Produced in partnership with

To meet their AI ambitions,

organizations must shift from pilots
and experiments to enterprise-wide
deployment.
A playbook for
crafting AI strategy

MIT Technology Review Insight s

Preface
fiA playbook for crafting AI strategyfl is an MIT Technology Review Insights report
spons ored by B oomi. To produce this report , MIT Technology Review Insights
conducted a global survey of C-suite and s enior data executives across countries
and industries. The report also draws on in-depth interviews conducted with
business
leaders on data and AI.
Adam Green was the author of the report , Teres a Elsey was the editor, and Nicola
Crepaldi was the publisher. The res earch is editorially independent , and the
views
express ed are those of MIT Technology Review Insights.
We would like to thank the following executives and experts for their time and
insights:
Kevin C ollins
, Founder and Chief Executive O˜cer, Charli AI
Amy Machado
, Senior Research Manager, IDC
Matt McLar t y
, Chief Technology O˜cer, Boomi
SP Singh
, Senior Vice President and Global Head, Enterprise Applic ation Integration
and Services, Infosys
About the survey
The survey forming the basis of this report was conducted by MIT Technology Review
Insights in March 2024. The survey s ample consists of 205 executives and data and
technology leaders. Eleven industries are repres ented: ˚nancial s ervices,
manufacturing,
IT and telecommunications, consumer goods and retail, pharmaceutic al and health c
are,
government , travel and hospitality, professional services, energy and utilities,
transport
and logistic s, and media and marketing.
Nearly all survey respondents (88%) come from the C-suite. Thes e include chief
executive o˜cers (20%), chief information o˜cers (18%), chief technology o˜cers
(19%), and chief data o˜cers (15%). The respondents™ organizations are
headquartered
in North America (31%); Europe, the Middle East, and Afric a (25%); Asia-Paci˚c
(26%);
and Central and South America (18%). All respondents work at organizations with
more
than U S $500 million in global annual revenue, with 73% repres enting
organizations
generating more than US $1 billion, and 34% more than US $10 billion.

3
MIT Technology Review Insight s
CON TEN TS
01

Executive summary

...................................................................................
...............
4
02

Par tnering for succes s

...................................................................................
........
6

Selecting a vendor

...................................................................................
.....................
7

Finance-friendly AI

...................................................................................
....................
7
03

Counting the cost

...................................................................................
....................
8

Spending expectations

...................................................................................
..........
9

Measuring return on investment

.......................................................................
10
04

Building a data core

...................................................................................
...............
11

Data management: Tips and tactics

................................................................
11
Reckoning with legacy infrastructure

............................................................
13

Data lineage and liquidity

...................................................................................
....
13

Metadata

...................................................................................
........................................
13
05

Acceleration versus caution

.............................................................................
14

Hallucinations, errors, and bias

..........................................................................
14

Cyber risk

...................................................................................
......................................
14

Data privacy and protection

.................................................................................
14

Rising regulatory tide

...................................................................................
.............
15

Compliance challenges

...................................................................................
........
16
06

C onclusion

...................................................................................
..................................
17
4

01

01
Executive
summary
G
iddy predictions about AI, from its
contributions to economic grow th to the
ons et of mass automation, are now as
frequent as the releas e of powerful new
generative AI models. The consultancy
P wC, for example, predicts that AI could boost global
gross domestic product (GDP) 14% by 2030, generating
U S $15.7 trillion.
1
Forty percent of our mundane tasks could be automated
by then, claim res earchers at the University of Oxford,
while Goldman Sachs forec asts US $200 billion in AI
investment by 2025.
2,3
fiNo job, no function will remain
untouched by AI,fl s ays SP Singh, senior vice president
and global head, enterpris e applic ation integration and
s ervices, at technology company Infosys.
While these prognostic ations may prove true, today ™s
businesses are ˚nding major hurdles when they seek
to graduate from pilots and experiments to enterpris e-
wide AI deployment . Just 5.4% of U S businesses, for
example, were using AI to produce a product or s ervice
in 2024.
4
Moving from initial forays into AI use, such as code
generation and customer s ervice, to ˚rm-wide
integration depends on strategic and organizational
transitions in infrastructure, data governance, and
supplier ecosystems. As well, organizations must weigh
uncertainties about developments in AI performance
and how to measure return on investment .
If organizations s eek to sc ale AI across the business
in coming years, however, now is the time to act.
This report explores the current state of enterprise
AI adoption and o˛ers a playbook for crafting an AI
strategy, helping business leaders bridge the chasm
bet ween ambition and execution. Key ˚ndings include
the following:
AI ambitions are subst antial, but few have scaled
beyond pilots
.
Fully 95% of companies surveyed
are already using AI and 99% expect to in the future.
But few organizations have graduated beyond pilot
projects: 76% have deployed AI in just one to three
us e c ases. But bec ause half of companies expect to
fully deploy AI across all business functions within two
years, this year is key to establishing foundations for
enterpris e-wide AI.
AI readiness spending is slated to rise
signi˜cantly.

