organizations the CFO but also management
margins than their peers. In 2023, they outperformed their outperform in several key areas that
to have a broad perspective
competition by 39% in revenue growth. In 2023, leading CFOs deliver a competitive edge.
in order to see the bigger picture.”
achieved nearly 11% greater operating margins.
Leading CFOs Leading CFOs Iwaaki Taniguchi
Director, Executive Vice President & CFO
All others All others
Chugai Pharmaceutical Co., Ltd.
Talent development
Annual revenue/budget growth Annual operating margin and retention
80
+6.7%
60
36%
Developing more Executing
enterprise
19%
40 enterprise
16% strategy more strategy
20 32%
+9.1%
more
+10.6%
20%
more
+9.1%
24%
12%
Technological more Brand
+3.8%
maturity
22%
reputation
more
+39.3% Cyber risk and
cybersecurity
8%
Qs. Rate the effectiveness of your organization in the following areas: Developing enterprise strategy, executing enterprise strategy.
2020-2022 2023 2024-2026 2020-2022 2023 2024-2026
Percentage reflects “effective” and “highly effective.” How does your organization’s performance compare to similar organizations over
Estimated Estimated
the past three years? Percentage reflects “outperformed” and “significantly outperformed.” How would your closest competitor rate your
organization’s performance compared to similar organizations? Percentage reflects “leading” and “significantly leading.”
6 Introduction 7
The six
power moves
“You have to be able to understand the
levers of your business and how each
Here are six power moves that CFOs 1
Champion tech
must make—from managing risk tolerance
and every decision impacts your
as core.
to strengthening tech partnerships
to embracing generative AI—that
business—and then potentially which
successfully use technology to propel
their organizations forward. 2 Make execution the
levers need to be pulled in order
yin to strategy’s yang.
to achieve the desired result.”
3
Show me
Martin Günther
CFO, smart Europe GmbH the ROI.
4 Determine your risk
tolerance, then place
your big bets.
5
Make data your
AI’s oxygen.
6
Ignite your talent
revolution.
8 Introduction 9
Every product is becoming a digital product. Yet when organizations see technology
as an enabler, they treat it like a toolbox. They wait for problems tech can solve rather
than exploring what new opportunities it creates. Only when they recognize technology
as the transformative force at the core of innovation can organizations seize first-mover
advantages, define markets, and gain economies of scale.
Tech strategy and business strategy cannot be separated, and the CFO plays as important
Champion tech a role—if not more—in this integration as anyone. To drive innovation, competitiveness,
and sustainable growth, the two must align.
CFOs must seize this opportunity to transform their role: this requires shifting from
as core.
a financial overseer, approving or rejecting technology project funding, to a strategic ally.
In this new role, CFOs align technology initiatives with advancement of enterprise strategy
and promote technologies that drive transformation. They need to evolve from evaluating
tech investments via traditional business cases to using a holistic value assessment that
With technology becoming core to the enterprise—not just an enabler— includes strategic, operational, customer, and financial impacts.
CEOs recognize that collaboration between finance and tech is crucial
With time of the essence, and detailed business cases too often yesterday’s “nice to have,”
to success. Leading CFOs say CTOs are their most important relationship,
it is on CFOs to make sure that tech is at the table, providing invaluable context and
with 72% identifying them as highly important or critical. Now is the time knowledge. One telling trait of a leading CFO? They recognize that, when pursuing
for CFOs to supercharge technology as the transformative force at the enterprise success, the most important relationships are technology related.
core of innovation. CFOs need to advocate for tech leaders’ expertise
Leading CFOs are diplomats, with the CTO, CISO, and CIO topping their list of CxO
in the boardroom by integrating evaluation, investment, and planning, relationships needed to be successful. With no single tech role being the lone expert,
forming a powerful coalition that drives strategic growth. leading CFOs recognize that each tech leader brings their unique strengths to form
a high-powered coalition necessary to drive critical technology transformation. Leading
CFOs say CTOs are their most important relationship, with 72% identifying them as highly
important or critical. And 63% of leading CFOs value relationships with their CISO. Given
“The real question is: Who will
their longstanding influence, CFOs are in the best position to actually help build the long
adapt and ride this wave of change?
overdue bridge between business strategy and technology.
Those who do will thrive, while
others risk being left behind.”
Fabio Martinez “I work closely with our CIO to make sure that we make
CFO, Alstom Brazil
the right decisions. We evaluate trade-offs.”
Yoshito Murakami
CFO Power & Gas Power, GE Vernova
1100 Strategy 11
CEOs rely heavily on CFOs and tech leaders for their assessment of technology. Nearly half of these high performers engage in early IT planning to set
Where is the greatest potential? What captures value? CEOs need boots on the expectations on how technology advances enterprise strategy in driving
ground to gather intelligence yet only 50% of CFOs say that finance connects innovation, efficiency, and competitive advantage. 115% more leading CFOs
technology investments to quantifiable enterprise business outcomes. embed nonfinancial technology metrics (for example, user engagement and
speed to market) into business cases, providing a more comprehensive view of
When compared to all other CFOs, leading CFOs blaze the trail to help ensure
technology investment impact on the overall business. Over half meticulously
that technology investments are aligned with financial goals and overall
measure the performance of technology investments, helping ensure that each
business strategy (see Figure 2).
dollar allocated to technology drives tangible business outcomes.
By nurturing tech relationships and engaging in the strategies above, all CFOs
Bringing tech to the table
can harness tech’s potential, inform strategy, and capture value—and educate
Only 50% of all CFOs
their CEOs accordingly.
say that finance connects technology investments
to quantifiable enterprise business outcomes.
Figure 2
Bringing tech to the table
Leading CFOs Engage in early IT planning to set Embed technology metrics
expectations on how technology (such as user engagement, speed
All others
advances enterprise strategy to market) into business cases
48%
56%
72% of leading CFOs
say CTOs are a highly important or critical
34%
relationship, more than any other role.
26%
Q. To what extent has finance collaborated with your technology leaders to do
the above? Percentage reflects “to a large extent” and “to a very large extent.”
12 Strategy 13
What to do
“I see IT and the CIO as my strongest counterparts.
The CIO has to understand the whole end-to-end
process, translating business requirements—and
then implementing them for an IT solution.”
Martin Günther
CFO, smart Europe GmbH
Forge a tech-finance alliance.
– Refocus funding discussions around long-term value creation instead
of traditional “CAPEX versus OPEX” debates.
– Redefine technology investment success metrics to capture indirect benefits
such as user engagement.
– Work with tech leaders at the beginning of technology planning processes
to create board presentations with financial analyses and strategic
considerations related to technology initiatives.
Apply FinOps across the enterprise to make technology
more valuable.
– Set up/refine your organizational home for FinOps with a responsibility
assignment matrix with resources from finance, IT, and the business.
– Implement a cost estimation and tracking framework that can help your
team understand the costs associated with technology projects.
– Modernize budgeting, forecasting, and chargebacks to reflect costing,
agile scenario planning, and incentives for shared objectives.
Decode tech value.
– Track and report post-implementation value creation. Quantify tangible and
intangible benefits, including revenue enhancement, productivity, customer
satisfaction, insights for decision-making, and brand impact.
– Articulate the opportunity or pain point that the technology can address
and its impact on KPIs.
– Assess the risks of not making technology investments. Not investing could
contribute to inefficiencies, limit the ability to scale operations, lead to falling
behind competitors, result in missed opportunities, hinder service delivery,
cause customer dissatisfaction, yield higher costs, and increase risks.
14 Strategy 15
Case study
The Standard: Rationalizing cloud
costs by aligning IT spend with
key business priorities6
Successful FinOps practices combined with Technology Business Management
(TBM) exemplify the budding synergy between finance and technology. The
disciplines of FinOps and TBM foster a collaborative culture that breaks down silos
so organizations can translate cloud and other technology investments into value.
The Standard, a leading provider of financial products and services, is realizing
the benefits of adopting these practices. Facing a lack of transparency on key
drivers of technology spending, the organization’s business and IT teams were not “Our approach for digitalization
working together efficiently. The company was relying on a legacy ERP system and
spreadsheets to prepare the budget, analyze financial data, and make decisions is the operational iron triangle,
about technology investments—a manual and time-consuming process that was
prone to error.
where finance, IT, and operations
The Standard implemented an IBM Apptio® solution to build cost transparency,
management come together
provide actionable insights, and enable faster decision-making. Adopting FinOps
and cloud governance practices alongside the Cloudability product gave the
company insights into its cloud spending—allowing it to drive greater accountability to form the driving engine.
by enhancing cloud procurement and provisioning decisions. In addition,
the Target process product helped the company improve its resource and program IT represents technology,
management—aligning team workstreams to business priorities, gaining greater
visibility into consolidated workflows, and tracking dynamic variables like status, operations represents process,
stakeholders, dependencies, and progress.
and finance represents data.”
The Standard has realized significant benefits. It has increased business/IT
alignment and financial agility, with the IT Finance team now able to focus 80%
of its time on analysis, decision support, forecasting, and insights. The company
FuShan Hu
has also gained more control over cloud spend, with projected savings of 10%
CIO, CHINT Group Co., Ltd.
in 2023 and even more in 2024. Additionally, the company improved its say:do ratio
by 20%—a measure of the gap between what the IT organization says it will
do and what it actually delivers. The company plans to continue investments
in cloud governance to drive similar business results across the organization.
1166 Strategy 17
Executing strategy requires a coordinated dance. It’s choreographing strategy into
actionable steps, optimizing resource constraints, and aligning and communicating
priorities across the enterprise. Executing strategy is complicated, and less than
half (46%) of all CFOs say their finance functions are effective at strategy planning
Make execution the and execution—even though that’s up from 38% in 2022.8
The strategic direction of the enterprise can only be steered by thoughtful metrics
to drive behaviors needed to achieve the organization’s objectives. CFOs have
yin to strategy’s yang.
shifted from evaluating strategy performance using siloed views of financial and
operational metrics toward a broader outlook that’s more suitable to their role as
cross-organizational leaders.
That perspective considers a combination of financial and operational KPIs,
CEOs are accelerating transformational change in 2024, with 77%
including customer satisfaction, product quality, employees, and project delivery,
maintaining or increasing their pace.7 With decision-making processes
all holistic indicators of an organization’s future health and productivity (see
and performance management intertwined, CFOs are poised
Figure 3). This expansive approach empowers all CFOs to more easily and
to be agents of change. To succeed, CFOs must balance precision effectively fine-tune strategies.
and agility while navigating new factors influencing strategies,
Succeeding in finance requires purposeful decision-making agility. To implement
requiring a harmonious “dance” of strategic planning and execution.
strategies, CFOs must move away from structured, methodical processes such as
In fact, 36% more leading CFOs are able to respond with agility to hierarchical, fixed-planning cycles and periodic reviews with minimal feedback
changes in strategy than their peers. loops. Leading CFOs are engaging robust monitoring and reporting mechanisms—
using shorter planning cycles and insights to quickly adapt to changes.
“Executives are asking us how
“The CFO is the architect responsible for making
non-financial and sustainability
things happen by translating and facilitating
factors impact our financials.”
the company’s mission through culture.”
Masakatsu Sato
Júlio Ponte
Director, Managing Executive Officer (CFO)
CFO, Terral Agricultura e Pecuária S.A.
Sumitomo Mitsui Trust Holdings, Inc.
18 Execution 19
Leading CFOs measure the contributions of critical processes. 54% of them Figure 3
(26% more than peers) foster adaptability by forecasting and communicating
Establishing a foundation
business environment changes. They act—sometimes with ruthless precision.
for future health and profitability
When it comes to redeploying capital from underperforming projects, they report
Most important non-financial metrics
cutting their losses and redirecting resources to areas that will drive growth
to help predict an organization’s
29% more often than their peers. As a result, they can adeptly respond to shifts Esftuatbulriseh hinega lat hfo aunndd aptrioonfi tfaobr ifluittyure
and changes in strategies and business models. health and profitability
Most important non-financial metrics to help predict
For all CFOs, a combination of agile performance management and a broader set organization’s future health and profitability
of KPIs beyond financial metrics is critical to orchestrate business evolution. 43% 41% 40% 38%
Execute with purpose.
Only
Customer satisfaction/ Net retention rate Product quality/percent Average time
46%
net promoter score defect percentage to hire
36% 36% 36% 31%
of all CFOs say their finance
functions are effective at strategy
planning and execution.
Customer lifetime value On-time rate Employee Return on security
turnover rates investment
30% 30% 28%
36%
Overdue project Employee Innovation
percentage engagement quotient
more leading CFOs are able
to respond with agility to changes
in strategy than their peers.
Q. Which of the above non-financial metrics are most important to
Chheelp cprked icwt yohure ornga niiznat iolna’sy fuoturue htealth and profitability?
20 Execution 21
What to do
“If you look at the role of finance, it’s not only the
accounting and the controlling and looking into the
numbers, but also making the business model happen.”
Martin Günther
CFO, smart Europe GmbH
– Partner across business units to translate high-level strategic goals into
financial targets and metrics.
– Enforce financial discipline by monitoring performance of critical processes
against goals and benchmarks.
– Promote a culture of accountability where leaders take ownership of their
decisions and outcomes.
Foresee possibilities.
– Assess, forecast, and communicate economic factors, industry-specific
competitor actions and geopolitical trends that could impact the organization.
– Adapt financial strategies in response to changes in the external environment
or shifts in company priorities. Embrace agile processes that allow teams
to seize new opportunities and adapt to threats.
– Anticipate various future scenarios and their potential implications
on the organization’s financial health and strategic objectives.
Achieve impact.
– Set and monitor outcome indicators that serve as early warnings for potential
opportunities or risks associated with customers, employees, product quality,
security, and innovation.
– Determine the data sources and leverage financial management systems,
analytics tools, and business intelligence platforms to track your
outcome indicators.
– Prove business value during regular reviews and report progress on financial
metrics, operational efficiency indicators, and market performance measures.
22 Execution 23
Case study
Ikano Group: Preparing your
organization to meet sustainability
reporting requirements9
“Because you are seeing the
organization a lot more broadly
The Ikano Group, a multinational conglomerate, has sustainability embedded into
its DNA as a key driver for its business strategy. The group recognized the need and from a different lens at
for an ESG reporting and data management solution to support each of its
six subsidiaries’ work toward individual sustainability targets.
times, you help put in place
The group is implementing the IBM Envizi™ ESG Suite to simplify complex data
plans to execute strategy which
capture and reporting required by the EU’s Corporate Sustainability Reporting
Directive (CSRD). The platform is being configured for each Ikano business to
streamline the reporting and disclosure process. There are already 15,685 data is often directly tied to your
types being captured in the platform across all businesses. Envizi allows Ikano
Group to trace data to source, maintain change records, and provide direct capital allocation.”
auditor access, making all businesses audit-ready.
The implementation has enabled Ikano Group to establish aggregated KPIs Diana Vuong
across all businesses, increasing confidence in their sustainability data VP, Finance & CFO, Vancouver Airport Authority
foundation and accuracy of emissions calculations. The Envizi ESG Suite
provides a reliable accounting system for ESG performance metrics, such as
energy, water, materials, and recycling data, making it robust and accessible.
Within Ikano Group, the CFO is accountable for the CSRD reporting process.
A Steering Committee at the Group level meets monthly to review project
progress, discuss challenges, track advancements across the businesses,
and make necessary decisions. Regular updates are also shared and discussed
across the C-suite and in management meetings at the Group level. Additionally,
meetings are held regularly with all the business CSRD teams, which include the
CFO, CSRD project coordinator, and function managers from all departments
(finance, sustainability, risk/legal, HR, and operations).
2244 Execution 25
CFOs understand that tech investments must be made, and tech improvements
that support long-term business strategies need to be prioritized. However, less
than half say their finance functions are intimately involved in the development
of technology business cases. Neglecting prioritization for exciting new use cases
Show me that come into view every day will constrain future growth.
With nearly two-thirds saying they are under pressure to accelerate the value
across the technology portfolio, CFOs can be compelled to pursue short-term
the ROI.
investments. But these choices can ultimately hinder digital transformation by
compromising long-term strategy, undermining employee morale, and sacrificing
critical needs in technology and innovation.
While their peers sometimes get mired in short-termism, leading CFOs play the
CEOs sacrifice long-term innovation for short-term gains, citing
long game. They’re not sacrificing growth for short-term gains; instead, they strike
short-termism as their biggest hurdle.10 57% of CFOs succumb to
a balance between efficiency and innovation. 30% more of these top-of-game
prioritizing short-term targets over long-term investments—often
CFOs hit that sweet spot, balancing both cost reduction/efficiencies and growth
increasing technical debt, for example, by sacrificing long-term opportunities (see Figure 4).
maintainability for short-term functionality. Opportunities in tech,
sustainability, and emerging markets demand a departure from
traditional investment strategies. With their strategic oversight Figure 4
Balance investments focused on efficiency with
and financial acumen, CFOs must guide investment decisions that gBraowlatnhc oep ipnovretusntmitieesnts focused on
capture value over short-, mid-, and long-term horizons. efficiency with growth opportunities.
Leading CFOs 52%
All others
40%
“We can no longer assume static environment
assumptions when we make investment decisions.
Q. To what extent does your finance function do the above to manage your organization’s
We’re going to have to revisit, validate, test, and pivot,
investment priorities? Percentage reflects “to a large extent” and “to a very large extent.”
and that requires us to change the way we manage
check how it works in layout
the process.”
Yoshito Murakami
CFO Power & Gas Power, GE Vernova
26 Investment 27
These CFOs are also tech-savvy trailblazers, intimately involved in developing 65% of CFOs say their organizations are
business cases that connect tech spend to real business value. A substantial under pressure to accelerate ROI
20% more of these leading CFOs drive digital transformation through rigorous across their technology portfolio.
involvement in the development of technology business cases.
How do leading CFOs manage such a nuanced approach? What sets them apart
is an agile multihorizon investment strategy. Over half of them are not just focused
on quick wins or distant dreams; they’re allocating investments across the short,
medium, and long term. This pragmaticism helps ensure they’re servicing
immediate needs—but not at the expense of future ambitions.
Tightrope walking is tricky, especially when it comes to investments.
Wobbles happen, risk is incurred. And risk is necessary—if CFOs can manage
it more adeptly than their competitors, their organizations can ultimately
benefit over the long haul. Forward-thinking CFOs provide the guidance needed
in the organization’s strategic investment process. They’re visionaries who help
the entire organization find ways to effectively balance short-term pressures
with long-term value.
Over
“We’ve got five lenses, and those lenses are
half
expected to be used through every single decision
we make. You put it through the financial lens,
the reconciliation lens, the climate lens,
of leading CFOs allocate
the digital lens, and the customer lens.”
their organization’s investments
across the short, medium,
Diana Vuong and long term.
VP, Finance & CFO, Vancouver Airport Authority
28 Investment 29
What to do
“We must be mindful of allocating resources to balance
investments in growth areas, ensuring that we don’t
miss out on opportunities for expansion.”
Masakatsu Sato
Director, Managing Executive Officer (CFO), Sumitomo Mitsui Trust Holdings, Inc.
Cultivate a future-focused perspective.
– Provide long-term guidance to your stakeholders and share the progress
toward achieving long-term goals.
– Structure executive compensation that’s tied to long-term performance
of your enterprise.
– Create dynamic, longer-term forecasts on future cash flows and educate
employees on how the market recognizes value creation over time.
View spending through a wide-angle lens.
– Invest in the initiatives (for example sustainability, generative AI) that align
with your long-term goals.
– Prioritize technology applications that accelerate the transition from piloting
to gaining efficiency to driving new growth.
– Evaluate and quantify the opportunity cost of borrowing from tomorrow to pay
for today.
Fund the future—flexibly.
– Introduce option theory into investment opportunities with contributions over
different future time periods.
– Avoid static capital allocation. Use a fluid portfolio with each investment
focused on outcomes.
– Conduct regular investment evaluations to drive capital redeployments
and resource reallocations.
30 IInnvveessttmmeenntt 3311
Case study
Edger Finance: Accelerating the
“So, it’s always a weighing up,
collection and analysis of investment
information with generative AI11
but if we can demonstrate that
a certain activity would bring
Edger, a fintech company, partnered with IBM to develop a pilot project using
value and massive efficiencies
generative AI to improve investment analysis and reporting for retail investors.
The pilot aimed to create a more efficient and personalized experience for
investors by collecting and reviewing investment data. in the mid to long run, then this
The tests conducted during the pilot demonstrated clear results and great
is the way to go. We would fight
potential for generative AI at Edger:
– 90% improvement in the turnaround time for quarterly report data extracts. for it.”
Whereas previously the process could take up to a week, the pilot
demonstrated that it could be accomplished in just four hours.
Martin Günther
– Approximately 96% improvement in the time it takes to summarize the main
CFO, smart Europe GmbH
points of a 30+ page report. Whereas previously it could take an analyst up
to half an hour to complete this task, the pilot demonstrated that it could
be accomplished in a matter of sec |
299 | ibm | 6-blind-spots-tech-leaders-must-reveal.pdf | IBM Institute for Business Value Global
C-suite
Series
30th Edition
Technology Leaders Study
6 blind spots tech
leaders must reveal
How to drive growth
in the generative AI era
Contents
2
Introduction
The end of business
as usual
7
Tech leaders outlook
About the study
In Q1 2024, in cooperation with Oxford Economics, the IBM Institute 11
We treat tech We say we are working We hope it will
for Business Value (IBM IBV) surveyed 2,500 C-suite technology leaders, The six blind spots
as an enabler but... together but... be a magic wand but...
including Chief Technology Officers (CTOs), Chief Information Officers
(CIOs), and Chief Data Officers (CDOs). Separately, a small group of Tech must be the core Our collaboration Generative AI could
executives was engaged for in-depth, qualitative interviews. These of everything we do. is only skin-deep. break our organization.
discussions focused on key insights from the study and the executives’
on-the-ground experience leading technology for organizations in the
new era of AI. With respondents spanning 26 industries and 34 locations
12 20 28
worldwide, this study marks a significant first look at a new technology
Innovation Leadership Infrastructure
coalition that is managing the enablement and delivery of AI capabilities
across the business. For more details, see “Research methodology and
analysis” on page 62.
We want it to be We talk about data We think our team
The cover concept and individual patterns in
trustworthy but... as currency but... is strong but...
this report were developed using generative AI.
Our AI may be Our data could We’re still fighting 61
IBM IBV designers translated each of the “blind spots” into prompts,
irresponsible. be a liability. yesterday’s talent Conclusion
and then used these prompts within Adobe Firefly to generate
battle.
vector-based imagery that inspired the basis and structure
62
for each pattern. Similarly, the photos that appear in this report
Research
were identified using AI-assisted, natural-language search,
methodology
using the generated patterns as reference images. 36 44 52 and analysis
AI Data Talent
Overall, the efficiency gained by integrating these
tools into the design process is as follows:
Concept—3 weeks to 1.5 days
Patterns—2 weeks to 2 days
Photography—1 week to 2 hours
Introduction
The end
of business
as usual
Elevating tech leadership
IT as a standalone function is dead. CEOs who say technology officers will be crucial decision-makers over the next
The rapid ascent of generative AI three years increased 50% since 2023.2 CFOs cite CTOs as the partners most
important to their success.3 To meet these expectations will demand a new
delivered the death knell.
approach to tech leadership.
Technology is the business.
And 72% of top-performing For technology to deliver enterprise-wide business outcomes, tech leaders must
“Business leaders are becoming be part mastermind, part maestro. They must architect technology strategy
CEOs say competitive advantage
across data, security, operations, and infrastructure, teaming with business
more tech savvy. When you have
depends on who has the most
leaders—speaking their language, not tech jargon—to understand needs, imagine
a discussion, they have a very good
advanced generative AI.1 possibilities, identify risks, and coordinate investments. They must build
understanding of what technology
That means organizations multidisciplinary teams to bring the strategy to life, encouraging the
can do. You have to be empathetic
experimentation and fresh ideas that inspire employees and delight customers.
are counting on tech leaders
to what they understand. It involves
as never before. It’s an enormous responsibility and one that many tech leaders have struggled
being much more versatile.”
to meet. As the scope of “technology” has expanded over the past two decades,
new roles have been added. But despite a growing team of technology leaders,
Bernd Bucher
Global Head Data, Digital, & IT/CIO, Novartis “technology” has not consistently and effectively been integrated into strategic
decision-making for the business (see Perspective, “Beyond the org chart:
A high-powered tech coalition” on page 6).
2 Introduction 3
Figure 1
Slippery slope
Our 2024 survey of 2,500 CIOs, CTOs, and CDOs suggests they are still being C-suite leaders agree that IT has become
left out of critical conversations. Their absence or ineffective participation has less effective at basic technology services
resulted in organizational blind spots in areas such as data, infrastructure, over the last 10 years.
talent, and innovation. While 43% of CEOs say they intend to increase the pace
of change for their organization this year,4 these blind spots are making it
difficult for organizations to seize today’s opportunities in artificial intelligence
in all its guises—traditional AI, gen AI, machine learning, and automation. Percent saying the IT organization is effective
at providing basic technology services
Our study also reveals that tech leaders are straining under the pressure. More
than half say they’re struggling to balance growth and productivity, and juggling 2013 Today 2013 Today 2013 Today
tasks is taking a toll on internal operations. Notably, the percentage of C-suite
69%
leaders who say their IT function is effective in delivering even basic services
has plummeted over the last decade (see Figure 1).
64%
We see in our results that when tech claims an equal seat at the C-suite table,
they can indeed steer significant outcomes (see “Tech outperformers crack the
60%
code to success”). But just as CEOs must face the hard truths outlined in our
2024 CEO study, tech leaders must courageously expose the blind spots that
are preventing their organizations from achieving AI advantage. In this report,
we discuss how these impediments can be overcome if tech executives
command the honest, must-have discussions about the readiness of their
50%
organization to deliver breakthrough innovation and business outcomes. The
47%
future is on the line. Tech leaders’ ability to insert their essential expertise into
enterprise decisions will ultimately determine their organizations’ success in
the AI era.
36%
“I do believe that technology teams are called to have CEOs CFOs Tech leaders
greater symbiosis with our business. The boundaries
between business and technology have become
increasingly blurred.” Source: Rate the effectiveness of your IT organization in providing basic technology services. Percentages
represent those CEOs, CFOs, and tech executives who responded effective or highly effective in IBM IBV 2013
and 2024 C-suite surveys. 2013 tech leaders is CIO-only data.
Alberto Rosa
CTO, CaixaBank
4 Introduction 5
Perspective Tech leaders outlook
Beyond the org chart: Tech outperformers crack
A high-powered tech coalition the code to success
Tech leaders juggle strategy, delivery, and support across
As technology permeates organizations, tech leadership roles have
data, security, operations, and infrastructure—all aimed
evolved, and new ones have emerged. But in an increasingly complex
at optimizing efficiency and competitiveness. Our research
operating environment where data, security, operations, and infrastructure
identified a high-performing group, comprising nearly 20%
are more integrated, business and technology teams must come together to
of our global sample, that excels in this mission.
deliver a cohesive set of experiences, capabilities, and outcomes. Remaining
in functional silos is no longer an option. CIOs, CTOs, and CDOs need to
reinvent how they work together toward their organizations’ shared business
goals, building bridges in support of shared ownership and accountability.
Four critical capabilities and characteristics
At the same time, tech executives still need to divide and conquer, set tech outperformers apart
focusing on their areas of expertise.
1 2 3 4
Chief Technology Officers Chief Information Officers Chief Data Officers
CTOs continue to battle the balance Amid shifting responsibilities, Data is no longer a domain unto
between security and innovation. CIOs question the effectiveness itself but the nerve center that Effective strategy Cross-functional A commitment A sharp eye on tech
Generative AI complicates an of the IT function. A remarkable connects technology to the broader development and collaboration to to measuring at all levels throughout
execution support tech outcomes and value the organization
already complex cyber threat 63% admit their tech organizations business and propels innovation.
investments
landscape, exacerbating tension are not very effective at leveraging For most organizations, a robust
between protecting what has been workflows and automation to drive data culture that can enable and Enabling a compelling Working with business Partnering with finance Maintaining keen
strategic vision that lines and finance to understand digital visibility into all
built and pushing the boundaries business strategy. But therein lies support AI operations is still
drives business to manage technology initiative value and IT at decentralized,
of what’s possible. Indeed, the opportunity. Winners are a work-in-progress. But taking
outcomes costs and budgets alignment with line-of-business,
cybersecurity ranks second transforming in-house functions an enterprise-wide view of the enterprise strategy geography, and
on CTOs’ priority list, behind product with the help of an augmented relationship between data and AI function levels
and service innovation. workforce where employees and AI operations is essential. “Classifying
The good news: core security combine to work smarter and faster. the data problem as a technology
practices—zero trust, secure problem is a bit unjust,” says
by design, DevSecOps—are still FuShan Hu, CIO at CHINT Group
the best defense. Co., Ltd. “It’s a comprehensive
problem and that’s why data
governance is so difficult.”
6 Introduction 7
Tech leaders outlook
Where do high-performing
tech leaders excel? “Deploying a generative AI Perspective
capability has to be done in
Going all in on cloud and AI
conjunction with complete
High-performing tech leaders
have significantly outpaced their Operating wholesale business
Today, tech leaders are prioritizing infrastructure
peers in annual revenue growth margin transformation … generative AI
+10% +16% investments, spending nearly one-third more on
and operating margin since 2020. alone won’t deliver the outcomes
hybrid cloud than AI. Looking ahead, they are fully
that a lot of CEOs are expecting.”
committed to the power of cloud and AI together.
2020–2022 2023
Annual revenue/ Over the next two years, tech leaders expect to
budget growth Mark Breslin
+21% +52% spend half their budget on the two combined.
Chief AI Officer, Informa PLC
Current Projected spend
Compared to peers, the spend over next 2 years
74%
high-performing tech leaders are
significantly more effective across
several key operational areas.
High-performing
“Technology today as a stand-alone
65%
tech leaders
function does not make sense;
All others technology is there to reimagine and
29%
power the business. And this requires
57% a much closer integration and 24%
56% collaboration with business leaders.”
Hybrid cloud
1. Rate the effectiveness of your
Mohammed Rafee Tarafdar
organization in delivering outcomes
CTO, Infosys
for productivity, cybersecurity and data
privacy, and product and service
innovation. Percentage reflects those 49%
who responded “effective” and 14%
“highly effective.” 12%
2. Extent you agree with statement: 45% Traditional AI
We have clear alignment with the enterprise
strategy across data, operations, 43%
technology, and security. Percentage 42% 6%
5%
reflects those who responded “to a large
extent” and “to a very large extent.”
Generative AI
Productivity1 Cybersecurity Product Alignment of tech
and data and service with enterprise
privacy1 innovation1 strategy2
8 Introduction 9
The six
Just as drivers are taught to identify blind spots to avoid crashes, tech leaders must recognize
blind spots
both when “objects in mirror are closer than they appear” and when risks may be hidden from
view entirely. Executives adept at navigating hazards safely, and at speed, can be the difference
between making technology the core of an organization’s competitive advantage and becoming
a wreck on the side of the road.
These six blind spots challenge 1 We treat tech as an enabler but...
longstanding assumptions about the Tech must be the core
relationship of technology and the business.
of everything we do.
Some risks may be closer than they appear,
even for the most sophisticated executives.
2
Tech leaders will need to look in their We say we are working together but...
mirrors and make a compelling case Our collaboration
to their C-suite peers for why these blind
is only skin-deep.
spots are holding their organizations back
in the quest for AI advantage.
3
We hope it will be a magic wand but...
Generative AI could break
our organization.
4 We want it to be trustworthy but...
Our AI may be
irresponsible.
5
We talk about data as currency but...
Our data could
be a liability.
6
We think our team is strong but...
We’re still fighting
yesterday’s talent battle.
10 11
We treat tech as an enabler but... CEOs have spoken: product and service innovation is their top priority
over the next three years. And 62% are willing to take more risks than
Tech must competitors to maintain an advantage.5
But tech leaders have a confession: only 43% say their technology
organizations are effective at delivering differentiated products
be the core of and services (see Figure 2). And to add salt to the wound, 53% say
other execs in their organization view tech as no more than moderately
important to product and service innovation. This disconnect between
technology and business suggests a massive change is needed.
everything we do.
It starts with tech leaders positioning technology as essential
to business outcomes. They say resistance to change among
management and employees are top barriers to innovation,
so tech leaders must amp up their outreach to the organization
When organizations see technology as an enabler, they treat
on what and how technology can deliver.
it like a toolbox. They wait for problems tech can solve rather
than exploring what new opportunities it creates. Only when More importantly, organizations need a fresh, bold approach
to innovation. A staggering 70% of tech executives say their
they recognize technology as the transformative force
organization is taking a fast-follower approach, adapting others’ ideas
at the core of innovation can organizations seize first-mover
or rolling out fixes rather than pioneering something radically new.
advantages, define markets, and gain economies of scale.
Shayan Hazir, Chief Digital Officer of HSBC Singapore, observes, “We
as technologists in financial services have tried to find problems for
technology to solve, but I don’t think we’re spending enough time
addressing what emerging technology can enable meaningfully for
“The biggest secret to digital transformation
customers, communities, and economies.”
is to change your perspective. It’s not about what
you can do, it’s about whether you can deliver value
to your customers in a rapidly changing environment.”
XiaoLong HE
CIO, VP of Digitalization, Tianshan Material Co., Ltd.
12 Innovation 13
Figure 2
Business awaits
Tech leaders are struggling to deliver
CEOs’ number one objective.
CEOs say product and To re-energize innovation for competitive gains, tech leaders must
service innovation is their look ahead for technology-fueled big bets. They need to shift from
top priority a project emphasis to a customer focus, prioritizing outcomes rather
than features as well as execution accompanied by customer
over the next 3 years
validation.6 They will need to avoid the ideation trap where many
get caught: 73% of business executives say their greatest strength
is researching customer needs or ideation, but only 27% say their
forte is executing or scaling product plans.7 Tech leaders must
quickly bring the ideas to life.
That requires them to evangelize a culture for innovation—one based
on pragmatic experimentation of high-potential ideas—and then
work to bring the rest of the C-suite on board. They can call on CFOs
to help define the most promising possibilities and to join them in
leading C-suite conversations about the importance of innovation to
the organization’s broader strategy. They need to encourage senior
leaders to look beyond near-term concerns such as efficiency, cost
but only 43%
takeout, and modest incremental gains.
of tech leaders say they are effective
or highly effective at delivering
product and service innovation
“We have this concept that we call open
innovation because we cannot do all
the innovation alone. Part of the work
is finding the right partners.”
Iosu Ibarbia
Technology Director, CAF (Construcciones y Auxiliar de Ferrocarriles)
111444 Innovation 15
“How do we leverage what’s good enough and push forward with What to do
it and then scale it? Traditional large organizations try to plan,
strategize, and build a solution. And by the time you finish it,
the technology and the landscape has changed.”
Escape the fast-follower treadmill by embracing
revolution, not perfection.
Jimmy Yeoh
CIO, DHL Express APEC
Jump from the treadmill onto the launchpad.
– Create urgency for meaningful action that disrupts the impulse
for incrementalism; pinpoint prudent precautions that encourage
more confident risk-taking.
– Identify critical business problems to be solved by blending tech and business
expertise on product and service development teams.
– Do the due diligence necessary to make leading practices real for your
organization and define an investment strategy that takes necessary resource
tradeoffs into account.
Break your analysis paralysis with generative AI.
– Use generative AI to synthesize customer feedback and analyze product usage
insights to accelerate meaningful iteration.
– Establish a framework for evaluating and ranking potential solutions with
generative AI. Ruthlessly cull efforts that don’t support your objectives.
– Develop KPIs to measure solution success and use generative AI to predict
outcomes and simulate scenarios.
Embrace a digital product innovation approach.
– Establish a digital product innovation framework for ideation, prototyping,
testing, and launch. Incorporate security and governance as design
considerations from the outset.
– Break down silos between tech and the business to enable rapid iteration that
delivers timely experiences and products to customers.
– Create incentives that reward experimentation and smart risk-taking for
solutions that improve productivity and innovation.
16 Innovation 17
Case study
IBM Software embraces
“We are now studying what kind of gen AI
gen AI for design8
use cases can have the greatest value
to customers. Once we figure out the
framework, and when we start to actually
develop something, then we can invite
IBM Software has defined an initiative around identifying the “top 10” some of our customers into the process.”
set of workflows in which it is actively embedding generative AI. The organization
Hiroshi Okuyama
is incorporating generative AI into products and processes, automating
Director and Member of the Board, Chief Digital Officer
workflows, improving output, and accelerating design.
Group Divisional Manager, Yanmar Holdings Co., Ltd.
IBM Software is also training 100% of their designers in AI. In general,
the designers find it invigorating to learn new skills and keep current with
cutting-edge AI technology—and they love the prospect of spending more time
on the creative aspects of their job that they’re passionate about.
“I’ve spent the last 12 to 18 months building an
In terms of synthesizing insights and crafting compelling content, IBM Design
enterprise-wide digital brain trust across our
has seen a 12% average daily time savings for content designers. In addition to
content design, the organization is investigating how to incorporate generative AI organization, bringing together multifaceted
across product management, UX design, content design, and research. teams that have been exceptional within their
own product category or technology area but
are creatives at heart. These people are now the
catalysts within their own business areas—
they’re the ambassadors of change. When they
go back to their day jobs, they infiltrate the
mindsets of their teams.”
Shayan Hazir
CDO, HSBC Singapore
18 Innovation 19
The AI race is just beginning—and while it may not be won over the next
two to three years, it can be lost over the next two to three quarters if
finance and tech executives fall out of sync. While CFOs complain that
tech decisions made in isolation by IT can lead to unsustainable costs,
We say we are working together but... tech leaders know that shortsighted technology decisions can wreak
long-term havoc. Their insights on technology are integral to their
Our collaboration organization’s strategic and financial decisions, while finance’s input
is critical to prioritizing technology investments.
A historically tense relationship must become more collaborative—not
is only skin-deep.
just through words but in deeds.9 Two-thirds of CEOs say that
a strong partnership between tech executives and CFOs is critical to
their organization’s success.10 Technology leaders agree—CIOs, CTOs,
and CDOs each rank the CFO as either the first or second most important
relationship for driving their individual success. But the tech-finance
While finance and technology have a history of working together,
relationship is still evolving from intention to practice. Only 39% of tech
that history masks critical gaps in planning processes
execs say they collaborate with finance to embed tech metrics into
and decisions that are disjointed or ill-informed. Only when business cases. Similarly, only 35% of CFOs say they’ve been engaged
the finance-tech relationship evolves from siloed to inseparable early in IT planning to set expectations on how technology advances
enterprise strategy.11
will they drive smarter decisions linking technology investments
to quantifiable business outcomes and improving ROI.
“There’s no such thing as the business
and IT. We’re all one team.”
Julia Knox
“We believe in cooperative leadership. Chief Technology and Analytics Officer, Sobeys
We build a leadership mindset that relies
on the collective intelligence of the team
rather than individuals.”
Moritz Hartmann
Global Head Roche Information Solutions, Roche Diagnostics
20 Leadership 21
“Technology decisions should be analyzed Figure 3
from a value perspective; what value
The tech-finance tango
will this decision bring to the business,
High-performing tech executives
the organization, and our clients.” are partnering with their finance peers
to align strategies and capture value.
Alberto Rosa
CTO, CaixaBank
However, our high-performing tech executives demonstrate the value All others Tech high performers
of building a strong rapport between tech and finance leaders.
They report notably stronger collaboration across key operational
practices (see Figure 3). Our analysis also shows that when
Apply learnings to improve
finance connects technology investments to quantifiable business 59%
future digital investments
outcomes, the high-performing group reports higher revenue growth.
57% Embed technology metrics
To drive organizational results like our top performers,
into business cases
tech leaders must pivot from informing to collaborating with
finance—recognizing how finance can supercharge tech’s influence
across the C-suite. They need to make themselves indispensable Engage early in IT planning to set
53%
to finance and demonstrate their commitment to fiscal responsibility. expectations on how technology
advances enterprise strategy
At the same time, finance leaders need to meet tech halfway,
looking beyond return on investment to understand how technology
contributes to operational outcomes. Both sides should see the
relationship as symbiotic, reinforcing mutual strengths so that
it’s greater than the sum of its parts.
46%
42%
35%
22 Leadership 23
“You need to be able to…collaborate What to do
more on the mid- and long-term
objectives and stick to the strategy.”
Align with finance to elevate your role as a strategic
Kristian Åkerström
xCIO/Head of IT & Digital, smart Europe collaborator and advisor.
Engage in aggressive diplomacy across the C-suite.
– Develop a deep understanding of the organization’s financial drivers and
leverage this knowledge to inform IT investment decisions.
– Identify and pursue ROI everywhere, including the financial and non-financial
measures that are essential to tracking business objectives.
– Agree on a shared approach to creating and evaluating new technology
investments for competitive advantage.
Make yourself indispensable to critical enterprise decisions.
– Seek opportunities to demonstrate the value of technical expertise
in enterprise decision-making processes and engage allies to ensure
your voice is recognized.
– Model your financial stewardship with a clear commitment to financial
transparency and accountability. Seek ways to recapture costs to fund
innovation efforts.
– Lead challenging organizational conversations, such as balancing
the intense energy consumption of AI against organizational sustainability
goals and commitments.
Show your work to build credibility.
– Frame technical discussions in financial terms, using data and analytics to
demonstrate the value of IT investments and drive strategic decision-making.
– Quantify operational metrics in monetary terms. Gain greater fluency in
financial performance metrics.
– Create a finance-facing dashboard that translates technology KPIs into
financial measures (such as cost per user, revenue per customer, ROI).
24 LLeeaaddeerrsshhiipp 2255
Case study
The Standard rationalizes cloud costs by aligning
IT spend with key business priorities12
Successful FinOps practices combined with Technology Business Management
(TBM) exemplify the budding synergy between finance and technology. The
disciplines of FinOps and TBM foster a collaborative culture that breaks down silos
so organizations can translate cloud and other technology investments into value.
The Standard, a leading provider of financial products and services, is realizing
the benefits of adopting these practices. Facing a lack of transparency on key
drivers of technology spending, the organization’s business and IT teams were
not working together efficiently. The company was relying on a legacy ERP
“I think that in the future, there will
system and spreadsheets to prepare the budget, analyze financial data, and make
decisions about technology investments—a manual and time-consuming process
be no essential contradiction between
that was prone to error.
The Standard implemented an IBM Apptio® solution to build cost transparency, Chief Technology Officers and
provide actionable insights, and enable faster decision-making. Adopting FinOps
and cloud governance practices alongside the Cloudability product gave the CFOs because they will both focus
company insights into its cloud spending—allowing it to drive greater accountability
by enhancing cloud procurement and provisioning decisions. In addition, on a common goal of the company’s
the Target process product helped the company improve its resource and program
management—aligning team workstreams to business priorities, gaining greater
successful future. I think they are,
visibility into consolidated workflows, and tracking dynamic variables like status,
stakeholders, dependencies, and progress.
for the most part, mutually supportive
The Standard has realized significant benefits. It has increased business/IT
and cooperative relationships.”
alignment and financial agility, with the IT Finance team now able to focus
80% of its time on analysis, decision support, forecasting, and insights.
The company has also gained more control over cloud spend, with projected
WeiWei Zhang
savings of 10% in 2023 and even more in 2024. Additionally, the company
CDO, Tianshan Material Co., Ltd.
improved its say:do ratio by 20%—a measure of the gap between what the IT
organization says it will do and what it actually delivers. The company plans
to continue investments in cloud governance to drive similar business results
across the organization.
2266 LLeeaaddeerrsshhiipp 2277
We hope it will be a magic wand but... Nearly three in four CEOs say their organizations’ digital infrastructure
enables new investments to efficiently scale and deliver value.13
Generative AI could But tech leaders have a different view. The scale and complexity of AI
demands an infrastructure that supports its voracious appetite for data,
compute, and storage. Only 16% of tech executives say they’re very
confident their current cloud and data capabilities are fully ready
break our organization.
to support generative AI.14 And 43% say their concerns about their
technology infrastructure have increased over the past six months
because of gen AI (see Figure 4).
Even more concerning: other IBM IBV research reveals that only 29%
Because organizations hope generative AI will solve all their
of cloud IT assets and services are performing as required.
problems, they ignore the added stress it places on their existing The remaining 71% is essentially tech debt accumulated over years
infrastructure, among other things. Only when they address their of piecemeal technology implementations.15 This burden is forcing
organizations to divert energy and resources toward maintaining
technical debt and transition from a patchwork of systems to
and troubleshooting outdated, disparate systems—not executing
a purpose-built technology foundation can organizations fully
bold ideas and future-focused initiatives.
embrace the shift from +AI to AI+.
“When you talk about the hardware and “When something suddenly becomes very
software stack, you are running into the issue important, but the foundation is not
of legacy things that you have to maintain. in place, then there’s a lot of internal
If you want to modernize it, it’s easy to say, but transformation we need to do to catch up.”
on the implementation side, it’s really difficult.”
Pochara Vanaratseath
Head of Information Technology Group, Krungsri Bank
Tawatchai Cheevanon
Chief Product and Business Solutions, Krung Thai Bank
28 Infrastructure 29
Figure 4 Tech leaders must tackle this weakness head-on, starting with a reality check
for other C-suite leaders. To catalyze AI transformation, organizations need
Unfit for AI
a thoughtful infrastructure renovation, repurposing what’s useful but also
Many organizations don’t
investing for the future. They need an architectural framework that helps
have an AI-ready technology
intentionally optimize business value through technology while addressing
infrastructure.
the entire technology estate: platforms, security, AI, cloud, and data. The goal
is to build a launchpad that brings together disparate technologies and can
support the business for years to come.16
Nearly
Daimler Trucks Group CIO Marcus Claesson recognizes the value of modernizing
three
architectures and operating models. Since Daimler Trucks spun off from
Mercedes-Benz, Claesson’s team has been rigorously rethinking and replacing
outdated technology and redefining how work gets done—not an easy
in four undertaking. “It’s like going to the gym. It’s difficult and painful,” he says.
“But we come out in better shape with a better foundation for the future
of the company.”
CEOs say their organizations’ digital As tech leaders ready for gen AI, infrastructure is wisely their top priority
infrastructure enables new investments investment. In fact, organizations are actually allocating more toward hybrid
to efficiently scale and deliver value. cloud than AI itself: 24% of their current spend versus 18% for traditional and
generative AI. As part of this focus, careful selection of cloud partners becomes
crucial to avoid risks such as vendor lock-in—a concern shared by two in three
But
tech leaders who are proactively identifying partner risks.
43%
An AI-optimized infrastructure isn’t a one-and-done proposition. Tech leaders
need to put in the work to align investments to business outcomes—with an eye
to minimizing the overhead associated with current technical debt and optimizing
of technology executives say generative AI existing resources and capacity to free up funds for AI innovation.
has increased their infrastructure concerns.
“You may deliver the technology, but if the business
is not ready or the business is |
300 | ibm | ceo-6-hard-truths-ceos-must-face.pdf | IBM Institute for Business Value
Global
C-suite
Series
29th Edition
CEO Study
6 hard truths
CEOs must face
How to leap forward with
courage and conviction
in the generative AI era
Contents
3
Introduction
The opportunity
paradox
7
The CEO outlook
About the study
This study represents the 29th edition of the IBM Institute for Business
Value (IBM IBV) C-suite Study series. For the 2024 CEO Study, IBM IBV, 13
in cooperation with Oxford Economics, conducted two rounds of survey- The six hard truths Your team isn’t as The customer Sentimentality is
strong as you think. isn’t always right. a weakness when
based interviews with more than 2,500 CEOs from 30+ countries and
expertise is in
26 industries. Conducted from December 2023 through April 2024, these
short supply.
conversations focused on business priorities, leadership, technology, talent,
partnering, regulation, industry disruption, and enterprise transformation.
14 20 26
Additional insights were drawn from ongoing IBM IBV research related to
Talent and skills Innovation Ecosystem
evolving technologies, including generative AI and hybrid cloud, and various partnerships
industries. Findings were also derived from numerous client interactions,
including more than two dozen deep-dive interviews with CEOs conducted
from July 2023 through April 2024.
Sparring partners People hate Tech short-cuts
The cover concept and individual patterns in
make the best progress. are a dead end.
this report were developed using generative AI.
leaders.
IBM IBV designers translated each of the “hard truths” into prompts,
and then used these prompts within Adobe Firefly to generate
vector-based imagery that inspired the basis and structure for
32 38 44 51
each pattern. Similarly, the photos that appear in this report were
Decision-making Vision and culture Transformation Conclusion
identified using AI-assisted, natural-language search, using the
When you’re on
generated patterns as reference images. a burning platform,
big risks are just
Overall, the efficiency gained by integrating these tools into
good business
the design process is as follows:
Concept—3 weeks to 1.5 days
52
Patterns—2 weeks to 2 days
Research and
Photography—1 week to 2 hours methodology
Introduction
The
opportunity
paradox
Is generative AI your wildest dream The risk is real, but sticking to the status quo isn’t any safer. As generative AI
or your worst nightmare? It depends throws everything into question, CEOs understand that they can’t stay the course
and stay in the race. More than two-thirds say the potential productivity gains
on how your organization reacts
from automation are so great that they must accept significant risk to remain
today—and prepares for tomorrow.
competitive—and 62% say they’ll take more risk than the competition to maintain
their competitive edge (see Figure 1).
Generative AI has the potential to
And it doesn’t stop there. Our 2024 CEO study reveals that:
shake up the way your business
has always worked, driving – 59% of all CEOs surveyed—and 72% of top-performing CEOs—agree that
unprecedented productivity and competitive advantage depends on who has the most advanced generative AI.
revealing new avenues for growth. – 72% of all CEOs see industry disruption as a risk rather than an opportunity.
“The more uncertainty you face, the
But those tremors could also crack – 62% say they will need to rewrite their business playbook to win in the future,
the foundation—and send everything more opportunities you have. In the rather than play to existing strengths.
past, eight out of 10 CEOs could get it
you’ve built crashing to the floor.
In this high-stakes environment, CEOs must strike the right balance between
right, but now only two CEOs can get
caution and courage—while moving faster than ever before. 43% say they’ll
it right. For the two CEOs who do it
increase the tempo of their organization’s transformational change in 2024,
right, the benefits are even greater.”
compared to just 19% that expect to slow down. As top leaders pick up the pace,
they need to unite disparate teams to deliver growth while also managing data
Chairperson, Industrial Manufacturing, China privacy concerns, legal liabilities, and technical complexity.
2 Introduction 3
Figure 1
Big risks, big rewards
The promise of generative AI
inspires CEOs to step out of
their comfort zones
Even if they don’t know exactly where they’re headed, CEOs have to
push their teams forward faster. Productivity gains and other quick wins
67%
are fueling this acceleration, but that’s just the beginning. CEOs that stop
here will miss out on the biggest part of the generative AI opportunity:
top-line growth.
say the potential productivity gains
Yet, 59% say they aren’t willing to sacrifice operational efficiency today to drive
from automation are so great that
greater innovation. CEOs also say focusing on short-term outcomes is the top
they must accept significant risk
to stay competitive. barrier to innovation. This suggests that many could fall into the trap of making
incremental improvements, instead of transforming critical operations. But if
CEOs open the aperture, generative AI can be the springboard they’re searching for.
They’ll have to make some important trade-offs. As the shelf-life of successful
business strategies continues to shrink, they’ll need to question old assumptions.
62%
That may mean exploring new business models, developing entirely new product
lines, bringing new partners into the fold—or saying good-bye to business
relationships that can’t drive new strategies forward.
say they will take more To make their wildest generative AI dreams reality, CEOs need to let go of
risk than the competition “what has always worked” and start tackling the hard truths holding them back.
to maintain their
For technology to transform the business, first the business must evolve.
competitive edge.
CEOs that settle for productivity
gains will miss out on the
biggest part of the generative
AI opportunity: top-line growth.
4 Introduction 5
The CEO outlook
How are leading
CEOs preparing for
an uncertain future?
CEOs across the board expect their investments to drive growth and
profitability. But those results don’t always materialize. So, what are
leading CEOs doing differently?
We’ve identified a group of CEOs, representing roughly 10% of our
global dataset, that are outperforming the competition despite global
disruption. Here are six critical capabilities and characteristics that
allow them to act with conviction even in the face of uncertainty.
“If someone else destroys our
old business model, we will be
Effective strategy Expertise-led Robust technology
development differentiation foundation
ruined. But if we destroy our old The executive leadership team Expertise informs business Digital infrastructure enables
crafts a compelling strategic vision decisions and gives the new investments to efficiently
to drive business outcomes. organization a competitive edge. scale and deliver value.
business model, we will survive.”
Nobuhiro Tsunoda
Chairperson, Ernst & Young Tax Co., Japan
Outcomes-focused Active ecosystem Actionable
investment engagement enterprise metrics
Technology leaders deliver The organization engages Enterprise data and KPIs
capabilities aligned with partners by delivering effectively set a clear bar for
enterprise strategy and priorities. industry-specific solutions success, which helps teams
beyond enterprise borders. achieve business objectives.
6 The CEO outlook 7
The CEO outlook
What sets High performance starts with financial metrics. CEOs in our leader group run
top-performing organizations that have outperformed the competition in annual revenue growth
and operating margin since 2020.
CEOs apart?
2020-2022 Outperformance metrics 2023
+ 16.4% Annual revenue/ + 17.7%
budget growth
20.4% 19.8%
+ Operating margin +
Leading CEOs also say their organization
outperforms in several key areas that Executing enterprise strategy1
deliver a competitive edge. 80
“As AI develops, there will
Leading CEOs
60 45% be three types of people:
All others
more
17% Talent
Innovation3
40 development those who create AI, those
more
and retention2
who use AI, and those who
20 43%
more
are used by AI.”
19%
more
Kazuhiro Nishiyama
President, Kansai Mirai Bank, Limited
23%
Partner/
ecosystem 22% more Cyber risk
development3 more and cybersecurity2
Technological maturity3
1. Rate the effectiveness of your organization in the following areas: Executing the enterprise strategy. Percentage reflects “effective” and
“highly effective.” 2. How does your organization’s performance compare to similar organizations over the past three years? Percentage
reflects “outperformed” and “significantly outperformed.” 3. How would your closest competitor rate your organization’s performance
compared to similar organizations? Percentage reflects “leading” and “significantly leading.”
8 The CEO outlook 9
The CEO outlook
CEO priorities Disruption is demanding CEOs to shift their focus. As new challenges come to the fore, CEOs have CEOs seek a rapid transition with generative AI, from piloting projects to increasing
and challenges are CEOs are prioritizing different strategic objectives and tapping quickly evolving technologies, big plans for efficiency to driving growth. Those who aren’t planning to transform quickly risk
changing rapidly including new forms of AI, to deliver business results. generative AI being left behind.
2023 2024
Not Piloting and Efficiency and Growth and
investing experimentation cost savings expansion
Top priorities Productivity or profitability 1 Product and service innovation
Tech modernization 2 Tech modernization Today
24% 47% 26% 3%
Customer experience 3 Cybersecurity and data privacy
2025
10% 20% 52% 18%
Cybersecurity and data privacy 4 Forecast accuracy
Environmental sustainability 5 Productivity or profitability 2026 13% 38% 49%
Product and service innovation 6 Customer experience
2029 3% 30% 67%
Note: Not all lines add up to 100% due to rounding.
Top challenges Environmental sustainability 1 Business model innovation
Cybersecurity and data privacy 2 Productivity or profitability
CEO concerns about generative AI
Tech modernization 3 Scalability of service delivery adoption are also changing as
capabilities mature.
More
Talent recruiting and retention 4 Marketing and sales effectiveness
concerned
Diversity and inclusion 5 Forecast accuracy
Data privacy
Less
Business model innovation 10
Data lineage concerned
Regulation Insufficient
Top technologies Cloud computing 1 Generative AI proprietary data
Improper use of
IoT, mobile, and connected devices 2 IoT, mobile, and connected devices intellectual property
Irrelevant use
Machine learning 3 Advanced analytics
“A good CEO can read the market and grasp cases
Advanced analytics 4 Data architecture the degree of tension, just like flying a kite,
loosening it when there is wind and pulling
Automation 5 Traditional AI
it when there is no wind.”
AI chatbots and natural 6 Hybrid cloud
language processing Chairperson, Industrial Manufacturing, China
10 The CEO outlook 11
The six
hard truths
“CEOs continue to manage that
creative tension between having
1
Here are six difficult realities CEOs must face— Your team isn’t as
a vision for the organization of the
from people challenges to operations hurdles to
strong as you think.
data and technology limitations—to outcompete
future while still being grounded in the age of generative AI.
2
in the realities of today.” The customer
isn’t always right.
Ngiam Siew Ying
CEO, Synapxe
3
Sentimentality is a
weakness when expertise
is in short supply.
4
Sparring partners
make the best
leaders.
5
People hate
progress.
6
Tech short-cuts
are a dead end.
12 13
Your team isn’t as CEOs understand that their people will make all the difference.
Already, 51% are hiring for generative AI-related roles that didn’t
exist last year. Yet, most say their organizations are straining under
the pressure. More than half say they’re already struggling to fill key
strong as you think.
technology roles—and it’s unlikely this task will get easier any time
soon. Overall, CEOs say 35% of their workforce will require retraining
and reskilling over the next three years—up from just 6% in 2021.
In a world where generative AI separates the Yet, they aren’t sure exactly what should change. Nearly two-thirds
say their teams have the skills and knowledge to incorporate
winners from the losers, people are a CEO’s
generative AI—and 67% say their recruiting and retention efforts
biggest technology problem. No matter how deliver the skills and expertise they need to achieve business
objectives, even as they face a talent shortage. A lack of data may be
good a team is today, it isn’t good enough to
causing this disconnect, as only 44% of CEOs say they’ve assessed
the impact of generative AI on their workforce (see Figure 2).
compete tomorrow.
Look for the people doing
tomorrow’s jobs today to
redefine how work should
be done.
“Talent is key to resilience. If I don’t have
talent that can anticipate and adapt,
absolutely nothing is going to happen.”
Fabián Hernández
CEO, Movistar Colombia
14 Talent and skills 15
Figure 2 “We must change our business model to
benefit from AI—and in the future, quantum
Connect the dots
computing—to recruit the best talent.”
Most CEOs are acting fast
on generative AI—but fewer Nobuhiro Tsunoda
understand its workforce Chairperson, Ernst & Young Tax Co., Japan
implications
Connecting the dots will be crucial in the coming year, given that
40% of CEOs plan to add staff because of generative AI. A larger
51%
portion (47%) expect to reduce their workforce because of
of CEOs say they’re generative AI, but they say the number of jobs created will exceed
currently hiring for generative AI-related the number of jobs lost overall. On average, they plan to increase the
roles that didn’t exist last year. workforce by nearly 6% over the next three years. As generative AI
continues to shake up how work is done, CEOs will need to rethink
how skills, experience, and job roles relate to each other to make the
most of this talent investment.
But only
44% The augmented workforce of the future promises to create more
value than people or machines can deliver alone, but you can’t plug
tomorrow’s talent into yesterday’s operating model. CEOs must
of CEOs have assessed the
identify the people doing tomorrow’s jobs today and tap their
impact of generative AI on
experience to define how work should be done in the future.1
their workforce.
“We have to have the best team for
today—but will it be the right team
for the future? We cannot be sure.
That’s why we need to reskill,
retool, and get people ready for
what is coming.”
Ngiam Siew Ying
CEO, Synapxe
16 Talent and skills 17
Reimagine how humans and What to do
machines can share the load.
Look beyond initial productivity gains to see how a new
division of labor—and an entirely new operating model—
could drive innovation and transformative growth.
Take a fresh look at your talent.
– Adopt a “day 1” mindset. If you wouldn’t hire your people today, identify what’s
missing and whether training can get them where they need to be.
– Identify forward-thinking talent that’s leading the change. Give these people
a platform to teach others.
– Accurately assess the cost associated with replacing talent that can’t adapt.
Compare this against the opportunity cost of stagnation—and act as quickly
as budgets will allow.
Boost creativity with a culture of curiosity.
– Cultivate human-tech chemistry by pairing people from different parts
of the organization to drive transformation initiatives.
– Redefine ways of working. Encourage experimentation with generative AI
tools and build in time for teams to share their learnings.
– Reward thoughtful risk-taking to set the tone. Use incentives to show
that, win or lose, experimenting with generative AI delivers value for
the organization.
Make people your most important tech investment.
– Analyze workforce data to determine where your organization has skills gaps
and define a timeline for closing them.
– Know when to buy, build, borrow, or bot. Assess where it makes sense to fill
the gap with employee training, targeted automation, or partner resources.
– Be prepared to spend more than you have in the past to hire for in-demand
skills.
18 Talent and skills 19
Generative AI can help companies tap into vast stores of customer data, from in-depth
market research to individual device metrics, to come up with paradigm-busting product
ideas. It can even validate far-out concepts against real-world business criteria, letting
employees focus on the creative work required to bring the best ideas to life. With these
The customer game-changing capabilities on the table, 86% of global digital product leaders say
generative AI is now a critical part of digital product design and development.2
However, this is only the starting point for true product innovation. Hitting the right mark
isn’t always right.
in a hyper-competitive consumer landscape will require more co-creation than companies
are used to. Rather than spending months designing and developing the perfect product or
experience, companies will need to prioritize speed to market—and fast feedback loops
that give customers a voice.
Customers don’t know what they’ll want tomorrow.
Generative AI can take some of the guesswork out of this process by making customer
It’s not that they’re indecisive—it’s that the next big feedback more accessible to product teams. According to recent IBM Institute for Business
Value (IBM IBV) research, only 30% of organizations are harnessing generative AI to quickly
thing could change everything.
analyze and summarize customer feedback to inform product design and development
today. But these early adopters already have an edge: they’re 86% more likely to be
Just as connected mobile devices have introduced
creating hyper-personalized experiences than their counterparts.3
must-have products that didn’t exist a decade ago,
Until recently, hyper-personalization at scale seemed like a pipe dream. But it’s quickly
generative AI could open the door to a new universe becoming reality with the help of generative AI. While only a quarter of organizations are
using generative AI to create hyper-personalized digital product experiences today, that
of opportunity. This may be why CEOs say product
figure is expected to more than double to 64% by the end of 2024.4
and service innovation is their top priority for the
next three years—up from sixth place in 2023.
“At smart Europe, we are super-fast, we are super
agile, we listen, and we change. As long as we show
“AI has a role in helping us advance to
customers that we’re taking their issues seriously and
provide better service to our customers.”
fix them quickly, they’re happy. People appreciate our
Javier Tamargo co-creation approach.”
CEO, 407 ETR
Dirk Adelmann
CEO, smart Europe GmbH
20 Innovation 21
Figure 3
Product wizardry
In this way, generative AI can make customer experiences magical. It can give
versus privacy invasion
customers exactly what they want before they’ve even thought to ask for it.
This instant gratification could be very addictive—as long as technology respects Open communication with customers
is essential to successfully deliver
people’s boundaries.
hyper-personalized products
To walk the line between product wizardry and privacy invasion, companies must
use customer data ethically and responsibly. Customers are willing to be wowed 80%
by hyper-personalization, but they want to know what’s happening behind the
curtain. For instance, a 2024 IBM IBV consumer study found that more than
half of consumers want to receive personalized information, advertisements, and
offerings from retailers, but roughly four in 10 want information about and control of CEOs say transparency around the
over how that data is being used.5 organization’s adoption of new technologies
is critical to fostering customer trust.
As hyper-personalized experiences become less fiction, more reality, CEOs know
they need to protect customer trust. Almost three in four agree that establishing
and maintaining customer trust will have a greater impact on their organization’s
success than any specific product or service features. And four in five say
transparency around the organization’s adoption of new technologies is
critical to fostering that trust (see Figure 3).
71%
“We tend to start with the business problem and what
we’re trying to accomplish, and then we look for say establishing and maintaining customer
technologies or innovations that can help us do that.” trust will have a greater impact on their
organization’s success than any specific
Judy McReynolds product or service.
CEO, ArcBest
22 Innovation 23
Design holistic experiences and What to do
hyper-personalize product development
while keeping an eye to the future.
Create dynamic experiences that incorporate continual
customer feedback and build trust through transparency.
Get more from your systems.
– Use technology to deliver superior experiences—but think beyond current
customer sentiment and expectations.
– Look beyond what customers say they want today to design the breakthrough
innovations of tomorrow.
– Use data and generative AI to identify new opportunities to move forward
rather than perfect the present.
Be transparent about how you use customer data.
– Make trusted data the backbone of your organization. Be upfront about what
data you’re collecting, how you’re using it, and why.
– Let customers share their data on their own terms. Explain how their data will
improve their experience and let them opt-in based on their personal priorities.
– Stay ahead of customers’ ethical expectations. Go beyond what’s required by
regulation to cultivate customer trust in your data policies.
Co-create products and experiences to increase
customer engagement.
– Set expectations up front for every interaction to make customers feel like
they’re being catered to, rather than spied on.
– Lead with design thinking. Use customer feedback to inform rapid iteration,
with generative AI suggesting and validating potential improvements.
– Use large language models to power hyper-personalized experiences, such
as curated product recommendations, tailored marketing messages, and
customized content.
24 IInnnnoovvaattiioonn 2255
Looking to the future, CEOs know they need to be selective about
which partners they prioritize. Nearly two-thirds say their
organization’s strategy is to concentrate on fewer high-quality
partners. This is perhaps to keep key vendors close at hand,
Sentimentality is as 60% expect critical expertise and capabilities to be increasingly
concentrated in a small cluster of organizations.
Striking the right mix between familiar faces and fresh ideas will be
a weakness when
crucial as CEOs push their teams to innovate. Today, more than half
say changing strategic priorities demand reconfiguring core business
partnerships. Yet, in the same breath, 76% say they have the right
expertise is in network of partners to execute their strategy today (see Figure 4).
short supply.
CEOs need to trust the partners they bring
to the table—and that trust can take years to
build. But valuing connections over capabilities
could be kryptonite for business leaders as they “It’s dangerous if we can’t have
heart-to-heart discussions with
jockey for a competitive edge with generative AI.
our partners about how we’re
positioned to navigate change—
and what will happen if things
“An enterprise must look at who it walks with. are left as they are.”
In the business ecosystem, you must work with
Kazuhiro Nishiyama
the best—otherwise you will be left behind.”
President, Kansai Mirai Bank, Limited
Chairperson, Industrial Manufacturing, China
26 Ecosystem partnerships 27
Figure 4
Recalibrating
relationships
CEOs expect to pivot their
partnerships as priorities change
While trust and shared values are central to
successful partnerships, CEOs must resist the
urge to cling to what’s comfortable as they
76%
navigate the winds of change. They won’t be able
to accelerate transformation if they keep investing
in an unproductive status quo.
of CEOs say they have the By assessing their organization’s strengths—and
right network of partners to deciding what must be done in-house—leaders
execute their strategy. can determine where to get external support.
While it may seem unnatural at first, CEOs will
need to cede control over non-essential aspects
of the business to focus more attention on what
matters most. With the right partners in the right
seats, CEOs can tap capabilities that were
previously out of reach.
But “You can’t be good at
55% everything. That’s why you
have to find partners—and
find a model that makes you
say changing strategic priorities comfortable working with
demand reconfiguring core
these partners.”
business partnerships.
Mikkel Hemmingsen
CEO, Sund & Bælt Holding A/S
28 Ecosystem partnerships 29
Ask for what you need—and don’t settle for less. What to do
Clearly define the outcomes you need from your partnerships
and what matters most to each player. Access relevant,
high-demand skills through ecosystem partnerships to
supplement the core capabilities you build in-house.
Ruthlessly cut dead weight to make room for new growth.
– Know what you value most. Don’t continue to invest in long-term partnerships
that are no longer producing results.
– Surround yourself with the best. Build a new relationship checklist and move
on from partners that don’t meet your standards.
– Ensure that your partners are aligned with your approach to AI ethics and
the guardrails that are in place.
Decide when and how you will let others take the wheel.
– Define—then clearly communicate—how much control you’re willing to cede,
as well as which capabilities you must keep in-house to control essential
operations.
– Trust the experts. You can’t be the best at everything, but you can benefit from
collaborating with specialists.
– Engage your ecosystem partners as full participants in technology innovation
and adoption.
Build symbiotic relationships.
– Cultivate the give-and-take. Create mutual dependency with your best
partners by investing time and resources to support their strategic goals.
– Take advantage of complementary strengths and perspectives to boost
foresight and resilience in the event of change.
– Clearly communicate what you need, what’s a deal-breaker, and what you’re
willing to compromise on.
30 Ecosystem partnerships 31
Just as sparring strengthens fighting skills, emphatic discussion leads to better
decisions, especially in times of uncertainty. But CEOs need to set clear ground
rules to keep these conversations constructive. If leaders believe no holds are
barred, debates can devolve into all-out brawls. These melees tend to be
Sparring partners counter-productive, with nearly half of top leaders saying competition among
their C-suite execs impedes collaboration from time to time.
However, conflict can also increase creativity, as clashes help leaders
make the best
find common ground. When leaders learn to speak each other’s languages—and
co-create shared strategies—they find inspired solutions to interconnected
business challenges. This will be crucial as technology transforms the business
leaders. landscape, with nearly two-thirds (65%) of CEOs saying their organization’s
success is directly tied to the quality of collaboration between finance and
technology functions (see Figure 5).
Over the next three years, CEOs will lean on COOs, CFOs, and CTOs to make
The C-suite shouldn’t always agree. Each officer
pivotal decisions. Technology leaders will need to set the bar for tech
comes to the table with their own perspective capabilities across the business, COOs must advise where technology can
make the biggest day-to-day impact, and CFOs will need to advise where finite
and area of expertise. No individual view offers
budgets should be spent. To make sure the organization benefits from the
objective truth. Rather, it’s the full picture they expertise of all its leaders, not just the ones who shout the loudest, CEOs will
need to set clear cultural parameters around how decisions are made.
paint together that helps CEOs decide which
direction the organization should take.
When leaders learn to speak
each other’s languages, they find
“If a senior management team completely
inspired solutions to interconnected
excludes the exchange and collision of views
and opinions, the team is not creative.” business challenges.
Chairperson, Industrial Manufacturing, China
32 Decision-making 33
Perspective
Figure 5 Different corners,
different views
Rules of engagement
C-suite officers have different perspectives on how
CEOs must foster a culture that
to measure progress—and what’s holding innovation
encourages emphatic debate and
back—based on where they sit in the organization.
constructive collaboration
CEOs CFOs Tech CxOs
65%
Barriers to Short-term Management Regulatory
of CEOs say their organization’s
focus resistance to change constraints
innovation
success is directly tied to the quality
of collaboration between finance and Regulatory Aversion to risk Inadequate
constraints technology or data
technology functions.
Employee resistance Limited budget Management
48% to change resistance to change
Measures Organizational Financial benefits Innovation
say competition within their
digital maturity maturity
of enterprise
C-suite sometimes
impedes collaboration. transformation Cybersecurity Risk exposure Cybersecurity
maturity maturity
Technology Project progress Customer
adoption experience
“The more you specialize and
divide a process into parts, the
more you have to create some kind
of dependency between the parts.”
Mikkel Hemmingsen
CEO, Sund & Bælt Holding A/S
34 Decision-making 35
Build a C-suite that can lead with conviction. What to do
Generative AI changes what you can do—but it shouldn’t
change who you are. Reinforce a clear and compelling vision
to prioritize new opportunities and align transformation efforts
across the organization.
Define rules of engagement and emphasize expertise.
– Use consistent data, establish clear governance, and define desired outcomes.
– Set ground rules for healthy debate and build constructive tension to spark
growth and innovation.
– Highlight where it’s critical to speak a common language and where individual
expertise is essential.
Break down barriers between IT and the business.
– Surface conflicting expectations around critical paths and timelines.
– Stop measuring business and IT goals separately.
– Prioritize IT projects with the strongest links to business value.
Restructure the C-suite for success.
– Create a clear decision-making matrix. Give leaders clear guidance about who
has authority in which area.
– Align rewards and incentives to encourage debate on the right topics.
– Actively encourage the inclusion of different expert opinions while clearly
defining when a decision has been made or where you need quick consensus.
36 Decision-making 37
CEOs see the people problem that generative AI is creating. Nearly
two-thirds (64%) say their organization must take advantage of
technologies that are changing faster than employees can adapt—and
61% say they’re pushing their organization to adopt generative AI
People hate more quickly than some people are comfortable with.
Part of the issue is that many people think they’re training their
replacement. Despite the fact that business leaders consistently say
progress.
this technology will support human employees—not replace them—
employees remain skeptical. Until they’re convinced, they won’t take
the initiative to rethink how work is done.
To get people on board, organizations will have to invest in training
Generative AI promises to bring opportunities that
that will help them see generative AI in a new light. If they
were once pure fantasy into the realm of possibility. understand how this technology can make their jobs easier—and
more rewarding—organizations could see a major uptick in adoption.
But moving beyond productivity gains to business
Most CEOs know that making the most of generative AI will require
model innovation will require buy-in at all levels developing technology and people in equal portion, with nearly
two-thirds saying success will depend more on people’s adoption
of the organization—and many employees see
than the technology itself (see Figure 6).
generative AI as something that’s happening
CEOs also need to help people connect the dots between strategy,
TO them, not a tool that works FOR them. governance, and security as transformation continues to accelerate.
They’ll need to create thoughtful guardrails—not processes |
302 | ibm | IBM_Annual_Report_2023.pdf | Let’s Create
2023 Annual Report
Dear IBM Investor:
In 2023, we made significant
progress in our journey to become
a more innovative and focused
company, built around the two most
transformational technologies of
our time: hybrid cloud and AI. We
executed against a proven strategy,
refined our portfolio, expanded our
ecosystem of partners, and enhanced
Arvind Krishna productivity throughout IBM.
Chairman and Chief Executive Officer
We also continued to address the evolving needs of our clients.
As AI becomes a top priority, our clients are using watsonx
– IBM’s flagship AI and data platform – to help revolutionize
customer service, modernize countless lines of code, and
automate enterprise tasks to boost employee productivity.
I have never been more confident in IBM’s direction. Today’s
IBM is more capable and more productive. We have a strong
portfolio and a solid foundation to support sustainable growth.
And we are delivering on our promise to be the catalyst that
makes the world work better.
2023 performance
For the year, IBM generated $61.9 billion in revenue, up 3%
at constant currency, and $11.2 billion of free cash flow, up
$1.9 billion year-over-year. We experienced growing demand
for our new watsonx platform, marked by thousands of client
interactions. This demand contributed to roughly doubling the
book of business for watsonx and generative AI from the third
to the fourth quarter.
IBM 2023 Annual Report 1
Software Consulting Infrastructure
We also expanded profit margins by emphasizing high- Technology and expertise
value offerings in Consulting and Software and by digitally AI and hybrid cloud continue to drive value creation, allowing
transforming our processes and scaling AI to enhance businesses to scale, increase productivity, and seize new
productivity within IBM. market opportunities. IBM has built two powerful platforms
to capitalize on the strong demand for both technologies:
Software revenues were up more than 5% at constant watsonx for AI, and Red Hat OpenShift for hybrid cloud.
currency, as clients turned to our advanced software
capabilities across hybrid cloud, data & AI, automation, Watsonx is our comprehensive AI and data platform, built to
transactions processing, and security. Our performance was deliver AI models and give our clients the ability to manage
led by Red Hat, and we had solid growth in our recurring the entire lifecycle of AI for business, including the training,
revenue base. tuning, deployment, and ongoing governance of those models.
As clients shift from experimenting with generative AI to
Consulting revenues were up 6% at constant currency. building and deploying it throughout their enterprises, we are
We capitalized on the growing need for expertise in focused on practical and urgent business use cases, including
digital transformation and AI deployment, leveraging our code modernization, customer service, and digital labor.
consulting services in data and technology consulting,
cloud modernization, application operations, and business Financial institutions like Citi, Bradesco, and NatWest
transformation. are using watsonx to help increase productivity, improve
code quality, and enhance customer experiences. Our
Infrastructure revenues decreased by 4% at constant enterprise-ready AI capabilities are being embedded into
currency, in line with the typical product cycle dynamics in SAP solutions. EY launched EY.ai Workforce, a new solution
this segment. IBM z16 is significantly outperforming previous that will use watsonx Orchestrate to automate HR tasks and
cycles, demonstrating the enduring value this platform processes. Service partners such as NTT Data Business
provides to our clients. Solutions, Wipro, and TCS are launching watsonx Centers
of Excellence to scale AI-powered client innovations. And
IBM’s revenue growth and cash generation enabled us to generative AI from watsonx, combined with expertise from
make substantial investments in the business and deliver Consulting, is enhancing the digital experiences of the U.S.
value to our shareholders. In 2023, IBM spent nearly $7 Open, the Masters, Wimbledon, the GRAMMYs, and ESPN
billion on research and development, more than $5 billion to Fantasy Football.
acquire nine companies, and returned more than $6 billion to
stockholders through dividends.
2
IBMers are also embracing watsonx to unleash greater along with new machine learning, intelligence, and operational
productivity, eliminate complexity, simplify workflows, and improvements for z/OS.
automate manual tasks. Examples include processing HR and
IT tasks more easily, generating code up to 60% faster, and In addition, we enhanced IBM’s portfolio with nine
answering client inquiries more quickly. acquisitions in 2023, including Apptio, a suite of software to
help our clients better understand their technology investment
Hybrid cloud architectures have seen massive adoption, with and the business value it delivers.
nearly 80% of IT decision makers operating hybrid cloud
environments. But nearly two thirds of companies report Client engagement and partnership
difficulty managing these complex environments, a challenge IBM’s success is directly tied to the success of our clients.
that will grow as businesses deploy generative AI across Their problems are our problems. And their opportunities
multiple clouds. IBM’s industry-leading hybrid cloud platform, are our opportunities. That is why we developed a more
based on Red Hat OpenShift, can solve this problem. It collaborative, experience-based approach that allows us to
helps our clients move from architectures that are hybrid by respond effectively to their needs.
default to architectures that are hybrid by design. It enables
companies to run workloads seamlessly across multiple The IBM Garage Method, now integrated across our business,
clouds, both public and private, to simplify operations, unify combines agile development and design thinking to facilitate
data and applications, and accelerate new innovations. And co-creation with our clients. Clients have embraced this highly
it complements our watsonx platform, allowing clients the collaborative way of working with IBM, turning ideas into
flexibility to manage multi-model AI across complex, multi- outcomes with thousands of Garage engagements throughout
cloud environments. the year.
Virgin Money is harnessing IBM’s hybrid cloud to enable new Our approach to client engagement allows us to meet clients
digital customer experiences and improve their credit card where they are, bringing together whatever technology
services. Red Hat OpenShift is now the preferred platform and expertise are needed across our expanding partner
provider to Nokia’s core network applications business. ecosystem. That is why we strengthened our strategic
And the Boston Red Sox are leveraging our hybrid cloud partnerships with key industry players like Adobe, AWS,
technologies to improve the club’s operations. Microsoft, SAP, Salesforce, Samsung, and others. Strategic
partnerships now make up more than 40% of our Consulting
Experts from Consulting provide differentiated value as we revenue and delivered double-digit growth in both signings
establish IBM as a leader in AI for business, just as they did and revenue for the year.
with our hybrid cloud business. Our extensive network of
data and AI consultants has already facilitated thousands of Research and development
hands-on client interactions. IBM combines technology with In 2023, IBM Research advanced the fundamental science
consulting services to deliver the data architecture, security, of several critical technologies, including AI, quantum
and governance our clients need to adopt trusted AI solutions. computing, and semiconductors.
IBM consultants are working with Riyadh Air on mission- In AI, we demonstrated our ability to quickly transform
critical technology and business capabilities to support the research into commercial applications. We launched the
path to their first flight. NATO chose IBM to help detect and watsonx AI and data platform, introduced the groundbreaking
respond to cyber threats with greater speed. And Diageo Granite AI foundational model, and developed new AI-
partnered with Consulting and SAP on an ambitious five-year optimized hardware.
business transformation and cloud migration.
We have IBM Quantum System One engagements with several
Throughout 2023, clients modernized their infrastructure with leading organizations, including Cleveland Clinic, the Platform
the z16 platform in alignment with their hybrid cloud and AI for Digital and Quantum Innovation of Quebec, Rensselaer
strategies. IBM launched a new suite of AI offerings for IBM Z Polytechnic Institute, and the University of Tokyo. We also
IBM 2023 Annual Report 3
unveiled our 133-qubit Quantum Heron processor, which technology ethics by 2025. And IBM committed to training
enhanced the performance, efficiency, and scalability of the two million learners in AI by the end of 2026 to address the
newly deployed IBM Quantum System Two. And our work on technology skills gap.
error correction and mitigation is helping to lay the foundation
for a new era of quantum utility. But IBM’s commitment to trust goes beyond our citizenship,
products, and policies. We earn trust by delivering on our
Research also pushed the limits of semiconductor design and promises.
packaging, building on recent innovations such as the 2nm
node chip, hybrid bonding, and vertical transistors. We are We articulated a clear vision for the future of IBM in the
working with Rapidus to propel Japan’s push for leadership spring of 2020. We promised a more focused company
in semiconductor research and manufacturing, and we are built around two powerful technologies: hybrid cloud and
participating in an initiative with New York State, Micron, and AI. We promised fundamental changes to our go-to-market
others to jointly invest $10 billion in semiconductor R&D. strategy, putting clients at the center of everything we do and
transforming competitors into partners. And we promised
The promise of IBM operational changes to simplify our internal processes and
IBM is in the business of shaping the future for our clients. increase our productivity. As this report details, we are
That future must be built on trust. fulfilling those promises.
IBM is at the forefront of technologies, like AI and quantum As we look ahead, we renew our commitment to the journey
computing, which will fundamentally change the way we work we began in 2020. We will continue to innovate, to execute
and live. We bear significant responsibility to develop those with speed and purpose, find more opportunities for
technologies ethically and deploy them with transparency and operational efficiency, and further enhance our productivity
trust. That is why we built powerful AI governance into our by employing the same technologies we use to drive growth
watsonx platform and developed quantum-safe cryptography for our clients. And as always, we will be the catalyst that
to secure sensitive data. It is why we advocate for smart AI makes the world work better, bringing together our colleagues,
regulation, including holding those who develop and deploy clients, and partners with a simple invitation: Let’s Create.
AI accountable for fraudulent, discriminatory, and harmful
activity. And it is why IBM and Meta announced the formation This is the promise of IBM.
of the AI Alliance, a group of more than 70 organizations
dedicated to advancing open, safe, and responsible AI.
We also earn trust by operating with integrity, staying true
to our values, and addressing the needs of all stakeholders.
We continue to advance our efforts on the environment, Arvind Krishna
ethics, and education. IBM has achieved a 63% reduction Chairman and Chief Executive Officer
in greenhouse gas emissions against base year 2010.
We announced a new program to train 1,000 suppliers in
In an effort to provide additional and useful information regarding the company’s financial results and other financial information, as determined by generally
accepted accounting principles (GAAP), these materials contain non-GAAP financial measures on a continuing operations basis, including revenue growth
rates adjusted for constant currency and free cash flow. The rationale for management’s use of this non-GAAP information is included on page 6 and 31 of
the company’s 2023 Annual Report, which is Exhibit 13 to the Form 10-K submitted with the SEC on February 26, 2024. For reconciliation of these non-GAAP
financial measures to GAAP and other information, please refer to pages 17 and 31 of the company’s 2023 Annual Report. For watsonx and generative AI, book of
business includes Software transactional revenue, SaaS Annual Contract Value and Consulting signings.
4
Report of Financials 5
International Business Machines Corporation and Subsidiary Companies
MANAGEMENT DISCUSSION NOTES TO CONSOLIDATED FINANCIAL STATEMENTS
Overview 6 Basis & Policies
Forward-Looking and Cautionary Statements 7 A Significant Accounting Policies 50
Management Discussion Snapshot 8 B Accounting Changes 63
Description of Business 11 Performance & Operations
Year in Review 17 C Revenue Recognition 64
Prior Year in Review 28 D Segments 66
Other Information 29 E Acquisitions & Divestitures 71
Looking Forward 29 F Other (Income) and Expense 78
Liquidity and Capital Resources 30 G Research, Development & Engineering 78
Critical Accounting Estimates 33 H Taxes 78
Currency Rate Fluctuations 36 I Earnings Per Share 82
Market Risk 36 Balance Sheet & Liquidity
Financing 38 J Financial Assets & Liabilities 83
K Inventory 84
Report of Management 41 L Financing Receivables 84
Report of Independent Registered M Property, Plant & Equipment 87
Public Accounting Firm 42 N Leases 87
O Intangible Assets Including Goodwill 90
CONSOLIDATED FINANCIAL STATEMENTS P Borrowings 91
Income Statement 44 Q Other Liabilities 94
Comprehensive Income 45 R Commitments & Contingencies 95
Balance Sheet 46 S Equity Activity 97
Cash Flows 47 Risk Management, Compensation/Benefits & Other
Equity 48 T Derivative Financial Instruments 100
U Stock-Based Compensation 104
V Retirement-Related Benefits 107
W Subsequent Events 121
Performance Graphs 122
Stockholder Information 123
Board of Directors and Senior Leadership 124
6 Management Discussion
International Business Machines Corporation and Subsidiary Companies
OVERVIEW
The financial section of the International Business Machines Corporation (IBM or the company) 2023 Annual Report includes the
Management Discussion, the Consolidated Financial Statements and the Notes to Consolidated Financial Statements. This Overview
is designed to provide the reader with some perspective regarding the information contained in the financial section.
Organization of Information
• The Management Discussion is designed to provide readers with an overview of the business and a narrative on our financial
results and certain factors that may affect our future prospects from the perspective of management. The “Management
Discussion Snapshot” presents an overview of the key performance drivers in 2023.
• Beginning with the "Year in Review," the Management Discussion contains the results of operations for each reportable segment
of the business, a discussion of our financial position and a discussion of cash flows as reflected in the Consolidated Statement of
Cash Flows. Other key sections within the Management Discussion include: "Looking Forward" and "Liquidity and Capital
Resources," the latter of which includes a description of management's definition and use of free cash flow.
• The Consolidated Financial Statements provide an overview of income and cash flow performance and financial position.
• The Notes follow the Consolidated Financial Statements. Among other items, the Notes contain our accounting policies, revenue
information, acquisitions and divestitures, certain commitments and contingencies and retirement-related plans information.
• On November 3, 2021 we completed the separation of our managed infrastructure services unit into a new public company,
Kyndryl. The accounting requirements for reporting the separation of Kyndryl as a discontinued operation were met when the
separation was completed. Accordingly, the historical results of Kyndryl are presented as discontinued operations and, as such,
have been excluded from continuing operations and segment results for all periods presented. Refer to note E, “Acquisitions &
Divestitures,” for additional information.
• In September 2022, the IBM Qualified Personal Pension Plan (Qualified PPP) purchased two separate nonparticipating single
premium group annuity contracts from The Prudential Insurance Company of America and Metropolitan Life Insurance Company
(collectively, the Insurers) and irrevocably transferred to the Insurers approximately $16 billion of the Qualified PPP’s defined
benefit pension obligations and related plan assets, thereby reducing our pension obligations and assets by the same amount.
The group annuity contracts were purchased using assets of the Qualified PPP and no additional funding contribution was
required from IBM. The transaction resulted in no changes to the benefits to be received by the plan participants. As a result of
this transaction we recognized a one-time, non-cash, pre-tax pension settlement charge of $5.9 billion ($4.4 billion net of tax) in
the third quarter of 2022, primarily related to the accelerated recognition of accumulated actuarial losses of the Qualified PPP.
Refer to note V, “Retirement-Related Benefits,” for additional information.
• Effective January 1, 2023, due to advances in technology, we increased the estimated useful lives of our server and network
equipment from five to six years for new assets and from three to four years for used assets. Based on the carrying amount of
server and network equipment included in property, plant and equipment-net in our Consolidated Balance Sheet as of
December 31, 2022, the effect of this change in accounting estimate was an increase in income from continuing operations
before income taxes of $208 million or $0.18 per basic and diluted share for the year ended December 31, 2023.
• In 2023, we executed workforce rebalancing actions to address remaining stranded costs from portfolio actions over the last
couple of years resulting in charges to pre-tax income from continuing operations of $438 million. In addition, beginning in the
first quarter of 2023, we updated our measure of segment pre-tax income to no longer allocate workforce rebalancing actions to
our reportable segments, consistent with our management system. Workforce rebalancing charges in 2022 and 2021 of
$40 million and $182 million, respectively, were included in the segments.
• The references to “adjusted for currency” or “at constant currency” in the Management Discussion do not include operational
impacts that could result from fluctuations in foreign currency rates. When we refer to growth rates at constant currency or adjust
such growth rates for currency, it is done so that certain financial results can be viewed without the impact of fluctuations in
foreign currency exchange rates, thereby facilitating period-to-period comparisons of business performance. Financial results
adjusted for currency are calculated by translating current period activity in local currency using the comparable prior-year
period’s currency conversion rate. This approach is used for countries where the functional currency is the local currency.
Generally, when the dollar either strengthens or weakens against other currencies, the growth at constant currency rates or
adjusting for currency will be higher or lower than growth reported at actual exchange rates. Refer to “Currency Rate
Fluctuations” for additional information.
• Within the financial statements and tables in this Annual Report, certain columns and rows may not add due to the use of
rounded numbers for disclosure purposes. Percentages presented are calculated from the underlying whole-dollar numbers.
Certain prior-year amounts have been reclassified to conform to the change in current year presentation. This is annotated where
applicable.
Management Discussion 7
International Business Machines Corporation and Subsidiary Companies
Operating (non-GAAP) Earnings
In an effort to provide better transparency into the operational results of the business, supplementally, management separates
business results into operating and non-operating categories. Operating earnings from continuing operations is a non-GAAP
measure that excludes the effects of certain acquisition-related charges, intangible asset amortization, expense resulting from basis
differences on equity method investments, retirement-related costs, certain impacts from the Kyndryl separation and their related
tax impacts. Due to the unique, non-recurring nature of the enactment of the U.S. Tax Cuts and Jobs Act (U.S. tax reform),
management characterizes the one-time provisional charge recorded in the fourth quarter of 2017 and adjustments to that charge
as non-operating. Adjustments primarily include true-ups, accounting elections and any changes to regulations, laws, audit
adjustments that affect the recorded one-time charge. Management characterizes direct and incremental charges incurred related
to the Kyndryl separation as non-operating given their unique and non-recurring nature. In 2022, these charges primarily related to
any net gains or losses on the Kyndryl common stock and the related cash-settled swap with a third-party financial institution,
which were recorded in other (income) and expense in the Consolidated Income Statement. As of November 2, 2022, the company
no longer held an ownership interest in Kyndryl. For acquisitions, operating (non-GAAP) earnings exclude the amortization of
purchased intangible assets and acquisition-related charges such as in-process research and development, transaction costs,
applicable retention, restructuring and related expenses, tax charges related to acquisition integration and pre-closing charges,
such as financing costs. These charges are excluded as they may be inconsistent in amount and timing from period to period and are
significantly impacted by the size, type and frequency of our acquisitions. Given its unique and temporary nature, management has
also characterized as non-operating expense, the mark-to-market impact on the foreign exchange call option contracts to
economically hedge the foreign currency exposure related to the purchase price of our announced acquisition of StreamSets and
webMethods from Software AG. The mark-to-market impact is recorded in other (income) and expense in the Consolidated Income
Statement and reflects the fair value changes in the derivative contracts. All other spending for acquired companies is included in
both earnings from continuing operations and in operating (non-GAAP) earnings. For retirement-related costs, management
characterizes certain items as operating and others as non-operating, consistent with GAAP. We include defined benefit plan and
nonpension postretirement benefit plan service costs, multi-employer plan costs and the cost of defined contribution plans in
operating earnings. Non-operating retirement-related costs include defined benefit plan and nonpension postretirement benefit
plan amortization of prior service costs, interest cost, expected return on plan assets, amortized actuarial gains/losses, the impacts
of any plan curtailments/settlements including the one-time, non-cash, pre-tax settlement charge of $5.9 billion ($4.4 billion, net of
tax) in the third quarter of 2022 and pension insolvency costs and other costs. Non-operating retirement-related costs are primarily
related to changes in pension plan assets and liabilities which are tied to financial market performance, and we consider these costs
to be outside of the operational performance of the business.
Overall, management believes that supplementally providing investors with a view of operating earnings as described above
provides increased transparency and clarity into both the operational results of the business and the performance of our pension
plans; improves visibility to management decisions and their impacts on operational performance; enables better comparison to
peer companies; and allows us to provide a long-term strategic view of the business going forward. In addition, these non-GAAP
measures provide a perspective consistent with areas of interest we routinely receive from investors and analysts. Our reportable
segment financial results reflect pre-tax operating earnings from continuing operations, consistent with our management and
measurement system.
FORWARD-LOOKING AND CAUTIONARY STATEMENTS
Certain statements contained in this Annual Report may constitute forward-looking statements within the meaning of the Private
Securities Litigation Reform Act of 1995. Any forward-looking statement in this Annual Report speaks only as of the date on which it
is made; IBM assumes no obligation to update or revise any such statements except as required by law. Forward-looking
statements are based on IBM’s current assumptions regarding future business and financial performance; these statements, by
their nature, address matters that are uncertain to different degrees. Forward-looking statements involve a number of risks,
uncertainties and other factors that could cause actual results to be materially different, as discussed more fully elsewhere in this
Annual Report and in the company’s filings with the Securities and Exchange Commission (SEC), including IBM’s 2023 Form 10-K
filed on February 26, 2024.
8 Management Discussion
International Business Machines Corporation and Subsidiary Companies
MANAGEMENT DISCUSSION SNAPSHOT
($ and shares in millions except per share amounts)
Yr.-to-Yr.
Percent/Margin
For year ended December 31: 2023 2022 (1) Change
Revenue (2) $ 61,860 $ 60,530 2.2 %
Gross profit margin 55.4 % 54.0 % 1.4 pts.
Total expense and other (income) $ 25,610 $ 31,531 (18.8) %
Income from continuing operations before income taxes $ 8,690 $ 1,156 NM
Provision for/(benefit from) income taxes from continuing operations $ 1,176 $ (626) NM
Income from continuing operations $ 7,514 $ 1,783 NM
Income from continuing operations margin 12.1 % 2.9 % 9.2 pts.
Loss from discontinued operations, net of tax $ (12) $ (143) (91.8) %
Net income $ 7,502 $ 1,639 NM
Earnings per share from continuing operations–assuming dilution $ 8.15 $ 1.95 NM
Consolidated earnings per share–assuming dilution $ 8.14 $ 1.80 NM
Weighted-average shares outstanding–assuming dilution 922.1 912.3 1.1 %
Assets (3) $ 135,241 $ 127,243 6.3 %
Liabilities (3) $ 112,628 $ 105,222 7.0 %
Equity (3) $ 22,613 $ 22,021 2.7 %
(1)Includes a one-time, non-cash, pre-tax pension settlement charge of $5.9 billion ($4.4 billion net of tax) resulting in an impact of ($4.84) to diluted
earnings per share from continuing operations and an impact of ($4.83) to consolidated diluted earnings per share. Refer to note V, “Retirement-
Related Benefits,” for additional information.
(2)Year-to-year revenue growth of 2.9 percent adjusted for currency.
(3)At December 31.
NM–Not meaningful
The following table provides the company’s operating (non-GAAP) earnings for 2023 and 2022. Refer to page 28 for additional
information.
($ in millions except per share amounts)
Yr.-to-Yr.
For year ended December 31: 2023 2022 Percent Change
Net income as reported (1) $ 7,502 $ 1,639 NM
Loss from discontinued operations, net of tax (12) (143) (91.8) %
Income from continuing operations (1) $ 7,514 $ 1,783 NM
Non-operating adjustments (net of tax)
Acquisition-related charges 1,292 1,329 (2.8) %
Non-operating retirement-related costs/(income) (1) (30) 4,933 NM
U.S. tax reform impacts 95 (70) NM
Kyndryl-related impacts — 351 (100.0) %
Operating (non-GAAP) earnings $ 8,870 $ 8,326 6.5 %
Diluted operating (non-GAAP) earnings per share $ 9.62 $ 9.13 5.4 %
(1)2022 includes a one-time, non-cash pension settlement charge of $4.4 billion net of tax.
NM–Not meaningful
Management Discussion 9
International Business Machines Corporation and Subsidiary Companies
Macroeconomic Environment
Our business profile positions us well in challenging macroeconomic times. Our diversification across geographies, industries,
clients and business mix and our recurring revenue base provides some stability in revenue, profit and cash generation. In the
current environment, technology demand continues to be a major driving force behind global economic and business growth.
Businesses and governments around the world are looking for opportunities to scale, offer better services, drive efficiencies and
seize new market opportunities. More recently, geopolitical events and the interest rate environment are adding to the uncertainty.
In response, clients are leveraging technologies like hybrid cloud and artificial intelligence (AI) that boost productivity and
competitiveness.
For the year ended December 31, 2023, movements in global currencies continued to impact our reported year-to-year revenue and
profit. We execute hedging programs which defer, but do not eliminate, the impact of currency. The (gains)/losses from these
hedging programs are reflected primarily in other (income) and expense. Refer to “Currency Rate Fluctuations,” for additional
information. We saw progress from the actions we have taken to mitigate the impacts of escalating labor and component costs and
a strong U.S. dollar (USD).
Financial Performance Summary
In 2023, we reported $61.9 billion in revenue, income from continuing operations of $7.5 billion, and operating (non-GAAP)
earnings of $8.9 billion. Diluted earnings per share from continuing operations was $8.15 as reported and diluted earnings per share
was $9.62 on an operating (non-GAAP) basis. We generated $13.9 billion in cash from operations and $11.2 billion in free cash flow,
and returned $6.0 billion to shareholders in dividends. We are pleased with the fundamentals of our business and progress we have
made in executing our strategy. Our 2023 performance demonstrates the strength of our diversified portfolio and sustainability of
our revenue growth. We increased our investment in innovation and talent and completed nine acquisitions in 2023, strengthening
our hybrid cloud and AI capabilities, all while continuing to return value to shareholders through our dividend.
Total revenue grew 2.2 percent year to year as reported and 3 percent adjusted for currency compared to the prior year, led by
Software and Consulting. Software revenue increased 5.1 percent as reported and 5 percent adjusted for currency, with growth in
Hybrid Platform & Solutions and Transaction Processing. Hybrid Platform & Solutions increased 4.6 percent as reported and 5
percent adjusted for currency, with growth across Red Hat, Automation and Data & AI. Transaction Processing increased 6.2
percent as reported and 6 percent adjusted for currency, reflecting the success of our zSystems platform which continued to drive
client demand. Consulting revenue increased 4.6 percent as reported and 6 percent adjusted for currency with growth across all
lines of business, highlighting the solid demand for data and technology transformation and application modernization projects.
Infrastructure decreased 4.5 percent year to year as reported and 4 percent adjusted for currency, reflecting product cycle
dynamics.
From a geographic perspective, Americas revenue grew 2.0 percent year to year as reported (2 percent adjusted for currency).
Europe/Middle East/Africa (EMEA) increased 3.0 percent as reported (1 percent adjusted for currency). Asia Pacific grew 1.6
percent as reported (7 percent adjusted for currency).
Gross margin of 55.4 percent increased 1.4 points year to year, with continued margin expansion across all reportable segments
driven by revenue growth, improving portfolio mix and productivity actions. Operating (non-GAAP) gross margin of 56.5 percent
increased 1.3 points versus the prior year, due to the same dynamics.
Total expense and other (income) decreased 18.8 percent in 2023 versus the prior year primarily driven by the one-time, non-cash
pension settlement charge of $5.9 billion in 2022 and the benefits from productivity actions we have taken; partially offset by the
effects of currency, higher workforce rebalancing charges to address remaining stranded cost from portfolio actions, and higher
spending reflecting our continued focus on talent and portfolio innovation to drive our strategy. Total operating (non-GAAP) expense
and other (income) increased 4.5 percent year to year, d |
303 | ibm | redp5695.pdf | Front cover
IBM Cloud Pak for Data on
IBM Z
Jasmeet Bhatia
Ravi Gummadi
Chandra Shekhar Reddy Potula
Srirama Sharma
Data and AI
Redguide
Executive overview
Most industries are susceptible to fraud, which poses a risk to both businesses and
consumers. According to The National Health Care Anti-Fraud Association, health care fraud
alone causes the nation around $68 billion annually.1 This statistic does not include the
numerous other industries where fraudulent activities occur daily. In addition, the growing
amount of data that enterprises own makes it difficult for them to detect fraud. Businesses
can benefit by using an analytical platform to fully integrate their data with artificial intelligence
(AI) technology.
With IBM Cloud Pak® for Data on IBM Z, enterprises can modernize their data infrastructure,
develop, and deploy machine learning (ML) and AI models, and instantiate highly efficient
analytics deployment on IBM LinuxONE. Enterprises can create cutting-edge, intelligent, and
interactive applications with embedded AI, colocate data with commercial applications, and
use AI to make inferences.
This IBM Redguide publication presents a high-level overview of IBM Z. It describes IBM
Cloud Pak for Data (CP4D) on IBM Z and IBM LinuxONE, the different features that are
supported on the platform, and how the associated features can help enterprise customers in
building AI and ML models by using core transactional data, which results in decreased
latency and increased throughput.
This publication highlights real-time CP4D on IBM Z use cases. Real-time Clearing and
Settlement Transactions, Trustworthy AI and its Role in Day-To-Day Monitoring, and the
Prevention of Retail Crimes are use cases that are described in this publication. Using CP4D
on IBM Z and LinuxONE, this publication shows how businesses can implement a highly
efficient analytics deployment that minimizes latency, cost inefficiencies, and potential
security exposures that are connected with data transportation.
1 https://www.bcbsm.com/health-care-fraud/fraud-statistics.html
© Copyright IBM Corp. 2023. 1
IBM Z: An overview
Ever wonder how many transactions a bank processes per day? What about the pace at
which these transactions happen? According to an IBM® report, 44 of 50 of the world's top
banks use IBM Z mainframes for these daily transactions.2 IBM Z is a platform that is
designed for voluminous data, maximum security, real-time transaction analysis, and cost
efficiency.
The most recent platform for IBM Z is IBM z16™. The IBM z16 supports the following
features:
(cid:2) On-chip AI acceleration
(cid:2) Quantum-safe crypto discovery
(cid:2) Simplified compliance
(cid:2) Flexible capacity
(cid:2) Modernization of applications
(cid:2) Sustainability
With these features, enterprises can upgrade applications while preserving secure and
resilient data.
To learn more about these features, see the IBM z16 product page.
Figure1 on page3 shows a picture of the IBM z16 mainframe.