Overall, AI spending in 2022 and 2023 was modest

or ˝at for most companies, with only one in four
increasing their spending by more than a quarter. That
is s et to change in 2024, with nine in ten respondents
expecting to increas e AI spending on data readiness
(including platform modernization, cloud migration,
and data quality) and in adjacent areas like strategy,
cultural change, and business models. Four in ten
expect to increase spending by 10 to 24%, and one-
third expect to increase spending by 25 to 49%.
fiNo job, no function will
remain untouched by AI.fl
SP Singh
, Senior Vice President and
Global Head, Enterpris e Applic ation
Integration and Services, Infosys

5
Dat a liquidity is one of the most import ant attributes
for AI deployment
.
The ability to s eamlessly access,
combine, and analyze data from various sources

enables ˚rms to extract relevant information and

apply it e˛ectively to speci˚c business scenarios.
It also eliminates the need to sift through vast data
repositories, as the data is already curated and tailored
to the task at hand.
D at a
quality is a major limitation for AI deployment.

Half of respondents cite data quality as the most

limiting data issue in deployment . This is especially
true for larger ˚rms with more data and substantial
investments in legacy IT infrastructure. Companies
with revenues of over U S $10 billion are the most likely
to cite both data quality and data infrastructure as
limiters, suggesting that organizations presiding over
larger data repositories ˚nd the problem substantially
harder.
C ompanies
are not rushing into AI.
Nearly all
organizations (98%) say they are willing to forgo being
the ˚rst to use AI if that ensures they deliver it safely
and s ecurely. Governance, s ecurity, and privacy are the
biggest brake on the speed of AI deployment , cited by
45% of respondents (and a full 65% of respondents
from the largest companies).
Governance, s ecurity, and
privacy are the biggest
brake on the speed of AI
deployment , cited by 45%
of respondents.
6

02

02
Partnering for
success
F
ew companies are ˝ying solo in the AI age. The
cost and complexity of creating large language
models (LLMs) and generative models from
s cratch is prohibitive and an abundance of
platforms and tools are already on the market .
fiFor most organizations, building their own large language
models is very expensive and the value is time-limited,fl
s ays Kevin Collins, founder and CEO of Charli AI, an AI
s olutions provider. fiIf you don™t have the necess ary
expertis e and resources to make a signi˚c ant investment ,
it™s better to ˚ne-tune and optimize o˛-the-shelf models.fl
AI is also increasingly being integrated into existing
s oft ware platforms, including those from giant
technology providers such as Micros oft and Adobe.
fi We are infusing AI into every thing we do within our
organization and similarly taking that s ame mindset to
our clients,fl says SP Singh of Infosys. Companies, of
cours e, can als o pay to use proprietary models from
providers such as OpenAI or ˚ne-tune those models
as needed. They c an also build their own generative AI
tools using open-source models.
Figure 1: Companies seek AI use cases tailored to industr y needs
Is your organization developing the following types of AI us e cas es?

(Organizations that are developing AI use c as es.)