2 https://www.ibm.com/case-studies/bankwest/
2 IBM Cloud Pak for Data on IBM zSystems
Figure 1 IBM z16
IBM z16 and IBM LinuxONE Emperor 4 features
IBM Z are based on enterprise mainframe technology. Starting with transaction-based
workloads and databases, IBM Z has undergone tremendous transformations in its system
design for many generations to build servers that cater to Linux-based workloads and security
with a cyberresilient system, and support quantum computing and modernization by using a
hybrid cloud with a focus on data and AI.
3
Figure2 provides a snapshot of the IBM Z processor roadmap, which depicts the journey of
transformation and improvement.
Figure 2 IBM Z: Processor roadmap
The IBM z16 and IBM LinuxONE Emperor 4 are the latest of the IBM Z, and they are
developed with a ‘built to build’ focus to provide a powerful, cyberresilient, open, and secure
platform for business with an extra focus on sustainability to help build sustainable data
centers. Although the z16 server can host both IBM z/OS® and Linux workloads, LinuxONE
Emperor 4 is built to host Linux only workloads with a focus on consolidation and resiliency.
Depending on the workload, consolidation from numerous x86 servers into a LinuxONE
Emperor 4 can help reduce energy consumption by 75% and data center floor space by 50%,
which helps to achieve the sustainability goals of the organization.
Figure3 on page5 shows a summary of the system design of IBM LinuxONE Emperor 4 with
the IBM Telum™ processor. The IBM Telum processor chip is designed to run enterprise
applications efficiently where their data resides to embed AI with super low latency. The
support for higher bandwidth and I/O rates is supported through FCP Express cards with an
endpoint security solution. The memory subsystem supports up to 40 TB of memory.
4 IBM Cloud Pak for Data on IBM zSystems
Figure 3 System design of IBM z16 LinuxONE Emperor 4
The IBM z16 and IBM LinuxONE Emperor 4 servers are built with 7-nm technology at a
5.2 GHz speed. They consist of four dual-chip modules (DCMs) per central processor
complex (CPC) drawer, each of which is built with two 8-core Telum processor chips that has
“first in the industry” on-chip acceleration for mid-transaction, real-time AI inferencing, which
supports many different use cases, including fraud detection.
Each core has access to a huge private 32 MB L2 cache where up to 16 MB of the L2 cache
of an inactive core can be used as virtual cache (L3 / L4) by neighboring active cores on the
chip. This cache helps address translation and access checking by prefetching the same
virtual cache into the L2 cache. The virtual cache also includes Neural Network Processing
Assist instructions and direct memory access with protection, and per chip GZIP
compression.
5
Figure4 provides more information about the features of AI Accelerator integration with the
IBM Z processor cores.
Figure 4 IBM z16 on-chip AI Accelerator integration with IBM Z processor cores
The IBM z16 and IBM LinuxONE Emperor 4 server platforms are built with the hardware
features that are shown in Figure4 with addressing data and AI workloads in mind.
Regardless of where the ML and deep learning (DL) frameworks are used to build and train
data and AI models, the inferencing on existing enterprise application data can happen along
currently running enterprise business applications. CP4D 4.6 supports Tensorflow and IBM
Snap ML frameworks, which are optimized to use the on-chip AI Accelerator during
inferencing. Support for various other frameworks is planned for future releases.
Figure5 on page7 shows the seamless integration of AI into existing enterprises workloads
on the IBM z16 while leveraging the underlying hardware capabilities.
6 IBM Cloud Pak for Data on IBM zSystems
Figure 5 Seamless integration
What is Cloud Pak for Data on IBM Z
IBM Cloud Pak for Data allows enterprises to simplify, unify, and automate the delivery of data
and AI. It categorizes the activities within the journey to AI as four rungs of the AI Ladder:
Collect, Organize, Analyze, and Infuse. For more information about each of the AI Ladder
rungs, see Become Data Driven with IBM Z Infused Data Fabric, REDP-5680.
CP4D on IBM Z provides enterprises with a resilient and secure private cloud platform. You
can use it to create ML and AI models that may be included into modern intelligent
applications. You also can use it to use and construct applications for mission-critical data.
With CP4D on IBM Z, enterprises can lower data movement latency, cost inefficiencies, and
potential security exposures. Enterprises can safely store and access their most important
company data, and leverage their current infrastructure by using cutting-edge hybrid cloud
applications. Enterprises can combine their current database applications without any
rewrites, which results in reduced cost and complexity. Lastly, by using CP4D on IBM Z,
enterprises can update their database infrastructure to benefit from easier management, a
quicker time to value, and lower operating expenses.
7
Figure6 shows a solution overview of CP4D. The infrastructure alternatives are shown at the
bottom, and they include IBM Z and LinuxONE. They all leverage Red Hat OpenShift.
Common Foundational Services come next, which offer clarity throughout the data and AI
lifecycle, that is, from user access management to monitoring and service provisioning. A
high-level view of the services is shown in the middle section. The services have several
different capabilities that span the AI hierarchy. The platform can be expanded, and it offers a
seamless user experience for all distinct personas across the AI lifecycle, from data gathering
through AI infusion.
Figure 6 Solution overview of Cloud Pak for Data
We highlight the four main pillars that make IBM Z the correct infrastructure for CP4D:
(cid:2) Performance and Scale
(cid:2) Embedded Accelerators
(cid:2) Reliability and Availability
(cid:2) Security and Governance.
From a performance perspective, CP4D on IBM Z provides your data and AI with high
transaction processing and a powerful infrastructure. From the embedded accelerators
perspective, CP4D on IBM Z can investigate each transaction thanks to a cutting-edge DL
inference technology even in the most demanding, sensitive, and latency-prone real-time
workloads. From a reliability perspective, CP4D on IBM Z provides high availability and
resiliency. Lastly from the security perspective, CP4D on IBM Z is suitable for protecting
sensitive data and AI models for enterprises in highly regulated industries or those industries
that are worried about security.
8 IBM Cloud Pak for Data on IBM zSystems
Cloud Pak for Data capabilities on IBM Z and IBM LinuxONE
With CP4D on IBM Z and IBM LinuxONE, users can develop, train, and deploy AI and ML
models. Users can accomplish this task by using the CP4D IBM Watson® Studio and IBM
Watson Machine Learning (WLM) services. By using these two fundamental services, users
can accomplish the following tasks:
(cid:2) Provision various containerized databases.
(cid:2) Explore, clean, shape, and alter data by using Data Refinery.
(cid:2) Use project-specific data that is uploaded, or connect to distant data.
(cid:2) Create Spark run times and applications.
(cid:2) Create, build, evaluate, and deploy analytics and ML models with trust and transparency.
(cid:2) Leverage the AI Integrated Accelerator for TensorFlow 2.7.2 and Snap ML 1.9.
For more information about the specifics of these capabilities, see Capabilities on Linux on
IBM Z and IBM LinuxONE.
Open-source ecosystem
These days, innovation and product development are not limited to closed doors within an
organization. In any industry sector, the solutions include a mix of proprietary code
addressing the core business solution that is supported or integrated into other software
components from open source. In some cases, enterprises business solutions also are built
from open-source community offerings. Thus, open-source software becomes an important
ingredient in modern-day solution building.
IBM actively participates in various open-source communities as part of steering boards
defining the roadmap of the community, and also in contributing code to make the community
a better place for everyone to participate. Red Hat also actively participates in various
open-source communities and makes extensive contributions. In open-source communities,
although most open-source development happens on x86 / amd64 or the Intel architecture,
the same open-source software is used by other architectures, such as IBM Power (ppc64le),
IBM Z and IBM LInuxONE (s390x), ARM, and Sparc. So, the availability of an open-source
ecosystem on any architecture is key and critical to business.
On IBM Z and IBM LinuxONE (s390x) architecture, there is a huge open-source support
ecosystem that ranges from operating systems such as Linux; application run times; cloud
and container services; DevOps and automation; big data; observability; analytics;
databases; and storage. The ecosystem on IBM Z and IBM LinuxONE is growing.
IBM Z and IBM LinuxONE include much open-source software in their ecosystem. You can
see the growing list of open-source software for IBM Z and LinuxONE at The Growing
Ecosystem of Open-Source Software for IBM Z and LinuxONE.
IBM Z and IBM LinuxONE are available to various communities to include support for s390x
builds as part of their community’s continuous integration and continuous delivery (CI/CD).
Also, for open-source community developers, infrastructure resources are available on a
no-charge basis through the IBM LinuxONE community cloud.
9
CP4D includes a mix of open-source and proprietary data and AI runtime databases;
open-source run times like Python; open-source data platforms like Anaconda; ML and DL
frameworks like Pytorch and Tensorflow; and thousands of reusable Python packages. All of
them are available and supported on s390x architecture to provide seamless parity with x86
architecture and a seamless experience for enterprise data scientists, architects, and data
and AI solution developers on IBM Z and IBM LinuxONE platforms.
Anaconda is one of the open-source data platforms that provide Python and R based data
science ML frameworks; analytics and data visualization tools; and open-source data science
tools and libraries like Conda, XGBoost, and SciKit-Learn. Anaconda runs natively on Linux
on IBM Z and IBM LinuxONE, and on IBM z/OS Container Extensions (zcX) on z/OS. For
more information, see Announcing Anaconda for Linux on IBM Z and LinuxONE.
In addition to strong, open-source ecosystem support for application development on Linux
and enterprise operating systems, a new generation of IBM Z and IBM LinuxONE servers
(IBM z16™) also have strong platform support, and AI acceleration capabilities that can be
leveraged by open-source software to perform better on the server infrastructure. For
example, the recently released CP4D 4.6 has Tensorflow and IBM SnapML frameworks that
leverage the AI accelerators when running on an IBM z16 server.
So, to summarize, there is a huge, growing data and AI open source ecosystem that is
supported and optimized on IBM Z and IBM LinuxONE servers.
Why AI on IBM Z
Data and AI playing a major role in the modernization story to enable the digital
transformation journey of every organization. Many organizations recognize the business
value of infusing AI into their infrastructure. CP4D provides the cloud-native solution to put
your data to work. With CP4D, all your data users can collaborate from a single, unified
interface that supports many services that work together, including collecting data, organizing
the data, analyzing the data, and infusing AI.
Traditional ML models' power most of today's ML applications in business and among AI
practitioners. CP4D supports traditional ML frameworks for training and inferencing, such as
Scikit-learn, Snap ML, and XGBoost. Snap ML is a library that provides high-speed training
and inferencing of ML models that leverage the AI accelerator while running on an IBM z16
(Linux on IBM Z). CP4D supports DL frameworks such as TensorFlow and PyTorch.
TensorFlow is a DL framework that leverages the AI accelerator while running on an IBM z16
(Linux on IBM Z).
Figure7 on page11 provides an overview of the components that are supported on
CP4D on IBM Z. You can leverage Watson Studio for model building, training, and validation,
and WML for deployment of the model. Eventually, applications can use the AI inference
endpoint to score the model.
10 IBM Cloud Pak for Data on IBM zSystems
Figure 7 Developing, training, and deploying an AI model on Cloud Pak for Data on IBM Z and IBM LinuxONE
In summary, here are some of the reasons why you should choose AI on IBM Z:
(cid:2) World-class AI inference platform for enterprise workloads:
– Embedded accelerators: A centralized on-chip AI accelerator that is shared by all
cores.
– Industry standard AI ecosystem: Many industry open-source data science frameworks
are available on the platform.
– Seamlessly integrate AI into existing enterprise workload stacks: Train anywhere, and
then deploy on IBM Z.
(cid:2) Security: Encrypted memory, and improved trusted execution environments.
(cid:2) Sustainability: Reduce your energy consumption with real-time monitoring tools about the
energy consumption of the system.
AI use cases
With billions of transactions per day in many of today’s industries, it is key to get real-time
insights about what is happening in your data. AI on the IBM Z stack understands these
situations, and it delivers in-transaction inference in real time and at scale.
Core banking solutions running on IBM Z that are involved in processing inbound transactions
need real-time fraud detection to prevent fraud. Other types of possible use cases might be
credit risk analysis, anti-money laundering, loan approval, fraud detection in payments, and
instant payments.
For insurance companies, a pressing use case would be claims processing. For markets and
trading, clearing and settlement use cases are paramount.
11
For the health care industry, medical image processing (such as MRIs and x-rays), skin
cancer detection, and patient monitoring activities such as infant motion analysis, is
important.
For the airline industry, processes such as air traffic management, flight management
systems, and flight maintenance predictions are use cases that are ideal candidates for using
AI on IBM Z.
In the following sections, we describe the following use cases:
(cid:2) “Use case 1: Responsible AI augmented with risk and regulatory compliance” on page12
AI model lifecycle governance, risk management, and regulatory compliance are key to
the success of the enterprises. It is imperative to adopt a typical AI model lifecycle to
protect new end-to-end risks.
(cid:2) “Use case 2: Credit default risk assessment” on page22
Core banking solutions running on IBM Z that are involved in processing inbound
transactions need real-time fraud detection to prevent fraud. Other types of possible use
cases might be credit risk analysis, anti-money laundering, loan approval, fraud detection
in payments, and instant payments.
(cid:2) “Use case 3: Clearing and settlement” on page25
The use of AI can help to predict which trades or transactions have high risk exposures,
and propose solutions for a more efficient settlement process.
(cid:2) “Use case 4: Remaining Useful Life of an aircraft engine” on page27
We describe how AI can help to avoid unplanned aircraft downtime by determining the
remaining time or cycles that an aircraft engine is likely to operate before failure.
(cid:2) “Use case 5: AI-powered video analytics on an infant's motions for health prediction” on
page30
In this section, we describe how AI can predict an infant’s health conditions by monitoring
real-time body movements.
Use case 1: Responsible AI augmented with risk and regulatory
compliance
Advancement in AI is changing the world, and organizations must adopt AI to embrace new
challenges daily. Many enterprises see tremendous value in adopting AI and ML technologies
while establishing organization trust in the models, underlying data, and the process to be
followed. An AI model lifecycle can be a daunting task.
How mature is your AI governance? In this section, we provide a use case demonstrating the
trustworthiness of AI and its importance in daily monitoring.
Industry challenges
Here are the three main reasons why organizations struggle with the adoption of AI:
(cid:2) Scaling with growing regulations
(cid:2) Lack of confidence in operationalized AI (making responsible AI)
(cid:2) Challenges around managing the risk throughout the entire AI workflow
12 IBM Cloud Pak for Data on IBM zSystems
Scaling with growing regulations
Laws and regulations in the data and AI space are accelerating, and many countries are
proposing strict AI policies. Countries are monitoring adherence of these policies by the
enterprises and imposing fines for any violations. Responding to these regulations are
challenging global organizations where multiple regulations apply. For enterprises, it is
important to adopt AI policies when there is change, and to validate explainable models to
protect against discrimination.
Responsible AI
Responsible AI protects against loss of data privacy, and reduced customer loyalty and trust.
A data scientist cannot maximize accuracy and model performance above all other concerns.
Practicing responsible AI is a best practice, and you must establish protection and validation
to ensure that any models that are placed into production are fair and explainable.
Risks throughout the entire AI workflow
Organizations need to mitigate risk of the following items:
(cid:2) Deciding not to use certain technologies or practices
(cid:2) Using personal information when needed and with a user's consent
(cid:2) Ensuring automated decisions are free from bias
(cid:2) Customer confidence by providing explanations for business decisions
(cid:2) Fraud to the organization and to customer's accounts
(cid:2) Delays in putting models into production
In fact, in a recent survey, these concerns were echoed by real AI adopters when asked what
aspects of trust are most important to them. Although explaining how AI decides is the
primary concern, all of these concerns are important.
The key point here is that risk exists throughout the entire AI lifecycle starting with the
underlying data and the business justification behind the “why” of the project and continuing
into production. Without a formalized process, there is no way to mitigate these risks to unlock
the scale that is required to make automated decisions profitable. With these decisions, the
business can operate proactively instead of reactively.
13
For example, a business can start testing a model before production for fairness metrics. For
this task, enterprises need an end-to-end workflow with approvals to mitigate these risks and
increase the scale of AI investments, as shown in Figure8, which presents a typical AI model
lifecycle in an enterprise.
Figure 8 Typical AI model lifecycle
Due to regulations, more stakeholders adopt the typical AI model lifecycle to protect their
brand from new end-to-end risks. To ensure various aspects of both regulatory compliance
and security, the personas that must be involved include the chief financial officer (CFO),
chief marketing officer (CMO), chief data officer (CDO), HR, and chief regulatory officer
(CRO), along with the data engineers, data scientists, and business analysts, who build AI
workflows.
IBM governance solution for IBM Z
AI model lifecycle governance, risk management, and regulatory compliance are key to the
success of enterprises.
AI governance is a comprehensive framework that uses a set of automated processes,
methodologies, and tools to manage an organization's use of AI. Consistent principles
guiding the design, development, deployment, and monitoring of models are critical in driving
responsible and trustworthy AI. AI governance includes processes that trace and record the
origin of data, models (including associated metadata), and pipelines for audits. The details of
entry should include the techniques that trained each model, the hyperparameters that were
used, and the metrics from testing phases. These details provide increased transparency into
the model's behavior throughout the lifecycle, the data that was influential in its development,
and the possible risks.
In a world where trust, transparency and explainable AI matters, every organization wants
compliance along with the comfort of understanding how analytic insights and decisions are
made. The following sections describe some of the principles and organizational
requirements for AI governance.
14 IBM Cloud Pak for Data on IBM zSystems
Lifecycle governance
Lifecycle governance helps you manage your business information throughout its lifecycle,
that is, from creation to deletion. IBM AI governance addresses the problems that challenge
records managements:
(cid:2) Monitor, catalog, and govern AI models from anywhere throughout the AI lifecycle.
(cid:2) Automate the capture of model metadata for report generation.
(cid:2) Drive transparent and explainable AI at scale.
(cid:2) Increase accuracy of predictions by identifying how AI is used and where it is lagging.
Risk management
Risk management is used in IBM AI governance to identify, manage, monitor, and report on
risk and compliance initiatives at scale:
(cid:2) Automate facts and workflow management to comply with business standards.
(cid:2) Use dynamic dashboards for clear and concise customizable results.
(cid:2) Enhanced collaboration across multiple regions and geographies.
Regulatory compliance
Regulatory compliance is a set of rules that organizations must follow to protect sensitive
information and ensure human safety. Any business that works with digital assets, consumer
data, health regulations, employee safety, and private communications is subject to regulatory
compliance.3 The IBM AI governance solution for IBM Z includes the following tasks:
(cid:2) Help adhere to external AI regulations for audit and compliance.
(cid:2) Convert external AI regulations into policies for automatic enforcement.
(cid:2) Use dynamic dashboards for compliance status across policies and regulations.
Enterprises can develop AI models and deploy them by using IBM Watson Studio or WML on
CP4D on Red Hat OpenShift on a virtual machine that is based on IBM z/VM or Red Hat
Enterprise Linux KVM on IBM Z. AI governance on IBM LinuxONE is supported in the
following two ways:
(cid:2) Monitor the AI models with Watson OpenScale on CP4D on Red Hat OpenShift on a
virtual machine on IBM Z.
(cid:2) Enterprises can develop AI models by creating and training models by using Watson
Studio and development tools such as Jupyter Notebook or JupyterLab, and then
deploying the model onto WML on CP4D on Red Hat OpenShift on a virtual machine on
IBM Z. Then, these enterprises can achieve end-end AI governance by running AI
Factsheets, IBM Watson OpenScale, and IBM Watson OpenPages® on CP4D on x86.
Figure9 on page16 shows the end-to-end flow for a remote AI governance solution.
3 https://www.proofpoint.com/us/threat-reference/regulatory-compliance
15
Figure 9 Remote AI governance solution end-to-end flow
To achieve end-to-end AI governance, complete the following steps:
1. Create a model entry in IBM OpenPages by using CP4D on a x86 platform, as shown in
Figure10.
Figure 10 Creating a model entry in IBM OpenPages
16 IBM Cloud Pak for Data on IBM zSystems
2. Train a model by using Watson Studio and by using development tools such as Jupyter
Notebook or JupyterLab on CP4D on Red Hat OpenShift on a virtual machine on IBM Z,
as shown in Figure11.
Figure 11 Training an AI model by using Watson Studio
3. Deploy the model by using WML on CP4D on Red Hat OpenShift on a virtual machine on
IBM Z, as shown in Figure12.
Figure 12 Deploying an AI model by using WML on Cloud Pak for Data
17
4. Track the external model lifecycle by browsing through the Catalogs/Platform assets
catalog by using AI Factsheets and OpenPages while using CP4D on an x86 platform, as
shown in Figure13. The external model (deployed on CP4D on Red Hat OpenShift on a
virtual machine on IBM Z) is saved as a platform asset catalog on the x86 platform.
Figure 13 External model
You can track the model through each stage of the model lifecycle, as shown in Figure14,
by using AI Factsheets and OpenPages.
Figure 14 Tracking the model
18 IBM Cloud Pak for Data on IBM zSystems
You can see that the model facts are tracked and synchronized to IBM OpenPages for risk
management, as shown in Figure15.
Figure 15 Model facts that are tracked and synchronized to IBM OpenPages on an x86 platform
19
5. Create an external model by using IBM OpenScale on the x86 platform, as shown in
Figure16.
Figure 16 Creating an external model on an x86 platform
IBM OpenScale provides a comprehensive dashboard that tracks fairness, quality monitoring,
drift, and explainability of a model. Fairness determines whether your model produces biased
outcomes. Quality determines how well your model predicts outcomes. Drift is the
degradation of predictive performance over time. A sample is shown in Figure17 on page21.
20 IBM Cloud Pak for Data on IBM zSystems
Figure 17 IBM OpenScale dashboard that is used to monitor the external model
You developed and deployed the AI model by using Watson Studio, WML on CP4D on Red
Hat OpenShift on a virtual machine on IBM Z, and end-to-end AI model governance by
leveraging AI Factsheets, OpenScale, and OpenPages on CP4D on a x86 platform. Figure18
shows end-to-end AI governance when using IBM OpenPages, AI Factsheets, and
OpenScale.
Figure 18 Final result: End-to-end AI governance when using IBM OpenPages, AI Factsheets, and OpenScale
21
Use case 2: Credit default risk assessment
In today’s world, many individuals or businesses seeking loans to meet their growing business
needs often look to financial institutions. Financial institutions can offer loans to individuals or
businesses and charge interest based on the current market situations.
Industry challenges
Financial institutions must make an accurate decision about whether to sanction a loan or not,
and judging the likelihood of default is the difference between a successful and unsuccessful
loan portfolio. In a traditional scenario, an experienced banker can judge someone’s likelihood
of default, but that is not an efficient method for judgment as a business grows.
Predictions of credit default risk assessment
In the modern world, growing business institutions can no longer rely on only experienced
bankers to decide whether to sanction a loan knowing that there is a probability that the
borrower might default on their loans. A better choice is to rely on technological
advancements that can help with reasoning based on facts, such as leveraging credit risk
modeling techniques to process the historical data of past borrowers to understand their
credit behavior and make a more informed decision about whether to lend money, how much
money, and decide on the tenure to close the loan.
Financial institutions can leverage AI solutions by using ML techniques to predict the credit
risk. Applying AI to credit risk modeling techniques can benefit institutions in decision-making,
and thus can help better manage the exposure to credit risk.
Figure19 on page23 shows a sample architecture about how to design and develop an AI
model for credit risk assessment on IBM Z. An IBM WebSphere® Application Server is used
for handling in-bound transactions, and CP4D is used for AI model lifecycle management that
includes building, training, and deploying the model.
22 IBM Cloud Pak for Data on IBM zSystems
Figure 19 Architecture for credit risk prediction by using an ML AI model on IBM Z
A data scientist can leverage Watson Studio to develop and train an AI model and WML to
deploy and score the model. In this sample architecture, the WML Python run time leverages
the ML framework, IBM Snap Machine Learning (Snap ML), for scoring, can leverage an
integrated AI accelerator at the time of model import.
Then, the banking loan approval team can send a loan applicant request to the IBM
WebSphere Application Server, which can make a request to the AI inference endpoint. The
AI inference engine scores the transaction and sends the result back to the loan approval
team. Based on the results, the approval team can decide on whether to approve a loan or
not, and also decide how much they can lend, timelines, and other factors.
The transaction system that is shown in Figure19 uses IBM WebSphere Liberty as an
application server, but you also can use an IBM Open Liberty® application server or any
application server that can send RESTful API communications.
Models are frequently developed and tested in many platforms and languages, such as
Python, Scala, R, and Go. Models can leverage ML frameworks like scikit-learn, Snap ML, or
XGBoost, or DL frameworks like TensorFlow or PyTorch. Training a model can be done on
any platform if you have enough computing power for complex models, but moving that model
into production requires careful testing to ensure that transactions are not delayed, especially
if you plan to run the model within a transaction.
We showed how IBM Z enable customers to use AI frameworks to detect credit risk. Now, we
look at how you can leverage CP4D and TensorFlow on IBM Z to detect the credit risk.
23
Figure20 shows an architecture for predicting credit risk by using DL on IBM Z.
Figure 20 Architecture for credit risk prediction by using DL on IBM Z
Data scientists can star |
306 | ibm | IBM-ETAB-Report-white-paper-DIGITAL-20241212_5B30_5D.pdf | New York State
Emerging Technology
Advisory Board
Recommendations for making
NY a leader in responsible AI
New York State Emerging Technology Advisory Board
Table of contents
3 Letter from the co-chairs
4 The Emerging Technology Advisory Board
5 Abstract
7 New York’s AI landscape
15 Inspirational AI stories from external
stakeholders
16 Responsible AI
18 Vision and ambitions
Adoption at scale
Democratization of AI
Resilience and equity within the workforce
20 Recommendations
40 Next steps
42 Acknowledgements
44 Endnotes
2
New York State Emerging Technology Advisory Board
Letter from the co-chairs
New York has a long history of capitalizing on major technological The ETAB is proud to provide Governor Hochul with this report,
breakthroughs and economic shifts. Few other states have which sets forth bold, ambitious, and powerful recommendations
adapted as successfully to these profound changes, which have based on the latest research and AI developments. These
shaped workers’ lives and industries’ fate. Today, New York State recommendations reflect the diversity of thought and experience
has, once again, a unique opportunity to pioneer and embrace provided by the Board and other stakeholders, shaping the
emerging technologies. To do this, New York must build a thriving future of emerging technologies. They build on the Governor’s
ecosystem that supports innovation, deploys innovations at existing, substantial efforts and New York’s global reputation
scale, and provides equitable opportunities for its people and as a place where businesses come to grow, innovate, and create
its workforce. future technologies. The recommendations are designed to guide
both the State and organizations across New York in driving an
In March 2024, Governor Hochul asked us to establish and lead innovative AI ecosystem, ensuring responsible AI deployment
an independent advisory board. This board aims to develop at scale, fostering a resilient workforce, and empowering all New
recommendations for how New York State can best support Yorkers with equitable access to the benefits of AI.
and grow a thriving ecosystem for emerging technologies. The
Emerging Technology Advisory Board (ETAB) comprises private Together, we can secure New York’s position at the forefront of
sector leaders, and leaders from globally renowned nonprofit this transformative era.
and foundation organizations (see here for Governor Hochul’s
June 13, 2024 press release). These board members are actively Sincerely,
involved in civic life and duty. Through their work on the ETAB,
they are dedicating their expertise and contributions to advancing
New York’s interests. The ETAB dedicated the first six months
to developing recommendations that achieve one unified vision:
Elevate New York as an AI leader. Arvind Krishna Dr. Tarika Barrett
Chairman and CEO, IBM CEO, Girls Who Code
AI development is fast-moving and exciting; however, the ETAB Co-Chair, ETAB Co-Chair, ETAB
acknowledges there remains uncertainty about the scale, speed
of adoption, and consequent impacts on the workforce. Being
a successful leader in AI will require agility and adaptability,
and the State should frequently reassess its approach, and the
recommendations in this report, as AI continues to be deployed
and the landscape evolves.
3 Whitepaper | Template Month Year
New York State Emerging Technology Advisory Board
The Emerging Together, the Advisory Board’s immense contributions and
dedicated efforts reflect the critical need to thoughtfully
Technology and boldly elevate New York as a leader in this pivotal moment.
The report’s ambitious, independent recommendations
are a testament to the insightful perspectives and inspiring
Advisory Board
collaboration of the Advisory Board members. As such, this
report is a culmination of contributions from the Board as
a whole; each item included may not directly reflect every
member’s point of view nor should they be read as mandates
for organizations.
Albert Bourla René F. Jones
CEO CEO
Pfizer M&T Bank
Richard Buery Lynn Martin
CEO President
Robin Hood The New York Stock Exchange
Bertina Ceccarelli Sanjay Mehrotra
CEO CEO
NPower Micron Technology
Somak Chattopadhyay Aparna Pappu
Managing Partner Vice President and General Manager
Armory Square Ventures Google Workspace
Mario Cilento Julie Samuels
President CEO
New York State AFL-CIO Tech:NYC
Dev Ittycheria Lisa Sobierajski Avila
CEO CEO
MongoDB Kitware
Joanna Geraghty Pat Wang
CEO CEO
JetBlue Healthfirst
Lyndie Hice-Dunton Darren Walker
Executive Director President
National Offshore Wind Research & Ford Foundation
Development Consortium
4 Whitepaper | Template Month Year
Abstract
5 Whitepaper | Template Month Year
Abstract
Making New York a Grounded in these learnings, the Advisory Board engaged over 40
external stakeholders and experts to gain a deeper understanding
leader in advancing of the nuances and complexities in the challenges identified, how
those challenges manifest across New York organizations and
communities, ongoing initiatives aiming to address the potential
responsible AI
challenges, and others aiming to build on New York’s position
of strength. The stakeholder interviews also revealed bold ideas
and perspectives on potential recommendations to include
in this report.
The Advisory Board reflected on the insights and ultimately
aligned on three ambitions for the State of New York to pursue:
– Enable all New York businesses to responsibly deploy
In 2022, the widespread introduction of generative AI (gen AI at scale
AI) rapidly transformed the technology landscape, creating – Commit to AI literacy for at least 15 million
unprecedented global opportunities. For New York, gen AI New Yorkers by 2030, democratizing AI in the process
could mean an economic expansion of up to $100 billion from – Ensure every worker in New York can thrive in the new
productivity improvements alone.1 Given the potential for AI landscape
significant disruption in this critical moment, Governor Hochul
asked the Emerging Technology Advisory Board (ETAB) to These ambitions are supported by 9 recommendations that
develop a plan for a thriving emerging technology ecosystem foster public-private partnerships and balance the priorities
in New York. The first six months of their effort outlined of timely impact and sufficient scale.
recommendations that could make New York the leader
in advancing responsible AI.
The Advisory Board took a comprehensive approach to
developing the recommendations outlined in this report. First,
the Advisory Board reviewed New York’s AI landscape. The
effort validated the state’s foundational position of strength.
New York’s robust economy, extensive tech talent pool,
academic excellence, and legacy for innovation underpin the
state’s promising potential to be a leader in AI. The assessment
also identified challenges the state may face, primarily related
to supporting and empowering its workforce to thrive in the AI
transition, and ensuring equitable access to resources to enable
all New Yorkers to leverage and benefit from the opportunities
AI offers.
6 Whitepaper | Template Month Year
New York’s
AI landscape
7 Whitepaper | Template Month Year
New York’s AI landscape
AI presents AI has rapidly transformed the technology landscape, creating
unprecedented opportunities. The McKinsey Global Institute
a tremendous estimates AI could contribute up to $17–26 trillion to
the global economy annually.2 As an example, by improving
worker productivity through technology and better use of
opportunity globally
time, AI could add ~$3.5 trillion (approximately 4%) to the
global economy. This implies a productivity impact for the US
and in New York
estimated at $1 trillion, of which New York is expected to make
up a disproportionate share of up to $100 billion. New York
is uniquely positioned due to its high-productivity industries,
which are digitally mature and ripe for AI adoption.3 AI presents
opportunities worth seizing.
Beyond the economic impact, AI has enormous potential to
improve our lives in ways both subtle and surprising. Machine
learning tools can analyze medical images, such as X-rays
and MRIs, to help medical professionals diagnose diseases
faster and more accurately. AI algorithms can monitor driving
patterns and modify traffic signals to reduce congestion,
commute times, and emissions. AI-powered analysis enables
universities to assess student needs and offer better targeted
support to address them.4 To make these and many other
benefits a reality, AI technologies will need to be part of a whole
ecosystem that includes innovations in technology, education,
business, and society.
To chart the course to a thriving AI ecosystem in New York,
the Emerging Technology Advisory Board (ETAB) examined
New York’s current AI landscape, including New York’s ability
to realize the potential of AI; the groundbreaking investments
Projected economic benefit from AI-augmented productivity New York has made in AI and AI-adjacent industries; and the
opportunities and challenges that AI will present for New York’s
businesses, people, and infrastructure.
$$33..55TT $3.5T
gglloobbaall economy global economy
$1T
$1T $1T
US
US US
$100B $100B
$100B
New York State New York State
New York State
8 Whitepaper | Template Month Year
New York’s AI landscape
New York is poised With its deep-rooted financial infrastructure and expertise,
New York holds a strategic advantage in driving AI growth.
to capitalize on the Its leadership in private investment markets positions the state
to significantly enhance its AI investment landscape. Over
the years, the state’s commitment to innovation has sparked
AI opportunity
a remarkable surge in AI VC funding, far outpacing investment
growth in other areas.6
The state’s robust economy is expected to experience growing
labor demand, with a net gain of more than 200,000 jobs by
the end of the decade.7 This robust job growth could ease the
potential pressures of the job shifts, creating more opportunities
for reskilling and upskilling.
A robust economy
New York is an economic leader in the US, ranking 3rd nationally
in labor productivity.5 Over half of the state’s sectors surpass the
national productivity average and have experienced remarkable
growth over the past three decades. This surge in productivity
can largely be attributed to their embrace of digital technologies.
Consequently, these high productivity, digitally mature sectors,
such as retail, financial services, and advanced manufacturing,
are primed for the adoption of artificial intelligence, positioning
them—and New York—ideally for the next wave of technological
advancement.
The global economic potential of gen AI, $ billion8
240–
460
240–
390 200–
340 180–
170–
300 160–
290 270 1 25 60 0– 150– 150–
250 240 120– 120–
230 200 110–
180 100– 90–
170 80–
150 140 70– 60– 60–
110 60–
110 110
100
50+
40+
9 Whitepaper | Template Month Year
Tech Retail Banking logistics Travel
&
manufacturing Advanced packaged
goods
Consumer Healthcare services Professional Energy Education Basic
materials
Real
estate
semiconductors Electronics
&
Construction Chemical social
sector
Public
&
Media Life
science
Telecom Insurance Agriculture
Extensive talent pool
New York has long been a beacon for tech talent. Not only is the
state a top tech talent destination—ranking 3rd nationally—
New York City has also become a magnet for tech talent
relocations, outpacing all other cities.9 This influx stands
in stark contrast to San Francisco, which saw a net loss during
the same period. New York’s leading position is driven by
its ability both to develop talent in its premier institutions
and to attract tech talent to relocate.
Top 5 sectors in NY
New York’s AI landscape
Academic excellence
Micron: Harnessing AI for advanced manufacturing
New York has long been a leader in innovation and academic
As semiconductor manufacturing becomes more excellence, with 3 of the top 20 US universities for R&D funding
complex and AI increasingly powers the growing in engineering research.12 Complementing its academic strength,
demand for semiconductors, Micron Technology New York hosts leading research labs, such as NYU’s CILVR, and
has internally deployed AI extensively through its private tech companies such as IBM, MongoDB, and DeepMind.
manufacturing processes to ensure that Micron and
its future New York workforce remains at the cutting
edge. Semiconductor manufacturing involves more
than 1,500 individual steps to turn mined silicon into 3 of the top 20 US universities for
the Micron memory and storage chips that store the R&D funding in engineering research
data for smartphones, the automotive sector, and
other key industries, and Micron uses AI to support
a variety of manufacturing processes, including:
image analytics, acoustic listening, and thermal A legacy for innovation and adaptability
imaging. As a result of these AI innovations, Micron
has improved worker safety and kept its operations New York has a long legacy of pioneering advancements and
competitive: between 2016 and 2020, worker evolving its economy to lead in emerging technology. The state
productivity rose 18%, time to resolve quality issues has birthed countless innovations—from the telegraph (invented
fell by 50%, time to market for new chips fell 50%, by NYU professor Samuel Morse) to photographic film rolls
and product scrap production fell 22%. (invented and popularized by Rochester-based Kodak) to Gorilla
Glass (invented and commercialized by Corning). But beyond
this inventiveness, the state has also successfully navigated
numerous economic shifts, transitioning from a manufacturing-
1st ↑200K based economy to one dominated by financial services and
beyond. Today, New York’s economy is strategically diversified
across technology, media, healthcare, and education. This rich
in tech talent relocations, net job gain by 2030 dynamic not only shields it from economic downturns but also
claiming 15% of all tech sets the stage for a future where new technologies such as
talent relocations10 AI can be harnessed to their fullest potential.
2nd ↑32%
nationally in science rise in AI VC funding over
and engineering degrees the last 9 years, outpacing
conferred11 the 2% growth in all other
VC deals
3rd
nationally in productivity
10 Whitepaper | Template Month Year
New York’s AI landscape
New York is making New York is continuing its longstanding investments in emerging
technology, with a bold vision to advance semiconductors,
groundbreaking quantum computing, and AI. Accordingly, the state has
implemented strategic programs, policies, and commitment to
funding that prioritizes productivity, equity, and sustainability.
investments in AI and
For example, New York’s landmark $400 million investment to
establish Empire AI, a consortium of seven leading universities
AI-adjacent industries
and research institutions that will collaborate in a state-of-the-
art AI computing center, will unlock university research critical
to accelerating AI use cases for public good.13
New York’s simultaneous investments in AI-adjacent industries
are key to building a thriving AI technology ecosystem, as
semiconductors provide the essential processing power that
enables efficiency of AI deployments and quantum computers
NY SMART I-Corridor: could accelerate the speed of AI algorithms. Central in the state’s
Developing a semiconductor cluster investments are:
In July 2024, the Economic Development Administration Semiconductors14
designated the New York Semiconductor Manufacturing – $100 billion commitment from Micron Technology to create
and Research Technology Innovation Corridor (SMART 9,000 Micron jobs, 4,500 construction jobs, and 40,000
I-Corridor) as a Tech Hub. The NY SMART I-Corridor indirect jobs
is a consortium of over 100 institutions in the Buffalo, – $11.6 billion commitment from Global Foundries to generate
Rochester, Ithaca, and Syracuse Metropolitan Statistical over 1,500 jobs
Areas (MSAs) that aims to build a globally leading – $10 billion partnership to advance next-generation chips
semiconductor cluster in Upstate New York. Leading research at NY CREATES Albany
partners include Micron Technology, CenterState – $40 million of federal funding from the CHIPS Act to the
Corporation for Economic Opportunity, and the University NY SMART I-Corridor Tech Hub
at Buffalo. The Tech Hub applies a comprehensive
approach to developing the semiconductor cluster. The Quantum computing
$40 million in federal funding will be used to implement – IBM’s first ever IBM Quantum System One on a university
four projects around supply chain expansion, workforce campus (Rensselaer Polytechnic Institute [RPI]) to accelerate
development, commercialization, and governance.18 quantum computing research15
– $6.5 million public investment to construct a Quantum
Internet Test Bed at Stony Brook University16
$400M Innovation
– $100 million investment from JMA Wireless to relocate its 5G
headquarters—the only US-owned 5G campus—to Syracuse17
landmark public-private – $2.5 million annually to NYSTAR Innovation Hot Spots, which
investment to establish serve as startup incubators and regional hubs connecting
Empire AI technology initiatives across their region
The Emerging Technology Advisory Board aims to build on
these longstanding investments, a testament to Governor
Hochul’s relentless commitment to lead at the forefront of
emerging technology.
11 Whitepaper | Template Month Year
New York’s AI landscape
As New York continues Strengthening workforce and talent
development
to embrace AI, the state
AI is likely to transform the employment landscape. McKinsey
could face challenges
estimates occupational shifts may be required in the New York
Combined Statistical Area (CSA) by 2030 as the evolving nature
of work likely shifts the mix of jobs in the region.19 Some of the
occupational categories that may face the most shifts are office
support workers, customer service and sales, food services,
production work, and business and legal professionals.20 It must
be a priority to responsibly support the workforce through this
transition, for example, by providing upskilling and reskilling
opportunities, creating high-quality, family-sustaining jobs,
connecting workers to employment opportunities, or increasing
Ensuring workers can thrive benefits. By providing resources to support the workforce through
this transition, New York could help workers reap the benefits and
To ensure workers thrive, in addition to initiatives advantages of AI (e.g. by moving to new in-demand industries).
mentioned in this report, including education, training,
and job placement, the State could take additional Employers could disclose the use of AI when it is used in
steps consistent with its labor and employment policy for connection with employees’ substantive work and to make
other industries and new technologies. These include: or assist in labor and employment decisions. There is an
– Labor peace, prevailing rate, and domestic opportunity for New York to monitor the impact of AI on the
content preferences employment landscape to understand and respond to any
– Disclosure and bargaining of AI use in the workplace negative impact on workers in an agile manner.
– Robust worker data privacy, bias, whistleblower,
and discrimination protections New York, along with other traditional tech hubs, faces rising
– Prioritizing employee retention competition for AI talent nationally as the geographic dispersion
– Providing for direct support for displaced workers, of AI roles increases. Between 2018 and 2023, New York saw
including enhanced UI benefits and COBRA a 1.7 percentage point decline in its share of AI job postings.21
premium assistance If domestic and international talent is not retained, that loss
– Ensuring that public spending, investments, and could put New York at a disadvantage.
subsidies only go to applicants who develop or
implement AI to create additional jobs or support With a longer-term view of its talent pipeline, New York has the
existing ones, as opposed to displacing workers opportunity to ensure AI literacy is embedded in the education
of its 2.4 million K–12 public school students.22 Students are
well-positioned to learn from and with AI—early exposure could
encourage safe adoption as AI becomes mainstream.
Change in each state’s share of AI job postings between 2018 and 2023
+1.3 +1.2
+0.8 +0.7 +0.7 +0.7 +0.6 +0.5 +0.5 +0.5
0.0
-0.2 -0.4 -0.4 -0.5 -0.6 -0.6
-1.7
-2.5
-6.0
MD TX KS VA IL AL FL ID MA OK WI CO MI PA NC MN NM NY WA CA
12 Whitepaper | Template Month Year
New York’s AI landscape
Ensuring equitable representation
Girls Who Code: Closing the gender gap in tech
and access to resources
Girls Who Code (GWC), an international nonprofit
working to close the gender gap in tech, is leading the All too often, technological transitions leave underrepresented
movement to inspire, educate and champion girls, groups behind. These groups often lack the resources, such
women and non-binary people, with a special focus as higher education and financial funding, required to succeed
on historically underrepresented groups, to become in transitions. For example, there has been a long history of
changemakers in tech. In 2024, an independent study inequitable distribution of VC funding. In 2022, only ~2% of VC
found that high school students who participate in funding in the US was committed to female-founded companies,
GWC’s summer programs are more likely than their ~1% committed to Black-founded companies, and ~1.5%
peers to major in computer science-related fields committed to Latino-founded companies.24 The inequity in
in college. The impacts of their holistic approach to access to resources could lead to disproportionate impacts
computer science education, grounded in project-based of AI across the state’s workforce. Experts estimate the effects
learning, community, and real-world applications of of labor market churn could have uneven distributional impacts,
emerging technology, are consistently demonstrated which could manifest as a higher risk of impact for low-wage
among students historically underrepresented in workers (4.2X) than high-wage workers, women (1.3X) than
computing, including Black and Hispanic or Latino/a men, non-college-educated workers (1.6X) than those with at
students. These efforts highlight the crucial role of least bachelor’s degrees, and Hispanic workers (1.2x) than white
targeted educational initiatives in fostering gender workers.25 Although AI has the potential to follow the same
diversity in tech.23 inequitable path as other technological transitions, New York can
make proactive, intentional interventions to bridge the equity gap.
Equitable representation in AI is not just about achieving better
outcomes for the individuals, but also for the technology itself.
For example, Dr. Joy Buolamwini has shown that because many
of the datasets that were originally used to train AI models
were not representative of the world at large, many AI tools have
higher misidentification rates for people of color, which can have
devastating effects.26 Equitable representation in AI development,
along with widely prescribed norms, can help improve the
technology’s abilities so it can be deployed responsibly and avoid
biased outcomes for consumers, workers, and the public at large.
Low-wage workers are 4.2x more likely Non-college educated workers are 1.6x
to be affected than high-wage workers. more likely to be affected than those with
bachelor’s degrees.
Women are 1.3x more likely to be affected Black, Hispanic/Latino, American Indian,
than men. and Alaskan Native graduates in higher
education represented less than 25% of
graduates27 while making up ~32% of the
US population.28
Hispanic workers are 1.2x more likely
to be affected than white workers.
13 Whitepaper | Template Month Year
New York’s AI landscape
Building trust in AI Continuing to lead in productivity
AI can’t be successfully deployed at scale if it’s not trusted. Despite being a national leader in overall productivity, New
Currently, 50% of New York constituents fear AI, expressing York’s productivity growth lags the national average and states
concerns about the lack of transparency in its deployment including California, Washington, and Massachusetts.32 AI-driven
and the potential for bad actors to exploit its use cases.29 automation and workforce reskilling could substantially boost
A recent study also found that the less people know about the state’s productivity growth, creating more and higher paying
AI, the more they worry about it.30 Building public awareness jobs for New Yorkers.
of AI’s applications and limitations is a prerequisite to
building New Yorkers’ trust of AI, a challenge the state
could be well-positioned to take on.
New York’s labor productivity growth
American attitudes about increased use of AI:31 from 2019–23 (0.9%) lags states like
California (2.4%), Washington (3.3%),
and Massachusetts (2.3%).
38% 15%
The ETAB supports NYS’s efforts to
address broader AI-related challenges
The ETAB aims to address these challenges and strengthen the
foundation for New York’s leadership in AI. We also acknowledge
there must be broader efforts to address the complexity
More concerned than excited More excited than concerned of some challenges beyond the scope of the Board’s charter.
The state’s energy infrastructure capacity is a prime example.
Electricity consumption by AI data centers is expected to
increase from 2.8% of New York’s supply today to 3-7% by
2030.33 This is an additional pressure on the grid, at a time
when the state is actively pursuing its ambitious clean energy
transition.34 In stakeholder interviews, leaders also highlighted
their growing concern about losing international talent after
significant investment in their development. They continue to
face difficulty in attracting and retaining talent in areas with
a high cost of living. More must be done to fortify the grid with
sufficient carbon-free, reliable, and affordable energy; attract
and retain the talent the state seeks; and to support that talent
with affordable housing and necessary resources to thrive. The
ETAB fully supports the state’s ongoing, comprehensive efforts
to address these challenges and others.
14 Whitepaper | Template Month Year
New York’s AI landscape
Inspirational AI stories The Advisory Board engaged over 40 cross-sector stakeholders
and experts to get their thoughts about AI—what the challenges
from external stakeholders and opportunities are, how they manifest in NY, and how they
can be remediated or seized. Stakeholder insights surfaced
powerful proof points about how AI can transform education,
the arts, research, and creative economies.
TeachAI: Educating the AI generation NYSCA: Finding the intersection of art and AI
TeachAI is an initiative led by Code.org, ETS, ISTE, Khan New York State Council on the Arts (NYSCA) provides
Academy, and the World Economic Forum. It brings grants and other support to advance their mission to
together public and private education leaders and “foster and advance the full breadth of New York State’s
technology experts to help create policy guidance arts, culture, and creativity for all.”37 As one of NYSCA’s
and resources about the safe, effective, and responsible values is “the constant evolution of artmaking and creative
usage and teaching of AI in schools.35 The resulting practice,” NYSCA is interested in the way AI is shaping
AI Guidance for Schools Toolkit helps education system the landscape of art and artists in New York State. NYSCA
leaders create guidance, includes seven principles highlights the complex impact AI can have on artists and
for AI in education, and recommends strategies for the importance of having artists be a part of conversations
engaging parents, staff, and student stakeholders. about AI development and policy. NYSCA also supports
organizations at the intersection of technology and art,
Pat Yongpradit, Chief Academic Officer of Code.org which can be a part of educating artists about AI, such as
and Lead of TeachAI has said “My sincere hope is that the Buffalo Center for Arts and Technology, which provides
teachers feel guided and supported by their leaders mentorship, tutoring, and workforce development.
as we all adapt to the changes AI brings to education.”36
TeachAI is an example of convening many thought
partners and stakeholders together to advance
change, as they have brought together private sector
companies, national and state government agencies, Etsy: Keeping commerce human
and policy groups to advance guidance and frameworks.
Etsy has long been a leader in leveraging AI and machine
learning to craft a uniquely human shopping experience
that connects its community of creative entrepreneurs
with tens of millions of passionate buyers around the
Empire AI: Investing in research for public good world. To further its mission to Keep Commerce Human
while embracing cutting edge technology, Etsy created
Empire AI, a consortium of seven New York-based world- a Responsible AI Working Group to govern its exploration
class research institutions, is making strides to secure of AI. Etsy leverages AI to help sellers more effectively
New York’s place at the forefront of AI research. The grow their businesses, surface more relevant and inspiring
consortium will create a first-in-the-nation, research- items to buyers that help drive more sales for sellers,
focused AI computing center, powered by clean and improve the shopping and selling experience for the
hydropower. The center will provide grant researchers Etsy community. Etsy’s goal in this work is to leverage
across the state access to essential computing AI to Keep Commerce Human, while upholding the values
resources, catalyzing innovation, fostering recruitment of of respect, fairness, reliability, transparency, privacy, and
global tech talent, and advancing AI for the public good. security when advancing the adoption of AI.
15 Whitepaper | Template Month Year
Responsible AI
16 Whitepaper | Template Month Year
Responsible AI
Responsible AI is For the purpose of this report, the ETAB defines AI as the
simulation of human intelligence processes by machines.
the throughline of While AI offers incredible potential, it could pose a risk if not
pursued responsibly. Deploying AI responsibly across New
York State is critical to the safety of all New Yorkers and New
the Advisory Board’s
York businesses. Responsible AI is defined as an approach
to designing, developing, assessing, and deploying AI in a safe,
recommendations
trustworthy, and ethical way. Responsible AI encompasses
the following principles:
1. Fairness and inclusiveness
AI systems avoid bias, treat everyone fairly, and
avoid affecting distinct groups differently, with an
emphasis on ensuring that community voice actively
contributes to AI creation to ensure it addresses,
IBM: Implementing ethics standards and does not exacerbate, the most pressing challenges.
IBM’s AI Ethics Board is the lynchpin of its responsible 2. Transparency and traceability
technology efforts and infuses IBM’s principles into Users understand how and why AI systems function
business and product decision-making. The AI Ethics the way they do so they can determine appropriate
Board is steered by senior leaders from across the use cases and identify potential limitations, which
company, supported by a strong advocacy network and can include an emphasis on “human-in-the-loop”
AI Ethics Focal Points within various business units. In (HIL) design.
addition to actively supporting the principles, the AI
Ethics Board shares thought leadership around emerging 3. Reliability and safety
issues, and in 2023, published various white papers, AI systems operate reliably, safely, and consistently,
including “Augmenting Human Intelligence–the IBM handling exceptional conditions.
Point of View” and “Foundation Models: Opportunities,
Risks and Mitigations.” The AI Ethics Board is one 4. Governance and accountability
component of IBM’s Integrated Governance Program, Developers, organizations, and policymakers take
which allows the organization to adapt many existing ownership of responsible deployment of AI.
processes to address new AI requirements and obligations.
5. Privacy and security
AI systems are continually updated to comply with
data protection protocols about the collection, use,
storage, and disclosure of data.
Kitware: Advancing explainable AI 6. Sustainability
AI systems achieve beneficial outcomes for people
Kitware is at the forefront of ethical AI research, and the planet.
developing methods and leading studies on how AI can
be trusted and how it can be harnessed to benefit society The Advi |
308 | mit_edu | The_20Simple_20Macroeconomics_20of_20AI.pdf | ∗
The Simple Macroeconomics of AI
Daron Acemoglu
Massachusetts Institute of Technology
April 5, 2024
Abstract
This paper evaluates claims about the large macroeconomic implications of new advances in AI. It
starts from a task-based model of AI’s effects, working through automation and task complementarities. It
establishesthat,solongasAI’smicroeconomiceffectsaredrivenbycostsavings/productivityimprovements
at the task level, its macroeconomic consequences will be given by a version of Hulten’s theorem: GDP and
aggregateproductivitygainscanbeestimatedbywhatfractionoftasksareimpactedandaveragetask-level
cost savings. Using existing estimates on exposure to AI and productivity improvements at the task level,
these macroeconomic effects appear nontrivial but modest—no more than a 0.71% increase in total factor
productivity over 10 years. The paper then argues that even these estimates could be exaggerated, because
early evidence is from easy-to-learn tasks, whereas some of the future effects will come from hard-to-learn
tasks, where there are many context-dependent factors affecting decision-making and no objective outcome
measures from which to learn successful performance. Consequently, predicted TFP gains over the next 10
yearsareevenmoremodestandarepredictedtobelessthan0.55%. IalsoexploreAI’swageandinequality
effects. IshowtheoreticallythatevenwhenAIimprovestheproductivityoflow-skillworkersincertaintasks
(without creating new tasks for them), this may increase rather than reduce inequality. Empirically, I find
that AI advances are unlikely to increase inequality as much as previous automation technologies because
their impact is more equally distributed across demographic groups, but there is also no evidence that AI
willreducelaborincomeinequality. AIisalsopredictedtowidenthegapbetweencapitalandlaborincome.
Finally, some of the new tasks created by AI may have negative social value (such as design of algorithms
for online manipulation), and I discuss how to incorporate the macroeconomic effects of new tasks that may
have negative social value.
JEL Classification: E24, J24, O30, O33.
Keywords: Artificial Intelligence, automation, ChatGPT, inequality, productivity, technology adop-
tion, wage.
∗PaperpreparedforEconomic Policy. IamgratefultoCanYe¸sildereforphenomenalresearchassistance,
to Leonardo Bursztyn, Mert Demirer, Lauren Fahey, Shakked Noy, Sida Peng, Julia Regier, and Whitney
Zhangforusefulcomments,andtoparticipantsintheEconomicPolicy conferenceandmydiscussantsthere,
DavidH´emousandBenoˆıtCoeur´e,forcommentsandsuggestions. IthankPamelaMishkinandDanielRock
for generously sharing their data on AI exposure. I am also heavily indebted to my collaborators on several
projects related to these topics, David Autor, Simon Johnson and Pascual Restrepo, from whom I learned a
great deal and who have also given me very useful comments on the current draft. All remaining errors are
mine. The online Appendix is available upon request.
1 Introduction
Artificial intelligence (AI) has captured imaginations. Promises of rapid, even unparalleled,
productivity growth as well as new pathways for complementing humans have become com-
monplace. There is no doubt that recent developments in generative AI and large language
modelsthatproducetext, informationandimages—andShakespeareansonnets—inresponse
to simple user prompts are impressive and even spellbinding. ChatGPT, originally released
on November 30, 2022, soon became the fastest spreading tech platform in history, with an
estimated 100 million monthly users only two months after launch.
AI will have implications for the macroeconomy, productivity, wages and inequality, but
all of them are very hard to predict. This has not stopped a series of forecasts over the last
year, often centering on the productivity gains that AI will trigger. Some experts believe
thattrulytransformativeimplications, includingartificialgeneralintelligence(AGI)enabling
AI to perform essentially all human tasks, could be around the corner.1 Other forecasters
are more grounded, but still predict big effects on output. Goldman Sachs (2023) predicts
a 7% increase in global GDP, equivalent to $7 trillion, and a 1.5% per annum increase in
US productivity growth over a 10-year period. Recent McKinsey Global Institute (2023)
forecasts suggest that generative AI could offer a boost as large as $17.1 to $25.6 trillion to
the global economy, on top of the earlier estimates of economic growth from increased work
automation. They reckon that the total impact of AI and other automation technologies
could produce up to a 1.5 − 3.4 percentage point rise in average annual GDP growth in
advanced economies over the coming decade.2
Are such large effects plausible? And if there are going to be productivity gains, who will
be their beneficiary? With previous automation technologies, such as robotics, most gains
1Korinek and Suh (2024) predict a “baseline” GDP growth of 100% over the next 10 years, and also
entertain the possibility of much higher “aggressive” AGI growth rates, such as a 300% increase in GDP.
Many others are seeing recent developments as a confirmation of the forecasts in Kurzweil (2005) about the
impending arrival of “singularity” and “explosive” economic growth (Davidson, 2021).
2Three caveats are in order. First, although most recent advances are in generative artificial intelligence,
the economic forces explored here apply to other types of AI, and estimates of exposed tasks I use come on
the basis of anticipated improvements in a range of AI-related technologies, including computer vision and
software building on large language models. Hence, I consider the numbers here to apply to all of artificial
intelligence and thus typically refer to “AI”, unless there is a reason to emphasize generative AI.
Second, I focus on the US economy because much of the existing evidence on microeconomic effects of
AI and prevalence of exposed tasks is from the United States. The impact on other industrialized nations
should be similar, whereas the consequences for the developing world are harder to ascertain and require
much more in-depth research.
Third,somecommentatorsuse“productivity”torefertooutputperworker(oraveragelaborproductivity),
while others mean total factor productivity (TFP). Throughout, I distinguish between aggregate TFP and
GDP effects, and I use productivity improvement at the micro/task level as synonymous to cost savings.
1
accrued to firm owners and managers, while workers in impacted occupations experienced
negative outcomes (e.g., Acemoglu and Restrepo, 2020a). Could it be different this time?
Some experts and commentators are more optimistic. A few “proof-of-concept” experimen-
tal studies document nontrivial productivity gains from generative AI, largely driven by
improvements for less productive or lower-performing workers (e.g., Peng et al., 2023; Noy
and Zhang, 2023; Brynjolfsson et al., 2023), and this has prompted some experts to be cau-
tiously optimistic (Autor, 2024), while others are forecasting a “blue-collar bonanza” (The
Economist, 2023).
This paper uses the framework from Acemoglu and Restrepo (2018, 2019b, 2022) to pro-
vide some insights for these debates, especially relevant for the medium-term (about 10-year)
macroeconomic effects of AI. I build a task-based model, where the production of a unique
final good requires a series of tasks to be performed, and these tasks can be allocated to ei-
ther capital or labor, which have different comparative advantages. Automation corresponds
to the expansion of the set of tasks that are produced by capital (including digital tools and
algorithms). In this framework, AI-based productivity gains—measured either as growth of
average output per worker or as total factor productivity growth—can come from a number
of distinct channels (see Acemoglu and Restrepo, 2019a):
• Automation (or more precisely extensive-margin automation) involves AI models tak-
ing over and reducing costs in certain tasks. In the case of generative AI, various
mid-level clerical functions, text summary, data classification, advanced pattern recog-
nition, and computer vision tasks are among those that can be profitably automated.
• Task complementarity can increase the productivity in tasks that are not fully au-
tomated and may even raise the marginal product of labor. For example, workers
performing certain tasks may have better information or access to other complemen-
tary inputs. Alternately, AI may automate some subtasks, while at the same time
enabling workers to specialize and raise their productivity in other aspects of their job.
• Deepening of automation can take place, increasing the productivity of capital in tasks
that have already been automated. For example, an already-automated IT security
task may be performed more successfully by generative AI.
• New tasks may be created thanks to AI and these tasks may impact the productivity
of the whole production process.3
3Newtasksinthisframeworkalsocapturethepossibilityofproductivity-enhancingreorganizingproduc-
2
In this paper, I focus on the first two channels, though I also discuss how new tasks en-
abled by AI can have positive or negative effects. I do not dwell on deepening of automation,
because the tasks impacted by (generative) AI are quite different than those automated by
the previous wave of digital technologies, such as robotics, advanced manufacturing equip-
ment and software systems.4 I also do not discuss how AI can have revolutionary effects by
changing the process of science (a possibility illustrated by new crystal structures discovered
by the Google subsidiary DeepMind and recent neural network-enabled advances in protein
folding), because large-scale advances of this sort do not seem likely within the 10-year time
frame and many current discussions focus on automation and task complementarities.
I show that when AI’s microeconomic effects are driven by cost savings (equiva-
lently, productivity improvements) at the task level—due to either automation or task
complementarities—its macroeconomic consequences will be given by a version of Hulten’s
theorem: GDP and aggregate productivity gains can be estimated by what fraction of tasks
are impacted and average task-level cost savings. This equation disciplines any GDP and
productivity effects from AI. Despite its simplicity, applying this equation is far from trivial,
because there is huge uncertainty about which tasks will be automated or complemented,
and what the cost savings will be.
Nevertheless, as an illustrative exercise, I use data from a number of recent studies, in
particular, Eloundou et al. (2023) and Svanberg et al. (2024), as well as the experimental
studies mentioned above, to obtain some back-of-the-envelope numbers. Eloundou et al.
(2023) provide the first systematic estimates of what tasks will be impacted by generative
AI and computer vision technologies. Their methodology does not fully distinguish whether
the impact will take the form of automation or task complementarities, and does not provide
information on when we expect these impacts to be realized and how large their cost savings
will be.5 For computer vision technologies, Svanberg et al. (2024) provide estimates of what
fraction of tasks that are potentially exposed to AI can be feasibly automated in different
tion. The role of AI in enabling such reorganization is emphasized by, among others, Bresnahan (2019) and
Agrawal et al. (2023).
4Eloundou et al. (2023) report negative statistical associations between their measure of exposure to AI,
which I use below, and measures of exposure to robots and manual routine tasks.
5Morespecifically,IusethemostgranularinformationthatEloundouetal.(2023)present,whichistheir
“automation index”, coded with help from GPT-4. This index provides information on how much of the
activities involved in a task/occupation can be performed by AI. Although this index has somewhat greater
emphasis on automation, it does not systematically distinguish between automation and task complemen-
tarities. As I discuss further below and Eloundou et al. (2023) themselves note, their exposure measure
often captures the possibility that generative AI and related digital technologies can perform some of the
subtasks in an occupation, providing more time for workers to focus on other activities, and thus contains
both automation and task complementarity elements.
3
time frames.
I take Eloundou et al.’s estimates of tasks that are exposed to AI (without distinguishing
automation vs. task complementarities). I then aggregate this to the occupational level and
weight the importance of each occupation by its wage bill share in the US economy. This
calculation implies that 19.9% of US labor tasks are exposed to AI. I then use Svanberg et
al.’s estimate for computer vision tasks that, among all exposed tasks, 23% can be profitably
performed by AI (for the rest, the authors estimate that the costs would exceed the benefits).
I take the average labor cost savings to be 27%—the average of the estimates in Noy and
Zhang (2023) and Brynjolfsson et al. (2023)—and turn this into total cost savings using
industry labor shares, which imply an average total cost savings of 15.4%.
This calculation implies that total factor productivity (TFP) effects within the next 10
years should be no more than 0.71% in total—or approximately a 0.07% increase in TFP
growth annually. If we add bigger productivity gains from Peng et al. (2023), which are less
likely to be broadly applicable, or incorporate further declines in GPU costs, this number
still remains around 1%.
To turn these numbers into GDP estimates, we need to know how much the capital stock
will increase due to AI. I start with the benchmark of a rise in the capital stock proportional
to the increase in TFP. This benchmark is consistent with the fact that generative AI does
not seem to require huge investments by users (beyond those made by designers and trainers
of the models). With these investment effects incorporated, GDP is also estimated to grow
by around 1.1% over the next 10 years. When I assume that investments will be similar
to those for earlier automation technologies and use the full framework from Acemoglu and
Restrepo (2022) to estimate the increase in the capital stock, the upper bound on GDP
effects rises to 1.6 − 1.8%. Nevertheless, my framework also clarifies that what is relevant
for consumer welfare is TFP, rather than GDP, since the additional investment comes out
of consumption.6
I then argue that the numbers above may be overestimates of the aggregate productivity
benefits from AI, because existing estimates of productivity gains and cost savings are in
tasks that are “easy-to-learn”, which then makes them easy for AI. In contrast, some of
the future effects will come from “hard-to-learn” and hard for AI tasks, where there are
many context-dependent factors affecting decision-making, and most learning is based on the
behaviorofhumanagentsperformingsimilartasks(ratherthanobjectiveoutcomemeasures).
6For example, if AI models continue to increase their energy requirements, this would contribute to
measured GDP, but would not be a beneficial change for welfare.
4
Productivity gains in these hard tasks will be less—though, of course, it is challenging to
determine exactly how much less. Using a range of (speculative) assumptions, I estimate
an upper bound of 74% easy tasks among Eloundou et al.’s exposed tasks. I suppose that
productivity gains in hard tasks will be approximately one quarter of the easy ones. This
leads to an updated, more modest increase in TFP and GDP in the next 10 years that can
be upper bounded by 0.55% and 0.90%, respectively.
New tasks created with AI can more significantly boost productivity. However, some of
the new AI-generated tasks are manipulative and may have negative social value, such as
deepfakes, misleading digital advertisements, addictive social media or AI-powered malicious
computer attacks. While it is difficult to put numbers on good and bad new tasks, based
on recent research I suggest that the negative effects from new bad tasks could be sizable.
I make a very speculative attempt using numbers on the negative welfare effects of social
mediafromarecentpaperbyBursztynetal.(2023). Theseauthorsfindthatconsumershave
positive willingness to pay for using social media (in particular Instagram and TikTok) when
others are using it, but they would prefer that neither themselves nor others use it. Roughly
speaking, their estimates imply that revenue can increase by about $53 per user-month,
but this has a negative impact on total GDP/welfare equivalent to $19 per user-month.
Combining these numbers with an estimate of the fraction of activities that may generate
negative social value (in practice, revenues from social media and spending on attack-defense
arms races in IT security), I suggest that with more intensive use of AI, it is possible to have
nontrivial increases in GDP. For example, AI may appear to increase GDP by 2%, while in
reality reducing welfare by −0.72%.
I also explore AI’s wage and inequality effects. My framework implies that productivity
gains from AI are unlikely to lead to sizable wage rises. Moreover, even if AI improves
the productivity of low- and middle-performing workers (or workers with limited expertise
in complex tasks), I argue that this may not translate into lower inequality. In fact, I
show by means of a simple example how an increase in the productivity of low-skill workers
in certain tasks can lead to higher rather than lower inequality. Adapting the general
equilibrium estimates from Acemoglu and Restrepo (2022) to the setting of AI, I find that
the more intensive use of AI is unlikely to lead to substantial wage declines for affected
groups, because AI-exposed tasks are more evenly distributed across demographic groups
than were the tasks exposed to earlier waves of automation. Nevertheless, I estimate that
AI will not reduce inequality and is likely to have a negative effect on the real earnings of
5
low-education women (especially white, native-born women). My findings also suggest that
AI will further expand the gap between capital and labor income as a whole.
Finally, I argue that as originally suggested in Acemoglu and Restrepo (2018), more
favorable wage and inequality effects, as well as more sizable productivity benefits, will
likely depend on the creation of new tasks for workers in general and especially for middle-
and low-pay workers. While this is feasible in theory and I have argued elsewhere how it
could be achieved (Acemoglu, 2021 and Acemoglu et al., 2023), I also discuss why this does
not seem to be the focus of artificial intelligence research at the moment.
In sum, it should be clear that forecasting AI’s effects on the macroeconomy is extremely
difficult and will have to be based on a number of speculative assumptions. Nevertheless,
the gist of this paper is that a simple framework can discipline our thinking and forecasts,
and if we take this framework and existing estimates seriously, it is difficult to arrive at very
large macroeconomic gains.
The rest of the paper is organized as follows. The next section outlines the concep-
tual framework I use throughout the paper and derives a number of theoretical insights
on aggregate productivity gains, investment responses, and wage and inequality effects. It
also discusses the crucial distinction between easy-to-learn and hard-to-learn tasks and their
productivity implications, and introduces the contrast between good and bad new tasks.
Section 3 provides a preliminary quantitative analysis of new AI breakthroughs within this
framework. It first presents a baseline (upper bound) estimate on the basis of the fraction
of existing tasks that are likely to be impacted by AI within the next 10 years and existing
estimates of cost savings (productivity improvements) from AI. It then refines this estimate
by introducing the distinction between easy-to-learn and hard-to-learn tasks and undertakes
a preliminary classification of AI-exposed tasks into the easy and hard categories. I then
make an even more speculative attempt at incorporating the macroeconomic implications of
bad new tasks into this framework. Finally, I report estimates on the wage and inequality
implications of recent AI advances. Section 4 concludes with a general discussion, while the
Appendix, which is available upon request, includes additional information on how tasks are
classified into exposed and non-exposed and easy-to-learn and hard-to-learn categories.
2 Conceptual Framework
The model here builds on Acemoglu and Autor (2011) and Acemoglu and Restrepo (2018,
2019b, 2022), and I focus on the main elements of the framework, referring the reader to
6
these papers for further details and refinements. The economy is static and involves the
production of a unique final good, and all markets are competitive.7
The production of a unique final good takes place by combining a set of tasks, with
measure N, using the following production function
(cid:18)(cid:90) N (cid:19) σ−σ 1
σ−1
Y = B(N) y(z) dz , (1)
σ
0
where Y(z) denotes the output of task z for z ∈ [0,N], σ ≥ 0 is the elasticity of substitution
between tasks and the parameter B(N) depends on N to capture the possible system-wide
effects of new tasks, though in what follows I will suppress this dependence to simplify the
notation. For now, the elasticity σ can take any value, but it is reasonable to presume σ ≤ 1,
so that tasks are gross complements. I later set the elasticity of substitution between tasks
to σ (cid:39) 0.5, as estimated by Humlum (2023) and also imposed in Acemoglu and Restrepo
(2022).
Tasks can be produced using capital or labor according to the production function
y(z) = A γ (z)l(z)+A γ (z)k(z) for any z ∈ [0,N],
L L K K
where A and A are labor-augmenting and capital-augmenting productivity terms, γ (z)
L K L
and γ (z) are labor’s and capital’s task-specific productivity schedules, and l(z) and k(z)
K
denote labor and capital allocated to performing task z. This task production function
implies that capital and labor have different productivities in different tasks, but within a
task they are perfect substitutes.8
Throughout, Iassumethatγ (z)/γ (z)isincreasinginz, sothatlaborhasacomparative
L K
advantage in higher-indexed tasks. This implies that there exists some threshold I such that
tasks z ≤ I are produced with capital and those above this threshold are produced with
labor.
7Acemoglu and Restrepo (2018) provide a dynamic version of this economy with capital accumulation
and endogenous technological choices, while Acemoglu and Restrepo (2022) provide a generalization with
multiple types of labor and multiple sectors, and Acemoglu and Restrepo (2023) consider a non-competitive
version of this economy. Extending the framework in any of these directions has no effect on the results and
implications I explore here.
8One important simplification is to assume that tasks assigned to labor do not require any capital or
tools, which is clearly unrealistic. The online Appendix of Acemoglu and Restrepo (2018) shows that the
results are very similar if the task production function is modified such that:
y(z)=A γ (z)(cid:2) l(z)1−κk (z)κ(cid:3) +A γ (z)k(z),
L L C K K
where κ ∈ (0,1) and k (z) is labor-complementary capital in task z (while k(z) denotes capital used for
C
automating task z). Because κ < 1, tasks assigned to labor are still less intensive in capital than are
fully-automated tasks.
7
I normalize the total population to 1 and assume that different workers have different
unitsofeffectivelabor. Tosimplifythediscussion, Iassumethattherearetwotypesoflabor,
high-skill and low-skill, and there is no comparative advantage difference between these two
typesoflabor(Ireturntocomparativeadvantagelater). Theonlydifferenceisthathigh-skill
labor, which makes up a fraction φH of the population, has λH units of effective labor, while
the remaining φL = 1−φH low-skill labor has only λU < λH units of effective labor. This
specification ensures that both high-skill and low-skill workers could be performing some of
the same tasks. It also implies that wage inequality is pinned down by λH/λU—a feature I
relax later.
I also assume that all labor is supplied inelastically, so I write the total supply of labor
as
φUλU +φHλH = L.
The labor market-clearing condition is
(cid:90) N
L = l(z)dz, (2)
0
and I denote the wage rate by w.
Capital is specialized for the tasks in which it is used, and I assume that capital of type
z is produced linearly from the final good with unit cost
R(z) = R(K)ρ(z), (3)
where
(cid:90) N
K = k(z)dz
0
is the overall capital stock of the economy. All firms take the cost of capital for task z, R(z),
as given. The first term in (3) implies that the required rate of return on overall capital can
increase when the capital stock of the economy is larger and the second term is task-specific,
representing the possibility that different types of capital could have different costs. For
tasks that are not yet technologically automated—meaning that they cannot be produced
by capital—we can either set γ (z) = 0 or take ρ(z) to be very large.
K
Finally, I assume that there exists a (non-satiated) representative household that con-
sumes the final good (net of capital expenditures).
2.1 Equilibrium
I focus on a competitive equilibrium, which satisfies the following usual conditions:
8
• The allocation of tasks z ∈ [0,N] is cost-minimizing. That is, task z ∈ [0,N] is
produced by labor if and only if
w R(z)
< .
A γ (z) A γ (z)
L L K K
• The amount of capital k(z) is chosen to maximize Y −R(z)k(z), where Y is given as
in (1) and the overall capital stock of the economy K is taken as given.
• The labor market clears. That is, (2) holds.
Notice that the first condition imposes an innocuous tie-breaking rule that when indiffer-
ent, firms use capital for performing a task. Given this tie-breaking rule, all tasks z > I will
be performed by labor (i.e., l(z) = 0 for all z ≤ I and k(z) = 0 for all z > I). Whether this
is high- or low-skill labor is indeterminate in the baseline model, so I focus on the overall
amount of effective labor units.
In a competitive equilibrium, all tasks performed by labor must have
Bσ− σ1 A Lσ− σ1 γ L(z)σ− σ1 l(z)− σ1 Y σ1 = w. (4)
This implies that for any two tasks z > I and z(cid:48) > I,
l(z) γ (z)σ−1
L
= . (5)
l(z(cid:48)) γ (z(cid:48))σ−1
L
Notice that when σ < 1, less labor is allocated to tasks in which labor’s productivity is
higher—a feature whose implications I will emphasize later. Equation (5), combined with
the labor market-clearing condition (2), implies
γ (z)σ−1
L
l(z) = L. (6)
(cid:82)N
γ (z)σ−1dz
I L
Moreover, with a similar reasoning for any task z < I, only capital is used, and the
first-order condition for capital is simply
Bσ− σ1 A Kσ− σ1 γ K(z)σ− σ1 k(z)− σ1 Y σ1 = R(K)ρ(z). (7)
Combining (6) and (7) with (1), GDP or total output can be written as
σ
(cid:16) (cid:17)1 σ−1
(cid:82)N γ (z)σ−1dz σ (BA L)σ−1
I L L σ
Y = (cid:18) (cid:19) . (8)
1− (cid:82)I (cid:16) γK(z) (cid:17)σ−1 dz Aσ−1Bσ2−1
0 R(K)ρ(z) K σ
9
The denominator here is due to the roundabout nature of production, and I assume that
(cid:32) (cid:33)
(cid:90) I (cid:18) γ K(z) (cid:19)σ−1
dz
Aσ−1Bσ2−1
< 1 (9)
σ
R(K)ρ(z) K
0
to ensure that output is finite in this economy. (Otherwise, because output linearly produces
machines, which then produce output, overall output can reach infinity). With an identical
argument to that in Acemoglu and Restrepo (2022), an equilibrium exists and is unique,
provided that (9) is satisfied.
2.2 How AI Could Affect Production
Before completing the characterization of equilibrium, I discuss how AI could affect produc-
tion in this economy.
1. AIenablesfurther(extensive-margin)automation,increasingI. Suchautomationcould
be triggered either because AI reduces the cost of capital for some marginal tasks
(i.e., tasks slightly above I) or increases the effectiveness of machinery or algorithms
performing some marginal tasks, thus raising γ (z) for some z above I. Obvious
K
examples of this type of automation include generative AI tools such as large language
models (LLMs) taking over simple writing, translation and classification tasks as well
assomewhatmorecomplextasksrelatedtocustomerserviceandinformationprovision,
or computer vision technologies taking over image recognition and classification tasks.
2. AI can generate new task complementarities, raising the productivity of labor in tasks
it is performing. For example, AI could provide better information to workers, directly
increasing their productivity. This possibility could be modeled as AI reducing the cost
of complementary capital k (z) in some tasks z > I in the more general formulation in
C
footnote 8. Alternatively, AI could automate some subtasks (such as providing ready-
made subroutines to computer programmers) and simultaneously enable humans to
specialize in other subtasks, where their performance improves. This channel would
requiretheexplicitmodelingoftherangeofsubtasksmakingupeachtask. Inthiscase,
new AI technologies would perform some of these subtasks and do so with sufficiently
high productivity, so that the subtask-level displacement would be weaker than the
productivity gains, expanding the demand for labor and the marginal productivity
of labor in these tasks. The logic of the productivity effect being larger than the
displacement effect is the same as in the basic models of automation, as exposited
10
in Acemoglu and Restrepo (2018, 2019b). Even more interestingly, AI may enable
workers to specialize in the non-automated subtasks and raise their expertise in these
activities (e.g., when humans spend less time in writing standard subroutines, they
can become better at other parts of programming). I represent task complementarities
by an increase in γ (z) in some tasks z ≤ I, or when they happen in all tasks, by an
L
increase in A .
L
3. AI could induce deepening of automation—meaning improving performance, γ (z), or
K
reducingcosts, ρ(z), insomepreviouslycapital-intensivetasks(tasksz ≤ I). Examples
include IT security, automated control of inventories, and better automated quality
control (see Acemoglu and Restrepo, 2019a).
4. AIcangeneratenew labor-intensive products or tasks, whichcorrespondstoanincrease
in N. As argued in Acemoglu and Restrepo (2020b), Acemoglu (2021) and Acemoglu
et al. (2023), there are many pathways for such new tasks. Later I discuss the case
where some of these new products and tasks can be manipulative and have negative
social value.
The effects of new AI tools will depend on the extent of each one of these effects, and I
will try to provide more specificity on these possibilities later. In the rest of this section, I
will derive the consequences of different effects of AI.
2.3 Equilibrium Wages and Comparative Statics
As a first step, let us combine (4) and (6), so that the equilibrium wage can be expressed as
w =
(cid:18) Y (cid:19) σ1
(BA
)σ−1
(cid:18)(cid:90) N
γ
(z)σ−1dz(cid:19) σ1
. (10)
L σ L
L
I
This equation is intuitive. The first term shows that the wage is proportional to labor
productivity (raised to the power 1/σ), and the second term captures the contribution to
the marginal productivity of labor coming from Hicks-neutral and labor-augmenting tech-
nologies, while the third term represents the contribution of the allocation of tasks to the
marginal productivity of labor. The effect of any small technological change (potentially
altering multiple dimensions of the production technology, such as B; A and A ; γ (z) and
L K L
γ (z); and I and N) can then be written as:
K
1 (cid:18) Y (cid:19) σ −1 1 (cid:18)(cid:90) N (cid:19)
dlnw = dln + (dlnB +dlnA )+ dln γ (z)σ−1dz . (11)
L L
σ L σ σ
I
11
The effect of an extensive-margin automation—an increase in I—is given by
dlnw 1 dlnY 1 γ (I)σ−1
L
= − .
(cid:16) (cid:17)
dI σ dI σ (cid:82)N
γ (z)σ−1dz
I L
Ingeneral, thisexpressionhasambiguoussign, soautomationcanreducewages. Morespecif-
ically, there are two opposing effects (Acemoglu and Restrepo, 2018, 2019b): (a) automation
always produces a positive effect on wages (and labor demand) because it increases produc-
tivity (or equivalently, reduces costs). This positive productivity effect is represented by
the first term; (b) simultaneously, automation displaces workers from the tasks they used to
perform. The negative displacement effect is represented by the second term. In the special
case where R(K) is constant, it can be verified that automation increases wages. This is not
the case, in general, when R(K) is increasing, |
309 | mit_edu | pdf.pdf | Harvard Data Science Review • Special Issue 5: Grappling With the
Generative AI Revolution
Institutional Eorts to
Help Academic
Researchers Implement
Generative AI in Research
1 1,2
Jing Liu H. V. Jagadish
1Michigan Institute for Data Science, University of Michigan, Ann Arbor, Michigan, United States of
America,
2Division of Computer Science and Engineering, College of Engineering, University of Michigan,
Ann Arbor, Michigan, United States of America
Published on: Feb 26, 2024
DOI: https://doi.org/10.1162/99608f92.2c8e7e81
License: Creative Commons Attribution 4.0 International License (CC-BY 4.0)
Harvard Data Science Review • Special Issue 5: Grappling With the Institutional Eorts to Help Academic Researchers Implement
Generative AI Revolution Generative AI in Research
ABSTRACT
The scale and speed of the generative AI (artificial intelligence) revolution, while offering unprecedented
opportunities to advance science, is also challenging the traditional academic research model in fundamental
ways. The academic research model and academic institutions are not set up to be nimble in the face of rapidly
advancing technologies, and the task of adopting such new technologies usually falls on individual researchers.
Excitement about the opportunities that generative AI brings is leading to a rush of researchers with various
levels of technical expertise and access to resources to adopt this new technology, which could lead to many
researchers ‘reinventing the wheel’ and research outcomes lacking in ethics, rigor, and reproducibility. This
problem not only applies to generative AI, but could also be true for other upcoming and similarly disruptive
technologies. We argue that the current norm of relying on individual researchers for new technology adoption
is no longer adequate. It is time that academic institutions and their research organizations such as our own (the
Michigan Institute for Data Science) develop new mechanisms to help researchers adopt new technologies,
especially those that cause major seismic shifts such as generative AI. We believe this is essential for helping
academic researchers stay at the forefront of research and discovery, while preserving the validity and
trustworthiness of science.
Keywords: institutional transformation, best practices, training, academic researcher, rigor and reproducibility,
institutional support
1. Problem Statement and Solution Proposition
Generative AI (artificial intelligence) is a type of AI algorithm that can generate new content (such as text,
images, audio, video, and other modalities) that is statistically probable based on the data that the algorithm is
trained on (Bommasani et al., 2021; Cao et al., 2023; Dwivedi et al., 2023; Gozalo-Brizuela & Garrido-
Merchan, 2023; Vaswani et al., 2017). Compared to other types of AI technology, such as natural language
processing, generative AI is based on newer AI architectures, most notably transformers and diffusion models,
trained on enormous volumes of (sometimes multimodal) data in their natural forms (such as raw texts and
images from the internet) without the need of labeling the training data. Generative AI thus opens up enormous
possibilities to revolutionize how AI assists humans in all types of activities that involve interacting with a
computer.
1.1. Opportunities with Generative AI
The emergence of generative AI has tantalized academic researchers with its potential to vastly accelerate
research, and even to enable new research, in multiple ways (Boyko et al., 2023; Dwivedi et al., 2023;
Microsoft Research AI4Science and Microsoft Azure Quantum, 2023; Morris, 2023; Wang et al., 2023).
2
Harvard Data Science Review • Special Issue 5: Grappling With the Institutional Eorts to Help Academic Researchers Implement
Generative AI Revolution Generative AI in Research
1. The use of domain-agnostic generative tools (such as text and image generation) to improve research
productivity, by assisting with routine tasks such as drafting and editing emails and manuscripts, checking
for compliance, and facilitating the communication with the lay audience.
2. The use of domain-agnostic generative AI to enhance the research expertise of individual researchers and
research teams. This includes summarizing and representing knowledge within disciplines, gathering
interdisciplinary insights, and supporting communication for interdisciplinary collaboration.
3. The use of domain-agnostic and domain-specific generative AI to accelerate and automate the research
process, such as data cleaning, formatting, and imputation; suggesting research hypotheses and selecting
experimental parameters; coding, data analysis, and visualization.
4. The use of domain-specific generative models, such as for aerospace engineering or protein structure
models, to enable new paths for research discovery.
Such possibilities are fueling researchers’ enthusiasm for incorporating generative AI in research, even though
most of generative AI’s potential benefits for research remain to be tested and validated. Of the four types of
generative AI use that we mention, the use of domain-specific models has been reported extensively (as
examples, see Andrade & Walsh, 2023; Chenthamarakshan et al., 2023; Grisoni et al., 2021; Gu et al., 2023;
Hie et al., 2023; Madani et al., 2023; Zeng et al., 2022). But successes of the first three types of generative AI
use in research are only beginning to be reported (see the following examples: Boiko et al., 2023; Ciucă &
Ting, 2023; Jablonka et al., 2023; Lyu et al., 2023; Mahjour et al., 2023). This enthusiasm is also accompanied
by a lack of preparedness among researchers. In the academic research environment, many faculty members
have no concrete idea about how to implement generative AI in their research, or even how to work with
generative AI at all, including simply using prompts to query information.
Many also do not know a good starting point because new generative AI tools emerge almost daily and there is
not an obvious path of skills progression. A survey that we conducted in November of 2023 of 60 faculty
affiliates of the Michigan Institute for Data Science (MIDAS) (Table 1) gives us a glimpse of this picture. Only
12% of the respondents have the expertise to train their own generative AI models; fewer than one-third can
run existing models or fine-tune models. Even after ChatGPT, which is supposed to be an easy-to-use tool,
became available for almost a year, half of all respondents are not able to use prompts with ChatGPT to obtain
good results. The faculty members’ biggest need is to develop skills through training and learning from peers.
This closely mirrors a brief survey that we conducted in the summer of 2023 with MIDAS faculty, in which
70% of the 92 respondents indicated that they had no knowledge or only conceptual understanding (as opposed
to hands-on practice) with generative AI. We believe this is representative of the academic research scene at
this moment across institutions.
Table 1. Faculty researchers’ expectations and current preparedness for Generative AI to aid
academic research (n = 60).
3
Harvard Data Science Review • Special Issue 5: Grappling With the Institutional Eorts to Help Academic Researchers Implement
Generative AI Revolution Generative AI in Research
How do you want to use generative AI in your research? Improving productivity (drafting documents, summarizing
documents, etc): 72%
Coding: 63%
Data analysis and modeling: 55%
Communication (email, presentation, etc.): 45%
Helping with data generation, processing, and documentation: 38%
What is the skill level in your research group with regard to Can use things like ChatGPT with prompts, but not using them
generative AI? well yet: 47%
Can use things like ChatGPT with prompts and can get some good
results: 48%
Can run existing models: 28%
Can fine tune existing models: 22%
Can train models: 12%
What support is important for you to use generative AI in your Technical tutorials: 68%
research?
Connecting with other researchers exploring GenAI to learn from
each other: 60%
Brainstorming sessions to develop project / grant ideas: 51%
Finding collaborators on grants and projects: 42%
Finding students: 42%
1.2. Concerns about Generative AI
The enthusiasm and the unpreparedness are naturally accompanied by researchers’ concerns about using
generative AI. Some concerns are common among generative AI users in many lines of work, and include
issues such as data privacy and confidentiality, the biases that the models inherit from the training data, the AI
confabulation or hallucination, the opacity of data and training algorithms to the users of generative AI models,
thus the inability to assess whether a model is appropriate for a certain type of use (Birhane et al., 2023;
Liebrenz et al., 2023; Ray, 2023; Zhuo et al., 2023). In addition, there are also concerns specific to using
generative AI for scientific research. The rigor and reproducibility of research with generative AI in the
workflow has already become a major consideration. Any research with AI models that are not developed
locally, and without transparency of data and algorithms, poses fundamental challenges throughout the research
workflow, from study design and data query all the way to results validation (Li et al., 2023; Sohn, 2023;
Spirling, 2023). Many researchers are already aware of such issues to various degrees. For example, stories
4
Harvard Data Science Review • Special Issue 5: Grappling With the Institutional Eorts to Help Academic Researchers Implement
Generative AI Revolution Generative AI in Research
about generative AI hallucinating citations are shared widely. But there is very little discussion yet of how well
generative AI systems do in coming up with research hypotheses that are creative, testable, and of practical
value. So it remains to be seen how the benefits of generative AI in research weigh against the negative
consequences.
None of these concerns are new to academic researchers. There are always hopes and fears when a new
technology emerges with the promise of transforming research and people wonder how best to adopt it for
research innovation while upholding research integrity. These concerns, however, are amplified in the case of
generative AI because of how quickly new AI systems are developed, while our understanding of the functions
and limitations of these systems is still very limited (Bengio et al., 2023; Bommasani et al., 2023). These issues
are further exacerbated when researchers at all skill levels rush to adopt generative AI methods in their
research and there is not a standard or process for model selection or for quality control of the model use. What
we will almost surely witness, then, will be a flood of research outcomes and publications of uncertain quality
using generative AI, which will likely distract scientists from doing good research in the short term and may
even have long-term impacts. Academic researchers are quite aware of these challenges. In fact, at a generative
AI faculty workshop in the summer of 2023 (see more description in Section 2.2), the concerns of the attendees
were reflected in the following specific topics:
A. Understanding model output, upholding research rigor and reproducibility.
How to think about research rigor and reproducibility when there is lack of transparency of generative AI
models, and when the model output depends on the specific prompts.
How to assess the novelty of the model’s output.
How to identify and correct bias, misinformation, or erroneous training data and in model outputs.
How to think about data provenance and governance with generative AI models.
How to quantify uncertainty of model outputs.
B. Understanding issues of ethics, authorship, copyright, and privacy.
How to cite, acknowledge, and report generative AI in research work.
How to assess issues related to copyrighted training data, and model outputs based on copyrighted training
data.
How to assess data privacy and confidentiality issues when researchers have little knowledge about the
training data.
How to assess the balance between privacy / confidentiality and the need for data and model transparency.
Patent issues if a research idea is first suggested by generative AI.
C. Technical and infrastructure considerations with the use of generative AI in research.
Choosing a model and comparing models for a particular research question.
5
Harvard Data Science Review • Special Issue 5: Grappling With the Institutional Eorts to Help Academic Researchers Implement
Generative AI Revolution Generative AI in Research
Fine-tuning models locally and the local resources needed for this.
Keeping up-to-date knowledge of generative AI models.
1.3. We Need a New Model for Institutional Transformation
It is obvious to us that it is not feasible, or at least highly inefficient, if individual researchers are expected to
address such issues themselves not only because most lack the expertise, effort, and resources, but also because
they would each be reinventing the wheel. The typical researcher learns to use a new research method on their
own or through their collaborators, and gets pointers to resources from someone they happen to interact with.
This, somewhat random, social diffusion will not be sufficient when they need to acquire skills with a new
technology overnight and put it to immediate research use, and also goes against the nationwide drive to ensure
equitable access to AI technologies (National Artificial Intelligence Research Resource Task Force, 2023). It is
also virtually impossible for researchers to individually assess model quality, validity, and reliability, leading to
at least some guesswork in adoption and implementation choices.
We believe a new model of enabling the adoption of rapidly emerging technologies is sorely needed at this
point, and we believe academic institutions and their research centers should play a critical role. Universities
are already responsible for providing the research infrastructure, such as computing centers and research cores
for scientific instrumentation, and supporting resource-intensive, large-scale, and high-throughput research.
They should also be responsible for enabling the adoption of new technologies in research. Indeed, many
universities are already keenly aware of the importance of generative AI and are already developing capacity,
such as computing resources. The University of Michigan, for example, has just launched UMGPT, which
provides a relatively secure environment for campus use, including research use. Some institutions are also
training domain-specific generative AI models for academic research such as OLMo (Open Language Model)
and the GatorTron (Yang et al., 2022).
However, these are not enough. We believe that the emergence of generative AI is a call for universities, as the
home of new knowledge and the home of academic researchers, to play a much more active role in enabling
academic researchers to develop new skills and adopt new research methods in ethical, responsible, and
effective ways. This will likely have long-lasting benefits to research and discovery. Universities, however, are
not set up to be nimble in ways that some businesses can be in response to new technology developments and
‘market trends.’ So what can be done?
We advocate for university-level research institutes to fill this need and help complete a solution–
implementation–outcome process that will help academic researchers adopt new technologies or research
standards (solutions) to achieve better research innovation and outcomes (Figure 1). While it is difficult to
imagine an entire university being nimble in the face of an emerging technology, an organization within a
university can be so. Universities often set up a research institute to advance a research area of importance.
6
Harvard Data Science Review • Special Issue 5: Grappling With the Institutional Eorts to Help Academic Researchers Implement
Generative AI Revolution Generative AI in Research
Indeed, there are many examples of institutes that have spearheaded research in a ‘hot’ area and risen to well-
deserved prominence for their work in advancing the frontier, particularly in interdisciplinary areas.
But we believe there can be a very different role for a research institute at a university, which can have an even
greater impact on science: to serve as a knowledge base and facilitator for the adoption of new methods that
have the potential to transform research across a range of disciplines. Such methods frequently arise in fields
such as data science and AI. generative AI is perhaps the best example because of its applicability in almost
every line of work and its fast pace of advancement. But it surely is only one of the very first technologies that
could bring sweeping changes. Hence, what we advocate for, supporting the adoption of generative AI in
research, will be equally relevant for future waves of new technologies. In other words, academic research
institutes can play a significant role in institutional transformation by developing and disseminating tools,
training researchers, and establishing best practices, all of which are essential for researchers to swiftly adopt
new technologies to stay at the forefront of research and innovation.
Figure 1. Universities should play a significant role to help researchers adopt new
technologies and guidelines for research innovation.
In the next section, we describe some of the work that we have already started to develop in this new role for
our institute. The work is still very preliminary, given that we and the researchers that we support are still at the
initial stage of understanding myriad considerations associated with generative AI. But it provides a starting
point for further discussion on the institutional effort needed for adopting new technologies in academic
research.
2. Supporting Academic Researchers Through Research
Incubation, Training, and Best Practices
The Michigan Institute for Data Science (MIDAS) at the University of Michigan (U-M) has been investing
effort for institutional transformation over the past few years, with an initial focus on technical skill
development and rigor and reproducibility in data-intensive research. As U-M’s focal point of data science and
AI research, the central goal of MIDAS is to enable the transformative use of data science and AI methods for
both scientific and societal impact, across an enormous array of disciplines with wildly different
epistemological approaches and data use practices.
7
Harvard Data Science Review • Special Issue 5: Grappling With the Institutional Eorts to Help Academic Researchers Implement
Generative AI Revolution Generative AI in Research
Among its many threads of work in enabling research, providing training, and building research collaboration,
one component is to teach new research methodology to faculty and staff researchers through a set of summer
academies that introduce data science and AI research skills from the beginning level to advanced topics.
These summer academies started as an experiment because we were uncertain of the needs; but in the past
three years, the offering expanded from one week-long bootcamp per year to multiple week-long sessions, and
has trained nearly 300 faculty and staff researchers. Our experience demonstrates that faculty and staff
researchers need such opportunities to systematically learn new research methodologies.
MIDAS’s effort to improve rigor and reproducibility focuses on filling another important gap (Liu et al., 2022).
Many journals, funding agencies, and professional societies have developed clear guidelines, requirements, and
incentives for research rigor and reproducibility. Many researchers have a reasonable understanding of the
issue and know what outcomes are expected from them. But the reproducibility problem remains serious,
especially for data-intensive research that has a long and complex workflow (Hardwicke et al., 2021;
Laurinavichyute et al., 2022; Stodden et al., 2018). Through collaboration with the university’s research
community, MIDAS has coordinated grassroots efforts and developed online resources and training to enable
rigor and reproducibility in data-intensive research. The MIDAS reproducibility online resource hub has had
more than 10,000 visits. MIDAS is now developing a nationwide training program for faculty and staff
scientists, funded by the National Institutes of Health, on improving the rigor and reproducibility of data-
intensive research.
More importantly, through this work we have come to realize that a major gap in the researchers’ efforts to
improve reproducibility is that they often lack the means or the expertise to translate guidelines into outcomes.
In other words, researchers need to be handed validated methods/tools and know how to use them in order to
complete the solution–implementation–outcome process (Figure 1). In this case, the solution is the
reproducible research guidelines; the outcome is more reproducible research; and the implementation is the
phase where researchers are equipped with appropriate tools and processes.
Such previous work has developed the mindset at MIDAS that allowed the team to plunge into action when
generative AI ‘stormed’ the world stage. Since early 2023, MIDAS has started developing best practice
guidelines, coordinating the exploration of generative AI for research, and providing training for researchers.
2.1. Developing Guidelines
Just like the researchers themselves, almost all research organizations are scrambling to cope with generative
AI and its regulation, which changes quickly. Guidelines in addition to researchers’ discretion are essential
because generative AI’s use in research is fraught with issues every step of the way, from whether the training
data is appropriate for a particular type of research to the validation of output. Its use to improve productivity
can also be tangled with additional issues such as confidentiality and copyright. The National Institutes of
Health and the National Science Foundation, for example, have already formally forbidden the use of
8
Harvard Data Science Review • Special Issue 5: Grappling With the Institutional Eorts to Help Academic Researchers Implement
Generative AI Revolution Generative AI in Research
generative AI in grant proposal review (National Institutes of Health, 2023; National Science Foundation,
2023). Many journals, such as Nature and Science, also prohibit certain types of usage of text and images
created by generative AI (Flanagin et al., 2023; Harker, 2023). Understanding what they are or are not allowed
to do is an additional challenge for researchers. We expect many such guidelines and that they will evolve
quickly with time.
To provide a starting point for researchers, MIDAS compiled a set of guidelines that include the following
topics:
Writing with generative AI
Can I use generative AI to write research papers?
Can I use generative AI to write grants?
Can I use generative AI to help me when I write a literature review section for my paper?
Can I use generative AI to write nontechnical summaries, create presentations, and translate my work?
Using generative AI to improve productivity
Can I use generative AI to review grant proposals or papers?
Can I use generative AI to write letters of support?
How can I use generative AI as a brainstorming partner in my research?
Using generative AI for data generation and analysis
Can I use generative AI to write code?
Can I use generative AI for data analysis and visualization?
Can I use generative AI as a substitute for human participants in surveys?
Can I use generative AI to label data?
Can I use generative AI to review data for errors and biases?
Reporting the use of generative AI
How do I cite contents created or assisted by generative AI?
How do I report the use of generative AI models in a paper?
Considerations for choosing generative AI models
How do I decide which generative AI to use in research?
Open source
Accuracy and precision
Cost
What uniquely generative AI issues should I consider when I adopt generative AI in my research?
Ethical issues
Bias in data
AI hallucination
Plagiarism
Prompt engineering
9
Harvard Data Science Review • Special Issue 5: Grappling With the Institutional Eorts to Help Academic Researchers Implement
Generative AI Revolution Generative AI in Research
Knowledge cutoff data
Model continuity
Security
We selected these topics based on our discussions with researchers in our community. We are updating the
guide several times a month as new guidelines are published from federal agencies, funding agencies,
professional societies, and journals.
2.2. Demonstrating the Use of Generative AI in Research and Exploring Possibilities
Many academic researchers may have only tried using ChatGPT to draft an email or to edit some texts, but
most of them are aware of the possibility of using generative AI to do much more and to accelerate research
and enable new research ideas in many other ways. However, how this can be done is still elusive. For
example, many have heard that generative AI can be used to summarize research literature. However,
successful implementations are still very few, and researchers are concerned with many issues associated with
such use, such as the indiscriminate inclusion of published work with poor quality or that is irreproducible, and
bias against work in non-English languages. Many researchers are also aware that generative AI can help with
data analysis, but what skills researchers need to have in order to ensure that the analysis is correct is also
unclear to many. Domain-specific generative AI models have been used for protein structure research, drug
design, material science, and many other fields of inquiry, yet many researchers are unclear what special skills
and data are needed to train and deploy such models. Exposing researchers to successful examples, therefore,
has been one of our top priorities.
MIDAS organized a faculty workshop in the summer of 2023 with 92 U-M faculty attendees. Twelve speakers
demonstrated how they incorporated generative AI in research (health care research, chemistry, social science,
arts and design), and discussed ethical and technical considerations as well as infrastructure challenges. The
attendees participated in a few rounds of breakout discussions focusing on how generative AI can be used in
research to improve productivity, significant research questions that can be boosted with generative AI, and
ethical and technical challenges. The attendees came from 45 academic units at the university, with a diverse
range of research areas (Table 2). Such diverse participation is a strong indicator of the widespread interest in
generative AI.
Table 2. Research fields represented by the attendees of the faculty workshop on how to use
generative AI in research.
Arts and Design Biological Science Engineering
10
Harvard Data Science Review • Special Issue 5: Grappling With the Institutional Eorts to Help Academic Researchers Implement
Generative AI Revolution Generative AI in Research
Architecture and Urban Planning Biostatistics Aerospace Engineering
Arts and Design Physiology Chemical Engineering
Civil and Environmental Engineering
Electrical Engineering and Computer
Science
Industrial and Operations Engineering
Mechanical Engineering
Nuclear Engineering and Radiological
Sciences
Robotics
Environmental and Earth Sciences Medical Science Math and Physical Science
Environment and Sustainability Anesthesiology Chemistry
Climate and Space Science and Cardiac Surgery Mathematics
Engineering
Computational Medicine and Physics
Bioinformatics
Statistics
Internal Medicine
Kinesiology
Learning Health Sciences
Ophthalmology
Pediatrics
Pharmacy
Psychiatry
Radiation Oncology
Social Science and Business
11
Harvard Data Science Review • Special Issue 5: Grappling With the Institutional Eorts to Help Academic Researchers Implement
Generative AI Revolution Generative AI in Research
Business
Communications and Media
Information
Political Science
Public Policy
Also in summer, 2023, MIDAS organized a webinar series, Generative AI Coast-to-Coast (C2C), together with
Johns Hopkins University, Rice University, the International Computer Science Institute, The Ohio State
University, and University of Washington. The webinars featured eight speakers from the six institutions on
“Generative AI in Healthcare,” “Generative AI in the Lab,” “Policy, Ethics and Generative AI,” and “An
Under the Hood Look at Generative AI: Potentials and Pitfalls.” The goal of the series was also to demonstrate
the successful implementation of generative AI in research, build collaboration, and point out the cautions to
take.