Source: MIT Technology Review Insights survey, 2024
“‘†
““†
˜˚˛˝˙ˆˆ
““†
‘“†
“š†
‘€†
‘’†

7
Matt McLarty, CTO at the s oft ware company B oomi,
concurs that building LLMs will not be the path forward
for most companies. fiMost organizations are not going
to build large language models, which are expensive
and require a massive amount of infrastructure,fl
he s ays. fiInstead, they need to become experts in
applying the new technologies in their own business
context . The companies that don™t put the AI cart
before the business problem horse are going to be
better positioned.fl
Our survey results suggest that business es are taking
that advice, seeking AI use c as es that address their
unique business problems. While many ˚rms are
deploying general-purpos e uses of AI (generative AIŒ
powered chatbots for customer s ervice, for example),
a larger share s ay they are developing industry-speci˚c
us e c ases, or even use c as es unique to their particular
business (s ee Figure 1).
Sectoral nuancesŠincluding di˛ering data formats,
regulatory requirements, and us ages Šsupport the c ase
for industry-tailored approaches to AI. From AI tools
designed to document patient-doctor convers ations to
those leveraged to optimize dry wall ˚nishing, o˛erings
tailored to the needs of speci˚c industries or business es
are most likely to o˛er game-changing results.
5 ,6
Selecting a vendor
Companies have plenty of AI vendors to choos e from;
if any thing, the choice can be daunting for executives
who must balance cost , security, and diversi˚cation.
fiIt feels overwhelming for companies Šthere is a lot of
nois e and questions around who to trust ,fl says Amy
Machado, senior res earch manager at IDC, a market
intelligence ˚rm.
For privacy reasons, many are wary of relying on
publicly available AI tools that may retain and reus e
the data users enter, with s ome companies restricting
employee use of certain LLMs.
7, 8
fiIf you™re shu˙ing
your data o˛ to a general-purpose AI, you have to do
your due diligence on how they ™re treating your data
and what they™re doing with it behind the scenes,fl says
Collins. More specialized LLM-based tools are more
likely to protect organizations™ proprietary data, come
with support, and be consistently updated to address
vulnerabilities.
Ultimately, executives may ˚nd that stitching together
a fimulti-AIfl environment best meets their business
needs and captures the distinctive value propositions
of di˛erent providers.
Finance-friendly AI
Financial services ˚rms operate in a highly regulated
environment with high standards for data privacy,
compliance, and s ecurity. fiIn the ˚nancial services
s ector, trust and s ecurity are paramount ,fl s ays
Kevin Collins at Charli AI. fiOur customers are major
investment banks, hedge funds, and big banks that
are analyzing S&P 500 companies. Every thing we do
needs to be trusted and fact-checked.fl
In ˚nancial services, valuable information is often
locked away in contracts, legal documents, PDF
˚les, spreadsheets, and PowerPoint pres entations.
Extracting and integrating this data for analysis c an
be a daunting task. Specialized s ervices like Charli
AI have opened new possibilities for unlocking this
trapped data: the company ™s tools mine and curate
a range of ˚nancial data, freeing up time for analysts
to perform higher-value tasks.
fiNew AI has the ability to understand ‚locked-up™
content far better than ever before, s o we™re unlocking
a goldmine,fl s ays Collins. fiOur customers refer to us
as the automated analysts for Wall Street Š we c an
do in 30 minutes what it may take an analyst 80 hours
of poring over a 500-page SEC ˚ling to do.fl
While companies are aware of the potential value of
their data, many struggle to leverage it e˛ectively.
fi They ™re very data literate, but they also understand
how challenging it is to get that data into a format
that™s suitable for analy tics and AI,fl says Collins.
Our survey found that ˚nancial services respondents
were least likely to be currently developing AI for
industry-speci˚c use c ases, but companies like
Charli are aiming to change this.

03

03
Counting
the cost
F
rom the electricity requirements for building
and running AI models to the s oft ware,
s ervices, and s alaries needed to manage
s olutions at s c ale, cost is a major
consideration for AI. Google™s U S $8 billion of
spending in the third quarter of 2023 alone was
overwhelmingly driven by AI.
9
Training OpenAI™s GPT-4
required U S $78 million of compute, while Google™s
Gemini Ultra had a compute cost of U S $191 million.
10
Hardware is a major source of price uncertainty,
including demand for graphics processing units
(GP Us). fiAI is very expensive, not just for the
development and training of models, but also for the
operations and hardware costs, particularly GPUs,fl
s ays Collins. Newer GP Us are more expensive, harder
to obtain, and have a limited lifespan requiring a
recurring c apital cost . He adds, fiAI is not cheap. A
lot of our customers are shocked when they realize
how much infrastructure they have to have to get the
performance that they need.fl
While tech companies are largely shouldering
hardware and computation spending in their race for
market supremacy, ˚rms inevitably face their own
costs in areas like data management and monitoring,
which are rarely built into lower-cost general-purpose
LLMs. They must als o invest to ensure compliance with
˜˚
˛˝˚
˙ˆ˚
ˆ˝˚
ˇ˚
˜˚
ˇ˘˚
ˆ˛˚
˙ˆ˚
ˇ˚
Source: MIT Technology Review Insights survey, 2024
Figure 2: Changes in AI readiness spending from 2 022 to 2 023
How did your organization™s investment in AI readiness change in the last year
(from 2022 to 2023)?

9
˚rm-speci˚c regulations, fund maintenance to ensure
the systems they build remain up-to-date and relevant
to the us e cas e, and pay for the increas ed energy
us ed by AI.
11
Talent costs include either hiring AI-skilled
workers or upskilling the existing workforce.
Spending expectations
Polls show that larger ˚rms, by revenue, are more
likely to have AI in production, suggesting that ˚nancial
constraints may be slowing smaller and medium-sized
companies.
12
Analysis from CCS Insights suggests cost
will contribute to a slowdown in generative AI adoption
throughout 2024.
13
As the hype cools, decision-makers
may become more wary, especially as the productivity
gains of generative AI may not be apparent for years.
14
Such ˚ndings are corroborated by our survey, which
found a link bet ween company size and spending.
Budgetary constraints were the biggest obstacle to
speed of AI deployment for medium-sized ˚rms (annual
revenue of U S $500 million to $1 billion), with 47 %
citing this issue compared to a survey average of 22%.
This indicates a fis queezed middlefl of companies that
need to adopt AI to stay competitive in their markets
but for whom the costs are prohibitive.
To date, most ˚rms have kept AI readiness spending
˝at , with larger companies more likely to have ramped
up (s ee Figures 2 and 3).
However, nearly all companies expect to boost
spending in the coming year, with 9 in 10 expecting
to increas e investment by at least 10% and one third
anticipating spending up to 49% more both in data and
adjacent areas like strategy and cultural change (s ee
Figure 4).
Figure 3: Large companies boosted AI spending more aggressively
Companies with revenue greater than $10 billion: How did your organization™s
investment in AI readiness
change in the last year (from 2022 to 2023)?
˜˜˚
˛˝˚
˛˙˚
ˆˇ˚
˜˚
˘˚
˛˘˚
˛˚
˜˚

Source: MIT Technology Review Insights survey, 2024

fiAI is not cheap. A lot of our customers are shocked when
they realize how much infrastructure they have to have to
get the performance that they need.fl
Kevin Collins
, Founder and Chief Executive O˜cer, Charli AI

10

As companies dial in their AI investments, AI cost

dynamic s will of course als o evolve. Innovation will
doubtless bring down costs through e˜ciency gains.
The cost curve for computation will als o fall as startups
and large companies address the shortcomings of
existing hardware and models.
Measuring return on investment
To weigh the ˚nancial outlays of AI, organizations
need to develop robust return on investment (ROI)
methodologies capturing not just the e˜ciency gains
of automating fibusiness as usualfl tasks, but also the
new value AI can create.
Companies are already quantif ying AI™s cost s avings.
Motorola has developed a framework that tracks hours
taken to complete a task with and without generative
AI.
15
Swedish ˚ntech company Klarna has already
c alculated generative AI™s output as equivalent to 700
customer service agents.
16
But fithe minds et is shifting
to using AI as an enabler for revenue grow th, not just
cost s avings,fl says Machado.
As employees become more productive, new
opportunities emerge that may be hard to quantif y. For
example, AI allows data s cientists to explore new ideas
and expand into new products. The automation of
routine tasks by copilots c an alleviate workloads and
reduce burnout, allowing employees to dedic ate more
time to more creative and innovative projects.
As enterpris e use of AI spreads, employee expectations
will shift and AI-enhanced workplace experiences will
become necessary to attract talent . AI professionals
already have di˛erent expectations than traditional job
s eekers, ranking interesting job content and work on
cutting-edge projects higher among their job priorities.
17
The consultancy PwC compares fihard ROI,fl such as
time s aved and productivity boosts, with fisoft ROI,fl
including improved employee experience and ability
to respond ˝exibly to new opportunities.
18
Accurately
ass essing AI™s ROI requires metric s beyond traditional
˚nancial measures. For instance, Collins points out
that tools like Charli AI™s unlock huge value by helping
companies make s ens e of their data more e˜ciently.
˜˜˚
˛˝˚
˙˜˚
˜ˆ˚
˜˚
ˇ˚
˛˘˚
˙ˆ˚
˜˚
˜˚
˛ˇ˝˘ˇ˙˚˚˙˝ˆ˛˝