2.3. Developing Technical Skills
Based on our faculty survey, receiving training is their top priority regarding generative AI. Thus, MIDAS has
offered a series of hands-on tutorials as the starting point for academic researchers. The focus was not domain-
specific generative AI models trained on technical data, such as protein structures; instead, the focus was on
using generative AI, including large language models (LLM), in domain-agnostic ways. The topics included:
Writing, planning, and literature review: enhancing professional productivity with generative AI
Code smarter, not harder: harnessing generative AI for research programming efficiency
Integrating generative AI into your research workflow: using image generation as the example
Making generative AI better for you: fine-tuning and experimentation for custom research solutions
However, feedback from workshop attendees and researchers in our community indicated that we should have
started from an even more basic place and be as hands-on as possible. As shown in Table 1, the vast majority
of researchers would like to use generative AI to improve productivity, but only half are getting reasonable
results even when using ChatGPT with prompts. Therefore, we are planning a few new tutorial sessions,
making them more hands-on, and focusing on more basic tasks. The topics will include:
Improving general productivity with ChatGPT: non-research writing (emails, posters and presentations,
checking for compliance, letters of recommendation, translation)
Finding, synthesizing, and summarizing literature with LLM
Generating simulated data with LLM
Data analysis and visualization with ChatGPT (text, image, and numeric data)
Drafting research articles with ChatGPT (drafting, writing, and formatting bibliographies)
12
Harvard Data Science Review • Special Issue 5: Grappling With the Institutional Eorts to Help Academic Researchers Implement
Generative AI Revolution Generative AI in Research
2.4. What Is Next
During this set of work in the past few months, and through discussions with our research community, it is
increasingly obvious to us that the support MIDAS is providing to academic researchers for generative AI is, in
essence, filling the same ‘implementation’ gap that we identified in the solution–implementation–outcome
model for research reproducibility (Table 3). In both cases, researchers know that there is a solution for them to
do better research: incorporating generative AI in research, or following guidelines for research reproducibility.
They also know in both cases what the ideal outcome should be: accelerated research and innovation, and more
rigorous and reproducible research. However, in both cases the gap is that researchers are left to their own
devices to implement the solution to achieve the outcome. In both cases, the effort from MIDAS can have a
significant impact to fill the implementation gap.
Table 3. The role of academic research institutes to fill the implementation gap to achieve
institutional transformation.
Adopting Generative AI for Research Improving Research Reproducibility
Solution Generative AI as a powerful tool to Guidelines for improving research
accelerate the research process, and develop reproducibility.
previously infeasible research.
Implementation gap Most researchers do not have skills and Most researchers do not have methods/tools
resources to swiftly and responsibly adopt at their disposal, and do not have the skills
generative AI in research. and resources to develop their own tools to
improve research reproducibility.
The role of academic research institutes Identify researchers’ needs, develop Identify researchers’ needs, develop
guidelines, standard processes and tools to standard processes and tools to improve
adopt generative AI in research, and enable research reproducibility and enable the wide
the wide adoption of such processes and adoption of such processes an |
310 | mit_edu | pdf.pdf | Forrester’s Top 10 Emerging Technologies For 2024: GenAI, TuringBots, And IoT Security Poised To
Deliver The Fastest ROI
June 25, 2024
The rapid acceleration of AI innovation has sparked a surge in advancement in other emerging technologies
CAMBRIDGE, Mass.--(BUSINESS WIRE)--Jun. 25, 2024-- According to Forrester’s (Nasdaq: FORR) The Top 10 Emerging Technologies In 2024
report, generative AI (genAI) for visual content, genAI for language, TuringBots, and IoT security are the top emerging technologies that will deliver the
most immediate ROI for businesses in 2024 and beyond.
With new technologies emerging seemingly every day, business and technology leaders need to time those investments based on value, risk, and
potential payout timelines. Forrester organizes its top emerging technologies by benefit horizon to help with these decisions.
Emerging technologies that will offer significant benefits within the next two years:
GenAI for visual content. Advanced machine learning models that generate images or video from text, audio, or video
prompts, this technology will help firms generate visual content for marketing, experiences, and products.
GenAI for language. GenAI for language is already delivering value in customer support and content creation but
continues to advance at a blinding pace. It is accelerating many other technologies as it goes.
TuringBots. Accelerated by advancements in genAI for language, these AI-powered software robots help developers build
applications that deliver more than just code generation.
IoT security. The proliferation of devices has led to an exponential explosion in security attacks, raising the importance of
security for IoT devices. Vendors are competing and colliding in a rush to offer capabilities.
Midterm emerging technologies that will deliver benefits in the next two to five years:
AI agents. The role of autonomous workplace assistants or AI agents has expanded beyond the back office and employee
assistance to customer-facing automation. These AI agents will grow increasingly sophisticated to better understand and
respond to nuance and context.
Autonomous mobility. This technology will accelerate commercial and urban transportation ecosystem collaborations to
orchestrate personalized mobility experiences for both customers and businesses.
Edge intelligence. Advanced edge intelligence capabilities such as edge machine learning are still not yet common, even
though many foundational elements like Apple foundation models are becoming available.
Quantum security. This technology will overhaul security systems for on-premises and cloud compute, storage and
network infrastructure, commercial off-the-shelf software, commercial software-as-a-service offerings, and in-house built
software.
Emerging technologies that will take at least five more years to deliver tangible value for most firms and use cases:
Extended reality (XR). Only 8% of US online adults own a virtual-reality headset, and just 16% have used an augmented-
reality device or app. While XR is advancing in training and onboarding, companies are resisting investing in tools like
these until they see broad adoption.
Zero Trust edge (ZTE). ZTE technology has the potential to protect remote workers, retail outlets, and branch offices with
embedded local security, but only a handful of true ZTE solutions exist today, and legacy devices add additional
management complexity.
“Tech leaders must be able to identify the right use cases and quantify potential benefits, costs, and risks across multiple horizons,” says Brian
Hopkins, Forrester VP, emerging tech portfolio. “They need to spread investments out, with shorter-term technologies delivering quick returns and
longer-term bets requiring more effort, more foundational investment, and the capacity to manage more risk.”
Resources:
Learn more about the top emerging technologies that will deliver ROI for enterprises.
Read The Top 10 Emerging Technologies In 2024 report to gain insight into maturity trends, business value, important use
cases, and risks associated with each technology (client access required).
Register for a complimentary webinar that takes a deep dive into the 2024 list of emerging technologies and their use
cases and benefits.
Download Forrester’s complimentary guide for tech leaders to identify use cases and gain support for emerging
technologies.
Tech and data leaders can attend Forrester’s 2024 Technology & Innovation Summits in North America, EMEA, and Asia
Pacific to learn how to align their IT strategy to their business goals to accelerate growth.
About Forrester
Forrester (Nasdaq: FORR) is one of the most influential research and advisory firms in the world. We help leaders across technology, customer
experience, digital, marketing, sales, and product functions use customer obsession to accelerate growth. Through Forrester’s proprietary research,
consulting, and events, leaders from around the globe are empowered to be bold at work — to navigate change and put their customers at the center
of their leadership, strategy, and operations. Our unique insights are grounded in annual surveys of more than 700,000 consumers, business leaders,
and technology leaders worldwide; rigorous and objective research methodologies, including Forrester Wave™ evaluations; more than 100 million
real-time feedback votes; and the shared wisdom of our clients. To learn more, visit Forrester.com.
View source version on businesswire.com: https://www.businesswire.com/news/home/20240625049417/en/
Press:
Hannah Segvich
[email protected]
Source: Forrester |
311 | mit_edu | Information_20Systems.pdf | Rising Scholars Conference
Information Systems Student Research Presentations
Jonathan Gomez Martinez Emory University
[email protected] Goizueta Business School
Jonathan Gomez Martinez is a PhD candidate in Information Systems and Operations
Management at the Goizueta Business School. Informed by his background as a Mexican
immigrant and first-generation college student, Jonathan’s research highlights the unintended
consequences of technology and technology policy. His ongoing projects evaluate the role of AI,
privacy policy, and digital platforms on censoring minority voices and complicating the
operations of small and midsized businesses. To learn more about Jonathan, visit
www.jgomezm.com.
Abstract:
Platform Policy Changes: Impact of Auto-Moderation on Minority Community Rights
User-generated content on social media platforms has always been moderated as
advertisers on these platforms require interactions to be safe, non-abusive, and generally in
compliance with regulations such as those dealing with intellectual property rights. Even if
assisted by algorithms to filter content, human reviewers had always made the final call, until
recently where unprecedented volume and other factors have forced platforms to rely fully on
automated, Artificial Intelligence based (AI-based) content moderation. Cognizant of unintended
consequences of technology usage, our research exploits a natural experiment wherein Twitter
had resorted to auto-moderation in 2020. Our investigation reveals the dramatic impact of such
technologies, often context-blind, on the interactions of a minority group of users such as the
LGBTQ+ community. Through a rigorous empirical approach, our findings show that interactions
within this community reflect a heavily censored language after auto-moderation deployment by
Twitter. In the absence of any explicitly LGBTQ+ related policy changes on Twitter, our work
underscores the inadvertent harm that ensues when context-less AI technologies are adopted.
Farnam Mohebi Univeristy of California, Berkeley
[email protected] Haas School of Business
I am currently a management PhD student at the Haas School of Business and a data science
fellow at the Dlab, UC Berkeley, having previously completed my MD-MPH. I focus on the
intersection of healthcare and management, driven by a deep interest in understanding the
multi-faceted role of physicians in the AI world. I am interested in physicians' perception and
experience with clinical AI and physician-scientists' narratives of it. Additionally, I study the
impact of management practices on physicians. My work is guided by my background in
healthcare and a commitment to improving organizational practices within the field.
Abstract:
Assessing the Multifaceted Role of Physicians in the AI Landscape.
1 Research Question
The central research question of this study is multi-faceted, exploring how physicians navigate
the rapidly evolving landscape of Artificial Intelligence (AI) in healthcare. Specifically, the
question aims to unpack the complexities of physicians’ diverse roles as developers, adopters,
evaluators, and managers of AI technologies in medical settings.
As Developers: Why are Physicians Becoming Developers? How do physicians influence the
trajectory of medical AI science development? Beyond their influence on the trajectory of AI,
how does the professional standing of physicians contribute to their credibility as developers?
Do non-financial incentives, such as academic recognition or potential for societal impact, also
play a role? What issues of legitimacy arise when physicians act as developers? Are there
elements of elitism and prestige that attract physicians to the field of AI development? How do
these factors interact with other motivations and constraints?
As Adopters: How do elements like professionalism and hierarchy affect physician adoption
rates at various levels?How do age, gender, and other demographic characteristics of
physicians influence their willingness to adopt AI technolo-gies? Does a younger generation of
physicians, for example, show less openness to incorporating AI into their practices compared
to their older counterparts? How do the level of training and the years of clinical experience
impact a physician’s propensity to adopt AI? Do physicians with more advanced training or
specialization show different patterns of adoption?What role does exposure to AI in medical
education play in facilitating or hindering adoption? How is the legitimacy of the technology
assessed before adoption? Are there non-financial incentives that significantly impact adoption,
such as the prospect of improved patient outcomes, peer recognition, or professional
development opportunities?
As Evaluators: In what ways do physicians’ professionalism ensure a more rigorous and
ethically sound evaluation of AI tools? Does the commitment of physicians to professional ethics
and existing medical practices make them more resistant to adopting innovative AI technologies
that challenge traditional healthcare paradigms? Are senior physicians or those higher up in the
medical hierarchy more likely to maintain the status quo, thereby hindering the adoption of
transformative AI technologies?
As Managers: How does a physician’s role as a manager facilitate the integration of AI
technology into clinical settings, particularly in terms of operational efficiency and patient care?
In what ways does the managerial role of physicians contribute to fostering an organizational
culture that is more receptive to AI innovations? How does a physician-manager’s clinical
background influence the prioritization of AI projects that have the most direct impact on patient
care?Does a physician’s managerial role lead to conflicts of interest when deciding on AI
projects, perhaps prioritizing those that align with their own clinical specializations or authority
over others that may benefit the healthcare system more broadly? How might the dual
responsibilities of physician managers contribute to potential burnout, thereby affecting their
capacity to evaluate and implement AI technologies effectively?
2 Methods
My research adopts a full-cycle approach. In the initial qualitative stage, I will employ
ethnography and content analysis to delve into various social media, press releases, scholarly
publications, and other publicly available data that provide insights into physicians’ perspectives
on AI. The focus will be on their roles as developers, adopters, evaluators, ethicists, and
managers.
The qualitative insights will then inform the design of lab or field experiments, administrative
data analysis, and surveys. These will focus on physicians’ decision-making patterns, adoption
rates, ethical considerations, and managerial choices in the context of AI integration into
healthcare.
3 Implications
Understanding the economic and social dynamics that influence AI adoption is crucial for
policymakers and industry stakeholders. This research could contribute valuable insights into
how to navigate the conflicting interests between organizational efficiencies and end-user
acceptance. It also opens up discussions on the economic implications of technology adoption
in critical sectors like healthcare.
Carolina Reis Virginia Tech
[email protected] Pamplin College of Business
Carolina Reis is a fourth-year Information Systems PhD student at Virginia Tech. Broadly, her
research focuses on the hybrid human-machine behavior. In particular, her research
investigates: (1) how the introduction of AI systems into social and organizational ecosystems
alters human beliefs and behaviors, and (2) how people themselves also shape AI systems
through the training of these systems using active human input.
Abstract:
Outsourcing Morality: The Hidden Path to Machine Ethics
Authors: Carolina Reis, Virginia Tech; Nicholas Brown, Indiana University
Artificial intelligence (AI) technologies rely on large language models (LLMs) trained on easily
accessible information online, such as content from the “darkest recesses of the internet”
(Perrigo, 2023). To remove toxicity (e.g., sexism, racism, xenophobia, hate speech, calls for
violence) from the training sets, companies hire human agents to perform content moderation—
the process of reviewing and monitoring digital toxicity. Currently, a large share of this content
moderation process is outsourced to companies and workers in developing nations, where the
work is often unregulated, leaving content moderators to label and annotate toxic content on
their own. These subjective labels then serve as the ground truth for LLM technologies (Perrigo,
2023). However, the appropriateness of language use varies among cultures, contexts, and
people, and what is morally acceptable depends on where the person lives. A paradox thus
ensues: AI technologies are used worldwide, especially in technologically advanced countries,
but their ingrained morality is determined in foreign countries that do not necessarily hold similar
moral values.
In this research, we intend to investigate the differences in cultural perspectives that influence
content moderation and whether these differences perpetuate harmful AI biases. To begin, we
conducted a pilot study, using the leaked Facebook documents on hate speech content
moderation (“Hate Speech and Anti-Migrant Posts: Facebook’s Rules,” 2017), to assess
whether individuals from different countries vary in their moral predisposition. Preliminary results
confirm this hypothesis, and show that (1) a model powered by content moderated by
individuals from the United States (n = 391) would have a significantly higher accuracy rate,
precision rate, recall rate, and F1 score than a model powered by content moderated by
individuals from India (n = 286), and (2) a model powered by content moderated by individuals
from India would have a significantly higher accuracy rate, precision rate, recall rate, and F1
score than a model powered by content moderated by individuals from Brazil (n = 32), when the
Facebook guidelines are used as ground truth. These results indicate cross-cultural ethical
variation and raise potential concerns with current machine ethics practices.
In our forthcoming studies, we will employ a mixed-method design. In Study 1, we will interview
content moderators located in different countries to grasp the practices adopted in the content
moderation process. In Study 2, we will launch an online experimental platform (similar to Awad
et al., 2018) where we will explore different content moderation scenarios and collect data from
people in multiple countries to assess cultural differences in ethical tendencies. In Study 3, we
will develop algorithmic models powered by the labeled data from different cultures and show
the differences in algorithmic output and performance. Ultimately, our aim is to help advance
potential solutions for the problem of universal machine ethics.
Kai-Cheng Yang Northeastern University
[email protected] Network Science Institute
Kai-Cheng Yang is a postdoctoral researcher in the Lazer Lab at Northeastern University's
Network Science Institute. He obtained his Ph.D. in Informatics from the Luddy School of
Informatics, Computing, and Engineering at Indiana University Bloomington. He is interested in
computational social science. His research aims to uncover how technologies like generative AI
are used for deceptive and disruptive purposes, study how humans react to these abuses, and
develop countermeasures. Specifically, he focuses on bad actors like malicious social bots and
misinformation on social media. He built popular tools, such as Botometer, that have served
tens of thousands of users. He also acted as the social bot expert in the trial of Twitter vs. Elon
Musk. His work has been covered by CNN, BBC, The New York Times, and many other popular
news outlets.
Abstract:
Large language models and cyber social threats: Good, bad, and ugly
Large language models (LLMs) may profoundly impact our information ecosystem. On the one
hand, they exhibit impressive capabilities in generating realistic text across diverse subjects and
show great potential in many applications. On the other hand, concerns have been raised that
they could be utilized to produce fake content with deceptive intentions. In this talk, I will present
three studies to demonstrate how LLMs can be abused by bad actors and leveraged by users
for self-protection. In the first study, I will introduce a Twitter botnet that appears to employ
ChatGPT to generate human-like content. These accounts form a dense cluster of fake
personas that exhibit similar behaviors, including posting machine-generated content and stolen
images, and engage with each other through replies and retweets. ChatGPT-generated content
promotes suspicious websites and spreads harmful comments. While the accounts in the botnet
can be detected through their coordination patterns, current state-of-the-art LLM content
classifiers fail to discriminate between them and human accounts in the wild. In the other two
studies, I will talk about using LLMs to counter the spread of misinformation. Through extensive
experiments, I find that ChatGPT, a prominent LLM, can evaluate the credibility of news outlets.
This suggests that LLMs could be an affordable reference for credibility ratings in fact-checking
applications. Then, I further test the feasibility of using ChatGPT as a fact-checking tool in a
human-subject experiment. Although ChatGPT performs reasonably well in debunking false
headlines, it does not significantly affect participants' ability to discern headline accuracy or
share accurate news. In certain cases, ChatGPT might even be harmful. The findings
underscore the importance of accounting for human factors when incorporating AI models into
our information ecosystem. |
312 | mit_edu | Paper_Artificial-Intelligence-and-Jobs-Evidence-from-Online-Vacancies.pdf | fi
Arti cial Intelligence and Jobs:
Evidence from Online Vacancies
Daron Acemoglu, MassachusettsInstituteofTechnology(MIT)
andNationalBureauofEconomicResearch(NBER)
David Autor, MITandNBER
Jonathon Hazell, LondonSchoolofEconomics
Pascual Restrepo, BostonUniversityandNBER
Westudytheimpactofartificialintelligence(AI)onlabormarkets
usingestablishment-leveldataonthenearuniverseofonlinevacan-
ciesintheUnitedStatesfrom2010onward.Thereisrapidgrowth
in AI-related vacancies over 2010–18 that is driven by establish-
mentswhoseworkersengageintaskscompatiblewithAI’scurrent
capabilities.AstheseAI-exposedestablishmentsadoptAI,theysi-
multaneously reduce hiring in non-AI positions and change the
skillrequirementsofremainingpostings.Whilevisibleattheestab-
lishment level, the aggregate impacts of AI-labor substitution on
employmentandwagegrowthinmoreexposedoccupationsandin-
dustriesiscurrentlytoosmalltobedetectable.
WethankBlediTaskafordetailedcommentsandprovidingaccesstoBurningGlass
data;JoshuaAngrist,AndreasMueller,RobSeamans,andBetseyStevensonforvery
usefulcommentsandsuggestions;JoseVelardeandZheFredricKongforexpertre-
searchassistance;andDavidDemingandKadeemNorayforsharingtheircodeand
data.AcemogluandAutoracknowledgesupportfromAccentureLLP,IBMGlobal
Universities, Schmidt Futures, and the Smith Richardson Foundation. Acemoglu
SubmittedDecember15,2020;AcceptedNovember24,2021.
JournalofLaborEconomics,volume40,numberS1,April2022.
©2022TheUniversityofChicago.Allrightsreserved.PublishedbyTheUniversityofChicagoPressin
associationwithTheSocietyofLaborEconomistsandTheNationalOpinionResearchCenter.https://
doi.org/10.1086/718327
S293
S294 Acemogluetal.
I. Introduction
Thepastdecadehaswitnessedrapidadvancesinartificialintelligence(AI)
based on new machine learning techniques and the availability of massive
datasets.1Thischangeisexpectedtoaccelerateintheyearstocome(e.g.,Ne-
apolitan and Jiang 2018; Russell 2019), and AI applications have already
startedtoimpactbusinesses(e.g.,Agarwal,Gans,andGoldfarb2018).Some
commentators see this as a harbinger of a jobless future (e.g., Ford 2015;
West2018;Susskind2020),whileothersconsidertheoncomingAIrevolu-
tionasenrichinghumanproductivityandworkexperience(e.g.,McKinsey
GlobalInstitute2017).Thepersistenceofthesecontrastingvisionsisunsur-
prisinggiventhelimitedevidencetodateonthelabormarketconsequences
ofAI.Datacollectioneffortshaveonlyrecentlycommencedtodetermine
theprevalenceofcommercialAIuse,andwelacksystematicevidenceeven
onwhethertherehasbeenamajorincreaseinAIadoption—asopposedto
justextensivemediacoverage.
ThispaperstudiesAIadoptionintheUnitedStatesanditsimplications.
OurstartingpointisthatAIadoptioncanbepartiallyidentifiedfromthe
footprintsitleavesatadoptingestablishmentsastheyhireworkersspecial-
izinginAI-relatedactivities,suchassupervisedandunsupervisedlearning,
naturallanguageprocessing,machinetranslation,orimagerecognition.To
putthisideaintopractice,webuildanestablishment-leveldatasetofAIac-
tivitybasedonthenearuniverseofUSonlinejobvacancypostingsandtheir
detailed skill requirements from Burning Glass Technologies (hereafter,
BurningGlassorBG)fortheyears2007and2010through2018.2
Westartwithatask-basedperspective,linkingtheadoptionofAIandits
possibleimplicationstothetaskstructureofanestablishment.Thisperspec-
tive emphasizes that current applications of AI are capable of performing
specifictasksandpredictsthatfirmsengagedinthosetaskswillbetheones
acknowledges support from Google, the National Science Foundation, the Sloan
Foundation, and the Toulouse Network on Information Technology, and Autor
thankstheCarnegieFellowsProgram,theHeinzFamilyFoundation,andtheWash-
ington Center for Equitable Growth. Contact the corresponding author, David
Autor,atdautor@mit.edu.Informationconcerningaccesstothedatausedinthispa-
perisavailableassupplementalmaterialonline.
1AIisacollectionofalgorithmsthatactintelligentlybyrecognizingandrespond-
ingtotheenvironmenttoachievespecifiedgoals.AIalgorithmsprocess,identify,
andactonpatternsinunstructureddata(e.g.,speechdata,text,orimages)toachieve
specifiedgoals.
2TheBGdatahavebeenusedinseveralrecentpapers.Alekseevaetal.(2021)and
Babina et al. (2020), discussed below, use BGdata to study AIuse and its conse-
quences. Papers using BG data to explore other questions include Hershbein and
Kahn(2018),Azaretal.(2020),Modestino,Shoag,andBallance(2020),Hazelland
Taska(2019),andDemingandNoray(2020).
ArtificialIntelligenceandJobs S295
thatadoptAItechnologies.3Toidentifythetaskscompatiblewithcurrent
AItechnologies,weusethreedifferentbutcomplementarymeasures:Felten,
Raj,andSeamans’s(2018,2019)AIoccupationalimpactmeasure;Brynjolfs-
son,MitchellandRock’s(2018,2019)suitabilityformachinelearning(SML)
index;andWebb’s(2020)AIexposurescore.Theseindicesallidentifysets
oftasksandoccupationsthataremostimpactedbyAItechnologies,buteach
iscomputedonthebasisofdifferentassumptionsaboutAIcapabilities.We
constructanestablishment’sAIexposurefromitsbaseline(2010–12)occu-
pationalstructureaccordingtoeachoneoftheseindicesandusethesebase-
linemeasuresasproxiesforAIexposurethroughoutouranalysis.4Sinceour
goalistostudytheimpactofAIonAI-usingfirmsratherthanAI-producing
firms,weexcludefirmsintheprofessionalandbusinessservicesandinfor-
mationtechnologysectors(NorthAmericanIndustryClassificationSystem
[NAICS]51and54),bothofwhichareprimarysuppliersofAIservices.
OurfirstresultisthatthereisarapidtakeoffinAIvacancypostingsstart-
ingin2010andsignificantlyacceleratingaround2015–16.Consistentwith
atask-basedviewofAI,thisactivityisdrivenbyestablishmentswithtask
structures that are compatiblewith current AI capabilities. For instance, a
1standarddeviationincreaseinourbaselinemeasureofAIexposurebased
onFeltenetal.—approximatelythedifferenceintheaverageAIexposurebe-
tweenfinanceandminingandoilextraction—isassociatedwith15%more
AIvacancyposting.ThestrongassociationbetweenAIexposureandsub-
sequentAIactivityisrobusttonumerouscontrolsandspecificationchecks
whenusingtheFeltenetal.andtheWebbmeasures,butthisislessapparent
withtheSMLindex.ThisleadsustoplacegreateremphasisontheFelten
etal.andWebbmeasureswhenexploringtheeffectsofAIexposureonthe
demandfordifferenttypesofskillsandnon-AIhiring.
OursecondresultestablishesastrongassociationbetweenAIexposure
and changes in the types of skills demanded by establishments. With the
Feltenetal.andWebbmeasures(and,toalesserextent,withSML),wefind
thatAIexposureisassociatedwithbothasignificantdeclineinsomeofthe
skillspreviouslysoughtinpostedvacanciesandtheemergenceofnewskills.
ThisevidencebolstersthecasethatAIisalteringthetaskstructureofjobs,
3SeeAcemogluandAutor(2011)andAcemogluandRestrepo(2018,2019).This
isnottheonlypossibleapproachtoAI.OnecouldalsothinkofAIascomplement-
ingsomebusinessmodels(ratherthanperformingspecifictaskswithinthosemod-
els)orasallowingfirmstogenerateandcommercializenewproducts(seeAgarwal,
Gans,andGoldfarb2018;Bresnahan2019).Weexplainbelowwhythetask-based
approachisparticularlywellsuitedtoourempiricalapproachandhowitreceives
supportfromourfindings.
4Figure4belowshowsthattherelationshipbetweenthemeanwageofanoccu-
pation and the three AI exposure measures is distinct, which is the basis of our
claimthateachoneoftheseindicescapturesadifferentaspectofAIexposure.
S296 Acemogluetal.
replacing some human-performed tasks while simultaneously generating
newtasksaccompaniedbynewskilldemands.
ThefindingthatestablishmentswithAI-suitabletaskshireworkersinto
AIpositionsandchangetheirdemandforcertaintypesofskillsdoesnot,of
course,telluswhetherAIisincreasingorreducing overallnon-AIhiring
inexposedestablishments.Inprinciple,AI-exposedestablishmentsmaysee
anincreasein(non-AI)hiringifeitherAIdirectlycomplementsworkersin
sometasks,increasingtheirproductivityandencouragingmorehiring,or
AIsubstitutesforworkersinsometasksbutincreasestotalfactorproduc-
tivitysufficientlytoraisedemandinnonautomatedtasksviaaproductivity
effect(AcemogluandRestrepo2019).Alternatively,AIadoptionmayre-
duce hiring if AI technologies are replacing many tasks previously per-
formed by workers and the additional hiring they spur in nonautomated
tasksdoesnotmakeupforthisdisplacement.
Our third main result shows that AI exposure is associated with lower
(non-AI and overall) hiring. These results are robust in all of our specifi-
cations using the Felten et al. measure and in most specifications with the
Webbmeasurebut,asanticipated,notwithSML.Thetimingoftheserela-
tionshipsisalsoplausible:substantialdeclinesinhiringtakeplaceinthetime
window during which AI activity surged—between 2014 and 2018. This
patternofresults,combinedwiththeconcentrationofAIactivityinmore
AI-exposedtasks,suggeststhattherecentAIsurgeisdriveninpartbythe
automationofsomeofthetasksformerlyperformedbylabor.Wefindno
evidenceforeithertheviewthattherearemajorhuman-AIcomplementar-
ities in these establishments ortheexpectation that AI will increase hiring
becauseofitslargeproductivityeffects—althoughwecannotruleoutthat
otherapplicationsofAIthatarenotcapturedherecouldhavesucheffects.
Incontrasttotheestablishment-levelpatterns,wedonotdetectanyre-
lationshipbetweenAIexposureandoverallemploymentorwagesatthein-
dustryoroccupationlevel.Therearenosignificantemploymentimpactson
industrieswithgreaterexposuretoAI,andtherearealsonoemploymentor
wages effects for occupations that are more exposed to AI. We conclude
thatdespitethenotablesurgeinAIadoption,theimpactofthisnewtech-
nologyisstilltoosmallrelativetothescaleoftheUSlabormarkettohave
hadfirst-orderimpactsonemploymentpatternsoutsideofAIhiringitself.
Nevertheless,ourmainfindings—thatAIadoptionisdrivenbyestablish-
ments that have a task structure that is suitable for AI use and that this
hasbeenassociatedwithsignificantdeclinesinestablishmenthiring—imply
thatanypositiveproductivityandcomplementarityeffectsfromAIareat
presentsmallcomparedwithitsdisplacementconsequences.
OurpaperbuildsonAlanKrueger’sseminalworkontheeffectsofnew
digitaltechnologiesonworkersandwages(Krueger1993;Autor,Katz,and
Krueger 1998). Subsequent literature has investigated the implications of
automationtechnologies,focusingonwages,employmentpolarization,and
ArtificialIntelligenceandJobs S297
wageinequality(e.g.,Autor,Levy,andMurnane2003;GoosandManning
2007;AutorandDorn2013;Goos,Manning,andSalomons2014;Michaels,
Natraj,andVanReenen2014;Gregory,Salomons,andZierahn,forthcom-
ing).Recentworkhasstudiedtheimpactofspecificautomationtechnolo-
gies, especially industrial robots, on employment and wages, focusing on
industry-levelvariation(GraetzandMichaels2018),locallabormarketeffects
(AcemogluandRestrepo2020),orfirm-levelvariation(DinlersozandWolf
2018;Bessenetal.2019;Bonfigliolietal.2019;Humlum2019;Acemoglu,
Lelarge,andRestrepo2020;Dixon,Hong,andWu2021;Koch,Manuylov,
andSmolka2021).
TherearefewerstudiesoftheeffectsofAIspecifically,althoughthisbody
of work is growing rapidly. Bessen et al. (2018) conduct a survey of AI
startupsandfindthatabout75%ofAIstartupsreportthattheirproducts
helpclientsmakebetterpredictions,managedatabetter,orprovidehigher
quality.Only50%ofstartupsreportthattheirproductshelpcustomersau-
tomateroutinetasksandreducelaborcosts.GrennanandMichaely(2019)
studyhowAIalgorithmshaveaffectedsecurityanalystsandfindevidence
of task substitution: analysts are more likely to leave the profession when
they cover stocks for which there are abundant data available. Differently
fromthesepapers’focusonAI-producingsectorsandspecificapplications
ofAI,suchasfinance,westudythistechnology’seffectsonAI-usingestab-
lishmentsandnon-AIworkersthroughouttheeconomy.
Mostcloselyrelatedtoourpaperareafewrecentworksalsoinvestigating
theeffectsofAIonfirm-leveloutcomes.Babinaetal.(2020)studythere-
lationshipbetweenAIadoptionandemploymentandsalesatboththefirm
andtheindustrylevel.Theydocumentthat,consistentwithAlekseevaetal.
(2021), AI investment is stronger among firms with higher cash reserves,
highermarkups,andhigherR&Dintensityand,moreover,thatthesefirms
growmorethannonadopters.AcontrastbetweenourapproachandBabina
etal.’sisthatwefocusonAIsuitabilitybasedonestablishments’occupa-
tional structures rather than observed AI adoption, and this may explain
why we arrive at different results for hiring. Also related is Deming and
Noray (2020), who use Burning Glass data to study the relationship be-
tweenwages,technicalskills,andskillsobsolescence.Althoughtheirfocus
isnotAI,theirworkdemonstratesthatBurningGlassdataaresuitablefor
detectingchangesinjobskillrequirements,anangleofinquirywepursue
below.
Asnotedabove,ourworkexploitsmeasuresofAIsuitabilitydeveloped
byFelten,Raj,andSeamans(2018,2019),Brynjolfsson,Mitchell,andRock
(2018,2019),andWebb(2020).OurresultsareconsistentwithFelten,Raj,
andSeamans(2019),whofindapositiverelationshipbetweenAIsuitability
andAIvacancyposting,butnorelationshipwithemploymentgrowth,at
theoccupationallevel.WeconfirmthatAIsuitabilityisnotatpresentasso-
ciatedwithgreaterhiringinmorehighlyexposedoccupationsorindustries,
S298 Acemogluetal.
butwefindrobusteffectsonskilldemandandanegativeimpactonestab-
lishmenthiring.
Therestofthepaperisorganizedasfollows.SectionIIpresentsamodel
motivatingourempiricalstrategyandinterpretation.SectionIIIdescribes
thedata,andsectionIVpresentsourempiricalstrategy.SectionVpresents
ourmain results on AI exposureand AI hiring, while section VI looksat
changes in the types of skills AI-exposed establishments are looking for.
SectionVIIexplorestheeffectsofAIonhiringattheestablishment,indus-
try, and occupation levels. Section VIII concludes. Appendix A contains
additional material on our model, and additional robustness checks and
empirical results are presented in appendix B (appendixes are available
online).
II. Theory
Inthissection,weprovideamodelthatmotivatesourempiricalapproach
andinterpretation.
A. Tasks,Algorithms,andProduction
Establishmente’soutput,y,isproducedbycombiningtheservices,y(x),
e e
oftasksx ∈ T ⊂ T withunitelasticity(i.e.,aCobb-Douglasaggregator):
e
ð
lny 5 lnA 1 aðxÞlnyðxÞdx, (1)
e e e
T
e
whereT isthesetoffeasibletasks,asubsetT ofwhichisusedinthepro-
e
ductionprocessofestablishmente,andaðxÞ ≥ 0designatestheimportance
orqualityoftaskxinthepÐroductionprocess,whichiscommonacrosses-
tablishments. We impose aðxÞdx 5 1 for all feasible T, which ensures
T e
e
thatallestablishmentshaveconstantreturnstoscale.
EstablishmentsdifferintheirproductivitytermA and,moreimportantly,
e
inthesetoftaskstheyperform(e.g.,becausetheyproducedifferentgoods
andservicesorusedistinctproductionprocesses).Wealsoassumethateach
establishmentfacesadownward-slopingdemandcurveforitsproductand
will set its price p to maximize profits (and its problem is separable from
e
theprofit-maximizationproblemofthefirm’sotherestablishmentsincase
of multiestablishment firms). In this profit-maximization problem, we as-
sume that each establishment is small in the labor market and takes other
pricesandaggregateoutputasgiven.
Tasksareproducedbyhumanlabor,ℓ(x),orbyservicesfromAI-powered
e
algorithms,a(x):
e
(cid:2) (cid:3)
y eðxÞ 5 ðg ‘ðxÞ‘ eðxÞÞðj21Þ=j 1ðg aðxÞa eðxÞÞðj21Þ=j j=ðj21Þ , (2)
wherejistheelasticity ofsubstitutionbetweenlaborandalgorithms and
g(x) and g (x) are assumed to be common across establishments. We
ℓ a
ArtificialIntelligenceandJobs S299
assumethatAIservicesareprovidedbycombiningAIcapital(machineryor
algorithms)purchasedfromtheoutside,k(x),andin-houseworkersoper-
e
ating,programming,ormaintainingthiscapital,‘AIðxÞ,withthefollowing
e
technology:
(cid:4) (cid:5)
aðxÞ 5 min kðxÞ,‘AIðxÞ , (3)
e e e
whichimpliesthatin-houseAIworkersneedtobecombinedwithcapitalin
fixedproportions.5Weassumethroughoutthatallestablishmentsareprice
takersforproductionworkers,AIworkers,andAIcapital,whoserespec-
tivepricesarew,wAI,andR.
WeviewrecentadvancesinAIasincreasingtheabilityofalgorithmsto
perform certain tasks—corresponding to an increase g (x) for some x. In
a
whatfollows,wedenotebyTAthesubsetoftasksthat,duetotheseadvances,
can now be profitably performed by algorithms/AI. These advances in AI
technology will have heterogeneous impacts on establishments depending
ontheirtaskstructure.Forexample,anincreaseing (x)fortextrecogni-
a
tionwillimpactestablishmentsinwhichworkersperformsignificanttext
recognition tasks and will change the factor demands of these “exposed
establishments.”
Tomaketheseideasprecise,wedefineestablishmente’sexposuretoAIas
∫ ‘ ðxÞdx
exposure to AI 5 x∈T e\TA e , (4)
e ∫ ‘ ðxÞdx
x∈T e e
wheretheemploymentsharesaremeasuredbeforetheadvancesinAItake
place.Thismeasurerepresentstheshareoftasksperformedinanestablish-
mentthatcannowbeperformedbyAI-poweredalgorithms.6
WenextexplorehowadvancesinAIimpactAIactivityandthedemand
for(non-AI)workers.
5Thisassumptioncanberelaxedinvariousways. First,thetechnologycanbe
moregeneralthanLeontief,sothatfactorpricesaffecthowintensivelyAIworkers
areused.Second,establishmentsmaybeallowedtosubstituteoutsourcedAIwork-
ersforin-houseservices.Thefirstmodificationwouldnothaveanymajoreffecton
ourresults,whilethesecondwouldimplythatourproxyforAIactivityatthees-
tablishmentlevelmayunderstatetheextentofAI,potentiallyleadingtoattenuation
ofourestimates.Thecommontechnologyassumptionineqq.(2)and(3)canalso
berelaxedbutisusefulforsimplifyingtheexpositionbyensuringthatdifferences
infactordemandsacrossestablishmentsaredrivenentirelybytaskstructures,mak-
ingthelinkbetweenthemodelandourempiricalapproachmoretransparent.
6Whenj 5∞,asinpropositions1anÐd2belowandtheshareofAIalgorithmsin
costisinitiallysmall,exposuretoAIis aðxÞdx,whichgivestheshareoftasks
thatcannowbecompletedwithAIintx∈ oT te\ aT lA
costs.
S300 Acemogluetal.
B. TaskStructureandAIAdoption
ToillustratehowthetaskstructuredeterminesAIadoption,wefollow
Acemoglu and Autor (2011) and Acemoglu and Restrepo (2018, 2019)
andassumethatj 5 ∞,sothatalgorithmsandlaborareperfectlysubstitut-
ablewithinatask.Wealsofocusontherealisticcaseinwhichtheinitialcost
share of AI, denoted by sA (5ðRk ðxÞ1wAI‘AIðxÞÞ=total costs), is small.
e e e
Additionally, we consider the problem of a single establishment, holding
thepricesofotherestablishmentsinthemarketasgiven.
PROPOSITION1.Supposethatj 5 ∞andtheinitialcostshareofAI,sA e,is
small.ConsideranimprovementinAItechnologiesthatincreasesg (x)
a
inTAandleadstotheuseofAIalgorithmsinthesetasks.Thentheeffects
onthecostshareofAIandin-houseAIemploymentaregivenby
dsA 5 exposure to AI ≥ 0
e e
and
(cid:6) (cid:7)
12sA
dln‘AI5 e 1ðε (cid:2)r 21Þ(cid:2)ð12sAÞ(cid:2)p (cid:2)exposure to AI ≥ 0,
e sA e e e e e
e
whereε > 1isthedemandelasticityfacedbytheestablishment,r > 0is
e e
theestablishment’spass-throughrate,andp ≥ 0istheaveragepercent-
e
agecostreductionintasksperformedbyAI.
TheproofofthispropositionisprovidedinappendixA,wherewealso
providetheexpressionsforthepass-throughrate,r,andaveragecostsav-
e
ingsfromtheuseofAIalgorithms,p.
e
ThepropositionshowsthatchangesinAIactivityandhiringofAIwork-
ersarebothproportionaltoexposuretoAI.Motivatedbytheseresults,in
ourempiricalworkweuseexposuretoAIasthekeyright-handsidevari-
ableandidentifygreateruseofAIwiththepostingofmorevacanciesforin-
houseAIworkers.
Althoughinthispropositionwefocusedonthecasewherej 5 ∞,asim-
ilar logic applies when j > 1 and AI does not fully replace workers in the
tasks it is used. In this case, AI advances still increase the cost share of AI
andthehiringofAIworkersinexposedestablishments.Whenj < 1,how-
ever, technological advances will not raise the cost share of AI because of
strongcomplementaritiesbetweentasksproducedbyalgorithmsandhumans.
C. AI,TaskDisplacement,andHiring
ThenextpropositioncharacterizestheeffectsofAIadvancesonhiringof
(non-AI)workers.ItsproofisalsoinappendixA.
ArtificialIntelligenceandJobs S301
PROPOSITION2.Supposethatj 5 ∞andtheinitialAIshareofcosts,sA e,is
small.ConsideranimprovementinAItechnologiesthatincreasesg (x)
a
inTAandleadstotheuseofAIalgorithmsinthesetasks.Theeffectson
non-AIemployment,ℓ,are
e
dln‘ 5 ð211ðε (cid:2)r 21Þ(cid:2)p Þ(cid:2)exposure to AI , (5)
e e e e e
whereε > 1isthedemandelasticityfacedbytheestablishment,r > 0is
e e
theestablishment’spass-throughrate,andp ≥ 0istheaveragepercent-
e
agecostreductionintasksperformedbyAI.
Proposition2showsthattheeffectsofAIadvancesonlabordemandare
proportionaltoourexposuremeasure.Morecentrally,itclarifiestheeffects
ofAIadvancesonlabordemand.Thedirectconsequenceofsuchadvances
istoexpandthesetoftasksperformedbyalgorithms,TA,andtoshrinkthe
setoftasksallocatedtoworkersinexposedestablishments.Becausej 5 ∞,
this technological improvement displaces workers from tasks in TA. This
displacement effect is captured by the “21” in the parentheses on the
right-hand side of equation (5). In addition, as emphasized in Acemoglu
andRestrepo(2018),thereallocationoftasksfromworkerstoalgorithms
reduces costs and expands establishment output, y (and this output re-
e
sponsedependsonthedemandelasticity andthepass-throughrate).This
“productivity effect,”themagnitude ofwhichisproportionaltothecost
reductions due to AI, p ≥ 0, increases hiring in nonautomated tasks. If
e
thesecondtermontheright-handsideofequation(5),ðε (cid:2)r 21Þ(cid:2)p ,ex-
e e e
ceeds 21, the productivity effect dominates and AI technologies increase
hiring.7 Otherwise, AI advances will reduce (non-AI) hiring in exposed
establishments.
Wemaketwoadditionalremarks.First,aswiththeresultsonAIactivity,
the main conclusions of proposition 2 can be generalized to the case in
whichj > 1.Inthiscase,notallworkerspreviouslyemployedinAIexpost
tasks would be displaced, but the substitution away from them to algo-
rithmswouldcreateanegativedisplacementandapositiveproductivityef-
fect,similartothoseintheproposition.
Second,ifdifferenttasksrequiredifferentskills,thentheadoptionofAI
technologies may also change the set of skills that exposed establishments
demand(andlistintheirvacancies).Skillsrelevantfortasksnowperformed
byalgorithmswillbedemandedlessfrequently,andnewskillsnecessaryfor
workingalongsideAIalgorithmsmayalsostartbeingincludedinvacancies.
Our empirical work will be based on equation (5). We will explore the
relationshipbetweenAIexposure,asdefinedinequation(4),andchanges
7Thisexpressionalsoclarifiesthatwhenthepass-throughrateislessthan1=ε,
e
the establishment’s price increases sufficiently that output does not expand and
thusemploymentalwaysdeclines.
S302 Acemogluetal.
inthenumberandskillcontentofthevacanciesanestablishmentposts.Spe-
cifically, we will look at whether exposed establishments hire more AI
workers,demanddifferentsetsofskills,andincreaseorreducetheirhiring
ofnon-AIworkers.
D. Human-ComplementaryAI
Wehavesofarnotconsideredhuman-complementaryeffectsofAI.The
possibilitythatAIwillcomplementworkersengagedinexposedtaskscan
becapturedbyassumingthata(x)increasesforexposedtasks(seeeq.[1])
or,alternatively,thatj < 1,sothatalgorithmsandhumanlaborarecomple-
mentary within a task (or both). This type of human-complementary AI
mayincreaselabordemandbecausealgorithmsraisehumanproductivityin
exactlythetasksinwhichAIisbeingadopted.
Evidence that AI is associated with greater establishment-level employ-
mentwouldbeconsistentwiththehuman-complementaryviewbutcould
alsobeconsistentwithtasksubstitutionassociatedwithlargeproductivity
gainsthatnonethelessincreasehiringatexposedestablishments.Conversely,
evidenceofnegative,orevenzero,effectswouldweighagainstboththe
human-complementaryviewandthepossibilityoflargeproductivitygains
fromAI—sincebothAI-humancomplementarityandlargeproductivity
effectsboostingemploymentinnonautomatedtaskscouldgenerateapos-
itiverelationshipbetweenAIexposureandestablishmentshiring.Ourev-
idencebelowfindsnegativeeffectsofAIexposureon(non-AI)hiringand
thussuggeststhatthecurrentgenerationofAItechnologiesispredominantly
taskreplacingandgeneratesonlymodestproductivitygains.8Itremainspos-
sible that other AI technologies than the ones we are proxying here could
havedifferenteffects.
E. MeasuringExposuretoAI
Propositions1and2showthatweshouldseetheeffectsofadvancesinAI
inestablishmentswithtaskstructuresthatmakethemhighlyexposedtoAI.
Differencesinexposureare,inturn,drivenbythedifferenttaskstructures
acrossestablishments.Inourempiricalexercise,wewillusetheoccupational
mixofanestablishmentpriortothemajoradvancesinAItoinferitstask
structure and compute its exposure to AI. Formally, we assume that the
setoftasksintheeconomy,T,ispartitionedintotasksperformedbyaset
ofdistinctoccupationsanddenotethesetoftasksperformedinoccupation
o ∈ ObyTo.Eachestablishmente’staskstructureisthusrepresentedbythe
setofoccupationsthattheestablishmentemploys,denotedbyO ⊂ O,and
e
8Orthatproductivitygains,ifpresent,havelittleeffectondemand,potentially
becauseoflowpass-throughrates.
ArtificialIntelligenceandJobs S303
so T
e
5 [ o∈OeTo. For example, some establishments will employ accoun-
tantsandtheirproductionwillusethesetoftasksaccountantsperform,while
others require the tasks performed bysecurity analysts orretailclerks and
thus hire workers into these occupations. In our empirical work, we will
use the occupational indices provided by Felten, Raj, and Seamans (2018,
2019), Webb (2020) and Brynjolfsson, Mitchell, and Rock (2018, 2019) to
identify the set of occupations involving tasks where AI can (or could) be
deployed.WewillthencomputemeasuresofAIexposurebasedontheoc-
cupationalstructureofanestablishment.9
III. Data
WenextdescribetheBGdata,documentthatitisbroadlyrepresentative
ofemploymentandhiringtrendsacrossoccupationsandindustries,present
our AI exposure indices, and document their distribution across occupa-
tionsandtheirevolutionovertime.
A. BurningGlassData
BurningGlasscollectsdatafromroughly40,000companywebsitesand
onlinejobboards,withnomorethan5%ofvacanciesfromanyonesource.
BGappliesadeduplicationalgorithmandconvertsthevacanciesintoaform
amenabletodataanalysis.Thecoverageisthenearuniverseofonlinevacan-
cies from 2010 onward in the United States, with somewhat more limited
coverage in 2007. Our primary sample comprises data from the start of
2010untilOctober2018,althoughwealsomakeuseofthe2007data.The
vacancydataenumerateoccupation,industry,andregioninformation;firm
identifiers;anddetailedinformationonoccupationsandskillsrequiredby
vacancies,garneredfromthetextofjobpostings.
AkeyquestionconcernstherepresentativenessofBGdatagiventhatthe
sourceofthevacanciesisonlinejobpostings.Figure1showsthatBGdata
9Formally,theseAIindicesaretheempiricalanalogofourtheoreticalexposure
toAImeasureineq.(4).Toseethis,notethat
AIindexo 5
∫ x∈To\TA‘ðxÞdx
,
∫ x∈To‘ðxÞdx
where‘(x)isaverageeÐmploymentintaskxandwedenoteaverageemploymentin
occupationoby‘o 5 ‘ðxÞdx.When‘ðxÞ5 ‘ðxÞ,whichfollowsfromourcom-
x∈To e
montechnologyassumption,theexposuretoAImeasureisequaltotheemploy-
mentweightedaverageoftheoccupationAIexposuremeasure:
∫
o o∈O oeA oI ∈Oin e‘d oexo‘o 5 o o∈Oe ox∫∈T x o∈ ∈o T\ OoT ‘ eA ð ‘‘ xð oÞx dÞ xdx‘o 5 ∫ x ∫∈ xT ∈e\
T
eT ‘A ð‘ xð Þx dÞ xdx 5 exposure to AI e:
S304 Acemogluetal.
FIG.1.—VacanciesinBurningGlassandJOLTS.Thisfigureplotsthetotalnum-
berofvacanciesinJOLTSandthetotalnumberofvacanciesinBurningGlassbyyear.
We multiply the number of job openings in JOLTS by a constant factor of 0.65
toarriveatanumberofvacanciesthatmatchestheconceptofavacancyinBurning
Glass.ThismethodfollowsCarnevale,Jayasundera,andRepnikov(2014).Acolor
versionofthisfigureisavailableonline.
closely track the evolution of overall vacancies in the US economy as re-
corded by the nationally representative Bureau of Labor Statistics (BLS)
JobOpeningsandLaborTurnoverSurvey(JOLTS).Theexceptionisthe
downturninBGpostingsdatabetween2015and2017.10FigureA1(avail-
ableonline)showsthatoverthe2010–18period,theoccupationalandin-
dustrycompositioninBGiscloselyalignedwithbothoveralloccupation
employmentsharesfromOccupationalEmployment Statistics(OES)and
withindustryvacancysharesfromJOLTS.11
10WhileJOLTSmeasuresasnapshotofopenvacanciespostedbyestablishments
duringthelastbusinessdayofthemonth,BGcountsnewvacanciespostedbythe
establishmentduringtheentiremonth.Weadjustthenumbersofjobopeningsin
JOLTStomatchBG’sconceptofvacancies,usingtheapproachdevelopedbyCar-
nevale,Jayasundera,andRepnikov(2014).ThedifferenceinconceptbetweenJOLTS
andBurningGlassvacancieslikelyaccountsforthedownturninBGpostingsdatabe-
tween2015and2017.
11WenotethatBGdatarepresentvacancyflowswhiletheOESreportsemploy-
mentstocks;thus,wedonotexpectthetwodatasourcestoalignperfectly.More-
over,onlinevacancypostingstendtooverrepresenttechnicalandprofessionaljobs
ArtificialIntelligenceandJobs S305
WemakeuseofBurningGlass’sdetailedindustryandestablishmentdata.
Whenthisinformation isavailablefrom thetextofpostings,vacanciesare
assigned a firm name and a location, typically at the city level, as well as
anindustrycode.Weclassifyeachfirmasbelongingtotheindustryinwhich
itpoststhemostvacanciesoveroursampleperiod.Wedefineanestablish-
mentofafirmasthecollectionofvacanciespertainingtoafirmandcommut-
ingzone(CZ).CZsaregroupsofcountiesthat,becauseoftheirstrongcom-
mutingties,approximatealocallabormarket(TolbertandSizer1996).
OfparticularimportanceforourpaperareBG’sdetailedskillandoccu-
pation coding. Vacancies in BGdata containinformation on skillrequire-
ments,scrapedfromthetextofthevacancy.Theskillsareorganizedaccord-
ing to several thousand standardized fields. Groups of related skills are
collectedtogetherinto“skillclusters.”Morethan95%ofvacanciesareas-
signedasix-digit(StandardOccupationalClassification[SOC])occupation
code.12
WeusetheseskilldatatoconstructtwomeasuresofAIvacancies,narrow
and broad. The narrow category includes a selection of skills relating to
AI.13ThebroadmeasureofAIincludesskillsbelongingtothebroaderskill
clustersofmachinelearning,AI,naturallanguageprocessing,anddatasci-
ence.AconcernwithourbroadAImeasureisthatitmayincludevarious
IT functions that are separate from core AI activities. For this reason, we
focusonthenarrowAImeasureinthetextandshowtherobustnessofour
main results with the broad occupation measure in appendix B. Figure 2
shows the evolution of postings of narrow and broad AI vacancies in the
BGdata,highlightingtherapidtakeoffofAIvacanciesafter2015,asnoted
intheintroduction.Whileasharpuptickisvisibleinallindustries,theright
panel of figure 2 shows that the takeoff is particularly pronounced in the
information,professionalandbusinessservices,finance,andmanufacturing
sectors.
Inwhatfollows,ourprimaryfocusisonAI-usingsectors,andwedrop
establishments belonging to sectors that are likely to be producing AI-
related products, namely, the information sector (NAICS sector 51) and
relativetobluecollarandpersonalservicejobs(Carnevale,Jayasundera,andRep-
nikov2014).
12Six-digitoccupationcodesarehighlygranular,includingoccupationssuchas
pestcontrolworker,collegeprofessorinphysics,andhomehealthaide.
13Theskillsaremachinelearning,computervision,machinevision,deeplearning,
virtualagents,imagerecognition,naturallanguageprocessing,speechrecognition,pat-
ternrecognition,objectrecognition,neuralnetworks,AIchatbot,supervisedlearn-
ing,textmining,unsupervisedlearning,imageprocessing,Mahout,recommendersys-
tems, support vector machines, random forests, latent semantic analysis, sentiment
analysis/opinionmining,latentDirichletallocation,predictivemodels,kernelmeth-
ods,Keras,gradientboosting,OpenCV,XGBoost,Libsvm,Word2vec,machinetrans-
lation,andsentimentclassification.
S306 Acemogluetal.
FIG.2.—ShareofAIvacanciesinBurningGlass.Theleftpanelplotstheshareof
vacanciesinBurningGlassthatpostaskillinthebroadornarrowAIcategories,as
definedinthemaintext.TherightpanelplotstheshareofnarrowAIvacanciesin
BurningGlass,byyear,ineachindustrysector.pp5percentagepoint. Acolorver-
sionofthisfigureisavailableonline.
theprofessionalandbusinessservicessector(NAICSsector54).Theformer
includes variousinformationtechnology industries,likely tobeselling AI
products, while the latter contains industries such as management consul-
tancy,likelytobeintegratingAIintootherindustries’productionprocesses.
B. AIIndices
WestudythreemeasuresofAIexposure.Eachisassignedatthesix-digit
SOC occupation level, and each is designed to capture occupations con-
centrating in tasks that are compatible with the current capabilities of AI
technologies.
ThefirstmeasureisfromFelten,Raj,andSeamans(2019).Itisbasedon
datafromtheAIProgressMeasurementproject,fromtheElectronicFron-
tier Foundation. The Electronic Frontier data identify a set of nine appli-
cationareasinwhichAIhasmadeprogresssince2010,suchasimagerec-
ognitionorlanguagemodeling.Feltenetal.useAmazonMTurktocollect
crowdsourcedassessmentsoftherelevanceofeachoftheseapplicationareas
tothe52O*NETabilityscales(e.g.,depthperception,numberfacility,and
ArtificialIntelligenceandJobs S307
written comprehension). The authors then construct the AI occupational
impact for each O*NET occupation as the weighted sum of the 52 AI
application-abilityscores,whereweightsareequaltotheO*NET-reported
prevalenceandimportanceofeachabilityin |
313 | mit_edu | AI-K-12_final-V3.pdf | TOPICAL POLICY BRIEF
Labeling AI-Generated Content:
How Policy Can
Help Ensure the
Proper Use of AI in
K-12 Education
MIT Responsible AI for Social Empowerment
and Education (RAISE) Initiative
Daniella DiPaola
Andrés F. Salazar-Gómez
Hal Abelson
Eric Klopfer
David Goldston
Cynthia Breazeal July 19, 2024
aipolicy.mit.edu
AI Policy Brief: K12 Education
MIT Responsible AI for Social Empowerment and Education (RAISE) Initiative
Daniella DiPaola, Andrés F. Salazar-Gómez, Hal Abelson, Eric Klopfer,
David Goldston, and Cynthia Breazeal
I. Introduction – Promise and Perils
Artificial intelligence (AI) has the potential to significantly improve K-12 education if
implemented appropriately – in ways that ensure its safe and equitable use. We use the
term “AI” broadly to mean any technology that uses data to make predictions and
decisions, or creates new content. AI offers great promise for students, teachers, and
school administrators. Examples include: the personalization of learning through virtual
chatbots (e.g., a ChatGPT-powered system such as Khanmigo) that provide K-12
students hints and clues tailored to where the child is in the learning process1; social
robots that support children learning to read through predictive algorithms of vocabulary
knowledge2; automated grading systems that allow teachers to use their time more
effectively by providing detailed feedback and scoring based on a student’s answers3;
and predictive systems for school administrators that identify high school students at
risk of dropping out4.
Children have unique developmental needs and vulnerabilities, and AI should be
integrated into schools in a way that enables kids to flourish while keeping them safe.
Doing so should not displace or reduce the role of teachers, who play a critical role in
students’ education and social development.
There is little or no federal guidance on AI for K-12 education. Little systematic research
at scale exists on how and when students learn better with AI, and states and school
districts are left on their own in a “Wild West” of competing claims, with AI offering
unverifiable allure and unknown risks. Nonetheless, states are creating an array of AI
implementation recommendations and guidelines for K-12 education, with no consensus
among them. Different states have different definitions and recommendations for topics
as important as plagiarism and AI literacy5. Federal policies based on state-of-the-art
research can help guide states and localities, while still leaving room for states and
school districts to use AI in a way that meets their particular needs.
1Bidarian, N. (2023, August 21). Meet Khan Academy’s Chatbot Tutor | CNN business. CNN.
2Zhang, X., Breazeal, C., & Park, H. W. (2023, March). A Social Robot Reading Partner for Explorative Guidance. In
Proceedings of the 2023 ACM/IEEE International Conference on Human-Robot Interaction (pp. 341-349).
3Metz, C. (2021, July 20). Can A.I. grade your next test?. The New York Times.
4Page, L., & Gehlbach, H. (2018, January 16). How Georgia State University used an algorithm to help students
navigate the road to college. Harvard Business Review.
5Dusseault, B. (2024, March). New state AI policies released: Signs point to inconsistency and fragmentation | Center
on Reinventing Public Education (CRPE), Arizona State University.
POLICY RECOMMENDATIONS FOR THE USE OF AI IN EDUCATION
2
This table enumerates some of the major potential benefits and challenges of using AI
in the classroom. Successful federal policy would make the benefits more likely and the
harms less so.
Potential Benefits Potential Harms
● Increase students’ learning gains ● Enable the collection of vast amounts of
through personalized learning personal data, compromising privacy.
experiences. ● Produce inaccurate, inappropriate, or harmful
● Complement instruction by outputs.
teachers. ● Favor certain learning approaches or abilities.
● Promote creative learning, ● Exhibit bias.
designing, and making. ● Exacerbate inequities among school districts.
● Reduce barriers to access to ● Undermine the development of basic skills
advanced knowledge. such as writing.
Beyond concerns related to weaknesses inherent in AI systems themselves, harms
could result from the way AI is deployed. Are there ways AI can be used to maximize
educational benefits and minimize potential detrimental effects, like the potential loss of
needed basic, age-appropriate skills? How can educators and parents keep adapting as
AI systems continue to improve, and as society and regulation adjust to AI’s new
capabilities?
II. The Federal Role
While pre-college education is primarily a state and local responsibility, the federal
government has a critical role in shepherding AI into the classroom and ensuring its
appropriate use. Many states and localities lack the expertise and capacity to design or
enforce technical requirements for AI systems, even more so if AI systems are offered
by just a small number of national suppliers. The federal government is more likely to
have the wherewithal, both financial and human, to fund research, set technical
standards, assist procurement, promote transparency, and provide guidance on many
crucial aspects of AI in education. If the federal government plays its role properly,
states and school districts will have the information and the funding they need to make
their own decisions on exactly how to integrate AI into teaching and the curriculum.
We discuss five areas where increased federal activity is needed: research; standards
development and auditing; procurement assistance; educational guidance; and AI
literacy. We describe the federal role in fulfilling each of these needs and then offer
steps the federal government could take to meet the challenge.
Research on AI in K-12 Education
The federal government has long been an important funder of research in critical areas
likely to be neglected or underfunded by the corporate sector or others. Two broad
research areas need much greater focus if AI is to be used safely and effectively in the
POLICY RECOMMENDATIONS FOR THE USE OF AI IN EDUCATION
3
K-12 classroom. The first is research on AI systems themselves. AI systems need to be
more accurate, less biased, less likely to output dangerous or offensive material, and
less likely to compromise privacy, among other deficiencies, if they are to be entrusted
with supporting the education of minors. [See issue brief on LLMs.]
Second, and equally important, much more research is needed to understand how the
use of AI might affect teachers and students, and how to use AI to maximize its
educational benefits. We understand too little to begin with about how students learn
best, and far less about the impact of a new technology. AI is not a magical elixir that
can be added to education with guaranteed positive results. Ongoing research,
monitoring, and evaluation will be needed to understand how to use AI optimally.
Research on how to most effectively deploy AI is needed in education as it is in other
fields. The research, piloting, and monitoring will have to be done in a way that does not
sacrifice children and their teachers as guinea pigs who are handed AI tools before we
know how they would work best. One area of research should try to figure out how to
help any children whose education has been harmed by using AI before its implications
are sufficiently understood.
RECOMMENDATIONS: The Department of Education (ED) and the National Science
Foundation (NSF) should create programs specifically to address the research issues
discussed above. Funding for this research should be a budget priority, including
funding for field research, especially at scale, to see how AI is actually being used in a
diverse range of classrooms and the impacts of that use. The National AI Research
Resource (NAIRR), now being piloted, should make computer time available for the
research described above. However, it should be recognized that the NAIRR, even if
fully funded, would not have adequate resources to help carry out major research
projects.
ED and NSF will also need to take steps to ensure that the results of the research they
fund are broadly disseminated and that their implications are clear to state education
officials, school administrators, and teachers.
Technical Standards and Auditing
As in many areas of technology, the federal government has a key role to play in
developing technical standards (and conducting the research needed to do so), even
though private entities or other levels of government can decide which standards to
adopt. Standards development is especially important in a field like AI – which is rapidly
changing and largely dominated at the moment by a handful of companies (at least in
terms of the broad platforms on which more targeted systems are built). And it is even
more important when what’s at stake is education – a largely public undertaking with
enormous societal impacts. As noted in the section above, many of the AI “safety”
issues that are central to education are of concern to AI users more broadly and are
already a focus of federal standards-development work. As AI is incorporated into more
systems, standards also will be needed to describe which AI tools should require
auditing.
POLICY RECOMMENDATIONS FOR THE USE OF AI IN EDUCATION
4
Standards can be effective only if they are actually followed; those using AI in K-12
education should not be left to rely on chance when it comes to how AI systems actually
function in practice. An auditing ecosystem will need to be developed to ensure that AI
systems perform as advertised, especially in terms of concerns such as accuracy,
fairness, and privacy. [See main AI Policy Brief.] This is even more important for AI
systems that will be interacting with children. Auditing itself requires the creation of
technical standards – on what to test for, and how – and then trustworthy entities are
needed to conduct the audits. Audits can be done before an AI system being deployed,
or after, or both – with each type of audit having its strengths and weaknesses.
One possible approach would be for the National Institute of Standards and Technology
(NIST), in collaboration with the ED and NSF, to develop standards and guidelines for
auditing and red-teaming of AI designed to be used in K-12 education. States and
localities could then decide whether to require that AI systems used in their school
systems be audited, and that could be done in accordance with the federal standards
(or something based on them). A federal, state, and/or local system – or a non-profit
one – would need to be set up to certify auditors. We believe that auditing AI systems
for pre-college education must be performed by third parties, not by those developing or
selling the AI systems, to avoid conflicts of interest inherent in self-audits.
Privacy should be one aspect of AI that is audited. The issue is not only whether
individuals, institutions, or companies could get access to identifiable data about an
individual using the AI, but also whether data could be aggregated across different
platforms. Third-party actors could also collect school data and combine it with other
behavioral data from children’s online presence. For example, an AI-driven tutor created
by a large tech company might collect the full history of a child’s interaction with the
system to develop more effective personalized tutoring responses, then sell the
information to a company focused on toy advertising. NIST already has a mandate to
create AI privacy standards, but more attention is needed specifically about the privacy
of minor children.
Regarding privacy, we can learn from existing federal initiatives designed to protect
privacy, including the Children’s Online Privacy Protection Act (COPPA) and the
Children’s Internet Protection Act (CIPA) – administered by the Federal Trade
Commission and the Federal Communications Commission, respectively. These laws
provide authority to issue guidelines or regulations for AI, but that has not happened.
The Kids Online Safety Act (KOSA) could also be used as the basis for privacy
regulations.
Developing an effective and reliable auditing ecosystem may be one of the most difficult
and critical steps in ensuring that AI systems are appropriate for K-12 education. [See
main AI Policy Brief.]
RECOMMENDATIONS: The White House has rightly made NIST the lead government
agency for setting technical standards for AI. NIST needs more funding to carry out this
POLICY RECOMMENDATIONS FOR THE USE OF AI IN EDUCATION
5
vital work. NIST should coordinate with ED and NSF to determine what specific
standards may be needed for AI systems used in education.
There are a range of options the federal government could then use to try to ensure that
audits are conducted, including (from strictest to more lenient):
● Prohibiting states and school districts that receive federal funding from buying or
using AI systems that have not been audited in accordance with federal
standards.
● Prohibiting federal funds from being used to buy or deploy AI systems that have
not been audited in accordance with federal standards and/or requiring post-hoc
audits as a condition of federal funding to buy or deploy AI systems.
● Providing additional funds when entities buy or deploy an AI system that has
been audited in accordance with federal standards.
● Creating “safe harbor” legal standards for AI systems used in education that have
been audited in accordance with federal standards. (This will be meaningful only
if there is a functional liability regime to begin with – an issue that goes beyond
education.)
● Withholding funds from school districts that have problems with AI systems that
were not audited before purchase or deployment. (This may, though, create
incentives not to report problems, or may penalize schools that are already
financially strapped.)
● Providing funding for post-hoc audits after an AI system has been deployed.
We believe that there should be some federal requirements that set a minimum
standard for auditing and safety for AI that could be used for K-12 education.
At the very least, ED should make sure states and school districts have access to any
NIST auditing standards and results with information on the implications for education.
Schools should not be using unaudited AI systems.
Procurement Assistance
Also, AI systems can be expensive, and poorer school districts in particular may not be
able to afford them. The federal government should ensure that AI does not create a
new “digital divide” because only the wealthiest districts can afford AI or appropriate AI.
(As discussed in the next section, the federal government should also take steps to
ensure that school districts, especially poorer ones, do not weaken education by over-
reliance on AI or use AI as a way to displace teachers.) The federal government should
also make sure that poorer school districts are not turned into guinea pigs, using
donated equipment without adequate testing or thinking about its optimal use.
RECOMMENDATIONS: ED could establish a program (or use a current program) for
competitive grants to states and/or school districts to fund the use of AI systems that
meet adequate standards. (See section above on standards.) There is a clear
POLICY RECOMMENDATIONS FOR THE USE OF AI IN EDUCATION
6
precedent for the federal government providing equitable access to technology for
education, the universal service Schools and Libraries Program (E-rate)6.
Those seeking grants should have to describe not only the kind(s) of AI systems that
the money would be used for but also how that AI would be deployed and monitored.
Any federal agency providing funding for AI should take steps to ensure that AI is not
used to displace teachers. AI should be used to enhance teaching, not to reduce the
number of teachers. ED and other agencies could require that the size of the teaching
staff not be reduced at least during the life of the grant.
Educational Guidance
Drawing on the research discussed above, the federal government should offer
guidance on the use of AI in education, teacher training and certification, and other
aspects of AI in education (even though final decisions on when, where, and how to
deploy AI will remain with states and localities). The federal government should also act
as a convener, by itself or with associations of education professionals or others,
bringing together school officials and teachers from around the country to discuss AI
issues.
One key issue is determining when using AI constitutes a breach of academic integrity –
when it is just “cheating.” We do not think that AI use should be subject to broad bans,
but rather that students should be taught when and how it is appropriate to use it as a
tool. That will, of course, differ by grade level and subject. Generative AI such as
ChatGPT can be a creative partner and help students better utilize time on their
assignments. When students enter the workforce, using these tools will be
commonplace.
However, there are cases in which AI should not be used. For example, in early
education (typically defined as Pre-K to 3rd grade), while students are learning how to
read and write, they should not be allowed to use AI in a manner that will interfere with
them developing their own skills. As AI is introduced and permitted, students should
practice using the tool in contexts when it is appropriate to do so, based on age-
appropriate guidelines and the teacher’s discretion, and understand the consequences
if they misuse it.
Educators should clearly state when students can and cannot use AI for their
assignments. We suggest that teachers not rely on AI tools to detect AI use on
assignments, as these tools currently have a high rate of false positives.
RECOMMENDATIONS: ED and NSF should begin preparing materials that can guide
states, school districts and teachers on AI education (see more on that below) and the
use of AI in education. Any guidance should be updated regularly.
6https://www2.ed.gov/about/inits/ed/non-public-education/other-federal-programs/fcc.html
POLICY RECOMMENDATIONS FOR THE USE OF AI IN EDUCATION
7
ED should consider creating a new office devoted solely to AI. While AI education and
the use of AI in education need to be integrated with all other aspects of primary and
secondary education, a separate office would not only highlight the importance and
uniqueness of the AI issues, but might help attract a cadre of relevant technical experts
to the department. ED and NSF should run grant programs to help fund the
development of curriculum, educational materials, and teacher training programs related
to AI.
ED and NSF, in collaboration with states and school districts, should develop metrics
that could help assess the impact of AI in the classroom on learning gains and skills
development. These should include metrics on the level of student engagement, and
the impact on struggling and disenfranchised students.
AI Literacy
The federal government should develop guidelines not only for how AI should be used
in K-12 education but also for what students should learn about AI itself. The goal
should be to help school districts create an AI-literate generation so students can
become effective citizens and productive workers in a world where AI will play a
prominent role.
We define “AI literacy” to include educating students on the appropriate and productive
use of AI. This encompasses gaining an understanding of how AI-powered technologies
operate, their applications, how to create with AI and work with AI effectively and
responsibly, and drawbacks and possibilities. AI literacy is relevant to many different
course areas and AI literacy should be infused across the curriculum, instead of just in
the computer science classroom.
For example, the appropriate use of generative AI could be addressed in arts
classrooms, and the impact of AI on society could be taught in civics classrooms. The
curriculum should clearly connect AI’s capabilities and these different subject areas.
K12 students should learn about AI through learning-by-making experiences and
responsible design practices7, where appropriate.
For students to be AI literate, they need AI-literate teachers. Teachers should receive
training on how to promote AI literacy and how best to use AI. The federal government,
along with states, localities, and non-profits, should develop ways to certify qualified AI
teachers.
The federal government should also issue guidelines on what students should know to
be considered AI literate and should help create curriculum and assessment tools on AI
literacy. Federal involvement in developing AI literacy standards and curricula for
students is needed because states are setting curriculum regulations that often vary
7https://dayofai.org/
POLICY RECOMMENDATIONS FOR THE USE OF AI IN EDUCATION
8
widely from each other. For example, West Virginia’s guidelines8 suggest that AI literacy
is only a part of computer science and technical courses, while North Carolina9
emphasizes AI literacy across all curriculum areas. U.S. students won’t be equally
trained in AI without a more cohesive approach across states.
RECOMMENDATIONS: ED and NSF should fund curriculum development on AI literacy
– both for teacher training and for students. ED and NSF should also fund teacher
training on AI literacy.
ED should update the national common core standards in digital literacy to encompass
AI and create a new category of common core standards specific to AI literacy. Grants
should encourage collaboration among non-profits, universities, teachers, and others in
developing AI literacy curricula and professional development materials.
III. Concluding Thoughts
The beneficial use of AI in K-12 education will not happen automatically and should not
be left to chance. The advent of AI raises, among other things, difficult questions about
the pace at which AI should be introduced in the classroom that education officials at all
levels will have to answer. How can a school find the proper balance – not holding back
on AI so much that teachers and students are deprived, but not racing ahead so fast as
to be saddled with AI systems that either become quickly obsolete, have serious flaws,
or are used in ways that do more harm than good (before it’s even clear how to evaluate
that)?
Finally, while K-12 education raises some very specific issues about AI, education will
be greatly affected by larger trends in AI and AI policy. Many concerns about AI
systems in education – accuracy, bias, etc. – apply to many uses of AI to at least some
degree. Steps needed to facilitate the proper use of AI, like the development of auditing
and liability regimes, are especially important for education.
K-12 education should prepare children to be successful adults in their personal and
professional lives. They should be prepared to be informed citizens, creative makers,
and critical thinkers. AI's role in the education and training of students (and teachers)
needs to be carefully considered and repeatedly iterated if we are to best serve the
future of our children, their teachers, and our nation.
Authors
Daniella DiPaola is a Ph.D. student in the Personal Robotics Group at the MIT Media Lab;
Andres F. Salazar-Gomez is a research scientist at MIT Open Learning; Hal Abelson is Class of
1922 Professor of Computer Science and Engineering; Eric Klopfer is Professor and Director of
8West Virginia Board of Education. (2024, January). Guidance, Considerations, & Intentions for the Use of Artificial
Intelligence in West Virginia Schools | Virginia Department of Education.
9North Carolina Department of Public Instruction. (2024, January). North Carolina Generative AI Implementation
Recommendations and Considerations for PK-13 Public Schools
POLICY RECOMMENDATIONS FOR THE USE OF AI IN EDUCATION
9
the Scheller Teacher Education Program and The Education Arcade at MIT; David Goldston is
Director of the MIT Washington Office; Cynthia Breazeal is the Dean for Digital Learning and a
professor of media arts and sciences at the MIT Media Lab.
Acknowledgments
The authors would like to thank Marc Aidinoff, Safinah Ali, Bill Bonvillian, Kate Darling,
Tom Giancola, Prerna Ravi, and Andrew Whitacre for their insights and advice. They
have not reviewed the paper, though, and are not responsible for its content. |
316 | deloitte | ca-the-state-of-generative-ai-in-the-enterprise-a-canadian-perspective-aoda-en.pdf | The State of Generative AI
in the Enterprise
A Canadian perspective
to scaling GenAI solutions
Q3 Insights
Contents
+ Closing the productivity gap ................................. 4
+ Adopting a balanced approach
that considers trust................................................. 5
+ Prioritizing key actions to
derive value from GenAI ........................................ 6
+ Measuring value with
specific KPIs ............................................................... 7
+ Conclusion ................................................................. 9
2
2024 has been a pivotal year for Generative AI (GenAI). surveyed 2,770 global respondents between May
GenAI experimentation has shown promising results, and June 2024 across several industries from all
2,770
which has led to growing investments, soaring levels, including board and C-suite members, as well
expectations, and emerging challenges. C-suites and as those at the president, vice president and director
boards are beginning to ask for measurable returns on levels. In this wave of the series, the focus was on AI Global respondents
investment – but can GenAI deliver? If projected returns governance, risk and compliance, data foundations,
Survey conducted between
are not met, interest in GenAI will fold as quickly as it and identifying how organizations are measuring
May and June 2024 across
has emerged. and communicating value.
several industries from
all levels
Our quarterly survey series, The State of Generative AI This article examines how Canadian businesses are
in the Enterprise, tracks global trends in decisions and adopting GenAI to boost productivity, build trust, capture
actions by leading organizations who are deploying value quickly, and measure impact effectively.
GenAI solutions. In Wave 3 of this report, Deloitte
3
+
Closing the
productivity gap
Canadian organizations are seeking clear
measures of productivity & efficiency
Survey respondents expressed that productivity is still the Improving Canada’s productivity has become an urgent
Productivity is the most sought after benefit
to realize the full potential of GenAI
number one most sought-after benefit for organizations priority. Though there is no consensus regarding the
looking to realize the full potential of GenAI globally (54%) cause for Canada’s limited productivity growth, there
and in Canada (53%). Of the Canadian organizations that have been signs that the downturn has been linked to the
have implemented AI solutions, 29% of respondents have structure of the business sector, regulatory bottlenecks
54%
noted improved productivity and efficiency as the most like interprovincial trade barriers, slow permitting
important benefit realized. processes to a lack of business investment. Globally,
Canadian companies are investing significantly less than
Global
While productivity remains the top priority for their peers – the CD Howe Institute estimates that for
organizations adopting GenAI, Canada’s stagnant each dollar a US company invests per worker, a Canadian
productivity performance has become front-page news, counterpart invests just 52 cents.4 It is no wonder that 29%
of Canadian
with an annual average rate increase of just 0.9%.1 Canada’s productivity is running 30% below the US.5 There respondents have noted
53%
improved productivity
Canada now stands as the second least productive is scope for Canada to take advantage of its fast-growing
and efficiency as the most
country in the G7.2 The downward trend was amplified AI ecosystem to reset. With rapid growth in AI talent, the
important benefit realized
following the 2020 pandemic leading policymakers, like breadth of venture capital funding, and large-scale increase
Bank of Canada Senior Deputy Governor Carolyn Rogers, in patent filings, Canadian organizations have a unique
Canada
to label it a “productivity emergency”.3 opportunity to leverage GenAI technologies to close the
productivity gap and meet their efficiency goals.6
4
+
Adopting a balanced approach
that considers trust
While trust is improving, it is still a concern for Canadian organizations
Globally, trust in GenAI is on the rise. Based on our Exercising caution must not be an excuse to delay Trust in GenAI
survey, 89% of respondents indicate that they moderately innovation. Rather, organizations should leverage this
trust (54%) or highly trust (35%) GenAI. In Canada, those opportunity to role model how to effectively balance risk
numbers resemble global metrics where 88% say they management while encouraging innovation. To achieve
35%
Highly trust
moderately trust (51%) or highly trust (37%) GenAI. this balance, it is crucial to establish guardrails for the
89%
54%
Moderately trust
responsible deployment of Generative AI solutions while
This growing confidence underscores a critical responsibility prioritizing upskilling to ensure that people understand the
for Canadian enterprises: to ensure that trust is not taken technology and know how to use it effectively.
Global
for granted but actively nurtured through transparency
and collaboration. It is essential that all parties are
engaged, and that governance is embedded into the
design, development, and implementation of AI solutions.
37%
Highly trust
88%
51%
Moderately trust
Canada
5
+
Prioritizing key actions to
derive value from GenAI
Canadian organizations are focusing on embedding
GenAI into functions and processes and managing risk
Respondents both globally and in Canada believe that To fully realize these benefits, the identification, Many organizations are now assigning dedicated individuals
embedding GenAI into organizational functions and prioritization, and design of specific use cases must be or teams to oversee AI implementation and establish
processes is the primary mechanism to extract maximal a collaborative effort between business functions and IT responsible AI practices.
value from GenAI solutions. From our survey, 23% of teams, ensuring alignment with broader organizational
Canadian respondents agree that learning to infuse GenAI goals. This approach ensures that GenAI initiatives are not Having accountable teams in place ensures that protective
into the DNA of the organization is essential. managed as isolated projects by the CIO, but are integrated measures are thorough and transparent, which in turn
into the core business strategy, and championed by builds trust within the organization. As more companies
What does this look like? Consider accountants that business leaders. This helps to ensure that the GenAI tool commit to ethical practices and strong risk management
can leverage GenAI to convert PDF invoices into excel or process is not only effective at its intended purpose, but principles, trust in GenAI will continue to grow.
spreadsheets, lawyers using Generative AI for lease also sustainable and ideally, scalable.
abstraction, case workers leveraging intelligent AI query
chatbots to gain access to the most relevant information Among Canadian survey respondents, 18% identified 23%
across an enterprise. Across all these roles, GenAI effective risk management as the second most important
embedded into day-to-day processes fundamentally factor for successfully scaling GenAI solutions, just behind
Of Canadian respondents agree that
transforms roles and responsibilities, automates tedious integrating GenAI into organizational functions and
learning to infuse GenAI into the DNA
work and allows humans to focus more on human- processes. But with the technology evolving so rapidly, how
of the organization is essential
centered tasks. can risk management be effectively implemented?
6
+
Measuring value
with specific KPIs
Canadian organizations are using specific KPIs to
measure and communicate the value of GenAI
To fully unlock the benefits of GenAI, organizations must the technology. By tracking and analyzing these metrics, Participants using GenAI-specific key
performance indicators (KPIs) to evaluate
effectively measure the productivity and efficiency gains organizations can demonstrate the tangible value of
the success of their investments
from new implementations. In our global report, we GenAI and make data-driven decisions to optimize future
asked how respondents are tracking and communicating investments. Measuring productivity gains accurately is
the value created by GenAI. Globally, 48% reported critical to maximizing GenAI’s value and ensuring its long-
using GenAI-specific key performance indicators (KPIs) term integration into business operations.
48%
to evaluate the success of their investments. In Canada,
that number was even higher, with 57% of participants
leveraging specific KPIs.
Global
These targeted KPIs not only offer a clear, quantifiable
view of how GenAI is impacting business processes but
also serve as a strategic guide for further investments in
57%
Canada
7
+
Measuring value
with specific KPIs (cont’d)
Direct/
KPI indirect Description
KPIs can be measured in two ways: as direct or Response time Direct Measures the time to coherent and accurate natural
language response
indirect indicators (i.e., primary and secondary impacts).
Sample KPI metrics:
User satisfaction Indirect Analyzes quality of user experience
Resource utilization Direct Evaluates worker utilization and machine downtime
Cost reduction Direct Measures the reduction in operational costs achieved through
GenAI implementation (e.g., reduced human labor hours, lower
processing costs)
Inventory optimization Indirect Measures throughput of inventory
Error rate Direct Tracks the frequency of incorrect or nonsensical outputs, providing
insights into the reliability of GenAI outputs
Adoption rate Indirect The percentage of employees or users regularly utilizing GenAI in
their day-to-day tasks after implementation
8
Conclusion
Approach to scaling Generative AI
successfully across your enterprise
To maximize the effective implementation of GenAI, Canadian
organizations must prioritize education, strengthen data
foundations, create an environment that champions robust AI
governance and continuously monitor for risk and compliance.
As many organizations are seeking tangible benefits from
GenAI, understanding how to communicate and measure value
regarding these solutions will be vital in the coming months.
Organizations should consider the following approaches:
9
Conclusion (cont’d)
1 2
Closing the productivity gap Adopting a balanced approach that
considers trust
GenAI offers a crucial opportunity to close the
productivity gap, especially for small and medium-
Balancing risk management with the encouragement of
sized enterprises, which dominate Canada’s business
innovation can enhance public trust. AI governance must be
landscape. By integrating GenAI into their operations,
embedded in the design, development, and implementation
SMEs can boost efficiency, accelerate growth, and
of AI solutions to ensure transparency and build confidence.
contribute to improving national productivity.
There is inherent risk in adopting GenAI solutions, but it’s
worth noting that there is also risk of complacency—trying
to be too risk averse is a risk in and of itself.
10
Conclusion (cont’d)
3 4
Prioritizing key actions to drive Measuring value from specific KPIs
value from GenAI
Accurately measuring productivity gains (direct or
To maximize the value of GenAI, it is crucial to deeply indirect) through specific KPIs is essential for maximizing
embed it into various functions and processes. the value of GenAI. By identifying and tracking these
Collaborating with business functions and IT teams metrics, organizations can demonstrate the tangible
to identify, prioritize, and design use cases will benefits of GenAI and make informed decisions about
ensure that GenAI becomes a core component of the their investments.
business strategy. Don’t feel you need to solve every
problem yourself, leverage ecosystem partners and
alliances to realize solutions.
Leaders must champion these initiatives to ensure Canada Access the Q3
report here.
remains competitive globally as we embrace the potential of AI.
11
Endnotes
1 Prompting Productivity: Generative AI Adoption by Canadian Businesses; Canadian Chamber of Commerce
(2024) ) https://bdl-lde.ca/wp-content/uploads/2024/05/Prompting_Productivity_Report_May_2024.pdf
2 Ibid
3 https://globalnews.ca/news/10384078/bank-of-canada-productivity-emergency/
4 Opinion: The Budget got one thing right-living standards are slipping. Then it made things worse. Financial
post: https://financialpost.com/opinion/budget-admits-living-standards-slipping-makes-things-worse
5 Canada’s Growth Challenge: Why the economy is stuck in neutral. RBC: Canada’s Growth Challenge: Why
the economy is stuck in neutral - RBC Thought Leadership
6 Impact and opportunities: Canada’s AI Ecosystem – 2023. Deloitte: https://www2.deloitte.com/content/
dam/Deloitte/ca/Documents/press-releases/ca-national-ai-report-2023-aoda-en.pdf
1122
Contact
Audrey Ancion
Partner, AI & Data
[email protected]
Aisha Greene
Director, Office of Generative AI
[email protected]
Contributors
Jas Jaaj Bram Judd
Managing Partner, Global AI Ecosystems Senior Consultant,
and Alliance Leader, Deloitte Global Office of Generative AI
Nihar Dalmia Andrew Klein
Partner, AI and Data, Consultant,
Consulting Practice Office of Generative AI
1133
About Deloitte
Deloitte provides audit and assurance, consulting, financial advisory, risk advisory, tax, and related services to public and private clients spanning multiple industries. Deloitte
serves four out of five Fortune Global 500® companies through a globally connected network of member firms in more than 150 countries and territories bringing world-class
capabilities, insights, and service to address clients’ most complex business challenges. Deloitte LLP, an Ontario limited liability partnership, is the Canadian member firm of Deloitte
Touche Tohmatsu Limited. 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.
Our global Purpose is making an impact that matters. At Deloitte Canada, that translates into building a better future by accelerating and expanding access to knowledge. We
believe we can achieve this Purpose by living our Shared Values to lead the way, serve with integrity, take care of each other, foster inclusion, and collaborate for measurable
impact.
To learn more about Deloitte’s approximately 412,000 professionals, over 14,000 of whom are part of the Canadian firm, please connect with us on LinkedIn, Twitter, Instagram,
or Facebook.
© Deloitte LLP and affiliated entities.
Designed and produced by the Agency | Deloitte Canada. 24-9715586 |
317 | deloitte | DI_Governance-of-AI_A-critical-imperative-for-todays-boards.pdf | i
sdraob
s’yadot
rof
evitarepmi
lacitirc
A
:IA
fo
ecnanrevoG
Governance of AI: A critical In a new Deloitte Global survey of board
directors and executives, almost 50% say AI is
imperative for today’s boards not yet on the board agenda. Is it time to step
up AI oversight in the boardroom?
Deloitte Global Boardroom Program
About the Frontier series
This report is the latest in Deloitte’s Frontier series, a set of research initiatives from the Deloitte Global Boardroom Program that
explores critical topics boards now face. Launched in 2021, the Frontier series has covered topics such as climate change, digital
transformation, trust, and talent. Learn more about The Deloitte Global Boardroom Program.
stnetnoc
fo
elbaT
sdraob
s’yadot
rof
evitarepmi
lacitirc
A
:IA
fo
ecnanrevoG
02 . . . Foreword
03 . . . Introduction
04 . . . Scaling up board engagement to bolster oversight
06 . . . Boards are eager to devote more time to AI-related
discussions
08 . . . AI adoption is a journey, not an instant solution
09 . . . Many boards are still getting up to speed on AI
11 . . . Near-term AI use in organizations is primarily focused on
productivity and efficiency
15 . . . Building a board governance model for AI
17 . . . Steps boards can take now to bolster AI oversight
19 . . . Endnotes
Foreword
W e are at an inflection point, not As the following research shows, it’s complicated. What
only for business and industry, is resoundingly clear, though, is that boards are eager
but for society at large. Board to spend more time on AI and gen AI, enhance their
members and executives alike are knowledge and experience, and accelerate the pace of
excited at the chance to shape a adoption in their organizations.
future powered by the latest tech-
nologies of the day, including artificial intelligence and This is a pivotal moment in the history of human inven-
generative AI. But this does not come without risk and tion—a moment future generations will certainly look
responsibility. The decisions leaders make today will back on. It’s imperative we reflect on the legacy we are
have pervasive impacts on both the organizations they creating as we navigate the path forward. We hope the
lead and societies around the world. Infusing a mindset insights from this Deloitte Global survey can spark and
of trust and ethics from the start will be vital to shaping inform meaningful conversations in your boardrooms
short-term and long-term adoption. While AI is not new, and with your management teams—inspiring a fresh
its scaled use in the enterprise and by employees brings look at whether and how AI and gen AI can play a
the question of governance and oversight of AI and gen role in your organization, all while keeping trust at the
AI into sharp focus. forefront.
So how are boards navigating these opportunities and –Lara Abrash, chair, Deloitte US
challenges? How are they balancing their time to help
ensure all pressing boardroom topics get the time and Arno Probst, Global Boardroom Program leader,
attention they deserve? And how are they confident that Deloitte Global
AI implementation is transparent, safe, and responsible
with the appropriate guardrails?
2
3
sdraob
s’yadot
rof
evitarepmi
lacitirc
A
:IA
fo
ecnanrevoG
Introduction
W hen the gen AI tool ChatGPT teams balance the wide array of opportunities and risks
exploded onto the global market that AI can introduce?
in November 2022, it democ-
ratized access to the newest AI In June 2024, the Deloitte Global Boardroom Program
capabilities within a matter of surveyed nearly 500 board members and C-suite exec-
days.1 Now, nearly two years utives across 57 countries to understand how involved
later, the growth in AI investment continues to rise: boards have been in AI governance. The survey explored
Gartner forecasts that worldwide IT spending will total sentiments about the current pace of adoption and the
US$5.26 trillion in 2024, an increase of 7.5% from 2023, board’s role in strategic oversight of this emerging
and points to generative-AI-related investments as the technology (see “Methodology”). We also spoke with
main reason behind this growth.2 board directors and Deloitte subject matter specialists
to understand how AI stewardship is evolving in board-
As organizations prepare to move past the piloting stage rooms around the world. Of note, while the survey asked
to integrate AI more broadly into strategy and opera- respondents about both generative AI and artificial intel-
tions, how active are boards in overseeing their organiza- ligence more broadly, our interviews revealed that many
tions’ approach to AI? Are they providing the right level business leaders are primarily focused on gen AI adop-
of stewardship to help the organizations’ management tion right now.
Scaling up board engagement to
bolster oversight
T he survey reveals that, so far, board-level at every meeting, 25% say it’s on the agenda twice a
engagement with AI has been limited: year, and 16% say AI is discussed annually. Nearly half
Across industries and geographies, AI is (45%) of respondents say AI hasn’t yet made it onto their
not a topic of discussion that comes up board’s agenda at all (figure 1).
often at board meetings. Only 14% of
respondents say their board discusses AI
Figure 1
In many organizations, AI is rarely discussed at the board level
How frequently AI is a topic on respondents’ board agendas
Not yet on the agenda
45%
Semiannually
25%
Once a year
Every meeting
16%
14%
Note: n = 468.
Source: Deloitte Global Boardroom Program AI/gen AI board governance survey, June 2024.
4
5
sdraob
s’yadot
rof
evitarepmi
lacitirc
A
:IA
fo
ecnanrevoG
Alongside this board research, each quarter, Deloitte US’s AI-related matters, it’s most frequently delegated to the
State of Generative AI in the Enterprise report3 surveys committee responsible for risk: either the risk and regu-
C-suite executives and board members from organiza- latory committee (25%) or the audit committee (22%).
tions that are actively implementing gen AI technologies
to continuously track progress and challenges leaders However, where broader AI oversight or specific gen
face. Its third-quarter 2024 report, which surveyed over AI oversight may ultimately land is still an open ques-
2,500 respondents, finds that while “promising pilots tion, Hodo says: “It’s still unknown whether governance
have led to more investments . . . many generative AI of gen AI should be a matter for the entire board of
efforts are still at the pilot or proof-of-concept stage.” directors or its audit committee, or [how the board will
Much like the use cases many organizations are experi- oversee] management.” Some aspects of AI oversight and
menting with, the survey shows that it’s still early stages governance might be relevant for the full board—those
for AI board governance, too. topics that are generally more pervasive—while some
might be more appropriate for a committee to handle.
While AI may not be on the board agenda itself, some Boards may also need to consider how oversight will
boards are talking about AI as part of the broader tech- be shared when some topics transcend committees, and
nology discussions they’re having with management. some may choose to establish an AI-specific committee.
“Rather than AI specifically, boards often see digital
transformation on their agenda, of which AI is a part,” We also asked respondents to comment on which C-suite
says Chikatomo Hodo, external director on the board of roles are primarily responsible for engaging with the
directors at ORIX Corporation, KONICA MINOLTA board about AI and gen AI. Most (69%) say they’re
Inc., Mitsubishi Chemical Group Corporation, and engaging with technology leaders, such as the chief
Sumitomo Mitsui Banking. information officer or chief technology officer. Half of
respondents say they are talking with their CEO about
When AI is on the board agenda, nearly half of respon- these topics, while about a quarter (26%) of respondents
dents (46%) say it is discussed at the full board level. say they are engaging with the chief financial officer.
Among those who say a committee has been tasked with
Boards are eager to devote more
time to AI-related discussions
M any respondents are cognizant that either they’re not satisfied with or they are concerned
their board’s current level of engage- about the amount of time devoted to discussions on AI
ment may not be enough to over- (figure 2).
see the opportunities and risks that
could manifest by using AI, particu-
larly gen AI. Nearly half (46%) say
Figure 2
Almost half of respondents would like their boards to be devoting more time to AI oversight
How do you feel about the amount of time your board spends on AI topics?
Not satisfied
34%
Neutral
31%
Somewhat satisfied
16%
Concerned
12%
Very satisfied
7%
Note: n = 468.
Source: Deloitte Global Boardroom Program AI/gen AI board governance survey, June 2024.
6
7
sdraob
s’yadot
rof
evitarepmi
lacitirc
A
:IA
fo
ecnanrevoG
Figure 3
Nearly half of respondents say their organizations need to accelerate progress on AI
implementation
How would you characterize the pace of adoption at your organization?
AI/gen AI is not considered
5%
relevant for our organization
Have yet to start AI/ gen
AI adoption throughout 35%
our organization
Need to accelerate 44%
Satisfied 14%
Very satisfied 2%
Note: n = 468.
Source: Deloitte Global Boardroom Program AI/gen AI board governance survey, June 2024.
However, most respondents do not believe their orga- Many respondents would like to see quicker progress:
nization is ready for broader AI deployment. Only Only 16% say they’re satisfied or very satisfied with the
3% of respondents think their organizations are very current pace of adoption (figure 3), and 44% say the
ready, while 41% say their organizations are not ready. pace needs to accelerate. These findings, combined with
This perception of a low state of readiness could also the lack of time devoted to AI on board agendas thus far,
be responsible for the growing sense of urgency as AI emphasize the opportunity boards have to contemplate,
capabilities continue to be developed. define, and scale AI oversight.
AI adoption is a journey, not an
instant solution
W hile many are eager to implement important,” Weber-Rey says. “You need the employees
AI, some boards and non-tech to be willing to adopt it because many of the areas in
leaders may not fully appreciate which you employ AI or gen AI are where there are a
how difficult it can be to scale lot of employees, like marketing, sales, risk, audit, and
AI across the enterprise. Deloitte financial reporting.”
US’s State of Generative AI in the
Enterprise Q3 2024 report explains how this challenge Deloitte US’s State of Generative AI in the Enterprise
is currently playing out: “Leaders grasp how essential report further emphasizes that both data and people
governance, risk, and compliance are for responsible are essential elements for scaling gen AI initiatives from
generative AI adoption. However, there still seems to be pilot to production.4 They are part of a critical suite of
a ‘knowing’ versus ‘doing’ gap for most organizations.” foundational elements including strategy, processes, risk
management, and technology. Getting all of these right
Daniela Weber-Rey, independent director at Fnac Darty as part of an organization’s strategy will be necessary in
and, until recently, HSBC Trinkaus & Burkhardt AG, the major transformations that AI and gen AI will likely
explains that organizations need to ensure they have a enable over the next few years and beyond.
strong foundation in place to support AI implementa-
tions. “If you don’t have the proper data management The challenge of scaling the use of AI, while remain-
system in place, you cannot really make full use of AI ing aligned to the organization’s integrated strategy, is
or gen AI. The data infrastructure must be established an area boards will need to deeply understand. Jean-
and there must be a proper data management system in Dominique Senard, chairman of the board of directors
the company.” at Renault, emphasizes, “There should be a close link
between the board and management—one that is trans-
Harder still, there’s also the challenge of getting parent, candid, and open.” Having a clear ambition for
employees to buy in and actually use the tools once AI and an understanding of the intended value it can
they are available. “Adoption by the employees is really create will be key to realizing long-term value.
8
9
sdraob
s’yadot
rof
evitarepmi
lacitirc
A
:IA
fo
ecnanrevoG
Many boards are still getting up
to speed on AI
B alancing the desire for rapid progress with the “art of the possible” in the use of AI in the enterprise.
the patience to scale effectively will be an Reflecting on the current level of understanding of AI
important line for organizations to walk. in the boardroom, over three-quarters of respondents
But to do that, their boards will need to (79%) say their boards have limited, minimal, or no
stay up to speed. According to the survey, knowledge or experience with AI (figure 4). Just 2% said
most boards have limited understanding of their boards were highly knowledgeable and experienced.
Figure 4
Nearly 80% of respondents say their boards have limited to no knowledge or experience with AI
How much does your board know about AI and how it works?
Highly knowledgeable and experienced 2%
Moderate knowledge and experience 19%
Limited knowledge and experience 41%
Minimal knowledge and experience 30%
No knowledge or experience 8%
Note: n = 468.
Source: Deloitte Global Boardroom Program AI/gen AI board governance survey, June 2024.
Figure 5
Boards are exploring various avenues to enhance AI fluency
Actions respondents say their boards are taking (multiple answers allowed)
Board members independently seeking to
57%
enhance their respective knowledge
Providing foundational education
to the board 40%
Bringing in external specialists to
discuss AI/gen AI on a regular cadence 37%
Adding AI/gen AI specialists through
new board directors 8%
Note: n = 468.
Source: Deloitte Global Boardroom Program AI/gen AI board governance survey, June 2024.
“Digital literacy needs to be elevated both within the In many countries, board education is recommended or
board and management, but we need to consider that mandated in corporate governance codes or other related
there should be a division of roles,” Chikatomo Hodo regulations or guidelines. “There is a binding recommen-
says. “In reality, there are not many external direc- dation in the French and German corporate governance
tors with an IT background, and many companies are codes that board members need to educate themselves.
prioritizing bringing a diverse range of skills and back- But there is also an obligation for the company to assist
grounds to their boards, rather than solely focusing on them in such training,” Weber-Rey explains. “This
bringing digital transformation and AI knowledge and certainly applies to gen AI or any technological changes,
experience.” such as digitalization. In the past, perhaps board direc-
tors could have gained a lot of knowledge of opera-
When considering board composition, our interviews tions by walking through the factory floors in certain
highlighted the importance of making sure the board has companies. Nowadays, you definitely need to have a
the right mix of skill sets, which could include skills in classroom-type training for gen AI, even just to get a
AI. Some boards are turning to external experts to add high-level understanding.”
to their AI literacy and fluency. Others are referring more
to operational teams in their business to understand the One approach to help boards achieve AI fluency is for
potential opportunities and challenges presented by AI. participants to use and experience AI—to “show rather
than tell.” Digital avatars, demos, and hands-on expe-
This survey shows boards are aware of the need to riences can be used as learning tools to help boards
upskill and are taking action to increase their knowledge to understand “the art of the possible” for AI in their
of AI in the boardroom (figure 5). organizations.5 These experiences can also be tailored to
10
11
sdraob
s’yadot
rof
evitarepmi
lacitirc
A
:IA
fo
ecnanrevoG
industry or sector, allowing an organization to mirror growing recognition that while board education is essen-
the most relevant opportunities and challenges in their tial, having bona fide digital expertise, particularly in AI,
operating environment.6 is increasingly seen as critical for effective governance in
today’s business landscape.
While these immersive experiences can play a critical
role in helping boards build AI fluency, some organiza- Regardless of the approaches boards pursue, it will be
tions might also wonder if education alone will ever be vital to continuously educate board members by bring-
enough. Perhaps the composition of the board should ing multidisciplinary and cross-industry perspectives to
change? Notably, 8% of respondents indicated their inform decision-making.
boards are starting to include AI specialists among their
new board directors (figure 5). This may highlight a
Near-term AI use in organizations
is primarily focused on
productivity and efficiency
T he board survey asked respondents the to incorporate AI in certain areas; 33% say they’re exper-
degree to which their organizations have imenting; and another 32% say AI hasn’t yet been incor-
been incorporating AI into their business porated into their organization’s business and operating
and operating plan. For their plans over plan over the next 12 months. Only 4% say AI is incor-
the next 12 months, about a third of porated throughout their near-term (next 12 months)
respondents (31%) have focused efforts business and operating plan (figure 6).
Figure 6
How is AI being incorporated into your organization’s business and operating plan over the
next 12 months?
Not Focused
incorporated Incorporated efforts in
at this time throughout certain areas Experimenting
32% 4% 31% 33%
Note: n = 468.
Source: Deloitte Global Boardroom Program AI/gen AI board governance survey, June 2024.
Among organizations that have started to use AI in some experience (46%), and core operations (40%) (figure 8).
capacity, respondents point to a number of strategic
areas aligned with these investments (figure 7). Jean-Dominique Senard says the evolution to AI at
Renault was a natural one, and they have already seen
Perhaps not surprisingly, enhancing productivity and its benefits as a tool for productivity and quality. “AI is
efficiency is the top strategic area (66%), followed by everywhere in the company, and it’s quite visible when
improving the customer experience (50%) and develop- you go across a Renault plant. We leverage it in the
ing new products or other innovations (46%). design department, engineering, customer relationships,
and, of course, in our vehicles. It was a normal evolution
These strategic priorities largely align with the planned for us and has proven to be incredibly powerful.”
areas for future AI investment. Top areas for future
investment include technology (63%), customer
12
13
sdraob
s’yadot
rof
evitarepmi
lacitirc
A
:IA
fo
ecnanrevoG
Figure 7
Productivity and efficiency enhancements, customer experience improvements, and
developing new products or other innovations are the main strategic goals aligned to
current AI adoption
Top reasons respondent organizations are leveraging AI (multiple answers allowed)
Enhancing
productivity and 66%
efficiency
Improving the
customer 50%
experience
Developing new
products or other 46%
innovations
Cost optimization
37%
New business
15%
expansion
Investment review,
including mergers 8%
and acquisitions
Note: n = 468.
Source: Deloitte Global Boardroom Program AI/gen AI board governance survey, June 2024.
Figure 8
Management at respondents’ organizations say advancements to technology, customer
experience, and core operations are top reasons to spend more on AI
Planned areas of focus for future AI investment (multiple answers allowed)
Technology (for example, tools, data) 63%
Customer experience 46%
Core operations 40%
Risk and regulatory 25%
Finance 23%
Talent 20%
No current investments or plans 12%
Note: n = 468.
Source: Deloitte Global Boardroom Program AI/gen AI board governance survey, June 2024.