Source: MIT Technology Review Insights survey, 2024

04

04
fi The minds et is shifting to
using AI as an enabler for
revenue growth, not just
cost savings.fl
Amy Machado
, Senior Res earch Manager,
IDC
Figure 4 : AI readiness spending set to rise signi˜cantly into 2 024
How will your organization™s investment in AI readiness change in the next year
(from 2023 to 2024)?

11
04

04
Building a
data core
D
ata quality, infrastructure, and governance
are all ess ential to deliver AI workloads
e˜ciently and at s c ale. While AI
technology c an be transformative, without
the data foundations in place,
organizations will struggle to both collect the right data
and derive insight from it . fiPeople still have the
perception that AI today is magic or super intelligent,
and that™s far from the truth,fl s ays Collins. fiAI is a
s cience and a tool. You still have to do all of the hard
work around data governance and ˚guring out the right
applic ation of thes e tools.fl
Organizations that overlook these supporting
dimensions may struggle to deploy AI successfully.
Data quality problems, cited by half of respondents
as their most limiting data issue, c an severely hinder
AI performance. fiGoing forward, the convers ation
will be around getting data AI-ready and in the right
format to actually utilize it e˛ectively,fl s ays Machado.
Data infrastructure or pipelines, data integration
tools, and cloud migration were also mentioned by
many respondents as major barriers to deployment,
indic ating the importance of a robust IT architecture.
Thes e issues are particularly prominent for companies
with revenues of over U S $10 billion, which are
the most likely to cite both data quality (52%) and
data infrastructure (55%) as obstacles, compared
with overall survey averages of 49% and 44% (s ee
Figure 5). Organizations presiding over larger data
repositories and legacy IT infrastructure may be
˚nding greater complexity and costs in transitioning to
an AI-ready architecture.
Data management: Tips and tactic s
Ł

Track and manage the lineage of your

organization™s data to maintain data quality and
integrity as it moves from one analytic al model
or application to the next .
Ł

Optimize communication across the

organization on AI, and on data aggregation
and requirements, ensuring cross-functional
teams collaborate on business cas es.
Ł

Instead of focusing on data centralization, adopt

a strategy that prioritizes data contextualization,
such as extracting metadata and storing it in an
appropriate metadata language.
Ł

Identif y your organization™s core capabilities

and competencies, ensuring you are able to
adapt to rapid changes and are not overly
reliant on legacy infrastructure.
fiAI is a s cience and a tool.
You still have to do all of the
hard work around data
governance and ˚guring out
the right application of
thes e tools.fl
Kevin Collins
, Founder and Chief
Executive O˜cer, Charli AI

12

Figure 5: Dat a bottlenecks in AI deployment

Which aspects of your organization™s data are most limiting the speed to deploy AI?
Source: MIT Technology Review Insights survey, 2024
˜˚˛˝˙ˆˆ
˜˚˛˚˝˙ˆ˚ˇ˘˛

13
Reckoning with legacy infrastructure
Transitioning away from legacy IT infrastructure c an
be daunting for organizations of all sizes. fiLegacy
systems and sunk costs are de˚nitely an inhibitor
for all kinds of data management initiatives,fl s ays
Machado. fiMany organizations don™t want to touch
certain legacy systems bec aus e ‚it™s kind of working™
and only one pers on knows how to maintain that
older technology.fl Thes e legacy systems are often
bas ed on fragmented, outdated architectures and
programming languages that are hard to integrate
with modern AI s olutions.
To reap the bene˚ts of AI, organizations must develop
an IT architecture capable of accommodating both
structured and unstructured data across the entire
lifecycle, from the s ource to processing, analytics,
and storage. Unstructured data c an yield considerable
insight for AI systems, but its raw information c an be
di˜cult to assimilate into existing systems. That c an be
problematic since, according to Machado, 90% of the
data in an enterprise content management system is
unstructured.
A centralized repository s olution in the form of a data
lake c an be useful, but it must be well organized and
structured to prevent it becoming ˚lled with irrelevant
and unus able data. fiIt™s not enough to just have a big
data lake, becaus e that™s going to become a data
swamp; you have to have good governance over the
data lineage and how it ˝ows from one model to the
next ,fl s ays Collins.
Nearly half (46%) of survey respondents als o cite
manual, non-digitized, or non-automated process es
for data handling and management as a limiting factor
for AI deployment . Without strategies to fully digitize
process es across the business, organizations will
leave valuable data s ources untapped and limit AI™s
automation potential.
Data lineage and liquidity
Deciding which data to leverage requires it to be
recombined and contextualized. Implementing data
lineage tracking and quality-assurance measures can
support this contextualization process, pres erving
the integrity of data as it undergoes processing and
analysis.
According to McLarty, data liquidity, or the ability
to s eamlessly access, combine, and analyze data
from various sources, is one of an organization™s
most important AI ass ets. Data liquidity empowers
organizations to understand the relationships,
dependencies, and nuances within their data,
enabling them to extract relevant information and
apply it e˛ectively to speci˚c business scenarios.
fiOrganizations with high data liquidityŠthe ability to
get the right data at the right time and placeŠwill be
most successful with AI,fl says McLarty.
Data liquidity helps ensure data is readily accessible,
us able, and shareable. Nurturing liquid data ass ets requires
a ˝exible, harmonized and s c alable IT infrastructure that
c an seamlessly handle increasing volumes and varieties
of data without disrupting existing operations.
Metadata
Extracting metadataŠor data about dataŠhelps to
contextualize data for us e in AI models. fiOrganizations
want to understand their data upfront to see what is
valuable for real business purpos es, and metadata holds
every thing together to make this possible,fl s ays Machado.
Within a content management system, data sources
from di˛erent systems can be connected by a common
metadata language, and it is important , Machado says,
to s elect a language suitable for the organization™s us e
c as e. For sharing information on biologic al specimens, for
instance, the Darwin Core language might be considered
appropriate. VRA Core is us ed for works of visual culture.
By curating metadata, enterpris es can harness the
power of their previously untapped unstructured data
res erves. This approach allows them to optimize their
data for AI solutions that will drive business value.
fiOrganizations with high data
liquidityŠthe ability to get the
right data at the right time
and placeŠwill be most
successful with AI.fl
Matt McLarty
, Chief Technology O˜cer,
B oomi

14

05

05
Acceleration
versus caution
A

signi˚cant barrier to the rapid adoption of AI

is the need to ensure robust governance,
s ecurity, and privacy measures are in place,
especially in the c as e of new risks pos ed by
generative AI. Thes e issues s erve as the
biggest brake on the speed of AI deployment for 45% of
companies Šand for 65% of companies with over $10
billion in revenue. That di˛erence may be due to the
increas ed risk pro˚le of larger organizations, including
their larger fiattack surfaces,fl making them more
vulnerable to potential breaches and regulatory scrutiny.
Hallucinations, errors, and bias
Generative AI comes with a range of new risks,
including producing inaccurate information or
fihallucinationsfl; legal risks such as plagiarism,
copyright infringement , and liability for errors;
challenges around privacy and data ownership; and
systemic bias. These risks c an harm customers and
expos e companies to legal liability. Air Canada was
recently held liable for damages caus ed by misleading
information provided by an AI chatbot .
19
Inaccurate information and hallucinations can
contribute to heightened bias in AI models. In
˚nancial services, for instance, AI-powered credit
risk assessments could help mitigate systemic
dis crimination in credit applications Šbut only if the
data us ed to train thes e AI systems is itself reliable,
accurate, and free of distortions and biases. An AI-
powered s olution built on discriminatory data and
algorithms, however, will amplify existing biases,
resulting in even less fair outcomes. Similar biases c an
crop up in applications including health care, policing,
and advertising, where algorithms can accentuate
gender biases or overlook underrepres ented minority
groups.
Cyber risk
The us e of generative AI tools also comes with new
opportunities for cyberattacks. In a prompt injection
attack , bad actors craft prompts to elicit unintended
respons es from an LLM, to manipulate its behavior,
achieve unauthorized access, or bypass security
measures. In February 2023, a student at Stanford
University discovered s ensitive details about Bing Chat
using this technique, instructing it to fiignore previous
instructionsfl and reveal hidden information.
20
In data pois oning, the data us ed to train an AI model
is intentionally targeted and injected with malicious
information. Leakages of data from databases and
pipelines that feed LLMs are another s ource of concern.
Data privacy and protection
Data privacy and protection demand signi˚cant
investment in the AI age. This is of particular concern
for high-risk sectors and for thos e to whom security
and regulatory compliance are paramount . In our
survey, respondents from ˚nancial services were 13
points more likely to cite governance, s ecurity, and
privacy as factors inhibiting AI adoption (58% of
respondents, compared to the survey average of 45%).