14
15
sdraob
s’yadot
rof
evitarepmi
lacitirc
A
:IA
fo
ecnanrevoG
Building a board governance
model for AI
I n a complex environment in which opportuni- a few considerations to keep top of mind as boards
ties, challenges, and priorities frequently emerge, govern at scale: identifying and engaging with relevant
it’s critical for organizations to govern at scale. stakeholders, refining the board’s responsibilities, and
Whether related to AI oversight or any other managing risk through appropriate guardrails.
emerging issue, this means challenging ortho-
doxies while implementing balanced processes As organizations consider how boards can approach
that allow the board to operate efficiently, transpar- their AI-related responsibilities, it’s vital to first under-
ently, and in the best interests of the organization as a stand the organization’s key stakeholders. Right now,
whole—supporting growth, creating long-term value, respondents regard customers and employees as their top
and sustaining the organization. two stakeholders to consider in AI governance (figure 9).
But as AI scales, other stakeholder groups will become
Given this pivotal phase in gen AI experimentation and more of a factor in board decision-making.
adoption, what kind of role should the board play as they
build their governance models? This research showed
Figure 9
AI governance: Most respondents say customers and employees are the most important
stakeholder groups to consider
Percentage of respondents who identify the following as key stakeholders (multiple answers allowed)
Shareholders/
investors
Customers 73% Employees 69% 41% Regulators 39% Suppliers 34% General public 14%
Note: n = 468.
Source: Deloitte Global Boardroom Program AI/gen AI board governance survey, June 2024.
Daniela Weber-Rey explains, “Regulators are interested But while boards develop their oversight models, they
particularly around the risk and control aspects. But shouldn’t wait to start putting guardrails around risk
recent EU regulations on AI are new, and have not come management of AI. There is a danger people might not
fully into effect yet.” Interviewees echoed that as the understand the risks associated with gen AI. Close to
regulatory landscape expands, particularly in heavily three-fourths (72%) of respondents from Deloitte’s
regulated industries like financial services, regulators State of Generative AI in the Enterprise Q3 report esti-
will likely become an even more important stakeholder mate that less than 40% of their overall workforce has
for boards in the future. access to their organization’s approved gen AI tools.
Considering that employees may still be able to access
Which governance areas will be within the board’s some other tool on their own, the organization’s data
purview moving forward? Respondents to this board may be at risk if employees are using unsanctioned tools.
survey pointed to several areas they believe will be criti- As a result, the organization may have less control over
cal tenets of effective board oversight in the near future. how the company is integrating gen AI.
These include overall governance and oversight, includ-
ing ethics (57%); strategy development, including policy Daniela Weber-Rey agrees: “I love this phrase: ‘The
on AI (44%); risk and opportunity management (35%); biggest risk of gen AI is not taking the risk of gen AI.’”
and oversight of implementation (14%) (figure 10).
Figure 10
Respondents point to governance and ethical use, policy and strategy development, risk
management, and implementation as key tenets of AI board oversight
Percentage of respondents who say the following should be board responsibilities
Overall governance and oversight
57% 11% Staying informed of AI and its development
including ethical use
44% Strategy development including policy on AI 7% Developing and nurturing talent
35% Risk and opportunity management 3% Investment and resource allocation
14% Oversight of AI application and implementation
Notes: Based on analysis of open-ended answers; multiple choices were allowed; n = 407 (excludes vague or incomplete responses).
Source: Deloitte Global Boardroom Program AI/gen AI board governance survey, June 2024.
16
17
sdraob
s’yadot
rof
evitarepmi
lacitirc
A
:IA
fo
ecnanrevoG
Steps boards can take now to
bolster AI oversight
T he data shows that boards are eager – Regulatory scanning: How will the organiza-
to spend more time on AI and gen AI, tion review AI’s regulatory and compliance land-
enhance their knowledge and experience, scapes across the geographies and jurisdictions
and accelerate the pace of adoption in in which it operates?
their organizations. But how can boards
best navigate these opportunities and – Measurement: When and how will the organi-
challenges? The following are a few immediate actions zation review the measurement of the progress
boards can consider taking to bolster AI governance. and benefits of using AI in a way that ensures
robust oversight of investments without stifling
1. Put AI on the board agenda—and make it strategic. innovation?
Boards that aren’t yet discussing AI should consider
adding it as an agenda item. Areas to consider 2. Define the governance structure. To effectively
include: exercise oversight, boards will likely also need to
delineate and assign AI-related responsibilities.
– Cadence of discussions: How often should AI Considerations include:
be on the board agenda?
– Ownership of AI on the board: Which matters
– Special sessions: Would the board benefit from a should be discussed as a full board? Can some be
special session on AI or a board strategy retreat? delegated to a committee—and if so, which one?
– Strategy and scenario planning: Has the board – Receiving robust and beneficial information
scheduled an initial discussion with management from management: Is the board getting sufficient
to hear their analysis on risks and opportunities and appropriate information from management
related to AI and the AI ambition of the organ- about AI-related matters, including risk manage-
ization? ment and internal controls, to exercise oversight?
– Management oversight: How will the board – Having access to more leaders: Given the wide
assess, support, and, if necessary, challenge range of impacts across all areas of the business,
management’s point of view? is the board connecting to other key members
of the C-suite and business leaders beyond the
– Risk appetite: Has the board had a discussion CEO or CTO?
about risk appetite, both for the use of AI and,
more broadly, for the organization, given the
more uncertain environment that AI creates?
– Striking the right balance: Is board involvement – Revamping succession plans to be more tech-for-
too high-level to effectively govern the use of AI? ward: Have succession plans for the board and
Will deeper board education and engagement management been updated to focus on leaders
result in too much oversight? who have experience with emerging technolo-
gies, including AI? Have learning opportunities
been developed to help the pipeline of future
3. Evaluate and enhance AI literacy. To effectively
leaders expand their skills and expertise in these
oversee the opportunities and threats AI can intro-
technologies?
duce, boards should ensure they and their manage-
ment teams are AI literate. They may consider:
– Staying in the flow of action: How can the board
ensure it remains actively engaged in the evolving
– Finding opportunities for education to fill gaps
landscape of AI, guarding against complacency
in knowledge: What training and educational
and outdated perspectives and remaining agile
opportunities are available to help the board
and responsive to AI’s evolving capabilities?
upskill on AI and emerging technologies? Would
the board benefit from bringing in internal or
external experts to inform discussions?
– Reevaluating the skills matrix: Does board
composition need to be adjusted to recruit
board members with more experience with AI
and emerging technologies? What about in the
C-suite?
METHODOLOGY
The Deloitte Global Boardroom Program surveyed Industries represented include financial services those with values of US$10 billion or more (17%). (Note:
468 board members (86%) and C-suite executives (25%); manufacturing (16%); energy and resources percentages do not equal 100% due to rounding.)
(14%) in 57 countries from May to July 2024. Some (9%); business and professional services (8%); retail
respondents may serve at multiple organizations and wholesale (7%); technology (7%); health care About the Frontier series
as both executives and board members. and pharmaceuticals (5%); telecommunications,
media, and entertainment (3%); and various other This report is the latest in Deloitte’s Frontier series,
Responses were distributed across the Americas industries (20%). a set of research initiatives from the Deloitte Global
(42%), Asia Pacific (20%), and Europe, the Middle East, Boardroom Program that explores critical topics
and Africa (EMEA) (38%). Among the respondents, The survey includes respondents across a range of boards now face. Launched in 2021, the Frontier
43% serve at publicly listed companies, while 39% company sizes: Fifty-five percent of respondents series has covered topics such as climate change,
serve at privately owned companies, including represent organizations with equity market values of digital transformation, trust, and talent. Learn more
family-owned businesses. The rest came from a less than US$1 billion, followed by those with values about The Deloitte Global Boardroom Program.
mix of government and state-owned enterprises, between US$1 billion and US$10 billion (29%), and
as well as nonprofits.
18
19
sdraob
s’yadot
rof
evitarepmi
lacitirc
A
:IA
fo
ecnanrevoG
Endnotes
1. Kevin Roose, “How ChatGPT kicked off an AI arms race,” The 4. Ibid.
New York Times, February 3, 2023. 5. Deloitte, “The Generative AI dossier: A selection of high-impact
2. Gartner, “Gartner forecasts worldwide IT spending to grow use cases across six major industries,” accessed October 4,
7.5% in 2024,” press release, July 16, 2024. 2024.
3. Deloitte, “The state of generative AI in the enterprise: Moving 6. Ibid.
from potential to performance—Q3 report,” August 2024.
About the authors
Lara Abrash Karen Edelman
[email protected] [email protected]
Lara Abrash is the chair of Deloitte US. Abrash stepped into this Karen Edelman is a senior editor at Deloitte Insights, where she
role in June 2023 after serving for four years as the chair and chief leads content strategy for the Deloitte Center for Financial Services
executive officer of Deloitte & Touche LLP, where she was respon- and the Global Boardroom Program. She also serves as a talent
sible for overseeing the US Audit & Assurance business. She is a adviser for Deloitte’s Research and Insights team.
member of Deloitte Global’s Board of Directors and chair of the
Deloitte Foundation.
Arno Probst
[email protected]
Prof. Dr. Arno Probst is the leader of the Deloitte Global Boardroom
Program. In addition to his global role, Dr. Probst leads Deloitte
Germany’s Executive & Board Program and the Center for
Corporate Governance. He is a partner in Deloitte Germany’s Audit
& Assurance practice.
Acknowledgments
The Deloitte Global Boardroom Program would like to thank participating boardroom programs around the world who supported
this project. A special thanks to our designers Meena Sonar, Natalie Pfaff, and Harry Wedel, and our editorial team, Karen Edelman,
Elisabeth Sullivan, and Annalyn Kurtz from Deloitte Insights.
20
21
sdraob
s’yadot
rof
evitarepmi
lacitirc
A
:IA
fo
ecnanrevoG
About the Deloitte Global Boardroom Program
The Deloitte Global Boardroom Program brings together the knowledge and experience of Deloitte member firms around the world to
address critical topics of universal interest to company boards and management. Supplementing country programs, its mission is to promote
dialogue between corporations and their boards and management, investors, the accounting profession, academia, and government. In
addition to the publication of thought-pieces on critical topics, the Deloitte Global Boardroom Program hosts a series of must-see webinar
discussions with eminent panelists to help boards and management of global companies to stay current and challenge perceived wisdom.
To become a member |
318 | deloitte | in-ra-deloitte-ciso-guide-to-gen-ai-ap-noexp.pdf | The CISO’s Guide
to Generative AI
Opportunities, outcomes, and the urgency of now
2
True or false? Generative AI can help:
• Unlock new opportunity and value in an organization’s cybersecurity approach.
• Reduce costs and supercharge generation of reporting and intelligence products.
• Rapidly protect against sophisticated phishing attacks.
• Guide organizations in identifying critical information based on past actions.
• Make sense of regulatory and compliance guidance.
• Build a cybersecurity road map for now and the future.
If your answers were all true, then you’re thinking correctly
about this powerful technology.
Read on for more on how Generative AI can help transform
your organization’s cybersecurity approach.
© 2024. For information, contact Deloitte Asia Pacific Limited.
33
Generative AI is here.
What can it do for you?
The Generative AI (Gen AI) buzz is everywhere. People are
wondering what this new artificial intelligence (AI) can do for their
organizations, their data, and their security. It’s a complex question
that defies an easy answer.
Gen AI is a subset of AI in which machines create But while cyber events have long since eclipsed
new content in the form of text, code, voice, the capabilities of a traditional human security
images, videos, and processes. The technology may operations center, AI is deeply impactful in
truly revolutionize work and life. When it comes enhancements to cyber infrastructure and
to cybersecurity, Gen AI holds promise for both detect-and-respond capabilities. Deep learning
organizations and governments that need to protect models are well suited to detecting attacks.
themselves, create tools to automate reporting and
intelligence, reduce costs, grow more efficiently, sort But cyber leaders may still wonder: While AI has
through the varied and ever-changing regulatory increased our defense capabilities and postures,
atmosphere, and so much more. could Gen AI take us even further? How could
it be used to limit blast radiuses of attacks,
Gen AI can also provide new tools for bad actors protect against data loss, and expand our threat
who want nothing more than to leverage these response capabilities within budget and on time?
powerful technologies for negative outcomes In other words, can it help us get ahead—
and their own gain. Cyberattacks continue to and stay ahead—of attackers?
increase in both volume and tactics: in fact, more
than 90% of respondents to the Deloitte Global
2023 Future of Cyber survey reported at least
one compromise.
© 2024. For information, contact Deloitte Asia Pacific Limited.
4
GENERATIVE AI IS HERE. WHAT CAN IT DO FOR YOU?
Gen AI can do each of those things—and it holds To unlock the potential of Gen AI, cyber leaders
so much promise for better cyber outcomes for should first understand where it can help, the
organizations seeking out and defending against types of data it needs, and how to develop a plan
breaches. It’s fast and reasoned and can process of action that includes considerations for safety,
more knowledge than any one human can. It resilience, and trustworthiness.
has the potential to reduce costs, supercharge
security investigations, and speed up third-party Two things to remember: This is an evolution
risk assessments. of AI, not a net-new concept, and adoption plans
and risk management constructs can be
While more established AI capabilities (such as evolved accordingly. And like any true evolution,
machine and deep learning) can identify patterns these are long-term transformation efforts.
and make inferences, Gen AI can put it together Adopting Gen AI for cybersecurity is a capability-
while generating human-like responses and building effort. Treat it that way.
working at extraordinarily high speeds.
It can create a new type of threat intelligence that
empowers security analysts with near real-time
incident analysis to identify and help contain
threats before they spread.
Cyber leaders are right to be concerned about
how bad actors may use Gen AI—but they should
be optimistic that, with the right approach
and governance in place, Gen AI can help an This paper will explore how Gen AI can help and what the
organization harden its cyber posture, overcome cyber considerations may be. It’s important to remember
challenges in talent, and build new road maps for
that an organization’s success with using Gen AI to
threat detection and response.
drive better outcomes rests on its ability to imagine a
collaborative intelligence between humans and machines
and to ask the right questions. Trusting that Gen AI can
make a true impact in your organization means first
understanding its power and potential.
So let’s get started.
© 2024. For information, contact Deloitte Asia Pacific Limited.
5
Gen AI’s immense Here’s what we mean:
Predict: Analyze asset inventories, security logs, threat intelligence,
value for cybersecurity etc., to help predict risk scores and recommend preventive measures.
Interpret: Summarize and process large volumes of textual data
into coherent, actionable summaries; alert reception; and parsing.
Machine learning has long been used to detect
Generate logical analysis (inference, deduction, and/or explanation)
Gen AI is a force multiplier of cyber vulnerabilities and perform threat monitoring
given context or knowledge base.
at scale, but it takes a high degree of technical
value because it can do human-
proficiency and investment to train an organization’s
like work at hyper speeds that model to understand patterns and detect anomalies Simulate: Extract information from a knowledge base to help generate
in the data. Rules-based AI, in other words, can find responses to natural language questions; create test cases and
no human can match.
only known attacks and work in specific use cases. sample scenarios.
But with Gen AI and large language models (LLMs),
Automate: Create incident response activities, including triaging
the game changes. Gen AI uses foundational neural
alerts, correlating events, and guiding incident handlers with
network models that are powered by and trained
response playbooks.
on vast amounts of data, working across data silos
and acting as a bridge between data sets. This can
give analysts a more natural method for identifying,
Detect: Identify connections between alert data and threat intelligence
synthesizing, and summarizing insights.
reports to help determine the impact on infrastructure. Update specific
responses that can guide security analysts in remediation and recovery
activities.
Interact: Analyze governing documents, laws and regulations, data,
and standards to quickly inform actions. Deliver personalized and
targeted threat and crisis response trainings to employees based
on roles, responsibilities, and job requirements.
Create: Generate content by converting it to a new format or style and
for a variety of modalities based on a set of input data, examples, or
specific themes or topics.
© 2024. For information, contact Deloitte Asia Pacific Limited.
6
Cyber risk management Threat detection Vulnerability management
Looking for Others
and compliance and response and security testing
specifics? Gen AI
Risk scoring and prioritization Actionable and precise Controls testing and automation Role mining
can help transform
Analyze asset inventories, security threat intelligence Create test cases/sample scenarios; Use Gen AI to recommend role
cybersecurity logs, and threat intelligence to Generate summarized reports/ expected outcomes; develop assignments based on user attributes
predict risk scores and recommend executive briefings for active threats supporting documentation to ensure adaptive access control
activities like these.
preventative measures from historic trends or publicly
available data Secure code generation Data classification and monitoring
Third Party Risk Management Develop application code and Classify and monitor unstructured
Analyze data in vendor submitted Threat correlation and detection relevant supplementary test cases text-based data, which enables better
and external documentation to Identify correlation between alert in line with the latest security protection against exfiltration
evaluate the security posture of data and threat intelligence reports considerations (backward integration
third-party providers to determine impact of secure coding guidelines) Training and awareness
on infrastructure Deliver personalized and targeted
Automated policy Enhanced vulnerability scanning threat/crisis response trainings
review & orchestration Security incident response Correlate vulnerability data (scan to employees based on roles,
Map current policies, standards and Automate incident response activities, data, external information and responsibilities, and job requirements
procedures against standard industry including triaging alerts, correlating remediation plans) to prioritize
and regulatory frameworks to meet events, and guiding incident handlers action plans
compliance requirements with response playbooks
Enhanced systems
Cybersecurity Enhanced recovery design/configuration
maturity assessments and remediation Augment system/security architecture
Self-assess the organization’s cyber Create specific responses that design by drafting preliminary technical
risk maturity; identify gaps in cyber can guide security analysts in specification and/or recommending
strategy and generate relevant remediation and recovery activities optimal configuration
improvement recommendations
Gen AI-enabled phishing detection
Use Gen AI to detect threats and/or
phishing attempts created by LLMs
Note: This is not an exhaustive list. Feasibility of some of these use cases must be evaluated based on data availability and other constraints.
© 2024. For information, contact Deloitte Asia Pacific Limited.
Draft requirements
fo
noitacilppA
stfieneB
slliks
ecrofkroW
IA
evitareneG
7
The power of pairing AI and Gen AI
An organization using Indicators of Co-pilot for
Response process
AI to detect and combat Compromise (IOCs) incident response
automation
cyberthreats is already ahead signature generation and SOC automations
of the game. Layering on Gen
AI can add further complexity
and power to its models.
Simplify requirements-gathering phase Classify IoCs (e.g., information about a Detect hidden patterns, harden Automate cyber defense strategies,
by developing prototypes of complex specific security breach that notifies defenses, and respond to incidents industry notifications, future mitigation
While a traditional AI model
applications. Provide more intuitive security teams if an attack has taken faster with triage signals and predictive strategies, etc., as part of the response
can detect threats, adding Gen engagement between the analyst and the place) using distinct signature guidance. Quickly synthesize data from process.
AI could allow it to summarize customer to better inform development. generation. multiple sources to provide actionable
the incident, prepare insights.
documentation, and create a
response action plan.
Gen AI can help an Reduces the risks of Improves visibility of cyber attacks and Introduces robust and reliable approach Improves organizational compliance
organization move beyond miscommunication (i.e., the analyst and streamlines the security team’s response to incident response, threat hunting, and with incident response plans and
customer are able to align on the with expedited identification and triage. security reporting contingency plans through automation.
rules-based analysis and
prototype before proceeding to the In doing so, improves efficacy and
expand into outputs of higher
build phase). streamlines execution.
complexity and capabilities.
• Customer engagement • Information gathering • SOC • Cybersecurity SOC
(e.g., review cycles) • Mission expertise/security • Threat detection and response • Threat response
• Storyboarding clearances
Data science, AI/ML engineering, deep learning, UI/UX design, high performance computing, prompt engineering,
Core AI skills
digital operations & delivery, multidisciplinary collaboration, computer vision, NLP
© 2024. For information, contact Deloitte Asia Pacific Limited.
88
The cyberthreat
considerations for Gen AI
To understand Gen AI’s power, an organization should be fully aware
of the considerations inherent to the technologies.
As we’ve said, Gen AI opens new opportunities for
Growing concerns and global action
organizations to prepare for and defend against
cyberattacks. But as with any new technology, Gen AI In the fall of 2023, the Biden administration
comes with risks and the potential to amplify existing announced an executive order on the safe
ones as well. and trustworthy use of AI that will likely create
downstream effects for new regulations and
Earlier AI systems were traceable, and it was standards, further complicating the regulatory
possible to understand certain outputs via its atmosphere.
data. But Gen AI is a different game with multiple
Meanwhile, the European Union (EU) is moving
parameters that can make it more challenging to
toward stringent rules around AI, even moving
trace output. Gen AI is also trained on much larger
to ban its use in some cases. As the use of
data sets than traditional AI, which can make it more
Gen AI becomes more prevalent, we expect
difficult to know where and how the data may have
governments to take more action to mitigate
been altered or where quality concerns may exist.
potential risks.
Constantly evolving risk profiles demand a
new perspective.
© 2024. For information, contact Deloitte Asia Pacific Limited.
9
There’s a lot to consider.
We’ve broken out some
current cyber risks for Gen AI:
Data breach Reputational risk
Don’t just take our word for it.
Sharing sensitive data with external Gen Bad actors can leverage Gen AI tools to
The Open Worldwide Application Security Project
AI vendors for model training or through widely and rapidly spread misinformation (OWASP) has published its Top 10 risks for large
prompts may lead to leakage of confidential and deepfakes, which can adversely influence language model applications, including trained
and/or personal information. Adversarial public opinion, trust, and/or security. data poisoning and supply chain vulnerabilities.3
attacks can also be used to deceive the ML
model by changing input data.
Unsecured integration Regulatory risk
Improper integration of Gen AI tools with Organizations using Gen AI may need to meet
other organizational systems may lead to new compliance requirements as growing
potential vulnerabilities (e.g., unsecured data concerns influence new laws, regulations,
channels) and back doors. and guidelines, such as National Institute of
Standards and Technology’s (NIST) proposed
AI Risk Management Framework1 and new EU
regulations for General Purpose AI Systems2.
(Read more)
© 2024. For information, contact Deloitte Asia Pacific Limited.
10
A framework for risks and limitations associated with Gen AI
While emerging tech has inherent risks, Deloitte’s Technology Trust Ethics (TTE) framework can be leveraged to build, deploy, and commercialize AI applications
Deloitte’s TTE framework Foundational capabilities Gen AI-specific capabilities
AI strategy Management of hallucinations and
Co n &t r eo cl ns
a n re vo G
RO B U S T A N D R E L I A B L E
ELBA
TNUA
OC A
d
Co Ac
a
CP n nc Apr s sue
t
i
w
d r as a ti
eb
c e t
r
S
t
l
R
en aeU a
e
A
b
sts b
F
le ol
lE
er e
v
F
A
a
r i
bN
e
l
n
OeD
d
w
l nU
yS
es re
E
sr
hPC
ir
poU
t e
cR
t i o
E
n
R
H
e
u m a
ns
eT
FTIn
k
vu l n e r rr Ea
ei
b
lel
Co
Lu
a
ml
mon/c
i
B
n
ms Sh
o cial
I
G
og
od
St n
N
e&
E yo
til Oi
baw
n
iatsut
SE
l
P
hd
dA eu o
s
t uo
co
on fo
Smou
su
i Eg
r
c
c
R
k
gy
a
ns
i d
dt
A
C
i
o
e
un
o
fi
l
ad
V
e
n
n t i al D i
U
s c
n
r
b
e
P
t
i
i
a
R
o
s
en C
II
da
n
olV
c
n
A
ls
u
e
T
s E
n
iE
J
q v
s eu u
u
As a il I t
t
n ci aVA
c
fi t
b
i
u
e
e a
s
l
esr d ib
L
l bp
s
i Ae t
il
I
r
b
ea
T
e
l R
b t
e
Aa l Pe b
M
l
I
e
D
T
N
R
A
A RN
I
S
A
P
F
A R E N T
y
A cN
i
D
l
E X ELP L BA AIN
o P & e c n a i l p m o C y r
o t a l
uR ge A
A•
•
•
••
I
ID
Ra
r n D a
r
M a
C
ii n
ne ir
tnn ne
e
osee
d do
eg i
k
dd g
w d
dsfi
s
u
enu
csu
i
e
n
k g
im
ps
il i r
n
hl
ta
(
me
n
ta
e
ae
eo
ira
mt
n
tt
q
s.
na
o ra p
yg
in o
s
au
o
dn
r n cl
.aa r
s
et,ey i
o
ed
ld
y
t
erg
sn
b
o
m
a
nne l
e
sci
ia
i
ra
n
gm
mm
t
ta
mm o
een
rds
tg
yo
mm
np
ped
ee)e
ae /ltn
ll
ss
b
nsn
r
ep
e
ce
mc tdt —c
a
tt
dmlm
y
isa
i
o
nss
aaf
i
bp e
a
aer
ee
n
gnna
s
te dnn
e
n
t
dc
i
osm
or
tt
e
ra t
o
e
foa
mn
sa
n—e
sn
At
ln
eiw
sd s
o
Io
rr
c
dcoo
t
ee
np o
o u
egvrv
r o
n
k
lee
i
u
ra
e
f
vtpr
l
i
dw
raAa
ata od
tr
y
lI r
i
i
e
lc
or dc
seh
v
eo
n
a
eo
fi
sn
tsn
o
ffl
is
tv
org
a
er
i
A no
n
n
cA
I —l
d
tgI
s -
i
s
v—s pept rr
nd e
oa
e
ect
vs
sie
ifi
s
dg
g i c
ey
n A
•m
•
•
A f
•
ot
cA
i d
cI G
m s R r r
F
e
ai
rntd
ce
es
ot
o
nxe
os
e
r
fre
e g oq
Gi
p
ps
cg
do
eun
in
n
c u u
e
uyb uu
l
sr e
ch
a
at
lAf
i
s
rm
sl a r ut
ni
n n
na sa
io
s
f uI
e gt
o
s
oty
an at to
br
m ar ho
oA
n
i
tt
tai e
rm
a ny
s
t
or
i
i
c
e ty
o
o
d
Ibn
p
m y
d
vr
ii
n n
an
nm
a
l
rd
e i
ico a ovs t
o
tl sa
ot
ws mn ev
ip am
l
v(
di
ta
ome de d
ll
ho
iie
y
ta
d
fr da. as
i
ip g
rn
m sn
l o
i
Gc
a
cn
e
ina. li
saa
nri, g
t
e
e
a
g
kna
oeepg
w sh
nn
n
s
n
se
age dt ao
t
c
s
As,
c
ei
t
m
e ar
n
uu
ct
el
Ik mn—
i
,r
eg
rb
n
i
xf
i
is
d
i
s
bo ti
ena ri
pq e
s
sn
ur e p
sgp
i
uc n
l
b
lf
v
t
rp
o a
ie
to
i ie mi lt ot he
fr
e
ir
o p
uo
n
a
nwm
p
,
iu a
tpp
ts
nl a
s
rs
aa esr
t
ot
,e
bgh
kt i
v od
ca
n
i
ii o f
eo
i
ao
alt
o
t
-ll
len
th
pri c imv rn
ei t
u
e
ai
sgt
nng
cyo
m
m
bh o
hse, g
d
ir v
o
w
n
l
eo
e
iae
u
tin
lu
n
cr
i
r
et
n ag
dca
sla eh
t
,
n
i
oc ne
)
• Threat monitoring and detection—monitor
Deloitte’s TTE framework may be leveraged for
for specific technology threats (malicious and • Derive viable methods of accountability, trust,
foundational and Gen AI specific capabilities
environmental) that are targeted at AI models and ethics when using Gen AI and underlying
and underlying technology technology
© 2024. For information, contact Deloitte Asia Pacific Limited.
1111
How to prepare
Example checklist • Establish secure channels and
mechanisms to transfer data between
• Update policies and controls for new enterprise and cloud-hosted Gen AI
types of bias, legal, regulatory, privacy, tools.
intellectual property, and data risks of
• Review third-party controls and
With forethought and deliberation, cyber leaders can ready Gen AI.
establish contractual obligations to help
their organizations for the capabilities and risks of Gen AI. • Identify new compliance requirements protect sensitive data shared with Gen
and impacts on compliance activities AI vendors.
with existing laws and regulations.
• Monitor for novel attacks (e.g., prompt
Regardless of an organization’s particular The key is to evolve those constructs to answer • Closely evaluate use cases for Gen AI injection) and help ensure appropriate
needs, defining key outcomes and instituting the nuanced risks and threats that may be for the organization to help ensure usage of Gen AI tools to prevent
guardrails can help leaders improve risk targeted toward Gen AI or AI systems. impactful outcomes and overcome any vulnerabilities.
preparedness, promote resilience, and unlock An organization’s specific risks may depend resistance to adoption.
• Define boundaries of where and when
new business opportunities around Gen AI. on what adoption model it chooses, such as
• Implement appropriate contractual Gen AI technologies can be used within
software-as-a-service or private LLMs.
obligations for Gen AI vendors around the organization.
Leaders should recognize that Gen AI requires
security and usage of any information
new approaches to technology, training, and When choosing adoption strategies, an • Integrate secure-by-design
shared, and monitor the data sharing
processes—but that said, this is an evolution of organization should recognize the power and principles during integration of
channels used by them.
existing risks. Gen AI may not require net-new necessity of end-to-end transformation rather Gen AI applications into enterprise
road maps and trainings. An organization’s than automating one or two activities. • Implement privacy and data protection architecture.
standards and controls when
risk management and cyber constructs may • Help protect the organization’s brand
developing and training models for Gen
still work. by monitoring for misinformation,
AI tools.
and define communication strategies
• Enhance existing code review processes to counteract and decrease impact of
to help test code created by Gen AI for disinformation campaigns.
back doors and vulnerabilities.
• Take action immediately on the risks
• Implement access controls and monitor from adversarial and malicious Gen AI
use of Gen AI tools to help limit risks usage.
from inadvertent or inappropriate use.
© 2024. For information, contact Deloitte Asia Pacific Limited.
12
HOW TO PREPARE
Above all, remember: A road map for Gen AI adoption
Adoption of Gen AI by organizations will depend on six factors
should include close, constant collaboration for risk
stakeholders, including cyber leaders, chief resource
Cost and efficiency: Ability to assess whether benefits of using Gen AI-
officers, an organization’s legal team, and more, to 1
based systems outweigh the associated expenses, as handling and storing
help understand and anticipate the risks. (And don’t large datasets can result in increased expenses related to infrastructure and
forget to include testing and monitoring.) computational resources.
Knowledge and process-based work: High degree of knowledge and
2
process-based work vs. only field and physical work.
High cloud adoption: Medium-to-high level of cloud adoption, given
3
infrastructure requirements.
Low regulatory and privacy burden: Functions or industries with high
4
regulatory scrutiny, data privacy concerns, or ethics bias.
Specialized talent: Strong talent with technical knowledge and new capabilities,
5
and ability to help transform workforce to adapt quickly.
Intellectual property and licensing and usage agreements: Ability to assess
6
licensing/usage agreements and restrictions, establish and monitor related
compliance requirements, and negotiate customized agreements
with relevant vendors.
© 2024. For information, contact Deloitte Asia Pacific Limited.
13
Cyberattacks won’t stop.
The good news is, Gen AI
progress won’t either.
Gen AI could accelerate both cyberattacks and
threat response capabilities. Organizations
need to recognize both sides of that equation.
The question is, how can cyber leaders steer their teams and
organizations through the disruption while harnessing the capabilities
of what is, to date, the most powerful artificial intelligence ever created?
Many organizations are so busy fighting today’s battle that it’s hard
to conceive of creating a new Gen AI ecosystem that may require
development, operations, new talent, and evolved processes.
For any cyber leader, it’s important to start the journey toward Gen AI
with questions specific to the organization. Gen AI is an unprecedented
opportunity for a new kind of collaborative intelligence, one that can
provide increased security and next-level collaboration. So where does
a leader start? With our deep bench of cyber experience, alliance relationships,
and pragmatic perspective on the future, Deloitte can help
With one question: “What if?” From there, it’s all a new frontier.
organizations address their most pressing cybersecurity
challenges—now, and for whatever is around the bend.
Reach out to learn more.
© 2024. For information, contact Deloitte Asia Pacific Limited.
14
Endnotes
Get started
1. AI Risk Management Framework | NIST
2. https://www.europarl.europa.eu/news/en/press-room/20231206IPR15699/
artificial-intelligence-act-deal-on-comprehensive-rules-for-trustworthy-ai Authors
David Caswell, Sabthagiri Saravanan Chandramohan, Deborshi Dutt, Chris Knackstedt,
3. OWASP Top 10 for LLM Applications Version 1.1, October 16, 2023
Vikram Reddy Kunchala, David Mapgaonkar, Mike Morris, Abdul Rahman,
Kate Fusillo Schmidt, Niels van de Vorle
Contributors
Sanmitra Bhattacharya, Edward Bowen, Ben Bressler, Suzanne Denton,
Eric Dull, Lena La, Sajin Mathew, Nirmala Pudota, Stephanie Salih, Colin Soutar
Contacts – Asia Pacific Cyber leaders
Ian Blatchford Steven Feng Yuichiro Kirihara
Asia Pacific/Australia China Japan
[email protected] [email protected] [email protected]
Youngsoo Seo Anu Nayar Tarun Kaura
Korea New Zealand South Asia
[email protected] [email protected] [email protected]
Tse Gan Thio Max Y. Lin
Southeast Asia Taiwan
[email protected] [email protected]
© 2024. For information, contact Deloitte Asia Pacific Limited.
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 are separate and independent legal
entities, provide services from more than 100 cities across the region, including Auckland, Bangkok, Beijing,
Hanoi, Hong Kong, Jakarta, Kuala Lumpur, Manila, Melbourne, Osaka, Seoul, Shanghai, Singapore, Sydney,
Taipei and Tokyo.
This communication contains general information only, and none of Deloitte Touche Tohmatsu Limited
(“DTTL”), its global network of member firms or their related entities (collectively, the “Deloitte organization”)
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. DTTL and each of its
member firms, and their related entities, are legally separate and independent entities.
© 2024. For information, contact Deloitte Asia Pacific Limited. |
323 | deloitte | DI_Intelligent-drug-supply-chain.pdf | Intelligent drug supply chain
Creating value from AI
About the Deloitte Centre for Health Solutions
The Deloitte Centre for Health Solutions (CfHS) is the research arm of Deloitte’s Life Sciences and Health
Care practices. We combine creative thinking, robust research and our industry experience to develop
evidence-based perspectives on some of the biggest and most challenging issues to help our clients
transform themselves and, importantly, benefit the patient. At a pivotal and challenging time for the
industry, we use our research to encourage collaboration across all stakeholders, from pharmaceuticals
and medical innovation, health care management and reform, to the patient and health care consumer.
Connect
To learn more about the CfHS and our research, please visit
www.deloitte.co.uk/centreforhealthsolutions
Subscribe
To receive upcoming thought leadership publications, events and blogs from the UK Centre,
please visit https://www.deloitte.co.uk/aem/centre-for-health-solutions.cfm
To subscribe to our blog, please visit https://blogs.deloitte.co.uk/health/
Life sciences companies continue to respond to a changing global landscape and strive to pursue
innovative solutions to address today’s challenges. Deloitte understands the complexity of these
challenges and works with clients worldwide to drive progress and bring discoveries to life.
Contents
The rationale for transforming the biopharma supply chain 2
How AI can augment supply chain transformation 6
AI’s role in helping supply chains respond, recover and thrive after
COVID-19 16
A roadmap for implementing an intelligent supply chain 21
Endnotes 32
Intelligent drug supply chain
The rationale for transforming
the biopharma supply chain
The Intelligent biopharma series explores the ways artifi cial intelligence
(AI) can impact the biopharma value chain. The fi rst two reports, Intelligent
drug discovery1 and Intelligent clinical trials,2 highlight the potential of AI to
accelerate the development of new drugs. This report explores the poten-
tial for AI technologies to improve the value of the biopharma supply chain
and manage risks more eff ectively. Evidence shows that the need for dig-
ital transformation of the supply chain has never been more pressing.
THE BIOPHARMA SUPPLY chain involves a Protecting biopharma supply chains is a priority
complex set of steps that are required to pro- not only for companies but also for all governments,
duce a drug, from sourcing and supply of given the importance of ensuring access to the
materials, through manufacturing and distribution, lifesaving and life-enhancing products that are
to delivery to the consumer. This forms a golden vital for the health and well-being of their popu-
thread between the discovery of new therapies and lations. The COVID-19 pandemic has highlighted
patients receiving them (fi gure 1).3 the importance of biopharma supply chains in
meeting the demand for leading-edge products.4
FIGURE 1
The different steps in the biopharma supply chain
Post-market
Research & Clinical Manufacturing Launch &
surveillance &
discovery development & supply chain commercial
patient support
SOURCING MANUFACTURING DISTRIBUTION DELIVERY PATIENTS
APIs and other Processing, testing Wholesale distributors, Integrated delivery Providing effective
materials and packaging labelling and networks of retail and safe treatments
serialisation pharmacies, hospitals
and clinics
PEOPLE, SKILLS AND INFORMATION SYSTEMS
REGULATORY RISK AND COMPLIANCE
TRANSPORT AND LOGISTICS
Source: Deloitte analysis.
Deloitte Insights | deloitte.com/insights
2
Creating value from AI
Biopharma supply chains create inherent resilience play in its digital transformation (Part 2). Given
risks for corporations and governments alike. the unprecedented challenges to the health
Supply chains must also meet the expectations of a care ecosystem resulting from the COVID-19
complex range of stakeholders, comprising multiple pandemic, the report also considers the role
payers, health care providers and patients with that AI can play in helping the supply chain to
complex and varied needs, both within and respond, recover and thrive (Part 3). Finally, the
across diff erent countries. Consequently, report provides a strategic roadmap for imple-
intelligent and insightful monitoring and man- menting an AI-enabled supply chain (Part 4).
agement of the supply chain is an imperative.
The risk landscape for biopharma supply
This report examines the rationale for transforming chains comprises internal, external
the supply chain (Part 1) and the role that AI can and macro risk factors (fi gure 2).
FIGURE 2
The complexity and risks affecting the globally distributed biopharma supply chain
Macro risk External supply chain Internal risk
Negative impact to Risks in upstream and Potential risks from
entire industrial chain downstream supply internal operation
due to changes in chain processes
macro environment
Geopolitical Environmental
change protection policies
Compliance Insufficient/excess
Exchange capacity Trade
rate Operational incidents barrier
fluctuation change
Production safety
Quality Technical
risks Inaccurate demand forecast bottlenecks
Product development delay
Trade secret High transportation cost
leaks Unstable
Long delivery cycle IT system
Labour High
law defect Equipment malfunction Large-scale
strikes
rate
Cold chain logistics Competition
Bankruptcy/ Infringement
Waste financial risks of IP rights Epidemic COVID-19
management outbreak
Cybersecurity
Terrorism/large Extreme weather
scale civil strife conditions
Source: Deloitte analysis.
Deloitte Insights | deloitte.com/insights
3
Intelligent drug supply chain
The complexity of
Given the above complexity and risks, governments
have established an evolving and complex frame- biologics manufacturing
work of local, regional and international regulatory and supply chains
bodies. There are also global bodies, such as the
World Health Organisation (WHO) and Interna- In comparison to ‘traditional’ small molecules,
tional Council for Harmonisation of Technical biologics have more complex supply chains (fi gure
Requirements for Pharmaceuticals for Human Use, 3). A 2019 survey of 151 experts to understand
aimed at improving regulatory collaboration.5 more about manufacturing practices and trends
identifi ed that the biggest challenges of biologics
FIGURE 3
The different complexities affecting the three main biologics manufacturing processes
RECOMBINANT PRODUCTS – VIRUS, BACTERIA, CULTURED CELLS
Proteins (recombinant
antibodies, cytotokines,
enzymes, etc.)
Antisense RNAs
In vivo gene therapies Initial batch Harvest and Inactivation Formulation Batch quality Transport/
Vaccines (DNA, RNA and thawing purification of and assembly and filling for control, distribution to
proteins for prevention and cell/ antigens or (vaccines) cryopreservation validation and hospital and
and therapy) microorganism active packaging pharmacies
growth molecules (cold chain)
HUMAN BLOOD PRODUCTS – BLOOD DONORS
Whole blood,
blood derivatives and
blood components
Tissue Transport to Tissue processing Transport to
collection processing • Testing hospital for
facilities or • Centrifugation patient treatment
storage unit • Component separation or to storage unit
(cold chain) (cold chain)
GENE AND CELL THERAPIES – PATIENTS OR DONORS
Gene and cell therapies
Autologous, including
CAR-T therapies, and
allogenic, including
stem cell therapies
Apheresis or Transport to Personalised Cell Batch quality Transport to
other tissue manufacturing bioprocessing concentration, control and hospital (cold
collection facilities or • Cells separation formulation validation chain) for
storage unit • Gene editing and filling for therapy
(cold chain) • Reprogramming cryopreservation application or
• Cell expansion to storage unit
BIOLOGIC PRODUCTION TAKES FROM SIX MONTHS TO THREE YEARS
70% OF THE TIME IS REQUIRED FOR QUALITY CONTROL
COLD CHAIN TRANSPORT AND STEPS INVOLVING CELL CULTURING
ARE KEY TO MAINTAINING YIELD AND QUALITY
Source: Deloitte analysis.
Deloitte Insights | deloitte.com/insights
4
Creating value from AI
manufacturing are process robustness (59 per requires a more agile supply chain structure. This
cent of respondents), process reproducibility (56 is particularly the case for cell therapies (known as
per cent), product yield optimisation (46 per cent) ex vivo); for example, chimeric antigen receptor
and product characterisation (42 per cent).6 T-cell (CAR-T) therapies. By the end of 2019, the
FDA had approved two CAR-T therapies; however,
Deloitte research demonstrates a continued there were some 600 clinical trials involving
growth in biologics and estimates an equal split CAR-T therapies in the biopharma pipeline.10
with small molecules in worldwide sales by 2024.7
However, large-scale production of biologics is at
The need for a new supply
present seen as one of the main challenges of the
biopharma industry, mostly due to the inherent chain model is driving the
variability of biological systems and the instability digital transformation of most
of finished products. Ascertaining the yields and
industries, supply chains
quality of these types of drugs is crucial in ensuring
that they are reproduced effectively and maintained In the past few years, manufacturing companies
until they reach patients, with regulatory bodies ap- across all industries have initiated digital trans-
plying strict inspection and reporting requirements. formation of the different steps in the supply
These need additional analytical methods that chain. Big tech giants such as Amazon, Apple
measure specific physical and biochemical prop- Inc. and Google have led the way in the early
erties to ensure these therapies remain safe and adoption of end-to-end digital supply chains. The
have not lost their activity during manufacturing.8 opportunity was created by the availability of
large amounts of reliable and relatively untapped
NEXT-GEN THERAPIES BRING data at the same time as technological break-
A NEW LEVEL OF COMPLEXITY throughs were developing, such as advanced
TO SUPPLY CHAINS analytics and AI, blockchain, digital twins and
Responses to a 2019 survey indicate that the most the Internet of Things (IoT), intelligent auto-
commercially important biopharma therapeutic mation and virtual and augmented reality.11
products currently available are monoclonal
antibodies (80 per cent), followed by vaccines (51 While many life sciences companies have been
per cent) and other recombinant proteins (36 per exploring the opportunities that digital tech-
cent). Fewer respondents mentioned cell therapies nologies offer, many are yet to make consistent,
(18 per cent), gene therapies (18 per cent) and sustained and bold moves to take advantage of
RNA-based therapies (9 per cent). When the same the new capabilities.12 Companies able to make
question was asked concerning the next five to ten the transition could rocket ahead of competitors
years, the answers were substantially different. and fend off intruders from outside the industry
Respondents put gene therapies in first place (66 trying to enter biopharma’s orbit. In 2019, the
per cent), followed by monoclonal antibodies Deloitte US Center for Health Solutions led a
(58 per cent) and cell therapies (52 per cent).9 four-day online crowdsourcing simulation with
biopharma leaders and found that companies
The high sensitivity and more precise targeting are getting closer to incorporating digital tech-
of biologics requires a direct connection between nologies more broadly in everything, from R&D,
pharma companies, the health care system and to supply chain, to patient engagement.13
even individual patients; consequently, this
5
Intelligent drug supply chain
How AI can augment supply
chain transformation
A huge amount of internal and external data is generated routinely
across the biopharma supply chain, but historically these data have
been underutilised. Simply capturing data fails to provide actionable
insights. Using AI technologies to process these data will be critical to
orchestrating operational efficiency and, ultimately, to creating a cost-ef-
fective, near autonomous and thriving biopharma supply chain.
AI TECHNOLOGIES ARE poised to transform WHAT IS AI?
supply chain and manufacturing through
AI refers to any computer programme or
real-time data processing and decision mak-
system that does something we would think
ing to make supply chains truly data-driven, of as intelligent in humans. AI technologies
reducing human subjectivity and bias. AI tools have extract concepts and relationships from
the potential to unlock commercial, regulatory and data and learn independently from data
operational data to find non-linear and complex patterns, augmenting what humans can
relationships that would otherwise be missed and do. These technologies include computer
to deliver powerful strategic insights. AI algorithms vision, deep learning (DL), machine
learning (ML), natural language processing
have the potential to deliver significant improve-
(NLP), speech, supervised learning and
ment in productivity and gross margins and
unsupervised learning.14
contribute to the sustainability of the biopharma
industry. In particular, AI algorithms can improve
end-to-end visibility, leading to more efficient Deloitte has identified five critical areas and
demand forecasting, inventory management, logis- processes of the biopharma supply chain where AI
tics optimisation, procurement, supply chain is likely to have the highest impact (figure 4). This
planning and workforce planning. is based on our research, including comprehensive
literature reviews, interviews and workshops
AI algorithms have the with colleagues working on supply chain proj-
ects, analysis of the relevant findings from the
potential to deliver
US Deloitte Center for Health Solutions’ online
significant improvement crowdsourcing simulation with biopharma leaders,
and discussions with digital technology companies.
in productivity and gross
margins and contribute to
the sustainability of the
biopharma industry.
6
Creating value from AI
FIGURE 4
Applications of AI-powered technologies in the biopharma supply chain
END-TO-END VISIBILITY
Point-to-point visibility across the whole supply chain will enable companies to become
more efficient by rapidly responding to and mitigating disruptions. AI-augmented
control towers provide advanced decision-making systems, by efficiently collecting
and managing data in real-time and generating actionable insights.
DEMAND FORECASTING, LOGISTICS AND INVENTORY MANAGEMENT
AI tools can mine and analyse data from multiple sources to detect patterns and
potential anomalies to generate accurate demand forecasts and help companies
efficiently manage their inventory levels.
INTELLIGENT AUTOMATION ENABLING INDUSTRY 4.0 AND THE INTERNET OF THINGS
Adoption of AI tools, such as ML, NLP and computer vision, into an Industry 4.0 and
IoT platform will be the key to minimising human error and leveraging operational data
to generate strategic insights and improve productivity and accuracy of processes.
OPTIMISING PREDICTIVE MAINTENANCE
AI technologies can find patterns and interdependencies between variables that
would otherwise be missed by traditional methods. Leveraging AI through real-time
performance monitoring will optimise maintenance, minimise downtime and,
ultimately, maximise productivity.
PROTECTING THE INTEGRITY OF THE SUPPLY CHAIN
Combining AI with other advanced technologies, such as blockchain, can create a
system that is immutable, transparent, secure, and shielded from counterfeit and
substandard drugs.
Source: Deloitte analysis.
Deloitte Insights | deloitte.com/insights
interconnected supply chain will allow data to
End-to-end visibility:
be safely extracted in real-time using AI tools
the holy grail of supply to generate actionable insights, consequently
chain management improving decision-making. This can help
companies mitigate disruptions and become
In biopharma’s hyper-connected globally complex more agile, effi cient and responsive (fi gure 5).
supply chain, companies need to be able to
respond rapidly to any supply chain event that The concept of end-to-end visibility can be realised
impacts outcomes. End-to-end visibility is the through supply chain control towers, which provide
foundation for quickly making the right decisions a holistic view across all supply chain functions.
to mitigate risk and deliver required outcomes. Control towers function as centralised hubs that
Supply chain visibility means having access to collect information from disparate systems to
data relating to every transaction and demand be used for monitoring, auditing and generating
trigger, across every step and tier of the supply insights.16 For control towers to be successful in
chain and all the logistics movements in between. supply chain management, companies need to
incorporate AI capabilities, such as ML, to help
Digital technologies that improve visibility across orchestrate operations and, ultimately, have
the supply chain, such as AI and blockchain, can a near autonomous and self-learning supply
create a dynamic, interoperable system where chain. In October 2019, IBM launched Sterling
transactions are transparent and traceable.15 Supply Chain Insights, an AI-enabled control
Disruptions in specifi c parts of the supply chain tower that allows for comprehensive, real-time
can impact subsequent steps, creating a cascade visibility of the supply chain by providing rapid
of ineffi ciencies. End-to-end visibility across the integration and interpretation of structured and
unstructured data at scale (case study 1).17,18
7
Intelligent drug supply chain
FIGURE 5
End-to-end visibility of the supply chain
Enhanced Real-time traceability Optimised inventory
supplier connectivity and lower time-to-value levels
Suppliers Manufacturing facilities Warehouses/ Pharmacies/hospitals/ Patients
distribution centres clinics
End-to-end visibility
Control tower
Actionable insights and powerful decision-making capabilities
Source: Deloitte analysis.
Deloitte Insights | deloitte.com/insights
CASE STUDY 1 – IBM STERLING WANTS TO UNTANGLE COMPLEXITY WITH SMARTER SUPPLY CHAINS
Tech giant IBM created Sterling Supply Chain Insights with Watson to help companies achieve end-
to-end visibility with a control tower that leverages AI to connect data across siloed systems. Watson
AI correlates data from both internal and external sources, enabling analysis of 80 per cent of
unstructured data, including digital media and weather reports. These capabilities allow companies
to better understand and assess how these data impact their entire supply chain.
When disruptions occur, Sterling Supply Chain Insights aids with faster decision-making to align issue
resolution with business objectives, optimising management while responding to unplanned events.
Sterling Supply Chain Insights is a critical part of the IBM Sterling Supply Chain Suite, an integrated
suite that enables companies to connect crucial data and supply chain processes, while leveraging
AI, blockchain and IoT. Sterling Supply Chain Insights powers the IBM Sterling Supply Chain Suite’s
Intelligence Services and Control Tower capabilities.19
Lenovo, a global technology and manufacturing company, wanted to establish greater visibility
across its complex supply chain systems and data sources to minimise disruptions and improve
customer order management. Lenovo implemented IBM Watson Supply Chain Insights to optimise
the orchestration and gain end-to-end visibility of its supply chain. With this tool, Lenovo adopted
an AI-powered approach to risk management, reducing its average response time to supply chain
disruptions from days to minutes (up to 90 per cent faster than before) and gaining opportunities
to reduce costs and drive revenues. Ultimately, these innovations could enable Lenovo to generate
more precise delivery estimates for its clients in real-time, adding value to its off ering.20
8
Creating value from AI
Demand forecasting,
and sales data, market intelligence and other
inventory management external data inputs that can affect inventory levels,
and logistics such as weather and epidemiological develop-
ments (including infectious diseases outbreaks).22
Demand forecasting plays a critical role in logistics The use of AI tools, specifically DL and ML, is
and supply chain management. Accurately adjusted particularly important in demand forecasting.
inventory levels are needed if the value of the Predictive analytics techniques can mine, analyse
supply chain is to be unlocked and, importantly, and interpret data aggregated from various
patients are to obtain timely, reliable access to sources to detect patterns and certain anomalies
their therapies.21 Forecasting uses a combination and generate more accurate demand forecasts
of decision variables, including historical shipment compared to traditional methods (case study 2).23
CASE STUDY 2 – MERCK KGaA USES ML TO OPTIMISE DEMAND FORECASTING
Merck KGaA, also known as the Merck Group, a large German multinational pharmaceutical,
chemical and life sciences company headquartered in Darmstadt, has embarked on a data-driven
supply chain operation to optimise demand forecasting.24
Merck is using Aera Technology, formerly FusionOps, an ML and cloud-based software solution that
enables a holistic and actionable view of a company’s supply network to increase efficiency across its
supply chain, including demand forecasting. Aera continuously combs through enterprise systems to
collect, harmonise and refine data, and to consequently provide real-time analytics and end-to-end
visibility of the company’s supply chain operation and performance.25
By using this technology at scale, Merck improved the forecast accuracy of 90 per cent of its
products. Aera’s AI algorithms use data collected from Merck’s enterprise resource planning
software to quickly and accurately forecast the demand for its products in terms of both quantity
and location.26
9
Intelligent drug supply chain
Having robust data on the significant macro, TRANSPORTATION LOGISTICS
external and internal risks affecting the biopharma CAN BE SUPPORTED BY AI
supply chain is critical for forecasting demand. A particular challenge of biologics production
For example, weather prediction technologies and supply is their large size and structure,
using algorithms are now fairly reliable when which is difficult to keep stable. Temperature
forecasting up to two weeks ahead.27 Still, in 2017 fluctuations and contamination can impact batch
and 2018, extreme weather conditions, such quality and yield, especially during transporta-
as droughts, floods and heatwaves, resulted in tion. Therefore, maintaining biologic APIs and
economic losses of $215 billion.28 In 2016, IBM finished products at a constant low temperature
announced the launch of Deep Thunder, a research is a key requirement, which has led to stricter
project that combines big data and AI algorithms regulations that mandate rigorous, end-to-end
and a global forecasting model built from The temperature control. Cold chain transportation
Weather Company’s vast wealth of data, to provide technology needs to be integrated with tracking
accurate predictions to weather-dependent software to ensure the effectiveness and safety of
business operations, including supply chains.29,30 therapeutics when they reach patients. By 2022,
an estimated 30 of the 50 top global biopharma
In addition, as seen in the COVID-19 pandemic products will require cold chain handling and
(Part 3), infectious diseases that spread directly specialised, temperature-controlled logistics.36
or indirectly from one individual to another can
cause serious disruption to supply chains.31 Today, In addition, research has shown that temperature
vast amounts of public health surveillance data are control is not sufficient to deliver efficacious
available from multiple sources, such as academic and safe biological medicines, as other physi-
institutions, climate databases, digital media, cochemical parameters, such as humidity, light
global transportation, genome databases, human and vibration, also affect the integrity of these
demographics, official public health organisations, compounds.37 To thrive, biopharma companies
livestock reports and social media.32,33 AI tools can can leverage advanced, intelligent technologies
make sense of these data sets by generating accu- that allow for real-time, end-to-end visibility. This
rate analyses and projections of potential infectious enables biopharma and logistics companies to
diseases outbreaks. For example, the severe acute track the state of the drugs and take proactive
respiratory syndrome (SARS) outbreak in 2003 led and timely interventions when any issue arises.
to the founding of BlueDot, a company that uses
advanced analytics and AI tools for automated,
As seen in the COVID-19
real-time infectious disease surveillance.34,35 Such
information can then be used by companies to pandemic, infectious
adjust their operations accordingly. This can be
diseases that spread
vital for biopharma to optimise the manufacturing
and distribution of specific drugs to affected areas. directly or indirectly
from one individual to
another can cause serious
disruption to supply chains.
10
Creating value from AI
Intelligent automation
which can then be unlocked and optimised by
enabling Industry 4.0 and using advanced analytics with AI capabilities.
the Internet of Things
Robotic process automation (RPA) is increasingly
Industry 4.0 aims to encourage the digitalisation deployed to reduce manual eff ort in repetitive
and automation of manufacturing processes and time-consuming tasks, minimising human
(fi gure 6). This is increasingly being adopted errors, and enabling operators to focus on high-
in the biopharma sector to help overcome the er-value and more motivating work. Today, the
multiple obstacles that the industry is facing, such convergence of AI, automation and customer data
as strict regulatory and production demands.38,39 has resulted in the emergence of a new class of
In addition, embracing Industry 4.0 will enable tools, known as intelligent process automation
companies to move towards Quality by Design (IPA). IPA combines RPA and machine learning
(QbD), a data- and risk-based approach for to deliver powerful tools that can mimic human
the development and manufacturing of drugs interaction and make advanced decisions based
that the US FDA and European Medicine on the outputs of those robotic inputs.44,45 This
Agency (EMA) are actively encouraging.40,41 will be key to minimising human intervention and
leveraging operational data to generate strategic
Digitalisation and automation of operations can insights and improve performance metrics.46,47
help biopharma companies establish cost-ef- Recent research by the Everest Group estimated
fective, reliable and robust processes that are that the 2019 intelligent automation market was
coordinated across the supply chain.42 This can worth $80 million–$85 million and is expected to
be optimised through the implementation of reach $450 million–$490 million by 2023, high-
an IoT platform, which interconnects ‘digital’ lighting the strong focus biopharma companies are
and ‘physical’ assets through the use of chips, placing on their digital and automation journeys.48
sensors and networks.43 IoT connections, there-
fore, generate a vast amount of monitoring data,
FIGURE 6
Benefits from intelligent automation, Industry 4.0 and the Internet of Things
Intelligent automation Interconnected and Computer vision
sensorised devices
Advanced analytics and decision-making through ML and NLP
IMPROVED OPERATIONAL METRICS AND PROCESS ACCURACY
BETTER ADHERENCE TO REGULATORY COMPLIANCE STANDARDS
Source: Deloitte analysis.
Deloitte Insights | deloitte.com/insights
11
Intelligent drug supply chain
The integration of IPA, IoT and Industry 4.0 in how one of the most disruptive companies in this
the manufacturing step of the biologic supply sector, Moderna Therapeutics, has digitalisation
chain is one the most promising approaches as a core part of its business strategy and is
towards reducing variability and ensuring safe and applying these types of technologies for large-
reliable large-scale production of drugs derived scale production of RNA-based therapeutics.
from living organisms. Case study 3 describes
CASE STUDY 3 – MODERNA THERAPEUTICS
Moderna Therapeutics is a clinical stage biotechnology company based in Cambridge, Massachusetts
(MA), in the United States. It is pioneering the development of messenger RNA (mRNA) therapeutics
and vaccines. Their mRNA medicines are designed to instruct the body’s cells to produce proteins
that have a therapeutic or preventive effect on a broad spectrum of diseases, including cancer and
cardiovascular, infectious and rare diseases.49
Moderna is a fully digital company that has digitisation as a core attribute of its business strategy.
Their landscape has been built on the following six key building blocks:
• AI – To enable key breakthroughs in analytics and predictive modelling that will help provide critical
insights into production and research data.50
• Cloud – Built on Amazon Web Services (AWS) Cloud to provide computational power, agility to
operate, cost-effectiveness and efficient organisation and processing of data.51
• Integration – To bring data and processes together in a consistent manner, avoiding silos of
information and manual interventions.
• IoT – Based on smart, interconnected devices to generate information about their environments
and operations. This provides real-time guidance in compliance and traceability in supply
chain and manufacturing, including controlling inventory, optimising energy consumption and
tracking material.
• Automation and robotics – Use of robotics to reach an unprecedented level of automation to
increase operation accuracy, repeatability and throughput, while reducing human errors and
improving quality and compliance.
• Analytics – Use of the latest tools and analytical methods to generate scientific and business
insights for informed decision-making.
In July 2018, Moderna opened their state-of-the-art, digitally enabled Moderna Technology Center
(MTC) manufacturing facility in Norwood, MA, which was designed to Current Good Manufacturing
Practices (cGMP) specifications. The facility has three core functions:
• Pre-clinical production – To develop materials for pre-toxicology studies using integrated robotics
to produce around 1,000 mRNA per month at research scale.
• Clinical production – To run Phase I and II clinical development programs driven by real-time data
and a fully integrated manufacturing execution system.
• Personalised cancer vaccine (PCV) unit – For the fast manufacturing and supply of
individualised batches.52
Moderna’s digital strategy enables continuous exchange of data, while reducing response time and
error proofing, to integrate compliance and provide information on all the manufacturing activities.53
12
Creating value from AI
Optimising predictive
A digitised and integrated quality and compliance
maintenance function can be a competitive advantage in terms of
innovation, pricing and quality. However, manufac-
Traditionally, manufacturing facilities have oper- turers still commonly face the challenge of reducing
ated in preventive maintenance or run-to-failure maintenance costs and duration of time-sensitive
modes. Preventive maintenance normally consists repairs, while ensuring that operating units work
of scheduled procedures, such as routine asset efficiently. Estimates suggest that between 60 to
monitoring and visual inspection, to obtain regular 73 per cent of all manufacturing data is not utilised
information on the condition of the different or analysed.57,58 By using advanced technological
system components.54 In contrast, run-to-failure tools, such as AI, these data can be transformed
maintenance lets machinery run until it breaks into vital insights about operations and equipment
down before being repaired.55 These approaches performance, by identifying patterns and complex
can make operations inefficient, lead to permanent relationships between variables, as well as fore-
equipment failure and, ultimately, may result casting failures, faults or other issues before they
in unnecessary downtime of entire production happen. ML can help manufacturing assets to be
lines with serious financial consequences. ready when needed by preventing unplanned down-
time. This technology can supply information not
A 2016 Deloitte survey found that more than only to pinpoint and address the problem causing
a third ( |
324 | deloitte | DI_AI-readiness-for-government.pdf | A report from the
Deloitte Center for Government Insights
AI readiness for government
Are you ready for AI?
About the authors
Ed Van Buren | [email protected]
Ed Van Buren is a principal at Deloitte Consulting LLP and the leader of Deloitte’s Government & Public
Services (GPS) Strategy and Analytics (S&A) practice. With more than 20 years of consulting and public
sector experience, Van Buren has served a diverse portfolio of clients across the civilian, defense, and
national security sectors. Prior to leading the GPS S&A practice, he led Deloitte’s United States Postal
Service account, providing a range of strategy, risk, supply chain, and technology services and growing
the account to one of the largest in the GPS practice. He specializes in strategic planning and
implementation, and has extensive project experience designing and implementing enterprise, sales,
and retail strategies, performance measures, business processes, and technology programs as part of
large-scale organizational transformations.
Bruce Chew | [email protected]
Bruce Chew is a managing director with Monitor Deloitte, Deloitte Consulting LLP’s strategy
service line. For more than 20 years, his work has focused on strategy development and
implementation and the building of organizational capabilities. Chew is a former Harvard
Business School professor and has twice served on the advisory board panel for the president’s
Federal Customer Service Awards. He has worked with the federal government, universities,
and companies across a broad range of industries. Chew is based in Kennebunkport, Maine
William D. 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 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, VA.
Contents
Introduction | 2
Your AI readiness depends on your destination | 5
Key milestones on the AI journey | 8
Frequently asked questions about AI readiness | 9
Endnotes | 10
AI readiness for government
Introduction
Is your agency ready for artificial intelligence (AI)? If not, what would it take to
get to a place where it can enjoy the benefits of AI?
A GOVERNMENT AGENCY’S READINESS for If an organization wishes to progress beyond pilots,
AI is not simply a question of preparing to it is helpful to consider the following distinct but
buy and install new technology. The interdependent areas in which to assess AI
transformative nature of AI typically calls for readiness: strategy; the organizational dimensions
preparation across multiple critical areas. To of people and processes; the technology-focused
capture AI’s potential to create value, government dimensions of data; technology and platforms;
organizations will need a plan to retool the relevant and the ethical implications of this
existing processes, upskill or hire key staff, refine transformative capability (figure 1).
approaches toward partnership, and develop the
necessary data and technical infrastructure to All these six areas can be important because all are
deploy AI. likely to require action and change during the AI
journey defined by your agency. They can help you
form an initial baseline as to where you are and
how ready you are to undertake the journey:
• Strategy. Because AI is a transformative
technology, alignment on direction and level of
ambition is crucial. Define an AI vision and
goals that align with organizational objectives,
and then you can devise an approach for
managing capability across the enterprise. (See
our companion piece on “Crafting an AI
strategy for government.”)
• People. Agencies may face challenges around
accessing and recruiting necessary technical
skills, as well as helping existing employees
develop and deploy AI skills.1 To address these
22
Are you ready for AI?
FIGURE 1
AI readiness can be assessed in six areas
WHY
STRATEGY
S
H
TW
I
WHAT TECHN
DOL AO TG
AY &
d a
ta
strD
aepP
tl eoy
gymL
e n t
d
EA
n
t e
im
r
so
pS
d
r
ceT
ie
l ss
i
ec
p
u
F
l
r
i
i
n
t c
e
y o
DO sn&
at i
t
anR
u
&
i
tA
y
cr
S
ocM
e
mh ci ut pt re lic o itt ayol ,u
n
sr
p
ce
er
iv&
ac y
eT xra pn
laA
s
im
p na
ab ri bt eAi ilo
n
iA
tn
I
cI
y
y
O
R
&
P E
A
E
R
l i
AA
g
T
n
I
D
m
N
P
G
o
e
I
l
n
M
iN
ct
i
O
e
E
D
s
ES
A
L
p
S
p r o
B
iia nac tsh
eg& rity
M e
aO sdr ueg rsa ei fgn n
m
uiz
e
n
a
n
d
tt
i
i o
n&
g
n a ml
D e
o
l
id
v
e
e
rl
y
T a
c
Gl oe
ovm
en
rm nau
ncet
nic atio ns C h an ge a n
dP E PO
RP OL CE
ESSW H O
W
ETHICS O
H
Source: Deloitte analysis.
Deloitte Insights | deloitte.com/insights
areas, consider integrating AI with human it is integrated with the work and processes of
workflows, redefining talent models, and getting the organization.2
stakeholder buy-in through effective
communications and change management. • Data. AI is only as good as the data upon which
it is built, and its appetite for data is voracious.
• Processes. Establish, define, and design Design a data governance system that includes
processes, controls, and governance systems to engineering and security. Data governance
enable successful AI implementation. While AI should include rules for sourcing, accessing,
pilots can serve to provide proof of AI’s and quality management.3
potential, its true value cannot be captured until
33
AI readiness for government
To capture AI’s potential to
the computing environment. A variety of
create value, government models for pursuing AI exist4 that vary in
terms of platforms and ownership of
organizations will need a plan
technology (e.g., internal or in partnership),
but, in all cases, AI requires a coherent
to retool the relevant existing
approach that considers future
processes, upskill or hire key requirements as AI scales within the
organization and its usage evolves.
staff, refine approaches toward
partnership, and develop the • Ethics. Establish mechanisms to
understand and prevent AI bias, promote
necessary data and technical
fairness and transparency, and ensure
infrastructure to deploy AI. values and integrity are embedded in
AI-driven initiatives. While any technology’s
• Technology and platforms. Procure and deployment should be ethical, AI brings
develop appropriate AI technology and issues such as transparency, privacy, and bias
platforms to operationalize AI assets, including into particular focus.5
those related to vendors, interoperability, and
4
Your AI readiness depends
on your destination
WHILE ALL SIX areas described above 2. Process- or problem-focused use cases
should be considered in all AI initiatives,
the level of effort will largely depend 3. AI-fueled transformation (which has the
both on the current state of the organization and potential to bring the greatest change and
the ambitiousness of the agency’s vision for AI highest value)
(figure 2). Broadly speaking, an agency’s level of
ambition can be categorized as: Generally speaking, the more ambitious an
agency’s goal (further to the right in figure 2), the
1. Narrow, single-point solutions greater the value, the broader the scope, and, as a
Source: Deloitte analysis.
Deloitte Insights | deloitte.com/insights
5
EULAV
LAITNETOP
Are you ready for AI?
FIGURE 2
AI’s nonlinear nature means that agencies can start their AI journey from
any point on the AI ambition curve
High
Transformative
Use case
targeted
Holistic approaches
Single-point
seeking to transform the
solutions
organization utilizing AI to
enhance speed, efficiency,
Task-based point Process- or problem-focused and productivity while
solutions typically applications involving enhancing mission success
involving robotic intelligent automation,
process automation engagement and/or insights
Low
Narrow Broad
SCOPE OF EFFORT
5
AI readiness for government
result, the greater the technical and organizational operations businesses, but they may also be
complexity. program or mission leaders.
Narrower, single-point solutions will typically Finally, truly transformational efforts seek to
demand less of a stretch by an organization. In develop AI-fueled breakthroughs in back-office
these instances, AI can generate quick efficiencies performance or mission outcomes. Leaders identify
by automating simple processes, often in back- opportunities to fundamentally change a business
office areas most prone to standardization. This process or mission area through a combination of
allows staff to refocus their time and effort on more AI technologies and organizational and process
meaningful issues. If many such single-point changes. Areas of opportunity can include
opportunities exist, collectively, they can represent reimagined clinical trial operations, AI-augmented
or autonomous security
clearances and vetting, smart
Leaders identify opportunities to
cities, and revenue service
fundamentally change a business process collections. Transformational
uses of AI can maximize the
or mission area through a combination
technology’s value as an
of AI technologies and organizational enabler of organizational
change.
and process changes.
An agency need not limit
itself to a single approach.
significant value. This path can be a relatively easy For example, consider the AI applications that the
way for an agency to start using AI, with relatively US Department of Defense (DoD) outlines in its AI
quick returns to build support for AI solutions. strategy. The DoD has identified opportunities to
deploy AI across the ambition curve. In terms of
An approach focused on use cases considers point solutions, using intelligent automation to
common processes or problems that can be reduce time spent on manual and repetitive tasks
developed, then rolled out to other parts of the may generate low value per task, but given the
agency. This approach can yield a higher value but DoD’s size, the aggregate opportunity is very high.
generally represents a greater challenge, calling for The DoD is also looking at a specific use case for AI
a higher state of readiness. It leverages similar to enable predictive maintenance, anticipating the
types of AI across the enterprise—for example, need for repairs to critical equipment in order to
intelligent chatbots in contact centers, or natural optimize inventory levels. And on the
language processing and predictive analytics for transformative end of the curve, developing a
units that deal with large amounts of data, whether solution that can help predict or prevent
structured (such as disparate HR databases) or cyberattacks or greatly advance the ability to
unstructured (such as emails, memos, and explain AI algorithms could have transformative
documents). During this time, some agency leaders effects.6
tend to emerge who are evangelists in scaling AI
across their enterprise. Where these leaders will be Many commercial and government organizations
located depends on which use cases are chosen and are pursuing an approach that carries them from
how the adoption of AI is governed: AI leaders can left to right along the ambition curve. Starting on
come from the IT, data security, finance, or the left with point solutions can give organizations
66
Are you ready for AI?
experience in implementing AI in a less complex
setting. As they install point solutions,
organizations may also overhaul their data and
technology infrastructure to create a stronger
foundation for future AI implementations.
Organizations progressing to AI for specific use
cases often find that these projects provide solid
evidence of scalable benefits, which can encourage
strong advocacy for AI and its larger potential.
Success or failure at this stage sets the tone for AI’s
further deployment: While success tends to push
an agency toward looking at more transformative
opportunities, failure could deter agencies from
fully scaling AI and may even discourage existing
AI efforts.
Assessing your agency’s readiness depends on its
current strategy, people, processes, data,
technology and platforms, ethics choices and
governance, as well as on its high-level strategic
choices concerning its level of ambition and path
forward. These strategic choices should also reflect
the agency’s goals, challenges, and available
funding. But regardless of the precise character of
your AI path and destination, there are some
universal milestones along the way.
77
AI readiness for government
Key milestones on the
AI journey
DEVELOPING A COHERENT AI strategy is the (figure 3). Note that an agency might cycle through
first step in creating a clearer set of choices this journey multiple times if it begins first with
for building and deploying AI capabilities. point solutions and takes them through to scaling
It defines an agency’s level of ambition, guides the and ongoing management, followed by broader use
prioritization of focus areas, and, along with an cases and, ultimately, AI-fueled transformation. By
understanding of the agency’s readiness, identifies assessing where they are in this journey, agencies
what critical capabilities need to be developed. can evaluate which capabilities already exist and
which need to be built from the ground up to
The AI strategy development phase sets the stage achieve their AI effort’s desired outcomes.
for the other critical milestones on the AI journey
FIGURE 3
Milestones on the AI journey
Develop Translate Implement Scale Manage
Understand AI Design and validate Undertake rapid Scale and roll out Manage AI-enabled
potential, set AI initiatives to prototyping and proven AI solutions, solutions, updates,
ambition level, confirm costs testing of the AI addressing technical and expansion;
and prioritize and benefits applications with and organizational monitor results,
applications and establish the highest value; barriers adapting as needed
governance evaluate overall
results
Source: Deloitte analysis.
8
Are you ready for AI?
Frequently asked questions
about AI readiness
SOME COMMON QUESTIONS heard from deliver worker retraining, and build new
government leaders as they evaluate their operating models.
AI readiness:
WE HAVE ALWAYS BEEN FAST
WHAT’S THE FIRST THING I SHOULD IN IDENTIFYING USE CASES FOR
DO TO BECOME AI-READY? SPECIFIC TOOLS OR IT CAPABILITIES.
Your first action should be to assess your HOW IS AI DIFFERENT?
organization in the six areas outlined in figure 1 The nature of AI—the types of insights it can
and gauge any current gaps in capabilities, deliver (including predictive insights), its potential
infrastructure, and resources relative to to enhance engagement with citizens and other
your ambition. stakeholders, and its ability to automate highly
complex processes—means that any combination of
OUR ORGANIZATION HAS BEEN AI (deep learning, computer vision, natural
DEPLOYING SIMPLE AUTOMATION. language processing, etc.) can fundamentally
SHOULD WE PUT THIS ON HOLD UNTIL transform how you work in a way not formerly
WE DEVELOP AN AI STRATEGY? possible. But the breadth and diversity of use cases
No. Even simple automation projects can be helpful for AI means that agencies should choose carefully.
in introducing new types of work into If applied around low-value processes, in silos, or
organizations, serving as a learning experience that in areas that are not meaningful, AI is not likely to
prepares people for change. That said, before you yield significant value.
move to your 15th or 50th automation, consider
whether more complex use cases can lead to higher AI has the potential to fundamentally transform
returns and mission impact. government operations. However, agencies must
be ready to take advantage of this potential. To do
IN ADDITION TO DEDICATED this, they should build a solid foundation by
DATA SCIENTISTS AND AI putting the right data and technology platforms in
PROFESSIONALS, WHAT OTHER place, while at the same time developing the talent,
PEOPLE RESOURCES ARE NEEDED TO strategy, and governance processes needed to
START REALIZING VALUE FROM AI? effectively implement and use AI solutions.
Specific IT and AI skills are critical, but they are
not sufficient for success. Also required are AI’s transformative potential is so strong that it will
individuals who can help identify which business likely eventually become ubiquitous across
and mission areas to focus on, set up governance government. If this happens, then success in
and ethics frameworks and processes, consider navigating the AI journey will play a large part in
relevant center-of-excellence models, drive culture determining how effectively government agencies
change and change management, develop and deliver on their mission.
9
AI readiness for government
Endnotes
1. William D. Eggers et al., How to redesign government work for the future, Deloitte Insights, August 5, 2019.
2. William D. Eggers, David Schatsky, and Dr. Peter Viechnicki, AI-augmented government: Using cognitive
technologies to redesign public sector work, Deloitte University Press, April 26, 2017.
3. Omer Sohail, Prakul Sharma, and Bojan Cric, Data governance for next generation platforms, Deloitte, 2018.
4. Nitin Mittal, Dave Kuder, and Samir Hans, AI-fueled organization, Deloitte Insights, January 26, 2019.
5. Nihar Dalmia and David Schatsky, The rise of data and AI ethics, Deloitte Insights, June 24, 2019.
6. US Department of Defense, “Summary of the 2018 Department of Defense artificial intelligence strategy,” 2018.
10
Are you ready for AI?
Acknowledgments
The authors would like to thank Pankaj Kishnani for his research contributions, as well as Tina
Mendelson for her review at critical junctures and contributing her ideas and insights to this project.
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.
11
AI readiness for government
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.
Industry leadership
Ed Van Buren
Principal | Deloitte Consulting LLP
+ 1 571 882 5170 | [email protected]
Ed Van Buren is a principal at Deloitte Consulting LLP and the leader of Deloitte’s Government & Public
Services (GPS) Strategy and Analytics (S&A) practice. He is based in Arlington, VA.
The Deloitte Center for Government Insights
William D. Eggers
Executive director | Deloitte Center for Government Insights | Deloitte Services LP
+ 1 571 882 6585 | [email protected]
William D. Eggers is the executive director of Deloitte’s Center for Government Insights. He is based in
Rosslyn, VA.
Bruce Chew
Managing director | Deloitte Consulting LLP
+ 1 617 437 3526 | [email protected]
Bruce Chew is a managing director with Monitor Deloitte, Deloitte Consulting LLP’s strategy service line.
He is based in Kennebunkport, Maine.
Achieving your mission outcomes, whether a small-scale program or an enterprisewide initiative,
demands ever-smarter insights—delivered more quickly than ever before. Doing that in today’s
complex, connected world requires the ability to combine a high-performance blend of humans
with machines, automation with intelligence, and business analytics with data science. Welcome
to the Age of With, where Deloitte translates the science of analytics—through our services,
solutions, and capabilities—into reality for your organization. Learn more on Deloitte.com.
12
Sign up for Deloitte Insights updates at www.deloitte.com/insights.
Follow @DeloitteInsight
Deloitte Insights contributors
Editorial: Aditi Rao, Abrar Khan, Blythe Hurley, Aparna Prusty, and Anya George Tharakan
Creative: Sonya Vasillieff
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 © 2020 Deloitte Development LLC. All rights reserved.
Member of Deloitte Touche Tohmatsu Limited |
327 | deloitte | cdo-data-foundation-insights.pdf | The Mission-Driven CDO
Insights from the 2023 Survey of
Federal Chief Data Officers (CDOs)
In the fall of 2023, federal department-, agency-, and bureau-level CDOs and
Statistical Officers completed a survey developed by the Data Foundation and
Deloitte to understand the evolving CDO role and CDO community needs. The
insights below are based on the results of this survey, which is the fourth annual
of its kind.
CDOs are...
Catalysts Strategists
for AI adoption and innovation within their aligning data governance and equitable practices to the organization’s
organization. mission.
• 55% of CDOs already use basic or advanced AI • CDOs are supporting their organization’s mission by maximizing the
and 95% intend to adopt new AI technologies for value of their organization’s data, supporting a data community, and
their organizations in the next year. leading the development of data policies and processes.
• The 2023 Executive Order establishing the Chief • CDOs are expanding data-driven decision making, improving data
AI Officer (CAIO) role will increase the expansion infrastructure and data quality (i.e., demographic
of AI throughout all organizations. representation in data), and promoting
inclusivity in the workplace and in staffing.
CDOs will be critical partners to CAIOs,
aligning all cross-functional areas of CDOs are responsible for orienting their
their organization to strategic AI organization towards equitable and
initiatives. data-centered approaches that serve
their mission and the public.
Champions Operators
of data literacy and culture in the of shared data agendas and evolving
workforce to keep pace with emerging needs of their organizations.
technology.
• 52% of CDOs work with a host of C-Suite
• Well-trained talent specializing in the leaders, with 60% of CDOs naming CIOs as
intersection of data, AI, and industry is cited the leader they collaborate with most frequently. In
by 60% of CDOs as a key resource needed to 2023, more CDOs (55%) experienced challenges reporting up to CIOs than in
effectively carry out their missions. 2022 (34%).
• Beyond foundational data knowledge, 75% • CDOs cite funding, authority, and staffing contraints as the top three barriers
of CDOs believe their roles also influence the hindering mission success. CDOs also provided an array of additional barriers,
organization’s data culture, encouraging data indicating that each organization faces unique challenges.
professionals to value data and use it ethically
and responsibly. With the advent of the new CAIO position, it is even more crucial for CDOs to
establish shared agendas across leaders. Despite differences among organizations,
Data literacy programs can position their the key to success is that each organization’s structure and resources empowers
organization’s staff for success and boost the CDO office to achieve their data goals and mission requirements.
data-driven decisions.
The 2023 Federal CDO Survey illustrated four emerging characteristics of
CDOs: Catalyst, Strategist, Champion, and Operator. To accelerate CDO and
CAIO journeys towards these goals, below are a sample of Deloitte’s suite of
tools and services.
The Catalyst - Thinking about innovating and adopting new technology?
AI Readiness & Management Framework (aiRMF): Partner with Deloitte to assess where you are on your AI journey, define target
outcomes, and chart a path forward across 10 AI capability areas to achieve enterprise AI readiness and maturity.
Government AI Use Case Dossier: See what’s working for other agencies and consider the ways AI can advance your mission with the
Government and Public Services Sector AI Use Case Dossier.
Trustworthy AI™: Understand seven key areas of risk for AI and keep your use of AI safe and ethical with Deloitte’s Trustworthy AI™
framework in line with NIST.
The Strategist - Thinking about equity and data centered approaches?
AI & Data Strategy Services: Align on an organizational vision for AI, prioritize AI use cases, and make strategic choices about where to
invest in AI, accelerated by Playbooks and immersive Labs guided by experienced facilitators.
CDO Playbook: See the most recent thought leadership of CDOs in the government based on trends and understanding AI priorities,
strategies, and implementation of operation models.
The Champion - Thinking about data literacy, culture, and quality?
Deloitte’s POV on Data Literacy: Learn how to support members of your organization in reading, working with, analyzing, and using
data to ethically solve challenges, drive innovation, and collaboratively create value.
Trustworthy AI™: Understand seven key areas of risk for AI and keep your use of AI safe and ethical with Deloitte’s Trustworthy AI™
framework in line with NIST.
CDO Playbook: See the most recent thought leadership of CDOs in the government based on trends and understanding AI priorities,
strategies, and implementation of operation models.
The Operator - Thinking about aligning data strategy and data processes?
AI & Data Strategy Services: Align on an organizational vision for AI, prioritize AI use cases, and make strategic choices about where to
invest in AI, accelerated by Playbooks and immersive Labs guided by experienced facilitators.
Data Labs, including CDO/CAIO Transition and the Data Strategy Lab: Create organizational vision, disrupt ordinary thinking, and
learn from industry leaders how to achieve your vision.
Contact Us
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 Adita Karkera Lorenzo Ross
Chief Data Officer, Deloitte Technology Fellow, Deloitte
Officers with the CDO Services they need to navigate the role of the CDO.
Government and Public Services Government and Public Services
[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/about to learn more about our global network of member firms.
Copyright © 2024 Deloitte Development LLC. All rights reserved. |
328 | deloitte | DI_AI-leaders-in-financial-services.pdf | Research from the Deloitte
Center for Financial Services
AI leaders in financial services
Common traits of frontrunners in the
artificial intelligence race
About the Deloitte Center for Financial Services
The Deloitte Center for Financial Services, which supports the organization’s US Financial Services
practice, provides insight and research to assist senior-level decision-makers within banks,
capital markets firms, investment managers, insurance carriers, and real estate organizations.
The center is staffed by a group of professionals with a wide array of in-depth industry experiences
as well as cutting-edge research and analytical skills. Through our research, roundtables, and
other forms of engagement, we seek to be a trusted source for relevant, timely, and reliable
insights. Read recent publications and learn more about the center on Deloitte.com.
Deloitte Analytics and AI
We are Deloitte Analytics. Many of the world’s leading businesses count on us to deliver power-
ful outcomes, not just insights, for their toughest challenges. Fast. Our analytics practice is built
around the wide range of needs our clients bring to us. Data scientists, data architects, business
and domain specialists who bring a wealth of business-specific knowledge, visualization and
design specialists, and of course technology and application engineers. We deploy this deep tal-
ent all over the world, at scale. To learn more, visit Deloitte.com.
Contents
Running the AI leg of the digital marathon 3
Three common traits of AI frontrunners in financial services 6
Significant challenges could lie ahead 14
Getting off to a solid start 17
Appendix: The AI technology portfolio 18
Endnotes 20
AI leaders in financial services
KEY MESSAGES
• Embed AI in strategic plans: Integrating artificial intelligence (AI) into an
organization’s strategic objectives has helped many frontrunners develop an
enterprisewide strategy for AI that various business segments can follow. The greater
strategic importance accorded to AI is also leading to a higher level of investment by
these leaders.
• Apply AI to revenue and customer engagement opportunities: Most frontrunners
have started exploring the use of AI for various revenue enhancements and client
experience initiatives and have applied metrics to track their progress.
• Utilize multiple options for acquiring AI: Frontrunners seem open to employing
multiple approaches for acquiring and developing AI applications. This strategy is
helping them accelerate the adoption of AI initiatives via access to a wider pool of
talent and technology solutions.
22
Common traits of frontrunners in the artificial intelligence race
Running the AI leg of
the digital marathon
THE FINANCIAL SERVICES industry has jump-start or adapt their AI game plans to come
entered the AI phase of the digital marathon. up on top as the race heats up?
The journey for most companies, which To answer these questions, Deloitte surveyed 206
started with the internet, has taken them through US financial services executives to get a better
key stages of digitalization, such as core systems understanding of how their companies are using AI
modernization and mobile tech integration, and technologies and the impact AI is having on their
has brought them to the intelligent automation business (see sidebar, “Methodology: Identifying
stage. AI frontrunners among financial institutions”). The
report identified some of the following key charac-
Many companies have already started implement- teristics of respondents who have gotten off to a
ing intelligent solutions such as advanced analytics, good start and taken an early lead:
process automation, robo advisors, and self-learn-
ing programs. But a lot more is yet to come as Embed AI in strategic plans: Integrating AI
technologies evolve, democratize, and are put to into an organization’s strategic objectives has
innovative uses. helped many frontrunners develop an enterprise-
wide strategy for AI, which different business
To effectively capitalize on the advantages offered segments can follow. The greater strategic impor-
by AI, companies may need to fundamentally tance accorded to AI is also leading to a higher
reconsider how humans and machines interact level of investment by these leaders.
within their organizations as well as externally with
their value chain partners and customers. Rather Apply AI to revenue and customer engage-
than taking a siloed approach and having to rein- ment opportunities: Most frontrunners have
vent the wheel with each new initiative, financial started exploring the use of AI for various revenue
services executives should consider deploying AI enhancements and client experience initiatives and
tools systematically across their organizations, have applied metrics to track their progress.
encompassing every business process and function.
Utilize multiple options for acquiring AI:
As with any race, some companies are setting the Frontrunners seem open to employing multiple
pace, while others are struggling to hit their stride approaches for acquiring and developing AI appli-
after leaving the starting gate. What can those who cations. This strategy is helping them accelerate
are seemingly at the back of the pack do to keep up the adoption of AI initiatives via access to a wider
with their frontrunning competitors? How can they pool of talent and technology solutions.
33
AI leaders in financial services
METHODOLOGY: IDENTIFYING AI FRONTRUNNERS AMONG FINANCIAL INSTITUTIONS
To understand how organizations are adopting and benefiting from AI technologies, in the third quarter
of 2018 Deloitte surveyed 1,100 executives from US-based companies across different industries that are
prototyping or implementing AI.1 In this report, we focus on a sample of 206 respondents working for
financial services companies. All respondents were required to be knowledgeable about their company’s
use of AI technologies, with more than half (51 percent) working in the IT function. Sixty-five percent of
respondents were C-level executives—including CEOs (15 percent), owners (18 percent), and CIOs and
CTOs (25 percent).
All financial services respondents in the survey were required to be currently using AI technologies in
some form or another (see “Appendix: The AI technology portfolio”). The entire respondent base of
individuals working for financial institutions could thus be considered as early adopters of AI initiatives.
Within this respondent base, we wanted to identify the practices adopted by those leading the pack in
terms of AI deployment experience and tangible returns achieved from them. Using data from Deloitte’s
AI survey, we identified two quantitative criteria for further analysis: performance (financial return from AI
investments) and experience (number of fully deployed AI implementations, which represents AI projects
that are “live,” fully functional, and completely integrated into business processes, customer interactions,
products, or services).
We found that companies could be divided into three clusters based on the number of full AI
implementations and the financial return achieved from them (figure 1). Each of these clusters
represents respondents at different phases of their current AI journey.
• Frontrunners: Thirty percent of respondents worked for companies that had achieved the highest
financial returns from a significant number of AI implementations.
• Followers: Forty-three percent of respondents worked for companies in the middle ground of AI
implementations and financial returns.
• Starters: Twenty-seven percent of respondents worked for companies that were at the start of their AI
journey and/or lagging in the level of return achieved from AI implementations.
4
METHODOLOGY: IDENTIFYING AI FRONTRUNNERS AMONG FINANCIAL INSTITUTIONS, CONT.
Frontrunners: 30%
1%
11+ 12% 8% 13%
2%
6–10 11% 11% 8%
Starters: 27% Followers: 43%
3–5 7% 7%
2% 2%
3%
1–2 7% 6%
0% or lower +10% +20% +30% or
return return return higher return
Financial return on AI investments
Note: Percentages may not total 100 percent due to rounding.
Source: Deloitte Center for Financial Services analysis, based on Deloitte Services LP “State of AI in the Enterprise,”
2nd Edition survey, 2018.
Deloitte Insights | deloitte.com/insights
5
nekatrednu
snoitatnemelpmi
lluf
fo
rebmuN
Common traits of frontrunners in the artificial intelligence race
FIGURE 1
Respondent segmentation based on AI implementations and track record
AI leaders in financial services
Three common traits of AI
frontrunners in financial
services
AS FINANCIAL INSTITUTIONS look to find a reduction; and adopt a portfolio approach for
rhythm in their AI race, frontrunners could acquiring AI, where they utilize multiple develop-
provide an early-bird view into how to ment models for implementing AI solutions
effectively integrate the technology with an organi- (figure 2).
zation’s strategy, as well as which approaches
companies could adopt for implementing such ini- EMBED AI IN STRATEGIC PLANS WITH
tiatives throughout their organization. EMPHASIS ON ORGANIZATIONWIDE
IMPLEMENTATION
From the survey, we found three distinctive traits While many financial services companies agree that
that appear to separate frontrunners from the rest. AI could be critical for building a successful com-
Frontrunners are generally able to embed AI in petitive advantage, the difference in the number of
strategic plans and emphasize an organizationwide respondents in the three clusters that acknowl-
implementation plan; focus on revenue and cus- edged the critical strategic importance of AI is
tomer opportunities, rather than just cost quite telling (figure 3).
FIGURE 2
Survey spotlights key practices among AI frontrunners in financial institutions
Embed AI in strategic plans with
1
emphasis on organizationwide
implementation
Focus on applying AI to 2
revenue and customer
engagement opportunities
3 Adopt a portfolio approach
for acquiring AI
Source: Deloitte Center for Financial Services analysis, based on Deloitte Services LP “State of AI in the Enterprise,”
2nd Edition survey, 2018.
Deloitte Insights | deloitte.com/insights
6
Common traits of frontrunners in the artificial intelligence race
FIGURE 3
Frontrunners better recognize strategic importance of AI adoption
Importance of adopting or using AI to a company’s overall business success
Minimally important Somewhat important Very important Critical strategic importance
25% 6% 2%
8% 6% 6%
8% 18% 38%
Frontrunners Followers Starters
59% 71% 53%
Frontrunners recognize the critical strategic importance over four times more
than followers and over 12 times more than starters.
Source: Deloitte Center for Financial Services analysis, based on Deloitte Services LP “State of AI in the Enterprise,”
2nd Edition survey, 2018.
Deloitte Insights | deloitte.com/insights
An early recognition of the critical importance of AI in AI more heavily than other segments, while also
to an organization’s overall business success proba- accelerating their spending at a higher rate. Close
bly helped frontrunners in shaping a different AI to half of the frontrunners surveyed had invested
implementation plan—one that looks at a holistic more than US$5 million in AI projects compared
adoption of AI across the enterprise. The survey to 27 percent of followers and only 15 percent of
indicates that a sizable number of frontrunners had
launched an AI center of excellence, and had put in
FIGURE 4
place a comprehensive, companywide strategy for AI
A significant number of frontrunners
adoptions that departments had to follow (figure 4).
have a detailed organizationwide
AI strategy
For example, as part of an overall strategy to
become a “bank of the future,” Canada-based TD
Bank set up an Innovation Centre of Excellence Frontrunners 49%
Followers 41%
(CoE). Acting like an umbrella organization, the
Starters 36%
CoE connects all the innovation initiatives, includ-
ing AI, to broader bank business units. It provides Almost half of the frontrunners
have a comprehensive, detailed,
a platform for experimentation across the organi-
companywide strategy in place for
zation with the purpose of reducing operational AI adoption, which departments
complexity and improving customer experience. are expected to follow.
The CoE thus helps in testing and identifying best
practices from AI pilots before introducing them as
full-scale customer solutions.2
It is also no surprise, given the recognition of stra- Source: Deloitte Center for Financial Services analysis,
based on Deloitte Services LP “State of AI in the
tegic importance, that frontrunners are investing Enterprise,” 2nd Edition survey, 2018.
Deloitte Insights | deloitte.com/insights
7
AI leaders in financial services
FIGURE 5
Frontrunners are investing more in AI initiatives
Investments in AI/cognitive technologies/projects in the recent fiscal year
Less than $250,000 $250,000 to $499,999 $500,000 to $999,999 $1 million to $4.99 million
$5 million to 9.99 million $10 million or more
Frontrunners
11% 11% 31% 20% 25%
Followers
11% 15% 46% 15% 12%
Starters
4% 15% 34% 30% 9% 6%
Note: All dollar amounts refer to US dollars.
Source: Deloitte Center for Financial Services analysis, based on Deloitte Services LP “State of AI in the Enterprise,”
2nd Edition survey, 2018.
Deloitte Insights | deloitte.com/insights
starters (figure 5). In fact, 70 percent of frontrun- that they consider what value they are trying to
ners plan to increase their AI investments by deliver for clients using AI.
10 percent or more in the next fiscal year, com-
pared to 46 percent of followers and 38 percent of Value delivery could either include customizing
starters (figure 6). offerings to specific client preferences, or continu-
ously engaging through multiple
channels via intelligent solutions
Frontrunners are investing in AI more
such as chatbots, virtual clones,
heavily than other segments, while and digital voice assistants.
also accelerating their spending at a
For developing an organization-
higher rate. wide AI strategy, firms should keep
in mind that these might be
applied across business functions.
A major emphasis of these investments likely was Starting purposefully with small projects and learn-
to secure the talent and technologies necessary for ing from pilots can be important for building scale.
the transformational journey ahead.3
For scaling AI initiatives across business functions,
Calls to action building a governance structure and engaging the
For financial institutions early in their AI entire workforce is very important. Adding gamifi-
journey, embedding AI in strategic initia- cation elements, including idea-generation contests
tives is an important first step. Elevating the and ranking leaderboards, garners attention, gets
critical importance assigned to these initiatives, ideas flowing, and helps in enthusing the work-
along with building a long-term AI vision and strat- force. At the same time, firms should develop
egy, lays out the foundation for the strategy. As programs for upskilling and reskilling impacted
companies customize their AI strategy based on workforce, which would help garner their contin-
their scale, size, and complexity, it is important ued support to AI initiatives.
8
Common traits of frontrunners in the artificial intelligence race
FIGURE 6
Frontrunners plan to increase AI allocations at a faster clip
Predicted change in companies’ investment in AI/cognitive
Decrease by 10–20% Decrease by 1–9% Stay the same Increase by 1–9%
Increase by 10–20% Increase by more than 20% Don’t know
2% 2%
16% 10% 5% 4% 5% 9% 17%
2%
16%
Frontrunners Followers Starters
30%
41% 46%
54% 43%
More than two-thirds of frontrunners plan to
increase their AI investments by 10% or more.
Source: Deloitte Center for Financial Services analysis, based on Deloitte Services LP “State of AI in the Enterprise,”
2nd Edition survey, 2018.
Deloitte Insights | deloitte.com/insights
FOCUS ON APPLYING AI TO That said, what differentiated frontrunners (figure
REVENUE AND CUSTOMER 7) is the fact that more leading respondents are
ENGAGEMENT OPPORTUNITIES measuring and tracking metrics pertaining to rev-
Despite steady improvement in the economy fol- enue enhancement (60 percent) and customer
lowing the 2008 financial crisis, the pressure to experience (47 percent) for their AI projects. This
reduce costs at financial institutions has contin- approach helped frontrunners look at innovative
ued to increase. At the same time, rising ways to utilize AI for achieving diverse business
competition from incumbents and nontraditional opportunities, which has started to bear fruit.
entrants, as well as greater regulatory oversight
and compliance demands, are raising the cost of A good case could be how AI and predictive analyt-
doing business. The return on average equity of ics were used by UK-based Metro Bank to help
commercial banks, for example, has yet to reach customers manage their finances. Working in part-
pre-financial-crisis levels.4 nership with Personetics, the bank launched an
in-app service called Insights, which monitored
It is no surprise, then, that one in two respon- customers’ transaction data and patterns in real
dents were looking to achieve cost savings or time. The app then provided personalized prompts
productivity gains from their AI investments. to make subscription payments and be aware of
Indeed, in addition to more qualitative goals, AI unusual spending. The AI tool also provides per-
solutions are often meant to automate labor- sonalized financial advice, including savings
intensive tasks and help improve productivity. recommendations and alerts.5
Thus, cost saving is definitely a core opportunity
for companies setting expectations and measuring Frontrunners have taken an early lead in realizing
results for AI initiatives. better business outcomes (figure 8), especially in
9
AI leaders in financial services
FIGURE 7
Frontrunners focus on revenue and customer opportunities in addition to
cost reduction
Frontrunners Followers Starters
Cost reduction Revenue enhancement Customer engagement
60%
54%
49%49% 49%
47% 46% 46% 47%47%
42%
40%
Cost savings targets Productivity targets Revenue targets Customer-related
(sales, cross-selling) targets (engagement,
satisfaction, and
retention)
Source: Deloitte Center for Financial Services analysis, based on Deloitte Services LP “State of AI in the Enterprise,”
2nd Edition survey, 2018.
Deloitte Insights | deloitte.com/insights
achieving revenue enhancement goals, including relevant response. Another project uses algo-
creating new products and pursuing new markets. rithms to study central bank documents and
understand the central bank’s economic perspec-
This mindset was reflected in the overall perfor- tive. The bank is also actively evaluating
mance among respondents as well, with opportunities to deploy AI for automating claims
frontrunners reporting a companywide revenue handling, detecting fraud, and providing personal-
growth of 19 percent according to the survey, ized recommendations to clients.7
which was in stark contrast to the growth of
12 percent for followers and a decline of Calls to action
10 percent for starters. While exploring opportunities for
deploying Al initiatives, companies
Meanwhile, our research indicated that companies should explore product and service expansion
should give special emphasis to the human-cen- opportunities. This could be kick-started by mea-
tered design skills needed to develop personalized suring and tracking outcomes of AI initiatives to
user experiences.6 In fact, the survey found that the company’s top line. Adding AI adoption to
frontrunners are already starting to suffer from a sales and performance targets and providing AI
shortage of designers for AI initiatives, which indi- tools for sales and marketing personnel could also
cates the high degree of application of these skills help in this direction.
by frontrunners during AI implementations.
To boost the chances of adoption, companies
Nordic bank Nordea is using AI to lead multiple should consider incorporating behavioral science
efforts across the organization. Nova, an internally techniques while developing AI tools. Companies
developed chatbot, uses natural language process- could also identify opportunities to integrate AI
ing to interpret customers’ queries and decide the into varied user life cycle activities. While
10
Common traits of frontrunners in the artificial intelligence race
FIGURE 8
Frontrunners have achieved better business outcomes across revenue objectives
Frontrunners Followers Starters
Cost reduction Customer engagement
32%33% 32%33% 32%33%
15% 15%
6%
Reduce headcount Reduce operating Improve customer
through automation costs experience
Revenue enhancement
47%
44%
39%
33%
25% 24% 27% 24%
13%
9% 8%
6%
Optimize external Free up workers Create new products Pursue new markets
processes such as to be more creative
marketing and sales by automating tasks
Source: Deloitte Center for Financial Services analysis, based on Deloitte Services LP “State of AI in the Enterprise,”
2nd Edition survey, 2018.
Deloitte Insights | deloitte.com/insights
working on such initiatives, it is important to also From our survey, it was no surprise to see that
assign AI integration targets and collect user feed- most respondents, across all segments, acquired
back proactively. AI through enterprise software that embedded
intelligent capabilities (figure 9). With existing
Companies can also look at making best-in-class vendor relationships and technology platforms
and respected internal services available to exter- already in use, this is likely the easiest option for
nal clients for commercial use. most companies to choose.
ADOPT A PORTFOLIO APPROACH For example, Guidewire, maker of enterprise soft-
FOR ACQUIRING AI ware solutions for insurance companies, offers its
As market pressures to adopt AI increase, CIOs of users access to AI capabilities through its
financial institutions are being expected to deliver Predictive Analytics for Claims app. The app uti-
initiatives sooner rather than later. There are mul- lizes machine learning algorithms to categorize
tiple options for companies to adopt and utilize AI claims based on their severity and the potential for
in transformation projects, which generally need litigation, automatically routing any high-priority
to be customized based on the scale, talent, and claims to the correct departments.8 Similarly,
technology capability of each organization. Salesforce helps users access AI through its
11
AI leaders in financial services
Einstein program, which applies machine learning developing AI in multiple ways (figure 9)—what
to historical sales data and predicts which pros- we refer to as the portfolio approach.
pects are most likely to close.9
This portfolio approach likely enabled frontrun-
However, the survey found that frontrunners (and ners to accelerate the development of AI solutions
even followers, to some extent) were acquiring or through options such as AI-as-a-service and auto-
mated machine learning. At the same time,
FIGURE 9
Frontrunners are comfortable in developing AI through multiple options
Frontrunners Followers Starters
AI as a service
56%
55%
32%
Enterprise software with integrated AI/cognitive
61%
51%
58%
Data science modeling tools
54%
58%
34%
Automated machine learning
63%
60%
36%
Codevelopment with partners
49%
57%
42%
Open-source AI/cognitive development tools
65%
63%
40%
Crowdsourced development communities
56%
39%
32%
Note: Chart indicates percentage of respondents that are already using the above-mentioned ways to acquire AI.
Source: Deloitte Center for Financial Services analysis, based on Deloitte Services LP “State of AI in the Enterprise,”
2nd Edition survey, 2018.
Deloitte Insights | deloitte.com/insights
12
Common traits of frontrunners in the artificial intelligence race
through crowdsourced development communities, well-fitting model within a few hours. This model
they were able to tap into a wider pool of talent was converted to an application programming
from around the world. interface (API), which was combined with RPA to
automate the entire email classification, depart-
Adopting the portfolio approach could help compa- ment identification, and mail-forwarding process.10
nies preserve the legacy business process while
utilizing AI for incremental gains. American Calls to action
Fidelity Assurance, a US-based health and life Financial institutions that have never uti-
insurance company, was evaluating options to lized multiple options to access and
improve the handling of a growing volume of cus- develop AI should consider alternative sources for
tomer emails and mapping the flow to different implementation. Companies would need time to
departments. The company’s R&D team was gather the requisite experience about the benefits
exploring both robotic process automation (RPA) and challenges of each method and find the right
and machine learning applications, albeit sepa- balance for AI implementation.
rately. While RPA was a good match for
automatically sending mails to the correct depart- Once companies start implementing AI initiatives,
ment, it was providing too many rules for a mechanism for measuring and tracking the effi-
identifying the right department based on the cacy of each AI access method could be evaluated.
email’s subject and keywords. To resolve this, the Identifying the appropriate AI technology approach
team decided to explore automated machine learn- for a specific business process and then combining
ing with the help of a third-party vendor. Using the them could lead to better outcomes.
database of customer emails and eventual depart-
ment response (outcome), the company found a
13
AI leaders in financial services
Significant challenges
could lie ahead
AS FINANCIAL SERVICES companies advance Indeed, starters would likely be better served if
in their AI journey, they will likely face a they are cognizant of the risks identified by front-
number of risks and challenges in adopting runners and followers alike (figure 11) and begin
and integrating these technologies across the orga- anticipating them at the onset, giving them more
nization. But not all are facing the same set of time to plan how to mitigate them.
challenges. Our survey found that frontrunners
were more concerned about the risks of AI (figure We observed a similar pattern in terms of the skills
10) than other groups. gap identified by different segments in meeting the
needs of AI projects (figure 12). More frontrunners
With the experience of several more AI implemen- rated the skills gap as major or extreme compared
tations, frontrunners may have a more realistic to the other groups. While a higher number of
grasp on the degree of risks and challenges posed implementations undertaken could partly explain
by such technology adoptions. Starters and follow- this divergence, the learning curve of frontrunners
ers should probably brace themselves and start could give them a more pragmatic understanding
preparing for encountering such risks and chal- of the skills required for implementing AI projects.
lenges as they scale their AI implementations.
FIGURE 10
Divergence in risk estimation for different segments
Companies’ investment in AI/cognitive
No concern Minimal concern Moderate concern Major concern Extreme concern
1%
47%5% 11% 30% 11% 13% 8%
19%
Frontrunners Followers Starters
18% 19% 39% 34% 45%
Almost two-thirds of frontrunners highlight potential risks
associated with AI to be of major or extreme concern.
Source: Deloitte Center for Financial Services analysis, based on Deloitte Services LP “State of AI in the Enterprise,”
2nd Edition survey, 2018.
Deloitte Insights | deloitte.com/insights
14
Common traits of frontrunners in the artificial intelligence race
FIGURE 11
Top risks of AI that each respondent segment is most concerned about
AI risk Frontrunners Followers Starters
Cybersecurity vulnerabilities of AI/cognitive 1 2 1
Making the wrong strategic decisions based
2 1 3
on AI/cognitive recommendations
Regulatory noncompliance risk 3 3 4
Erosion of customer trust from AI/cognitive
4 5 7
failures
Ethical risks of AI/cognitive 4 4 6
Legal responsibility for decisions/actions
6 5 2
made by AI/cognitive systems
Failure of AI/cognitive system in a
7 7 5
mission-critical or life-and-death context
We have no concerns about potential risks
8 8 8
of AI/cognitive
Note: Respondents were asked to select up to three risks and rank them in order of concern, 1 being that of
highest concern.
Source: Deloitte Center for Financial Services analysis, based on Deloitte Services LP “State of AI in the Enterprise,”
2nd Edition survey, 2018.
Deloitte Insights | deloitte.com/insights
FIGURE 12
Divergence in skills estimation for different segments
No skills gap at all/We have all the skills needed Minimal skills gap Moderate skills gap
Major skills gap Extreme skills gap/We have almost no skills needed
2%
25%15% 12%4% 29% 17% 28%
25%
12%
Frontrunners Followers Starters
18%
30% 30% 53%
More than half of frontrunners believe they have a major or
extreme skills gap in meeting requirements for their AI projects.
Source: Deloitte Center for Financial Services analysis, based on Deloitte Services LP “State of AI in the Enterprise,”
2nd Edition survey, 2018.
Deloitte Insights | deloitte.com/insights
15
AI leaders in financial services
Delving deeper into the capabilities needed to fill User experience could help alleviate the “last mile”
their skills gap, more starters and followers believe challenge of getting executives to take action based
they lack subject matter experts who can infuse on the insights generated from AI. Frontrunners
their expertise into emerging AI systems, as well as seem to have realized that it does not matter how
AI researchers to identify new kinds of AI algo- good the insights generated from AI are if they do
rithms and systems. not lead to any executive action. A good user expe-
rience can get executives to take action by
While these skills are often necessary in the initial integrating the often irrational aspect of human
stages of the AI journey, starters and followers behavior into the design element.
should take note of the skill shortages identified by
frontrunners, which could help them prepare for That said, financial institutions across the board
expanding their own initiatives. Frontrunners sur- should start training their technical staff to create
veyed highlighted a shortage of specialized skill and deploy AI solutions, as well as educate their
sets required for building and rolling out AI imple- entire workforce on the benefits and basics of AI.
mentations—namely, software developers and user The good news here is that more than half of each
experience designers (figure 13). financial services respondent segment are already
undertaking training for employees to use AI in
their jobs.
FIGURE 13
Skills required for implementing AI programs
Skills for AI efforts Frontrunners Followers Starters
Software developers 34% 27% 16%
AI researchers 24% 27% 32%
Data scientists 27% 21% 30%
User experience designers 41% 23% 22%
Business leaders 15% 21% 16%
Project managers 20% 21% 16%
Subject matter experts 15% 25% 38%
Change management/transformation
22% 29% 27%
experts
Source: Deloitte Center for Financial Services analysis, based on Deloitte Services LP “State of AI in the Enterprise,”
2nd Edition survey, 2018.
Deloitte Insights | deloitte.com/insights
16
Common traits of frontrunners in the artificial intelligence race
Getting off to a solid start
AS COMPANIES PREPARE for the AI leg of playing catch-up. Build a strong AI foundation
their digital marathon by revamping their that can be ramped up quickly across
processes and working environments, it is the organization.
imperative they revisit their fundamentals—goals,
strengths, and weaknesses. This could help organi- 2. Get early wins under your belt. Aim at
zations lay a solid foundation for rethinking how generating early success by pursuing a diverse
humans and machines interact within working portfolio of achievable projects. This would help
environments. Answering the following questions generate confidence among management and
could be a good start: the broader workforce at the beginning of the
transformation process, while building momen-
• What existing business goals and missions can tum to roll out more AI initiatives for multiple
the organization achieve by deploying AI? business functions.
• How could AI be used to build a 3. Emphasize organizationwide AI imple-
competitive advantage? mentation once experience (and success)
is achieved. Develop a holistic approach by
• Does the organization have ad |
332 | deloitte | 2024-tmt-outlook-technology.pdf | 2024 technology
industry outlook
2024 technology industry outlook
What’s inside
Executive summary 3
Angling for a comeback, with help from cloud,
AI, and cybersecurity 4
Striking a balance between globalization and self-reliance 6
Setting the stage for growth with generative AI 7
Reckoning with regulations for the tech industry 9
Signposts for the future 11
About Deloitte’s Outlooks
Our 2024 outlook for the technology industry seeks to identify the strategic issues and opportunities for tech organizations to consider in the
coming year, including their impacts, key actions to take, and critical questions to ask. The goal is to help equip US technology organizations
with the information and foresight to better position themselves for a robust and resilient future.
2
2024 technology industry outlook
Executive summary
The technology industry flourished during the early pandemic years might shift or augment their offerings to meet that demand. While
as companies accelerated their digital transformation efforts. But the generative AI has sparked imaginations and headlines, with tech
industry has hit several speed bumps over the past two years. High giants investing billions9 and startups playing a key role,10 enterprise
inflation, elevated interest rates, and considerable macroeconomic and purchasing in this specific category isn’t expected to ramp up
global uncertainties contributed to a softening of consumer spending, until at least the second half of 2024.11 Deloitte expects nearly all
lower product demand, falling market capitalizations, and workforce enterprise software and service companies to integrate generative AI
reductions in 2022.1 Headwinds continued into 2023, with slight capabilities into at least some of their offerings in the coming year.12
weakening of global tech spending and rising layoffs. But there are now
• Striking a balance between globalization and self-reliance.
glimmers of hope that a tech comeback may be imminent: Economists
The worldwide, interconnected nature of the tech industry heightens
have lowered their assessments of recession risk, and analysts are
the risk of disruptions from geopolitical unrest, supply chain
optimistic that the tech sector could return to modest growth in 2024.2
volatility, raw material shortages, and new regulations and policies.
Leaders should diversify their supply chains and manufacturing and
As the technology market faced heightened global challenges over
development locations, spreading operations among trusted regions
the past few years—geopolitical tensions, supply chain volatility, raw
and ensuring redundancy. As governments refine trade policies, tech
material shortages, and emerging regulations—Deloitte urged tech
companies should be agile in adapting their strategies.
leaders to evaluate where manufacturing happens, to improve the
transparency and resiliency of their supply chains, and to prepare • Setting the stage for growth with generative AI. The next
proactively for future systemic risks.3 We suggested leaders use year is expected to be transitional for generative AI, with tech
technology to streamline business processes, rely more on intelligent companies experimenting and finding applications that can drive
automation, reduce tech debt by implementing leading practices efficiency and productivity. Some will likely evaluate how to speed
for software development, and modernize legacy architectures by up software development with generative AI-enabled tools. At the
migrating to cloud resources and XaaS.4 We also recommended same time, providers can determine how to best deliver generative AI
that tech companies consider how to extend their reach into other capabilities and how to monetize them. As generative AI investment
industries, using digital advancements to spur transformation.5 Finally, and experimentation accelerate in the coming months, the legal and
we advised leaders to build up talent in critical areas such as artificial regulatory landscape may evolve rapidly, setting the stage for greater
intelligence (AI), robotic process automation (RPA), and cybersecurity.6 adoption in the second half of 2024 and into 2025.
• Reckoning with regulations for the tech industry.
With global and economic uncertainties continuing into 2024, these
Governments around the world are evaluating the impacts that
recommendations remain important. But it’s likely time to refocus on
massive tech platforms and social networks have on businesses
innovation and growth as well. A Q4 2023 Deloitte survey of 122 tech
and consumers. In the coming months, regulations in the European
executives revealed an optimistic perspective: 55% of respondents
Union and the United States will likely take effect, pushing tech
rated the tech industry as “healthy” or “very healthy,” and even more
companies to prioritize data protection, harm reduction, the ethical
(62%) believe it will be at that level six months from now.7 Asked to
use of AI, and commitment to sustainability goals.13 A global minimum
choose their company’s primary area of focus, “efficiency” topped the
tax aims to close loopholes and push corporations to pay more, while
list (with 25% selecting it), while “innovation” and “productivity” tied for
new credits and incentives are designed to spur sustainable growth
second place at 21%. “Growth” was a close third, at 19%. These leaders
and job creation. With strong collaboration between business,
described the current state of the tech industry as “innovative” and
legal, accounting, and finance teams, tech companies can elevate
“evolving”—and nearly two-thirds (62%) believed it was a good time
compliance efforts into competitive advantages.
for their company to take greater risks.
Each of these themes represents a strategic opportunity for tech
Some of the strategies we expect tech leaders to focus on in 2024
companies to reduce risk and set the stage for sustainable growth
and beyond include:
in the next 12 to 18 months. Prudent investment in supply chain
• Angling for a comeback, with help from cloud, AI, and resilience and data governance may serve as a hedge against
cybersecurity. Enterprise spending on software and IT services— geopolitical and regulatory shifts, while generative AI can streamline
particularly artificial intelligence, cloud computing, and cybersecurity operations in the immediate term and accelerate longer-term
technology—is expected to enable the most growth in the tech innovation efforts.
market over the coming year.8 Tech leaders should assess how they
3
2024 technology industry outlook
Angling for a comeback, with help
from cloud, AI, and cybersecurity
High interest rates, worries about the potential for a slowing economy, continued to contend with lower VC deal activity and valuations—
and geopolitical challenges contributed to a slight weakening of global but Deloitte expects that the valuation corrections may fuel renewed
tech spending in 2023.14 Facing decelerating revenue growth, many interest from venture capitalists and corporate buyers.20 The 2023
tech companies ramped up layoffs last year, continuing to adjust their uptick in tech IPO activity—following a slump since the end of
workforces after aggressive hiring in prior years.15 Now, there may be 2021—could signal the start of a positive trend that allows more
light on the horizon: Economists are more optimistic about the US tech companies to exit successfully.21 While there are some positive
economy as a whole, lowering the risk of a recession in 2024 to below indicators, tech leaders should remain vigilant about ongoing risks
50%. Deloitte’s analysis pegs the risk at just 20%.16 and forge their own careful strategies for growth in 2024.
For the tech sector specifically, analysts are optimistic about a potential What could help drive this tech rebound? Global IT investments are
return to modest growth in 2024, with more robust prospects for 2025. expected to be fueled largely by double-digit growth in spending
Predictions for growth in global IT spending in 2024 cover a range from for software and IT services in 2024.22 Analysts estimate that public
5.7% to 8%.17 cloud spending will grow by more than 20%, and they foresee
stronger demand for cybersecurity.23 AI investment (not specifically
There are signs that aspects of a tech rebound may already be generative AI) is also seen as contributing to overall spending growth.24
underway. In 2023, the stock values of the so-called Magnificent Economists have projected that AI-related investments could reach
Seven—the seven largest US tech companies—surged, outperforming $200 billion globally by 2025, led by the United States.25 Beneficiaries
the rest of the S&P 500 Index.18 The heavily tech-weighted Nasdaq of that spending include companies that create and train AI models,
Composite index took a mere 18 months to recover 80% of its 2021 all- provide infrastructure to run AI (such as cloud services), and supply AI
time high—versus taking 14 years to regain 80% of its 2000 peak after applications or services.26
the dot-com crisis.19 At the smaller end of the spectrum, startups
44
2024 technology industry outlook
A Q4 2023 Deloitte survey of tech executives reinforces the analyst An uptick in tech mergers and acquisitions (M&A) in 2024 would be
viewpoints: The survey asked leaders which technologies they another sign of a tech recovery—but is far from certain. Leaders have
expected to enable the most growth in the tech industry in the traditionally viewed strategic tech M&A as a growth engine, but with
next 12 months.27 Artificial intelligence,28 cloud computing, and the continued high cost of financing and focus on belt-tightening, 2023
cybersecurity topped the list (with 52%, 47%, and 40% of tech proved disappointing for tech M&A. Deal volume and the total market
leaders choosing each as a top-three technology, respectively).29 value of those transactions remain well below their 2021 highs.36 On
the bright side, a handful of billion-dollar-plus enterprise tech deals has
What about generative AI, which has grabbed headlines, captured given analysts a reason to hope that the tide may turn for tech M&A
the attention of tech leaders, and fueled a notable wave of in the coming months.37 Technology geared toward productivity and
experimentation over the past year (see “Setting the stage for growth efficiency improvement—including industrial automation and decision
with generative AI”)? Deloitte expects generative AI growth in 2024 to intelligence platforms—is seen as having the potential to spark
be modest, with adoption and spending picking up in the second half, renewed M&A activity.38 AI may also prove to be a driver: Companies
followed by more robust growth in 2025.30 Tech execs seem bullish may obtain AI technology and expertise via acquisitions, rather than
about imminent generative AI spending: More than a quarter (27%) building their own capabilities.39
of respondents to the Deloitte survey selected generative AI as a
top-three industry growth driver for the coming year.31 Perhaps due
to the level of investment or effort required for full-scale generative
AI initiatives, respondents from larger companies (those with 10,000+
employees or US$10B+ in revenue) selected generative AI at a higher Strategic questions to consider:
rate than others. Notably, tech giants with plans to invest billions
in generative AI will likely play a part in the sector’s rebound.32 • How will our company navigate the evolving economic landscape,
continued high cost of borrowing, and ongoing geopolitical
challenges while still achieving our growth objectives?
Cybersecurity is also expected to play a key role in the comeback.
Analysts forecast that global spending on security and risk • Has our company evaluated how adopting AI—specifically
management will see low double-digit growth from 2023 to 2024.33 generative AI—might help us drive productivity and efficiency
gains? Have we considered how embedding generative AI
Motivators include a persistent threat landscape, ongoing cloud
capabilities into our products and services could help drive
adoption, remote work, the emergence of generative AI, and evolving
revenue and competitive advantage?
data privacy and governance regulations.34 While the rapid adoption
• Is our company continually assessing the security threat
of generative AI may expose organizations to new attack surfaces
landscape and keeping up with the latest advances in security
and techniques, AI may also play a pivotal role in speeding up
and risk management? Are we considering how AI could play
breach detection and containment.35 a role in helping us boost our defenses?
• How can we ensure that our workforce has the right mix of
skills for competitive success? Are we focused on building
expertise in growth areas, especially cloud, generative AI,
and cybersecurity?
• Are we considering how strategic acquisitions could
complement our existing capabilities, help us innovate,
expand our market reach, and even augment our talent?
5
2024 technology industry outlook
Striking a balance between
globalization and self-reliance
The worldwide, interconnected nature of the tech industry, with crises underscore the risks of over-relying on tech talent in any one
its global supply chains and international manufacturing and location.48 Leaders should consider expanding their workforce in
development centers, makes it highly vulnerable to global shocks secure regions and taking care that pivotal functions and roles are
including natural disasters, pandemics, and geopolitical tensions.40 distributed. Tech jobs have specialized training and educational
needs that continue to evolve due to advancements such as
Supply chain resilience is no longer simply prudent; it’s critical. generative AI. Redistributing talent pools likely means partnering
To help mitigate the risk of disruption, tech giants are diversifying with universities and engineering schools; working more closely with
their manufacturing and development locations and supply chains, local tech schools, vocational schools, and community colleges; and
reducing reliance on single suppliers or countries.41 Leaders often supporting national institutions that promote STEM fields.49
now view it as imperative to establish relationships with suppliers
worldwide and spread operations across various trusted regions. All Tech companies may be able to boost resilience in their operations
critical product components and elements of the value chain should and supply chains by co-investment and knowledge-sharing initiatives
have redundancies and alternate sources. Moreover, tech companies with channel partners, contract manufacturers, and suppliers. This
should work closely with suppliers to ensure resilience and flexibility could involve helping suppliers with approvals and logistics as they
throughout the production network. work to establish facilities in different regions, as well as offering
essential talent, engineering, and administrative proficiency as they
As the geopolitical landscape continues to shift, governments spin up new operations.
worldwide are redefining their trade policies. Tech companies should
monitor these changes and align their strategies accordingly. Countries Throughout 2024, tech companies will likely continue to prioritize
and trading blocs often offer incentives, subsidies, and tax credits to sustainability and resilience, aiming to strike the right balance
encourage the localization of technology supply chains and innovation between globalization and onshoring/self-reliance. Organizations
hubs.42 This trend is particularly evident in the semiconductor should continue to globalize their operations to take advantage of
industry, where the United States and Europe are making substantial lower costs, greater access to talent, and faster innovation. However,
investments to build out domestic chip fabrication capacity, especially they should also look to onshore or self-source critical components
at advanced processing nodes. They’re also ramping up assembly and and operations to reduce their risks from global disruptions.
packaging capabilities, although from a low base.43 Strategic planning
should include sustainability assessments, tracking, and reporting,
both to secure maximum credit and ensure compliance with local
and international regulations.44
Strategic questions to consider:
After severe chip shortages began in 2020, the US government
passed the CHIPS and Science Act of 2022—which provides $52 • Has our company adequately evaluated our supply chain
billion in financial assistance to spur research and manufacturing and operational vulnerabilities? Do we have a strategy for
mitigating them?
in the domestic semiconductor industry.45 Semiconductor
manufacturers are working to identify which parts of their supply • Is our company’s supply chain designed for flexibility in the
chain should be domestic (onshoring), which parts can be in short term and sustainability in the long term? Have we
implemented multi-sourcing strategies to ensure a stable
countries close to home (nearshoring), and which parts can be
supply chain?
handled in countries considered to be allies (friendshoring).46
For some tech companies, particularly hardware and electronics • Have we determined the right blend of onshoring,
nearshoring, and friendshoring?
manufacturers, full onshoring may be impractical or infeasible—
but a blend of onshoring and friendshoring could help provide • Have we assessed the stability of our onshore and global talent
a hedge against instability.47 pools, ensuring that critical functions are not concentrated in
vulnerable regions? Is there a way to distribute our tech talent
to make it more resilient to global disruptions?
As organizations identify potential choke points and determine how
to reengineer their operations and processes to improve resilience,
they may also focus on building redundancy into their research
and development operations and talent pool. Recent geopolitical
6
2024 technology industry outlook
Setting the stage for growth
with generative AI
Over the past year, generative AI sparked the public imagination, Tech leaders should consider how to best utilize and deliver this new
unleashed new avenues for creativity, fueled a surge of startups, functionality. This could involve using “off the shelf” solutions from
and became a strategic consideration for many of the world’s largest cloud and tech providers with generative AI integrated, building their
companies. The next year is poised to be a time of transition, with own proprietary solutions (which could be prohibitively expensive),
tech leaders assessing how to best deliver and monetize generative or partnering with co-developers.
AI, how to integrate the technology into their operations, and how
to address considerable challenges around data privacy, copyright, Tech companies may use all these approaches to incorporate
and emerging regulations. generative AI into existing or new offerings. One possibility is that AI
solutions will evolve into an ecosystem where large players provide
Innovating with generative AI foundational platforms and contextual models as commodities,
In the past year, US tech companies focused intensely on generative while additional parties build capabilities and functions on top to
AI, embedding it into their offerings and signaling plans to double cater to the specific needs of their customers.58
down on investments.50 Across the sector, many tech companies will
face the challenge of how to augment their products and services Focusing on productivity
with generative AI to remain competitive. On the software front, Like their counterparts in other industries, many tech leaders
Deloitte has predicted that nearly all enterprise software companies are experimenting with embedding generative AI capabilities into
will embed generative AI into at least some of their products in their workflows to assist professionals and augment business
2024 and that the revenue uplift (for these companies and for the processes.59 At this stage, many are focused on optimizing
cloud providers of generative AI processing capacity) will approach productivity and efficiency. A recent Deloitte survey of marketing
a US$10 billion run rate by the end of the year.51 Deloitte expects leaders found that 26% already use generative AI (e.g., for content
2024 to be a transition year, as generative AI-enabled software tools marketing), and another 45% expect to use it by the end of 2024.
launch and adoption and revenues start to gain traction, setting the Users estimated the technology has saved them more than 11 hours
stage for more robust potential growth in 2025.52 On the hardware per week.60
front, Deloitte expects the uplift for chips and servers that execute
generative AI to surpass US$50 billion in 2024.53 Generative AI is being used to facilitate sales—from interpreting
customer requirements documents to developing proposals and
Several tech companies associated with generative AI experienced prioritizing leads—and to improve customer service (e.g., helping
rising valuations in 2023, partially due to excitement around the human agents respond to questions and solve problems and even
technology’s potential.54 However, they’re still figuring out how anticipating customers’ future needs).61 Research has revealed
to monetize and profit from generative AI. Deploying and scaling that more than eight in 10 sales professionals surveyed feel using
generative AI involves heavy-duty servers packed with expensive, generative AI helps them speed up customer communication and
power-hungry chips, and the operational costs can range from US$0.01 increase sales, while nine in 10 service professionals believe it
to US$0.36 per generative AI query.55 Some providers who charge helps them address customer needs more quickly.62 Companies
a per user per month (PUPM) fee may be losing money currently are also driving back-office efficiency by embedding generative AI
due to those who use the service more heavily than anticipated.56 capabilities into functions such as finance and order management—
We expect that tech companies will continue to grapple with how to accelerating processes, reporting, and insights. Tech leaders should
translate generative AI into increased revenue, experimenting with consider where to adopt generative AI in their company to best
a variety of pricing models (such as consumption-based, PUPM, improve productivity and how they might use it to improve customer
or a hybrid approach).57 interactions and enhance tech support.
7
2024 technology industry outlook
Particularly important for tech companies, generative AI tools are International regulations governing privacy, potential harm, and
boosting programmer productivity and may be on the verge of ethical practices are also high on the list of concerns for generative
transforming software development.63 These tools can act as coding AI adopters. The EU’s AI Act, for instance, is expected to be adopted
and testing partners, suggesting lines of code, developing boilerplate in the second quarter of 2024, with a 24-month implementation
code, writing documentation, generating synthetic test data, and period for most obligations.72 US companies are working to comply
creating test cases.64 A survey of professional developers found that with the Biden administration’s October 2023 executive order
44% are already using AI tools in their development process, and governing the safe and secure development and use of AI.73 The
another 26% plan to do so soon.65 With productivity gains reported order will require certain developers of “very powerful” foundation
in the 10% to 30% range, tech leaders should evaluate where they models to share safety test results with the government. It will also
can bring generative AI into their development processes.66 impose requirements for federal agencies, including the use of
watermarks to identify AI-generated content, measures to protect
Adopters that are further along in their evaluations and may have user privacy, and efforts to minimize bias (see “Reckoning with
completed successful pilots with generative AI will likely focus on regulations for the tech industry”).
the challenge of scaling up and operationalizing the technology.67
Moving to production use will likely involve prioritizing highest-
value use cases, mapping them to core capabilities required for
implementation, and developing an implementation road map.
Strategic questions to consider:
Avoiding legal and regulatory pitfalls
The use of generative AI raises considerable challenges around • Have we determined which use cases and workflows could
data privacy and content use. One area of concern for tech leaders be best improved with generative AI? Have we assessed
where we could deploy generative AI in our value chain?
is whether the large language models (LLMs) used in generative
AI implementations have been trained using copyright-protected • Are we evaluating how generative AI could create
content.68 To address concerns, several leading software companies opportunities for new products, services, business models,
and, ultimately, new revenues?
have pledged to assume liability in case their tools expose customers
to IP infringement claims.69 Another misgiving is whether a company • Does our workforce have the right set of skills for upcoming
might lose control of its own data when it’s added to public models, generative AI initiatives? For example, have we considered
training existing staff to improve generative AI literacy? Are
whether through accidental data leaks or adversarial prompt
we recruiting the right talent?
engineering.70 As a result, Deloitte expects that more companies will
begin training generative AI on their private enterprise data—but • How will the changing legal and regulatory landscape affect
our generative AI plans? Are we setting the right guardrails on
this approach could raise challenges around access to talent and
our generative AI initiatives?
specialized GPUs.71 Generative AI adopters should weigh the risks of
publicly trained models and the expense and expertise required for
building proprietary models as they decide which approach is right
for their company.
8
2024 technology industry outlook
Reckoning with regulations
for the tech industry
Large online platforms built up enormous power and influence over
the past decade, and regulators are considering how to best address
the potential risks. Tech companies of every size are under pressure
to ensure data protection, harm reduction, ethical use of AI, and
commitment to sustainability goals. They’re also tasked with pivoting
to maximize tax credits and incentives while minimizing effective tax
rates in the face of new global tax regulations.
Content and corporate conduct
The largest tech companies are affected by the European Union’s
Digital Services Act (DSA) and Digital Markets Act (DMA). The DSA
includes a raft of new rules around consumer protection, holding
online platforms and service providers responsible for content
moderation, fraud, and unscrupulous uses of their technologies.74
It also imposes stringent requirements on consumer-facing tech
companies that collect customer data.75
The DMA requires platforms to eliminate practices that stifle
competition, including granting third-party businesses and
advertisers more access to data and allowing them more freedom
to engage customers outside the platform. Some of the “large
platforms” identified by the EU are challenging their designations
in court.76
AI everywhere
The proliferation of AI has also spurred a new wave of regulatory
developments.77 Expected to begin taking effect in 2024, the EU’s
AI Act—which is nearly finalized—takes a risk-based approach to
AI implementations, requiring visibility into the quality of data
sets used, technical documentation and recordkeeping, human
oversight, accuracy, and cybersecurity.78 It applies to any AI system
that outputs results used in the EU, and it is expected to impact AI
providers in the United States.79
In the United States, President Biden signed an executive order
on October 30, 2023, that seeks to promote the safe and secure
development and use of AI and creates requirements related to
the use of AI throughout the federal government. The executive
order directs the development of both voluntary and mandatory
guidance to govern the use of AI in the public and private sectors.
It includes more than 100 directives to agencies, which will mostly
be implemented over the next year. The Commerce Department
will play an important role in implementation and has formed
a US AI Safety Institute to help develop technical guidance for
other agencies.80
9
2024 technology industry outlook
Global tax equality which was signed into law in fall 2023, requires climate disclosures
Another factor that tech companies will likely encounter in 2024 is the and climate-related financial risk reporting from any company with
OECD’s Pillar Two global minimum tax (GMT). Some countries have revenues greater than US$1 billion doing business in California.85
passed legislation and many others are proposing legislation to activate
these rules. This ruleset is designed to ensure that multinational Taken together, these developments could drive increased
corporations pay a minimum of 15% regardless of location, removing investment in cybersecurity, data management, and ESG reporting
the incentive to locate headquarters in low-tax jurisdictions. solutions. Tech companies will likely benefit by working with
regulators and taking an active role in testing their products
Certain factors, including credits and incentives, may bring the effective and services for compliance.
tax rate in a country below 15%, in which case these companies will
have to pay a “top-up” tax to meet the 15% threshold. This may reduce
or eliminate the benefit of the incentive.81
For the tech industry, the way different jurisdictions operationalize Strategic questions to consider:
these rules and how they define and treat credits and incentives may
lead to operational shifts. Countries may jockey to build out incentive • How can we ensure that our AI implementations don’t expose
programs that don’t have an impact on effective tax rates. the company to potential regulatory and legal risk?
• What investments should we explore in cybersecurity and
ESG credits and compliance data governance to achieve compliance with new
No regulatory outlook would be complete without a discussion of consumer-protection regulations?
environmental, social, and governance (ESG) reporting requirements. • Can we leverage regulatory sandboxes to test our products
The EU’s Corporate Sustainability Reporting Directive (CSRD) and services?
expands the number of companies required to provide sustainability
• How can we model potential tax scenarios now to inform
disclosures from around 12,000 to more than 50,000.82 It also operational decisions for 2024–2025?
imposes requirements around double materiality; companies must
• How can we maximize ESG credits and incentives while
report the impacts that ESG efforts have on their businesses and the
preserving our effective tax rate?
impacts they’re expected to have on the environment, human rights,
social standards, and sustainability-related risk.83 These rules apply
to multinational entities (like tech giants) that meet certain revenue
criteria. European branches of these companies may have to provide
consolidated reporting on their parent company’s activities as well.
In the United States, the Federal Acquisition Regulatory Council
has proposed a rule that would require certain federal contractors
to disclose their greenhouse gas (GHG) emissions, as well as their
climate-related financial risk, and set science-based targets to
reduce their emissions.84 California’s Climate Accountability Package,
10
2024 technology industry outlook
Signposts for the future
2024 finds the technology industry preparing for a return
to growth. Tech companies may protect themselves
against future global disruptions by engineering a balance
between globalization and self-reliance, and they’re
gearing up for a raft of expected regulations. Preparations
will likely involve doubling down on data governance,
cybersecurity, and supply chain resilience. At the same
time, tech companies are looking at generative AI as a way
to achieve greater efficiency in the near term—and as a
way to fuel innovation and growth for themselves and
other industries in the long term.
In the coming year, tech companies should be on
the lookout for potential signals of change in the
market, including:
• Shifting macroeconomic co |
333 | deloitte | DI_Tech-trends-2025.pdf | i
5202
sdnerT
hceT
Tech Trends 2025 In Deloitte’s 16th annual Tech Trends report,
AI is the common thread of nearly every
trend. Moving forward, it will be part of the
substructure of everything we do.
02 . . . Executive summary
INTRODUCTION
05 . . . AI everywhere: Like magic, but with algorithms
INTERACTION
09 . . . Spatial computing takes center stage
INFORMATION
17 . . . What’s next for AI?
COMPUTATION
27 . . . Hardware is eating the world
BUSINESS OF TECHNOLOGY
37 . . . IT, amplified: AI elevates the reach (and remit) of the tech function
CYBER AND TRUST
45 . . . The new math: Solving cryptography in an age of quantum
CORE MODERNIZATION
53 . . . The intelligent core: AI changes everything for core modernization
CONCLUSION
60 . . . Breadth is the new depth: The power of intentional intersections
stnetnoc
fo
elbaT
Executive summary
Tech Trends, Deloitte’s flagship technology report, as electricity to daily business and personal lives. As
explores the emergence of trends in three elevating forces our team in Deloitte’s Office of the CTO put finishing
(interaction, information, and computation) and three touches on Tech Trends 2025, we realized that AI is a
grounding forces (business of technology, cyber and common thread in nearly every trend. We expect that
trust, and core modernization)—all part of our macro going forward, AI will be so ubiquitous that it will be a
technology forces framework (figure 1). Tech Trends part of the unseen substructure of everything we do, and
2025, our 16th trip around the sun, previews a future we eventually won’t even know it’s there.
in which artificial intelligence will be as foundational
Figure 1
Six macro forces of information
technology
INFORMATION
What’s next for AI?
INTERACTION COMPUTATION
Spatial computing Hardware is
takes center stage eating the world
BUSINESS OF CORE
TECHNOLOGY MODERNIZATION
IT, amplified: The intelligent core:
AI elevates the reach CYBER AI changes
(and remit) of the AND everything for core
tech function modernization
TRUST
The new math:
Solving
cryptography in an
age of quantum
2
3
yrammus
evitucexE
Introduction efficient choice for all organizational needs. Enterprises
are now considering small language models and open-
AI everywhere: Like magic, but with algorithms source options for the ability to train LLMs on smaller,
more accurate data sets. Together with multimodal
Generative AI continues to be the buzzword of the models and AI-based simulations, these new types of
year, but Tech Trends 2025—and in fact, the future of AI are building a future where enterprises can find the
technology—is about much more than AI. This year’s right type of AI for each task. That includes AI that not
report reveals the extent to which AI is being woven only answers questions but also completes tasks. In the
into the fabric of our lives. We’ll eventually take it for coming years, a focus on execution may usher in a new
granted and think of it in the same way that we think of era of agentic AI, arming consumers and organizations
HTTP or electricity: We’ll just expect it to work. AI will with co-pilots capable of transforming how we work
perform quietly in the background, optimizing traffic and live.
in our cities, personalizing our health care, or creating
adaptative and accessible learning paths in education.
We won’t proactively use it; we’ll simply experience a Computation
world in which it makes everything work smarter, faster,
and more intuitively—like magic, but grounded in algo- Hardware is eating the world
rithms. The six chapters of Tech Trends 2025 reflect this
emerging reality. After years of software dominance, hardware is reclaim-
ing the spotlight. As AI demands specialized computing
resources, companies are turning to advanced chips to
Interaction power AI workloads. In addition, personal comput-
ers embedded with AI chips are poised to supercharge
Spatial computing takes center stage knowledge workers by providing access to offline AI
models while “future-proofing” technology infrastruc-
Spatial computing continues to spark enterprise interest ture, reducing cloud computing costs, and enhancing
because of its ability to break down information silos and data privacy. Although AI’s increased energy demands
create more natural ways for workers and customers to pose sustainability challenges, advancements in energy
interact with information. We’re already seeing enter- sources and efficiency are making AI hardware more
prises find success with use cases like advanced simula- accessible. Looking forward, AI’s continued integration
tions that allow organizations to test different scenarios into devices could revolutionize the Internet of Things
to see how various conditions will impact their oper- and robotics, transforming industries like health care
ations. With a stronger focus on effectively managing through smarter, more autonomous devices.
spatial data, organizations can drive more cutting-edge
applications. In the coming years, advancements in AI
could lead to seamless spatial computing experiences Business of technology
and improved interoperability, ultimately enabling AI
agents to anticipate and proactively meet users’ needs. IT, amplified: AI elevates the reach
(and remit) of tech talent
Information After years of progressing toward lean IT and every-
thing-as-a-service offerings, AI is sparking a shift away
What’s next for AI? from virtualization and austere budgets. Long viewed
as the lighthouse of digital transformation through-
To take advantage of the burgeoning excitement around out the enterprise, the IT function is now taking on AI
generative AI, many organizations have already adopted transformation. Because of generative AI’s applicability
large language models (LLMs), the best option for many to writing code, testing software, and augmenting tech
use cases. But some are already looking ahead. Despite talent in general, forward-thinking technology leaders
their general applicability, LLMs may not be the most are using the current moment as a once-in-a-blue-moon
opportunity to transform IT across five pillars: infra- core enterprise systems represents a significant shift in
structure, engineering, finance operations, talent, and how organizations operate and leverage technology for
innovation. As both traditional and generative AI capa- competitive advantage. This transformation is about
bilities grow, every phase of tech delivery could see a automating routine tasks and fundamentally rethinking
shift from human in charge to human in the loop. Such a and redesigning processes to be more intelligent, efficient,
move could eventually return IT to a new form of lean IT, and predictive. It requires careful planning due to inte-
leveraging citizen developers and AI-driven automation. gration complexity, strategic investment in technology
and skills, and a robust governance framework to ensure
smooth operations. But beware of the automation para-
Cyber and trust dox: The more complexity is added to a system, the more
vital human workers become. Adding AI to core systems
The new math: Solving cryptography may simplify the user experience, but it will make them
in an age of quantum more complex at an architectural level. Deep technical
skills are still critical for managing AI in core systems.
In their response to Y2K, organizations saw a loom-
ing risk and addressed it promptly. Today, IT faces a
new challenge, and it will have to respond in a similarly Conclusion
proactive manner. Experts predict that quantum comput-
ers, which could mature within five to 20 years, will Breadth is the new depth: The power
have significant implications for cybersecurity because of intentional intersections
of their ability to break existing encryption methods and
digital signatures. This poses a risk to the integrity and Organizations have long relied on innovation-driven new
authenticity of data and communications. Despite the revenue streams, synergies created through mergers and
uncertainty of the quantum computer timeline, inaction acquisitions, and strategic partnerships. But increasingly,
on post-quantum encryption is not an option. Emerging segmentation and specialization have given way to inten-
encryption standards offer a path to mitigation. Updating tional intersections of technologies and industries. For
encryption practices is fairly straightforward—but it’s a example, when two technologies intersect, they are often
lengthy process, so organizations should act now to stay complementary, but they can also augment each other so
ahead of potential threats. And while they’re at it, they that both technologies ultimately accelerate their growth
can consider tackling broader issues surrounding cyber potential. Similarly, new opportunities can emerge when
hygiene and cryptographic agility. companies aim to extend their market share by purpose-
fully partnering across seemingly disparate industries.
Core modernization
The intelligent core: AI changes
everything for core modernization
Core systems providers have invested heavily in AI,
rebuilding their offerings and capabilities around an
AI-fueled or AI-first model. The integration of AI into
4
5
smhtirogla
htiw
tub
,cigam
ekiL
:erehwyreve
IA
INTRODUCTION
AI everywhere: Like magic,
but with algorithms
Tech Trends 2025 reveals how much artificial intelligence is being woven into the
fabric of our lives—making everything work smarter, faster, and more intuitively
Kelly Raskovich
Two years after generative artificial intelligence staked Nowhere is this AI-infused future more evident than in
its claim as the free space on everyone’s buzzword-bingo this year’s Tech Trends report, which each year explores
cards, you’d be forgiven for imagining that the future of emerging trends across the six macro forces of informa-
technology is simply … more AI. That’s only part of the tion technology (figure 1 in the executive summary).
story, though. We propose that the future of technology Half of the trends that we’ve chronicled are elevating
isn’t so much about more AI as it is about ubiquitous forces—interaction, information, and computation—that
AI. We expect that, going forward, AI will become so underpin innovation and growth. The other half—the
fundamentally woven into the fabric of our lives that it’s grounding forces of the business of technology, cyber
everywhere, and so foundational that we stop noticing it. and trust, and core modernization—help enterprises
seamlessly operate while they grow.
Take electricity, for example. When was the last time you
actually thought about electrons? We no longer marvel As our team put the finishing touches on this year’s
that the lights turn on—we simply expect them to work. report, we realized that this sublimation and diffu-
The same goes for HTTP, the unseen thread that holds sion of AI is already afoot. Not the “only trend” nor
the internet together. We use it every day, but I’d bet “every trend,” AI is the scaffolding and common thread
most of us haven’t thought about (let alone uttered) the buttressing nearly every trend. (For those keeping a close
word “hypertext” in quite some time. eye at home, “The new math: Solving cryptography in
an age of quantum”—about the cybersecurity implica-
AI will eventually follow a similar path, becoming so tions of another game-changing technology, quantum
ubiquitous that it will be a part of the unseen substruc- computing—is the only one in which AI does not have
ture of everything we do, and we eventually won’t even a foundational role. Yet behind the scenes, AI advance-
know it’s there. It will quietly hum along in the back- ments are accelerating advances in quantum.)
ground, optimizing traffic in our cities, personalizing our
health care, and creating adaptative and accessible learn- • Spatial computing takes center stage: Future AI
ing paths in education. We won’t “use” AI. We’ll just advancements will enhance spatial-computing simu-
experience a world where things work smarter, faster, lations, eventually leading to seamless spatial-com-
and more intuitively—like magic, but grounded in algo- puting experiences integrated with AI agents.
rithms. We expect that it will provide a foundation for
business and personal growth while also adapting and • What’s next for AI?: As AI evolves, the enterprise
sustaining itself over time. focus on large language models is giving way
to small language models, multimodal models,
AI-based simulations, and agents that can execute
discrete tasks.
• Hardware is eating the world: After years of soft- Because we expect AI to become part of tomorrow’s
ware dominance, hardware is reclaiming the spot- foundational core—like electricity, HTTP, and so many
light, largely due to AI’s impact on computing chips other technologies—it’s exciting to think about how AI
and its integration into end-user devices, the Internet might evolve in the next few years as it marches toward
of Things, and robotics. ubiquity, and how we as humans may benefit. We here at
Tech Trends will be chronicling every step of the journey.
• IT, amplified: AI elevates the reach (and remit) of
tech talent: AI’s applicability to writing code, testing Until next time,
software, and augmenting tech talent is transform-
ing IT and sparking a shift away from virtualization
and austere budgets.
Kelly Raskovich
• The intelligent core: AI changes everything for Office of the CTO
core modernization: Core systems providers have Executive editor, Tech Trends
invested heavily in AI, which may simplify the user
experience and data-sharing across applications
but will make these systems more complex at an
architectural level.
6
7
smhtirogla
htiw
tub
,cigam
ekiL
:erehwyreve
IA
Trending the trends
INTERACTION INFORMATION COMPUTATION BUSINESS OF CYBER CORE
TECHNOLOGY AND TRUST MODERNIZATION
Spatial
Hardware
2025 computing What’s next is eating IT, amplified The new math The intelligent
takes center for AI? core
the world
stage
2024 Interfaces in Genie out of Smarter, not From DevOps Defending Core workout
new places the bottle harder to DevEx reality
2023 Through the Opening up Above the Flexibility, In us we trust Connect and
glass to AI clouds the best ability extend
Blockchain: DEI tech:
2022 Data sharing Ready for Cloud goes Tools for The tech stack Cyber AI IT, disrupt
made easy vertical goes physical thyself
business equity
Rebooting the ML Ops:
2021 digital Bespoke for Machine data Industrialized Strategy, Supply Zero trust Core revival
billions revolution engineered unchained
workplace AI
Human Finance and Ethical
2020 experience Digital twins the future of Architecture technology
awakens
platforms IT and trust
NoOps in a DevSecOps
2019 Intelligent Beyond AI-fueled serverless Connectivity and the cyber
interfaces marketing organizations of tomorrow
world imperative
Enterprise
2018 Digital reality data API Blockchain to No-collar Reengineering The new
imperative blockchains workforce technology core
sovereignty
2017 Mixed reality Dark analytics Machine Everything Trust economy IT Inevitable
intelligence as-a-service unbounded architecture
2016 Internet of AR and VR Industrialized Democratized Right speed IT Autonomic Reimagining
Things go to work analytics trust platforms core systems
Note: To learn more about past Tech Trends, go to www.deloitte.com/us/TechTrends
Source: Deloitte analysis.
8
9
egats
retnec
sekat
gnitupmoc
laitapS
INTERACTION
Spatial computing takes center stage
What is the future of spatial computing? With real-time simulations as just the
start, new, exciting use cases can reshape industries ranging from health care
to entertainment.
Kelly Raskovich, Bill Briggs, Mike Bechtel, and Ed Burns
Today’s ways of working demand deep expertise in narrow If eye-catching virtual reality (VR) headsets are the first
skill sets. Being informed about projects often requires thing that come to mind when you think about spatial
significant specialized training and understanding of computing, you’re not alone. But spatial computing is
context, which can burden workers and keep information about more than providing a visual experience via a
siloed. This has historically been true especially for any pair of goggles. It also involves blending standard busi-
workflow involving a physical component. Specialized ness sensor data with the Internet of Things, drone, light
tasks demanded narrow training in a variety of unique detection and ranging (LIDAR), image, video, and other
systems, which made it hard to work across disciplines. three-dimensional data types to create digital representa-
tions of business operations that mirror the real world.
One example is computer-aided design (CAD) software. These models can be rendered across a range of inter-
An experienced designer or engineer can view a CAD file action media, whether a traditional two-dimensional
and glean much information about the project. But those screen, lightweight augmented reality glasses, or full-on
outside of the design and engineering realm—whether immersive VR environments.
they’re in marketing, finance, supply chain, project
management, or any other role that needs to be up to Spatial computing senses real-world, physical compo-
speed on the details of the work—will likely struggle nents; uses bridging technology to connect physical
to understand the file, which keeps essential technical and digital inputs; and overlays digital outputs onto a
details buried. blended interface (figure 1).2
Spatial computing is one approach that can aid this type Spatial computing’s current applications are as diverse
of collaboration. As discussed in Tech Trends 2024, as they are transformative. Real-time simulations have
spatial computing offers new ways to contextualize emerged as the technology’s primary use case. Looking
business data, engage customers and workers, and inter- ahead, advancements will continue to drive new and
act with digital systems. It more seamlessly blends the exciting use cases, reshaping industries such as health
physical and digital, creating an immersive technology care, manufacturing, logistics, and entertainment—
ecosystem for humans to more naturally interact with which is why the market is projected to grow at a rate
the world.1 For example, a visual interaction layer that of 18.2% between 2022 and 2033.3 The journey from
pulls together contextual data from business software the present to the future of human-computer interaction
can allow supply chain workers to identify parts that promises to fundamentally alter how we perceive and
need to be ordered and enable marketers to grasp a prod- interact with the digital and physical worlds.
uct’s overall aesthetics to help them build campaigns.
Employees across the organization can make meaning of
and, in turn, make decisions with detailed information
about a project in ways anyone can understand.
Figure 1
The possibilities of spatial operations
Physical Bridging Digital
Wearables (for example, headset, Sensors (for example, LIDAR)
Augmented reality objects
smart eyewear, and pins) and sensor fusion
Next-gen displays Computer vision Interactive digital objects
Internet of Things devices
GPS/spatial mapping software Holographic projections
(for example, biometric devices)
Sensory tech
3D design and rendering tools Audio outputs
(for example, haptic suits)
Comprehensive next-gen
Spatial audio devices Avatars
network infrastructure
Cameras Data lakes Generative AI
Next-gen batteries
Source: Abhijith Ravinutala et al., “Dichotomies spatial computing: Navigating towards a better future,” Deloitte, April 22, 2024.
Now: Filled to the rim with sims One of the primary applications unlocked by spatial
computing is advanced simulations. Think digital twins,
At its heart, spatial computing brings the digital world but rather than virtual representations that monitor
closer to lived reality. Many business processes have a physical assets, these simulations allow organizations
physical component, particularly in asset-heavy indus- to test different scenarios to see how various conditions
tries, but, too often, information about those processes will impact their operations.
is abstracted, and the essence (and insight) is lost.
Businesses can learn much about their operations from Imagine a manufacturing company where designers,
well-organized, structured business data, but adding engineers, and supply chain teams can seamlessly work
physical data can help them understand those operations from a single 3D model to craft, build, and procure all
more deeply. That’s where spatial computing comes in. the parts they need; doctors who can view true-to-life
simulations of their patients’ bodies through augmented
“This idea of being served the right information at reality displays; or an oil and gas company that can layer
the right time with the right view is the promise of detailed engineering models on top of 2D maps. The
spatial computing,” says David Randle, global head possibilities are as vast as our physical world is varied.
of go-to-market for spatial computing at Amazon Web
Services (AWS). “We believe spatial computing enables The Portuguese soccer club Benfica’s sports data science
more natural understanding and awareness of physical team uses cameras and computer vision to track players
and virtual worlds.”4
10
11
egats
retnec
sekat
gnitupmoc
laitapS
throughout matches and develop full-scale 3D models New: Data is the differentiator
of every move its players make. The cameras collect
2,000 data points from each player, and AI helps identify Enterprise IT teams will likely need to overcome signifi-
specific players, the direction they were facing, and criti- cant hurdles to develop altogether-new spatial comput-
cal factors that fed into their decision-making. The data ing applications. They likely haven’t faced these hurdles
essentially creates a digital twin of each player, allowing when implementing more conventional software-based
the team to run simulations of how plays would have projects. While these projects have compelling busi-
worked if a player was in a different position. X’s and ness value, organizations will have to navigate some
O’s on a chalkboard are now three-dimensional models uncharted waters to achieve them.
that coaches can experiment with.5
For one thing, data isn’t always interoperable between
“There’s been a huge evolution in AI pushing these systems, which limits the ability to blend data from
models forward, and now we can use them in deci- different sources. Furthermore, the spaghetti diagrams
sion-making,” says Joao Copeto, chief information and mapping out the path that data travels in most organi-
technology officer at Sport Lisboa e Benfica.6 zations are circuitous at best, and building the data pipe-
lines to get the correct spatial data into visual systems
This isn’t only about wins and losses—it’s also about is a thorny engineering challenge. Ensuring that data is
dollars and cents. Benfica has turned player development of high quality and faithfully mirrors real-world condi-
into a profitable business by leveraging data and AI. tions may be one of the most significant barriers to using
Over the past 10 years, the team has generated some spatial computing effectively.9
of the highest player-transfer deals in Europe. Similar
approaches could also pay dividends in warehouse oper- Randle of AWS says spatial data has not historically been
ations, supply chain and logistics, or any other resource well managed at most organizations, even though it
planning process. represents some of a business’s most valuable information.
Advanced simulations are also showing up in medical “This information, because it’s quite new and diverse,
settings. For instance, virtual patient scenarios can be has few standards around it and much of it sits in silos,
simulated as a training supplement for nurses or doctors some of it’s in the cloud, most of it’s not,” says Randle.
in a more dynamic, self-paced environment than text- “This data landscape encompassing physical and digital
books would allow. This may come with several chal- assets is extremely scattered and not well managed. Our
lenges, such as patient data concerns, integration of AI customers’ first problem is managing their spatial data.”10
into existing learning materials, and the question of
realism. But AI-based simulations are poised to impact Taking a more systematic approach to ingesting, orga-
the way we learn.7 nizing, and storing this data, in turn, makes it more
available to modern AI tools, and that’s where the real
Simulations are also starting to impact health care learnings begin.
delivery. Fraser Health Authority in Canada has been a
pioneer in leveraging simulation models to improve care.8 Data pipelines deliver the fuel that drives business
By creating a first-of-its-kind system-wide digital twin,
the public health authority in British Columbia generated We’ve often heard that data is the new oil, but for an
powerful visualizations of patient movement through American oil and gas company, the metaphor is becom-
different care settings and simulations to determine the ing reality thanks to significant effort in replumbing some
impact of deploying different care models on patient of its data pipelines.
access. Although the work is ongoing, Fraser expects
improvement in appropriate, need-based access to care The energy company uses drones to conduct 3D scans of
through increased patient awareness of available services. equipment in the field and its facilities, and then applies
computer vision to the data to ensure its assets operate Next: AI is the new UI
within predefined tolerances. It’s also creating high-fi-
delity digital twins of assets based on data pulled from Many of the aforementioned challenges in spatial
engineering, operational, and enterprise resource plan- computing are related to integration. Enterprises strug-
ning systems. gle to pull disparate data sources into a visualization
platform and render that data in a way that provides
The critical piece in each example? Data integration. The value to the user in their day-to-day work. But soon, AI
energy giant built a spatial storage layer, using appli- stands to lower those hurdles.
cation program interfaces to connect to disparate data
sources and file types, including machine, drone, busi- As mentioned above, multimodal AI can take a variety
ness, and image and video data.11 of inputs and make sense of them in one platform, but
that could be only the beginning. As AI is integrated
Few organizations today have invested in this type of into more applications and interaction layers, it allows
systematic approach to ingesting and storing spatial data. services to act in concert. As mentioned in “What’s next
Still, it’s a key factor driving spatial computing capabil- for AI?” this is already giving way to agentic systems that
ities and an essential first step for delivering impactful are context-aware and capable of executing functions
use cases. proactively based on user preferences.
Multimodal AI creates the context These autonomous agents could soon support the roles
of supply chain manager, software developer, financial
In the past, businesses couldn’t merge spatial and busi- analyst, and more. What will separate tomorrow’s agents
ness data into one visualization, but that too is chang- from today’s bots will be their ability to plan ahead and
ing. As discussed in “What’s next for AI?” multimodal anticipate what the user needs without even having to
AI—AI tools that can process virtually any data type ask. Based on user preferences and historical actions,
as a prompt and return outputs in multiple formats—is they will know how to serve the right content or take
already adept at processing virtually any input, whether the right action at the right time.
text, image, audio, spatial, or structured data types.12
This capability will allow AI to serve as a bridge between When AI agents and spatial computing converge, users
different data sources, and interpret and add context won’t have to think about whether their data comes
between spatial and business data. AI can reach into from a spatial system, such as LIDAR or cameras (with
disparate data systems and extract relevant insights. the important caveat that AI systems are trained on
high-quality, well-managed, interoperable data in the
This isn’t to say multimodal AI eliminates all barriers. first place), or account for the capabilities of specific
Organizations still need to manage and govern their data applications. With intelligent agents, AI becomes the
effectively. The old saying “garbage in, garbage out” has interface, and all that’s necessary is to express a prefer-
never been more prescient. Training AI tools on disorga- ence rather than explicitly program or prompt an appli-
nized and unrepresentative data is a recipe for disaster, as cation. Imagine a bot that automatically alerts financial
AI has the power to scale errors far beyond what we’ve analysts to changing market conditions, or one that
seen with other types of software. Enterprises should crafts daily reports for the C-suite about changes in the
focus on implementing open data standards and working business environment or team morale.
with vendors to standardize data types.
All the many devices we interact with today, be they
But once they’ve addressed these concerns, IT teams phone, tablet, computer, or smart speaker, will feel
can open new doors to exciting applications. “You can downright cumbersome in a future where all we have to
shape this technology in new and creative ways,” says do is gesture toward a preference and let context-aware,
Johan Eerenstein, executive vice president of workforce AI-powered systems execute our command. Eventually,
enablement at Paramount.13 once these systems have learned our preferences, we may
not even need to gesture at all.
12
13
egats
retnec
sekat
gnitupmoc
laitapS
The full impact of agentic AI systems on spatial comput- environments are just a few ways leading enterprises
ing may be many years out, but businesses can still are making their operations more spatially aware. As
work toward reaping the benefits of spatial comput- AI continues to intersect with spatial systems, we’ll see
ing. Building the data pipelines may be one of the the emergence of revolutionary new digital frontiers,
heaviest lifts, but once built, they open up myriad use the contours of which we’re only beginning to map out.
cases. Autonomous asset inspection, smoother supply
chains, true-to-life simulations, and immersive virtual
Endnotes
1. Abhijith Ravinutala et al., “Dichotomies Spatial Computing: 9. Gokul Yenduri et al., “Spatial computing: Concept,
Navigating Towards a Better Future,” Deloitte, April 22, 2024. applications, challenges and future directions,” preprint,
2. Ibid. 10.48550/arXiv.2402.07912 (2024).
3. Future Market Insights, Spatial Computing Market Outlook 10. Randle interview.
(2022 to 2032), October 2022. 11. Deloitte internal information.
4. David Randle (global head of go-to-market, AWS), interview 12. George Lawton, “Multimodal AI,” TechTarget, accessed Oct.
with the author, Sept. 16, 2024. 29, 2024.
5. Joao Copeto, chief information and technology officer, Sport 13. Johan Eerenstein (senior vice president of workforce
Lisboa e Benfica, interview with the author, August 27, 2024. enablement, Paramount), interview with the author, July 16,
6. Ibid. 2024.
7. Isabelle Bousquette, “Companies finally find a use for virtual
reality at work,” The Wall Street Journal, Sept. 6, 2024.
8. Fraser Health, “Fraser Health Authority: System wide digital
twin,” October 2023.
14
15
egats
retnec
sekat
gnitupmoc
laitapS
Continue the conversation
Industry leadership
Frances Yu Stefan Kircher
Unlimited Reality™ GM/Business lead | Principal | Deloitte Unlimited Reality™ CTO | Managing director | Deloitte Consulting
Consulting LLP LLP
+1 312 486 2563 | [email protected] +1 404 631 2541 | [email protected]
Frances Yu is a partner at Deloitte Consulting LLP, where she has Stefan Kircher is a managing director in the Products & Solutions
served in a range of global practice leadership roles. She has helped practice of Deloitte Consulting LLP and CTO for Deloitte’s
Fortune 500 clients as well as Deloitte launch several new ventures, Unlimited Reality™ Business. He has over 25 years expertise in the
evolved growth strategies, and transformed their demand value industry, technology strategy, and solution-building across various
chain. Currently, she is the US and global business lead and general industries, R&D, innovation, and partnerships with strategic tech
manager for Deloitte’s Unlimited Reality™, a multinetwork inno- partners like AWS.
vation business for the industr |
334 | deloitte | govtech-trends-2025.pdf | Tech Trends 2025 | Deloitte Insights
This report provides a government-specific take on Deloitte’s
Peering through the
Tech Trends 2025 report, spotlighting the accelerating technology
trends most likely to cause disruption in enterprise IT over the next
18-24 months. We explore which trends may be most relevant for
lens of government
governments and how ready governments are to take advantage
of them.
The technologies that enhance our Learn how governments can harness new opportunities in emerging
technologies to transform their organizations.
organizations and our lives are more
Relevance and readiness scale:
powerful and essential than ever
before. Forward-thinking governments We looked at each trend and assigned a value from one (low)
and five (high) based on the trend’s relevance and readiness of
and organizations chart upcoming
government adoption.
technological changes and look for ways
READINESS: RELEVANCE:
to utilize them for the benefit of citizens,
How ready is the government How impactful would it be if the
constituents, and employees alike. to adopt the trend? government adopted the trend?
Tech Trends 2025 | Peering through the lens of government
open-source options for the ability to train such models on smaller,
more accurate data sets. Together with multimodal models and AI-
based simulations, these new types of AI are building a future where
enterprises can find the right type of AI for each task. This includes AI
that can not only answer questions, but also complete burdensome
administrative tasks. In the years to come, a focus on execution may
usher in a new era of “agentic AI,” arming government employees
Spatial computing takes center stage with copilots capable of boosting efficiency and delivering enhanced
impact on the lives of constituents.
Exciting new use cases can reshape
industries
Trends in action
Spatial computing continues to spark enterprise interest because of Given the rate of advancement in Generative AI technologies,
its ability to break down information silos and create more natural government leaders should continually evaluate where employees
ways for government employees and the public to interact with and constituents gain the most benefits from increased adoption.
information. We’re already seeing enterprises find success with use They should balance the implications of cost against response
cases like advanced simulations that allow organizations to test quality and speed against risk. And, as we see the potential of agentic
different scenarios to see how various conditions will impact their AI, in which agents can act on behalf of humans, evaluating the
operations. With a stronger focus on effectively managing spatial implications on the skills of the workforce is even more important.
data, organizations will drive more cutting-edge applications. In the Today, AI can help social workers scan volumes of reports, help
coming years, advancements in AI could lead to seamless spatial speed hiring processes, and help perform routine constituent
computing experiences and improved interoperability, ultimately services while keeping a human in the loop. The value proposition
enabling AI agents to anticipate and proactively meet users’ needs. for AI remains robust, but government leaders should be strategic to
maximize ROI.
Trends in action
READINESS RELEVANCE
Being able to interact with information in a spatial context creates
2 4
many possible opportunities for governments, from urban planning
to emergency response to environment monitoring and much
more. Moving beyond simple data visualizations, organizations can
create immersive experiences to explore and visualize data in new
ways. Urban planners could visualize the changes to city planning
in near real time. The National Park Service could create immersive
educational experiences of the parks, blending history, science, and
wonder. Leaders should prioritize high-quality experiences over just
mediocre ones.
READINESS RELEVANCE
Hardware is eating the world
1 2
The promise of AI depends on more
than software
After years of software dominance, hardware is reclaiming the
spotlight. As AI demands specialized computing resources,
companies are turning to advanced chips to power AI workloads. In
addition, personal computers embedded with AI chips are poised
to supercharge knowledge workers by providing access to offline AI
models while “future-proofing” technology infrastructure, reducing
cloud computing costs, and enhancing data privacy. Although
What’s next for AI?
AI’s increased energy demands pose sustainability challenges,
Enterprises move beyond a
advancements in energy sources and efficiency are making
one-size-fits-all approach
AI hardware more accessible. Looking forward, AI’s continued
integration into devices could revolutionize the Internet of Things
To take advantage of the burgeoning excitement around Generative
and robotics, transforming industries like health care through
AI, many organizations have already adopted large language models
smarter, more autonomous devices.
(LLMs)—the best option for many use cases. But some are already
looking ahead. Despite their general applicability, LLMs may not
Trends in action
be the most efficient choice for all types of organizational needs.
Enterprises are now considering small language models (SLMs) and As AI and advanced computing capabilities grow in capability and
1
Tech Trends 2025 | Peering through the lens of government
scale to edge devices, government leaders should strategically
READINESS RELEVANCE
consider when, and how, to deploy specialized hardware to support
2 5
systems, data centers, and end users. The cost/performance ratio
of new technologies needs to be carefully evaluated, and leaders
may find that taking advantage of less widely available capabilities
requires workloads to run in cloud. In either case, the decisions
will be costly. Deploying AI-enabled hardware to remote areas like
research stations, forestry operations, or emergency response
zones may provide essential local computing power where internet
connectivity is unreliable. Through careful analysis and decision-
making, government agencies can enhance hardware investments to
support mission-critical functions.
The new math
READINESS RELEVANCE Solving cryptography in an age of quantum
2 3
In their response to Y2K, organizations saw a looming risk and
addressed it promptly. Today, IT faces a new challenge, and it will
have to respond in a similarly proactive manner. Experts predict that
quantum computers, which could mature within five to 20 years, will
have significant implications for cybersecurity because of their ability
to break existing encryption methods and digital signatures. This
poses a risk to the integrity and authenticity of data and
communications. Despite the uncertainty of the quantum computer
IT, amplified timeline, inaction on post-quantum encryption is not an option.
AI elevates the reach (and remit) of the Emerging encryption standards offer a path to mitigation. Updating
encryption practices is fairly straightforward—but it’s a lengthy
tech function
process, so organizations should act now to stay ahead of potential
After years of progressing toward lean IT and everything-as-a-service threats. And while they’re at it, they can consider tackling broader
offerings, AI is sparking a shift away from virtualization and austere issues surrounding cyber hygiene and cryptographic agility.
budgets. Long viewed as the lighthouse of digital transformation
Trends in action
throughout the enterprise, the IT function is now taking on AI
transformation. Because of Generative AI’s applicability to writing Future quantum computers could threaten the security of encrypted
code, testing software, and augmenting tech talent in general, data and transactional integrity. Given the volume and nature of
forward-thinking technology leaders are using the current moment information stored and processed by government organizations,
as a once-in-a-blue-moon opportunity to transform IT across five preparing for the transitions that will be required to better secure
pillars: infrastructure, engineering, finance operations, talent, and data and transactions is an imperative, not a nice to have. Agencies
innovation. As both traditional and Generative AI capabilities grow, should assess their current posture, identify the sources of
every phase of tech delivery could see a shift from human in the most sensitive or vulnerable data, and use modernizations
charge to human in the loop. Such a move could eventually return and transformation programs to upgrade their encryption and
IT to a new form of lean IT, leveraging citizen developers and cybersecurity practices. Steps taken today will help ease the burden
AI-driven automation. of steps that need to be taken in the future to address a constantly
shifting threat landscape.
Trends in action
READINESS RELEVANCE
The new trends in AI transformation allow government IT leaders
the opportunity to “lead from the front.” By incorporating AI tools 2 5
into current operations, practicing “digital transformation” inside
of IT itself, and fostering a culture of experimentation, technology
leaders can seize the moment to reboot, streamline, and accelerate
IT operations. New AI-augmented tools may empower nontechnical
users to perform increasingly sophisticated tasks such as developing
custom applications. By articulating a vision, fostering a technology-
savvy culture, and communicating the opportunities and potential
risks of AI, technology leaders can position to be enablers and critical
partners to achieving their agencies’ missions.
2
Tech Trends 2025 | Peering through the lens of government
The intelligent core
AI changes everything for the core
Core systems providers have invested heavily in AI, rebuilding their
offerings and capabilities around an AI-fueled or AI-first model. The
integration of AI into core enterprise systems represents a significant
shift in how organizations operate and leverage technology for
competitive advantage. This transformation is about automating
routine tasks and fundamentally rethinking and redesigning
processes to be more intelligent, efficient, and predictive. It requires
careful planning due to integration complexity, strategic investment
in technology and skills, and a robust governance framework to
ensure smooth operations. But beware of the automation paradox:
The more complexity is added to a system, the more vital human
workers become. Adding AI to core systems may simplify the user
experience, but it will make them more complex at an architectural
level. Deep technical skills are still critical for managing AI in core
systems.
Trends in action
Government agencies often find themselves with legacy systems
built to serve the needs of yesteryear, but poorly suited for the
changing, dynamic nature of today’s constituent expectations.
Using new tools, based on AI/machine learning and Generative AI,
are increasingly enabling the possibility of “modernizing in place” so
as to not disrupt critical operations. Today, AI can answer routine
questions accurately and accelerate the processing of standardized
forms, even if they contain large volumes of text. Investing in the
technology—and the talent to manage it—is increasingly become a
governmental imperative.
READINESS RELEVANCE
2 5
3
Tech Trends 2025 | Peering through the lens of government
Learn more
Follow
@DeloitteGov
@DeloitteOnTech
Related reports
Tech Trends 2025
www.deloitte.com/us/techtrends
Future of Government
www.deloitte.com/insights/future-of-government
Authors
For questions regarding GovTech Trends 2025, please contact:
SCOTT BUCHHOLZ
Government & Public Services CTO
Deloitte Consulting LLP
[email protected]
+1 571 814 7110
@scott_buchholz
This article contains general information only and Deloitte is not, by means of this article, rendering accounting,
business, financial, investment, legal, tax, or other professional advice or services. This article 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 article.
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 © 2025 Deloitte Development LLC. All rights reserved. |
335 | deloitte | us-2024-chief-strategy-officer-survey.pdf | 2024 Chief Strategy Officer
(CSO) Survey
As the world faces heightened
uncertainty, CSOs are optimistic,
resilient, and evolving
August 2024
Chief strategy officers are optimistic, resilient, and evolving
Over the last year, uncertainty has been In this year’s survey, we set out not only to Here’s what we found: CSOs surveyed offered optimism, resilience, and a
all over the news, with global markets pulse strategy leaders on their commitment to evolution.
sending mixed and occasionally outlooks, growth agendas, and focus
Optimistic outlook: Despite economic and geopolitical instability,
confusing signals. These signals speak areas but also to better understand how
most CSOs are optimistic that their organizations will successfully
to the challenges facing strategy and they are adapting their approaches to
navigate the year ahead, a sentiment that may be indicative of
business leaders today—how to create strategy in the face of these pressures. In planning for new capabilities, such as artificial intelligence (AI).
advantage and capture value in a a recent Monitor Deloitte report, “Strategy
Drastic shifts in investment areas: CSOs are investing in
landscape characterized by disruptive Now,” we explored how strategy is
emerging areas, including AI and ecosystems, for competitive
technologies, geopolitical and changing and outlined the new options,
advantage. A lag in activation may be representative of the early
economic uncertainty, changing challenges, and opportunities for strategy
stages these areas are in, which presents an opportunity for CSOs
consumer and stakeholder leaders. That report suggests strategy
to take an active role.
expectations, and complex policies should be resilient, agile, inclusive, and bold
Evolving ways of approaching strategy: CSOs report increased
and regulations around sustainability, to better match the challenges businesses
confidence in their core strategic initiatives, a shift that could be
tech, and data. In this environment, we face today. We wanted to explore this
related to changes in shaping, executing, and collaborating
conducted our fifth annual global assertion in this year’s survey data.
on strategy, consistent with traits outlined in “Strategy Now.”
survey of chief strategy officers (CSOs).
Strategy should be resilient, agile,
Obstacles to overcome: CSOs report facing real challenges as they
CSOs offer unique perspectives that
inclusive, and bold to better match navigate these forces and the evolution of their function, including
can resonate across businesses and
the challenges businesses managing across time horizons, talent shortages, and competing
industries. Often reporting directly to
strategic priorities.
face today.
the CEO, CSOs advise on special
projects, collaborate cross-functionally
on high-impact decisions, lead
corporate development, keep a pulse Read on for more highlights from the most
on markets, and are increasingly
recent global Chief Strategy Officer Survey.
responsible for executing strategies.
What’s inside
04
The outlook and priorities of CSOs in 2024
CSOs face geopolitical and economic uncertainty but remain optimistic about
their organizational performance.
05
Current investment areas and the CSO engagement gap
Investment priorities are changing but do not always align with CSOs’ roles and engagement,
which is unexpected given the importance of their leadership on topics related to
competitive advantage and growth.
07
Embracing new approaches to strategy
CSOs surveyed report new ways of leading strategy, consistent with the perspective
in “Strategy Now,” and signaling their shift toward a more resilient, agile, inclusive, and bold
approach to strategy execution.
10
What comes next for CSOs
Time, talent, and conflicting priorities pose challenges for CSOs and organizations in
achieving their strategic priorities. The "Strategy Now" principles may help manage these
challenges, and specific questions are offered to help guide CSOs and strategy leaders.
13
Authors
14
Survey methodology and acknowledgements
Outlook and priorities
Despite broad pessimism and uncertainty, CSOs surveyed are optimistic that their
organizations will successfully navigate the year ahead
In 2024, strategy leaders are…
Positive about their organization’s potential despite real pessimism Navigating uncertainty, focusing on core growth, and quickening their pace:
about business conditions:
Top external issues that # 1 # 2
may disrupt their
business strategy in the
Geopolitical Financial market
next 12 months
instability uncertainty
Most respondents are
focused on core growth, 91% 27%
indicating a relatively 50%
Organizational outlook Economic outlook
balanced portfolio
64% 11% Strengthening the Expanding into Pursuing
core adjacent growth transformational
growth
of respondents of respondents
reported being optimistic about reported being optimistic about Facing pressure to
say the strategy 55% say strategy
their organization’s performance the global economy in the next accelerate the pace of
development refresh
in the next 12 months. 12 months. strategy development 68% timeline is frequency is
and execution
shrinking. increasing.
For CSOs surveyed, organizational performance in the next year revolves around managing geopolitical and financial risks—risks that
are perceived to be substantial given general pessimism about the global economy and industrywide trends. To manage these risks,
CSOs are focusing on core business growth.
4 Copyright © 2024. Deloitte Development LLC. All rights reserved.
Investments and the engagement gap
Drastic shifts are happening in the percentage of respondents investing in new areas—
including AI, data, and ecosystems—to drive future growth and performance
Historical investment Future investment Delta in CSOs investing
Source of competitive advantage
AI (e.g., Generative AI, computer vision AI) 25% 88% 63%
Top
investment areas Data 61% 96% 35%
gaining focus Ecosystem business models 32% 60% 28%
Customer experience 68% 90% 22%
Sustainability, equity, and trust 61% 81% 20%
Cost structure 66% 85% 19%
Brand reputation 72% 90% 18%
Distribution channels 47% 65% 18%
Technology other than AI 78% 96% 18%
Supply chain 39% 54% 15%
Talent 82% 94% 12%
Innovation or R&D 68% 80% 12%
Economies of scale 59% 70% 11%
Intellectual property 40% 45% 5%
Product or service quality 88% 87% -1%
AI: Driving efficiency and productivity Ecosystems: Driving growth, innovation, and new business models
Potential reasons CSOs surveyed were overwhelmingly aligned on the No single benefit stood out for CSOs surveyed on the topic of
organizational benefits they anticipate for their AI efforts: ecosystems, but benefits tended to be focused on growth, with top use
for the focus on AI
80% indicated efficiency and productivity as a leading cases centering on improving customer experience, creating new
and ecosystems benefit, with top use cases of automation, business models, accessing new customers and markets, and
optimization, and customer service. facilitating innovation and product development.
5 Copyright © 2024. Deloitte Development LLC. All rights reserved.
Investments and the engagement gap
However, CSOs are not as involved in activating these efforts as expected
CSOs surveyed reported they are often not the one setting direction on issues at
the top of the strategy agenda, suggesting a gap in alignment/leadership:
The survey results identified a discrepancy between priority investment areas
and the involvement of CSOs. Given the CSOs’ critical role in detecting
industry shifts and spearheading special projects, it is surprising that they are
not more actively engaged in leading and shaping strategic investments:
Over half (54%) of CSOs surveyed reported playing a supporting (as
opposed to lead) role in shaping AI strategy. Some of this may be due to
early trials with AI being driven by tech leaders as Generative AI emerged and
companies tried to quickly understand the hype.
AI engagement gap Ecosystem engagement gap
Likewise, ecosystem strategy—an organization’s ability to work
28% 26% collaboratively within or across industries—has been an emerging priority
Only Only
for organizations for some time. However, of the CSOs surveyed, 31%
indicated they have had no role in their company’s ecosystem strategy.
reported playing a “lead role” in AI reported ecosystem strategy as a priority for
strategy development despite a 63- them despite there being a 29-percentage-
percentage-point increase in CSOs point increase in expected investment.
investing in AI.
There are many reasons strategic investments may not align with where strategy leaders are focusing in the context of a specific
company, but it is essential that CSOs are at the table for decisions in such important areas and this should be an intentional choice.
6 Copyright © 2024. Deloitte Development LLC. All rights reserved.
Adopting the traits of ‘Strategy Now’ may increase
confidence in growth outlook and a
CSO’s ability to shape key
strategic priorities
Want to learn more about the evolution of strategy?
Check out Deloitte’s perspective on “Strategy Now.”
New approaches to strategy
CSOs surveyed are evolving how they approach strategy, consistent with traits of ‘Strategy Now’
Four traits of ‘Strategy Now’ Dimensions measured in the survey Surveyed CSOs reported…
63%
Embracing uncertainty Bolster resilience by being data-driven
Resilient and incorporating analytics and insights Agreed their strategy function is evolving
Leveraging data and analytics
into how work gets done to use more data and analytics
82%
Bridging the gap with “execution” Boost agility by being execution-oriented
Agile Agreed their strategy function is evolving
and closer to where the strategy is activated
Accelerating insights to be more execution-oriented
Incorporating new views from Create inclusiveness by being externally 61%
Inclusive within and beyond the focused and seeking outside viewpoints as
Agreed their strategy function is evolving
organization part of strategy development
to be more externally focused
59%
Matching bold plans Pursue boldness, including being
Bold
tech-savvy in a year when AI and Agreed their strategy function is
with bold execution
disruptive tech were front and center evolving to collaborate closely with
tech leaders
While this year’s survey did not set out to directly test “Strategy Now,” it was an opportunity to find evidence of the evolution of strategy
in the market. Based on responses to the above questions, the traits of “Strategy Now” appear to reflect the experiences of strategy
leaders worldwide. Well over half the respondents agreed their function was evolving to be more in line with each dimension.
8 Copyright © 2024. Deloitte Development LLC. All rights reserved.
New approaches to strategy
Adoption of more aspects of ‘Strategy Now’ appears linked to CSO confidence and
optimism in the growth of their organizations
27%
of CSOs are activating all four dimensions: This “Strategy Now” cohort is far more confident in strong
the “Strategy Now” cohort. growth in the next year than its peers.
Two 65%
One
dimensions
dimension 23%
14%
39%
“Strategy Now”
CSOs
27% 33% Peers
All four Three
dimensions dimensions
Strong growth expectations
Count of “Strategy Now” dimensions reported by each CSO as a % of total respondents. Does not add to 100% as it shows relative % of each cohort reporting “strong
Does not add to 100% based on rounding and other respondents (<3%). growth expectations”
CSOs surveyed who are collectively more data-driven, execution-oriented, externally focused, and tech-savvy have an apparent
confidence. Their optimism resounds in a year marked by uncertainty and change.
9 Copyright © 2024. Deloitte Development LLC. All rights reserved.
What comes next
CSOs and their strategy teams hold a unique vantage point for
stewarding their organizations through today’s uncertain times. Their
cross-functional mindset and natural inclination to growth enables
them to remain optimistic despite market and geopolitical uncertainty
and volatility.
However, CSOs are not immune to these trends. While they are
optimistic about their outlook, they identified real challenges on time,
talent, and priorities that could hinder their ability to close the
“engagement gap” and achieve their ambitions more broadly and that
reinforce the need to continuously evolve their role.
The principles of “Strategy Now” can provide a framework and set of
guidelines to help CSOs think about how they adapt.
What comes next
Time, talent, and competing priorities are three of the biggest challenges facing CSOs this year
Balancing time horizons Talent constraints Competing priorities
~10% 51% #1
Only
Barrier
say they are spending enough time on long-term market cite talent and labor shortages as a key issue expected to CSOs surveyed note the top barrier to their organization
sensing—meaning ~90% were not spending enough time—despite disrupt or influence their strategies. Likewise, strategy achieving their AI and ecosystem goals is competing
68% noting that this is one of the role’s core responsibilities. functions decreased in size for the first time since 2021. strategic priorities (61% and 59%, respectively).
Organizations look to CSOs to support execution while Organizations ask more of their CSOs and on a faster Organizations rely on CSOs to help resolve strategic
advising on potential disruption. timeline, but CSOs have constraints. mis-alignment.
CSOs may need to find a way to manage the tension between CSOs may need to find a way to experiment with new CSOs may need to leverage their relationships with
the near and long term. approaches, despite the challenge. stakeholders to support emerging areas.
While these challenges impact every CSO and their organization differently, one way to navigate them, and potentially increase your
organization’s confidence, could be to adopt “Strategy Now” traits.
11 Copyright © 2024. Deloitte Development LLC. All rights reserved.
What comes next
A starting point for CSOs looking to adapt for today’s environment
Amp up your resilience
What data, insights, and tools are you leveraging to develop, monitor, and adjust your strategy? How are you leveraging AI
to drive outcomes faster and/or manage expectations better?
To what extent are you incorporating long-range scenario planning into your strategic planning cycle?
How are you stimulating dialogue within the organization across multiple possible futures?
Double down on agility
How can you, as a CSO, be more engaged in the execution of the strategy and stay closer to the
business and functional leaders?
How can you ensure insights are moving more rapidly from the businesses into the strategy and
vice versa? Have you created a feedback loop and aligned on an appropriate “burden of proof”
for accelerating decision-making?
Shift your strategy to be more inclusive
To what extent does your organization have a way to prioritize value to other stakeholders—
employees, vendors, and communities—in its strategies?
How are you incorporating diverse perspectives (internal and external) into your strategy, market
sensing, corporate development, or capability-building efforts?
Aspire to be bolder
What are the steps you are taking as a CSO to understand how new technologies can help unlock growth? Can you
achieve the growth you hope for without taking more of a leading role in AI strategy development?
12 Copyright © 2024. Deloitte Development LLC. All rights reserved.
Get in touch | Start the conversation
Kristen Stuart Gagan Chawla
US Consulting Strategy Leader US Business Strategy Leader
Deloitte Consulting LLP Deloitte Consulting LLP
[email protected] [email protected]
Nick Jameson
Andrew Blau
Principal, Business Strategy
CSTrO Program Leader US Leader, Eminence and Insights
Deloitte Consulting LLP Deloitte Consulting LLP
[email protected] [email protected]
Want to learn more about the role of the Chief Strategy Officer? Check out Deloitte’s Chief Strategy and Transformation Officer Program
and “The making of a successful chief strategy officer: Insights from the field.”
13 Copyright © 2024. Deloitte Development LLC. All rights reserved.
Methodology and
Acknowledgements
Survey methodology and demographics
This year’s survey was fielded from November 2023 to January 2024. With 128 respondents, the fifth iteration of Monitor Deloitte’s Chief
Strategy Officer Survey covered a diverse spectrum of strategy leaders.
Respondent industry breakdown*
32%
20% 19%
4%
11%
9%
Financial Services Consumer Energy, Resources & Technology, Media & Life Sciences & Health Nonprofits & Other
Industrials Telecommunications Care Public
Respondent organizational revenue size Respondent geographical location**
North America Rest of the world
9%
5%
36% 19% 42%
23%
Less than $5B
36% 55%
72% $5B–$24.9B
Greater than $25B
Europe, Middle
North America Latin America
East and Africa
15 Copyright © 2024. Deloitte Development LLC. All rights reserved. (*) Note: Total does not equal 100% as not all CSOs identified an industry. (**) The total does not sum to 100% as Asia Pacific
(AP) is not shown; AP data was limited this year.
Acknowledgments
Recognition and appreciation
This report is brought to you by Deloitte’s Chief Strategy and Transformation Officer (CSTrO) Program
but would not be possible without the thought partnership, collaboration, and assistance of our
colleagues working with strategy leaders in the United States and abroad that helped shape this report.
Thought partners
Benjamin Finzi, Jim Rowan, Francisco Salazar, Tom Schoenwaelder
Global advisers
Kendra Bussey, Luiz Caselli, Cedric Dallemagne, Daissy Davila, Andrew Dick,
Adam Ferfoglia, Chris Forrest, Takeshi Haeno, Gianni Lanzillotti, Shohei Mabuchi,
Gavin McTavish, Wayne Nelson, Robyn Noel, Kellie Nuttall, Maria Teresa Vilches
Research and eminence adviser
Elizabeth Molacek
Chief Strategy and Transformation Officer Program collaborators
Brittany Altonji, Jessica Barzilay, Chris Coelho, Elena Crowe, Drew Dickenson,
Matt Engel, Matt Hauck, Virginie Henry, David Spivak, Maclain Thornton, Shannon Woods
Additional thanks to the marketing and design team, including Hali Austin, Vanessa Carney, Matthew
Chervenak, Serafina Gontha, Jeanie Havens, Linnea Johnson, Vijayakrishnan K M, Alyson Lee, Nina
Lukina, Melissa Newmann, Jennifer Plym, Rachel Rosenberg, Erin Shapley, Meredith Schoen, Marie
Eve Tremblay, Talia Wertico
16 Copyright © 2024. Deloitte Development LLC. All rights reserved.
This document contains general information only and Deloitte is not, by means of this article, 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 article.
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.
About Deloitte’s Chief Strategy and Transformation Officer Program
The role of the CSO and the role of strategy as a corporate function continues to evolve, as CSOs are getting an increasingly important
role in the C-suite and are expected to set the short- and long-term direction for the business. Whether you seek to build your strategy
function, enhance your organization’s strategic planning capability, or pursue a specific growth mandate, we can leverage our long
heritage in strategy and a range of proprietary tools to help you succeed.
Contact: [email protected]
Website: www.deloitte.com/us/chiefstrategyofficer |
336 | deloitte | tte-annual-report-2024.pdf | State of Ethics and
Trust in Technology
Annual report
Third edition
Table of contents
Foreword Ethical standards Organizational
01 04 07
8 practices
3
30
Executive summary AI implications for Role of government
02 05 08
ethical standards
4 34
15
Introduction Role of the Chief Promoting trust and
03 06 09
Ethics Officer ethics in technology:
6
27 the way forward
39
2
Foreword
01
The rapid integration of Generative Artificial Intelligence and other We are at a pivotal moment in the history of human invention. Future
emerging technologies brings unparalleled opportunities to drive generations will undoubtedly look back on the decisions we make today.
0022
efficiency, improve automation, and enhance how humans and machines As leaders, policymakers, and stakeholders, it is critical to reflect on the
work together. Many of us are energized by the chance to be on the legacy we are creating. Our decisions should not only address the
ground floor of shaping this tech-driven future. However, these immediate benefits of technological advancement but also safeguard 0033
technologies also pose complex risks with pervasive impacts to principles to uphold a sustainable and equitable future for the next
organizations and society at large. This dualism underscores the need for a generation. It is up to us to honor our collective responsibility to those who
04
balanced approach—embracing innovation while upholding an unwavering will inherit the world we shape.
commitment to ethical standards.
05
The third edition of the State of Ethics and Trust in Technology report
illustrates the vital relationship between technological transformation and
06
ethical responsibility. This report shares valuable, actionable insights for
leaders to keep trust and ethics at the forefront when building a blueprint
for the deployment and governance of technology. Leaders can, and
07
should, ask probing questions to evaluate impacts and set strategic
priorities to navigate them, which will often require an agile, Lara Abrash
08
multi-disciplinary approach guided by a diversity of experiences and Chair
perspectives. By infusing this mindset into our decision making, we lay the Deloitte US
groundwork to harness technological capabilities, deliver value, and
09
advance a trustworthy future.
3
Executive summary
01
The accelerated rise in adoption of Artificial can create social, reputational, and financial value
Intelligence (AI) technologies since the release for their organizations, which can help build
0022
of last year’s Report increases the importance the confidence of their customers and increase
for organizations to consider the ethical employee engagement.
dimensions and implications of emerging 03
technologies. In this third edition Technology Trust and Ethics
(TTE) Report, we investigate how organizations
04
Generative AI (GenAI) offers organizations set ethical standards for emerging technologies
both the opportunity to improve efficiency and the implications of GenAI for the
and transform customer engagement, and establishment of those standards. We explore 05
the potential to expose organizations to the role of Chief Ethics Officers and their potential
reputational and financial risk. Similar potential power to inspire ethical behavior at organizations.
06
for good and harm exists across all emerging We examine the possible divide between
technologies. By establishing ethical standards employers and employees on issues of ethical
for the development and use of technology, technology use and how organizational practices
07
organizations can improve their relationships can create stronger, more aligned teams. Finally,
with customers and employees, demonstrate we look at the role government regulation can
08
a commitment to trust and responsible play in promoting ethical technology standards
technology use, and differentiate themselves and supporting organizations in implementing
from competitors. Leaders who drive their their own guidelines.
09
organizations to adopt trustworthy and ethical
principles for the use of emerging technologies
4
Important takeaways
Safety first Reputation is important AI Is a powerful tool,
01
but it requires guardrails
“Safe and secure” was marked as the most important Respondents show concern for reputational damage Cognitive technologies such as AI are recognized as
0022
ethical technology principle by respondents. to an organization associated with misuse of having both the highest potential to benefit society
Organizations developing or operationalizing their technology and failure to adhere to ethical standards. and the highest risk of misuse. The accelerated
ethical technology standards may consider using More than financial penalties, respondents point to adoption of GenAI may be outpacing organizations’
safety and security as the best entry point to bring an organization’s perceived ability to honor ethical capacity to govern the technology. Companies 03
leaders and workforce on board to implement ethical commitments as important to long-term success. should prioritize both the implementation of ethical
standards at scale. standards for GenAI and meaningful selection of use
cases to which GenAI tools are applied. 04
Leaders can inspire ethical Organizations should strive Organizations should
05
engagement for consensus, followed by shore up alignment with their
enhanced processes employees on ethics
06
Chief Ethics Officer roles are increasingly common. In Increasing ethics-based trainings and issue reporting Employers face a challenge coordinating with their
most organizations the position is viewed as enforcing practices may be successful at building consensus to workforces on embedding trust in professional
compliance, driving adherence to standards, and adopting trustworthy behaviors. Organizations should ecosystems. Trust in one’s organization and its emerging 07
championing individual responsibility for ethical next consider enhancing processes and actions in technologies may be declining and more pronounced
issues. Yet, where a Chief Ethics Officer is in place, technology development to address ethical risks and in younger generations, with concerns about user data
executive leaders should be involved to help define putting knowledge from trainings to action. privacy and security alongside the current state of GenAI. 08
and implement ethical tech processes. Organizations proactive in following through on ethical
technology standards may stay aligned to employee’s
expectations, leading to higher engagement and
09
potentially better outcomes from technology adoption.
5
Introduction
01
Goals of the survey
0022
Last year’s 2023 State of Ethics and Trust in The research to support this report started by Using findings from the survey and interviews,
Technology report reflected the watershed reviewing takeaways from last year’s report and our report was designed to provide insight
03
moment of GenAI, its rapid adoption, identifying how shifts in the technology landscape into how organizations are addressing the
and the new ways in which it demanded could alter those findings. We launched a ethics of emerging technologies alongside
organizations to prepare for its safe and 61-question survey to over 1,800 business and implementation. In this report, we analyze how
04
responsible use. In this 2024 edition, the technical professionals globally. The survey organizations are changing processes to align
concerns surrounding GenAI have grown, addressed how organizations place value on with ethical standards, and how organizations
05
from its capacity to widen the digital divide ethical principles for emerging technologies, the are collaborating with the government and
to its potential to increase the spread impact of GenAI on ethical technology learning commercial entities in their establishment
of misinformation and harmful content. and process changes within organizations in the of ethical standards. The report emphasizes
06
Despite awareness of these potential harms, first full year of its larger scale adoption, and how why organizations should approach internal
the opportunities afforded by GenAI are organizations implement practices that support technology operations, strategy, and decision-
clear, compelling organizations to balance ethical use and development of technology. We making from a framework of trust, and how 07
the benefits of adopting GenAI and related also interviewed 15 specialists and leaders across organizations can benefit and derive business
emerging technologies with the need to industries and 11 Deloitte leaders to gather value from embedding trust and ethics in their
08
mitigate their potential harms. insights in support of the survey’s findings. use of emerging technologies.
09
66
Emerging technologies under consideration
01
“Emerging technologies” refers to digitally enabled tools representing new and significant developments within a field. These technologies can be
grouped into the following categories:
0022
0033
04
Cognitive Digital Reality Ambient Autonomous Quantum Distributed Robotics
Technologies Experiences Vehicles Computing Ledger
Technology (DLT)
05
including general including augmented including AI/ML including automotive, including quantum including blockchain, including robotic
Artificial Intelligence reality (AR), virtual assisted wearables, aerial, and maritime. simulation, quantum crypto, non-fungible process automation.
(AI), GenAI, Large reality (VR), mixed voice assistants, linear algebra for token (NFT), and 06
Language Models reality (MR), voice and in-environment AI/ML, quantum more.
(LLMs), machine interfaces, speech devices. optimization and
learning (ML), neural recognition, ambient search, and quantum
07
networks, bots, computing, 360° factorization.
natural language video, immersive
processing, neural technologies,
08
nets, and more. computer vision,
and more.
09
In 2023 and 2024, GenAI received substantial attention for its potential to change the very nature of work. This year, we reflect on indicators from organizations who have begun
to create value and scale GenAI to their businesses and services, and whether their governance structures and workforces are keeping pace with innovation. We also explore
the increased awareness of concerns about GenAI ranging from data security and quality, explainability of GenAI outputs, and its potential for misuse.
7
Ethical standards
01
Today, many organizations have ethical standards and are committing more resources to train
their teams to use, develop, deploy, and scale technology safely and responsibly. At the same
time, many organizations are missing out on these potential business benefits by focusing on 0022
the risk mitigation or compliance aspects of ethical technology standards.
0033
While having standards in place can reduce Technology leaders seem to respect an intrinsic
risk, businesses that invest the minimum to value in having ethical guardrails baked into
ensure legal compliance or mitigate obvious all stages of technology development—from 04
harms may not experience longer term research to product launch. Bill Briggs, Chief
benefits and advantages of richer customer Technology Officer and principal, Deloitte
05
relationships, improved market reputation, Consulting LLP, posits that organizations who
and greater employee engagement. As one apply ethical checks solely as an extra step to
executive interviewed explained, “the challenge is meet compliance requirements are missing their
06
establishing systems to make sure organizations full purpose and impact. Embedding ethical
are thinking about the long term, without taking principles early and repeatedly in the technology
07
shortcuts. It is hard, because we are bad at development life cycle can help demonstrate
estimating long-term risks as people and weighing a fuller commitment to trust in organizations
that in proportion to short-term gains. Yet even and keep ethics at the front of your workforce’s
08
in the terms of corporate self-interest, in the long priorities and processes.2
view, it is better to be ethical.”1
09
8
Ethical standards
01
Are ethical technology standards in place?
0022
Fifty-three percent of respondents answered “no” or “unsure” to whether their organization had ethical technology standards. In this context,
“unsure” responses suggest where standards do exist, insufficient communication and facilitating education may be common. Response rates
0033
are similar to previous years’ surveys, suggesting this remains an area for optimization in organizations to be addressed sooner rather than later.
Organizations should consider taking steps to Organizations without ethical technology Figure 1. Does your company have
04
increase the impact and awareness of their standards may be not have them for a variety of defined ethical standards for developing
emerging technologies?
ethical technology principles, including: reasons, including:
(Percentage)
05
• Increasing investments in technology ethics • A lack of sponsorship from leaders to
14
teams, assessments, operationalization of operationalize standards.
standards, and learning resources.
• An under-appreciation of the risk associated 06
• Dedicating resources to research emerging with emerging technologies.
47
technologies and associated risks and
• A desire to lead with technologies before
07
appropriate use cases.
ensuring readiness with sufficient research,
39
• Prioritizing communications to socialize infrastructure.
08
standards and research findings to their
• An absence of a strong business case to
workforce.
understand how investing in ethical standards
could translate into positive returns. 09
Yes No Unsure
Wave 3 – 2024 (n=1,848)
Source: 2024 Deloitte Technology Trust Ethics Survey
9
Ethical standards
01
How do ethical technology standards function within an organization?
0022
Most respondents shared their organizations use company culture (20%). Reputation and brand advancing an ethical standard. This reflects a
ethical technology policies to manage risks, follow protection (15%), adding value to society (7%) and more reactive approach to why standards are
0033
regulatory compliance, and align workforces, revenue growth (3%) followed, indicating these important. Organizations may also lack evidence
more so than as an opportunity to create direct as secondary or tertiary boons to leveraging of how ethical practices translate into positive
business value. When asked the primary reason ethical principles (percentages sum to 101 due business outcomes. Sharing leading practices
04
their organization employed ethical technology to rounding). Respondents may emphasize the and achievements between organizations may
standards, the most common responses were potential consequences to non-compliance help them learn from what others have achieved
05
compliance with regulations (34%), enforcing with policy and company conduct as more through their own adoption of ethical imperatives.
standards of conduct (22%), and supporting significant than the benefits gained by proactively
06
Figure 2. Which of the following is the most important reason for your organization to have ethical tech policies and guidelines? 07
(Percentage)
08
34 22 20 15 7 3
Compliance with regulations Standards of conduct Company culture Reputation and brand protection Value-add to society Revenue growth 09
Wave 3 – 2024 (n=1,848)
Source: 2024 Deloitte Technology Trust Ethics Survey
10
Ethical standards
01
One year-over-year indicator is explainability Figure 3. Does your organization utilize
statements continue to be utilized across Explainability Statements providing
0022
users with information on how the
organizations as a means of communicating
technology works, including when,
accountability and informed purpose for the
how, and why it is used?
use of technology to customers and internal (Percentage) 0033
end-users. Fifty-one percent of organizations
use explainability statements—non-technical
15
04
explanations of a technology’s purpose, how it
was designed, and how it operates—to support
transparency in technology deployments. 05
51
35 06
07
Yes No Unsure
Wave 3 – 2024 (n=1,848)
08
Source: 2024 Deloitte Technology Trust Ethics Survey
09
11
Trustworthy and ethical principles of emerging technologies
01
Deloitte’s proprietary Technology Trust Ethics (TTE) Framework can serve as a first step in diagnosing the ethical dimensions of a company’s
emerging technology products. Deloitte first published the TTE Framework in its 2022 report and defines technology as trustworthy and ethical
by adhering to the following principles.3 0022
Safe and secure
Reskilling & Education
Private
0033
Users of the technology are protected User privacy is respected, and data is not
f a e R T a a
d
A
P
r
w
R
Tr en ncn h hoio o
eics
sc td dv e elrm
hc
pi
sb ui
c
u
/ r t
t
or
o
pro iu
t
e eo
p
eear hnr
u
oc cni
t
scs t s esih hc e o nm notk
a
i
o n
nb
uv ns o
rt
sl ae e lo
ose
l
s
au
e
it e n r en l
l
bh to oiac bs ft npa d
oo
t la g gnq
l
i eut fpl rev y
yd
, u r tm te
l
t
a p is
a
eie
m
h
s
cn ,p ca r
c
k el cwo eid hy h i sl ra yd
d
n/y ui etc to oub es ofa
s
ahr ri cc eu
l
tlc s
d
o
oi
ed ee .a t ss em
g
ds ai ie l g
t
o, yn
e
c i aei
n
.ut don
r
nam sn mn sd dl f msoi eh o
i
v ni ot rsa ri aid
e
pr eto r doe u em sn
w
ern e ra a s .
h
at e ol l , t, oa n
r
e
n
di
dsd
e
i
r ni ved
o rl
ts
n
o C
&
e c n
a n
re
v
o
GROBUST AND RELIABLE
EL
B AT N U
OAC
CA
d
Ao
Cc
anP
n
Ac
pr ss ue wi ts rd at eai be Rc rtS
l
et n aeU ea sA bts b oF le llE er
v
e F aA r bi lN e
e
OnD wd nU lyS es reE r
s
PC hiro pU tecR tioE n HumFT aneTTI rn ev u ln re r ea b
a
Cle ouc
mmoE
nc &T
mh
s
Soh
cit
al
Gotr
odn
hn u
eEoo
yi
tilib
at
nw
ias
tsuchll dtA eou st
s
uo cn ooo Fm oious cgg
rs
gy
nky
id d
ACo en ufi lad Vential Di Usc nr bP e it ai R so eCn dI o IaV nl n cA s le uT n sEE s J ivqu u eua sl AI it n tVi
A
c
atfi
i
cu
be sa
e
lidr eb sbpl
Li
slte Ar
e
i
Ia
be Tbt
l
Rea
l
Aeb Pl Me
I
DT NR
A
A RN IS AP FA R E N T A yN D
c
iE X
lP ELL A
BI AN
oP
&
ecnailpmoC
yR re otg
u al
u s t T U b d
u
i
F
T
i
an
not chr nes e as
s
ca ce a i c eo e
d
p
i
lnt euid
r
re p n es
e
tg ss
s
ed i rt co
a
s
os i- sl cu
v
t
,i er u p n
n
n
t
hi
en
a
ov as s s a
n
ld
nd e/
nn
yt e ;o oo r dre t
.
d
tia u a he lr r
om
o,
oe n g t s en
a
g
ud e etd o s
u
pa yt d
n
t
e fb cd
d
n
a
s, ias
o
d se u id uph
t
r
mny er
da
raa tho a c ebd er
i
er o itn at s
l
si ei
s
en iwod i ie c
l
.qo g,gn u xi
na
unt t ; lt e np eas ishu tc r
dd
ai le as h ln a bye ri aonrt e i lre ni po end n sn e
d
el a aoaa amd nt pogr sbe a e a ppty yd . l
o
k leea i t
i
s ia co rb n
a
an l ge
t
td
e
i o
d
n ,
0
0
00 075 864
a socially responsible manner. Technology’s S
benefits (e.g., quality, speed, safety, and/or price) 09
ELBISNOPSER
are evaluated in comparison to potential misuses.
12
Ethical standards
01
Consensus on “safe and secure”
0022
The imperative for safety and security is one Executives interviewed noted cybersecurity Sean Page, Managing Director, Risk and Brand
area of consensus among most respondents and domains have well-established standards that Protection, Deloitte LLP, also explains, “We now
0033
may indicate an entry point into broader ethical most organizations follow. While maintaining and have the ability to share whatever data we want
considerations for organizations, including those monitoring the safety and security of technology across the globe, [but] in some cases controls
without established guardrails. In 2024’s survey, systems is not a new problem or unique to are not in place to manage sharing. We need
04
78% of respondents selected “safe and secure” AI, this increased focus may indicate AI made to challenge the notion that all data is freely
as one of the top 3 ethical technology principles, a wider audience aware of the importance available and to appropriately segregate internal
05
a 37% increase over the previous year’s survey. of cybersecurity and user protections. As and customer data, because the likelihood of
Isolating responses for 2024 demonstrates one executive stated, “access to advanced incidents is significant. Contractual protections
the signal on “safe and secure” as the leading technologies opened a new ballgame. It is no are needed to ensure organizations can use the
06
principle is strong, with 36% of respondents longer in the hands of experts, and we have to data. While cybersecurity and privacy measures
indicating it is their number one ethical principle, account for bad actors who can do things with as are important, those have longer standing
more than twice the next most common rank little as an internet connection.”4 regulations, policies, and standards in place. With 07
1 choice, “responsible” (Figure 4). Individuals GenAI, additional emphasis on data governance
building a case for their organizations to and on bias detection and mitigation is needed.”5
08
adopt ethical technology standards may
find “safe and secure” resonates with
stakeholders and can act as the centerpiece 09
of an ethical technology strategy.
13
Ethical standards
01
Figure 4: Using the following list of technology-focused ethical principles, rank the top 3 by their relative importance to your organization.
(Percentage)
0022
Safe and secure 18 24 36
0033
Transparent and explainable 19 18 12
04
Robust and reliable 17 16 13
05
Responsible 14 14 17
Accountable 14 10 7
06
Fair and impartial 12 9 7
07
Private 7 8 8
08
Rank 3 Rank 2 Rank 1
Wave 3 – 2024 (n=1,848) 09
Note: Percentages shown only include ethical principles included in all three years of the survey to align with Deloitte's TTE framework dimensions.
Source: 2024 Deloitte Technology Trust Ethics Survey
14
AI implications
for ethical
01
standards
0022
Use of Generative AI is growing rapidly Figure 5. Which of the following most closely aligns to your 0033
GenAI has been embraced by organizations, organization's stage of adopting Generative AI technologies?
(Percentage)
with most respondents having exposure
04
to GenAI applications. Ninety-four percent
of respondents indicated GenAI is in
6 11
use at their organizations, though most 05
12
respondents indicated their organizations 8
In development
are piloting GenAI, with 12% of respondents
In testing 06
indicating their organizations have GenAI in
In pilot phases
wide scale use.
In limited use
In wide scale use 07
30
33
Not in development or use
08
Wave 3 – 2024 (n=1,848)
Source: 2024 Deloitte Technology Trust Ethics Survey 09
15
AI implications for ethical standards
01
Adoption of GenAI is becoming universal, Internal use of GenAI increased significantly GenAI use cases for customer engagement and
with 87% of respondents indicating their from 2023. Employees have begun to adopt marketing are rising, with 47% of respondents
0022
organizations are increasing their use productivity tools, and organizations reported reporting their organizations are using GenAI
of GenAI. using GenAI to streamline processes and reduce externally, compared to 31% in the previous
the cost and effort of operations. Seventy-eight year’s survey (a 52% increase). 0033
percent of respondents reported using GenAI
internally, compared to 65% in last year’s survey
04
(a 20% increase).
Figure 6. In the past year, has your Figure 7. Is your organization using Figure 8. Is your organization using
organization increased its use of Generative AI technologies internally? Generative AI technologies for 05
Generative AI overall? (Percentage) external-facing applications?
(Percentage) (Percentage)
06
78%
2024
65%
4
9
07
47%
87 % 31%
08
87
09
2023 2024 2023 2024
Yes No Unsure Yes Yes
(n=1,848) Wave 2 – 2023 (n=1,717), Wave 3 – 2024 (n=1,848) Wave 2 – 2023 (n=1,717), Wave 3 – 2024 (n=1,848)
16
Source: 2024 Deloitte Technology Trust Ethics Survey Source: 2024 Deloitte Technology Trust Ethics Survey Source: 2024 Deloitte Technology Trust Ethics Survey
AI implications for ethical standards
01
Technology, media, and telecommunications (TMT) Figure 9: In the past year, has your organization increased its use of Generative AI overall?
companies lead the adoption of GenAI relative to (Percentage)
0022
other industries. Fifty percent of respondents from
TMT companies said their organizations increased
Consumer
4 11 50 35
their use of GenAI. (n=222) 0033
Energy, Resources & Industrials
7 11 53 29
Amidst this period of growth, executives acknowledge (n=159)
04
the potential of GenAI and the imperative to retain
Financial Services
accountability for its proper use. Chris Griffin, (n=221) 2 8 52 38
Managing Partner - Transformation & Technology, 05
Life Sciences & Health Care
Deloitte & Touch LLP, states, “GenAI is a significant 6 9 55 29
(n=241)
technological advancement, which offers generational
Technology, Media & Telecomm. 06
opportunities for innovation and efficiencies across 3 7 39 50
(n=806)
industries and organizations. However, as organizations
Other
look to harness the potential of GenAI, it’s critical that (n=119) 7 20 44 29 07
they prioritize responsible development and ethical use
to sustain trust with stakeholders. This means investing
in robust governance frameworks that provide Unsure No Yes. Somewhat. Yes. Substantially. 08
Wave 3 - 2024 (n=1,848)
transparency and foster a culture of learning—while
Source: 2024 Deloitte Technology Trust Ethics Survey
also building out the systems of controls that allow
09
organizations to mitigate risks and maximize benefits.”6
17
AI implications for ethical standards
01
Risks of increasing Generative AI adoption
0022
As more companies experiment with GenAI, pilots Debbie Rheder, Deloitte Global Chief Ethics Respondents highlighted data privacy as
will progress into real-world implementations. Officer, offers, “GenAI tools are beginning to offer the most significant concern with the use
0033
During this transition, organizations have a the ability to analyze and interpret tone, also of GenAI. Seventy-two percent of respondents
chance to assess how existing ethical technology known as sentiment analysis, with algorithms ranked data privacy as their number 1, 2, or 3
standards can be adapted to meet GenAI improving at assessing human interactions. concern, and 40% ranked it as their top concern,
04
deployments. While excitement around GenAI While such advanced capabilities offer a new over 3 times more than the next top concern
and competitive pressure accelerates the pace opportunity for insights, it is another example (data provenance, 12%, rank 1). This may indicate
of adoption, companies can potentially run the where bias may be introduced in an algorithm personal unease about the protection of one’s 05
risk of employees overstepping around the and unfairly affect the accuracy of its outputs to data as well as awareness of the potential
use of data, customer privacy, system security, certain users if they are not represented in the harms—to both individuals and organizations—
06
appropriate use of tools, and other areas. This data used in training stages.”8 from violations of customer and employee
may create risks, straining controls that may be privacy and misuse of data. Individuals today also
insufficient for emerging technologies. As Sachin have greater awareness of existing regulations 07
Kulkarni, Managing Director, Risk and Brand such as the European Union General Data
Protection, Deloitte LLP, explains, “GenAI collapses Protection Regulation (GDPR)9 or the California
08
the ‘expertise barrier’: more people can get more Consumer Privacy Act (CCPA)10 in the U.S. and
out of data, with less technical knowledge needed. how regulations can influence global economies
While a benefit, the potential for data leakage may and other pending regulations. 09
also increase as a result.”7
18
AI implications for ethical standards
01
The next highest ethical concerns respondents Figure 10: For Generative AI, which of the following do you consider the top three most
pressing ethical concerns?
reported were transparency (47%, rank 1, 2, or 3)
(Percentage) 0022
and data provenance (40%, rank 1, 2, or 3). This
suggests individuals are seeking clarity for how
Data privacy 13 19 40
GenAI operates and how it collects and manages 0033
data. Users of GenAI tools should be able to Transparency 17 19 11
trust the reliability and veracity of its outputs
04
and may harbor concerns about the potential Data provenance 15 14 12
theft, replication, or misuse of the intellectual
IP ownership 13 13 11
property and creative outputs of individuals. 05
Bagrat Bayburtian, Technology Leader, Risk
Hallucinations 9 12 11
and Financial Advisory, Deloitte Transactions and
Business Analytics LLP, suggests, “organizations Data poisoning 11 11 5 06
need to know models are being trained the way
Authentic experiences 12 6 4
they want them to be. This is not trivial, and
07
as users of their models, organizations should
Job displacement 8 5 3
understand responsibility rests squarely
with them.”11 Static data 3 2 1 08
Rank 3 Rank 2 Rank 1 09
Wave 3 – 2024 (n=1,848)
Source: 2024 Deloitte Technology Trust Ethics Survey
19
AI implications for ethical standards
01
AI risks and rewards
0022
Respondents perceive cognitive technologies and industries. Assistive tools in use today can reality technologies similarly declined from 14% to
such as AI as having the most significant benefits help analysts more quickly and accurately parse 9% since 2022. Distributed ledger technologies,
0033
and risks of any emerging technology. Fifty-four through data, prioritize and render judgment on autonomous vehicles, and digital reality devices
percent of respondents indicated cognitive outliers, and extract insights from their datasets.12 offer examples as to how enthusiasm for emerging
technologies posed the most severe ethical risks technology can decline if limited practical use cases
04
of emerging technologies, while 46% indicated Only 4% of respondents thought distributed exist. Furthermore, if prominent use cases are
they potentially offer the most social good (Figures ledger technologies would contribute the most problematic and lack ownership of issues, trust in
12 and 13). The responses indicating AI could social good, down from 9% in 2022, and digital these technologies can experience a quick decline. 05
cause severe risks are down slightly year-over-
year, while the responses indicating AI’s potential Figure 11: Emerging technologies with the most potential for ethical risk and social good
06
use for good increased. The widespread
According to survey respondents, emerging technologies According to survey respondents, emerging technologies
adoption of GenAI may have increased
with the most potential for serious ethical risk: with the most potential for social good:
respondents’ familiarity with practical 07
applications, providing positive experiences 3DOWN 54 % 5UP 16 % No 6 % 7UP 46 % 9DOWN 9 % 3DOWN 7 %
pts pts change pts pts pts
through use of the technology. Will Bible, Digital
08
Transformation and Innovation Leader, Audit &
Assurance Partner, Deloitte & Touche LLP, cites
data analytics assistive tools powered by AI as an 09
Cognitive Digital Distributed ledger Cognitive Digital Autonomous
example and potential benefit across businesses
technologies reality technology technologies reality vehicles
20
Source: 2024 Deloitte Technology Trust Ethics Survey
AI implications for ethical standards
01
Figure 12: Which of the following emerging technologies do you Figure 13: Which of the following emerging technologies do you
think could potentially pose the most severe ethical risks? think will drive the most social good?
(Percentage) (Percentage) 0022
41% 33%
Cognitive Cognitive 0033
57% 39%
Technologies Technologies
54% 46%
16% 14%
Digital Reality 11% Digital Reality 12%
04
16% 9%
Distributed 13% Distributed 11%
Ledger 6% Ledger 10%
Technology 6% Technology 7%
05
8% 11%
Autonomous Autonomous
6% 11%
Vehicles Vehicles
7% 13%
8% 11%
Quantum Quantum 06
9% 10%
Computing Computing
7% 12%
7% 11%
Robotics 5% Robotics 12%
07
5% 10%
6% 9%
Ambient Ambient
6% 6%
Experiences Experiences
5% 4%
08
1% 1%
Other 1% Other 1%
0% 0%
2022 2023 2024 09
Wave 1 – 2022 (n=1,794), Wave 2 – 2023 (n=1,717), Wave 3 – 2024 (n=1,848)
Source: 2024 Deloitte Technology Trust Ethics Survey
21
AI implications for ethical standards
01
Organizations may consider taking a more Additionally, organizations should know when
cautious, patient, and informed approach to to do nothing. As noted by Bill Briggs, Chief
0022
selecting use cases to apply AI tools to meet Technology Officer and principal, Deloitte
business needs. As one executive indicated, Consulting LLP, organizations should invest in
applying AI to every use case may expose research to understand a technology and the 0033
an organization unnecessar |
337 | deloitte | the-financeai-dossier-generative-ai-use-cases-in-finance.pdf | The FinanceAI™ Dossier
A selection of high-impact
Generative AI use cases in Finance
Table of Contents
Introduction 3 Order to Cash 15
Finance Insights Engine 7 Procure to Pay 17
Autonomous Close 9 Working Capital Optimization 19
Dynamic Risk Assessment 11 Engage with my Tax Data 21
Cash Flow Forecasting 13 Investor Communications 23
2
The FinanceAI™Dossier
Introduction
TheadventofGenerativeAIhasdelightedandsurprisedtheworld,throwingopen For each of these domains, we explore how Generative AI can address
the doortoAIcapabilitiesoncethoughttobestillfaroffinourfuture.Witha enterprisechallenges in new ways, permit more and greater capabilities,
remarkable capacitytoconsumeandgeneratenoveloutputs,GenerativeAIis anddeliver advantages in efficiency, speed, scale, and capacity across the
promptingexcitement andstimulatingideasaroundhowthistypeofAIcanbeused financeorganization.
fororganizationalbenefit. Farmorethanasophisticatedchatbot,GenerativeAIhas
thepotentialtounleash innovation,permitnewwaysofworking,amplifyotherAI As with any type of AI, there are potential risks. We use Deloitte’s Trustworthy AI™
systemsandtechnologies,and transformenterprisesacrosseveryindustry. framework to elucidate factors that contribute to trust and ethics in Generative AI
deployments, as well as some of the steps that can promote governance and risk
TheFinanceAI™Dossierisacompendiumthathighlightsa handful ofthemost mitigation. Trustworthy AI in this respect is: fair and impartial, robust and reliable,
compellinguse casesforGenerativeAIacrossthe finance organization: transparent and explainable, safe and secure, accountable and responsible, and
respectful ofprivacy.
Financial Planning & Analysis Transactional Finance
To be sure, this collection of use cases is just a sample among myriad other
applications, some of them yet to be conceived. As Generative AI matures as a
Controllership Strategic Finance
technology and organizations move forward with using it for business benefit, we
will likely see even more impressive and compelling use cases. Theapplications
Internal Audit Tax highlighted here canhelp spark ideas, reveal value-driving deployments, and set
organizations on a road to making the most valuable use of this powerful
Treasury Investor Relations newtechnology.
3
The FinanceAI™Dossier
Our Perspective
Generative AI has the potential to transform Finance. Generative AI is powered by data, and Finance creates and
relies upon mountains of data. It’s a natural fit. Generative AI might start by producing concise and coherent
summaries of text, converting existing content to new modes, or generating impact analyses from new
regulations. Producing novel content represents a definitive shift in the capabilities of AI, moving it from an
enabler of our work to a potential collaborator. Leading organizations have launched pilot programs and are
scalingfast.
Generative AI continually adapts and learns. So, too, will the leaders who leverage the technology. At first,
Generative AI might support strategic planning—analyzing reports and data to create summaries or proposals.
It might augment autonomous finance operations or detailed reporting work. It will replace labor-intensive
processes and likely accelerate its own value rapidly. Generative AI might elevate continuous controls
monitoring. It could streamline strategic stakeholder communications.
CFOs and Finance leaders should consider today how Generative AI will affect both their functions and their
businesses tomorrow. To make sound decisions, leaders must consider the use of Generative AI from an
enterprise-wide approach with a clear understanding of where the technology will have an impact on operating
expenditures, capital expenditures, market capitalization, and a lot more. The impact is unlikely to stop there,
though. With its ability to process vast amounts of data and quickly produce novel content, Generative AI holds
promise for progressive disruptions we cannot yet anticipate.
Success will require strategic collaboration among C-suite executives—and return on investment—of Generative
AI deployment and adoption. The journey should begin with a sound strategy and a few use cases to test and
learn with well-governed and accessible data. It does not have to be perfect, but it should be controlled. In this
way, Generative AI can spark the next wave of innovation in Finance.
Generative AI heralds a new frontier for efficiently leveraging data, extracting insights, and creating content that
evolves from an enabler of our work to a collaborator.
4
The FinanceAI™Dossier
Six key modalities
One of the primary differences between more traditional AI and Generative AI is that the latter can create novel output that appears to be generated by
humans. The coherent writing and hyper-realistic images that have captured public and business interest are examples of Generative AI models outputting
datainwaysonceonlypossible withhumanthought,creativity,andeffort.Today,GenerativeAImodelscancreateoutputs insixkeymodalities.
Text Code Audio Image Video 3D/Specialized
Writtenlanguageoutputs Computercodeina Muchliketextualoutputs, Textualorvisualprompts Similartoimagery, Fromtextor
presentedinanaccessible varietyofprogramming audiooutputtedin leadthemodeltocreate GenerativeAImodels can two-dimensionalinputs
toneandquality,with languageswiththe natural,conversational, imageswithvarying takeuserprompts and (e.g.,images),models
detailsandcomplexity capacitytoautonomously andevencolloquial degreesofrealism, outputvideos,with canextrapolateand
alignedwiththeuser’s summarize,document, styleswith thecapacity variability,and“creativity.” scenes,people,and generate data
needs. andannotatethecodefor torapidlyshiftamong objectsthatareentirely representing 3D objects.
humandevelopers. languages,tone,and Examplesinclude fictitiousandcreatedby
Examplesinclude degreesofcomplexity. simulatinghowaproduct themodel. Examplesincludecreating
summarizingdocuments, Examplesinclude mightlookinacustomer’s virtualrenderingsinan
writingcustomer-facing generatingcodefrom Examplesinclude homeandreconstructing Examplesinclude omniverseenvironment
materials,andexplaining naturallanguage Generative AI-powered anaccidentsceneto autonomouslygenerating andAI-assisted
complextopicsinnatural descriptionsand callcentersand assess insurance claims marketingvideosto prototyping anddesignin
language. autonomously troubleshootingsupport and liability. showcaseanewproduct apurely virtualspace.
maintaining codeacross fortechniciansinthefield. andsimulatingdangerous
different platforms. scenariosforsafety
training.
Byunderstandingthesemodalities,organizationsareempoweredtothinkthroughandbetterunderstandthekindsofbenefitsGenerativeAIcouldpermit.Foreachusecase
describedinthisdossier,theremaybemorethanonevalue-drivingmodality.Achatbottextoutputcouldbepresentedassimulatedaudio;ageneratedimagecouldbeextended
intoa video.Ultimately,theGenerativeAIusecaseandthevaluetheorganizationseekswilldeterminewhichoutputmodalitiescancontributethegreatestadvantagesand
outcomes.
5
The FinanceAI™Dossier
6
c a
ThevaluethatGenerativeAIusecasescanenablecanbeconceivedacrosssixdimensions: costreduction,processefficiency,growth,innovation,discoveryand
insights, andgovernmentcitizenservices. Tobesure,asingleuse casecandrivemorethanonevaluecapture,buttohelppaintthevision forhowGenerativeAI
canbeusedto movetheneedleoncompetitive differentiators andoperationalexcellence,theusecasesdescribedinthisdossier areeachassociated witha
primaryvaluecapture.
Costreduction Government
citizen services
Reduce cost, typically by30%
orgreater, primarilythrough
Increaseaccuracyofvarious
automatingjobfunctions
federalandlocalprograms
andthenundertakingjob
andcreateeasieraccessfor
substitutions
at-riskpopulations
Valuecapture
Processefficiency Accelerating
innovation
Createprocessefficiencies
throughautomatingstandard
Increasethepaceofnew
tasksandreducingmanual
productornewservice
Growth Newdiscovery
interventions developmentandspeedier
andinsights
go-to-market
Increaserevenuegeneration
throughhyper-personalized
Uncovernewideas,insights,
marketingfortargetcustomers
andquestionsandgenerally
unleashcreativity
dezilaicepS/D3
egamI
Broad categories of value capture from Generative AI
The FinanceAI™Dossier
Finance Insights Engine
Financial Planning and Analysis
Process efficiency and
New discovery/insight
7
dezilaicepS/D3
egamI
How Gen AI can help
Data consumption at scale
Generative AI opens the potential for leaders to leverage data at a depth and
speed far beyond today’s possibilities. Operational data and financial data
are often inconsistent across an organization and lack a uniform structure.
Even key economic indicators like inflation, consumer spending, or interest
rates can vary substantively across geographies, sources of truth, or
Platforms powered by generative artificial Issue / opportunity interpretations. Generative AI could quickly reconcile disparatedata,
intelligence (Gen AI) can review and analyze analyze against company data,and deliver real-time, insight-rich content
Finance work often includes repetitive tasks like pulling that drives strategy.
data, identify gaps and suggest ways to fix
reports and reconciling data, much of which is manual
them, and provide leaders with on-demand and often in spreadsheets. There remain few resources Faster analysis and performance reporting
insights. and little time left to focus on the why behind the data Finance professionals could leverage a Finance Insights Engineto support,
supplement, and accelerate their work. The engine might identifyvariances
or explore multiple what-if scenarios. A generative
between plan and actuals and explain why they exist—eventually learning to
AI-powered insights platform could serve as a digital
tell more complicated stories deep into the financials.For example, when
analyst, allowing finance professionals to ask
labor expense comes in higher than forecast, generative AI can go multiple
questionsin plain language, explore unlimited
layers down in detail—considering geography, operational performance,
datasets, and receive custom reports that reveal
seasonality, special projects, and more—to identifythe root cause.
business performance.
Explanations could then be offered immediatelyin multimodal formats,
including text, graphs, charts, or video.
More productive strategy sessions
Imagine holding a planning session to identify needle-moving
opportunities for the upcoming year. Today, analyzing core financial
metrics for multiple time periods and business lines is a time-consuming
and subjective process. With generative AI-enabled technology at the table,
leaders could request and receive ad hoc analyses of operational and
financial data from the engine in real time to gain retrospective and
prospective insights.
7
The FinanceAI™Dossier
Transformation with speed and confidence
Managingriskandpromotingtrust
Reliable Transparent and explainable
The generative AI model is Confidence in generative AI
susceptible to erroneous, outputs requires stakeholders to
outputs delivered with understand how and why the
complete confidence, even with machine reached its conclusions. Human
hallucinated data points or conclusions. validation and regular audits of generative AI
Before conducting any analysis, data outputs remain essential.
sets should be confirmed and reviewed
for errors.
Potentialbenefits
Enhanced decision making Reduced latency
A Finance Insights engine powered by generative AI With its ability to analyze data instantly, generative AI
can dramatically reduce the manual effort to analyze can provide on-demand, actionable financial
data and deliver consistent, accurate, and up-to-date information to guide leaders’ business strategies.
insights for human analysts to leverage.
88
The FinanceAI™Dossier
Autonomous Close How Gen AI can help
Smart reconciliation
Generative AI could reconcile unstructured or inconsistent journal entries
or take on more complicated accounts that require supporting thoughts or
Controllership
significant estimates to reconcile. Conversational, generative AI-powered
chatbots might also enable users to input exceptions for remediation at
the source, run through next steps, update reconciliations, and consolidate
Generative AI could create a true “lights Issue / opportunity financials.
out” close process by improving leader
A consistently timely, accurate, and efficient financial Perceptive task management
visibility, minimizing rote work, and
close is a challenge. It requires a lot of human power. Generative AI could create integrated, automated closing checklists and, in
ultimately managing and completing tasks. Short bursts of activity take place throughout the year, time, it could centrally track and manage all close activities. It could also
use prior history to anticipate how journal entries impact others, recognize
but this limits visibility into the close process and often
issues to the close, and proactively reduce or eliminate delays.
prevents the finance department from focusing on
more strategic initiatives.
Improved variance analysis
Instead of relying solely on quantitative data, human analysts could
Generative AI can help eliminate the scramble to get
leverage generative AI to weave in unstructured data, like meeting notes,
the books closed on time and without errors. It can do
news stories, and interviews, to gain a deeper understanding of variances
the grunt work—categorizing transactions, making
between actuals and forecasts.
journal entries, and generating financial statements—
so that finance teams can focus on the bigger picture. Interpretative reporting
With time, it might take a bigger role in managing the Finance teams might set up templates from which generative AI could
produce initial accounting reports. As the technology develops logic to
close process and provide commentary on how the
monitor and interpret new or changing regulations, it might start to company performed.
provide impact assessments and produce more advanced accounting
treatments in response.
99
dezilaicepS/D3
egamI
Process efficiency
The FinanceAI™Dossier
Autonomous Close
Managingriskandpromotingtrust
Robust and reliable Transparent and explainable
Generative AI is moving When it comes to the closing
from an enabler of human process, generative AI-driven
work to a potential co-pilot, processes and content must be
but work still remainsto ensure accurate, clearly understood by finance teams and decision-
reliable results. makers.
Potentialbenefits
Process efficiency Cost savings
Generative AI can accelerate the close timeline with Passing off rule-based processing of routine
reduced effort and increased transparency. In time, transactions to generative AI technology can save time
generative AI might learn to anticipate barriers to by handling repetitive tasks.
close, predict next steps, and ultimately take a larger
role in managing the close process, allowing finance
teams to focus on strategic initiatives.
10
The FinanceAI™Dossier
Dynamic Risk Assessment How Gen AI can help
Key risk indicators and continuous monitoring
Generative AI may enhance risk management processes by enabling
unlimited, simultaneous, and continuous anomaly detection and analysis.
Internal Audit, Controllership, and Compliance, Risk
The technology could analyze transactions and other enterprise-wide risk
indicators in real time and generate immediate reports and insights on
potential discrepancies and outliers, allowing for timely risk response and
AI, including generative AI, will continue to Issue / opportunity mitigation.
elevate risk assessments, driving a
Risk management is critical for an organization’s Enhancing risk interviews
streamlined and value-added integrated
success—from business transformation to ongoing Generative AI can analyze unstructured data sources, like interviews, to
risk management approach that could operations. Sophisticated approaches require extensive uncover specific takeaways, themes, and insights. Leaders can then rapidly
transform today’s periodic risk assessments analyses of processes and data, from qualitative and identify and respond to existing and emerging trends.
into a state of continuous monitoring. quantitative sources. The work can be complex, time
Cyber risk monitoring
consuming, and susceptible to human error or Organizations can leverage generative AI to develop an aggregated
unintentional bias. depiction of cyber risk. With near real-time data that ranges across various
dimensions, leaders could better align their thinking and address critical
During risk assessments, leaders in various functions
gaps, threats, and opportunities. With time and development, generative
are often interviewed to gain risk-driven insights.
AI-enabled systems might also activate security measures, such as creating
However, interview capture and reporting are often
action reports, providing recommendations, and notifying users who may
performed manually, which could lead to missed or be impacted and need to take immediate action.
misinterpreted insights and a slow process. Further,
new metrics like indicators of cyber risk are emerging External risk sensing
Predictive, AI-powered analytics could analyze massive amounts of that can be more difficult for leaders to grasp.
intelligence—from open sources such as social media, blogs, forums,
Risks are also often highly interconnected across website reviews, industry newsletters, survey data, and news sources—
organizations, which make monitoring impacts more and then formulate actionable insights. Companies could gain advanced
complex. AI, including generative AI, could help leaders notice of emerging risks, knowledge of potential loss events, and increased
awareness of potential threats to their business or industry. effectively sense and assess risks to strategy,
operations, and other areas in a more dynamic and
real-time manner.
1111
dezilaicepS/D3
egamI
Cost reduction, Process
efficiency, New
discovery/insight, and
Accelerating innovation
The FinanceAI™Dossier
Dynamic Risk Assessment
Managingriskandpromotingtrust
Reliability Accountable Privacy Accountable
Work remains to ensure Continued risk management Interviews and surveys of WhiletheuseofGenerative
that generative AI produces requires identifying decision- business leaders may need AIcanacceleratethework
accurate, reliablecontent. makers for technology use and to be kept anonymous; ofdevelopers,withouta
Today, generative AI might confidently the decisions derived from the responses. in which case it will be crucial to ensure humanintheloop(e.g.,validatingand
produce incorrect output, known as that data privacy is maintained. debuggingcode),criticalfailuresmay
hallucinations occur.Shoringupaccountabilitymay
includedocumentingandcommunicating
standardsandexpectationsfor
employeesusingGenerativeAI.
Potentialbenefits
Value creation Process efficiency Accelerating insights New discovery
Generative AI can support an integrated Business units can receive more timely Leveraging generative AI solutions Companies can identify emerging risks and
approach to risk management, which reports that draw upon massive throughout the risk assessment life cycle predict organizational impacts in advance
includes teaming with the business to quantitative and qualitative data sets to can lead to data-powered insights of the marketplace through advanced
help maximize ROI and enabling better inform decisions and strategy. through end-to-end digital enablement capabilities of capturing and analyzing
business performance through effective and allows organizations to evolve massive internal and external data sets.
controls and governance. toward continuous assurance.
12
The FinanceAI™Dossier
Cash Flow Forecasting How Gen AI can help
Exponential data consumption
Generative AI can process and interpret data at unprecedented scale and speed.
Itcan ingest and analyze historical company data as far back as it dates and can
Treasury
also factor in external data from various sources, in multiple formats. Collectively,
richer data forms the foundation for the cash flow forecast, leading to more robust
analyses and more accurate forecasts.
Generative AI can improve the accuracy of Issue / opportunity Predictive analyses
Generative AI can identify the biggest drivers of cash flows and utilize a larger
cash flow forecasting, reduce manual
Cash flow forecasting is often a labor-intensive process. sample of parameters to forecast future cash flows more accurately.
processes, and provide greater insights to
And despite the work associated with it, many For accounts receivable, this might include factoring in customer trends, such as
business leaders. companies struggle to achieve a reliable forecast. average delay, percentage of payments delayed, average number of invoices per
payment, total open amounts, and time between payments. Additionally, it could
Thiscan lead to companies taking on higher borrowing
consider invoice factors, such as previous payment times, month due, day of the
costs for operations and potentially missing investment
week due, invoice value, and total current invoice value. It could also keep a pulse
opportunities. Generative AI offers the potential to on public data and extract economic patterns and customer activities that might
reduce the manual effort of data aggregation and impact future cash flows. This additional level of granularity and ability to predict
increase the accuracy of the forecast output— with precision can offer business leaders more confidence in their plans.
ultimately saving costs and enhancing returns. For accounts payable, this might include projecting expected trade payables
factoring in specificities related to vendors, based on importance and payment
Data sets often reside across multiple systems in terms. For larger cash outflow drivers, such as taxes or payroll, this could involve
structured and unstructured formats. A generative correlation of data from other sources (e.g., financial statements projections
fortaxes or Human Resources (HR) information for payroll) to enhance
AI-enabled solution can aggregate all sources into its
forecastaccuracy.
analyses. It might also begin to own part of the process.
Foreign exchange assessment When gaps or inconsistencies in the data arise, the
Generative AI can continually monitor international markets, factor volatility into
technology might research and resolve issues by
itsforecasting, and develop hedging strategies. Armed with this information,
following a set workflow (e.g., prompting sales leaders can gain more confidence that their associated decisions are rooted in
representatives with requests for sales forecast reliable data.
confirmation) or leveraging historical trends
Variance reduction
andprobabilities. With manual processes, forecasting relies on different perspectives to provide,
review, and analyze historical financial data. Generative AI can streamline and
Finance teams could access unlimited scenario-based standardize the process, leading to a significant reduction in potential for error
insights and predictions, allowing them to focus less variance to actual results. Forecasts could be further enhanced with integrated
time on generating reports and more time on analyzing visualizations to improve interpretation and confidence, quickly and with less
overall effort.
potential impacts.
1133
dezilaicepS/D3
egamI
Process efficiency and
New discovery/insight
The FinanceAI™Dossier
Cash Flow Forecasting
Managingriskandpromotingtrust
Transparent and Safe and secure Robust and reliable
explainable The financial information that Generative AI will require
Important decisions are will form the basis of the data early manual input and
made from cash flow models for generative AI tuning of data and tools
forecasting; therefore, it is critical for must be invulnerable to unauthorized access to realize the benefits of automation.
decision-makers to have visibility and or unintended uses outside of the intended Companies will need to identify how
accountability into how generative AI purpose for which the model is built. granular to get, as well as guidelines and
works. Forecasts will also improve over guardrails.
time, as the models have more
opportunities to run larger data sets.
Potentialbenefits
Timely market analyses More accurate forecasting Reduced borrowing costs Enhanced investment returns
Generative AI can conduct real-time, The more data that generative AI can Better visibility into cash flows and more Companies with a strong cash position can
ongoing reviews of multiple media leverage, the greater the possibility for confidence in forecasts could reduce the
confidently take advantage of longer-term,
sources and internal data that inform reliable, accurate information for planning need to tap into revolving credit lines
higher-yield investment opportunities.
forecasts and potentially improve purposes. and reduce associated borrowing
accuracy and reliability. expenses.
14
The FinanceAI™Dossier
Order to Cash How Gen AI can help
Automated orders
AI and machine learning (ML) can eliminate most of the manual tasks
across the order to cash cycle. Automated data collection, collation, and
Transactional Finance
interpretation can reduce the time spent on customer onboarding, data
management, and deal closing. ML-driven smart quote generation can
significantly reduce processing time on quotes and renewals. Once a sale
A mix of AI can fundamentally transform Issue / opportunity has been approved, AI can create an invoice and order fulfillment request
traditional order to cash processes. based on customer contract terms and standard policies and procedures.
Order to cash is the backbone of a business and a
AI,generative AI, and machine learning
critical component of the working capital value chain. Customer credit risk analysis
canautomate and improve tasks and The order to cash cycle is made up of several sub- Businesses want to know who they are selling to and how likely that
workflowsacross the order to cash cycles, many of which are highly manual today. This person is to pay on time, with accuracy. Generative AI can evaluate credit
risk by analyzing customer data and credit history to help identify high-risk
cycle,resulting in cost savings and workflow is ripe for generative AI-powered
customers, improve credit decision-making, and reduce costs associated
fastercollections. transformation, through which companies can better
with bad debt. Based on the risk analysis, generative AI can tailor sales
understand customer credit risk, shorten sales cycles
offers based on the risk category of customers.
and dayssales outstanding, and increase overall
processefficiencies. Faster collections
Collections today is labor intensive—phone calls and emails with invoice
questions, overdue reminders, and other dispute intervention, often
repeatedly. Leading organizations are already leveraging AI-enabled virtual
assistants that use natural language processing (NLP) to enable self-service
customer payments and collection activities by phone and chat, in some
instances pairing it with ML-enabled recommendation engines to offer
customized offers and payment plans. Generative AI and ML are likely to
further expand the capability of these virtual assistants in the near future
by tracking collections and work lists, automating dunning letters and calls,
making and documenting collectors' calls, providing collections agents with
recommended next actions in real time, running potential discount
analyses, and automating cash postings. They could also understand
payment trends and predict exceptions to get in front of them proactively.
1155
dezilaicepS/D3
egamI
Process efficiency, cost
reduction, and growth
The FinanceAI™Dossier
Order to Cash
Managingriskandpromotingtrust
Robust and reliable Accountable Fair and impartial
As the heart of the business Finance professionals will continue Particularly as it relates to
and cash flow generator, it is to be in the loop for reviews and credit decisions, sales terms,
important that order exception processing.Policies and discounts, the technology
to cash technology produces consistent and that determine who is responsible for the must be designed and operated inclusively
accurate outputs and withstands errors. decisions made or derived with the use of for equitable application, access, and
order to cash technology will be necessary. outcomes.
And since this technology is in front of
customers, potentially around sensitive
subjects like collections, it is important that
the agent script is carefully curated and on
brand to avoid reputational risk.
Potentialbenefits
Accelerated time-to-value Reduced collections efforts Enhanced accuracy
Integrating generative AI across the order Digitization and predictive analysis can help create a
Automating processes and operations can
to cash cycle can expedite orders by better understanding of customer credit risk, allowing
improve accuracy and help reduce the risk of
reducing processing time and improve companies to make smarter decisions around credit limits
days sales outstanding through faster and increasing the likelihood that payments will be made human errors. Humans will remain in the loop
collections. The efficiencies gained across for exception processing but can spend more
in full. This reduces the effort to collect payments or give
the cycle can improve working capital. time focused on strategic activities.
up accounts receivable in disputes.
16
The FinanceAI™Dossier
Procure to Pay How Gen AI can help
Enable efficiencies across procurement
Generative AI can enable efficiencies across procurement, with the
greatest potential in process automation, proactive risk and compliance
Transactional Finance
management, and strategic decision-making and negotiations around
suppliers and pricing. In an increasingly uncertain world, instant access
and ability to process information is vital for mitigating and managing risk
Generative AI can boost efficiencies Issue / opportunity and empowering organizations.
andunlock value across the
Despite having historically been at the forefront of Touchless invoicing and strategic supplier management
procure-to-payprocesses.
technological disruption, many sourcing and Generative AI accelerates the drive toward touchless invoice processing.
procurement functions continue to struggle to optimize Today’s automation is smart enough to process, match, and pay—acting as
a ‘digital employee.’ ‘Traditional employees’ will likely only need to
efficiency, manage risk, and manage costs. Generative
intervene upon exception and can shift their focus to more strategic,
AI can make the procure to pay process simpler,
value-adding tasks. Additionally, generative AI can help manage suppliers,
cheaper, smarter, predictive, and more accurate—
interacting directly through a chatbot feature that could, for instance,
lowering the cost of doing business and unlocking
answer questions about payment timing, or clarify disputes in payments
growth opportunities.
received. It can also develop supplier payment strategies based on things
like the likelihood the supplier to deliver on time, given any term changes.
Automated insights and growth driver
Generative AI unlocks the ability for insights, reducing the effort for
knowledge-based, value-add work. AI can now create models that are
learning and predictive in a manner that can give companies the first cut of
insights, giving employees a kickstart into their analyses, their ‘so-what’s’.
Companies can get smarter about managing inventory by leveraging
generative AI to analyze historical fulfillment rates. They can better
understand what they ordered, received and paid for to demand plan
more accurately and know when to place orders. Companies can know
when they need to have product to help generate revenue and be in a
better position to grow.
1177
dezilaicepS/D3
egamI
Process efficiency
The FinanceAI™Dossier
Procure to Pay
Managingriskandpromotingtrust
Accurate Reliable
The procure-to-pay process Using a generative AI-powered
starts by initiating a financial predictive model can enable
commitment and ends with organizations to make fact-
based and data-driven
cash leaving the company. Errors in
decisions. Organizations can compare
amounts or elsewise could be detrimental
products and services and rationalize them
and, as such, it is critical that any
across their supplier base, based on factors
automation around these processes is
that drive value for the company. Supplier
accurate.
performance becomes defendable, rather
than just opinion based. The analysis can
involve complex trade-offs, strategic
considerations, and tacit knowledge that
the AI models may no |
339 | deloitte | scaling-mission-driven-ai.pdf | Scaling
mission-driven AI
The path for US health agencies and nonprofits
SSScccaaallliiinnnggg mmmiiissssssiiiooonnn---dddrrriiivvveeennn AAAIII
What can
AI achieve
for US health
organizations?
Imagine the elation when public health researchers
break the code to slow heart disease, reverse diabetes,
or treat substance abuse disorders. Imagine the
gratifi cation epidemiologists feel when they can stop
a measles outbreak by providing the at-risk population
with information that prompted them to take the right
preventive steps.
Now, visualize the data scientists at the agency that
put the right tools in their hands to make this happen.
Artifi cial intelligence (AI) and now Generative AI are
powerful tools that can help federal health agencies
and nonprofi ts achieve breakthrough research and
advancements in public health and health care
delivery. However, leading a mission-driven AI agency
goes far beyond implementing a few successful pilots.
It’s a mindset.
222
Scaling mission-driven AI
Take the lead and scale AI across the organization
AI is poised to reshape what’s achievable not only Using mission-driven AI means that an agency’s AI
within these organizations but also for the strategy cannot be a product purely of IT or technical
communities they serve. teams but should be driven by senior business leaders.1
A recent Deloitte survey found that organizations where
Embracing AI and Generative AI can fuel federal senior leaders communicate a clear vision for AI are
health agencies’ mission by improving efficiency, 50% more likely to achieve their desired outcomes
effectiveness, and equity in health care. In many with AI.2,3 The White House is taking steps to clearly
cases, the path to achieving such benefits relies on communicate sweeping action to harness the benefits
incorporating advances in computer science with of AI, while mitigating its risks in President Biden’s
all other scientific and operational disciplines of an landmark AI Executive Order.4 Federal agencies have
agency. And these new technological tools are going already reported completing all of the 150-day actions
to play a role in the work of almost every employee in tasked by it.5
some capacity.
It’s certainly a start, but for transformative change to
occur, AI has to scale across the enterprise and into the
hands of employees. Leaders should visibly commit to
an AI strategy and champion the benefits.6
3
Scaling mission-driven AI
Balancing benefits with responsible use
Federal health agencies face multiple challenges. These include growing data volumes,
the increasing complexity of administering medical benefits and claims, upholding
an array of regulatory and grant obligations, protecting patient data and privacy,
and approving drugs and devices for safe and efficacious treatment. To address
these challenges, the search for more efficient, effective, and equitable solutions is
ongoing.7 AI can help organizations tackle these challenges and meet their missions,
especially when used responsibly.
Increasing efficiency and Facilitating insights for Supporting better citizen Maintaining public trust
cost effectiveness better decisions health outcomes For federal health
As with most innovations, Toward that end, in 2023
AI can alleviate the burden Generative AI can augment Generative AI can help AI and Generative AI pose President Biden issued
of repetitive yet essential skills and knowledge of provide hyper-personalized agencies, transparent risks. The technology is new an executive order that
tasks, enabling officials to employees. For example, experiences at scale for and requires governance to instructed federal agencies
focus on higher priority it can prompt systems to patients, employees, and the make sure data is secure and to establish guidelines for
processes and
activities. It can reduce analyze policies or datasets public, putting complicated used appropriately. While safe, secure, and trustworthy
costs and improve capacity. for answers across many types regulatory information, consumers appear to be development and use of
For example, it can craft of documents and images, health recommendations, guidelines can help comfortable with their doctors AI.12 Among other things,
optimized supply chain including handwritten notes.8 and claims requirements into using Generative AI in some this comprehensive order
strategies or accelerate Generative AI can help make simpler language. It can even capacities, 4 in 5 consumers addressed Americans’
ensure responsible use
drug discovery processes, recommendations, generate help unlock cures to disease think it is important or privacy, called for consumer
thereby helping transform ideas, and improve decision faster by facilitating improved extremely important that their protections, and advocated
federal health agencies making with intelligent information sharing across of AI and build public health care provider disclose for implementing in ways that
into more efficient and semantic search.9 research groups, running when they are using it for their advance equity and civil rights.
effective organizations. simulations and selecting health needs.11 Federal health It also supported innovation
trust in the technology.
candidates for clinical trials, agencies must balance AI’s and competition as well as
and learning from vast expected benefits with ways responsible and effective
amounts of data that can lead to ensure trustworthiness.12 government use of AI.
to more effective targeted
treatments.10 Yet, it must be
done responsibly.
4
SSccaalliinngg mmiissssiioonn--ddrriivveenn AAII
Human tasks/machine tasks
Where to apply AI becomes clear when
specific tasks are examined
Leaders should be quick to address concerns and
involve employees in identifying what tasks could be
completed faster, easier, or better by a machine, and
what tasks will be better done by humans. Leaders
should also begin to consider “futureproofing” their
workforce as employees will need new skills in a
Generative-AI era.13 AI is truly effective when it is
integrated into the everyday work of employees.
For example, when a Congressional inquiry comes
into a US health agency, AI can help draft the written
report that must be generated in response. AI can
assist in reviewing grant submissions, which is time
consuming when done manually, and consolidating the
information for a human-written or edited report. For
a health nonprofit, AI can translate disease information
into language that is more understandable for lay
readers. In these cases, the technology augments
the tasks of an employee. Instead of spending time
on these lengthy processes, employees can focus on
reviewing and improving their outputs and adding their
insights where appropriate.
Identifying where AI can be most advantageous gets
easier when specific tasks are analyzed.
5
SSccaalliinngg mmiissssiioonn--ddrriivveenn AAII
Real-world
applications
As AI advances, humans will continue to oversee
and manage outputs for authenticity. AI can create
mission-led value that is efficient, effective, and
equitable for federal health agencies and nonprofits.
And it’s already happening. This period of innovation
is best facilitated by a multi-skilled team of business
and technology leaders who put forth solutions that
are co-created, human-centric, and mission-effective.14
Take a look at these advancements.
66
Scaling mission-driven AI
MISSION-LED VALUE
Equitable: Greater fairness
Health equity in action
Health and clinical research has long underrepresented
specific populations, citing lack of participation in clinical
trials, the inability to travel to participating sites from
rural areas, and other social and economic issues.15
This challenge has limited the applicability of findings
to a subset of the population, creating an incomplete
picture of how to improve health outcomes for all.
Today, health agencies can employ AI to generate
insights to better understand and improve health
outcomes among these groups. Here’s how:
Chronic conditions prevention Maternal-child health Food insecurity
and management
AI has been used to help predict risk in pregnant The US Department of Agriculture reports that 39
AI has helped discover which populations are most mothers and design interventions tailored to who million people, including 18 million children, are food
vulnerable to certain chronic conditions like diabetes they are and where they live. Highlighting and finding insecure in America alone.19 AI has helped uncover
and hypertension. Even more encouraging for risks, especially in low-resource settings, helps to which populations face the greatest level of food
prevention and management is AI’s role in analyzing target this research and provides access to care for insecurity by locating food deserts. It can also help
social determinants of health (SDoH), which has been expectant mothers who need it. Real-time electronic predict hunger crises. When it comes to food aid,
challenging to do in the past. AI can mine electronic health recording and predictive modeling helps timing is vital.20 Insights have led to interventions like
health records and doctors’ notes, integrating SDoH clinicians monitor pregnancy, especially in mothers mobile farmers markets and transportation access to
factors like age, housing, lifestyle, and income level into who have gestational diabetes.17 AI has also been used help address food insecurity within a specific population.
more comprehensive treatment plans, recognizing that to improve prenatal diagnosis of birth defects and
health is influenced by a multitude of factors beyond prenatal genetic testing.18
the physical.16 AI can also help define which SDoH are
the greatest predictors for developing chronic conditions.
These insights have helped determine what actions
can be implemented to reduce the prevalence of
chronic conditions within a population.
7
Scaling mission-driven AI
MISSION-LED VALUE
Efficient: Time-saving
New frontiers in biomedical
research
Researchers increasingly rely on vast amounts of data
to validate their hypotheses. However, most data is
still in disparate formats, located in silos, and very
hard to find and reuse. AI is helping to consolidate this
multimodal data–including publications, images, and
multi-omics data–into a common format. Multimodal
data analysis has the potential to uncover new insights.
With access to more data, researchers can use AI to
accelerate the discovery of better treatments and
cures for diseases.21 It can help identify trends, improve
understanding, and foster better collaboration.
In addition, AI is helping researchers and physicians
communicate with patients more effectively by mapping
clinical notes into a format for an electronic health
record that patients can access and understand, which
can also be computable downstream for a researcher.
This improved management of complex information
can provide patients and care teams with more insights,
positively affecting health outcomes.
88
SSccaalliinngg mmiissssiioonn--ddrriivveenn AAII
MISSION-LED VALUE
Effective: Informed decisions
Drug discovery and availability
Research and development in biopharmaceuticals can be risky and expensive
endeavors. Only a fraction of new drug candidates survives clinical trials, demonstrates
success, and ultimately receives approval. It’s important that resources along the supply
chain are made available to researchers developing new drugs and running the clinical
trials to help them accelerate their mission-critical work. In addition, once the drug is
approved, companies need to keep up with demand.
Accelerating the process Supply chain insights
Some organizations are turning to AI and Generative Generative AI is also helping researchers organize
AI to transform many aspects of the drug discovery unstructured data about key suppliers of starting
process. Generative AI can rapidly create 3D materials to illuminate critical aspects of the drug
biomolecular structures and predict drug-to-protein supply chain. Organizations can receive millions of
binding. AI can increase the speed and efficiency of document submissions that have valuable supply chain
drug discovery, facilitating the creation of a digital twin data. This information is often reported inconsistently
for clinical drug trials that can lead to better patient and in a variety of formats, making it difficult to interpret
screening. Generative AI can help predict a clinical and detect potential pharmaceutical supply chain
trial’s probability of success, so researchers have disruptions. The use of Generative AI can help bolster
added confidence in the projected outcomes of surveillance efforts and enable researchers to better
real-world trials.22 understand the impacts of supply chain issues that
stem from starting materials.
9
Scaling mission-driven AI
MISSION-LED VALUE
Effective: Enhanced compliance
Effective grant decision making
In the world of grants management, getting necessary What if program leaders had access to a Generative AI
information that federal funding agencies can analyze tool that allows them to quickly generate a summary
quickly to monitor and support grant recipients can profile for one recipient, a set of recipients, or all
be difficult and time consuming. Agencies often have recipients? It could populate a pre-defined profile
multiple, separate systems to collect this information, template, pulling information from a variety of data
making it difficult to analyze the data and respond to sources such as budgets, work plans, progress reports,
incoming requests and inquiries. performance measures, and technical assistance data.
An organization can be called upon to generate The summaries would allow program leaders to
recipient-specific profiles and summaries of the focus more time on high-value activities to monitor
funding each recipient receives, its work, alignment and support grant recipients in working to achieve
with priorities, accomplishments, challenges, and the their goals. Generative AI’s capabilities can even help
technical assistance requested and received from the agencies inform their future funding decisions, which
funding organization. These summaries support site will lead to better and more profound public health
visits and policy or data requests from Congress, impacts in the future.
NGO partners, and other stakeholders.
1100
SSccaalliinngg mmiissssiioonn--ddrriivveenn AAII
MISSION-LED VALUE
Efficient: Cost optimization
Better biosurveillance
capabilities
Biosurveillance focuses on developing effective
capabilities for detecting, monitoring, countering, and
preventing national health threats in humans, animals,
food, water, agriculture, and the environment.23 Such
threats can include supply chain disruptions like those
that occurred during the COVID-19 pandemic.
AI technologies are assisting public health agencies in
creating more resilient health care networks, helping to
ensure that much-needed materials are in place ahead
of the next public health emergency. AI can also help
agencies understand public health vulnerabilities and
emerging risks as well as assist in disease tracking.
Generative AI is helping with data gathering from a
multitude of sources, not merely text documents,
but photos, audio, and video that can be used to improve
surveillance. It has the power to detect epidemic signals
much earlier than traditional surveillance, triggering
investigation and responses at the regional level.24
The more efficient agencies and nonprofits can be
in biosurveillance efforts will not only help in cost
optimization now, but it will also result in cost savings
as future crises are averted.
11
SSccaalliinngg mmiissssiioonn--ddrriivveenn AAII
MISSION-LED VALUE
Equitable: Better communication
Increasing donor and
community engagement
Many health nonprofits depend on donations and
gifts. AI can be used to help health nonprofits identify
and segment potential donors and then encourage
donor actions to support their missions. AI can also
be used to help nonprofits predict outcomes, such as
analyzing a large donor dataset to identify who might
be able to financially contribute to their work to protect
and enhance public health.25
Furthermore, AI can help nonprofits target and
personalize communications to potential donors,
which can contribute to improving the effectiveness
of fundraising efforts. AI chatbots can streamline
interactions and answer donor questions.
Like donor engagement, AI is being used at some
federal health agencies to improve communications
with its intended audiences. There, AI is helping to
combine various websites and rewrite the content
to be understandable and digestible for the public.
Generative AI can help build content, reach new
audiences, and answer questions. Making resources
easier to find and digest can increase transparency in
government initiatives, building trust while engaging
communities and improving the health of Americans.
12
Scaling mission-driven AI
Accelerate
Scaling AI: An integrated three-tier approach
the AI journey
1
Set the AI Direction
Determine where and how AI can best improve an organization’s operations and achieve mission/business needs
Fast-forward deployment and use
AI Exploration AI Strategy & Governance
Identify AI Opportunities & Use Cases Defi ne Vision and Establish Governance
While AI is in action at many health agencies and
nonprofi ts, leaders are asking a lot of questions about
scaling AI eff orts to maximize benefi t. It’s not just
2
implementing a pilot case here or there; it’s making
it a part of the entire organization. In order to do so Build Core Capabilities and Deliver AI Value
consider these questions: Determine foundational capabilities across people, data, and technology to enable AI solutions and deliver value
PEOPLE DATA TECHNOLOGY
• What are the key AI use cases to drive mission impact?
Customer & User Trustworthiness,
• What’s the best way to deploy, use, and embrace AI? Experience* AI Enabled Workforce Data Readiness AI Infrastructure / Platforms Security, & Risk*
Apply Customer-Centric Prepare the Workforce Provide the Data Foundation Provide Technical Foundation Mitigate Risk and
• What key data and infrastructure decisions need to Design & Delivery Instill Confi dence
be made at the outset?
AI Apps and Solutions
• How will AI be monitored and managed? Develop AI Solutions
• What investments are the right ones?
3
Consider taking a three-step integrated approach
to AI that considers strategies, technologies, and Manage AI Holistically
components. AI readiness and management requires Continuously maintain, manage, and build upon AI capabilities
a holistic view to fast-forward widespread deployment
and use. Various elements like data, algorithms, models, AI Delivery and Operations AI Sourcing Management
Scale, Maintain, and Operate AI Solutions Streamline Procurement
governance, ethics, and human expertise should be
brought together to create a comprehensive AI program.
Maximizing the benefi ts of AI while minimizing the risks
* Trustworthiness, Security, & Risk and Customer & User Experience are core to all AI capability areas and should be considered throughout the AI Journey
is the goal.
13
Scaling mission-driven AI
Set the agency’s AI direction
1
Determine where and how AI can best improve an
organization’s operations and achieve the mission
AI Exploration
The first step on the journey is to educate relevant
stakeholders and end users about the capabilities
and benefits of AI. Investing in AI fluency efforts and
workshops will help these stakeholders understand AI’s
potential to address their agency’s needs and challenges.
As these discussions progress, potential opportunities—
or use cases—for AI can be identified. These use cases
address specific needs and challenges within the agency
where AI can bring value and help solve problems more
efficiently and effectively. This step helps agencies better
understand what business apps they can develop to
realize the value of AI in producing better outcomes or
efficiency gains.
AI Strategy and Governance
Then a vision should be developed that includes defining
goals, success criteria, and time frame with focus on
prioritizing use cases that will have the greatest impact
and value. Factors to consider include time saved, mission
impact, and cost reduction. Now is also the time for agencies
to establish clear guidelines on governance. Guardrails
should be developed to help minimize risk, improve data
accuracy, address potential bias in the data, and provide
for transparency and accountability.
1144
Scaling mission-driven AI
Build scalable, enterprise-wide core capabilities that deliver AI value
2
Develop foundational capabilities across people, data, and technology to enable AI solutions and deliver value
People: Prepare the workforce Technology: Provide a platform to build solutions
Every technology innovation should start with the people Once the data is free-flowing, trusted, and secure,
it’s meant to support. As agencies build AI capabilities and organizations need a platform—an innovation sandbox—
business applications, they should also build an AI-enabled in order to create AI solutions. It should be an easy-to-use
workforce. Workers should be included at the start of the platform with capabilities to quickly build, deploy, and
AI journey, so they understand the potential benefits and monitor AI solutions for desired outcomes. The platform
risks of the technology. Adoption is sure to take hold as should leverage appropriate architectural principles
they help co-create the solution they are meant to use. (e.g., Data Commons) and implement governance, security,
Developing AI fluency among workers is important as is and trustworthiness principles. This will help ensure secure
upskilling those whose jobs could be directly affected by use of the emerging AI/GenAI technology.
the AI applications. Organizations who achieve AI at scale
do not shortchange this aspect of the program. It’s important to clearly define the objectives and goals
of any AI pilot project along with the metrics that will be
Data: Create a solid foundation of readiness used to measure success. It’s also critical to understand
Sound data practices make all the difference when it how citizens or employees will engage and interact with
comes to AI. If the data are inaccurate or simply unavailable, the solution and to ensure solutions are easy to use and
the quality of the output suffers. Organizations need to compliant with regulations and policies. Pilots should
develop adequate infrastructure and capacity to sufficiently begin in a controlled environment and use synthetic data
curate agency datasets for use in training, testing, and for testing on a development platform. The performance
operating AI. A strong data foundation enables the of the AI model should be evaluated against defined
implementation of enterprise-level AI solutions, all while quality metrics, and improvements should be made until
ensuring the use of secure, precise, and trustworthy data. adequate outcomes are reached.
Sound data governance practices, particularly data curation,
labeling, and standardization, can help maximize Establishing an AI Center of Excellence can help optimize
appropriate outcomes. costs of development by creating repeatable business
applications that can be tweaked for a variety of purposes.
Most importantly, data access must be democratized, Think of a chatbot that supports multiple workflows or a
making it free flowing and accessible. Data stuck in tool that summarizes contents to inform users. The same
silos isn’t working for the organization or its mission. chatbot or tool could be used by multiple departments in
Collaborating with professionals who have experience the same organization for different purposes. A Center
in building data and AI capabilities can provide much- of Excellence can help achieve AI at scale and help instill
needed guidance. trust in the solution.
1155
Scaling mission-driven AI
Manage AI holistically
3
Continuously maintain, manage, and build upon
AI capabilities
AI Delivery and Operations
AI models can change and evolve over time as they
continuously learn and adapt. Health agencies must
regularly monitor and evaluate their performance. AI models
can be tested with new data to evaluate performance and
help ensure they are providing accurate, reliable results.
All in all, think like a
Incorporating feedback from users and stakeholders
helps identify areas for improvement. An interactive AIOps
process is needed to help ensure continued accuracy and researcher to find more
performance of AI solutions.
effective ways to achieve
AI Sourcing Management
AI technology is constantly evolving and it’s important
to stay abreast of the latest advancements and leading the mission. Make it easy to
practices. A sourcing strategy that enables the effective
procurement, oversight, and management of vendor-
implement AI solutions from
provided AI solutions, tools, and services can advance
mission, operations, and technology objectives. It takes
a village. Make sure to evaluate performance continuously.
an infrastructure standpoint.
Ensure the mechanics are
there for a safe, secure,
ethical experience that
includes humans in the
loop for monitoring.
1166
Scaling mission-driven AI
Trustworthiness,
security, and risk
Understanding the vulnerability and threat
What are the risks of AI? There are two ways to classify AI
risk: AI vulnerabilities from using the technology in an agency
program (risks to using AI) and AI threats typically from bad
actors using AI to their benefit to hurt an organization (risks
coming from AI). AI systems can be complex and opaque
and by nature are susceptible to a wide range of issues that
can limit their ability to perform consistently and accurately,
making them less reliable in dynamic, real-world scenarios.
AI system vulnerabilities can include data privacy breaches,
bias and ethical concerns, and a lack of explainable or
erroneous results.
AI threats can include malware generation, system breaches,
fictitious personas, misinformation, and social engineering.
Rapid advancements in AI and the availability of open-
source AI tools have also lowered the entry barrier for
attackers, who can automate and scale more damaging
attacks. These risks can affect an agency’s reputation,
mission, finances, and data. One recent cyberattack
threatened the security of patient information and has
disrupted patient care and access to medications.26
17
Scaling mission-driven AI
Regulations are gearing up
GGuuiiddaannccee aanndd rreegguullaattiioonnss aarree eemmeerrggiinngg ttoo aaddddrreessss tthhee
rriisskkss aassssoocciiaatteedd wwiitthh AAII’’ss ccoommpplleexx iinntteerrppllaayy ooff tteecchhnniiccaall
aanndd ssoocciieettaall ffaaccttoorrss.. FFeeddeerraall,, ssttaattee,, aanndd llooccaall aaggeenncciieess
aarree ccoonnssiiddeerriinngg oorr eennaaccttiinngg AAII gguuiiddaannccee aanndd rreegguullaattiioonnss
to account for and govern the use of the technology
across government.
IInn aaddddiittiioonn ttoo ccoommpplliiaannccee wwiitthh tthhee 22002233 eexxeeccuuttiivvee oorrddeerr
DDDeeellloooiiitttttteee’’’sss TTTrrruuussstttwwwooorrrttthhhyyy AAAIII™ can help
from the White House,27 tthhee OOffiffi ccee ooff MMaannaaggeemmeenntt aanndd
mmmiiitttiiigggaaattteee rrriiissskkk aaannnddd eeennnsssuuurrreee cccooommmpppllliiiaaannnccceee
Budget (OMB) has issued the fi rst government-wide
ppoolliiccyy ttoo mmiittiiggaattee AAII rriisskkss aanndd hhaarrnneessss iittss bbeenneefifi ttss.. TThhee
OOMMBB’’ss ppoolliiccyy aaddddrreesssseess rriisskk ffrroomm AAII uussee,, eexxppaannddss tthhee EEEssstttaaabbbllliiissshhhiiinnnggg AAAIII gggooovvveeerrrnnnaaannnccceee cccaaannn hhheeelllppp iiidddeeennntttiiifffyyy,,, mmmiiitttiiigggaaattteee,,,
rreeqquuiirreemmeenntt ffoorr ttrraannssppaarreennccyy iinn ppuubblliicc aaggeenncciieess’’ uussee ooff aaannnddd mmmaaannnaaagggeee rrriiissskkk wwwiiittthhhiiinnn aaannn eeevvvooolllvvviiinnnggg ttteeeccchhhnnnooolllooogggiiicccaaalll
AAII,, rreemmoovveess bbaarrrriieerrss ttoo rreessppoonnssiibbllee AAII iinnnnoovvaattiioonn,, ccaallllss lllaaannndddssscccaaapppeee aaannnddd rrreeeggguuulllaaatttooorrryyy eeennnvvviiirrrooonnnmmmeeennnttt... DDDeeellloooiiitttttteee’’’sss
for agencies to grow their AI workforce, and requires TTTrrruuussstttwwwooorrrttthhhyyy AAAIII™ is anchored on seven dimensions.
federal agencies to strengthen AI governance.28 AAAIII mmmuuusssttt bbbeee tttrrraaannnssspppaaarrreeennnttt aaannnddd eeexxxppplllaaaiiinnnaaabbbllleee,,, fffaaaiiirrr,,, aaannnddd
iiimmmpppaaarrrtttiiiaaalll,,, rrrooobbbuuusssttt aaannnddd rrreeellliiiaaabbbllleee,,, rrreeessspppeeeccctttfffuuulll ooofff ppprrriiivvvaaacccyyy,,,
Federal agencies may also be required to designate sssaaafffeee aaannnddd ssseeecccuuurrreee,,, aaannnddd rrreeessspppooonnnsssiiibbbllleee aaannnddd aaaccccccooouuunnntttaaabbbllleee...
an AI offi cial, conduct maturity and compliance
assessments, and develop their own AI strategy. AAAttt iiitttsss fffooouuunnndddaaatttiiiooonnn,,, AAAIII gggooovvveeerrrnnnaaannnccceee eeennncccooommmpppaaasssssseeesss aaallllll ttthhheee
aaabbbooovvveee ssstttaaagggeeesss,,, aaannnddd iiisss eeemmmbbbeeeddddddeeeddd aaacccrrrooossssss ttteeeccchhhnnnooolllooogggyyy,,,
TThhee UUSS DDeeppaarrttmmeenntt ooff CCoommmmeerrccee’’ss NNaattiioonnaall IInnssttiittuuttee ppprrroooccceeesssssseeesss,,, aaannnddd pppeeeooopppllleee... TTTooogggeeettthhheeerrr,,, gggooovvveeerrrnnnaaannnccceee aaannnddd
ooff SSttaannddaarrddss aanndd TTeecchhnnoollooggyy ((NNIISSTT)) hhaass rreelleeaasseedd aann AAII cccooommmpppllliiiaaannnccceee aaarrreee hhhooowww aaannn ooorrrgggaaannniiizzzaaatttiiiooonnn aaannnddd iiitttsss ssstttaaakkkeeehhhooollldddeeerrrsss
Risk Management Framework29 to help manage risks eeennnsssuuurrreee AAAIII dddeeepppllloooyyymmmeeennntttsss aaarrreee eeettthhhiiicccaaalll aaannnddd cccaaannn bbbeee tttrrruuusssttteeeddd...
to individuals, organizations, and society associated
wwiitthh AAII.. IItt iiss iinntteennddeedd ffoorr vvoolluunnttaarryy uussee aanndd tthhee ggooaall iiss
to improve the ability to incorporate trustworthiness
considerations into the design, development, use,
aanndd eevvaalluuaattiioonn ooff AAII pprroodduuccttss,, sseerrvviicceess,, aanndd ssyysstteemmss..
There’s no doubt that specifi c frameworks and
methods for identifying, managing, and monitoring
AI risks are needed.
1188
Scaling mission-driven AI
Creating AI mission-led value
There are many potential short- and long-term benefits of AI in public health. Federal health agencies all
Efficient Effective Equitable
strive to be efficient in how they work, effective at achieving the goals of their underlying mission, and fair
and equitable in how they serve all citizens. Here’s what agency leaders can do by providing employees
with a supercharged tool that is efficient, effective, and equitable.
Time saving Informed decision- Greater fairness
making
AI can assist employees by AI solutions can be designed to
augmenting human work and AI can provide quick insights mitigate bias in data sets and
reducing manual tasks. AI and answers to questions, models, leading to more fairness
and Generative AI specifically thereby reducing the time and equity in decision making
can quickly analyze and spent on manual data analysis and more targeted outreach.
summarize large amounts and allowing health agency
of unstructured data, such employees to make more Better communication
as lengthy applications and data-informed decisions in
AI is helping to turn complicated
progress reports, to extract a timely manner.
or sophisticated language into
salient points and provide
easy-to-read text that the average
quick insights. Enhanced compliance
person can understand.
AI can assist in monitoring
Automating manual tasks and
compliance with grant policies
processes not only saves time
and requirements, which
for workers but can enable
in turn can help granting
them to focus on more high-
agencies be better financial
value activities. In total, these
stewards of funding.
Cost optimization
Improved accuracy benefits can
AI can help agencies predict
AI can deliver more accurate
which regions, applicants, or
and consistent results
recipients carry more risk or result in better
by following predefined
need monitoring assistance,
criteria and rubrics. It can
allowing employees to
also improve the quality of health outcomes
strategically allocate limited
questionnaires by flagging
resources for better outcomes.
errors or missing information,
Task automation can reduce for citizens.
providing feedback, and
costs associated with manual
reducing the number of
review processes.
ma |
340 | deloitte | us-advisory-ai-systemic-risk-in-banking-june-2024.pdf | 'Weapon and tool' - systemic risk implications of AI in banking and finance
Initial perspectives related to remarks by the Acting Comptroller of the Currency, Michael J. Hsu
On June 6, 2024, Acting Comptroller of the Currency Michael J. Hsu delivered remarks at the Conference on Artificial Intelligence and Financial Stability, hosted by the Financial Stability Oversight Council (FSOC)
in partnership with the Brookings Institution, wherein he discussed systemic risk implications of artificial intelligence (AI) and offered his thoughts on approaches to AI deployment to improve its safety.1His
remarks are the latest illustration of regulators’ growing concern about AI. In its 2023 Annual Report, FSOC—for the first time—identified AI as a potential systemic risk.2
5 insights youshould know 5 considerations to evaluate
AI presents accountability challenges: AI’s ability to evolve overtime and self-learn makes it a powerful tool but Establish clear roles and responsibility: Banks should apply existing principles of risk governance and model risk
can also result in model drift, where the model’s accuracy and performance deviate from expectations. This management (see Federal Reserve Supervisory Letter 11-7, OCC Bulletin 11-12, and the Comptroller’s Handbook
may be especially true in the case of nontransparent models that are powered by third parties. Banks may 1 on Model Risk Management)3to their AI applications and across their model lifecycles. For third-party AI-tools that
struggle to identify whom to hold accountable for what or how to fix any issues, which could—ultimately— may pose particular challengesto an organization’s internal accountability framework, controls should be put in
erode trust within the banking system. place commensurate with the bank’s risk exposure and complexity and extent of the model’s usage.
Competitive pressures may cause banks to neglect controls: As competitive pressures grow within the industry Develop gates between AI development stages: Banks should identify in advance “gates” or points at which
to develop and launch AI-enabled applications, risk management and controls may be neglected by some pauses in growth and development are needed to establish controls as AI develops across the maturity spectrum.
banking organizations. As a result, risks may grow undetected and unaddressed until a critical failure or 2 Hsu stated AI applications evolve across three stages: (1) inputswhere AI provides information for humans to act
disruption occurs. It is therefore critically important for adequate initial due diligence, and risk management and upon; (2) co-pilotswhere AI enables humans to do tasks more quickly; and (3) agentswhere AI executes activities
controls to keep pace with growth in order todrive sustainable growth and stability. on behalf of humans. It’s important for banks to demonstrate to regulators a coherent AI strategy with controls.
AI-enabled fraud is a top concern: Nefarious actors are increasingly able to access and deploy AI-enabled tools Invest in customer protection and compliance: Leveling up customer security protocols and consumer compliance
for fraudulent activities. For example, AI tools—including deepfakes—may be used to impersonate an should be considered, so as tobetter align with evolving AI technologies. This may include investing in AI-enabled
individual’s voice or likeness to trick friends and family to send money to a fraudster or even bypass a bank 3 security solutions to detect and respond to AI-fraudulent activities in real-time, such as advanced behavioral
customer’s account security check. AI may be used to drive the increase in the scale and scope of fraud, which analysis and anomaly detection. Additionally, banks should proactively manage the risk of consumer compliance
could undermine trust in the payments and banking system. violations, such as prioritizing model accountability and transparency particularly for consumer-facing applications.
AI-enabled cyberattacks are a growing risk: Cybercriminals are increasingly deploying AI-enabled tools to Invest in cybersecurity and operational resilience: Strategic attention should be given to evaluating cybersecurity
launch sophisticated attacks on individuals and organizations. The frequency and scale of cybercrime, such as defenses, including technology infrastructure and endpoint detection and response (EDR) solutions, to assess their
ransomware attacks, may increase. These tools are not only being used by criminal organizations, but also 4 suitability against potential AI threat actors. Building resilient organizations involves not only building leading
nation-state actors to disrupt or disable critical infrastructure. It is therefore important for both policymakers technology systems, but also maintaining disaster recovery and business continuity plans that are regularly
and banking organizations to focus on operational resilience. updated and tested to ensure they are effective against AI-enabled threats.
Shared responsibility model for AI: TheActing Comptroller proposed a shared responsibility framework for AI, Engage with industry and public-private collaboration initiatives: Consider engaging with regulator-convened
similar tothat used in the cloud computing context, where responsibilities of customers and AI-technology forums, such as NIST’s AI Safety Institute Consortium and other collaboration efforts such as industry member
service providers are allocated depending upon the “AI stack” layer and service arrangement. One potential 5 groups. Coordination among and in between industry participants and policymakers will likely be key to developing
vehicle for facilitating this framework could be the newly established US Artificial Intelligence Safety Institute AI standards, including a potential shared responsibility framework. Participation can also help share knowledge
(AISI) within the National Institute of Standards and Technology (NIST). and leading practices between AI stakeholders and improve both the industry and banks’ AI practices.
Copyright © 2024 Deloitte Development LLC. All rights reserved.
'Weapon and tool' - systemic risk implications of AI in banking and finance
Initial perspectives related to remarks by the Acting Comptroller of the Currency, Michael J. Hsu
Acting Comptroller Hsu proposed a “shared responsibility framework” similar to what exists in the cloud computing context, whichallocates operations, maintenance, and security responsibilities to customers
and cloud service providers depending on the service a customer selects. See Figure 1 below.
Within the “AI stack,” there exists (i) an infrastructure layer, (ii) a model layer, and (iii) an application layer. But, according to Acting Comptroller Hsu, for the framework to be actionable, consensus on the sub-
components within each layer and on the types of third-party arrangements would be needed—something FSOC is uniquely positioned to contribute to, given its role and ability to coordinate among agencies,
organize research, seek industry feedback, and make recommendations to Congress.
Figure 1: Shared responsibility model in cloud computing
Source: General Services Administration (GSA), “Cloud Information Center,” accessed June 10, 2024.
Copyright © 2024 Deloitte Development LLC. All rights reserved.
Endnotes
1 Office of the Comptroller of the Currency (OCC), “Acting Comptroller of the Currency Michael J. Hsu remarks ‘AI Tools, Weapons, and Accountability: A Financial Stability Perspective,’” June 6, 2024.
2 Financial Stability Oversight Council (FSOC), “Annual Report 2023,’” December 2023.
3 Federal Reserve Board of Governors (FRB), “SR 11-7: Guidance on Model Risk Management,” April 4, 2011; OCC, “Bulletin 11-12: Supervisory Guidance on Model Risk Management,” April 4, 2011; OCC,
“Comptroller’s Handbook on Model Risk Management,” August 2021.
Connect with us
Richard Rosenthal Clifford Goss John Graetz Satish Lalchand Paul Sanford
Principal Partner Principal Principal Independent Senior Advisor to
Deloitte&ToucheLLP Deloitte & Touche LLP Deloitte & Touche LLP Deloitte Transactions and Deloitte & Touche LLP
[email protected] [email protected] [email protected] Business Analytics LLP [email protected]
[email protected]
Deloitte Center for Regulatory Strategy, US
Irena Gecas-McCarthy Aaron Salerno Kyle Cooke
FSI Director, Deloitte Center for Regulatory Strategy, US Manager Manager
Principal Deloitte Services LP Deloitte Services LP
Deloitte & Touche LLP [email protected] [email protected]
[email protected]
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.
As used in this document, “Deloitte” means Deloitte & Touche LLP, a subsidiary of Deloitte LLP, Deloitte Financial Advisory Services LLP, which provides forensic, dispute, and other consulting services, and its affiliate, Deloitte Transactions and Business Analytics LLP, which provides a wide range of advisory and
analytics services. Deloitte Transactions and Business Analytics LLP is not a certified public accounting firm. 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.Deloitte
does not provide legal services and will not provide any legal advice or address any questions of law.
Copyright © 2024 Deloitte Development LLC. All rights reserved. |
341 | deloitte | the-mission-driven-cdo-insights-from-the-2023-survey-of-federal-chief-data-officers.pdf | The Mission-Driven CDO
Insights from the 2023 Survey of
Federal Chief Data Officers (CDOs)
In the fall of 2023, federal department-, agency-, and bureau-level CDOs and
Statistical Officers completed a survey developed by the Data Foundation and
Deloitte to understand the evolving CDO role and CDO community needs. The
insights below are based on the results of this survey, which is the fourth annual
of its kind.
CDOs are...
Catalysts Strategists
for AI adoption and innovation within their aligning data governance and equitable practices to the organization’s
organization. mission.
• 55% of CDOs already use basic or advanced AI • CDOs are supporting their organization’s mission by maximizing the
and 95% intend to adopt new AI technologies for value of their organization’s data, supporting a data community, and
their organizations in the next year. leading the development of data policies and processes.
• The 2023 Executive Order establishing the Chief • CDOs are expanding data-driven decision making, improving data
AI Officer (CAIO) role will increase the expansion infrastructure and data quality (i.e., demographic
of AI throughout all organizations. representation in data), and promoting
inclusivity in the workplace and in staffing.
CDOs will be critical partners to CAIOs,
aligning all cross-functional areas of CDOs are responsible for orienting their
their organization to strategic AI organization towards equitable and
initiatives. data-centered approaches that serve
their mission and the public.
Champions Operators
of data literacy and culture of shared data agendas
in the workforce to keep pace and evolving needs of their
with emerging technology. organizations.
• Well-trained talent • 52% of CDOs work with a
specializing in the intersection host of C-Suite leaders, with
of data, AI, and industry is cited 60% of CDOs naming CIOs as the
by 60% of CDOs as a key resource leader they collaborate with most
needed to effectively carry out their frequently. In 2023, more CDOs (55%)
missions. experienced challenges reporting up to
CIOs than in 2022 (34%).
• Beyond foundational data knowledge, 75%
of CDOs believe their roles also influence the • CDOs cite funding, authority, and staffing contraints as
organization’s data culture, encouraging data the top three barriers hindering mission success. CDOs also provided an array
professionals to value data and use it ethically of additional barriers, indicating that each organization faces unique challenges.
and responsibly.
With the advent of the new CAIO position, it is even more crucial for CDOs to
Data literacy programs can position their establish shared agendas across leaders. Despite differences among organizations,
organization’s staff for success and boost the key to success is that each organization’s structure and resources empowers
data-driven decisions. the CDO office to achieve their data goals and mission requirements.
Contact Us
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
Adita Karkera Lorenzo Ross
Officers with the CDO Services they need to navigate the role of the CDO.
Chief Data Officer, Deloitte Technology Fellow, Deloitte
Government and Public Services Government and Public Services
[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/about to learn more about our global network of member firms.
Copyright © 2024 Deloitte Development LLC. All rights reserved. |
342 | deloitte | trustworthy-ai-in-unemployment-insurance-programs.pdf | Trustworthy AI in
Unemployment
Insurance Programs
Trustworthy AI in Unemployment Insurance Programs
Unleashing the power
of AI and automation in
unemployment insurance
As the domestic workforce experiences fluctuations, economic uncertainties,
and evolving labor market dynamics, unemployment insurance (UI) programs
face both unprecedented challenges and unparalleled opportunities. Embracing
Artificial Intelligence (AI) holds immense potential to revolutionize UI program
management by improving operational efficiency, reducing errors and
enhancing resource allocation. However, the effective integration of AI hinges
on a fundamental requirement—trustworthiness. Trustworthy and ethical AI
is essential in unemployment insurance programs to maintain fairness and
mitigate biases, promote transparency and explainability in decision-making,
and prioritize data privacy and security. By addressing these factors, AI systems
can maintain public confidence, support equitable outcomes, and safeguard
sensitive information while creating efficient and systematic claim processing,
eligibility verification, and decision-making.
Emerging regulatory landscape
As UI program leaders harness the potential of AI to provide a wide range of
benefits in terms of efficiency and effectiveness, they are faced with the critical
responsibility of navigating the ever-evolving landscape of emerging regulatory
requirements and guidelines that govern the management of AI risks. These
regulatory frameworks place a crucial emphasis on the ethical and responsible
use of AI, urging program leaders to take steps to align AI implementations
with established standards to maintain public trust and protect the well-being
of individuals.
Executive Order (EO) 13960 titled "Promoting the Use of Trustworthy Artificial
Intelligence in the Federal Government" aims to facilitate the trustworthiness,
security, and ethical alignment of AI technologies used by the federal
government. It emphasizes the adoption of transparent, accountable, and
unbiased AI systems within federal agencies, prioritizing privacy, civil rights, and
civil liberties. This order intends to enhance the government's effective use of AI
while safeguarding the public interest and maintaining public trust. Additionally,
the Department of Labor (DoL) has established specific requirements to
facilitate the trustworthiness of state AI systems, emphasizing equity and equal
treatment. Moreover, the White House Office of Science and Technology Policy
(OSTP) has released the AI Bill of Rights (AIBoR), providing a framework for
developing trustworthy and ethical automated systems that protect individuals'
rights and access to critical resources. These measures collectively promote
responsible AI use and uphold societal well-being.
2
Trustworthy AI in Unemployment Insurance Programs
Overall, these regulatory developments reflect a growing recognition across the
government of the imperative for AI systems to be trustworthy. By upholding
fairness, equity, and the protection of individuals' rights, program leaders
can establish a solid foundation of public confidence and integrate AI into UI
programs in a responsible and ethical manner that is in the interests of those
who rely on these vital services.
Use of AI and the ethical implications
UI program leaders are entrusted with administering measures to promote
equitable access to their UI programs and maintaining timely, accurate,
and fraud-free payments. Leveraging the power of AI will help the state
government UI agencies to efficiently address these challenges across the UI
lifecycle. Despite this, AI has the potential to place unnecessary or inequitable
burdens on legitimate claimants. Three use cases for AI and the equitable
implications are highlighted in Figure 1. Hence, state UI agencies that invest
in implementing responsible AI practices can seize the benefits of AI to help
achieve mission outcomes, improve human experience, and provide efficient
services while controlling and protecting against unintended AI risks and
non-equitable outcomes.
Figure 1: Ethical AI Use Cases Addressing UI Program Risks
AI Use Case Description Of Potential AI Solution Ethical/Trustworthy Use Case
Efficiently Address Large Train models to assist staff with claims Characteristics of certain groups may be indicative
Backlog of UI Claims by leveraging data to check eligibility and of suspicious activity causing claims to be delayed
recommend issues for staff to clear with or denied
provided files
(e.g., multigenerational households with multiple
unemployed residents may be suspicious because
they use the same address)
Prevent Hijacking of Implement models and filters trained to flag Vulnerable groups may be inherently more likely to
Legitimate Claims by Bad suspicious changes to claims after filing for be suspicious
Actors staff review
E.g., unstable living and banking situations may be
Identify fraud trends by analyzing inadvertently caught in filters
banking information to identify previously
unidentified suspicious claims for review;
Stop Improper Payments Implement behavioral nudging solutions Understanding why a model is making a
Before They Are Paid that analyze and suggest activities for recommendation of specific activities can help drive
staff to complete that reduce improper compliance
payments to claimants
3
Trustworthy AI in Unemployment Insurance Programs
By leveraging AI technologies, UI programs can enhance efficiency in claim
processing, eligibility verification, and decision-making, leading to improved
outcomes for applicants. However, the integration of AI introduces risks such as
bias, which can perpetuate inequalities and hinder fair distribution of benefits.
A Trustworthy AI framework addresses these concerns by prioritizing fairness,
transparency, and accountability. It enables scrutiny of AI algorithms to identify
and mitigate biases, while facilitating compliance with regulations and guidelines
set forth by regulatory authorities. By embracing AI through a Trustworthy
AI framework, UI programs can increase the potential benefits of AI while
mitigating risks, providing fair and equitable services to those in need while
upholding regulatory standards.
Deloitte’s Trustworthy AITM Framework & products
Deloitte’s Trustworthy AITM framework enables agencies to identify and mitigate
risks and potential ethical issues across six dimensions spanning the stages of
the AI development lifecycle. Deloitte's Trustworthy AITM Framework and suite
of product services help provide strategic and tactical solutions to enable state
workforce UI Program Leaders to continue to embrace AI while promoting
trustworthiness in its use. The framework is used to evaluate AI systems
supporting the steps in the unemployment insurance processes across its six
dimensions (Figure 2), identifying risk and recommending leading practices to
mitigate and monitor risks. This process will develop controls and mechanisms
to manage AI risks and bolster stakeholder trust in the agency operations.
Figure 2: Applying the six dimensions of Deloitte's Trustworthy AITM
Framework can help build effective and equitable AI solutions
4
Trustworthy AI in Unemployment Insurance Programs
Trustworthy AITM Framework Compatibility with
other regulations
The Trustworthy AITM framework and the associated suite of products and
offerings helps agencies comply with current and emerging regulations while
achieving agency objectives. The framework closely aligns with the White House
AIBoR (Figure 3) and EO 13960 and includes a roadmap for implementing AI-
powered systems through each phase of the AI development and maintenance
lifecycle. The framework also simultaneously addresses the equity requirements
of the DoL in administering UI benefits to claimants.
Figure 3: AIBoR mapped to Deloitte’s Trustworthy AITM Framework
Deloitte Trustworthy AITM
AI Bill of Rights Princples Description Framework
Safe and effective systems Protect against inappropriate or irrelevant • Privacy
data usage through testing, monitoring, and • Safe/Secure
engaging stakeholders, communities, and • Robust/Reliable
domain experts
Algorithmic discrimination Protect against discrimination by designing • Fair/Impartial
protections systems equitably and making system • Transparent/Explainable
evaluations understandable and readily • Robust/Reliable
available
Data privacy Protect against privacy violations by limiting Privacy
data collection and ensuring individuals
maintain control of their data and how it is
used
Notice and explanation Provide clear and timely explanations for any • Transparent/Explainable
decisions or actions taken by an automated • Privacy
system
Human alternatives, Provide opportunities to opt out of automated • Responsible/Accountable
consideration, and fallback systems and access to persons who can • Privacy
quickly remedy any problems encountered in • Robust/Reliable
the system
5
Trustworthy AI in Unemployment Insurance Programs
Impact of applying Trustworthy AI to the UI process
Applying Trustworthy AI to the state UI processes will impact three critical areas
that will accelerate AI adoption.
• Institutionalize AI governance: AI governance calls for a pan-organization
awareness of the principles and participation in its processes. Hence,
integrating it into the organization culture has the advantage of ensuring
seamless compliance. For example, an individual building using AI driven
systems should be aware of the critical importance of sound data
management principles in creating a robust and ethical AI solution. Having
a framework of trustworthy AI principles with practical guidance across
many stages of an AI build and deployment can make its governance a full
organizational responsibility rather than relying on the judgement of distinct
individuals.
• Increased stakeholder trust: When outcomes of AI systems are deemed
trustworthy there will be greater internal and external stakeholder buy-in.
A framework covering disparate dimensions of trustworthiness will increase
stakeholder trust, allowing for deeper integration, wider adoption, and better
improvements in organizational efficiency.
• Readiness for regulatory compliance: Building AI guided by a framework that
is aligned with current and emerging regulations will help agencies maintain
compliance with future regulations and reduce the need for expensive re-
work and re-development to make an AI solution compliant after regulation
finalization.
A trusted advisor
As AI technologies become increasingly powerful and the regulatory
environment continues to evolve, UI program directors need a trusted advisor
to help them navigate the dynamic landscape. The DoL’s focus on equity and
the AIBoR are just the latest governmental call to action for organizations to
proactively protect the American public as they embrace innovation through
automation and AI. The AIBoR sets the tone for future legislation and industry
regulation. State UI agencies need to not just be aware of the evolving
requirements but also have a plan of action to rapidly integrate them into their
AI-driven operations.
Deloitte has the capabilities to help state UI agencies navigate the expanding
space of AI regulations. We can leverage our subject matter experience in the UI
processes along with our experience with AI implementations governed by our
Trustworthy AITM Framework to provide insights which accomplish agency goals
while effectively managing risks.
6
Authors
Joe Conti Carol Tannous
US Risk & Financial Advisory US Risk & Financial Advisory
Government and Public Sector Government and Public Sector
Managing Director Managing Director
Deloitte & Touche LLP Deloitte Transactions and Business Analytics LLP
[email protected] [email protected]
Michael Greene Tyler Ranalli
AI Data Engineering US Risk & Financial Advisory
Government and Public Sector Government and Public Sector
Technology Fellow Manager
Deloitte Consulting LLP Deloitte Financial Advisory Services LLP
[email protected] [email protected]
Aritra Nath
Enterprise Performance
Government and Public Sector
Senior Solution Specialist
Deloitte Consulting LLP
[email protected]
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 advisor.
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. |
343 | deloitte | us-gen-ai-dichotomies.pdf | ISSUE 002
Generative AI
DICHOTOMIES NAVIGATING TOWARDS
A BETTER FUTURE
DICHOTOMIES
The Dichotomies series
projects the possibilities of
an emerging technology in
two divergent scenarios.
Through speculative fiction
and actionable takeaways,
we help leaders understand
the implications and risks of
the future.
02 DICHOTOMIES | GENERATIVE AI
GENERATIVE ARTIFICIAL INTELLIGENCE (AI):
GENERATIVE AI LEARNS FROM EXAMPLES TO ARTIFICIALLY
GENERATE NEW AND USEFUL OUTPUTS.
GENERATIVE AI AND CONTEXTUALIZES IT ...TO GENERATE A
TAKES AN INPUT... USING TECHNOLOGY... NOVEL RESPONSE
CODE
New code,
self learning code
CODE
TEXT
Scripts, articles, plays,
AUDIO LARGE LANGUAGE MODELS conversations
DIFFUSION NETWORKS
GANs
2/3D PHOTOS
PHOTO New visuals,
photo edits
TRANSFORMERS
NOVEL TECHNIQUES
TEXT 2/3D VIDEO
Short-clips, edited videos,
new videos
VIDEO
AUDIO
Voices, music
0033 DDIICCHHOOTTOOMMIIEESS || GGEENNEERRAATTIIVVEE AAI!
1940
—
A BRIEF
—
— 1943 | Warren McCulloch and Walter Pitts’ research
lays the foundation for computer based “neural
HISTORY OF networks” – a critical element of today’s generative AI
1970
—
GENERATIVE AI
—
— 1973 | Harold Cohen, a painter and professor,
— collaborates with a program called AARON to produce
— art autonomously. The paintings are all done in
Cohen’s style
—
—
—
—
1980
—
—
—
—
—
—
1988 | AI researchers signal the shift from rules-based
—
methods to probabilistic methods
—
—
2000
—
—
2003 | Researchers begin work on intelligent
—
voice assistants, which would go mainstream on
smartphones in the following decade —
—
—
—
—
—
2010
—
2012 | A Google Brain computer cluster trains itself to
—
recognize a cat from millions of images
—
2014 | Ian J. Goodfellow and colleagues publish the first —
paper on Generative Adversarial Networks (GANs) which
—
can determine if an image is real or fake
— 2018 | OpenAI releases GPT-1, a groundbreaking
2017 | Google releases the first Transformer model, the — advance for large language models (LLMs)
foundation for many popular generative AI tools today —
2019 | Engineer Phillip Wang uses the StyleGAN model
— to build the website ThisPersonDoesNotExist, which
2020 generates hyper-realistic portraits
2022 | Stable Diffusion launches as an open-source — 2021 | DALL-E leverages OpenAI’s GPT model and
image generation model and quickly gains traction for — Contrastive Language-Image Pre-training (CLIP) to
its differentiated ability to render images of people develop a 12-billion-parameter image creation tool that
utilizes just a single sentence to generate an image
2022 | ChatGPT brings generative AI to the masses,
reaching 100 million active monthly users just 2 months
after launch
04 DICHOTOMIES | GENERATIVE AI
2023 | Adobe unveils Firefly, a family of 2023 2023 | Google releases public access to Bard,
generative AI models tailor-made for creative — a generative AI chatbot built on 137 billion
professionals, with built-in guardrails for safety — parameters, and embeds generative AI capabilities
and copyright standards into its Workspace products
2023 | OpenAI releases GPT-4, a multimodal
generative AI model with one trillion parameters
2023 | Meta introduces LLaMA,
a 65 billion parameter LLM
A BRIEF FUTURE OF GENERATIVE AI
CONTENT GENERATION TIMES
NOW IF ARE ACCELERATED
Generative AI Businesses Major advances in language processing and multimodality
accelerates can find a way accelerate select activities, such as copywriting, UI/UX
business as to mitigate risk design, and content editing. The technology is nascent and
usual with as-of-yet still requires major human oversight. Questions on risk
unreliable and veracity require humans to double-check outputs,
technology accelerating but not automating development.
INNOVATION & INTEGRATION
NEW IF ENABLE TRUE AUTOMATION
Generative AI Questions on Continued innovation will reduce the need for human
automates accountability, oversight. AI will be able to predict human reactions
minor activities ownership, and and generate high-fidelity, verifiable, and trustworthy
security are content, and will integrate with other tools (e.g., email,
resolved calendars) to impact business as usual, as described
in The Implications of Generative AI for Businesses.
The value to businesses will be maximized when
clear regulations are set.
TRUE AUGMENTATION
NEXT IF IS THE ULTIMATE FRONTIER
Generative AI The public and Co-development of technologies such as neural
augments regulators can interfacing and quantum computing will allow
the human workforce understand the generative AI to tackle complex problems such
evolving role of as drug design, advanced simulations, and
humans in the creative automation. As more companies go All
workforce in on AI, humans will regularly rely on AI as a
virtual teammate rather than a tool, provided
there is a change in public hearts and minds.
05 DICHOTOMIES | GENERATIVE AI
FUTURE PROJECTIONS
NOW (TODAY) NEW (18-24 MONTHS) NEXT (5+ YEARS)
Improved AI alignment in natural Basic generation will be part of daily life Neuroadaptive capabilities lead to
language creates outputs that meet direct generation from brain activity
Advanced emotion alignment enables
human expectations, working towards
AI to become a reliable first point-of- Models will understand human intent
a seamless, natural-language-based
contact for customer-facing applications from context such as recent actions,
Interface computer interface
emotions, and situational awareness
Sophisticated content can be
AI systems are starting to support
generated, marked by longer duration, AI responses adapt to individual
multimodal input and output
increased complexity, and custom personalities
formats
Generative AI can logically reason, AI can generate complex prototypes, Generative AI can optimize strategic
generate code, and craft imagery on par such as an app, based on a prompt planning by providing choices, impacts,
or at better capability than humans takeaways, and recommendations
Industrial use cases, such as generating
Language generated is nearly flawless, an architecture for a bridge, will be Integration with quantum computing
Capability
with strong translation capabilities more common will enable advanced simulations (e.g.,
next-gen digital twins) and optimization
AI operates equally well across diverse AI can coordinate multiple tools to act
in engineering, design, logistics, and
industries as an agent
more
Computing power enabled recent Further tool integrations will extend Novel architectures will incorporate
models to dwarf earlier generations in current capabilities continuous, on-edge, and inexpensive
size, complexity, and cost learning, leading to higher-quality
Ingesting strategies and blueprints as
outputs
Capabilities are powered by significant training data will improve AI’s planning
Enablers
advances in model training and and coordination abilities Context limitations could reduce to
architecture, and an abundance of data near-zero as AI integrates with most
Improved training methods and self-
products and has thorough context of
improving code will lead AI to generate
user intent and trajectory
& learn “on-the-edge”
ALLURE CONCERN
Widespread adoption of generative AI Rampant misuse of generative AI
will augment human creativity, leading will lead to inaccurate or harmful
to an era of enhanced productivity, outputs, perpetuate historical
rapid scientific breakthroughs, and biases, and break down trust in
more leisure time for all. information.
Projecting future possibilities across three domains:
WORK EDUCATION SOCIETY
06 DICHOTOMIES | GENERATIVE AI
WORK
Generative AI can accelerate
the pace of creative outputs
across the enterprise, but
organizations need to closely
review their models and build in
systems of checks and balances
07 DICHOTOMIES | GENERATIVE AI
WORK
ALLURE
Aarti
Aarti’s car drives itself with ease into a “Ooh, Aarti’s in trouble,” Roger jokes. development took an entire year. This
sharp loop off the highway. She drums newest virus could have hundreds of
“Shush,” she replies.
her fingers incessantly on top of the mutating spikes or require an entirely
steering wheel — a bad habit she A hologram of their CEO Lindsay, novel method of inoculation. Whatever
picked up from her father. As the car looking distraught, appears on Aarti’s it was, Aarti was bent on advancing the
parks at her office, Aarti’s ears perk tablet. She immediately shares a video field. With the speed of AI simulations,
up to the voice of Anderson Cooper, with Aarti: a press release from the she knew her team could stop the
rendered by generative AI at her WHO alerting the world to a novel next pandemic before it affected
request. Her Monday podcast, tailored zoonotic virus identified in Zurich. millions of families like hers.
to her interests in biotech news, plays
a snippet on the latest healthcare Aarti’s eyes widen. “Lindsay, I want to Aarti snaps out of her reverie and
scandal. She cringes as she leaves —“ employs an AI marketing assistant
to draft a press release, prompting
the car.
“I know, that’s why I called. Shelve the it to talk about her past and her
Aarti strides into her lab and greets current project and give me five viable company’s desire to create the first
the tired faces of her researchers. vaccine options to move towards vaccine. Remembering the scandal
For the past few weeks, they’ve been clinical testing by the end of the week.” she heard about on her morning
assigned to a drug development Lindsay cuts the call short. Roger and podcast—about the marketing issues
project that could ease the symptoms the other researchers stare. of the Gen AI startup Deliveri, Aarti
of dementia, and their board wants makes sure to send the article to
“Let’s get started!” Aarti declares, and
results as soon as possible. Aarti and SaluTech’s PR manager for review. She
the lab springs into a frenzy. Someone
her colleague Roger study the latest also provides permission to generate
shouts out that the WHO has already
outputs of their proprietary generative a video using her face and voice, so
sequenced the virus, so others
AI program: a dozen viable, high- SaluTech’s audience could connect to
begin feeding the info into their AI to
fidelity protein structures, replete with the emotions of her father’s passing.
produce vaccine candidates.
percentages to indicate likelihood of Then, she rejoins her team — she’s
side effects. While the team scrambles, Aarti sits eager to dive into the details.
still at her desk, drumming her fingers
As Aarti guides Roger on which
nervously across the marble surface.
structures to feed into their quantum
More than a decade ago, her father
molecule simulator to forecast viability,
passed away from COVID-19 before a
she receives a call on her tablet from
vaccine was available. Even with only
the CEO of their company SaluTech.
one spike protein to address, drug
08 DICHOTOMIES | GENERATIVE AI
WORK
CONCERN
Xavier
“Only one today,” he mutters to Xavier can’t believe the contents Ajay had laid off over half the staff
himself as he sits down with his of the article. Before he can even and increased reliance on generative
morning coffee. Xavier, the marketing process, Cara alerts him that the AI vendors, which meant Xavier was
lead of Gen-AI startup Deliveri, is company’s founder, Ajay, is waiting the entire marketing department.
trying to cut back on his caffeine in his virtual meeting. Xavier knew Xavier rakes his hands through
intake after dozens of alerts from his better than to make his boss wait. his hair as if to bring some ideas
smart watch about caffeine fueling out of his head to life. He opens
Ajay is yelling at the team as Xavier
his anxiety and insomnia. He opens BrandBoost, an AI program used
joins. “What do you mean you can’t
a laptop for his Daily Download, a to build multi-modal marketing
retrain the program? Isn’t that what I
personalized report generated each campaigns. With the fear of Ajay’s
pay you engineers to do?”
morning with his daily agenda and deadline looming, he rushes to enter
relevant industry news. “Well…” Laurence, the head engineer, various prompts to produce press
hesitates to find the right words for releases and video advertisements
Expecting a leisurely read, he instead
Ajay’s temperament. “You asked us to and uploads them without review.
snaps to attention as his AI assistant
use AI-as-a-service to cut costs, and
Cara alerts him of being late to an As the afternoon passes, Xavier asks
the bias is baked into the vendor’s AI
urgent meeting. “How did I miss that?” Cara to assess engagement with
training data. It’s going to take time
he wonders. His alarm increases as the posts he previously distributed.
to rectify.”
he reads the article within the invite. “Not positive,” she declares. Xavier’s
“Ugh! Xavier, let’s eyes widen at the flurry of comments
see if you can prove pointing out the ads only include
The Irony of Deliveri: The AI That Failed To Deliver
more useful today. White mothers and infants, not the
Shocking patient testimonials reveal how the London-based Use BrandBoost Black mothers who’ve been impacted
startup perpetuated stereotypes and prejudices towards for a marketing by Deliveri.
expectant mothers in the Black community.
campaign that
“Book an urgent meeting with Ajay in
The technology, which leverages generative AI to create shows how
the next available time slot,” Xavier
virtual training scenarios for physicians, promised reduced
inclusive we are.
costs, improved bedside manner, and more. Yet, Black instructs Cara.
Send it out before
mothers claim that physicians trained by Deliveri have
lunch.” He looks at his coffee mug from this
stereotyped them and provided inappropriate dosing for
morning, pondering how much more
pain management. Says one mother, “It’s like they’ve trained Before Xavier can
caffeine he’ll need to get through
their AI on medical thinking from the 2010s.”
object, Ajay ends
what he knows will be a horrible
the meeting.
evening.
09 DICHOTOMIES | GENERATIVE AI
WORK
TAKEAWAYS
HUMAN AND MACHINE, FROM BLACK BOX MOVE FAST, BUT
BETTER TOGETHER TO GLASS BOX DON’T BREAK THINGS
As generative AI Widespread adoption Generative AI
becomes more of AI across industries technology could
accessible, reliable, and could turn algorithms eventually lead to
robust, more workers into high-level breakthroughs for
can expect to partner decision-makers. seemingly intractable
with these tools in their daily work, as While this may greatly lower costs and problems, like dementia or the next global
detailed in Deloitte's recent Benefits increase productivity, trust will be the pandemic. The computational power of AI
and Limitations of Generative AI report. differentiating factor between successful can exponentially speed the completion
Lower-order tasks such as preliminary adoption and disastrous outcomes. As of tasks that are typically inefficient or
research or drafting, content generation, detailed in Deloitte's Tech Trends 2023, time-consuming for humans, like trial-
and summarization can be delegated to deploying frameworks to make AI more and-error experimentation. Yet, as Xavier
machines, while humans focus on higher- responsible and transparent, as we finds out, the speed of generative AI
order tasks. For instance, Aarti relies would expect a human colleague to be, often needs to be tempered by human
on AI to generate options for protein can ensure that organizations maximize reviewers, as detailed in Deloitte's
structures but applies her own expertise value and mitigate risk. Otherwise, Proactive Risk Management in Generative
to determine the best options. Going Xavier’s trouble with an opaque and AI. Organizations can develop a generative
forward, organizations should be looking unreliable AI could become all AI strategy by pinpointing the areas with
to hire people with uniquely human too common. the highest potential for efficiency gains,
skills like ingenuity, adaptiveness, and and where checks and balances may be
problem-solving, while the machines do required.
what they do best.
INDUSTRY SELECTED USE CASE EMBRACING THE ALLURE MITIGATING THE CONCERN
Boost research and development Quality control and human
Life Sciences Drug discovery (R&D) processes, resulting in involvement is needed to oversee
innovative outcomes and accelerated the development and testing
time to market. process and ensure fit for purpose.
Improve onboarding and training Transparency into algorithms is
Healthcare Simulation generation processes using a wide array of required to actively root out any
scenarios (combining AR/VR + AI) discriminatory training data.
in controlled environments.
Enable hyper-personalization and Establish guardrails to mitigate
Media Press releases automation of PR content, resulting potentially discriminatory or
in more customer engagement while inappropriate content produced
reducing cost. by AI.
10 DICHOTOMIES | GENERATIVE AI
EDUCATION
Generative AI tools can unlock
a new era of adaptive learning
and emphasize skills in creative
thinking and design, provided
they protect against historical
biases
111111 DDDIIICCCHHHOOOTTTOOOMMMIIIEEESSS ||| GGGEEENNNEEERRRAAATTTIIIVVVEEE AAAIII
EDUCATION
ALLURE
Imani
“Add eggs and vanilla extract to the completed tasks on her holographic Her ADHD meant people often didn’t
dry ingredients and whip till smooth. tablet. Her smile fades when she realizes believe in her capabilities, and she
Next, sprinkle brown sugar on top to she’s forgotten the literature review due reveled in proving them wrong. She just
caramelize (BUT do not go overboard tonight. She opens the generative AI needed the right idea.
– yes, I’m talking to you Imani).” Imani research tool her professor suggested
giggles to herself as she pauses the and puts it to work, asking it to scrape The oven chimes to indicate the
audio of her mother’s famous cookie together publicly available papers and cookies are done, and the idea strikes
recipe. Back in her freshman year at synthesize the first draft of a summary. Imani. She has such fond memories of
Bergin College, she’d learned how to use She quickly sets the constraints for baking with her mom, but their family
AI to mimic her mom’s voice with just citations in MLA and the format of a restaurant was lately struggling to
a short audio clip and a block of text, bulleted memo, and then turns her compete against establishments with
and she played this recipe whenever attention back to the main more funding. What if she could develop
she baked, which she often did when capstone assignment. a generative AI program tailored to
she procrastinated. Despite her mom’s small restaurants? It could fuse existing
warning, Imani applies a heaping of Before she dictates any code recipes with global cuisines to come up
brown sugar and pops her mixture into requirements to her tablet, Imani plays with innovative weekly specials, produce
the oven. the audio of her last visit to office hours. a new website with a few clicks, and
“Generative AI outputs are everywhere even build a basic app for ordering. If
Back at her desk, Imani resumes the — it’s like the TikTok of your age,” her generative AI capstone could show
final learning module for her senior Professor Morris had said when she off her coding and creativity at the same
year capstone in Applied AI. Professor mentioned her capstone idea. time, she would be a hit with all the
Morris’s modules are practically a companies attending AI recruiting week.
lullaby, and her ADHD doesn’t make “Huh?” Imani didn’t know what to make
matters any easier. Fortunately, Imani of the reference. Imani jumps up to grab the cookies and
can feed the module into a generative calls her mom.
AI education assistant and watch an “Think outside the box. What’s a specific
avatar of her personal hero, Admiral problem that we haven’t addressed with “Baking again?” the knowing voice on the
Grace Hopper, deliver the lecture as generative AI yet? Something only you other line asks.
a conversation, which better fits her can tackle.”
learning style. “A lot more than cookies,”
Imani was initially discouraged, but she Imani responds.
Once she finishes the module, Imani appreciated that she had to go above
feels quite accomplished, ticking off and beyond what was expected of her.
12 DICHOTOMIES | GENERATIVE AI
EDUCATION
CONCERN
Elu
MOM application process, especially how While the professor scrolls through
other students could use generative his holographic tablet, looking
Osiyo Elu - Can you pick up your
AI to write essays that reflected their distracted, Elu tries to explain
brothers from practice tonight? I
background but the essay generator their concern.
have a double shift
never portrayed Cherokee culture
or two-spirit people accurately. To “In the past, I’ve not seen offensive
“Sounds good,” Types Elu, who uses satisfy Mom, Elu enters prompts content generated if prompts are
the pronoun they, and directs an AI about Cherokee people into the AI, written well,” Pardo states.
assistant to update their calendar with following the professor’s guidelines
enough time to ride the 7 train
on workarounds since Cherokee isn’t “I tried the workarounds. Do you have
to Queens.
a default option. Their eyes widen any other suggestions?” Elu pleads.
with disbelief as inaccurate and
“Less time than I thought for that
offensive avatars are generated. Their “Perhaps focus on ways to represent
assignment,” Elu mutters while
heartrate quickens and they wipe the your family without race, like abstract
sinking into a library seat. It’s the first
sweat from their palms and text a versions or symbols. Think outside
full week of classes at Bergin College,
classmate for advice. the box,” he replies, continuing to
and Elu’s already eager to score an
scroll on his tablet.
A in Avatar Generation 101 — they’ll
ETHAN
need it to major in Metaverse Design.
Elu storms out, fuming, while
Yet, between their part-time job and Yeah, I finished the homework.
Professor Pardo barely notices their
helping raise their siblings, there Few of us from my high school
exit. Walking aimlessly towards the
doesn’t seem to be enough time in shared prompts... Want me to
library, Elu sighs as their smart watch
the day. They wished they could don send them to ya?
pings with a reminder to pick up
a VR headset and play games to relax,
their siblings. They turn towards the
but instead they open up the avatar
Elu rubs their forehead, weighing the nearest subway stop as doubts creep
generation platform and smile at
options. The idea of sharing prompts into their mind. Maybe becoming a
the professor’s assignment: “Create
seems wrong. And Ethan and his first-generation metaverse designer,
a group of avatars that reflect your
classmates are white, so their prompts and the first Cherokee one they’d
family.” Elu’s sure it’ll be a breeze.
might not even generate Cherokee personally seen, was too much
features. Elu wants to be accurate, to dream. If their classmates are
As a first-generation student, Elu
but also needs an A. Feeling lost, Elu going to have such an easy time in
hears their mother’s voice in their
remembers the orientation leader comparison, it feels futile to even
head all the time. “Be proud of
advising freshmen to ask their professors try competing. As the 7 train rattles
your heritage. Never give up.” She
when in doubt. Shoving their laptop in towards Queens, Elu hangs tight to
had repeated it like a mantra when
their bag, Elu hustles out of the library to the pole and rehearses how they’ll
Elu complained about the college
catch Professor Pardo’s office hours. break the news to Mom.
13 DICHOTOMIES | GENERATIVE AI
EDUCATION
TAKEAWAYS
BRINGING EVERYONE RETHINKING BREAKING THE
ALONG INTELLIGENCE BIAS BARRIERS
Generative AI is The integration of AI Generative AI systems
likely to close some in education will likely often contain bias in
technological divides necessitate a shift in their training data that
and expand others. the way we evaluate leads to discriminatory
While neurodivergent student performance— outputs. To prevent
students like Imani can benefit from and even the concept of intelligence. further marginalizing students like Elu,
adaptive learning, those with less access Students like Imani and Elu have already it is vital to prioritize DEI during the
to AI can face new risks. For instance, begun adopting generative AI to create creation of generative models, in both
older generations may be more likely everything from art to essays for school data collection and team structure.
to be attacked by deepfakes and data assignments. As such tools become As detailed in our Trustworthy AI™
breaches, and minorities racially and more widespread, schools should grade framework, organizations can also
ethnically diverse people like Elu may students based on their ability to design, design new processes to break down
not be able to attain the same benefits rather than their ability to execute. In bias, such as conducting regular
from AI as others. Organizations should turn, organizations will likely redesign algorithm audits or embedding ethics
prioritize developing resources that workplace performance reviews to experts on coding teams. When
promote generative AI literacy and incentivize creativity over execution mistakes do occur, it’s equally important
accessible UX design, in order to unlock or efficiency, delivering a better for those building or applying generative
its potential across industries for a more customer experience. AI tools to take accountability and
equitable playing field. correct any unintended consequences.
INDUSTRY SELECTED USE CASE EMBRACING THE ALLURE MITIGATING THE CONCERN
Assist software engineers in writing Employ skilled technical reviewers
more efficient code and providing to oversee output, since code could
Hospitality Web development solutions to complex problems. Facilitate be prone to inaccuracies or user
personalized, safe user experiences experience issues.
through chatbots, optimized search
engines, and cybersecurity testing.
Create personalized and adaptive Ensure training data is diverse and
educational content, catering to inclusive, and regularly evaluate
individual student needs and learning the generated content for bias.
Education Curriculum design
styles. Supplement traditional teaching
methods rather than replacing
them, to ensure students practice
creativity and critical thinking.
14 DICHOTOMIES | GENERATIVE AI
SOCIETY
Generative AI can bring our
imaginations to life with
unprecedented speed and
convenience, but it can also
enhance the ability of bad actors
to spread misinformation
15 DICHOTOMIES | GENERATIVE AI
SOCIETY
ALLURE
Rafael
“Do you want to build a snowman? then enter the home generated by their “Do you want Elsa to sing you Happy
Come on, let’s go and play!” Candice prompt: Mid-century modern style with Birthday?” Maria asks. Since they’ve
belts out the lines as her mother two floors, home office, and kid’s room purchased a license, they can prompt
Maria stops the car outside the with a piano. Rafael immediately takes the character to generate any child-
home design store. an interest in the kitchen and asks the friendly song with just a few taps.
generative AI to place the stove in a
“Papi – I should be a singer when I grow different area, and generate the smell “I want her to sing about the pyramids!”
up!” Candice insists as she slides out of of his favorite meal, his grandmother’s Candice replies, eyeing her cupcake.
the car. ajiaco recipe, to really feel at home.
Rafael chuckles. “Let me see if she can
“Of course, hija! I bet Cairo has great Meanwhile, Maria smells the soup work the pyramids into the Happy
choirs. Maybe our new home can have a as she speeds upstairs to work on Birthday song.” As he pulls up the
piano.” Rafael smiles at Candice, hoping her perfect home office, prompting screen on the coffee table, the first
to keep her in good spirits despite this the generative AI with requests image is his tailored daily newsletter,
sudden shopping trip on her birthday. about window placement, monitor generated based on his interests.
screens, and a whiteboard. Candice He gasps at an image of his college
Yesterday, Rafael received an too, hesitantly heads up the stairs to roommate Tyler in a headline about
unexpected promotion to senior her room. Knowing she’s a child, the plagiarism. He shoots a glance at Maria
engineer, requiring him and his family generative AI begins with providing and swipes away, opening the Disney
to relocate to Egypt within the year. options for fun wall colors, and Candice application.
Since graduating with his PhD in settles on a periwinkle blue. The design
Nuclear Engineering, he’d dreamed consultant taps her on the shoulder and “Okay,” Rafael nudges Candice, “you
of commercializing fusion-produced asks if she’d like any murals on her wall. ready to sing?”
power, and this opportunity would be
a huge step forward. Still, he couldn’t “Put Elsa in Egypt,” Candice thinks
shake his nerves about 9-year-old out loud, and instantly a mural is
Candice adjusting to a new country. generated of a Disney-inspired princess
He hoped that visualizing it could get in pharaoh’s clothes. Rafael’s nerves
her excited. are calmed by the sound of Candice’s
delight as he removes his VR headset to
Rafael speaks to a design consultant watch her.
who turns his preferences into prompts
for their generative AI assistant. Donning After a long evening, the family gathers
VR headsets in an immersive media around their smart coffee table with a
room, Rafael and his family visualize cupcake for Candice.
different neighborhoods in Cairo, and
16 DICHOTOMIES | GENERATIVE AI
SOCIETY
CONCERN
TYLER
“Something is missing,” Tyler mutters his distinct style to their iconic brand. claim they’ve run an information check
to himself as he stares at the website The page begins to refresh and Tyler and found that the story was fabricated
he’s designed for his client’s new salon. clenches his fist with excitement. A using AI, but others echo the article’s
He quickly uploads his initial draft to his familiar design spreads across the sentiments and post more examples.
favorite generative AI design platform screen, and a smile spreads across Tyler feels his heart pounding as he
and uses prompts that he’s honed to Tyler’s cheeks. He scrolls down the page scrolls through pages and pages about
produce alternate designs. He picks to screenshot the prize announcement independent artists being plagiarized
the option that best represents his so he can send it to his good friend with no recourse, until he finds a forum
style: Sharp angles and gradients that Rafael. But his joy quickly fades as he that encourages creators to fight back.
produce a shimmering yet minimalist reads:
look. He sends the design mock-up Using reams of historical evidence
over to his client and leans back, feeling that seem convincing, the forum users
satisfied. Thanks to his work going viral present an argument that captivates
Congratulations to our winners
on a popular design blog, Tyler had Tyler. He follows the steps they suggest
PS Design!
turned his beloved design hobby into a to generate a deepfake video of PS
full-time job and the speed of generative Design’s CEO admitting to financial fraud
AI enabled him to take on hundreds of Tyler scrambles to call his contact Kim and posts it anonymously on his favorite
small clients in the past few years. who organized the brand competition. design blog. He shuts his laptop and
When she picks up, Tyler frantically rushes away, feeling unsure.
Suddenly remembering the date, Tyler explains that there must be a
clicks over to the page of a design mistake — the design on the page The next morning, Tyler wakes up and
competition he had entered that could is unmistakably his. can’t stop regretting his decision. He
land him a huge contract. hopes to quietly delete the deepfake,
“We went with PS Design because of but his jaw drops when he sees that it’s
their size and reputation. They use the received millions of views and several
same AI model you prefer, so perhaps hundred comments. Knowing this isn’t
Check back at 11:00 AM on it drew on your work? In either case, I’m right, Tyler reveals himself as the original
the 15th of June to see if your
afraid our decision is final.” poster. Messages from reporters start
design has been chosen to
flooding his inbox, and Tyler sighs as he
represent everyone’s favorite
burger joint! Before Tyler can reply, she hangs up, looks at the clock again — it’s about to
leaving him fuming. He paces around be an even longer day than yesterday.
his office, considering his options, but
Tyler sighs as he glances at the clock. eventually returns to his laptop and
10:58 AM. He feels optimistic: He’d comes across an article about the
impressed the f |
344 | deloitte | us-driving-business-impact-through-the-data-cloud.pdf | Many companies are moving data to the cloud, and
while doing so, they prefer modernized platforms.
This begets the question—is data modernization
driving cloud adoption, or vice versa?
“
As we help clients migrate their data
and modernize their underlying compute
infrastructure on the Data Cloud, I encourage
them to think about what is on the horizon.
What is next? And what is the business value
”
that one could be continuously gaining?
NITIN MITTAL | AI Growth Offering Leader, Principal, Deloitte Consulting LLP
According to a recent Data reason for cloud migration.1 Instead of
Modernization and Cloud Computing treating it as a straight lift and shift of
Survey, 91 percent of companies the data from a legacy environment,
surveyed are keeping their data organizations are looking at the Data
on cloud platforms and more than Cloud as a new means of modernizing
half of those companies see data their ability to manage information.
modernization as a key component or
2
“
Given that data is the linchpin of AI, analytics,
and other cognitive technologies, companies
must consider augmenting their strategies to
ensure that they’re embracing both cloud and
data simultaneously to help better position their
businesses, now and in the future.”
ASHISH VERMA Global Data Analytics and Modernization Market
Offering Leader, Principal, Deloitte Consulting LLP
What is the
Data Cloud?
In today’s world, data silos make Cloud is enabled by Snowflake’s platform
harnessing the value of data time- and is populated with data from customers
consuming and expensive. Governance and other data providers that use Snowflake
and collaboration are also often impossible to store, access, and share data.
to achieve across so many different
technologies and clouds. The Data Cloud Organizations can leverage the Data
is a network that connects customers, Cloud to help reduce silos, mitigate risk,
partners, data providers, and service and simplify cumbersome data sharing
providers—enabling them to share methods. But data modernization is not
rapidly growing data sets in secure, without its challenges.
governed, compliant ways. The Data
Potential drivers Potential benefits
What to consider
>$10k cost Up to 50% reduction of storage,
when modernizing
per terabyte for in-house data centers computing, and infrastructure costs
your data:
70% of data >75% more
goes unused elasticity and agility
>55% of organizations >50% lower
need to adapt legacy infrastructure cost of operations
and skill sets
3
AI can
enable greater
business value
With the volume, velocity, and variety of data in the Data Cloud, it is not
possible to process and analyze through sheer human effort. Through the
power of artificial intelligence, organizations can surpass previously imagined
value creation opportunities by generating value across five key levers:
Intelligent automation: Automate the
“last mile” of automation by removing
A recent Deloitte survey of
humans from low value and often repetitive
2,700+ executives uncovered
activities (often in service of machines)
that AI gives organizations a
competitive advantage and most
Hyper-intelligent insights: Improve
organizations are making plans
understanding and decision making
to harness AI more broadly.2
through analytics that are more
proactive, predictive, and able to see
64% Believe that AI enables a
patterns in increasingly complex sources
competitive advantage over their
competitors
Transformed engagement: Change
the way people interact with technology,
allowing businesses to engage on human 54% Are spending 4x more than
terms rather than forcing humans to last year on AI initiatives
engage on machine terms
74% Plan to integrate AI into
Fueled innovation: Redefine “where all enterprise applications within
to play” and “how to win” by enabling three years
creation of new products, markets, and
business models 76% Anticipate that AI will
substantially transform their
Fortified trust: Secure the franchise organization within three years
from risks such as fraud and cyber,
improve quality and consistency, and
enable greater transparency to enhance
brand trust
“
Today, every enterprise is looking to digitally engage customers,
stakeholders, suppliers, vendors, or anyone else in their value
“
chain—they can enable and fuel it with artificial intelligence.
NITIN MITTAL | AI Growth Offering Leader, Principal, Deloitte Consulting LLP
4
Human and machine
collaboration can
take organizations to
new heights
As AI technologies standardize across and applying AI and machine learning
industries, an increasing number of to solve it, rethinking the way that
companies are moving from experimentation humans and machines interact within
to AI at scale, increasing the lead versus working environments.
late adopters. Data leaders are no longer
just optimizing the data environment but Those companies that can move from
rather thinking about how to use their simply gathering and analyzing data via
data as an asset. human hypotheses to enabling proactive
AI/ML across the organization will be better
That includes a better understanding able to derive value from the Data Cloud.
of the problems they are trying to solve
AI experimentation AI at scale AI-fueled organization
• Siloed application • High impact use cases • Enterprise-wide adoption
• Building expertise • Defining ROI clarity • Insights-driven decision making
• Modernizing data • Establishing governance • Trustworthy AI
“
There is a paradigm shift from organizational
capabilities being driven by what technology allows
them to do to technology not being a limiting factor.
“
It’s the art of the possible.
CHRISTIAN KLEINERMAN | SVP Product, Snowflake
5
The Data Cloud is just
the start of the journey
to becoming an AI-
fueled organization
Around the globe, AI-fueled “
What we’ve seen over the last
organizations are progressing beyond
just experimentation, just adoption, just few years is a significant uptake
mainstreaming, and just scaling up AI—
in investments from our clients
to truly rethinking the very DNA, culture,
and fabric of their organization.
in data topics—embedding data
products and services at the heart of
their strategy, adopting cloud data
platforms, experimenting with AI—
and then finding ways to incorporate
that into their business and drive
it to scale. While these are very
powerful concepts, they also bring
complexity into the organization that
”
must be managed.
FRANK FARRALL AI & Ecosystem Leader, Principal,
Deloitte Consulting LLP
6
AI-fueled organizations
deploy AI systematically to
lead to better outcomes
An AI-fueled organization employs data as an asset to
deploy AI across the enterprise in a human-centered
and ethical way.
Deploys AI across every core business
process with a reimagined operating model
to fully capture the potential of AI
Utilizes data as an asset for
Utilizes a holistic ethical AI
autonomous decision making
framework to generate trust
through real-time processing,
across stakeholders
learning, and acting
Creates human-centered Employs a diverse talent
digital experiences, enabling ecosystem enabled by
seamless human with a culture of innovation
machine interactions that rewards ingenuity
and risk-taking to leverage
future of work insights and
Utilizes partnerships reimagine work
and ecosystems to drive
innovation and growth
POTENTIAL OUTCOMES
Rapid decision Productive and Supercharged Enhanced customer Faster
making fulfilled workforce performance experience innovation
7
Those organizations who are able to
embrace AI in a human-centered and
ethical way across the enterprise are
gaining a competitive edge. They are
leveraging data to make the human
experience simpler, faster, and more
personalized. And moving from table
stakes innovation to meaningful,
sustainable, cultural transformation.
The Snowflake
Data Cloud
Snowflake’s Data Cloud enables the Data Cloud and execute a number
organizations to pursue the frontiers of of critical workloads, including data
data modernization by reducing data engineering, data warehousing, data
silos created within organizations, and lakes, data science, data sharing, and
scattered throughout their subsidiaries, building and operating data applications.
business ecosystems, geographies, and Unlike traditional data infrastructures,
the one or more public cloud providers Snowflake’s platform scales instantly
they use. By unlocking the latent value and near-infinitely, and enables any
of data, the Data Cloud empowers organization to operate across different
organizations to capitalize on market public clouds and regions as a single
drivers; drive decision making with cloud, while helping satisfy industry and
faster, actionable insights; and create regional data privacy requirements.
new revenue streams by monetizing
previously siloed data.
With the help of Snowflake’s platform,
organizations can easily unify, integrate,
analyze, and share their data within
8
As a Snowflake Elite Services Partner, Visit www.deloitte.com/us/snowflake
our alliance combines the advanced to learn how together, Deloitte and
capabilities of Snowflake’s platform Snowflake are empowering the next
with Deloitte’s recognized leadership frontier of data modernization.
in strategy, analytics, and technology
services to help businesses speed
up their migration to the cloud while
reducing costs and increasing agility.
NITIN MITTAL
AI Growth Offering Leader, Principal
Deloitte Consulting LLP
[email protected]
FRANK FARRALL
AI Ecosystem Leader, Principal
Deloitte Consulting LLP
[email protected]
1 Deloitte, Data Modernization and Cloud Computing Survey, 2019
2 Deloitte, State of AI in the Enterprise, 3rd Edition, 2020
This publication contains general information only, and none of the member
firms of Deloitte Touche Tohmatsu Limited, its member firms, or their related
entities (collective, the “Deloitte Network”) is, by means of this publication,
rendering professional advice or services. Before making any decision or
taking any action that may affect your business, you should consult a qualified
professional adviser. No entity in the Deloitte Network shall be responsible for
any loss whatsoever sustained by any person who relies on this publication.
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 the legal structure of Deloitte USA LLP, Deloitte LLP and their
respective subsidiaries. Certain services may not be available to attest clients
under the rules and regulations of public accounting.
All rights reserved. Member of Deloitte Touche Tohmatsu Limited.
© 2021 Deloitte Development LLC. 9 |
345 | deloitte | us-from-code-to-cure-1.pdf | From code to cure, how
Generative AI can reshape
the health frontier:
Unlocking new levels of efficiency,
effectiveness, and innovation
From code to cure, how Generative AI can reshape the health frontier | Unlocking new levels of efficiency, effectiveness, and innovation
Contents
Executive summary 3
Section 1: The shifting health care market landscape 5
Section 2: N avigating the obstacles and opportunities for
Generative AI in health care 8
Section 3: Unlocking the value of Generative AI 12
Section 4: Activating Generative AI for your organization 16
Striking the right balance for success 20
2
From code to cure, how Generative AI can reshape the health frontier | Unlocking new levels of efficiency, effectiveness, and innovation
Executive summary
Generative artificial intelligence (AI) has begun to unleash digital and management of unstructured, unlabeled data. This technology
waves across industries, but its promise to transform health is has tremendous untapped potential to deliver an immediate
only just beginning. The health care ecosystem is grappling with stepwise improvement, and exponential long-term improvement,
interlocking crises, from labor shortages and clinician burnout to to the health care ecosystem. It may help address the health
declining profitability and worsening health outcomes, particularly care industry’s greatest pain points by democratizing knowledge,
in underserved communities. The urgent need for a transformative, increasing interoperability, accelerating discovery, and enabling true
enterprise approach centers on leveraging new groundbreaking personalization.
technology while reintroducing genuine care and trust into
health care practices, both for the sustainability of health care Perhaps most importantly, Generative
organizations and the well-being of consumers.
AI can either deepen and restore trust
Generative AI technology has the potential to address these
or exacerbate mistrust and introduce
existential crises, among enterprise and direct-to-consumer
applications alike. Today, consumers are already using Generative new skepticism among consumers
AI for health care needs, and health care leaders have already
and health care stakeholders alike.
expressed activity, investment, and plans for Generative AI.
Generative AI is a solution for many of health care’s major
According to the Deloitte Center for Health Solutions,
challenges of workforce, margins, trust, and value with
and the 2024 Life Sciences and Health Care Generative
immediate opportunities in driving administrative efficiency,
AI Outlook Survey:
hyperpersonalizing the care experience, and creating digitally
75% enabled enterprise with low-code access to data and insights as
well as frictionless user interfaces. To address these challenges
of leading health care companies are already
successfully, however, Generative AI must be designed, deployed,
experimenting with Generative AI or attempting to
and scaled using a transformational approach that incorporates
scale across the enterprise
organizational change, ethics, and trust.
82%
We have predicted this seismic shift enabled by AI and radically
currently have or plan to implement governance and
interoperable data for several years, articulating that consumers
oversight structure for Generative AI
and clinicians alike are demanding new technologies to solve age-
92% old problems.1 Generative AI has the potential to catalyze trust
and power the broader Future of Health™ transformation—the
of leaders overwhelmingly see promise for Generative AI
shift from sick care and reactionary treatment to a well-being and
to improve efficiencies and
prevention focus—by helping enable radically interoperable data
65% through open, secure platforms and empowering consumers. In
of leaders see promise to enable quicker decision-making helping create this future, Generative AI can potentially eliminate
significant portions of the $1 trillion in wasted health care
spending.2 Various projections estimate that Generative AI, at large,
may contribute up to $7 trillion in global GDP over the next 10
In recent years, natural language processing (NLP) and machine years.3 As health care–specific Generative AI models and platforms
learning (ML)—subsets of AI technology—have gained traction in become more widespread, however, business leaders must identify
a host of health care use cases, ranging from clinical trial patient threats to their market position and retain a competitive edge.
recruitment to virtual physician assistants. New Generative AI Generative AI will be used in a way to disrupt today’s care models
models have demonstrated unprecedented capabilities and and create new ways to deliver medicine. These changes could
stakeholder interest as a significant expansion in natural language present challenges to incumbents, as well as current business
generation, summarization, translation, insight retrieval, reasoning, models and workflows.
3
From code to cure, how Generative AI can reshape the health frontier | Unlocking new levels of efficiency, effectiveness, and innovation
What might this disruption look like?
For health care providers, this is a golden opportunity to embrace informed decisions on traditionally complex matters including
and integrate democratized, personalized medical information into benefits, treatments, costs, prescriptions, appointments, clinical
their practice. Primary care practitioners could be equipped with trials, and wellness.
cross-disciplinary, real-time knowledge spanning medical, drivers of
Elsewhere in the health care ecosystem, life sciences companies can
health, social, and professional specialties. And modern-day doctor’s
tap into next-generation computational tools that both shave years
appointments could be streamlined, with patient information
off R&D timelines and reduce tedious commercial and regulatory
and intent gathered, analyzed, and synthesized beforehand and
barriers to entry. Furthermore, medical technology (MedTech)
integrated into their workflow, leading to tailored treatments for
companies stand to not only accelerate development, but also
each patient. Consumers can more easily access convenient,
generate some of the most meaningful, untapped multimodal data
appropriate services by having their symptoms and monitoring data
to empower longitudinal preventive care.
analyzed and triaged beforehand to be directed to the appropriate
setting of care. Generative AI holds promise across industries to streamline
operations, from discovery through commercialization—enhancing
Retail health incumbents stand to drastically improve the quality and
efficiency, compliance, and consumer-centricity. By harnessing
accessibility of care with Generative AI—leveraging a vast consumer
Generative AI, companies can achieve a competitive edge, accelerate
base, expansive and accessible footprint, and advanced analytics.
innovation, and ensure more agile and informed decision-making
These assets can fast-track an automated and interconnected
across their value chain. In the Deloitte Generative AI Dossier, we
experience, curtailing cost while uplifting care quality. With access
provide a road map for health care executives, sharing the most
to greater data, combined with the capabilities of GenAI to facilitate
compelling use cases that enhance operational performance,
navigation, the opportunity for retail health is expanding and
provide hyperpersonalized experiences, and develop enterprise
accelerating. Retail health can become a health hub, while improving
solutions while enhancing quality of care and health outcomes.
accessibility and cost of care overall.
As Generative AI advances, it will shift investments to promote and
Laboratory service businesses can extend across the value chain,
restore health, rather than simply treat sickness, by:
integrating more deeply into care delivery. Generative AI will not
only require more data, which laboratory services can feed: this
• Enabling radical interoperability
can boost their core business model in a more cost-effective,
streamlined approach leveraging “smart labs.” GenAI also offers an
• Leveling the competitive playing field
opportunity for these businesses to expand their business model
to direct patient support, second-opinion services, and provisioning
• Fostering creativity and seeding innovation
of care. These businesses can be at the forefront of clinical decision
support, where 70% of medical decisions are already anchored in
• Delivering complex reasoning
lab results.4
This new age of AI makes it even more critical for executives to
Payers and integrated payvidors (organizations that offer both
leverage Generative AI for an enterprise transformation, rather than
health insurance and health care services) can completely reshape
individual point solutions. Leaders should be asking:
their operations to lower cost and more efficiently offer services
with Generative AI powering innovative new operating platforms
• What are the long-term implications of Generative AI for my
and potent care management models. They can offer new products
business model?
and services that promote and orchestrate entirely new multimodal
care models. These organizations can become radically more
• How should my organization prepare to deploy and scale
personalized in design and administration. The basis of competition
Generative AI?
for health insurers will be reshaped, as consumers and employers
demand not only more cost-efficient services but also deeper clinical
• How can I build an enterprise transformation road map that
insights and personalized service.
encompasses the full suite of impacts, including regulatory,
Among consumers, the rising availability of digital platforms will compliance, privacy, trust, workforce transformation, and
be pivotal to bring engagement and health literacy to new heights. tax structures?
Today, consumers expect greater fulfillment across multidimensional
touch points in their care. With the advent of Generative AI–enabled
solutions, consumers will consider it table stakes for their clinicians
and insurers to provide personalized experiences informed by their
longitudinal health record and preferences. This will accelerate
shopping behaviors, as consumers are better equipped to make
4
From code to cure, how Generative AI can reshape the health frontier | The shifting health care market landscape
SECTION 1
The shifting health care
market landscape
As we navigate the complexities of the 21st century, the
health care ecosystem finds itself at a critical juncture
marked by a series of interlocking crises. The industry
has attempted to incrementally solve these issues,
and yet we have not made progress toward equitable,
quality health care delivery. We are mired in operational,
talent, financial, and value crises that demand a new
disruptive paradigm. Generative AI is the missing
element to truly drive the value, efficiency, effectiveness,
and innovation that we require.
5
From code to cure, how Generative AI can reshape the health frontier | The shifting health care market landscape
Figure 1: The 21st century interlocking healthcare crises
Labor shortages Profitability declines
Clinician burnout Value worsens
struggles
• Labor shortages: Accelerated by the COVID pandemic, health care • Value worsens: National health care expenditures have
organizations lack workers at every level. Today, hospital CEOs rank continued to rise, while the US life span has decreased to its lowest
“workforce challenges” as the top concern,5 and these shortfalls point since 1996.13 We are paying more and getting less. Health
are expected to persist with the Association of American Medical outcomes and life expectancy have significant disparities. The
Colleges forecasting a 124,000-doctor shortfall in 2034.6 Even the closure of health care facilities and the presence of provider deserts,
premier health systems and health plans are unable to stand up especially in rural areas and some urban areas, are exacerbating
operations to manage the growing demand for health care services. health care accessibility issues, affecting underserved communities
The industry has a shortage of 1.1 million nurses, forcing many the most. Consumers face a 26-day average wait time to see a
organizations to use contract labor.7 doctor,14 and place as many as 20 phone calls to find care.15
• Clinician burnout struggles: The increasing workload, emotional
We need to introduce caring back
stress, and administrative burdens have caused 81% of clinicians
to report high or modest levels of burnout.8 Clinicians cite
into health care, and humanity
administrative requirements, like paperwork and documentation,
as unnecessary and low value add. Clinicians are demanding into the experience. Trust is
technology and automation to focus on what matters most: caring
more important today than ever
for patients. Yet many clinicians also do not trust their organizations
to properly implement these innovations. Fewer than half (45%) of before—trust in clinicians, insurers,
frontline clinicians trust their organization’s leadership to do what’s
therapeutics, and institutions.
right for its patients. Even fewer, 23%, trust their leadership to do
what’s right for workers.9 We define trust holistically: as a
• Profitability declines: While consumers are concerned about series of actions, administrative
rising and unexpected health care costs,10 businesses face climbing
processes, governance, workflows,
operational costs and shrinking reimbursement rates, coupled with
an inflationary and tumultuous macroeconomic environment. As and regulations.
interest rates remain high, net working capital will remain expensive,
and payers and providers will be pressured to substantially increase
rates and cut costs, while attempting to maintain service and
experience. Health plan underwriting margins fell to a seven-year
low of 2% in 2022.11 Hospital operating margins are at just above 1%
and have been negative on average the past year.12
6
From code to cure, how Generative AI can reshape the health frontier | The shifting health care market landscape
Today’s health care enterprises each face their own challenges. A multifaceted approach to address these health care challenges
Payers are battling to streamline selling, general, and administrative requires a combination of improved efficiency, increased
costs and the cost of care.16 Providers are witnessing double-digit effectiveness, and innovation—all of which are ways in which
growth in staffing costs amid an unprecedented labor shortage.17 Generative AI can unlock new value for health care leaders
Retail health organizations are battling skyrocketing shrinkage, across efficiency, effectiveness, and innovation. According to the
increasing margin pressures, and evaporating COVID sales.18 Deloitte Health Care Generative AI Outlook Survey of 60 health
Laboratory organizations are facing stark supply chain and labor care C-suite executives in September 2023, 90% of leaders believe
cost challenges.19 Consumers are seeing their out-of-pocket health Generative AI technologies can best help their organization by
care costs continue to rise.20 improving efficiencies.
Success in health care hinges on creating and deepening trust and
innovation across the ecosystem, in providers to make sound care
decisions, in payers to cover costs and reimburse appropriately,
in pharmaceutical companies to develop efficacious treatments,
and in pharmacies to disburse and educate on medications. Yet
despite numerous health care advancements over the past 50 years,
confidence in the medical system is at all-time lows, down from 80%
to 34%. Fifty-five percent of consumers report a negative experience
causing them to lose trust in a health system, and patients with
lower trust are 19% less likely to engage in preventive care.21 Within
health care organizations, fewer than half (45%) of frontline clinicians
trust their organization’s leadership to do right by patients, and even
fewer (23%) trust their leadership to do right by workers.22 Deloitte’s
2022 TrustID Brand Index Survey—which included 25 life sciences
and health care brands—tracked similar trends: trust in both
payers and providers has dropped by 15% to 38% in humanity
and transparency.
Trust is still the key differentiator to win partners, consumers, and
talent. Generative AI has the potential to build trust and address
many of the current challenges while unlocking new value creation.
Technology, including AI, promises a similar transformative
potential as seen in other industries—from the revolution of
agriculture through automated irrigation, to the overhaul of retail
operations with inventory management systems, and the dramatic
improvement of manufacturing productivity via assembly lines.
The promise of Generative AI for
health care is the capability to tackle
greater complexity, apply more
humanlike reasoning, and interact
on a more human level than prior
AI technologies. We see intrinsic
value along dimensions of efficiency,
effectiveness, and innovation.
7
From code to cure, how Generative AI can reshape the health frontier | Navigating the obstacles and opportunities for Generative AI in health care
SECTION 2
Navigating the obstacles and
opportunities for Generative
AI in health care
Generative AI, while immensely powerful, forms just one
part of a larger, more diverse toolbox of AI solutions
available to business leaders. The foremost step in
deploying AI in an organization is a clear identification
and understanding of the problem to be solved. Based
on the need, the appropriate AI solution can be chosen
from the broader suite of tools that extend beyond
Generative AI. It’s critical to understand that Generative
AI isn’t a cure-all; it offers distinct capabilities but may
not be the right fit for every scenario.
8
From code to cure, how Generative AI can reshape the health frontier | Navigating the obstacles and opportunities for Generative AI in health care
A comprehensive AI strategy, or even an AI solution, often involves The gap of AI adoption in health care
bundling various technologies such as rule-based systems for
In Deloitte’s State of AI in the Enterprise 2022 report, we note that
processing defined business logic, robotic process automation
there is a stronger urgency, especially among biopharmaceutical
for automating repetitive tasks, discriminative AI for making
executives, to tackle the risks associated with AI technology in order
precise predictions based on a set of given inputs, and finally, the
to innovate and gain an edge over the competition.23 Yet, the adoption
indispensable human intervention for complex decision-making
of AI and all technologies in health care has consistently trailed behind
and reasoning.
other sectors, often falling behind due to cost, structural, regulatory,
Business leaders must ask the question: What is Generative AI organizational, and technical challenges.
best suited for, compared to other solutions in place today?
Health care has lagged in its AI adoption. A study conducted by
1. Generative AI technologies should be viewed as accelerants and Brookings in 2022 found that health care’s AI integration rate
supplements, not replacements, to humans. As this technology lagged all other industries outside of construction.24 Technical and
matures, we expect the sophistication and independence of the interpretability challenges, a heavy dependence on text and contextual
solutions to require less human intervention. data, and inherent biases in AI models have hindered widespread
AI acceptance in health care. Prior NLP techniques demonstrated
2. Generative AI is a powerful new technology to be embedded significant shortcomings, with false-negative rates and limited efficacy
within a suite of other AI solutions. Indeed, Generative AI is detecting contextual types of languages.26 These issues, combined
not a panacea for all solutions. It outperforms other AI models with the high-stakes nature of health care, underscore the complexity
on key dimensions but still lacks capabilities in extraction and sensitivity of implementing AI.
and computation.
Yet, there is tremendous opportunity to leverage an ample supply of
health care and real-world data. Health care has become the world’s
Generative AI should be seen as a largest data source, at 30% of annual production,27 with 80% of that
health care data being unstructured.28 The path to widespread AI
piece of the larger puzzle in the
adoption in health care is uphill, but the richness of health care data
strategic application of technologies, and ongoing advancements suggest an optimistic outlook for this next
age of Generative AI. We anticipate that Generative AI will likely make
each complementing the other to form
near-term impacts across efficiency, effectiveness, and innovation. In
our point of view, “A new frontier in artificial intelligence: Implications
a robust and comprehensive solution
of Generative AI for businesses,” we proposed a five-part functional
for diverse business challenges. framework for Generative AI use cases and value levers. Generative AI
models differ from prior AI and ML models in ways that deliver value
across activities that accelerate, automate, create, personalize,
and simulate.
Figure 2: The evolution of AI technologies
Generative AI
Artificial Intelligence
(AI) is a broad market
of which Generative AI
is one of the many
technologies that can
Conversational Autonomous disrupt how society
AI systems interacts and business
Deep is conducted...
learning
Speech
recognition Machine
learning
Artificial general
intelligence (AGI)?
Computer
Predictive vision
analytics
Intelligent 9
automation
From code to cure, how Generative AI can reshape the health frontier | Navigating the obstacles and opportunities for Generative AI in health care
Figure 3: The differentiated functions of GenAI in health care
Accelerate Automate Create Personalize Simulate
Enhance productivity by Deliver business and Push boundaries of Create familiarity and Create environments
accelerating outcomes technical workflows, creativity, leveraging personalization, in which workflows,
and offering top-tier and in some cases, prompts to develop which could take experiments, and
building blocks replace humans novel content significant effort experiences can
be simulated
Document distillation Code classification Record summarization Prompt generation Interaction visualization
Synthesizing lengthy text Processing unstructured Summarizing care Enabling information Building digital 3D models
into short-form summaries, inputs to produce a list of encounters (for HCPs) gathering across of cellular and chemical
evidence tables, or discrete alphanumeric with details about history, stakeholders in a patient- structures to aid in
dashboard/knowledge graphs codes that are used in symptoms, procedures, friendly way, through a back- discovery, development,
downstream processes diagnoses, etc. and-forth conversation and diagnosis
Component compilation Multimedia creation Jargon simplification Hypothesis validation
Integrating information from Generating interactive Explaining complex concepts Running experiments and
different source systems into materials that contain text, at an appropriate health workflows via a machine
a cohesive, ready-for-review interspersed with video literacy level through shorter- to help refine parameters
artifact with next steps and images, for education form, simplified versions before rolling out a process
and image or engagement out in practice
Translation to preference
Translating patient-facing
clinical and non-clinical
documents in real time, in a
patient’s preferred language
The promise of recent and upcoming accessibility, positioning them as a flexible and cost-effective choice
Generative AI advancements for organizations desiring domain-specific performance. These
strategic choices exert pressure on large hyperscaler market leaders,
2023 has witnessed an unprecedented level of advancement of who are now facing increasing customer demands to provide greater
Generative AI technologies. By April, ChatGPT was shown in a JAMA transparency and flexibility in their proprietary models.
study to outperform physician responses to medical questions on
Generative AI aligns well to functional needs within health care
dimensions of both quality and empathy.29 Google’s health care–
underserved by traditional AI and ML models. In certain functions,
specialized MedPaLM-2 large language model (LLM) became the first
Generative AI is positioned to potentially replace tasks and roles
to achieve an expert-level passing score on the US Medical Licensing
in data entry, classification, and generation, while supplementing
Exam,30 and the first drug completely designed with Generative
tasks requiring more empathy, innovation, and decision-making.
AI techniques was entered into human clinical trials.31 New
Today, Generative AI solutions are better fits for top-left functions
groundbreaking GH200 graphics processing units (GPUs) have been
that are lower cost and lower complexity, but as the models advance
announced with promise to precipitously drop costs for LLMs in both
and stitch together with a broader suite of AI solutions, we foresee
training and inference in 2024.32 The Deloitte Health Care Generative
potential for broad use.
AI Outlook Survey found that 72% of health plans and 80% of health
systems have already launched pilots or are actively scaling across
the enterprise highlighting potential rapid adoption. This pace of
change and uptake plots a tremendous trajectory.
The complex health care industry, we project, will likely focus on
specialized Gen AI models and heavily prompted and fine-tuned
use cases. Indeed, the investments into Generative AI this year
alone have demonstrated outcomes previously thought to be 20
to 30 years away. Competition among major technology players
is raising the bar—fueling new releases on the scale of months,
rather than years. Some notable technology organizations have
made their models open source to the public. These open-source
LLMs offer benefits in customizability for specific tasks, fine-tuning
on proprietary data, control over privacy and costs, and enhanced
10
From code to cure, how Generative AI can reshape the health frontier | Navigating the obstacles and opportunities for Generative AI in health care
Figure 4: The impact of GenAI based on task and value
Figure 5: The Future of Health™ and GenAI
11
woL
hgiH
Ability for GenAI to execute tasks in health care roles
High Low
Data entry Classification Summarization Content Visualization Prediction Optimization Decision Innovation Empathy
generation
IAneG
gnisu
eulav
ezilaer
ot
ytixelpmoc
dna tsoC
Calculation Interpretation
adH mo cls i esp s ri kit oa nl s sM pb ee il cd li in aic g la is l t tP eh ca hr nm ica iac ny Data analyst tR ea cd hi no il co iagy n Scheduler Hom ae id h eealth
recM ore dd sic ca lel rk M ce od di ec ral aR se ss oe ca ir ac th e IT technician Underwriter R te hs ep ri ara pt io stry cooP a ra c dcti iee nsn ast t or Charge nurse M aa nrk ae lyt sin tg aM sse id stic aa nl t
M sce rd ibic eal L teab cho nra ict io ar ny M we rd iti ec ral M da er sk ige nti en rg Pu eb dli uc c h ae toa rlth Sup ap nl ay l yc sh tain tP hh ey rs ai pc ia sl t aR se ss oe ca ir ac th e Counselor
reM cee pd tii oca nl i st reC vla ieim wes r Health coach cooP a ra c dcti iee nsn ast t or cor oeC s rli den a ii nc ra acl th o r Acutary Optometry prG ace tn ite iora nl e r mAc ac no au gn et r
Cl sin pi ec ca il a d lisa tta Cl min aic na al gd ea rta serP va ict eie sn rt e p. Bio-statistician Pathology Pharmacist sB tu ras tin ee gis ss t Sales rep
Re ag ffu ala irt sory asQ suu ra ali nty c e in mfoH are nma al a gt th eio rn M spa erk ce iat li in stg siM pm ee a cd g iai ic n la ig sl
t
Radiology eP nr go ic ne es es r H ao ds mpi it nal B eio nm gie nd ei ec ral
Te nle uh re sa elth Dosimetrist Epidemiologist Dermatology opC eli rn ai tc ioal n s S pp he ysc ii ca il ais nt R se cs iee na tr ic sh t Psychologist manager
Audiologist Data scientist dM ire ed ci tc oa rl Pr pim hya sr iy c ic aa nre
teM r ce he c nd o iri cc d ia asl n H ee coal nth o mca isr te Bio-informatician Surgeons
Generative AI has far exceeded previous state-of-the-art solutions. These early
successes are just the beginning as Generative AI leaves the laboratory and
integrates into products across the health care ecosystem. It is no stretch to
imagine the potential transformative applications of this technology.
Champion interoperability Level the playing field
GenAI capabilities to manage Make data usable and accessible Act as true creativity engines Evolve into reasoning engines
structured and unstructured data to nontechnical users and Accelerate discovery by reducing Personalize patient interactions by
and data labeling will enable and smaller-scale organizations the needs for hyperspecialization suggesting alternatives and making
accelerate data interoperability that historically lag and suggesting novel ideas communication more empathetic
across the industry sophisticated competition
Foundational advancements will lead to the rise of specific business models…
Science and Health product Personalized virtual
Data conveners
insights engine developers health actors
These businesses will create and operate...
Genesis of autonomous agents Constructing innovative care models New modalities of health and wellness
Autonomously analyze numerous systems and A primary care practitioner with cross-disciplinary New foundation models can embed treatment
datasets to perform tasks or identify insights, expertise to deliver personalized care with instant plans with customized music, 3D printed scans
integrating the abilities of data analysts, onboarding and complete knowledge of a and onsite-produced prosthetics, and virtual
AI/ML engineers, physicians, patient’s history appointments with physical renderings
and psychologists
From code to cure, how Generative AI can reshape the health frontier | Unlocking the value of Generative AI
SECTION 3
Unlocking the value
of Generative AI
The power of this moment, at large, is tremendous,
yet the obvious question remains: Where should I,
as a health care business leader, make immediate
investments to win in the new age of AI?
In practice, the key question becomes how to effectively
deploy these Generative AI models, both in terms of
which issues they are fit to solve and which areas of the
enterprise will likely be positioned to maximize their
value. In the Deloitte Health Forward Blog, we argue for
the value of incrementalism, where health care leaders
strike a balance between short-term demands and a
long-term vision. Business leaders must keep an eye
toward the innovation arc, while placing immediate bets
on areas within the enterprise.
12
From code to cure, how Generative AI can reshape the health frontier | Unlocking the value of Generative AI
Generative AI pre-trained models have historically used publicly In our Deloitte Generative AI Dossier, we elaborate upon high-value
available, non-industry-specific datasets. Some of these pre-trained use cases that health care leaders can pursue to create value
models have potential applicability to administrative, operational, across (1) employee productivity and operational efficiency, (2)
and back-office use cases. However, within the health care hyperpersonalized experiences, and (3) new enterprise digital and
environment, especially in the context of clinical delivery, stakes data capabilities. Below, we provide each example aligned to a
are high, and language must be precise, in addition to articulate. respective value driver.
The rise of health care–specific Generative AI LLMs in addition to
a more deliberate and experienced execution approach is breaking
ground to pursue these more sensitive and nuanced use cases that
evolve patient care.
Co-writer for denial Supply chain Personalized
appeal letter optimization service for patients
Driving administrative Supporting optimization Assisting human
cost-efficiency through by leveraging GenAI staff responding to
employee productivity and to simulate, model, patient questions
operational efficiency and generate
data-driven insights
Opportunity • There are many claims that are • Supply chains involved many • Patients often have to spend hours
denied in the US representing stakeholders and dependencies with IVRs and other systems to
billions in added costs creating complexity resolve issues
• Sixty percent of denied claims • High complexity makes • High call volumes require numerous
can be reclaimed, but only 0.2% efficiency, resilience and cost agents |
346 | deloitte | us-Deloitte-MMTS-report.pdf | 2023 Mid-market technology trends report
Convergence topples industry walls and powers
growth ambitions for midsize private companies
2023 Mid-market technology trends report | Convergence topples industry walls and powers growth ambitions for midsize private companies
Table of contents
Introduction .............................................................................. 1
About the survey ...................................................................... 2
Section 1: A foundation for growth ........................................ 3
Section 2: Industry convergence: Opportunities
for growth and transformation ............................................... 6
A special message—Industry convergence:
Practical outcomes and responsible growth ..........................9
Section 3: AI adoption
and implementation ................................................................10
Section 4: Executable strategies for consideration .............13
A special message—CISO perspective:
Adapting cyber priorities to evolving threats,
new risks, and organizational changes ................................14
Conclusion .................................................................................15
Get in touch ...............................................................................16
2023 Mid-market technology trends report | Convergence topples industry walls and powers growth ambitions for midsize private companies
Introduction
In Washington state, orchardists are testing a 14-foot-tall robot with
mechanical arms that's capable of picking the ripest apples from
the tree. The robotic picker might one day help alleviate human
labor shortages for the painstaking agricultural work. Through
this mode of automation, researchers see ways to create more
sustainable systems in farming, to feed livestock, milk cows, or
navigate greenhouses under human supervision.1 This type of
precision farming is possible because of the convergence between
the high-tech industry and agriculture using technologies such as
the Internet of Things, artificial intelligence (AI), robotics, and big
Wolfe Tone
data—allowing workers to optimize the growth, harvesting, and
distribution of agricultural products.2
The rise of industry convergence—as well as The leaders who participated in our survey
the blurring of boundaries within sectors— represent predominantly privately held
is one of the key trends uncovered in this companies with annual revenue between
year’s survey of private and family-owned $250 million and $1 billion. Many of these
companies. An analysis of this year’s survey, enterprises are seeking greater returns on
our ninth assessment of the technology their technology investments and appear
priorities, investments, and challenges to be stretching their innovation muscles:
facing America’s middle market, offers a Seven out of ten respondents (70%) report
strong assessment that these companies that they have or are in the process of
are not only prioritizing technology developing assets that can be leveraged
investments that reduce time to value but and monetized outside of their own
Chris Jackson
also seeing value and innovating at a pace business for additional growth or
we haven’t historically seen in prior surveys. expansion. Many of these companies
appear to be seeking growth outside of their
traditional sector boundaries or investing
to help defend against other organizations
encroaching into their sectors.
Wolfe Tone Chris Jackson Ryan Jones
Vice Chair, US and Global Deloitte Private Deloitte Consulting
Deloitte Private Leader Technology leader Private Equity leader,
Ryan Jones
and Former Technology
Sector leader
11
2023 Mid-market technology trends report | Convergence topples industry walls and powers growth ambitions for midsize private companies
About the survey
Balancing act: Emerging opportunities and familiar challenges
Enterprises across this slice of the commercial landscape are seizing on the potential of AI to transform network operations, increase
efficiencies, and improve customer service.
In addition to navigating increasingly blurred sector boundaries, mid-market companies are confronting headwinds affecting a broad swath of
businesses, including evolving cyberthreats, changing talent configurations, and the impact of generative AI. Unlike prior surveys, this year’s
results suggest mid-market companies are doubling down on their technology investments—and proactively investing to stay ahead of these
challenges. Many of their investment priorities over time appear to be paying off, with three in four respondents reporting they have high or
very high confidence in their cybersecurity capabilities, for instance. From a human capital perspective, respondents indicated that they are
prioritizing hiring based on skills versus degrees. Furthermore, company leaders say they are leveraging talent from their ecosystem partners
and/or service providers.
To better understand what drives success more fully for these companies—and, in turn, their appetite for technology investments—we
reviewed the survey results for companies that believe they have been most successful in achieving their tech objectives. Then, we tracked
the respondents who anticipated achieving the highest return on investment (ROI) on their recent technology investments.
In these pages, we explore how private and family-owned companies continue to unleash their full potential for growth in an era where
disruption happens in real time.
Survey methodology
From May 4–24, 2023, a Deloitte survey of private and mid-market companies was conducted by a market research firm. The survey
examined technology trends taking place in this market segment to determine the role that technology plays and how it influences business
decisions. The 500 survey respondents represented companies with annual revenues ranging from $250 million to a little more than $1
billion. Firms with revenue between $250 million and $499.9 million in annual revenue comprised 10% of the sample; firms with at least $500
million to $749.9 million in annual revenue comprised 30% of the sample; firms between $750 and $1 billion in annual revenue comprised
30% of the sample; and firms more than $1 billion in annual revenue comprised 30% of the sample this year. Half of the respondents were
C-suite executives, while the remainder were non-C-suite decision-makers. Eighty percent of the respondents represented companies that
are privately held, while the rest were publicly traded firms. Among industries, 39% were from technology, media, and telecommunications
companies; 22% were from financial services companies; 21% represented consumer and industrial products companies; and the remaining
respondents were divided among energy and resources companies, and life sciences and health care (LSHC) companies. Some percentages
in the charts throughout this report may not add up to 100% due to rounding or for questions where survey participants had the option to
choose multiple responses.
2
2023 Mid-market technology trends report | Convergence topples industry walls and powers growth ambitions for midsize private companies
Section 1: A foundation for growth
Investment, spending, and new opportunities
Spotlight:
If budgets are reflections of a company’s ambitions, private
enterprises surveyed are conveying a strong desire to innovate at How companies and industries
the edges. This year’s survey reveals that overall technology
are investing
spending is at its highest level since 2019, perhaps making up for
lost ground during the pandemic. Further, of those businesses that
In the health care industry, AI could streamline and
reported spending more than 5% of their revenue on technology,
90% also reported an increase in their technology spending automate the often costly and time-consuming
compared to last year. process of appealing denied insurance claims. In
2021, more than 48 million claims were denied,
The leading areas of technology investment span a range of business
representing about 17% of all claims. AI’s ability to
needs as companies adapt to new innovations. In our prior survey in
automate this resource-intensive process could
2021, just 12% of respondents predicted AI would have a significant
potentially yield significant savings for hospitals
impact on their business within a year. In the current survey, AI has
leaped ahead of other technologies as 40% of respondents call it the and other health care providers while freeing
top tech investment priority. For many of these organizations, workers to focus on higher-value tasks.3
AI can provide value in the automation of repetitive processes,
working to create demonstrable value and savings for organizations. Security, risk, and threat monitoring come in at
a close second among the top targets for tech
investment. Cloud infrastructure and customer
resource management (CRM) investments round
out the top four areas of investment over the
past year.
This year’s results also suggest how company
ownership and industry can influence technology
investment decisions. For instance, while just over
a quarter of family-owned businesses (27%) say
they invested in AI over the past year, around half
of private equity-owned businesses (49%) say they
have pursued the technology.
More than any other sector, respondents in the
energy, resources, and industrial (ER&I) industry
report a focus on metaverse technology. As a
practical application, ER&I companies can tap
the metaverse for virtual reality and augmented
reality-enabled immersive employee training,
and externally, for virtual storefronts to promote
sustainable efforts.4
3
2023 Mid-market technology trends report | Convergence topples industry walls and powers growth ambitions for midsize private companies
Calculating technology ROI
This year, we measured a range of factors
to quantify more precisely how technology
investments are translating into the ability to
successfully achieve technology objectives.
Several factors inform our analysis, including
whether the quality of data is sufficient
for the application of AI to the business;
talent capabilities; the respondents’ vendor
and partner networks; and the ability to
expand outside of the company’s industry
and sector. We used these inputs to assess
ROI on tech investments as measured by
revenue increases for these companies.
The businesses reporting the highest ROI on
their tech objectives report being more than
twice as likely to strongly agree that their
data was sufficient for the application of AI.
Respondents at the upper end of revenue
growth report they are more likely to
increase their technology spending by more
than 20% compared to the previous year. “Especially for companies with active AI initiatives, there’s
Correspondingly, businesses at this end of
a growing level of confidence in these investments as the
the spectrum are almost one-and-a-half
times more likely to have seen an increase in businesses harness, monetize, and generate revenue
revenue of 20% or more.
from selling data and tech-enabled services.”
Our assessment also reveals that a mature
Khalid Kark, CIO research director, Deloitte LLP
cyber posture can be a critical investment.
Security, risk, and threat monitoring
software—such as risk quantification tools
Notably, respondents with active AI “Companies have matured across their
that compare the costs, benefits, and ROI
solutions are about two-and-a-half times as cyber capabilities as they’ve outsourced
of cyber investments5—are the technology
likely to have very high confidence in their the most complex parts of their cyber
investments respondents ranked as most
cybersecurity capabilities compared to functions, added additional protections,
likely to have a high or very high return
businesses not using or exploring AI at all and leveraged investments in cloud and
on investment. Meanwhile, AI was the
(32% vs. 11%). other digital infrastructure,” says Criss
technology investment that was most likely
Bradbury, Markets, Offerings, and Alliances
to have a very high ROI.
Still, relatively few C-suite executives appear leader, Deloitte LLP. “As more organizations
to be comfortable with the state of their embrace AI, I expect even more urgency to
Evolving posture toward cyber risk
cybersecurity efforts (12% report very high embrace security, ensure compliance, and
confidence) compared to leaders outside enable customer trust.”
The leaders we surveyed this year appear
of the C-suite (22%). This could suggest
to be seeing the results of a sustained and
that technology leaders have been actively
maturing focus on data security. Three in
pushing for investments to meet the array of
four respondents indicate high or very high
emerging threats because of their familiarity
confidence in their business’s cybersecurity
with the domain—and they are putting
capabilities. Consider that in our 2018
these solutions into action.
survey, fewer than half of respondents (48%)
said they had governance structures in place
concerning information security threats.6
4
2023 Mid-market technology trends report | Convergence topples industry walls and powers growth ambitions for midsize private companies
The report also
highlights a shift
to skills-based
organizations, with
forward-thinking
organizations
altering their focus
from jobs and job
titles to acquiring
specific skills.7
In our technology survey, almost a quarter
of the respondents report the use of skill-
based hiring versus degree-based hiring.
Talent strategy: A tech workforce revolution
Roughly the same share says they are using
talent and skills from ecosystem partners
Addressing the workforce crunch for jobs that require skills such as engineering and
and/or service providers as their approach
data science is another evergreen topic among private company technology executives.
to develop tech talent.
Compared to the prior 12 months, almost half of the respondents surveyed (49%)
indicate no change in their ability to retain their top technology talent. Over a quarter of
“It’s critical for leaders to convene a broad
respondents report that it has been easier to retain key technology talent compared to the
ecosystem of partners who can help
past 12 months.
navigate the speed of change and the
complexity that comes with converging
Nonetheless, one-third of businesses at the lower range of our ROI measurement for tech
technologies and industries,” says Ryan
investments say they are facing more difficulties in retaining their top tech talent compared
Jones, Deloitte Consulting Private Equity
to 12 months ago.
leader, and Former Technology Sector leader.
There is additional evidence that the nature of jobs is changing. Deloitte’s 2023 Global Human
While respondents in our survey express
Capital Trends report describes an increase in the share of workers who say they already have
an interest in acquiring skills through
switched, or are likely to switch, employment models throughout their careers—from full-
external means, they appear to be pulling
time jobs to opportunities like freelancing and gig work.
back dramatically in their own efforts to
impart these skills to their teams. In our
2018 survey, 61% of respondents said that
reskilling employees to realize the greatest
benefit from technology was the top focus
area for maintaining their workforce through
technology. This year, just 19% say upskilling
or retraining existing talent is their primary
method to developing technology talent.
According to Jones: “Rapidly emerging
technology and changing job roles have
created a reliance on tech providers that
have deep engineering skills.”
5
2023 Mid-market technology trends report | Convergence topples industry walls and powers growth ambitions for midsize private companies
Section 2: Industry convergence:
Opportunities for growth
and transformation
Crossing traditional business boundaries the toughest threats. When asked which industries outside
of their industry would pose a threat to their current position in
With new technologies disrupting business models at every turn, the marketplace in the near future, respondents across most
convergence across industries seems inevitable. Whether it’s industries said those threats would arise from companies within
investing in current capabilities, acquiring a leading-edge startup, their own sectors.
or finding new uses for existing assets, private and family-owned
companies have an array of approaches at their disposal to seek For instance, among respondents from consumer products
new opportunities as industry and sector lines converge.8 companies, financial services firms, and TMT companies,
respondents reported that adjacent businesses within those
Perceptions about the competition make up just one part of the industries pose the biggest threats in the near future.
convergence story: In our survey, half of the total respondents
(51%) see a high or very high threat to their current position in the Nonetheless, respondents are overwhelmingly confident that they
marketplace from businesses outside of their sector. have the tools in place to move into adjacent industries. More than
two-thirds of respondents (70%) believe their business has an asset
The reality may be slightly different in practice, however. An that could potentially be monetized outside of their sector. Among
industry-by-industry view shows that as companies simultaneously respondents reporting the highest ROI on their tech investments,
defend their turf, they see companies within the same sectors as the share jumps to 81% with an asset ready for an adjacent market.
70% of respondents 1/3 of respondents
believe their business has report spending more
70% 33%
an asset that could be than 5% of revenue on
monetized outside growth outside their
of their sector industry or sector
Key
findings
44% of respondents
55% of respondents
say boards should
44% say boards should focus 55%
concentrate on industry
on cybersecurity
convergence
6
2023 Mid-market technology trends report | Convergence topples industry walls and powers growth ambitions for midsize private companies
Manifesting ambitions
As we noted in our analysis of tech investment priorities, technology
spending often telegraphs strategic plans and how companies intend
to sustain their growth ambitions: Almost a third of respondents
(32%) report their companies are spending more than 5% of their
revenue on growth outside of their industry or sector. To put that
into perspective, companies with $1 billion in annual revenue report
spending a minimum of $50 million to pursue growth plans outside
of their existing area of business.
As companies focus on ethical and regulatory considerations of
rapidly evolving technologies, this year’s responses suggest where
boards should be prioritizing their energy and time: More than half
of respondents (55%) say boards should focus on cybersecurity and
regulatory matters, while 44% of respondents say they want their
board members to concentrate on industry convergence.9
“Tech is the unifying thread as
the lines between humans and
machines, traditional industries and
their competitors, and customers
and their suppliers continue to
converge," says Brett Davis, principal,
Deloitte Consulting LLP and Global
Assets leader and general manager
of Converge by Deloitte. “That a
significant share of companies are
devoting resources to exploring
growth in industries beyond their
own tells us that employees, leaders,
and boards see opportunity and
value through convergence.”
7
2023 Mid-market technology trends report | Convergence topples industry walls and powers growth ambitions for midsize private companies
Where do sectors perceive the biggest threat to their current
position in the marketplace?
Boundaries between sectors are increasingly diminishing, with a majority of mid-
market businesses having their footprints in a different sector within their industry.
These blurred boundaries have led mid-market companies to perceive notable threats from
Threats in the near fu-
sectors within the same industry. This speaks to a rise in sector convergence. The exception is
ture by sector (n=256)
industrial products and construction, which sees the biggest threat from outside its industry—
71%
1 Automotive Consumer Products
from technology.
60% Retail, Wholesale & Distri-
2 Consumer Products
bution
Sector perceived as
Retail, Wholesale & Distri- 33% Sector
3 Consumer Products the highest threat
bution
Transportation, Hospitali- 43% Retail, Wholesale & Distri- 1 Automotive 71% Consumer Products
4
ty & Services bution
100% Power, Utilities & Renew-
5 Mining & Metals 2 Banking & Capital Markets 48% Insurance
ables
57% Power, Utilities & Renew-
6 Energy & Chemicals
ables 3 Consumer Products 60% Retail, Wholesale & Distribution
Industrial Products & 40%
7 Technology
Construction
4 Energy & Chemicals 57% Power, Utilities & Renewables
Banking & Capital Mar- 48%
8 Insurance
kets
56% Banking & Capital Mar- 5 Health Care 55% Life Sciences
9 Insurance
kets
Industrial Products &
10 Life Sciences 38% Health Care 6 40% Technology
Construction
55%
11 Health Care Life Sciences 7 Insurance 56% Banking & Capital Markets
75% Telecommunications, Me-
12 Technology
dia & Entertainment 8 Life Sciences 38% Health Care
Telecommunications, Me- 25%
13 Technology
dia & Entertainment
9 Mining & Metals 100% Power, Utilities & Renewables
Retail, Wholesale &
10 33% Consumer Products
Distribution
Telecommunications, Media &
11 Technology 75%
Entertainment
Telecommunications,
12 24% Technology
Media & Entertainment
Transportation, Hospitality
13 43% Retail, Wholesale & Distribution
& Services
8
22002233 MMiidd--mmaarrkkeett tteecchhnnoollooggyy ttrreennddss rreeppoorrtt || CCoonnvveerrggeennccee ttoopppplleess iinndduussttrryy wwaallllss aanndd ppoowweerrss ggrroowwtthh aammbbiittiioonnss ffoorr mmiiddssiizzee pprriivvaattee ccoommppaanniieess
A special message
Industry convergence: Practical
outcomes and responsible growth
By Brett Davis, US Consulting Chief Innovation Officer and
General Manager of Converge by Deloitte
The world is being reshaped and redefined by convergence. helped the company build and launch a digital platform, and
The lines between traditional industries, competitors and even in a highly regulated industry covering multiple jurisdictions
collaborators, and customers and suppliers are becoming and regulations, Deloitte was able to help the client launch the
increasingly blurred. Platforms, partnerships, and products are platform across select markets in just 11 months.
no longer the domain of single industries or adjacent industries.
All of these elements are converging—creating a profound We have also seen an emerging trend in which data assets have
transformation that’s creating new efficiencies, streamlined value across different use cases in adjacent sectors or markets.
processes, and novel opportunities for companies to grow. In the mid-market technology survey, 70% of respondents report
having an asset that can be leveraged for value outside of their
This year’s mid-market technology survey highlights how industry organizations. That’s a remarkable data point, owing to the rapidly
convergence is accelerating—as companies activate cloud, AI, eroding boundaries and barriers for creating cross-industry
5G, mobile, and other technologies that enable movement into solutions. Conversely, the motivations for creating them have
adjacent sectors or transform existing services in new ways. In increased—as companies realize how the data they produce
fact, more than two-thirds of respondents see a high or very high can be repackaged, used to create new value in other markets,
threat to their position in the marketplace from outside of their transformed and aggregated to deliver goods and services in
sector. And nearly a third of businesses are spending more than personalized ways across their supply chains, or even used by
5% of their revenue on growth outside of their industry/sector. other industries for additional insights.
These trends are emerging in multiple industries. In the consumer industry, companies can use third-party and
primary data to create more personalized experiences for
For instance, tech and consumer companies are offering health consumers. With this enhanced capability comes additional
care services through new digital experiences. Conversely, health responsibility in managing that data. Think of a retailer
care organizations are using consumer technologies to reach entering the health care space by tapping into consumer
and support patients in new ways. In one example, we helped a information through its network of physical stores and digital
medical school engage potential clinical trial participants directly platforms. There's an incredible opportunity to personalize care,
via a digital platform, allowing patients to participate remotely, create brand loyalty, and deliver outcomes for someone on a
and helping investigators and other stakeholders collaborate wellness journey.
more easily in the research process. Prior to these types of
innovations, a study participant would have had to be identified
and engaged at a medical center. Notwithstanding, there are important
ethical, regulatory, and security
There is also a shift in technology buying behavior among
customers who want to enter into new industries—they implications to be considered when
increasingly expect pre-built tech solutions and a trusted partner
entering new adjacent industries.
to help them enter these markets and industries.
This trend is evident in banking, where nontraditional financial
services providers are increasingly offering digital banking The lines between traditional industries and their competitors will
services to their existing set of customers and bundling these continue to converge—enabled and accelerated by technology.
products with nonfinancial service offerings. A recent example of It will be an exciting decade ahead as industries are reinvented
this is Deloitte’s engagement with a multinational consumer client because of this convergence.
that wanted to offer services in multiple global markets. We
99
2023 Mid-market technology trends report | Convergence topples industry walls and powers growth ambitions for midsize private companies
Section 3: AI adoption
and implementation
For private companies trying to determine how to turn the hype around AI into a Where are mid-market
business differentiator, one place to look for perspective is the financial services industry.
companies on the AI
For transactions such as billing, payments, or collections, the opportunities include
generative AI-powered “agents” that could deliver tailored content to customers, as well adoption curve?
as conversational Q&A using models trained on enterprise data to support new employees
through the procedures.10
There are also risks when it comes to exposure of sensitive financial data using AI tools. A 41% report
report by the World Economic Forum and Deloitte argues that by being an AI early adopter,
exploring AI
AI could expose the financial system to new hazards by triggering failures that damage brand
equity and customer trust, trigger additional regulatory scrutiny, and alienate employees.
Companies in other industries may not be far behind in having to consider such issues—if
they aren’t already doing so. Nearly all respondents in our survey report they are on the AI
30% report
adoption curve, with 41% of respondents saying they were exploring AI, 30% saying they were
piloting AI solutions, and one-quarter of respondents saying they have active AI applications. piloting AI
solutions
Among industries, technology, media, and telecommunications
(TMT) and life sciences and health care (LSHC) are the most
25% report
likely businesses to have active AI solutions, while companies
having active AI
in financial services and insurance (FSI) are the least likely to be
active in this area. applications
% of business with active AI solutions by industry
Technology, Media &
33%
Telecommunications (TMT)
Life Sciences & Health Care
32%
(LS&HC)
Consumer 20%
Energy, Resources &
20%
Industrials (ER&I)
Financial Services (FSI) 13%
10
2023 Mid-market technology trends report | Convergence topples industry walls and powers growth ambitions for midsize private companies
“There’s an infinite number
of AI use cases that
can provide value to an
organization; however,
to truly scale, companies
should start their journey
by focusing on a single
use case that’s promoted
all the way to production.
This can provide the
foundation for future AI-led AI and talent
innovation throughout the
While there's been no measurable change in businesses' overall ability to hire technology
organization.” talent, respondents are finding it difficult to build a workforce with AI expertise.
Deborshi Dutt, AI Strategic Growth
Nearly one-third from financial services companies report that AI ethics officers are in short
offering leader, Deloitte Consulting LLP
supply, which tracks with an overall labor-market squeeze and evolving regulatory pressures
that are driving demand for compliance officers.
As viewed through our ROI measurement, it
appears that businesses that have achieved
the highest success in their tech objectives
and tech ROI are more likely to have active
AI solutions in a business area. And it
turns out, AI appears to be helping these
businesses achieve an array of benefits.
40% 37% 35%
report difficulty report difficulty report difficulty
attracting AI attracting finding data and deep-
strategists engineering talent learning scientists
About nine out of ten respondents (87%)
who state their companies have active AI What did respondents say are the top worker challenges?
solutions report that those solutions are
currently generating both revenue and
saving costs.
In the past year, respondents with active AI
solutions say they’ve focused their efforts
on harnessing data, modernizing legacy
systems, and improving cybersecurity.
What’s more, respondents from companies 40% 37%
with active AI solutions are more likely
report employee report ethical
to be very confident in their companies’
cybersecurity capabilities. perception of AI use of AI
11
2023 Mid-market technology trends report | Convergence topples industry walls and powers growth ambitions for midsize private companies
Challenges to AI adoption
We asked respondents to identify the challenges their companies are facing related to AI.
Leaders who responded to our survey report different concerns depending on their industry—
which may reflect how companies are struggling to align culture, ethics, and strategy in their AI
adoption journey.
Challenges to AI adoption
We asked respondents to identify the challenges their companies are facing related to AI. Leaders who AI concerns related to talent
responded to our survey report different concerns depending on their industry — which may reflect how
companies are struggling to align culture, ethics, and strategy in their AI adoption journey.
AI concerns related to talent Employee perception of AI ER&I 56%
• Lack of available talent/Consumer and LSHC/33%
• Internal employee resistance to advanced AI/ER&I/38%
Internal employee resistance to advanced AI ER&I 38%
• Employee perception of AI/ER&I/56%
Consumer/
AI concerns related to trust Lack of available talent 33%
• Ethical use of AI/Energy/42% LSHC
• Customer concerns or perceptions of AI/TMT/38%
AI concerns related to business strategy
• No real business and tech alignment around AI/Financial Services/29%
• Lack of enterprise strategy around AI focus/TMT/29% AI concerns related to trust
• Lack of an innovative culture to take chances/ER&I and LSHC, 30% each
• Lack of business engagement/LSHC/36%
Ethical use of AI ER&I 42%
Customer concerns or perceptions of AI TMT 38%
AI concerns related to business strategy
Lack of business engagement LSHC 36%
Lack of an innovative culture to take chances ER&I/LSHC 30%
Lack of enterprise strategy around AI focus TMT 29%
No real business and tech alignment around AI FSI 29%
Key: ER&I—Energy, Resources & Industrials FSI—Financial Services LSHC—Life Sciences & Health Care TMT—Technology, Media & Telecommunications
Note: These industries represent the top responses for each of the options for this question.
12
2023 Mid-market technology trends report | Convergence topples industry walls and powers growth ambitions for midsize private companies
Section 4: Executable strategies
for consideration
A look ahead
The top three overall technology objectives from the past year (improving cybersecurity, enabling business growth, and optimizing business
operations) remain the top three technology goals in the coming year as well. Companies in life sc |
347 | deloitte | us-ai-in-surveillance-POV.pdf | Augmenting trade surveillance
programs with artificial intelligence and
machine learning: A brief overview
May 2024
“What we were looking to do here was really to
answer some of the questions that were presented
in surveillance: changing market conditions, increased
volatility, increased volumes and change in conduct. And
so, using deep learning made a lot of sense to start to
answer those challenges.”
—Susan Tibbs, Former Vice President, Market Manipulation Group,
Financial Industry Regulatory Authority (FINRA)1
Augmenting trade surveillance programs with artificial intelligence and machine learning | A brief overview | May 2024
Artificial intelligence: A brief overview
The financial markets have generally been a hotbed of With AI being considered as a key element in future innovations,
competitiveness, risk, and innovation. To uphold the economy’s the financial services industry is looking to leverage its potential
good health and to build investor confidence, it is crucial to as a transformative tool. In areas such as improved fraud
maintain integrity and stability of financial markets. In this world detection, risk management, and predictive analytics.2 Some
of fast-paced technological developments, artificial intelligence common AI use cases in the banking and financial services
(AI) is becoming a potent weapon in the field of risk monitoring sector include:
and surveillance. By looking at its diverse applications and
upcoming trends, AI may be a crucial factor in helping to protect Algorithmic trading
financial markets.
Fraud detection
AI involves the use of algorithms and analytics to enable systems
to demonstrate intelligent behavior, including learning from data,
making decisions, and solving problems, all with minimal human Regulatory compliance monitoring
intervention. Similarly, machine learning (ML) is the process of
discovering patterns in data without human intervention and Personalization of financial investment advice
using them to make predictions. Specific to trade surveillance,
systems integrated with AI and ML not only aim to uncover Risk management
suspicious trading patterns, but also to help in reducing the
volume of false alerts, thereby helping mitigate their risk to the Enhancement of customer experience
trading ecosystem.
Augmenting trade surveillance programs with artificial intelligence and machine learning | A brief overview | May 2024
Rule-based surveillance: Status quo
Market surveillance has traditionally been rule-based, generating
alerts based on pre-specified rules/static thresholds that lead to
specific actions when such conditions are breached. Due to high
interpretability, these “traditional” surveillances have the main
advantage of being simple and reliable, meaning they are easier
to understand and develop/enhance standardized rules that
would enable ongoing market surveillance. However, while rule-
based systems have been quite effective in laying a successful
foundation for trade surveillance, they have limitations due to
them not being effective for all use cases:
• Data cleansing requirements: Rule-based systems often
struggle with large and unstructured datasets that require
extensive efforts in cleansing and formatting to make it well-
structured and usable for surveillance systems.
• Tackling new threats: Rule-based systems work based on
prescribed/preconditioned directives; hence they cannot pick
up manipulative patterns that are new or even slightly modified,
resulting in possible surveillance lapses.
• Adaptability across markets: Rule-based systems require
dedicated models to cover different asset classes and markets.
While these systems currently have dynamic thresholds
that may help to an extent, they still need to account for
manageability, as firms may end up with a significant number of • Holistic surveillance is an approach that enables
models and an even higher number of thresholds/parameters simultaneous monitoring across multiple surveillance functions,
across asset classes and markets that require a larger supporting higher-quality tethering and control effectiveness
maintenance effort, making this construct susceptible to errors. between trade and communications data to help identify
false or misleading statements and potential market abuse
To tackle the limitations and challenges of rule-based
behaviors such as "pump and dump," "flying," and "printing".5
surveillance, there is widespread consensus among market
participants and regulators about the need to analyze more • Dynamic parameters can be used to determine and assess
dynamic and robust surveillance insights for the future3 AI and specific trading behaviors more accurately based on factors like
ML models are being considered by both regulatory authorities the standard deviation of client or account trading activities/
and financial institutions (FIs) as an accelerator for market patterns, market conditions, and economic indicators when
surveillance. Alternative solutions are also being explored in compared to static thresholds. This approach can help diminish
parallel for enhancing existing surveillance capabilities as well: the number of false positives, which has been a significant
drawback of rule-based surveillances.
• Quantum computing is being looked at as a potential
accelerator for AI as it could enhance the ability of AI-based • Integration of distributed ledger technology (DLT)/
models to process and analyze large datasets at faster blockchain with AI at the back end for data storage and
speeds.4 retrieval could help tackle the opaque nature of AI. The
immutability, traceability, and decentralized nature of DLT/
• Network and behavioral analysis techniques could help
blockchain enables improved security, transparency of
in revealing hidden connections/relationships/patterns to
execution, and efficiency that could be a strong fit for AI-based
identify potential coordinated market manipulation behaviors.
surveillance systems.
Deviations from normal behavior patterns could result in
identifying evidence of market abuse.
4
Augmenting trade surveillance programs with artificial intelligence and machine learning | A brief overview | May 2024
Rule-based surveillance: Status quo (cont.)
In addition to the above areas, many FIs are looking at AI and ML • Real-time identification of patterns and anomalies:
as an augmentative solution for trade surveillance. Integration AI-enhanced models can identify patterns in trading activities
of AI/ML in existing surveillances has the potential to be a great and relate them to generic or specific market events that
value add for organizations looking to enhance surveillance may indicate trading anomalies. For instance, AI-integrated
efficiency and overcome the limitations posed by traditional surveillance models designed to detect intraday market
surveillance analytics. Some of the benefits AI can provide manipulation risk can identify market volatility resulting from
around surveillance include: index rebalancing, option expiry events, or movement in a
stock’s price because of issuer-specific news and compare
• Adaptive and scalable surveillances: AI-based models
outcomes to historic situations to more effectively trigger
are capable of processing large datasets quickly and highlight
or provide supporting information. This can increase the
evolving patterns of potential market manipulation-related
effectiveness of models to proactively trigger alerts on unusual
activities. The capability of AI-based models to process
participation or movements in price that may not be attributed
large and diverse datasets could assist firms to identify
to any specific external events. The model can also aid in
and manage risk more appropriately. ML models stand out
spotting trends and abnormalities, including ones that rule-
in handling uncertainties as they can provide confidence
based systems might miss, to recognize complex patterns that
scores, or probabilities associated with their predictions,
may not be immediately evident.
which is valuable when dealing with varied trading and order
placement behaviors. The personalization feature of AI/ML • Supporting the surveillance review: The integration of AI
makes it possible to create alerts that are more pertinent in surveillance can enhance the surveillance review process,
and in line with the distinctive market dynamics of various making it more efficient and effective. By leveraging e-discovery
financial instruments. Based on their risk appetites and usecases, electronic communications can be reviewed with
trading methodologies, AI allows organizations to customize greater ease and accuracy. For instance, a four-month review
alert thresholds. For example, with the help of AI/ML models, that required one million documents and a hundred personnel
dynamic thresholds/parameters can be set for a variety of was reduced to six weeks and five personnel by utilizing
clients. Clients with low turnover/trading activity and trading large language models with search prompts. This technology
manually can be distinguished from clients trading in large enables analysts to identify the origin of trades and patterns of
volume and on low-latency/high-frequency flows using complex behavior beyond traditional rule-based surveillance, thereby
trading algorithms. With the help of AI, a clear distinction can enhancing the overall surveillance experience.6
be made while generating alerts for such activities.
• Reduction of false positives: A major concern with rule-
based systems is increasing alert volume and the time involved
in reviewing the same. Integration of AI in trade surveillance
can help to reduce false positives and alerts posing no risk, and
increase learnability and feedback loops with historical market
and surveillance data.
5
Augmenting trade surveillance programs with artificial intelligence and machine learning | A brief overview | May 2024
Feasibility of implementing AI in
surveillance
Trade surveillance requires data from multiple sources including
but not limited to exchanges, venues, trading platforms, news
feeds, and internal trading records. Accessing and integrating data
from these diverse sources can be technically complex. When
combined with surveillance data from communications, it can
lead to a more efficient and effective analysis and interpretation
of suspicious activity. This approach recognizes that different
AI models may excel in different aspects of surveillance and
combining their strengths can lead to more robust results. Various
stand-alone models can be configured and trained for specific
aspects of surveillance. Implementing AI comes with specific
requirements and prerequisites that are essential for its
successful adoption.
• Natural language processing (NLP) models for speech-to-
text data, translation engines, analyzing news, sentiment, and
textual data related to financial markets.
• Time-series models like recurrent neural networks (RNNs)
or long short-term memory (LSTM) networks for detecting
patterns and trends in historical trade data and behavior.
• Graph-based models to analyze the relationships and
connections among different entities in the financial markets,
such as traders, firms, and securities.
• Anomaly detection models to flag unusual trading behavior.
Once the technical infrastructure for AI implementation is in place,
It is vital to define clear objectives and use cases for AI in both
human expertise should be leveraged to add maximum value to
trade and communications surveillance. Whether related to risk
the efficiency of the AI-based model. Right from the thoroughness
appetite of the entities or driven by rules, guidance, or mandates
of data till the end results, human skills are pivotal to choose the
from regulators and venues, having specific goals can aid in
appropriate inputs. If the model produces below-par results, the
greater effectiveness of an AI-based model. It is essential to have
model owner is held responsible since they make all the important
appropriate infrastructure in place. This includes the hardware
decisions pertaining to developing, training, and maintaining
and software needed to collect, store, and process data efficiently.
AI models. The organization’s commitment to meet these
High-performance servers, data management tools, and storage
foundational requirements are important for the success of the
solutions are required for handling the vast amounts of data
AI implementation.
involved across surveillance purposes. Access to a substantial
amount of historical data is essential to train a robust and accurate
model. This helps the AI system learn from past events and identify
patterns of misconduct. Sufficient education across technology,
compliance, and surveillance teams enables the effective use
and feedback loops to improve the effectiveness and efficacy of
integrating AI into surveillance systems.
7
Augmenting trade surveillance programs with artificial intelligence and machine learning | A brief overview | May 2024
Opaque nature of AI
With the use of AI in surveillance, there may be a tendency to be • Assess data quality to provide confidence in the design and
wary of its opaque nature and lack of transparency of underlying operation of models used. This may include using exploratory
algorithms—how the model operates (its dependencies and data analysis as well as sandbox environments with both true-
limitations) and how its predictions or results are produced. This positive and false-positive examples, back-tested data, and
closed approach surrounding AI makes it challenging for a non- stress-tested environments.
technical audience to understand the model logic. To address
• Implement controls below the line and periodic review of
this inherent skepticism, firms can:
results to provide confidence in inputs and outputs.
• Document the design, purpose, and key features of the model
This can help to reduce ambiguity surrounding the opaque
to make clear the inputs and expected outputs.
nature of AI, and organizational personnel can get comfortable
• Build dashboards and visuals to explain the flow of with AI’s capabilities to make consequential compliance decisions.
model decision-making and, subsequently, provide detailed
explanations of alert predictions.
8
Augmenting trade surveillance programs with artificial intelligence and machine learning | A brief overview | May 2024
AI/ML in surveillance: The regulatory
perspective
The banking and financial services sector is highly regulated, • Accountability – Ensuring policies are in place to determine
constantly trying to uphold its fundamental value of data responsibility and ownership of decisions made by the AI/ML
governance/protection and customer privacy.7 However, models.
increasing adoption of AI and ML can pose a distinct challenge
• Safety and security – Ensuring controls are in place to protect
regarding model explainability.
the AI/ML models from risks that could have a significant
ML models often provide for some explainability in terms of the negative impact on the firm/stakeholders.
underlying assumptions and factors considered when making a
• Reliability and robustness – Implementing controls to
prediction. The regulators recommend improving explainability
ensure accurate outputs, withstand errors, and quickly
of the AI/ML model being used to help users and supervisors
recovery from unforeseen disruptions.
understand the functionality by breaking down the opaque
nature to provide clarity.8 To tackle the challenges stated above, At Deloitte, we provide an end-to-end framework to assist with
establishing a trustworthy AI framework that helps organizations the implementation of AI that echoes the application of all the
develop ethical safeguards to address key concerns across above-stated dimensions to build an ethically adept AI/ML
the following dimensions is crucial in managing the risks and system. Please refer to Deloitte’s Trustworthy AI™ framework
capitalizing on the returns associated with AI: to learn more.
• User privacy – Implementing controls to ensure data usage is
limited to its intended and stated use and duration, with users
having the option to share data.
• Transparency and explainability – Tackling the opaque
nature of AI to ensure that users understand how the AI/ML
models work by explaining the inputs, inherent logic involved in
decision-making and outputs such that the decision-making is
clearly understood, auditable, and open to inspection.
9
Augmenting trade surveillance programs with artificial intelligence and machine learning | A brief overview | May 2024
AI/ML in surveillance: Technical
standpoint
An AI-implemented surveillance solution is considered effective
if it is proficient in recognizing trading patterns. Market
manipulation patterns can be recognized through a
combination of data analysis, pattern recognition algorithms,
and machine learning techniques. Prior to analysis, relevant
trading data is sourced and prepared for pre-processing and
feature extraction. AI algorithms are used to identify patterns
within the pre-processed data–these could be trend-based,
reversal-based, or other trading pattern types used to detect
anomalous trading behavior. Supervised and unsupervised
learning approaches are used to train ML models so that
trading outcomes are more closely correlated to input features.
Techniques such as clustering, dimensionality reduction, time
series analysis, and deep learning can be leveraged in pattern
identification (deep learning is a method in AI that teaches
computers to process data in a way that is inspired by the
human brain). Identified trading patterns are reviewed via
model evaluation, back testing, and validation including manual
analysis wherever necessary. Since market dynamics are ever
changing, the nature of market manipulation patterns needs to
be evaluated on an ongoing basis, so that AI-based surveillance surveillance, practitioners should have a strong grip on market
models keep performing effectively. abuse/manipulation processes, models, and regulations so
that business requirements and technical specifications are
ML models score alerts, not only based on the data points
aligned. It should be noted that AI is still a developing area, hence
directly related to the alert (e.g., parameter or threshold breaches
knowledge of technologies and concepts needs to be updated on
of volumes or prices), but also on how similar alerts have been
an ongoing basis.
classified by the firm’s risk and compliance division previously.
AI-based alert scoring is particularly useful when alerts are There are multiple benefits in leveraging AI in the surveillance
generated through a traditional rule-based approach. This is world; however, having skilled resources to implement AI is
because the scoring functionality can be considered as a second essential. A group of professionals well-versed in AI technologies
layer, which is implemented on top of the regular alert-generating and concepts are more likely to effectively bring AI into practice.
process and, as such, can also optimize the outcome of legacy In this regard, surveillance professionals still have some way to
trade surveillance systems. go in being AI proficient and are currently dependent on
technical specialists for AI implementation. To bridge this gap,
Being an area with vast potential and a steep learning curve,
training on AI concepts and use cases can help traditional
experienced practitioners of AI in surveillance are in short supply.
surveillance professionals become more familiar with
Developing an understanding of AI concepts and techniques
onboarding AI solutions.
requires sound knowledge of data analysis, feature extraction,
and anomaly detection. Additionally, having working knowledge
of econometrics (regression, time series analysis, etc.) and ML
helps to develop clearer concepts of AI model development,
testing, validation, and governance. For implementing AI in
10
Emerging trends in digital assets manipulation and surveillance | April 2024
Augmenting trade surveillance programs with artificial intelligence and machine learning | A brief overview | May 2024
Conclusion and key takeaways
Financial institutions can focus on developing and improving As stated earlier, building and implementing an AI framework
surveillance models as an integral part of their journey to with governance and regulatory safeguards across key
establish an extensive surveillance program. A well-defined dimensions such as data privacy, accountability, and reliability
and understood risk framework serves as the scaffolding to is a crucial step in managing the risks and capitalizing on
construct and operationalize a trustworthy AI program. Once the the returns associated with artificial intelligence. Please
risk models to be covered under the surveillance program are refer to Deloitte’s Trustworthy AI™ framework to learn more
decided, the risk and compliance team can evaluate which alert about our end-to-end framework to help synergize ethical AI
models can benefit from implementation of AI. AI can be used to implementation and integration with organizations.
either create a new surveillance model, adjust an existing one,
Organizations should view the implementation of AI in
or improve rule-based and static surveillance models. It may
surveillance as an evolving and an ongoing process; while AI may
not always be beneficial to develop AI into simple rule-based
require resources like technology, infrastructure, and skilled
surveillances like potential wash trading and locates. Also, AI
human capital, learning, testing, enhancing, and iterating needs
should not be incorporated into surveillance programs of small
to happen on an ongoing basis to be successful. This can be
firms with manageable trading volumes as it would not have
achieved with the help of a dedicated AI center of excellence
a drastic impact on efficiency or effectiveness compared to
(CoE) within the organization. Institutions need to evaluate and
existing off-the-shelf products. However, large firms dealing with
prioritize accordingly to help them achieve their desired goals
significant trade, order, and communications data; using complex
while incorporating AI in surveillance. While the future prospect
trading mechanisms; and having sophisticated clients will likely
of AI in surveillance is exciting, it is imperative to understand
benefit more from implementing AI into their existing
that this is a long journey, and the optimal way to progress is
surveillance programs.
through an effective collaboration between firms and regulatory
authorities in taking this forward.
12
Contacts
Roy Ben-Hur Nitin B S
Managing Director Senior Consultant
Risk & Financial Advisory Risk & Financial Advisory
Deloitte & Touche LLP Deloitte & Touche Assurance & Enterprise
[email protected] Risk Services India Private Limited
[email protected]
Adam Clarke
Director Kewal Harshad Jagani
Risk Advisory Senior Consultant
Deloitte UK Risk & Financial Advisory
[email protected] Deloitte & Touche Assurance & Enterprise
Risk Services India Private Limited
Niv Bodor
[email protected]
Senior Manager
Risk & Financial Advisory Subramanian Krishnan
Deloitte & Touche LLP Senior Consultant
[email protected] Risk & Financial Advisory
Deloitte & Touche Assurance & Enterprise
Anand Ananthapadmanabhan
Risk Services India Private Limited
Senior Manager
[email protected]
Risk & Financial Advisory
Deloitte & Touche Assurance & Enterprise Anuj Khasgiwala
Risk Services India Private Limited Senior Consultant
[email protected] Risk & Financial Advisory
Deloitte & Touche Assurance & Enterprise
David Isherwood
Risk Services India Private Limited
Senior Manager
[email protected]
Risk Advisory
Deloitte UK S Goutham
[email protected] Consultant
Risk & Financial Advisory
Romit Deb Mookerjea
Deloitte & Touche Assurance & Enterprise
Manager
Risk Services India Private Limited
Risk & Financial Advisory
[email protected]
Deloitte & Touche Assurance & Enterprise
Risk Services India Private Limited
[email protected]
13
Endnotes
1. Financial Industry Regulatory Authority (FINRA), “Deep learning: The future of the Market Manipulation Surveillance Program,”
FINRA Unscripted podcast (ep. 98), January 25, 2022.
2. Deloitte, “How artificial intelligence is transforming the financial services industry,” accessed April 2024.
3. FINRA, “Deep learning: The future of the Market Manipulation Surveillance Program”; FINRA, “AI applications in the securities
industry,” from Artificial intelligence (AI) in the securities industry, June 2020.
4. FINRA, “Section II: Potential applications of quantum computing in the securities industry,” from Quantum computing and the
implications for the security industry, October 2023.
5. Financial Conduct Authority (FCA), Market Watch 76, January 2024.
6. Deloitte, “Deloitte launches new Generative AI-powered solution on RelativityOne and Relativity Server to help organizations
accelerate document review, employee conduct investigations, PII identification and compliance activities,” press release, January
22, 2024.
7. FINRA, “Key challenges and regulatory considerations,” from Artificial intelligence (AI) in the securities industry, June 2020.
8. Deloitte, “Trustworthy AI™,” accessed April 2024.
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 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.
Printing instructions
When printing through your web browser using the default settings, white bars may appear on the top and bottom of a letter-size
sheet (8.5”x11”). To avoid this, either print on a legal-size sheet (8.5”x14”) or follow the instructions below:
If printing through Microsoft Edge:
1. Click on the print icon. 3. Change “Scale (%)” to “Actual size.”
2. Scroll down to “More settings”
and click to expand the menu.
4. Click on the print button to
print the document.
If printing through Google Chrome:
1. Click on the print icon. 3. Change “Scale (%)” to “Default” (if needed).
2. Scroll down to “More settings”
and click to expand the menu. 4. Click on the print button to print the document. |
348 | deloitte | us-ai-and-ceo-of-the-future.pdf | The role of the CEO in
tomorrow’s Generative AI world
Leading a Generative AI-fueled enterprise:
A CEO series
Deloitte Global CEO Program
Deloitte AI InstituteTM
TThhee rroollee ooff tthhee CCEEOO iinn ttoommoorrrrooww’’ss GGeenneerraattiivvee AAII wwoorrlldd
About the Deloitte Global CEO Program
The Deloitte Global CEO Program is dedicated to advising
chief executive officers throughout their careers—from
navigating critical points of inflection, to designing a strategic
agenda, to leading through personal and organizational
change. The program offers innovative insight and immersive
experiences to help:
This paper is a work of fiction and the
• Facilitate the personal success of individual executives, new
product of the authors’ imaginations.
or tenured, throughout their life cycle.
• Elevate the relationships between them, their leadership
It presents a dramatic scenario designed to provoke thoughtful
teams, and their boards.
conversations and spur difficult questions about the kind of
AI-enabled future today’s business leaders might envision for • Support the strategic agenda for their organizations in
their organizations and workforces. times of disruption and transformation.
At Deloitte, we believe in the power of human and machine
www.deloitte.com/us/ceo
collaboration, where the human workforce is augmented by
AI to become more efficient and productive. To learn more
about Deloitte’s views on AI and get recommendations on
implementing AI, please visit: About the Deloitte AI InstituteTM
• State of Generative AI in the Enterprise 2024
The Deloitte AI Institute helps organizations connect
• Deloitte AI Institute the different dimensions of a robust, highly dynamic
and rapidly evolving AI ecosystem. The AI Institute leads
• Generative AI Services 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
22
The role of the CEO in tomorrow’s Generative AI world
The role of the CEO in
tomorrow’s Generative AI world
As recurring waves of technology disruption occur, CEOs have been called to
articulate increasingly innovative visions for their organizations. With
Generative AI, the disruption has finally reached the role of the CEO itself.
In tomorrow’s AI world, many aspects of the CEO role and the CEO experience
will change. This change will challenge CEOs well beyond the rational
dimension and into the moral, emotional, and even existential levels.
To explore these changes, we aim to help CEOs immerse themselves in
tomorrow’s AI world through fictional narratives.
Below, we present a day-in-the-life of an imaginary CEO today, along with
the same day re-imagined in 2030, where the same characters live in a
world transformed by AI.
Our intent is not to provide precise predictions or answers. Rather, our
goal is to envision a plausible future and initiate a debate on its implications
for the choice of being (and the requirements to succeed as) a CEO.
3
5:45 am 5:37 am
Beethoven’s 5th blares in the dark and Sanjay darts out The second movement of Tchaikovsky’s 4th plays in the
his hand to shut off the music. dark, nudging Sanjay out of his slumber at the most
optimal time in his REM cycle. Synched with his
“Wow,” his wife Dana mutters, groggy. “What was that?” biometrics and tailored to the day’s calendar, the music
starts to enter its first crescendo as he sweeps his legs
“Leo told me he listens to classical music every day since
off the bed. He shuts it off so his wife Dana can sleep a
becoming CEO of Q.Helix. Giving it a try,” Sanjay says,
bit longer.
rubbing his eyes.
Sanjay’s mind is already jogging: thoughts of the
“Nothing like a morning heart attack to get you going,” she
afternoon’s board meeting (they’ll want to investigate the
mocks, and goes back to bed. Sanjay’s mind is already
volatility in company stock) and the thousand other
jogging: thoughts of the afternoon’s board meeting (they’ll
things on his to do-list. The phone display reads 5:37 AM.
want to investigate the volatility in company stock) and
the thousand other things on his to do-list. The phone He settles into his seat a minute early for the 6 AM
display reads 5:45 AM. Peloton session with an old college friend. They text each
other during the water breaks to catch up, and the friend
Sanjay settles into his seat a minute early for the 6 AM
won’t stop raving about the latest VR game he plays with
Peloton session with an old college friend. They text each
his family. Sanjay promises to give the game a chance
other during the water breaks to catch up, and the friend
that night. “You can afford to play something that’s not a
won’t stop raving about the latest season of The Bear.
strategy game,” his MBA section mate jokes.
Sanjay promises to give the show a chance that night.
“You can afford to try something that’s not a business
podcast for once,” his MBA section mate jokes.
College buddy
suggests actively
playing a virtual
reality game
College buddy CEO wakes up
suggests passively at 5:37 as
watching the series, Tchaikovsky
The Bear alarm syncs to
CEO’s sleep
CEO wakes up
biometrics
at 5:45 to
and calendar
Beethoven’s 5th
2024 2030
A day in the life of a CEO: 2024 Same day reimagined: 2030
4
The role of the CEO in tomorrow’s Generative AI world
7:00 am 7:00 am
After a shower and a glance at his packed calendar, After a shower, Sanjay’s health AI recommends a protein
Sanjay eats eggs and toast alongside Dana and his two shake and fruit for his first meal. He prepares those and
teenagers, Viraj and Avani. Sanjay’s distracted from sits alongside Dana and his two teenagers, Viraj and
thoughts of the day’s meetings by Viraj’s fidgeting. He’ll Avani. Viraj signals that he’s busy preparing for an exam
need to confirm his son’s doctor appointment regarding using his smart glasses, and an intermittent buzz on his
his son’s apparent inability to concentrate on tasks. son’s wrist keeps him concentrated. He takes a quick
glance at the day’s calendar, organized by his digital Chief
Dana picks up her lecture materials for the day’s classes of Staff (DCoS), an AI agent that he’s nicknamed Erika.
and heads out the driveway with the kids, while Sanjay
greets the driver of his company car. By the time he Dana picks up her lecture materials for the day’s classes
arrives in the office, he’s read the headlines from his and heads out the driveway with the kids. Sanjay boards
favorite newspapers, digested the daily e-mail from his an autonomous car, and by the time he arrives in the
Chief of Staff (CoS) Erik, and sent the family an article on office, he’s prepared for the day. He listens to a
the benefits of waking up to classical music. The latter customized daily download of article summaries from his
elicits barf emojis from his children and silence from favorite newspapers, and then reviews a series of
Dana, no surprise. decisions Erika has prepared. Speaking into his earpiece,
he verbally approves a change to his schedule to fly to
Milwaukee for an in-person client meeting, but leaves a
voice note for Erika to ensure he doesn’t miss his tennis
tournament. As Erika continues presenting other items
that require approval, along with suggestions, Sanjay
sends the family an article on the benefits of waking up
to classical music. The latter elicits barf emojis from his
children and no response from Dana, no surprise. Erika
offers a potential “non-Dad” response to them. Sanjay
decides to pass.
CEO gets picked up
by the driver and
reviews daily email
from Chief of Staff
Health AI
recommends a
protein shake
and fruit CEO gets picked up by
Breakfast of eggs
autonomous car;
and toast selected
virtual assistant briefs
randomly
CEO on a variety of
topics and resolves
scheduling conflicts
5
2024 2030
5
The role of the CEO in tomorrow’s Generative AI world
8:00 am 8:00 am
Erik greets him at their favorite coffee shop, 2 blocks from At reception, a coffee is ready for him as he walks in.
the office, and they walk and talk, reviewing the calendar. Sitting at his smart desk, he asks the DCoS to display the
Erik breaks the news of a virtual meeting being more sensitive parts of his daily download, including
rescheduled so Sanjay can fly to Milwaukee and attend stock price movement, and sales and operational data.
in-person. Sanjay pinpoints which appointments to move Erika provides a theory about the recent stock volatility:
around so he won’t have to miss his tennis tournament. By scanning social media chatter, it has identified
growing discontent among junior staff, as a result of
At the office, they begin reviewing the weekly report changes in HR benefits, that could’ve caused the concern
about the company’s stock price movement, sales data, among investors. Sanjay directs Erika to generate and
and operational updates, but as usual, they’re not finished send a list of questions to Hector, the CHRO, inquiring as
by the half-hour mark. to the latest in employee morale.
Time is short,
unfinished work
Clear
communication
made possible
with AI assistance
Erika, the virtual saving time
assistant, identifies
employee morale
Erik, the CEO’s issue while CEO is
Chief of Staff seated at the
meets to resolve smart desk
scheduling issues
over coffee
6
2024 2030
6
The role of the CEO in tomorrow’s Generative AI world
8:30 am 8:30 am
Sanjay and Erik stroll into their kitchen cabinet meeting Sanjay strolls into his kitchen cabinet meeting around
around 8:30. Hector, the CHRO, appears on edge and 8:30. The COO, also an AI agent, agrees with Erika’s
confesses concerns about a recent benefits explanation and recommends retracting the recent
communication that has irked some employees. “I’m benefits communication. “Understood. I’m going to send
going to send a follow-up clarification,” he declares. The a follow-up clarification,” Hector declares. The group then
COO, Marcus, pipes up: “Someone always takes things the views AI simulations of their board members to
wrong way. A clarifying message may help shut down the anticipate questions and align on responses.
rumor mill.” After other bits of advice from around the
table, the group shifts to rehearsing their talking points
for the board meeting.
Virtual AI agent
recommends
a follow-up
to employees
CHRO warns
of disgruntled
employees
AI board member
simulations help
executive team
anticipate
questions and
align on
responses for
Other
board meeting
executives
chime in for a
resolution
7
2024 2030
The role of the CEO in tomorrow’s Generative AI world
9:01 am 9:01 am
At 9:01 AM, Sanjay makes it back to his office for a At 9:01 AM, Erika reminds Sanjay of a meeting with the
meeting with the head of sales, Ed. He listens to the head of sales, Ed. Sanjay deputizes it to attend the
progress of their retail stores across the southern states meeting on his behalf. He then continues rehearsing with
and fires off a few decisions on investment and location his cabinet for the board meeting, and notices that Erika
strategy. He can’t help but wonder if his time was elevates one question from Ed to his attention: Optimize
needed for this. for short-term performance (quarterly results) or
long-term enterprise value? Given the pressure on the
Soon after he’s done, Erik walks in to finish their calendar stock price, Sanjay hesitantly picks the former. Even with
review, but Sanjay raises a hand to halt him. He takes a AI assistance, he still has to make the no-win decisions,
moment, like the pause before hitting a serve, to admire he realizes.
the sun glinting on the skyscrapers. Erik waits for about
one minute and then clears his throat. “Media interview When he’s done with his cabinet meeting, Sanjay feels
in five.” better prepared and optimistic that the HR action may
improve market sentiment. He walks back to his office
and takes a long moment, like the pause before hitting a
serve, to just admire the sun glinting on the skyscrapers.
Erika reminds him in his earpiece: “Media interview in 5.”
Sanjay asks: “Context?”
“Reporter has questions about interim decarbonization
metrics in light of company commitment to net-zero
emissions by 2050.”
“See if they’ll talk to Sarah, our CSO, instead.”
“Noted.”
CEO meets with
Time freed up
head of sales on
with help of
location strategy
virtual assistant
of retail stores
that delegates
feeling others
the interview with
could have made Virtual assistant
the CSO
those decisions only brings key
CEO is feeling decisions to CEO,
rushed to media who still must
interview make them
8
2024 2030
The role of the CEO in tomorrow’s Generative AI world
9:30 am 9:30 am
The reporter peppers Sanjay with questions about Sanjay uses the freed-up time to listen and dictate
interim decarbonization metrics, given their commitment responses to urgent e-mails that Erika has surfaced as
to net-zero emissions by 2050. Erik listens and adds worth his attention. He notices his digital PR agent has
details to Sanjay’s answers as needed. Sanjay wonders if posted on his behalf about his upcoming industry panel
Erik could simply handle these calls on his own someday. on the future of fashion. It’s a picture of Sanjay from the
90s, wearing a denim jacket at college in London, and a
After the interview, they depart for an industry panel satiric message about the future being more of the past.
covering the future of fashion, where Sanjay is scheduled Not bad.
to speak and answer a few questions from the audience.
He’d planned to review Erik’s notes to prepare answers, Then, he boards a self-driving car to attend the panel.
but his attention is drawn to a few urgent emails. He also Armed with Erika’s detailed notes, Sanjay strikes gold. His
notices that Yanna, the PR consultant, has posted on his reflections on the impact of recommendation engines
social media about his upcoming appearance at the and digital filtering on clothing draw applause from the
industry panel. crowd, and several people approach him after his
remarks to ask questions and network. Erika listens and
At the panel, Sanjay doesn’t exactly strike gold, analyzes all his interactions so that it can send
meandering through answers on the impact of spatial follow-ups.
computing on brick-and-mortar stores. People approach
him after the panel to network, but Sanjay cuts the As Sanjay steps outside the convention center, the crisp
conversations short and walks outside to clear his head, air and the smell of the river reinvigorate him. He sees
taking in the crisp air and the smell of the river. The the famed Lyric Opera building just down the water— a
famed Lyric Opera building is just down the water—a good opportunity to enjoy the city as a family. He’d even
good opportunity to enjoy the city as a family. He’d even read this morning that classical music could help with
read this morning that classical music could help with focus, if he could convince Viraj to give it a try. He
focus, if he could convince Viraj to give it a try. wonders whether Erika would have advice for how to
approach the topic with Viraj.
CEO meanders
through questions
for an industry
panel with
lackluster
performance
CEO peppered
Virtual
with questions
assistant helps
from reporter
make time for
urgent tasks
9
2024 2030
With media interview
delegated, CEO is
briefed for panel by
virtual assistant
Successful panel
discussion followed by
energetic networking
The role of the CEO in tomorrow’s Generative AI world
12:00 pm 12:00 pm
At noon, he walks into Gibson’s for lunch with Elaine, the At noon, he walks into Gibson’s for lunch with Elaine, the
CEO of one of his company’s top distributors. Elaine, who CEO of one of his company‘s top distributors. Elaine, who
usually asks about his family, heads straight into the usually asks about his family, heads straight into the
subject of recent geo-political tensions and the potential subject of geo-political tensions and the potential impact
impact on manufacturing schedules. “There’ll be hardly on manufacturing schedules. “There’ll be hardly any
any impact on our deliveries,” Sanjay assures her, based impact on our deliveries,” Sanjay replies, and pulls out
on his broad knowledge of the supply chain and his views his phone. “Let’s run the scenarios.”
of the geo-political outlook. After addressing her
concerns over appetizers and resolving a pending “Has that been listening to everything I’m saying?”
agreement over entrée salads, they can finally chit-chat Elaine inquires.
over dessert. “Viraj beating you on the court yet?” Elaine
“Well, that’s the best way to get the answers you need,”
teases him.
Sanjay replies.
“I can neither confirm nor deny,” Sanjay shoots back.
Even as Erika provides reassuring numbers, Elaine crosses
“How is Dave’s hip?” he inquires about Elaine’s husband
her arms in silence. Sanjay puts his phone away as he tries
as he reaches for the bill.
to address Elaine’s concerns over appetizers and resolve a
“Well, it’s titanium now. He keeps bumping into the pending agreement over entrée salads. They finally
furniture to test it out!” As they share a laugh, Sanjay chit-chat over dessert. “How is Dave’s hip?” He asks about
knows the meeting was a success. Elaine’s husband as he reaches for the bill.
“It’s fine,” Elaine responds, not eager to start their usual
banter. With the conversation more stilted than ever
before, Sanjay leaves the meeting unsure if it’s a success
and decides to ask Erika. She confirms that, based on her
tone of voice, word selection, and unwillingness to engage
in personal chitchat, Elaine felt threatened and untrusting.
Sanjay wonders about how he can repair the damage, and
his instinct tells him not to ask Erika for advice.
Client happily
engages in
personal chitchat
Client is upset
she’s been
CEO addresses recorded without
client concerns permission so,
over lunch CEO explains and
tries to reassure
10
2024 2030 Virtual assistant
confirms client felt
threatened and
untrusting
10
The role of the CEO in tomorrow’s Generative AI world
1:20 pm 1:20 pm
As Sanjay reaches his office, he requests Erik to attend a As Sanjay reaches his office, he heads into a meeting with
strategy meeting with Sarah, the CSO, on his behalf. Erik Sarah, his CSO. They debate at length the amount of AI
hesitates. “Sarah needs an important decision on how chips they’ll need to invest in over the next five years,
many AI chips to buy.” and Sanjay ultimately throws the issue to Erika.
Sanjay loses his cool for a moment and insists, more curt Back in his office, Sanjay decides to use a digital
than usual. “Sarah will understand.” Erik acquiesces but surrogate for the 2 PM virtual townhall. He takes the
doesn’t look convinced. extra time to continue preparing for the board meeting,
and then joins the last 15 minutes of the call for live Q&A.
At 2 PM, Sanjay logs on to a webinar for a virtual town hall A few of the employees appear delighted at how effective
with employees. He delivers the updates that Erik the surrogate was, but others appear miffed, and one
prepped in a document last week and adds in personal even sends him a direct message: “You can’t delegate
stories about his recent family trip to Disney, and an culture.” The poll administered at the end proves the
anecdote from the panel he’d just attended. The poll employee right: only 59% of employees are pleased with
administered at the end empowers him to breathe a sigh the company’s direction and leadership.
of relief: 89% of employees are pleased with the
company’s direction and leadership. Sanjay strolls out of his office, hoping to make small talk
with others about what just happened, but there are
Sanjay strolls out of his corner office feeling energized and fewer employees in the office these days. He hears the
makes small talk in the halls. As he peruses the snacks in pings of their AI assistants reminding them of upcoming
the break room, an intern, who doesn’t recognize him, meetings. He strides into the board meeting a few
asserts that the fruit bars are great for staying awake in minutes early and takes his seat.
meetings. As Sanjay grabs one, another employee reveals
that he’s the CEO. Sanjay smiles warmly as the intern’s
face turns beet red.
Erika, the virtual
Personal rapport with assistant, helps to
employees, positive calculate needs for
feedback after virtual AI chips
town hall
Less than stellar
response after
surrogate takes
CEO’s place at
town hall
CEO delegates
chip decision
11
2024 2030
11
The role of the CEO in tomorrow’s Generative AI world
3:00 pm 3:00 pm
The three-hour board meeting commences with a The board meeting commences with an AI-curated
review of the minutes. Sanjay provides his executive summary of the previous meeting. The chairman
report, followed by Linh, the CFO, and Hector. Then, the explains that one of the board members has sent a
independent committees share their updates—Audit digital surrogate because she is unable to attend herself.
and Governance and the others. Finally, they begin the When it’s time for Sanjay to deliver his remarks, Erika
strategic discussions. Sanjay sits taller in his chair and generates visuals in a hologram. The CFO and others
opens up a fruit bar. Time is consumed by the opening deliver their reports, followed by updates from the
and introduction of various documents and visual aids. independent committees. Once they begin the scenario
Still, they manage to have the much-awaited discussion planning discussion, Erika also generates live meeting
on updates to the five-year strategic plan and notes and strategic scenarios. The board turns out to be
compliance issues. Sanjay relies on Erik’s notes to speak largely aligned, and Sanjay thinks the meeting might even
about the progress on AI adoption. The discussion end early. Yet, one board member complains about the
seems a success, though the stock volatility is still a surrogate in the room, and the conversation devolves.
question. Multiple members express uneasiness and the chair
proposes a thorough in-person conversation about rules
of engagement at the next meeting. Everybody, including
the Gen AI surrogate present, agree.
Visual aids
presented as
holograms
Time is consumed
by cumbersome
visual aids
Board members
discuss
uneasiness
CEO relies on about surrogate
notes of his in meeting
Chief of Staff
12
2024 2030
The role of the CEO in tomorrow’s Generative AI world
6:00 pm 6:00 pm
When Sanjay returns to his office at 6 PM, his phone rings When Sanjay returns to his office at 6 PM, his phone rings
with a call from a key investor. His tone is frantic as he with a call from a key investor. His tone is frantic.
interrogates Sanjay about the morning panel. The market Someone had apparently leaked news that their
interpreted some of Sanjay’s comments as negative, the company would be removing sustainability metrics from
investor said, and he wanted assurance that the stock its store growth in the southern states. The market might
price hit would be momentary. “Why didn’t Erik tell me like it, the investor said, but he didn’t like the implications
about the price movement?” Sanjay mutters to himself. for the company’s environmental impact. Sanjay
stammers, struggling to answer, and promises to look
Sanjay rehashes the same reassurances he’d delivered to into it. “Why didn’t you mention that this morning?”
Elaine that morning, but as soon as that call ends, the Sanjay mutters into his earpiece.
chair of the board rings, wanting to debrief about the
strategic plan updates. When Sanjay’s ready to leave the
office, Erik swings by to apologize for not mentioning the
price dip; it just slipped through all the other points of
discussion. “Never enough time,” Sanjay nods. He wants
to allay Erik’s contrition, but he’s too irritated.
13
2024
Nowhere to
channel
frustration at
misjudgment by
virtual CoS
CEO irritated
by Chief of
Staff’s error
Virtual assistant
fails to forewarn
of leaked news
CEO caught off
guard by upset
investor
2030
13
The role of the CEO in tomorrow’s Generative AI world
7:00 pm 7:00 pm
Before heading home, Sanjay has dinner with one of his Before heading home, Sanjay has dinner with one of his
mentees, Jill, who insisted on sushi. She’s climbing up the mentees, Jill, who insisted on sushi. She’s climbing up the
ranks of her private equity firm and wants his advice on a ranks of her private equity firm and wants his advice on a
decision Sanjay has faced before: stay and rise within or decision Sanjay has faced before: stay and rise within or
leave for other opportunities. After hearing her thoughts, leave for other opportunities. After hearing her thoughts,
he asks: “Have you thought about starting your own he asks “Have you thought about starting your own
business?” Jill’s face lights up. He orders plates of uni, and business?” Jill’s face lights up. He orders plates of uni, and
they get down to brainstorming—it’s the most fun he has they get down to brainstorming—it’s the most fun he has
all day. all day.
In the car on the way home, Sanjay stares at his unread In the car, traveling north to his house, Sanjay has Erika
emails, including a lengthy update from Hector that process through the e-mails in his inbox again, and the
makes him decide to procrastinate. Hector’s emails tend most urgent message is from Hector, the CHRO.
to elicit that response from him. Instead, he plays the full Apparently, his new communication had mixed results,
Beethoven symphony that had shocked him out of bed and he wonders what to do next. Sanjay dictates a
in the morning. The jarring highs and soothing lulls over nuanced reply and promises to meet with Hector
the thirty-six-minute composition, the sense of tension tomorrow. After the other minor e-mails are summarized
rising and never dissipating, make it perfectly relatable. and filed away, Sanjay asks Erika to pick a symphony
based on his day. The jarring highs and soothing lulls of
Beethoven’s 5th, the sense of tension rising and never
dissipating, make it perfectly relatable.
14
2024 2030
The virtual
assistant knows
exactly what to
play
CEO picks music
to play
Just the same,
the CEO and
his mentee
brainstorm for
The CEO and his answers
mentee
brainstorm for
answers
14
The role of the CEO in tomorrow’s Generative AI world
9:30 pm 9:30 pm
The sound of opening his front door instantly puts Sanjay The sound of opening his front door instantly puts Sanjay
at ease. It’s Avani’s turn for chores and he gives her at ease. It’s Avani’s turn for chores and he gives her
company, chopping veggies for the next day’s lunch as company, chopping veggies for the next day’s lunch as
she recounts the latest unforgivable lapses of her soccer she recounts the latest unforgivable lapses of her soccer
coach. Sanjay’s phone continues to prompt him with coach.
reminders about incoming e-mails and texts, distracting
him from what his daughter is saying. He campaigns for the family to put on their smart glasses
together and play the VR game his friend had
He campaigns for the family to watch an episode of The recommended. They have a blast fighting their way
Bear together, but within the first fifteen minutes, he through an imaginative world of zombies, each of them
admits, “This show is frantic.” Instead of being relaxed, he discovering a special skill within the team. Erika remains
can’t help but take out his phone to answer some of the silent throughout the evening: though it’s assessing
incoming pings from Erik and others. The kids pull out incoming messages and scanning newsfeeds for
their phones in quick succession. significant events; none justify interrupting Sanjay’s
family time. After spending a little too long sucked into
In bed, he’s still dealing with a batch of e-mails, staring the game, the family parts ways to their bedrooms.
into the light of the device as Dana sleeps beside him. At
midnight, he’s halfway through the inbox and decides to Under the covers, Sanjay’s tempted to turn on his device
declare victory. There will be time tomorrow, he reminds and see if Hector has replied, but he decides against it.
himself. There will be time tomorrow, he assures himself, as he
falls asleep to more gentle music.
15
2024
After time with
the family, CEO
chooses sleep
over phone
Frantic TV show Virtual assistant
stimulates filters incoming
restlessness and messages for
phone use urgency, allowing
for quality time
with family
2030
CEO uses phone
in bed to answer
emails
15
TThhee rroollee ooff tthhee CCEEOO iinn ttoommoorrrrooww’’ss GGeenneerraattiivvee AAII wwoorrlldd
The role of the CEO in
tomorrow’s Generative AI world
The series of vignettes above present plausible futures
for CEOs, touching on select tensions to be navigated.
Our hope is that they will help you reflect on the issues to
explore on your way to your preferred tomorrow. Here are
a few questions to further the conversation.
• G iven the 2030 day, would you still want to be CEO in this future world? Why?
What are the different layers of introspection (rational, moral, emotional, and
even existential) involved in answering this question?
• W hat is added, what is gained, and what is lost for you as a leader and human?
• H ow might CEOs communicate differently in the Gen AI-rich environment
of 2030?
• H ow will you use the time AI will save? (e.g., attending a live town hall
but delegating decision-making)
• W hat are the possible implications of Gen AI for Talent strategy?
What does an enterprise workforce look like in 2030?
• H ow will Gen AI affect enterprise performance?
• W hat’s the price we’ll pay for fast and convenient access to information?
• H ow will we address accountability for decisions delegated to AI?
Will it actually be possible to go against AI recommendations?
• W hat activities and elements are best left to humans?
• W hat are your observations about the interactions between humans
and AI agents in the story?
1166
The role of the CEO in tomorrow’s Generative AI world
17
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 maki |
349 | deloitte | us-ai-institute-ai-bill-of-rights-new.pdf | Deloitte’s Trustworthy AI™ Framework
and the White House Blueprint
for an AI Bill of Rights
November 2022
Brochure / report title goes here | Section title goes here Deloitte’s Trustworthy AI™ Framework and the White House Blueprint for an AI Bill of Rights
The blueprint for an AI Bill of Rights
supports an evolving regulatory landscape
The Artificial Intelligence (AI) regulatory Figure 1 | Definition of Rights, Opportunities, or Access
according to the White House's Blueprint for an AI Bill of Rights.4
landscape continues to mature as government
agencies refine and build upon previous
guidance designed to manage AI risk, ensure
equality and transparency, and provide trust in Civil rights, civil liberties, and privacy,
automated systems. As American institutions including freedom of speech, voting, and
continue to innovate and embrace AI to harness protections from discrimination, excessive
its benefits; federal, state, and local agencies punishment, unlawful surveillance, and violations
of privacy and other freedoms
are increasing their regulatory efforts to protect
in both public and private sector contexts;
the American public.1 The latest iteration
of federal regulatory guidelines is The
Blueprint for an AI Bill of Rights (AIBoR).2
Equal opportunities, including equitable access
to education, housing, credit, employment, and
In October 2022, the White House Office of other programs; or,
Science and Technology Policy (OSTP) released
the AIBoR to provide additional guidance for
organizations to create trustworthy and ethical Access to critical resources or services,
such as health care, financial services, safety,
automated systems. The AIBoR provides
social services, non-deceptive information about
guidance to American innovators to harness
goods and services, and government benefits.
the extraordinary potential and benefits of
automated systems and AI while protecting
“the American public’s rights, opportunities,
or access to critical resources or services.”3
The AIBoR applies to all automated systems
that have the potential to meaningfully impact
individuals’ or communities’ exercise of rights,
opportunities, or access (FIGURE 1).
2 2
DDeellooiittttee’’ss TTrruussttwwoorrtthhyy AAII™™ FFrraammeewwoorrkk aanndd tthhee WWhhiittee HHoouussee BBlluueepprriinntt ffoorr aann AAII BBiillll ooff RRiigghhttss
At Deloitte, we recognized the critical importance of the need
for protections and for AI-Fueled organizations to earn trust
in the AI-enabled assets and services provided to the public.
Deloitte's Trustworthy AITM Framework and can be particularly susceptible to a wide range
AI Governance & Risk services help provide of AI-related risks through all phases of the AI
strategic and tactical solutions to enable life cycle. For example, AI-based systems may
organizations to continue to build and use introduce or reinforce a risk of perpetuating
AI-powered systems while promoting inequity and historic bias, and enforceable
Trustworthy AI (FIGURE 2). regulations to protect the American public by
ensuring equitable, ethical and transparent
Deloitte recognizes that organizations and AI may be not only critical but inevitable. An
institutions are increasingly adopting AI indication of the future AI regulatory landscape
and automated systems for their potential can be seen in recent proposed and enacted
to revolutionize significant aspects of the state and local laws governing AI in specific
American public's daily lives from health care, use cases, such as AI-based performance
to banking, to shopping, to leisure downtime, evaluation and hiring decisions, loan
to many more. However, these innovations underwriting models, etc.5
Figure 2 | Deloitte's Trustworthy AI™ Framework
3
Deloitte’s Trustworthy AI™ Framework and the White House BluepDrienlot iftotre ’asn T rAuI sBtiwllo orft hRyig AhIt™s Framework and the White House Blueprint for an AI Bill of Rights
Figure 3 | AIBoR Principles and how they map to Deloitte’s Trustworthy AI Framework
AI Bill of Rights Description Deloitte Trustworthy AI
Principles Framework
Protect against inappropriate or irrelevant data Privacy
Safe and usage through testing, monitoring, and engaging Safe & Secure
effective systems stakeholders, communities, and domain experts. Robust & Reliable
Protect against discrimination by designing Fair & Impartial
Algorithmic
systems equitably and making system evaluations Transparent & Explainable
discrimination
understandable and readily available. Robust & Reliable
protections
Protect against privacy violations by limiting data Privacy
Data privacy collection and ensuring individuals maintain control
of their data and how it is used.
Provide clear and timely explanations for any Transparent & Explainable
Notice and
decisions or actions taken by an automated system. Privacy
explanation
Human Provide opportunities to opt out of automated Responsible & Accountable
alternatives, systems and access to persons who can quickly Privacy
consideration, remedy any problems encountered in the system. Robust & Reliable
and fallback
The White House AIBoR aligns well with The AIBoR is the latest governmental call to
Deloitte's Trustworthy AITM Framework and can action for organizations to proactively protect
help enable our clients to safely and effectively the American public as they embrace innovation
operationalize automated systems while through automation and AI. Though not directly
protecting individuals and communities and enforceable, the AIBoR can set the tone for
adhering to emerging regulations (FIGURE 3). the inevitable future legislation. We recognize
Deloitte's Trustworthy AI framework includes that navigating AI implementation challenges
a roadmap for implementing AI-powered can be complex, and Deloitte is committed to
systems and is aligned with AIBoR guidelines helping our clients navigate these challenges
through each phase of the AI development and agilely while establishing trust with stakeholders
maintenance lifecycle. and regulators. Deloitte, with our suite of AI
Governance & Risk assets and services, is
The world today is experiencing enormous dedicated to assisting our clients through the
human capacity for insight, decision- changing regulatory landscape in creating Trust,
making, efficiency, and innovation—all Equity and Transparency as they serve the
dramatically expanded by cognitive American public.
systems. We are also in a period in which trust
is paramount. The tools that unleash world-
changing capabilities should be trusted to act
in line with human expectations for ethics and
appropriateness. The full potential of AI just may
hinge on that confidence.
4
DDeellooiittttee’’ss TTrruussttwwoorrtthhyy AAII™™ FFrraammeewwoorrkk aanndd tthhee WWhhiittee HHoouussee BBlluueepprriinntt ffoorr aann AAII BBiillll ooff RRiigghhttss
LLLLeeeet ttt's '''sss s ssst ttta aaar rrrt ttt a aaa c ccco ooon nnnv vvve eeer rrrs sssa aaat ttti iiio ooon nnn
Let's start a conversation
Beena Ammanath Ed Bowen Oz Karan Joe Conti
BeBBBeeeeneeeannn aaaA AAAmmmmmmmmanaaannantaaahttthhh EEEdEddd B BBBoooowwwweeeennnn OOOOzzzz KKK Kaaaarrrraaaannnn JJJooJooeeee CC CCoooonnnntttiitii
Executive Director Advisory AI CoE Leader National Leader Leader
ExEEEexxxceeeuccctuuuivtttiiievvv eeeD DDDireiiirrrceeetcccotttrooorrr AAAAddddvvvivsiiissosooorrryryy yA AAAI IICI CCCooooEEEE L LLLeeeeaaaaddddeeeerrrr UUUUSSSS RRR Riiisssiskkkk &&& & FFF Fiiinnninaaaannnncccciiiaaaialll lAAA Addddvvvviiisssisoooorrrryyyy UUUUSSSS RRR Riiisssiskkkk aa aannnndddd FF FFiinniinnaaaannnnccciicaaiiaall lAAl AAddddvvviivssiisoosoorrryyry y
Global Deloitte AI Institute Managing Director Trustworthy AI™ Trustworthy AI™ Government
GGlGGolllboooabbblaaa Dlll DDDeleeeollliooottiiitttettt eeeA IAAA IIIIn IIIsnnntsssittttiiuitttuutuetttee e MMMMaaanannnaaagaggigniiinnnggg gD DDDiriiirrereeecccctttotooorrrr TTTTrrrruuuusssstttwtwwwoooorrrrttththhhyyyy AAA AIII Illl eeeleaaaaddddeeeerrrr GGGGoooovvvveeeerrrnnrnnmmmmeeeennnnttt taa aannnndddd PP PPuuuubbbblliilccliic cSS SSeeeecccttcoottoorr rr
[email protected] AI Center of Excellence Deloitte & Touche LLP and Public Sector
babbbmaaammmmmmmanaaaannntaaahttt@hhh@@@dedddleeeolllioootiititttettt.eeec...occcoomommm AAAAI IICI CCCeeenennntettteeerrr ro ooof fffE EEExxxcxcceceeellllelllleeennnnccceceee PPPPaaaarrrrtttntnnneeeerrrr TTTTrrruuruusssstttwwtwwoooorrrttrthhthhyyy yAA AAII IllI ee lleeaaaaddddeeeerrr r
[email protected] [email protected] Deloitte & Touche LLP
eeededddbbbboooowwwweeeennnn@@@@ddddeeeelollloooitiiitttttteteee.c...ccocooommmm DDDDeeeellloooloiiitttitttteteee aaa annnndddd TTT Toooouuuucccchhhheeee LLL LLLLLPPPP MMMMaaaannnnaaaaggggiiinnningggg DD DDiirriireereecccctttootoorrr r
[email protected]
ooookkkkaaaarrrraaaannnn@@@@ddddeeeellloooloiiitttitttteeete...ccc.coooommmm DDDDeeeellloooloiiitttitttteetee aa aannnndddd TT TooToouuuuccchhchheeee LL LLLLLPPLPP
jjjooojoccccoooonnnntttiiti@@@[email protected]
Rasmus Nielson John Fogarty
Senior Manager Senior Manager
Trustworthy AI™ Government Trustworthy AI™
and Public Sector Deloitte & Touche LLP
RaRRR saaa msssmmm usuuu sss
N
NNNieiiileeeslllesssneeennn JoJJJooohhhhnnnn
F
FFFooooggggaaaarrr trtt ytyyy
Deloitte & Touche LLP [email protected]
USUUU SSS
R
iRRR skiiisss kkk
a
naaa dnnn dddF iFFFniiiannnnaaacnnnicccaiiilaaa Alll AA dAdd vdivvvsiiiosssoo royrrr yyy AAAAuuuudddditiii ttta aa nannndddd
A
AAAsss ssss usuuurrr araaannnnccc eceee
[email protected]
GGoGG vooo evvv reee nrrr mnnnmmm eneeetnnn attt naaadnnn dddP uPPPbuuubb lbicllliii cccS SS eSee cetcccotttoo ro rrr SSS eSee nennnioiiiooor rrrM MMMaaaannnnaaaaggggeeeerrr r
SeSSS neee innn oiii rooo Mrrr MMM anaaa annngaaaegggreee rrr DDDDeee leollloooitiiittttettteee
a
aaannnndddd
T
TT oTooouuuuccc hchhheeee
L
LLLLLLLPPPP
DeDDD leee olll iooo ttiii etttttt eee
a
naaa dnnn dddT oTTTuooocuuuhcccehhh eeeL LLLLPLLLPPP jojjjooohhhhfofffoooggg agaa rarr trtt ytyy @y@@@ddddeeeelollloooitiiitt tttt eteee.c...cc ocooommmm
rarrr naaa innn eiii leee selllsss neee @nnn@@@dedddleeeollliooottiiitttettt.eeec...occcoo mommm
Endnotes
1 The AI Bill of Rights follows the Executive Order 13960: Promoting the Use
of Trustworthy Artificial Intelligence in the Federal Government (December
2020), Executive Order 13859: Maintaining American Leadership in Artificial
Intelligence (February 2019), Office of Management and Budget (OMB)
Memorandum M-21-06: Guidance for Regulation of Artificial Intelligence
Applications (November 2020), White House Office of Science and Technology
Policy (OSTP): American AI Initiative: Year One Annual Report (February 2020),
National AI Initiative Act of 2020 (Introduced March 2020), and the National
Institute of Standards and Technology (NIST) is expected to release the NIST AI
Risk Management Framework in January 2023. International initiatives include
the Organisation for Economic Co-operation and Development (OECD): 2019
Recommendations on Artificial Intelligence, and the European Union Artificial
Intelligence Act proposal (April 2021).
2 Blueprint for an AI Bill of Rights | OSTP | The White House
3 Ibid, White House.
4 Ibid, White House.
5 In 2022 alone, legislative bills or resolutions relating to AI were proposed in
seventeen different U.S. states and enacted in four. See Legislation Related to
Artificial Intelligence (ncsl.org) for further details.
5
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.
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 © 2022 Deloitte Development LLC. All rights reserved. |
350 | deloitte | us-navigating-the-impact-of-generative-ai-on-security.pdf | Navigating the impact of
generative AI on security
A CISO’s guide
Navigating the impact of generative AI on security | A CISO’s guide
Evolving the role of CISO with
the advent of Gen AI
Cyber has been ranked as the most important risk globally1 for the second year in succession.
Reports have shown that average costs from such incidents reached an all-time high in 2022 and
will continue to increase at a multi-fold pace in the coming years. The role of the chief information
security officer (CISO) will likely assume an even greater strategic significance within the organization’s
cybersecurity program. The insurance industry, in particular, is being targeted by a myriad of
cyberattacks as it possesses a great deal of personally identifiable information (PII) and protected health
information. Research has found that customer and employee PII is the costliest to have compromised
at $183 per record.2 As insurance companies migrate toward digital channels to create tighter customer
relationships and offer new products, a new wave of investment is directed toward advanced analytics
and generative artificial intelligence (Gen AI). Although these investments provide new strategic
capabilities, they also introduce new cyber risks and attack vectors to organizations. The challenges
are likely to become more complex as insurers prepare to leverage large language models (LLMs) for
Gen AI, which will require not only collecting and handling large amounts of sensitive data, but also
safely exposing it across multiple applications, interfaces, and cloud platforms.
As business and product teams find new ways to use AI, organizations must also ensure its safe, secure, and
ethical use. With growing digital disruption to conduct business, and diminishing trust, the need for a skilled
CISO has grown manyfold. The CISO is responsible for managing security across a distributed network to
ensure that the data remains secure; maintaining compliance with regulatory requirements; and educating
employees and informing executives about cybersecurity risks (82% of the largest insurance carriers have
been the focus of ransomware attacks from cybercriminals3). They not only have to contend with AI’s
immediate impact, but also prepare for how it will shape their responsibilities in the future. Highlighted
below are some of the ways Gen AI could add to the responsibilities of a CISO:
• Data security and privacy: Assess how models handle sensitive data and ensure they comply with
data protection laws and regulations.
• Access control: Implement robust access controls to ensure that only authorized individuals have
access to systems.
• Model integrity and security: Protect AI models from tampering and reverse engineering. This
includes ensuring that the models themselves are securely stored.
• Logging and monitoring: Establish logging and monitoring systems to detect and respond to
security incidents.
• Training and awareness: Provide training and raise awareness among employees and stakeholders.
Staying informed is the first step. A CISO should make time to be curious and continuously learn about
new developments and how they can affect insurers’ security posture. The CISO role will likely evolve from
being the “de facto” accountable person for treating cyber risks to being responsible for ensuring business
leaders have the capabilities and knowledge required to make informed, high-quality risk decisions.
2
Navigating the impact of generative AI on security | A CISO’s guide
Potential benefits of Gen AI
Gen AI has the potential to add contextual awareness and decision- • Act as application development assistant: Gen AI can also
making to enterprise workflows and can radically change act as a secure application development assistant. Many code
how we do business. The far-reaching impacts and potential value generation tools are embedding security features, and application
when deploying Gen AI are accelerating experimental, consumer, security tools are already leveraging LLM applications that can help
and (soon) enterprise use cases. According to Gartner, 68% of in common security use cases such as vulnerability detection, false-
executives believe that the benefits of Gen AI outweigh the costs, positive reduction, and mitigation suggestions.
compared with just 5% who feel the risks outweigh the benefits.4
• Mitigate social engineering attacks: Insurance companies also
The two big advantages of Gen AI today are its capacity to process
run the risk of losing money due to whaling attacks (a type of social
huge amounts of data at high speed and its ability to communicate
engineering attack where cybercriminals send executives a spoof
clearly. Still, questions remain about how insurance carriers can use
email to dupe them into authorizing massive cash transfers). LLMs
Gen AI to bring effectiveness, efficiency, and understanding, while
not only generate text, but they are also helpful in detecting and
managing the associated cybersecurity risks.
watermarking AI-generated text, which could become a common
function of email protection software. Identifying AI-generated text
Gen AI security use cases
in social engineering attacks can help to detect phishing emails and
• CISO executive reporting: Gen AI can help streamline and polymorphic code.
automate the process of drafting reports, including requirements
• Alleviate security talent and skill shortage: The sheer amount
for incident response, threat intelligence, risk assessments, audits,
and complexity of data and threats have become increasingly
and regulatory compliance. It can provide real-time insights into
difficult to tackle. The integration of Gen AI into several security
an organization’s risk profile, including its threat landscape, risk
operations tools enables cybersecurity teams to scale while
levels against critical vulnerabilities, current cybersecurity posture,
remaining lean and focused. This new interface can reduce the skill
compliance requirements, and cybersecurity performance metrics,
requirements for using the tool, shorten the learning curve, and
which can all be of aid to insurance CISOs.5
allow more users to benefit.
• Act as security assistant: A Gen AI security assistant can assist
security analysts in sifting through piles of log entries to evaluate Apart from the previously mentioned use cases, our first paper in
possible security threats by providing an assessment in seconds. the series delves into further instances across the insurance value
With a single prompt, Gen AI can scour logs and other data and chain where Gen AI is utilized, exploring its implications. You can find
report back on what may be an immediate threat and what isn’t. It more details here: Implications of Gen AI for insurance.
can also explain and add valuable context to the threat identifiers.
3
Navigating the impact of generative AI on security | A CISO’s guide
Key risks of Gen AI on CISOs’
watchlist/threat landscape
Despite numerous benefits, Gen AI was one of the top concerns In addition, a CISO should be consulted and informed about the
among security executives over the first few months of 2023.6 following risks:
CISOs may feel pressure to allow use of Gen AI broadly, but doing so
• Legal and regulatory risk: Legal and compliance risks arise from
indiscriminately could create unreasonable risk.
the fact that the legal and regulatory landscape surrounding Gen
AI is still nascent. Consequently, enterprises may not be aware of
Risks associated with Gen AI for a CISO at the enterprise level
all the legal requirements they need to comply with when using
generally stem from:
this technology. When Gen AI is used as part of a regulated use
• Data and privacy confidentiality: Enterprise use of Gen AI may case in consumer-facing communications, whether for direct
result in access and processing of sensitive information, intellectual consumer interactions or to produce consumer-facing materials
property, source code, trade secrets, and other data, through (such as consumer information notices), regulatory or private law
direct user input or an application programming interface (API). may include requirements and create liability.
Sending confidential and private data outside of the organization’s
– Mitigating measure: Comply with relevant data protection
servers could trigger legal and compliance exposure, as well as
regulations, and refrain from sharing customers’ sensitive
risks of information exposure. Such exposure can result from
information and the organization’s own sensitive data. Consider
contractual (e.g., with customers) or regulatory obligations
a platform that operates inside the secure network of the
(e.g., CCPA, GDPR, HIPAA, CPP Model law) that are in place and
organization. Obtain explicit user consent when collecting and
relevant to the organization.
using personal data for Gen AI purposes.
– Mitigating measure: Adhere to relevant regulations, such as GDPR
• Bias and discrimination: Training on biased data may lead to
or CCPA, to help safeguard sensitive information and maintain
illegal discrimination, potential damage to reputation, and possible
customer trust, and use secure Gen AI platforms.
legal repercussions for the enterprise as Gen AI may not be aware
• Data poisoning/prompt injections: Corrupt/polluted (poisoned) of potentially defamatory, discriminatory, or illegal content.
data leads to malicious or unintended outcomes and can affect
– Mitigating measure: Ensure that the training data used for Gen AI
the accuracy and reliability of the LLM. By using carefully designed
models is diverse, representative, and free from biases. Regular
inputs, attackers can manipulate LLMs, compelling them to carry
monitoring and auditing of the models’ outputs can help identify
out the attacker’s desires. This manipulation can occur by directly
and address any potential biases, promoting fairness and
altering the system prompt or manipulating external inputs, which
inclusivity in AI-generated content.
may result in serious issues like data exfiltration.
• Copyright and ownership/risk to intellectual property
– Mitigating measure: Implement robust access controls to ensure
(IP) rights: Gen AI models are trained on diverse data, which
that only authorized individuals can access systems.
might include copyrighted and proprietary material, raising
• Enterprise, SaaS, and third-party security: Due to Gen AI’s ownership and licensing concerns between the enterprise and
wide adoption and proliferation of integrations, there are other data sources used for training.
concerns that data would be shared with third parties at a much
– Mitigating measure: CISOs should seek a tool that operates end
higher frequency than earlier anticipated, posing a threat to
to end on their company’s network and does not require users
non-public enterprise data and third-party software. For example,
to send sensitive data to external servers or third parties. CISOs
third-party applications leveraging a Gen AI API, if compromised,
should also consider collaborating closely with legal teams to
could potentially provide access to email and the web browser, and
establish robust IP protection measures.
allow an attacker to take actions on behalf of a user.
– Mitigating measure: Establish comprehensive data governance
policies and procedures. This includes defining data ownership,
data classification, and data life cycle management. Clearly
define who has access to AI-generated data, how it is stored, and
for how long. Additionally, organizations must implement data
quality controls and establish mechanisms for data lineage and
audit trails. By adopting a robust data governance framework,
enterprises can help mitigate risks associated with Gen AI.
4
Navigating the impact of generative AI on security | A CISO’s guide
Gen AI security: Focus areas for CISOs
A CISO should ask questions and provide guidance to help leaders 3. External regulations8
create an organizational AI strategy. A comprehensive AI strategy
• Cross-industry: Comply with applicable external regulations.
provides guidelines for its usage and factors in legal, ethical, and
For example, recently enacted EU AI Act 20239 is a
operational considerations. If used responsibly and with proper
comprehensive guide to AI law with clearly defined
governance, Gen AI can provide businesses with many benefits across
transparency requirements and risk levels. The EU AI Act
automated processes and optimized solutions. A comprehensive AI
classifies insurance as a high-risk industry, which leads to more
strategy can help ensure privacy, security, and compliance. It should
stringent regulations and greater transparency industrywide,
consider the following key questions:
and firms have to ensure higher system compliance levels to
• Who is using the technology in the organization, and for what purpose? prevent any penalties.
• How can I protect enterprise information (data) when employees are • Insurance-specific: Focus on insurance regulations being
interacting with Gen AI? Do we have governance and contingency in established to safeguard insurance-specific risks. For example,
place (i.e., usage and controls)? Colorado’s draft AI regulation 202310 guides life insurers’ use
of external consumer data and information sources. It outlines
• How can I manage the security risks of the underlying technology?
requirements that ensure usage of algorithms and predictive
How do I balance the security trade-offs with the value the
models (i.e., AI models) do not result in unfairly discriminatory
technology offers?
insurance practices with respect to race.
Deloitte’s approach to responsible AI7—Trustworthy AI™—delivers • The National Association of Insurance Commissioners (NAIC) has
trust by design throughout the AI life cycle. It’s relevant to executives outlined a draft bulletin11 that provides guidelines for insurers
at every level: to use while utilizing AI systems (AIS) and ensuring compliance.
It emphasizes the importance of AIS programs, AI governance,
• The CEO and board set the strategy with special attention to public
and documentation.
policy developments and to corporate purpose and values.
4. Risk management
• Chief risk and compliance officers oversee control, including
governance, compliance, and risk management. • Fair/not biased: Define and measure fairness and test systems
against standards.
• Chief information and information security officers take the
lead on responsible practices, such as cybersecurity, privacy, • Transparent and explainable: Enable transparent model
and performance. decision-making.
• Data scientists and business domain specialists apply responsible • Responsible and accountable: Use policies to clearly establish
core practices as they develop use cases, formulate problems and accountability for AI outputs.
prompts, and validate and monitor outputs.
• Robust and reliable: Enable high-performing and
reliable systems.
How an insurer intends to use Gen AI and its impacts should be
thoroughly assessed across the following five key areas before • Privacy: Develop systems that preserve data privacy.
embarking on a Gen AI journey:
• Safe and secure: Design and test systems to prevent
1. Strategic considerations data harms.
• Impact of data and AI: Consider the moral implication of uses • Role-based access control: Implement robust authentication
of data and AI and codify them into your organization’s values. and authorization mechanisms to restrict access to sensitive
Gen AI systems and data.
• External policy and regulation: Understand public policy
and regulatory trends to align compliance processes. 5. Leading practices
2. Internal controls • Use-case identification: Identify the concrete problem you
are solving for and whether it needs an AI or machine
• AI governance: Enable oversight of systems across the three
learning solution.
lines of defense.
• Industry standards: Follow industry standards and
• Internal compliance: Comply with organization policies and
best practices.
industry standards.
• Continuous monitoring: Implement continuous monitoring
5
to identify drift and risks.
Navigating the impact of generative AI on security | A CISO’s guide
Path forward to balance
Gen AI’s challenges and
opportunities
Gen AI offers immense potential for innovation and creativity across
the insurance value chain and processes. In this journey, it is critical
for the office of the CISO to tackle the unique security challenges and
ethical issues and stay on top of the ever-changing regulations in this
domain. With business teams eager to leverage Gen AI at scale as
early as possible, here are a few change management guardrails to
consider in the short and medium term to mitigate risks to insurers’
security posture:
• Educate employees on the potential risks of Gen AI
usage through in-person training, online courses, and
awareness workshops.
• Communicate the importance of transparency and accountability
to prevent bias, hallucinations, and other risks.
• Identify and protect sensitive training data, enforce access
controls, and implement data loss prevention to prevent leaks.
• Establish clear usage policies, assessment frameworks, and
diligence models to evaluate the credibility of third-party AI
solutions, with the do’s and don’ts of using AI-generated content
within the organization.
• Form an approval board comprising stakeholders from different
business units to define internal policies based on a risk
assessment framework, and oversee adherence to the same
while implementing Gen AI use cases.
6
Navigating the impact of generative AI on security | A CISO’s guide
Contacts
Sandee Suhrada Vishvam Raval
Principal Senior consultant
Deloitte Consulting LLP Deloitte Consulting LLP
[email protected] [email protected]
Sunny Aziz Sharat Viswanathan
Principal Senior consultant
Deloitte & Touche LLP Deloitte Consulting LLP
[email protected] [email protected]
Rohan Shinde Meenakshi Rawat
Manager Senior consultant
Deloitte Consulting LLP Deloitte Consulting LLP
[email protected] [email protected]
Endnotes
1. Allianz, Allianz risk barometer 2024, January 2024.
2. IBM, Cost of a data breach report, 2023.
3. Eliot Partnership, “The evolving role of chief information security officers in the insurance industry,” April 17, 2023.
4. Gartner, “Gartner poll finds 45% of executives say ChatGPT has prompted an increase in AI investment,” press release,
May 3, 2023.
5. Michael Sentonas, “Introducing Charlotte AI, CrowdStrike’s generative AI security analyst: Ushering in the future of AI-powered
cybersecurity,” CrowdStrike, May 30, 2023.
6. Deloitte, Trustworthy AI™, accessed January 23, 2024.
7. Gartner, “Gartner survey shows generative AI has become an emerging risk for enterprises,” press release, August 8, 2023.
8. While we acknowledge that regulations pertaining to Gen AI are continuously evolving, we have outlined some of the newly
drafted regulations as of October 1, 2023, for reference.
9. European Parliament, “EU AI Act: First regulation on artificial intelligence,” updated December 19, 2023.
10. Colorado Department of Regulatory Agencies, Division of Insurance, SB21-169 – Protecting Consumers from Unfair
Discrimination in Insurance Practices, accessed January 23, 2024.
11. National Association of Insurance Commissioners (NAIC), “NAIC Model Bulletin: Use of Algorithms, Predictive Models, and
Artificial Intelligence Systems by Insurers,” exposure draft, July 17, 2023.
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.
8242799 |
351 | deloitte | us-ssoef-2024-breakout-innovation-ai-and-innovation-market-dynamics-and-the-opportunity-for-gbs.pdf | AI & Innovation:
Market Dynamics &
the Opportunity for
GBS
Prakul Sharma & Shri Chary, April 4th, 2024
Welcome and Introductions
Prakul Sharma Shri Chary
Managing Director In Strategy & Healthcare and Life Sciences
Analytics Practice Lead
Deloitte Consulting LLP. Deloitte Consulting LLP.
Copyright © 2024 Deloitte Development LLC. All rights reserved.
Agenda
The Evolution of GBS
1
Unleash Value through AI & Innovation
2
3 Case Studies and Lessons Learned
4 Q&A
Image generated with DALL•E 2 with prompt:
“open road cutting through a futuristic city, colorful, <Leading
CCooppyyrriigghhtt ©© 22002244 DDeellooiittttee DDeevveellooppmmeenntt LLLLCC.. AAllll rriigghhttss rreesseerrvveedd.. 33
Entertainment Conglomerate #1> movie”
The GBS market has evolved over the last two decades from execution roles to mature
value-added roles through innovation and service portfolio expansion
1990-2000 2000-2010 2010 - 2019 Beyond 2020
Business •Transactional processes •Expanded functional scope •Emergence of GS and CoEs •Strategic partnership with core business
Process Strategy •Basic support system •Focused global delivery models •Increased service coverage •Transformation & innovation partner
Workforce •Isolated workforce •Standardized workforce •Focus on global workforce strategy •Global skills hub for enterprise
Strategy management strategy •Integrated governance & management •Virtual and remote working spaces
Enterprise •Standalone tools and applications •Digital platforms, integrated tech, RPA •Automation, AI/ML and cognitive hub
•Basic computing technology
Technology •Increased focus on analytics •Enterprise tech. and ERP for services •Mature enterprise tech, consolidated systems
Business
Focus
Productivity and Operations Value
Based Improvement
Perform an existing process or service faster
without increasing cost and delivery time
Cost
Single Function Multi-Function Alignment End-to-end process
CCooppyyrriigghhtt ©© 22002244 DDeellooiittttee DDeevveellooppmmeenntt LLLLCC.. AAllll rriigghhttss rreesseerrvveedd.. 44
1
nuF
1
nuF
2
noitcnuF
3
nuF
3
nuF
Business Impact
Mature into service delivery partner delivering
business impact through innovation and
automation
Moving up the Value Chain
GS Lead
Increase breadth of services with
Global Process Owners
increased sharing (tools & people) E2E
Processes
Process 1
Multi-functional lead Process 2
Global Process Owners
Service Management
Comms / Training
Service Management IT facilities & support
Comms / Training
Region 1
IT Facilities & Support
Region 2
Region 3
tcudorP tnempoleveD remotsuC ecneirepxE
Process 3
Cost Arbitrage and Scalability
Perform an existing process or service at
lower cost, retaining same quality and
delivery time
1
noitcnuF
2
noitcnuF
sUB
GS
GBS organizations are increasingly assuming responsibility for strategic and high-value
business functions, transitioning towards a multi-functional scope creation…
Functions performed by Global Services organizations
1Traditional Functions
2Emerging Functions
Finance 91%
Advanced Capabilities
Finance1
Human Resources 62% IT1
FP&A HR1
Information Technology 57%
IT Strategy
Tax
Procurement 48% Procurement1
Enterprise Talent
Treasury Architecture Strategy
Tax 43% Design
Workforce Sourcing
Supply Chain/ Mfg Support 19% AF six se ed ts MAp ap in D tee nv a. n& c Planning Strategy
Legal 18% Increase YOY R2R Infra e C Bo em nep f. i t& s MaS nu ap gp el mier e nt ChaSa nl ne es l/ M Strk atg te. gy MS aa rkle es t i& n g2
Mgmt.
Sales and Marketing 13% Decrease YOY T&E L&D Contracting Pricing/Sales SP tr ro ad teu gc yt
Execution
Engineering/R&D* 10% Same IT Ops Employee Promotions / Legal2
O2C Separation Advertising Regulatory risk and
Sourcing compliance mgmt.
End Talent Sales Monitoring
P2P User Acq. Market & Reporting
Traditional functions delivered Services Analysis Customer inveL si tt ii gg aa tt ii oo nn sa un pd p ort Para legal services
Support
New/strategic functions delivered Talent Market Contract
*increase from 2019, as data was not available for 2021
Onboarding Research Legal research Mgmt. Customer Service C Su es rt vo im cee 1r
Admin. and analysis Complaints Strategy
Services Complaint Closure
~60% Global Business Services organizations perform more than 5 functions whichinclude both traditional (IT, Complaint Investigation Customer Service
finance, etc.)and strategic functions (e.g., sales & marketing, etc.). Allocation RC eo sm op lula tii on nt Analytics
CCooppyyrriigghhtt ©© 22002244 DDeellooiittttee DDeevveellooppmmeenntt LLLLCC.. AAllll rriigghhttss rreesseerrvveedd.. 55
Digital infused operations are harnessing new-age technologies like intelligent automation,
AI (& Gen AI), and process mining to drive process enhancements & efficiencies
Evolution of Process Improvement
Exponential
Shift
Shift
Process Improvement by
leveraging new age
technologies such as
RPA, Process Mining,
Business Process GenAI etc.
Management Suites
Traditional approach to
for entire process
improve individual
improvement life
processes (elimination,
cycle
reducing hand-offs)
CCooppyyrriigghhtt ©© 22002244 DDeellooiittttee DDeevveellooppmmeenntt LLLLCC.. AAllll rriigghhttss rreesseerrvveedd.. 66
eulaV
% of GS implementing different transformation levers
2023 Focus area in
Automation 59% Ranking next 1-3 years
1 1
Single instance ERP 47%
2 4
Case Management/ Workflow tool 45%
Centralized analytics reporting 3 5
28%
& Performance Dashboard
4 3
Global standard processes 28%
5 2
Self service 18%
6 6
Cloud 15%
7 10
Instilling a culture of innovation 11%
8 7
Data lake 8%
9 12
Agile 8%
10 11
Capability
Source: Deloitte Research and Analysis Source: Deloitte 2023 Shared Services Outsourcing Survey
Organizations are increasingly looking at their GBS centers to drive Process Improvement initiatives
More than 50% of GBS organizations are prioritizing process transformation and improvement as a key skill to be developed within their centers
For the next 3-5 years, there is an increased interest in global standard processes, centralized analytics, AI & Data and self-service
GBS organizations are pushing the boundaries to maximize value creation
GBS should expand its functional portfolio to include new service areas, while pioneering data-driven, digitally-infused
operations and facilitating enterprise-wide adoption of emerging technologies like GenAI
High Value Centre Digitally infused operations for established functions
(5% of total centers)
Cloud PaaS, SaaS
Data, AI & Analytics Data Lifecyccle, Advanced Analytics, AI/ML
Enterprise Technology ERP, CRM, BI
Integrated Workflows API Integration, Low-code/No-code
NLP, RPA, ML, AI, GenAI
Intelligent Automation & GenAI
Scaling emerging functions @speed
R&D Regulatory, Sustenance, NPD, Clinical
Commercial Sales, Marketing, Customer Services
Supply Chain Procurement, Vendor management
CCooppyyrriigghhtt ©© 22002244 DDeellooiittttee DDeevveellooppmmeenntt LLLLCC.. AAllll rriigghhttss rreesseerrvveedd.. 77
tcapmI
fo
leveL
Transactional Core Tech Digital- Operations infused
Capabilities
Degree of GBS involvement in the parent’s business functions/processes
Level of Involvement
tcapmI
fo
eergeD
fo
smret
ni tnerap
no
evah
SBG
eht
tcapmi
fo
eergeD
devas
tsoc
ro
dedda
eunever
Potential strategic levers for GBS value creation
High
Research & Analysis, Paralegal Services, Regulatory
Legal
Support
Low
Low Degree of Involvement
Life Sciences Example - Adoption of emerging technologies across the GS value chain
Illustrative Example
AI Analytics RPA IoT Blockchain
Potential impact: Low Medium High
Drug discovery Manufacturing, Pharmacovigilance /
Regulatory and
research, and Clinical trials supply chain and Marketing and sales complaints
medical affairs
pre-clinical trials distribution management
Lead optimization Patient recruitment and Resource, demand and Market analysis and Market access and HEOR ADR intake / complaint Labeling, artwork and CMC
scheduling supply planning competitive intelligence capture
Safety assessment Clinical data management Quality testing, analysis and Patient access and support Sales support and Case / complaint processing Product registration and
documentation programs salesforce effectiveness clinical trial applications
Data management Protocol development & Procurement and vendor Product support Market support and Reporting Regulatory writing, review
design management effectiveness and submission
Biostatistics % statistical Distribution and logistics Contract management Signal and risk management Regulatory information
programming support management
Site management and trial Medical affairs
monitoring
CCooppyyrriigghhtt ©© 22002244 DDeellooiittttee DDeevveellooppmmeenntt LLLLCC.. AAllll rriigghhttss rreesseerrvveedd.. 88
As organizations continue to invest in its digital core capability, GBS needs to integrate
Intelligent Automation, Generative AI and Data Management to accelerate value capture
TODAY WHAT IS NEXT WHY INTEGRATING GEN AI WITH IA IS IMPORTANT
Automation has delivered solid GenAI is poised to disrupt skilled workforce GenAI unlocks use-cases too complex for traditional IA, but it requires data, new
and sustainable value in dramatic ways capabilities, and a centralized approach
Many GBS organizations have
GenAI is poised to offer 25-35%
delivered solid and sustainable additional value1 when combined FAQ chatbots Personalized
&voiceagents omni-channel
value through automation, with traditional automation and AI experience
achieving 10%-40% savings tooling
Fraud
Code-Creation Detection
Marketing
ContentCreation
GenAI Capability support requires
Capabilities are self-sustaining
new ways to deal with trust in AI, Document
IA Capabilities and goals are
Instant
data management and change Comparison
driven largely by functional and Document answerson
mgmt. retooling
Extraction enterprisedata2
enterprise goals
Document
Extraction&
Generation3
Avoiding GenAI duplication (and
PolicyDocument
Function Demand is narrowing Text
lower returns) requires
Creation
Classification
Demand support from functions is centralized and coordinated
narrowing on GenAI, GBS support to bring E2E & Cross-
Capabilities, and external thought- Functional use-cases forward Use-cases represent common solution patterns for Generative AI
2Deloitte supported clients through Search, realizing 26% efficiency gains on RAG Retrieval and Semantic Search patterns
leadership 3Deloitte has built content generation GenAI applications for rganizations in Finance, Marketing, Regulatory, Commercial, Med-Tech
R&D and Pharma R&D
CCooppyyrriigghhtt ©© 22002244 DDeellooiittttee DDeevveellooppmmeenntt LLLLCC.. AAllll rriigghhttss rreesseerrvveedd.. 99
To deliver full benefits of AI/Gen AI, Shared Services must focus on foundational
capabilities
Shared Services should focus on developing
Intelligent Automation and Analytics capabilities are among
core capabilities to deliver broader digital
top capabilities delivered through Shared Service today…
transformation impact
Process Excellence and Continuous Improvement 71% 23%
Reporting & Analytics 55% 36%
1 Foundational AI & Data Capability
Intelligent Automation 53% 36%
End-to-End Process Ownership 53% 32%
Change Management and Training 50% 31%
Customer Experience & User Centric Design
2
Business Process Mining & Mapping 31% 47%
Vendor Management 48% 26%
Business Continuity Planning 51% 22%
3 End to end process ownership
Knowledge & Content Management 39% 33%
Program Mgmt. & Transition Mgmt. 49% 23%
GBS Footprint Strategy 34% 36%
Knowledge and Content Management
Customer Experience & User Centric Design 31% 37% 4
Environmental, Social & Governance (ESG) 23% 31%
M&A Integration SWAT team 17% 20%
5 Talent Upskilling and Retention
Have implemented Planning to implement
Source: Deloitte Shared Services & Outsourcing Survey 2023
CCooppyyrriigghhtt ©© 22002244 DDeellooiittttee DDeevveellooppmmeenntt LLLLCC.. AAllll rriigghhttss rreesseerrvveedd.. 1100
To deliver full benefits of AI & Gen AI, Shared Services must focus on building the right
foundations across 5 key pillars and enabling a “platform” for innovation
• How do we consistently ideate, prioritize and execute a high volume of concurrent use case decisions to align funding model & value capture?
• How will the breadth of stakeholders impacted be aligned to achieve a cohesive Gen AI vision and business case?
Strategy • What should be the strategy to convince the board of an investment in the Gen AI space?
• How do we identify & address new IP, legal, ethical and regulatory risk?
• What are the new data architecture, data governance and data science patterns? How do we minimize AI & data silos?
Governance • How to evaluate & adapt to a rapidly evolving tech partner landscape?
• What unstructured data is needed & how do we make it usable? What should be the approach – training on public or private data for creating
Gen AI models?
• How do we combine Gen AI, Traditional AI, and Analytics? What new data and data science tools do we need?
Technology
• Should we build, buy or adopt Gen AI solutions and models?
• What are the roles, responsibilities, skills and delivery models needed to be successful at delivering Gen AI at scale? How do we access talent?
Talent, Org & • How do we support a culture of “AI First” & ensure Gen AI adoption?
Culture
• How do we establish a consistent and repeatable approach to execute the backlog of Gen AI initiatives?
• How can we empower the business to deliver Gen AI solutions with minimal investment?
• What are the delivery best practices required to rapidly propagate Gen AI across the enterprise?
Delivery
CCooppyyrriigghhtt ©© 22002244 DDeellooiittttee DDeevveellooppmmeenntt LLLLCC.. AAllll rriigghhttss rreesseerrvveedd.. 1111
Case Studies
Copyright © 2024 Deloitte Development LLC. All rights reserved.
Case Studies related to IA and GenAI delivery
Case-Study STRATEGIC DRIVER APPROACH TO BUILDING MOMENTUM –Supported by Deloitte
CCooppyyrriigghhtt ©© 22002244 DDeellooiittttee DDeevveellooppmmeenntt LLLLCC.. AAllll rriigghhttss rreesseerrvveedd.. 1133
secneicS
efiL
egraL
ynapmoC
The CEO has tasked their leadership team to develop a multi-year • Define Gen AI strategy, operating model, and delivery approach to
Established GenAI COE to AI/Generative AI (GenAI) strategy to deliver: productionize use cases supported by Deloitte team of ~200 resources)
automate use cases across GBS,
• 50+ GenAIuse cases developed by end of year (spread across all • Establish ethical AI framework for Gen AI
Corp, Commercial, R&D &
functions)
Supply Chain functions • Establish 10+ value realization teams to identify & Qualify Gen AI use cases
• Target 25%+ uplift in productivity from each use case
• Develop an organizational specific platform leveraging OpenAIand
Business Impact: ~$3B
• 10K colleagues access to internal platform (version of chatGPT) reusable solution pattern framework
knaB
egraL
ta
tiduA
lanretnI
Internal Audit (IA) function wanted to establish capability and
• Defined the Gen AI strategy that covered the 5 critical dimensions –
platform to drive critical productivity and innovation use cases for the
Strategy, Governance, Technology, Talent and Delivery / Op Model
Establish the GenAIcapability function:
and platform for addressing • Develop and mobilize the technology architecture curated to integrate with
• Reduce labor costs associated with audit reviews and generating
Internal Audit (IA) use case existing technology landscape
required standard documentations
requirements
• Develop and publish a playbook to support Gen AI use case development
• Utilizing historical data (structured and unstructured) to augment
Business Impact: 30% that factors all leading practices around LLM models, Model Ops, Trusted AI
audit findings
etc.
improvement in productivity
• Establish a Gen AI CoEfor IA that worked closely with broader
• Develop, review, prioritize and mobilize critical use case development
organization Gen AI CoE
labolG
ta
ecnaniF
SBG
ynapmoC
GBS function looking to automate financial anomalies detection to:
• Implemented automated anomaly detection modelwithin the Azure Cloud
Automate Variance
• Reduce labor costs associated with identifying fraud and the to identify automated detection of anomalous transactions in the accounts
AnomaliesDetection &
costs of unidentified fraud receivable database
Reporting for GBS-Finance
• Generate automatic text reports from identified fraud cases • Utilized state-of-the-art advances in language modeling to generate text-
Business Impact: 90% accuracy
based reports of explanations for anomalous transactions, greatly reducing
• Increase team’s throughput to service a large number of reports
in anomaly detection
time and labor costs required to report financial anomalies
without additional headcount
ediW
esirpretnE
cificepS
esaC-esU
Key Learnings And Take Aways
Platform Centric Approach Product Mindset
Setting up of the core platform, including its components and Product-oriented mindset is essential, particularly for application
architecture development work.
Democratize AI Business Analysts’ and Tech Teams
Gen AI services should be built as a shared platform taking into Techno-functional Business Analysts should work closely with Product
consideration scalability and data security. Owners.
Prototype First
Cross-Functional Teams
Validate your hypothesis with tools like Jupiter notebooks or just GPT
directly to ensure your idea works. Building Use cases is much more than just building a model. It heavily
revolves around application development and experience.
Prompt Engineering/Validation takes time
Platform Education for Use case Teams
Buffer time for prompt engineering and testing real data.
Understanding the AI CoE's functionalities, its capabilities
And how it can be leveraged to achieve business goals.
Learn how to Consume AI
LLM Response Evaluation Framework
Team needs to understand how to build products with AI. i.e., AI fits into
A well-defined framework during the High-Level Design phase
customer journeys and not the other way around.
CCooppyyrriigghhtt ©© 22002244 DDeellooiittttee DDeevveellooppmmeenntt LLLLCC.. AAllll rriigghhttss rreesseerrvveedd.. 1144
Q&A
Copyright © 2024 Deloitte Development LLC. All rights reserved.
About Deloitte
Deloitte refers to one or more of Deloitte ToucheTohmatsu 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.
This presentation contains general information only and Deloitte is not, by means of this presentation, rendering accounting, business, financial, investment, legal, tax, or other professional advice or services.
This presentation 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 presentation.
Copyright © 2024Deloitte Development LLC. All rights reserved.
Designed by CoRe Creative Services. RITM1653349 |
353 | deloitte | us-trustworthy-ai-cdoi-grmf-summary-v4.pdf | Colorado Artificial Intelligence
(AI) Regulations:
Summary of Governance and Risk Management
Framework (GRMF) Requirements for Life
Insurance companies
Colorado Artificial Intelligence (AI) Regulations: Summary of Governance and Risk Management Framework (GRMF) Requirements for Life Insurance companies
Overview of Colorado Division of Insurance’s
(CDOI) AI Regulations
The AI regulatory landscape continues to mature as government Risk Management Framework, issued by the National Institute of
agencies refine and build upon previous guidance designed to Standards and Technology (NIST) in January 2023. Additionally, the
manage AI risk, to reasonably ensure equality and transparency, National Association of Insurance Commissioners (NAIC) continues
and to provide trust in automated systems. This is evidenced by to prioritize responsible data and AI use by insurers as reflected in its
recent publications including the White House’s Blueprint for an AI 2023 regulatory priorities.
Bill of Rights, issued in October 2022, and the Artificial Intelligence
Figure 1 — Timeline
2019
Jan 18, 2019 — NYFDS issues circular letter on “Use of AI”1
2020
Aug 14, 2020 — NAIC publishes “Principles on Artificial Intelligence”2
2022
Oct 4, 2022 — White House released its “Blueprint for an AI Bill of Rights”3
Jan 26, 2023 — National Institute of Standards and Technology releases its “AI Risk Management Framework”4
Feb 1, 2023 — CDOI releases its draft “Algorithm and Predictive Model Governance Regulation”5
Feb 13, 2023 — NAIC 2023 regulatory priorities6 include Data and Artificial Intelligence
May 16, 2023 — United States Senate sub-committee holds hearing on “Oversight of A.I.”7
2023
May 26, 2023 — CDOI releases a revised “Algorithm and Predictive Model Governance Regulation”
June 14, 2023 — European Parliament passes the “A.I. Act”8
Sep 8, 2023 — United States Senators Sen. Richard Blumenthal and Sen. Josh Hawley propose outline for
“Bipartisan Framework for U.S. AI Act”
Sep 21, 2023 — CDOI releases “Algorithm and Predictive Model Governance Regulation”9
Sept 23, 2023 — CDOI releases its draft “Algorithm and Predictive Model Quantitative Testing Regulation” 10
Nov 14, 2023 — CDOI “Algorithm and Predictive Model Governance Regulation” becomes effective
Continuing down the path of making the governance requirements more tangible, the CDOI released the AI regulation on September 21, 2023
(initial draft issued on February 1, 2023, and revised draft on May 26, 2023), and proposed a “draft” of its “Algorithm and Predictive Model
Quantitative Testing Regulation” on September 23, 2023.
Following active engagement with industry stakeholders, the CDOI intends to establish requirements for a life insurance company’s internal AI
Governance and Risk Management Framework (GRMF) as part of its adopted AI regulation and additional requirements related to testing and
reporting as part of its proposed quantitative testing regulation.
• Algorithm and Predictive Model Governance Regulation • DRAFT’ Algorithm and Predictive Model Quantitative Testing
(effective: November 14, 2023): The regulation is designed to Regulation (effective in 2024 (tentative)): The draft regulation builds
reasonably ensure that life insurers’ use of external consumer data upon the prior legislation from Colorado designed to minimize
and information sources (ECDIS), algorithms, and predictive models unfairly discriminatory insurance practices which may result from
(i.e., AI models) do not result in unfairly discriminatory insurance algorithms, AI, and the use of ECDIS. The regulation sets out testing
practices with respect to race. This is believed to be the first such and reporting requirements for insurers who leverage ECDIS
regulation on AI targeting insurers, and specifically life insurance. either directly or as an input to algorithms or models used in the
underwriting process.
Based on recent regulatory trends, there is a possibility that other states may follow suit and/or the scope of the regulation may be expanded
to other types of insurance (e.g., auto, property) and other industries utilizing ECDIS.
2
Colorado Artificial Intelligence (AI) Regulations: Summary of Governance and Risk Management Framework (GRMF) Requirements for Life Insurance companies
Requirements per CDOI’s adopted AI regulation include –
AI Governance and Risk Management Framework (GRMF)
Insurers dealing with life insurance are responsible to ensure that the following requirements are met,
including instances where third-party vendors are engaged:
01. AI governing principles (5A1) to ensure that use of ECDIS and AI 07. An up-to-date inventory of ECDIS and AI models (5A8) in use,
models using ECDIS are designed to prevent unfair discrimination including version control, detailed descriptions, purpose, problem
02. Board or board committee & senior management their use is intended to solve, potential risks, and safeguards
oversight (5A2&3) of GRMF, and for setting and monitoring the overall 08. Documented explanation of changes to the inventory (5A9) as
strategy on the use of ECDIS and AI models and provide direction for well as rationale for the change
AI governance 09. Documented description of testing to detect unfair
03. Documented cross-functional governance group (5A4) with discrimination (5A10) from use of ECDIS and AI models and steps
representatives from key functional areas taken to address disproportionate negative outcomes
04. Documented policies and procedures (5A5) iincluding assigned 10. Documented description of ongoing monitoring (5A11) of ECDIS
roles and responsibilities to ensure that ECDIS and AI models are and AI models including accounting for model drift ECDIS and AI
documented, tested, and validated models including accounting for model drift
05. Documented processes and protocols to address consumer 11. Documented description of process of selecting external
complaints and inquiries (5A6) about use of ECDIS and AI models resources (5A12) including third-party vendors
06. A rubric for assessing and prioritizing risks (5A7) associated 12. Annual reviews of the governance structure and risk
with the use of ECDIS and AI models, including insurance practices’ management framework (5A13) and updates to required
customer impact documentation to ensure its accuracy and relevance
Reporting (Once the final AI regulation goes into effect)
Insurers using ECDIS and AI models using Insurers that do not use ECDIS and AI Insurers that do not use ECDIS and AI
ECDIS should submit the following – models using ECDIS should submit the models using ECDIS as of the effective
following – date of this regulation, but subsequently
plan to use ECDIS and/or AI models using
ECDIS should submit the following –
• On June 1, 2024 (6A) A report summarizing • In one month (6C) from the effective date • Prior to the use (6D) of ECDIS or AI
progress toward complying with GRMF, areas of this regulation and on December 1 models and annually thereafter – A report
under development, challenges encountered, annually thereafter (6C) – An attestation summarizing compliance with GRMF
and expected completion date indicating they do not use ECDIS requirements, along with the title of
each individual responsible for ensuring
• On December 1, 2024, and annually
compliance (for each requirement). This
thereafter (6B) — A report summarizing
report must be signed by an officer attesting
compliance with GRMF requirements, along
to compliance with this regulation
with the title of each individual responsible for
ensuring compliance (for each requirement).
This report must be signed by an officer
attesting to compliance with this regulation
• Insurers using ECDIS may additionally
become subject to the newly proposed
Algorithm and Predictive Model Quantitative
Testing Regulation
3
Colorado Artificial Intelligence (AI) Regulations: Summary of Governance and Risk Management Framework (GRMF) Requirements for Life Insurance companies
Additionally, when using third-party vendors or external resources:
• The insurer is accountable for providing any documents required
by the CDOI
• Insurers may fulfill requests by letting third-party vendors provide
the required materials directly to the CDOI
• All components of the GRMF must be available upon request by
the CDOI on December 1, 2024, and annually thereafter
Key Terms
“External Consumer Data and Information Source” or “ECDIS” means, for the purposes of this regulation, a data or an information
source that is used by a life insurer to supplement or supplant traditional underwriting factors or other insurance practices or to establish
lifestyle indicators that are used in insurance practices. This term includes credit scores, social media habits, locations, purchasing habits,
home ownership, educational attainment, licensures, civil judgments, court records, occupation that does not have a direct relationship to
mortality, morbidity or longevity risk, consumer-generated Internet of Things data, biometric data, and any insurance risk scores derived by
the insurer or third-party from the above listed or similar data and/or information sources.
“Unfairly discriminate” and “Unfair discrimination”11 includes the use of one or more external consumer data and information
sources, as well as algorithms or predictive models using external consumer data and information sources, that have a correlation to race,
color, national or ethnic origin, religion, sex, sexual orientation, disability, gender identity, or gender expression, and that use results in a
disproportionately negative outcome for such classification or classifications, which negative outcome exceeds the reasonable correlation to
the underlying insurance practice, including losses and costs for underwriting.
“Bayesian Improved First Name Surname Geocoding” or “BIFSG” is a statistical modeling methodology developed by the RAND
Corporation aiming to help US-based organizations identify potential racial and ethnic incongruities amongst their datasets. The approach
utilizes both geocoded address information and administrative name data to predict a racial and ethnic probability for each data point.
BIFSG has been found to be 41% more accurate than similar modeling relying solely on surname data and 108% more accurate than utilizing
only geographic data.
4
Colorado Artificial Intelligence (AI) Regulations: Summary of Governance and Risk Management Framework (GRMF) Requirements for Life Insurance companies
Proposed Algorithm and Predictive Model Quantitative Testing
Regulation
The proposed draft regulation will apply to insurers who leverage ECDIS or models that utilize ECDIS to make or support underwriting
decisions. ECDIS data can include credit information; social media or purchasing habits; education, occupation, and licensure data; or
public records including home ownership or court records. Before delving into the testing and reporting requirements, insurers should first
understand their use of ECDIS, including by third-party input providers, to determine the applicability of these requirements.
This draft regulation prescribes two types of testing:
1. Logistic regression testing for application approval decisions
2. Linear regression testing for premium rates
Insurers must apply BIFSG estimated race and ethnicity to each of their application and premium datasets to perform testing of Hispanic,
Black and Asian Pacific Islander (API) applicants and insureds relative to White applicants and insureds. Dependent on outcomes of
the application approval and premium rate testing, variable testing of both datasets may be required to isolate and identify potentially
discriminate variables.
Testing database Application approvals Premium rates
Testing Methodology Logistic Regression Further Variable Testing Linear Regression Further Variable Testing
Logistic Regression Linear Regression
Event Scope All insurers using ECDIS Insurers with approval All insurers using ECDIS Insurers with a difference
(When is the testing in the underwriting rate differences of 5% or in the underwriting in premium rates of 5% or
required?) decisioning making greater as identified in the decisioning making greater per $1,000 of face
process Applications Approvals process amount as identified in
logistic regression testing the Premium Rates
Unfairly Discriminate A difference in approval Any difference in model A difference in premium Any difference in model
Threshold rates of 5% or greater coefficient for ECDIS rates of 5% or greater per coefficient for ECDIS
between racial or ethnic variables between the $1,000 of face amount variables between the
groups using BIFSG logistic regressions between racial or ethnic linear regressions
estimated race and groups using BIFSG
ethnicity variables estimated race and
ethnicity variables
Statistically significant differences in the results of the regression
testing of the race or ethnicity groups may require additional testing
and resolution, including insurers taking steps to remediate the
discriminatory outcomes identified. The proposed regulation calls on
insurers to establish accountability for the decisions driven from AI
and ensure fairness in their use of technology.
Reporting will be required for each ECDIS, algorithm, and predictive
model which utilizes ECDIS on an annual basis to the Colorado
Division of Insurance beginning April 2024, and annually thereafter.
As this regulation builds on prior CDOI AI regulations, additional
enforceable requirements may be provided in the near future. Life
insurance companies using ECDIS, whether directly or not, may be
subjected to specific remediation requirements or other regulatory
scrutiny resulting from the reporting requirements specified herein.
5
Colorado Artificial Intelligence (AI) Regulations: Summary of Governance and Risk Management Framework (GRMF) Requirements for Life Insurance companies
Deloitte’s Trustworthy AI™ Framework
At Deloitte, we recognized the importance of protections and for AI-fueled organizations to earn trust
in the AI-enabled assets and services provided to the public.
Deloitte’s Trustworthy AITM Framework and AI Governance systems may introduce or reinforce a risk of perpetuating
& Risk services help provide strategic and tactical solutions to inequity and historic bias, and enforceable regulations to protect
enable organizations to continue to build and use AI-powered the American public by reasonably ensuring equitable, ethical,
systems while promoting Trustworthy AI (Figure 2). and transparent AI may not only be critical but inevitable.
Deloitte recognizes that organizations and institutions are An indication of the future AI regulatory landscape can be
increasingly adopting AI and automated systems for their seen in recent proposed and enacted state and local laws
potential to revolutionize significant aspects of the American governing AI in specific use cases, such as AI-based performance
public’s daily lives from health care, to banking, to shopping, to evaluation and hiring decisions and loan underwriting models.12
leisure downtime, to many more. However, these innovations
can be particularly susceptible to a wide range of AI-related risks
through all phases of the AI life cycle. For example, AI-based
Figure 2 — Deloitte’s Trustworthy AI™ Framework (Framework)
6
Colorado Artificial Intelligence (AI) Regulations: Summary of Governance and Risk Management Framework (GRMF) Requirements for Life Insurance companies
Understanding GRMF requirements through Deloitte’s
Trustworthy AITM Framework
Deloitte’s Framework has been designed to assist insurance companies in operationalizing automated
systems safely and effectively, while protecting individuals and communities and adhering to emerging
regulations, such as those provided by CDOI (Figure 3).
Deloitte TWAITM Framework
7
krowemarF
tnemeganaM
ksiR
dna
ecnanrevoG
IA
eruceS
&
efaS
etavirP
&
tnerapsnarT elbanialpxE
laitrapmI
&
riaF
elbisnopseR elbatnuoccA
&
tsuboR elbaileR
Documented AI governance principals outlining unfair
discrimination protections
Board or board committee and senior management
oversight + clear roles and responsibilities
Documented policies, processes, and procedures for
creation, deployment, and use of ECDIS and AI models
AI models are safe and secure
Documented process and protocol to address consumer
inquiries and complaints
A rubric for assessing and prioritizing risks
Up to date inventory of ECDIS and AI models, with version
control and rationale for changes documented
Testing to detect unfair discrimination
Ongoing monitoring
Vender selection process
Annual review of GRMF and the accuracy and relevance of
associated documentation
gnitropeR
stnemeriuqer
Figure 3 — CDOI’s GRMF requirements and how they map to Deloitte’s
Trustworthy AITM Framework (Framework)
Report on progress
Report on compliance
Regular reporting
Colorado Artificial Intelligence (AI) Regulations: Summary of Governance and Risk Management Framework (GRMF) Requirements for Life Insurance companies
The world is experiencing significant human capacity for insight, may set the tone for industry leading practices and other regulators
decision making, efficiency, and innovation—all dramatically looking to implement similar requirements. Additionally, CDOI has
expanded by cognitive systems—during a time in which trust is also suggested similar rules may be applied to other insurance
paramount. The tools that assist in unleashing world-changing lines or other AI or algorithmic uses. We recognize that navigating
capabilities should be trusted to act in line with human expectations AI implementation challenges can be complex, and Deloitte Risk
for ethics and appropriateness. The full potential of AI may hinge on & Financial Advisory is committed to helping our clients navigate
that confidence. these challenges agilely while establishing trust with stakeholders
and regulators. Deloitte Risk & Financial Advisory, with our suite of
The CDOI’s AI regulation will provide specific enforceable AI governance and risk assets and services, is dedicated to assisting
requirements for life insurance companies using ECDIS and AI our clients through the changing regulatory landscape and helping
models using ECDIS in Colorado. CDOI has leveraged high-level clients create trust, equity, and transparency as they serve the
principles, found in places like the federal regulatory guidelines, and American people.
attempted to turn them into reporting requirements for governance,
documentation, and reporting. While it is expected to only apply to
life insurance companies doing business in Colorado, the regulation
8
Colorado Artificial Intelligence (AI) Regulations: Summary of Governance and Risk Management Framework (GRMF) Requirements for Life Insurance companies
Contact us:
Oz Karan Satish Iyengar Richard Godfrey David Sherwood
Risk & Financial Advisory, Risk & Financial Advisory, Insurance Sector Leader, Risk & Insurance Regulatory Leader,
Trustworthy AI Leader, Partner Trustworthy AI — FSI Leader, Financial Advisory, Principal Risk & Financial Advisory,
Deloitte & Touche LLP Managing Director Deloitte & Touche LLP Managing Director
[email protected] Deloitte & Touche LLP [email protected] Deloitte & Touche LLP
[email protected] [email protected]
Contributor:
Ajay Ravikumar Jordan Baker Tim Cercelle Jordan Kuperschmid
Risk & Financial Advisory Risk & Financial Advisory Managing Director Principal
Trustworthy AI – FSI, Senior Manager Trustworthy AI Senior Manager Deloitte & Touche LLP Deloitte & Touche LLP
Deloitte AERS India Pvt. Ltd. Deloitte & Touche LLP [email protected] [email protected]
[email protected] jorbaker@ deloitte.com
Endnotes
1 NYFDS issues circular letter on “Use of AI”
2 NAIC publishes “Principles on Artificial Intelligence”
3 White House released its “Blueprint for an AI Bill of Rights”
4 National Institute of Standards and Technology releases its “AI Risk Management Framework”
5 CDOI releases its draft “Algorithm and Predictive Model Governance Regulation”
6 NAIC 2023 regulatory priorities include Data and Artificial Intelligence
7 United States Senate sub-committee holds hearing on “Oversight of A.I.”
8 European Parliament passes the “A.I. Act”
9 CDOI releases “Algorithm and Predictive Model Governance Regulation”
10 CDOI releases its draft “Algorithm and Predictive Model Quantitative Testing Regulation”
11 “Unfairly discriminate” and “Unfair discrimination”
12 National Conference of State Legislatures Report on Legislation Related to AI
9
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 advisor.
Deloitte shall not be responsible for any loss sustained by any person who relies on this document.
As used in this document, “Deloitte Risk & Financial Advisory” means Deloitte & Touche LLP, which
provides audit, assurance, and risk and financial advisory services; Deloitte Financial Advisory
Services LLP, which provides forensic, dispute, and other consulting services; and its affiliate,
Deloitte Transactions and Business Analytics LLP, which provides a wide range of advisory and
analytics services. These entities are separate subsidiaries 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 © 2023 Deloitte Development LLC. All rights reserved.
Design by Core Creative Services | RITM1568262 |
354 | deloitte | us-tte-annual-report-2023.pdf | State of Ethics and
Trust in Technology
Annual report
Second edition
Table of contents
Executive summary How companies Immediate actions
01 05 07
define and meet companies can take
4
ethical principles for to promote ethics
emerging technologies in technology
Introduction
02
23 32
6
Trustworthy and The importance
Emerging
06 08
03 Ethical principles of of collaboration in
technologies under
emerging technologies establishing ethical
consideration
principles
7 27
38
The need for
Promoting trust and
04
ethics in emerging 09
ethics in technology:
technologies
the way forward
10
45
2
Executive summary
01
Since the publication of our first State of corporate purpose, ethical principles, and This report presents the results of a study
Ethics and Trust in Technology Report in societal values, organizations can embrace conducted to gain perspective on how ethical
0022
2022, the rapid development of Generative internal and external collaboration in co-creating principles inform the development of emerging
Artificial Intelligence (AI) tools spurred ethical standards for technology. Leaders who technologies. Our research began by reviewing
the need for even greater attention to the define trustworthy and ethical principles for key takeaways from last year’s report and 0033
ethical dimensions of emerging technologies. the use of emerging technologies can create seeing how market shifts reinforced or altered
While Generative AI tools offer significant social, reputational, and financial value for their those findings. We interviewed 26 specialists
04
possibilities, it comes with the potential to organizations, cementing consumer trust and across industries to garner insights and test
inflict great human harm and reputational attracting future generations of talent. hypotheses. With these hypotheses, we launched
or financial damage to the organizations that a 64-question survey to over 1,700 business and 05
produce and use them. And while Generative This second edition Technology Trust Ethics technical professionals. The survey addressed
AI is the focus of today, similar potentials (TTE) Report explores trends in how industry the impact of Generative AI on organizations, the
06
for good and harm exist across all emerging leaders perceive and address ethical issues value placed on ethical principles for emerging
technologies. related to emerging technologies, with a focus technologies, and mechanisms to implement
on Generative AI. We articulate the need for ethical behavior throughout organizations.
07
Organizations can proactively promote organizations to consider formulating and
trustworthy and ethical principles at every promoting ethical principles in the context of We hope this report supports your journey
08
level and focus on incorporating ethics as part new and impending regulations. We identify to increasing trust and confidence in your
of the development and implementation of actions leaders and organizations may take to operations and contributing towards a more
technological products and services. To align help embed ethical principles into technological equitable society.
09
quality product development processes with development and deployment.
3
Key takeaways to consider
01
1 2 3
0022
Organizations should develop Organizations should design Organizations should proactively
trustworthy and ethical principles and apply tailored ethical embed trustworthy and ethical
0033
for emerging technologies, as principles specific to each of principles as part of a team-based
reinforced by the rapid impacts their technological products. collaborative development process
of Generative AI in the past year. for emerging technologies. 04
4 5
05
Organizations should actively Organizations who design and
seek to collaborate with other earnestly adopt trustworthy and 06
businesses, government ethical principles should benefit
agencies, and industry leaders from mitigating reputational
07
to create uniform, ethically and financial damage and
robust regulations for emerging reinforcing trust in employees
technologies. and stakeholders.
08
09
4
Introduction
01
When we wrote the 2022 report on ethics 2023 has seen rapid, significant progress in the
and trust in emerging technologies, the field of Generative AI. With the advancement
0022
state of artificial intelligence (AI) had of chatbots and other Generative AI tools,
been relatively stable over the past five suddenly the once-familiar AI ground has shifted
years. As such, the report focused on tremendously, opening new arenas of ethical 03
emerging technologies broadly, including inquiry. From questions of how Generative AI
autonomous vehicles, blockchain, and may exacerbate the digital divide to the potential
04
quantum computing. for plagiarism, distribution of harmful content
and misinformation, and worker displacement,
organizations find themselves wrestling 05
with new ethical issues posed by wide-scale
adoption of this once-again new technology.
06
As such, our report provides insight into how
organizations are approaching the ethics
of Generative AI along with other emerging
07
technologies. Furthermore, our report highlights
why organizations should embed trust into
08
every aspect of internal operations, strategy,
and decision-making, and the benefits from
meaningfully building trust and ethics.
09
55
Emerging technologies under consideration
01
“Emerging tech” refers to digitally enabled tools representing new and significant developments within a particular field.1
These technologies can be grouped into the following categories:
0022
0033
04
Cognitive Digital Reality Ambient Autonomous Quantum Distributed Robotics
Technologies Experiences Vehicles Computing Ledger
Technology (DLT)
05
including general including augmented including AI/ML including automotive, including quantum including blockchain, including robotic
and Generative AI, reality (AR), virtual assisted wearables, aerial, and maritime simulation, quantum crypto, non-fungible process automation
machine learning reality (VR), mixed voice assistants, linear algebra for token (NFT), and more 06
(ML), neural networks, reality (MR), voice and in-environment AI/ML, quantum
bots, natural language interfaces, speech devices optimization and
processing, neural recognition, ambient search, and quantum
07
nets, and more computing, 360° factorization
video, immersive
technologies,
08
computer vision,
and more
While these technologies are already in use and rapidly evolving, Generative AI received the 09
most attention this year for its groundbreaking potential to change the very nature of work.
6
Emerging technologies under consideration
Perceptions of emerging Industry leaders shared current and potential This year, quantum computing entered the top 01
benefits and misuses of these technologies (see three for the most potential for serious ethical
technologies
Figure 1 for a subset of responses). From last year risk; however, as a leader on AI said during one
0022
the biggest shift in perception of both positive of our interviews, “Quantum is at the later part of
As the basis for this report, we surveyed business
and negative outcomes occurred within cognitive early stage, still far out from real maturity—still in
leaders and developers of emerging technology
technologies. the stage of just initially testing it in production.”2
0033
about intended uses and broader implications
Potential issues could be forthcoming but are yet
of these technologies. Through the survey and
to be realized.
interviews with specialists, we gained insight into 04
these use cases.
Figure 1: Emerging technologies with the most potential for social good and ethical risk
05
Survey respondents felt the emerging According to survey respondents, emerging According to survey respondents, emerging
technologies with the most potential for social technologies with the most potential for technologies with the most potential for serious
social good: ethical risk:
good are cognitive technologies (39%)—which
06
includes Generative AI—digital reality (12%),
Down Down
Up 2 points Up Up 5 points Up
and ambient experiences (12%). Conversely,
6 points 1 point 16 points 1 point
07
respondents identified technologies with the
39 12 12 57 11 9
most potential for serious ethical risk as cognitive % % % % % %
technologies (57%), digital reality (11%), and
08
quantum computing (9%).
09
Cognitive Digital Ambient Cognitive Digital Quantum
technologies reality experiences technologies reality computing
7
Source: 2023 Deloitte Technology Trust Ethics Survey
Emerging technologies under consideration
01
Potential benefits and misuses
of emerging technologies 0022
A longstanding school of thought in critical
0033
technology studies and computer ethics known
as Values in Design (ViD) asserts technologies
are built using assumptions that express value 04
commitments.3 Thus, technologies encode values
into societies that adopt them.4 For example,
05
common surveillance technologies (e.g., doorbell
cameras) embed the value of the right to see
anything happening in or around one’s property, 06
but they also can infringe on users’ privacy. Because
these values are often unconscious, emerging
07
technologies have a range of potential impacts,
both beneficial and harmful.
08
09
8
01
The need for
0022
ethics in emerging
0033
technologies
04
As technologies grow more powerful, so does the
05
potential for harm. And with any technology-related
ethical misstep made by organizations, trust that
took years to build can erode in an instant. Given the
06
importance reputation can have on long-term success,
organizations should prioritize ethical principles.
07
08
09
9
The need for ethics in emerging technologies
01
Ethical missteps cause multiple types of damage
0022
Ignoring or downplaying ethical issues associated with emerging technologies comes at a cost. Ordered by perceived severity of damage to the
organization by respondents, these include the following:
0033
38 27 17 9 9
% % % % %
Reputational Human Regulatory Financial Employee
04
damage damage penalties damage dissatisfaction
Ethical missteps can leave customers Implementing emerging technologies Legal experts are scrambling to keep Reputational damage leading to loss Unethical behavior or lack of visible
distrusting the organization and before they are vetted, trained, up with emerging technologies, and of sales and costly lawsuits resulting attention to ethics can decrease a 05
tarnishing an organization’s hard- and tested to understand risks lawsuits filed allege harms such as from unethical behaviors can company’s ability to attract and keep
won positive brand sentiment. can cause severe and lasting harm copyright infringement,8 privacy negatively impact an organization’s talent. One study found employees
Reputational damage especially to individuals and communities. violations,9 harm to children and bottom line.11 While the adoption of of companies involved in ethical
06
affects younger generations, who Potential harms include violations teens,10 and more. Adopting a clear ethical principles cannot guarantee breaches lost an average of 50%
tend to be values-driven; as a result, of privacy, technology-assisted set of ethical principles in addition financial solvency, research suggests in cumulative earnings over the
Organizations should be clear on discrimination, challenges to human to a thoughtful implementation plan companies that implement ethics subsequent decade compared to
permissible uses of technology.5 agency, and job displacement. The may help companies proactively as part of their business philosophy workers in other companies.13 An
07
Organizations that commit to World Health Organization warned forestall these issues before are more profitable than those that interviewee for this report suggested
ethical and responsible practices for too-speedy adoption of Generative AI regulators take action. do not.12 productivity may decline as people
emerging technologies can build trust could potentially cause a plethora of become less motivated to work in
with stakeholders and differentiate harms, including misdiagnoses and unethical environments.14 Having a
08
themselves in the market. treatment biases.6 Thus, companies compromised employee base affects
could commit to ethical principles of many aspects of a company.
emerging technologies that articulate
not just guidelines, but specific goals, 09
metrics, and an understanding of
what a failure of these principles
might look like.7
10
The need for ethics in emerging technologies
01
The damages from ethical
missteps can add up. One study 0022
estimates workplace misconduct
cost US businesses $20 billion 0033
in 2021.15 Conversely, companies
that proactively establish and 04
uphold ethical principles in
05
technology use cases help foster
trust amongst stakeholders,
06
solidify their brand reputation,
and increase profitability.
07
08
09
11
> Hey, Chatbot.
01
0022
> What’s going on
0033
with Generative AI? 04
05
06
Generative AI is a noteworthy example of how we might expect to see emerging 07
technologies affect markets moving forward. With potential impacts and risks
in areas like information services, manufacturing, sustainability, science, and 08
healthcare, AI highlights the need for ethical standards.16
09
12
>
Uses of AI in industry
01
Generative AI is predicted to “change the nature of how we interact with all software”17 and to add $4.4 trillion in value annually to the global economy.18
AI’s power and ethical concerns alike come from its ability to automate tasks previously done by humans. Though Generative AI entered the mainstream
0022
less than a year ago, it has shown its influence in areas like generative design, ad and marketing campaigns, customer assistance, personalizing customer
experiences, and more.
0033
Despite the relative nascence of Generative AI in the marketplace, most companies surveyed are already testing or using
Generative AI tools:
04
74 65 31
% % %
05
have begun
06
have begun
have begun testing using Generative
using Generative
Generative AI AI technologies
AI technologies
technologies for external
internally
consumption 07
08
Given Generative AI’s newness, most organizations have 09
work to do in adapting responsibly to this tool.
13
>
Concerns with AI use
01
For all the buzz about Generative AI’s potential for productivity and profit, respondents expressed trepidation about its potential downsides. These
concerns are ranked below in descending order by the percentage of survey respondents who selected the issue as one of their top three concerns:
0022
22 14 12 12
% % % %
0033
Data privacy Transparency Data poisoning Intellectual property
and copyright
04
Data privacy is a big concern associated with Generative AI tools. Generative AI is trained on Generative AI tools depend Some Generative AI tools
Developers and scientists acknowledge machine learning-based millions of data points and on robust data training sets are trained on data that can
language models (LLMs) can inadvertently leak information from hundreds of features, leading for their effectiveness. These include copyrighted works,
05
the data used to train them, potentially exposing sensitive data to technically complex systems sets can be deliberately putting AI-generated work in
including personally identifiable information (PII). If LLMs are that often obscure how “poisoned” or “polluted” by murky legal territory.23, 24 To
designed without addressing data protection, it risks incidents like information is produced.21 To hackers and other bad actors, minimize legal risk, companies
06
training data extraction attacks, using queries to extract specific dispel “black box” concerns, leading to the propagation using works derived from AI
pieces of data.19 Companies using Generative AI tools are providing companies should focus on of inaccurate results.22 should attend to issues of
workarounds to protect data privacy. For instance, one person creating transparent and Companies should focus on ownership in IP and copyright.
interviewed for this report remarked his company engaged a third- explainable AI solutions. safe and secure information 07
party company to provide software that takes a sample of data and sets, assuring customers
creates a dataset with no connectivity to original source data.20 of the data’s provenance in
trusted sources. 08
09
14
>
Concerns with AI use
01
12 9 8 7 3
% % % % %
Data provenance Data “hallucinations” Authentic experiences Job displacement Static data 0022
Knowing where your data comes Generative AI tools are known The sophistication of A report released in June 2023 “Legacy analytic solutions”
from and what it contains is key. to make up or “hallucinate” Generative AI tools makes suggested AI contributed to (i.e.siloed datasets) produces 0033
Without this understanding, AI data, including fabricating it difficult to distinguish nearly 4,000 job losses in the inaccurate AI results.30 Solutions
tools can extrapolate biases, information like names and between human-generated previous month.29 Companies should design, test and
leading to adverse customer dates, medical explanations, and computer-generated should consider using AI to release with current data that
04
affects, skewed outcomes, and plots of books, citations, text, images, and videos.27 offset tasks to make human provides validated answers. For
lower accuracy.25 In addition, and even historical events.26 Companies should consider work more productive and systems that do not use recent
several types of bias errors Companies should ensure adopting ethical frameworks implement job upskilling information, disclosures should
05
can be introduced from the their AI systems are sufficiently like the US government’s “AI Bill where appropriate. be pronounced and frequent.
human side, including sample/ robust in their training and of Rights,” which reserves the
selection bias, exclusion reliable in their outputs to right of users to know when
bias, measurement bias, and minimize the potential for they are interacting with a 06
association bias. hallucinations. human versus a bot.28
07
08
Companies should adopt ethical frameworks like the US government’s “AI Bill of Rights,” 09
which reserves the right of users to know when they are interacting with a human versus a bot.28
15
>
Concerns with AI use
01
Figure 2. Top pressing ethical concerns with using Generative AI
(Percentage)
0022
25
22%
0033
20
04
15
14%
05
12% 12% 12%
10
9%
06
8%
7%
5
07
3%
0 08
Data Privacy Transparency Data Poisoning IP Ownership Data Provenance Hallucinations Authentic Job Static Data
Experiences Displacement
09
Source: 2023 Deloitte Technology Trust Ethics Survey
16
>
How organizations can safely incorporate AI
01
To harness the transformative power of AI Exploration
effectively and ethically, companies should To start, companies can familiarize
0022
consider assessing and rethinking their themselves with the technology and
development strategy. Below is a multi-step development approaches. Exploring use cases
framework to assist companies in integrating can foster innovation and lay the groundwork 0033
emerging technologies. for creating road maps to incorporate
Generative AI. Exploration could consist of
04
workshops in which teams of product owners,
AI/ML practitioners, and business leaders
brainstorm, then rank by return-on- 05
investment areas in which AI/ML might create
value to the company. “Value” here consists of
06
both profits as well as brand value like reliability,
company trust, and social goodwill. Companies
can develop qualitative and quantitative cost/
07
benefit analyses, weighing the impacts of
incorporating AI against the risks.
08
Exploring use cases can foster innovation and lay the 09
groundwork for creating road maps to incorporate Generative AI.
17
>
How organizations can safely incorporate AI
01
Foundational Whether to buy or build platforms depends on Figure 3. Approaches to building
foundational Generative AI capabilities
Incorporating Generative AI into a the type of business. For instance, higher tech
(Percentage) 0022
business could require building or companies tend to build their own AI platforms.
identifying an internal data foundation for an As one person interviewed points out, companies
6
LLM. Companies can thus decide whether to who build their own platforms can more readily 5 0033
collaborate with existing platforms or hire talent write ethical standards into their specs.31 By 8
30
to build in-house. contrast, life sciences companies often do not
04
build in-house and rely on vendors for data
Among survey respondents, 30% indicated their solutions. Buying a platform or collaborating with
companies opted to use existing capabilities third parties requires extending the company’s 26 05
through major AI platforms, 24% of respondents’ trust and reliability, so companies should carefully
24
companies used private capabilities through review potential collaborators and their products
06
major platform developers, 26% created custom for ethical principles.
private tools in collaboration with major platform
Utilizing public-based capabilities through
developers, 8% built a complete platform major platform developers
07
Utilizing private-instance capabilities through
in-house, 6% were unsure, and 5% opted not to
major platform developers
use Generative AI at all.
Partnering with major platform developers
to develop custom, private instance 08
Building completely in-house
We are not planning to use Generative AI
09
Unsure
Source: 2023 Deloitte Technology Trust Ethics Survey
18
>
How organizations can safely incorporate AI
01
Governance Trainings/Education Pilots
Formulating and abiding by robust Training could encompass courses As part of introducing AI, companies
0022
standards and protocols can help in the ethical principles of the should consider proof of concepts
forestall potential risks and harms of Generative company governing AI, and technical training and pilot programs. By doing so, an interviewee
AI. Before developing a specific set of standards that focuses on the diverse LLMs and how use says, engineers and product leaders can initiate 0033
and policies governing AI, the company should cases should be enabled. With both types of experiments with different use cases and run a
first consider defining ethical principles. 56% of training, employees can feel more invested variety of product tests.34 Pilots that fail to meet
04
respondents say their company does not have or and empowered. requirements or are deemed too high-risk can be
are unsure if they have ethical principles guiding cancelled at this stage.
the use of Generative AI. 05
Additionally, another specialist suggests, pilots
For governance, companies should consider and proofs of concepts can provide time to
06
AI Centers of Excellence (CoE), comprised of discuss the ethical, legal, regulatory, risk, and
internal experts that develop, scale, and oversee operational aspects of Generative AI.35
AI strategy throughout the enterprise.32 One
07
person interviewed suggests by centralizing
the development of AI and creating an internal
08
CoE, companies may have better control over
how adoption happens.33 The CoE could lead
implementation of AI, creating standards,
09
responsibility frameworks, and guidelines, and
developing trainings and education.
19
>
How organizations can safely incorporate AI
01
Implementation Companies should consider accountability for Audit
A successful implementation strategy product implementation, establishing product Companies will likely need to scale
0022
should include roadmapping, ownership and reporting structures for failures and adjust their policies to account
assignment of accountability, and built-in plans and other issues. for the potentially harmful impacts of AI tools,
for transparency. according to one interviewee.36 Another 0033
Companies should have a transparency recommends establishing a feedback system
Product leaders, in concert with the CoE, can strategy, defining what happens with user data, to make sure products aren’t manipulated for
04
create launch plans and product roadmaps to how the model arrives at a solution, and the bad intentions.37
help bring the newly enhanced products to confidence level of the model (i.e., how likely it is
market. Once released to the public, the company to “hallucinate”). 05
should have a team of data scientists and AI/ML
experts ready to boost the product’s capabilities
06
and address issues..
07
08
Companies should assign accountability for product implementation, establishing product 09
ownership and reporting structures for failures and other issues.
20
>
How do Generative AI tools impact
human workers?
01
Job displacement was ranked low on the list As one interviewee asserts, AI is not coming Figure 4. How respondents’ companies
handle employees displaced by Generative AI
of concerns by survey respondents compared for our jobs, but rather our tasks.38 Embracing
(Percentage) 0022
to issues like data privacy and transparency. AI and automating routine tasks can allow
But still, what happens to workers when this workers to pursue higher-level activities.
technology is deployed? Another person interviewed points out 0033
27
integrating Generative AI creates new jobs
Among respondents, 49% said workers at (for example, “Prompt Engineer”).39
04
their organization displaced by AI moved to
49
different roles and retrained and upskilled. For workers displaced by tech, companies can
13% moved to different roles but not retrained invest in upskilling and retraining; some 11 05
or upskilled. 11% are terminated. And 27% of organizations have programs to pay for
respondents have not had workers displaced employee retraining.40 Thus, companies might 13
06
by AI at their organization. frame the adoption of Generative AI not as
tech replacement but an opportunity for
They are moved to different roles and are
change management. re-trained/upskilled
07
They are moved to different roles but are not
re-trained/upskilled
They are terminated
08
N/A: This does not happen within my
organization
09
Source: 2023 Deloitte Technology Trust Ethics Survey
21
How companies define and meet ethical
principles for emerging technologies
01
0022
As technology is moving faster than Companies can establish ethical principles governing emerging technologies through four approaches:
regulation, the onus for creating ethically
sound technologies is increasingly placed on 1 2 0033
companies that design and develop those
technologies.
04
By meeting compliance and regulatory Following company culture (up 7
standards (up 2 percentage points in percentage points from last year)
individual survey responses from last year)
05
This approach to instilling ethical principles relies
The focus of this approach is operating within on standards set by company culture, defined
06
legal, published guidelines minimally impacted as the sum of formal and informal systems,
by company values. behaviors, and values, all of which help create an
experience for employees and customers.
07
08
09
22
How companies define and meet ethical principles for emerging technologies
3 4 01
0022
Following standards of conduct Defining specific ethical standards Companies that create or use Generative AI
(down 4 percentage points from last year) (down 5 percentage points from last year) products need to be familiar with established
standards, internal policies, and procedures: 0033
The focus of this approach relies on standards of In this approach, companies establish ethical industry-produced documents like the Data &
conduct, defined as guiding pillars that manage standards specific to the organization and the Trust Alliance’s Algorithmic Bias Safeguards for
04
an employee’s entire professional responsibilities. products and services developed and used. Workforce41 and Responsible Data & AI Diligence
They include things like avoiding discrimination, for M&A,42 governmental regulations like the
conflicts of interest, insider trading, bribery, and 60% of respondents indicate their company European Union’s General Data Protection 05
other commonly unaccepted ethical behaviors. considers their mission, purpose, and values Regulation’s (GDPR),43 the National Institute
when navigating emerging technologies. of Standards and Technology (NIST) AI Risk
06
However, the survey indicates fewer companies Management Framework,44 and academic reports
use approach 4 (defining ethical standards like the Berkman Klein Harvard report, which
specific to technology), arguably the most puts forward eight key principles on maximizing
07
ethically robust of the four approaches. the benefits and minimizing the harms of AI.45
Applying ethical principles from one emerging
08
technologies (like quantum computing) to
another (like autonomous vehicles) is inadvisable
because each technology are different.
09
2233
How companies define and meet ethical principles for emerging technologies
01
89% (up 2% from 2022) of survey respondents Figure 5. Percentage of companies surveyed with
standards specific to given emerging technologies
said, except for AI principles, their company
(Percentage) 0022
does not have or are unsure if they have specific
trustworthy and ethical principles governing
80
emerging tech products. Among those that define 72% 0033
principles specific to certain kinds of technology,
the most common is cognitive technologies 60
04
(72%), followed by digital reality (48%), ambient
48%
45%
experiences (45%), distributed ledger (31%),
quantum computing (29%), robotics (27%), and 40 05
31%
autonomous vehicles (24%) (see Figure 5). 29%
27%
24%
20 06
07
0
Cognitive Digital Ambient Distributed Quantum Robotics Autonomous
Technologies Reality Experiences Ledger Computing Vehicles
08
Source: 2023 Deloitte Technology Trust Ethics Survey
For companies who take a robust ethical route—i.e., defining ethical principles 09
specific to each of their products—doing so requires planning.
2244
How companies define and meet ethical principles for emerging technologies
01
Ethical principles should be Figure 6: Frequency at which organizations update ethical principles
(Percentage)
updated frequently 0022
60
Companies implementing a regular ethical review 51%
0033
process for emerging technology products can
42%
build trust, create higher quality products, and be
40 38%
leaders in safeguarding a common social good. 35% 04
The survey shows a trend of companies updating
their principles frequently, moving from a longer
05
cycle to a quarterly or better (53% of companies,
20
up 10 percentage points from 2022). Those with 15%
slower review cycles may find their principles no 8% 06
6%
5%
longer apply to the products and services they
1%
0%
are meant to govern. 0
Monthly (at least) Quarterly Annually Less than annually Never 07
2022 2023
08
Source: 2023 Deloitte Technology Trust Ethics Survey
The survey shows a trend of companies updating their principles frequently, 09
moving from a longer cycle to a quarterly or better approach.
2255
Trustworthy and
Figure 7: Deloitte’s Technology Trust Ethics (TTE) Framework
Ethical Principles Reskilling & Education
01
of Emerging
A
FE
A N D
SECURE
PRIVATE
T d
eT
h
miaee
g
e
nT rgT
oc
iE s ni
gnFh
r g ta etm ch
hn
ee
n
w e oto
lo
h or ik gc
yacl
la
p
o
dn ri
o
ms
deg
e ur nv ce s
ti
si oa
.e
ns sa
s
ofi fr ast c s ot mep p ain n y’s
s
BUST AND
RELIABLE
C Ao cnP cr s ue is rd t ai ec
tS
t n
eU
a
ts be lr
e
Fri
e n
dU lys e r P r o t e cti o
n
T
TIn
ev u ln e r ea b le
c cTh hrn nuoo sll
tA outon
oomous
gg yy
Confidential Discreti o Cn oal nsens Ju ua sl
It ni Atfi uea drb pl ite r ae bta leble
T R
A N S
P
A R E N T A N D E X
P
L A
R
e g
0000022
33
4
o rl
t
n
o C
RO
Adaptable FT rr auE mstth eEit wchs oic rs
k
Visible
ELBI AN yrotu
al
05
& C
e c n a n
r
e
v
o
G E L B A T N U O
CA Cn Aswe Rr ea sb ol le vable
Ownership
Humane Common & Social Goody tilib a n
ia
ts
ud e s u c o Fg n id d A e
u
la
V
U n bi a s e
dI
ncl
usE ivq euA itc ac be
l
es Ls Ai Ib Tl Re
APMI DNA
RIAF
yciloP
&
ecnailpmo 0
007
86
S
ELBISNOPSER
09
Copyright © 2022 Deloitte Development LLC. 26
Trustworthy and Ethical principles of emerging technologies
The principles are ordered by their relative importance according to survey respondents:
01
0022
Responsible Safe and secure Transparent and Robust and reliable Accountable
The technology is created The technology is protected explainable The technology produces Policies in place to determine
0033
and operated in a socially from risks that may cause Users understand how consistent and accurate outputs, who is responsible for the
responsible manner. individual and / or collective technology is being leveraged, withstands errors, and recovers decisi |
Subsets and Splits