15
Our survey ˚nds that nearly all companies across
s ectors are taking safety and security s eriously, with
a remarkable 98% of respondents saying they are
willing to forego ˚rst-mover advantage if it ensures
AI is delivered s afely and securely (s ee Figure 6).
This c autious approach highlights the widespread
recognition that AI risks c an be potentially devastating
to brand reputation or prompt a stringent regulatory
respons e. Companies are taking the long view, striking
a balance bet ween progress and security.
In many c as es, ˚rms are leveraging AI its elf to
strengthen their cybers ecurity and resilience.
Companies now must account for threats at a fimachine
s c ale, rather than a human s c ale,fl and organizations
s ee the power of AI solutions for this work .
21
AI tools c an
participate in vulnerability s c anning, risk ass essments,
threat detection, and proactive threat hunting.
Despite the progress made, every new AI s olution can
als o create new threats. Human oversight will remain
an important part of the process, not least bec ause the
human element c an help improve these tools. fiGiven
trust and governance concerns, we™re still in a position
where we need to have lots of human guardrails in the
picture,fl says McLarty.
Rising regulatory tide
The regulatory response to AI thus far has focused on
establishing safeguards and governance frameworks
to supervise AI activity, control risk, and provide a level
playing ˚eld. An analysis of 127 countries shows that
the number of bills mentioning fiarti˚cial intelligencefl
pass ed into law worldwide grew from just 1 in 2016
to 37 in 2022.
22
While this re˝ects a recognition of
the transformative impact of AI, there is still a need
to accelerate the implementation of comprehensive
policies and safeguards on ethic s, privacy and security.
The European Union™s AI Act is one of the most
comprehensive initiatives currently being enacted. The
Act establishes a risk-based approach to c ategorizing
AI systems and imposes stricter requirements
on systems used in higher-risk areas such as
infrastructure, employment , and law enforcement .
AI used in thes e areas will face rigorous obligations
related to data quality, transparency, human oversight ,
accuracy, and other considerations before being
allowed on the market . Companies implementing
high-risk systems will need to complete a conformity
ass essment before being placed on the market, with
low-risk systems such as chatbots subject to voluntary
codes of conduct and transparency requirements.
23

˜˚˛˝˙ˆˆ

Source: MIT Technology Review Insights survey, 2024

Figure 6: Executives willing to delay AI use to assure safet y and security
Respondents who agree with the following statement: fiWe are willing to forego being
˚rst to us e AI in our
s ector if that ensures that we deliver it s afely and securely.fl

16

Certain uses of AI, such as social scoring and mass

surveillance, meanwhile, will be banned outright .
24
Similar initiatives are underway in the US, where
President Joe Biden issued an executive order s etting
out new standards for AI s ecurity and s afety in October
2023. The order imposes a degree of human oversight
through measures such as an advanced cybersecurity
program aimed at developing AI tools to ˚nd and ˚x
vulnerabilities in critical soft ware. It also establishes
an AI Safety and Security Board, and it requires that
developers of the most powerful AI systems share
s afety test results with the government .
25
Compliance challenges
The ˝urry of activity around AI laws and policies will
be an ongoing challenge for enterpris e. fiIt can take an
organization a couple of years to get up to speed on
how to comply with new regulations, given the amount
of reading required to understand all the policies and
guidelines,fl says Collins.
A key challenge will be establishing workable
systems for assessing compliance and performance.
AI assurance process es such as algorithm impact
ass essments and bias audits help organizations
monitor and evaluate the c apabilities of their
AI systems, playing a crucial role in building
con˚dence and trust .
26
Gaps in regulation make
devising frameworks for AI assurance challenging,
but standards like thos e set by the International
Organization for Standardization (ISO) o˛er guidance.
27
The current scope of AI regulation recognizes that risk
c an never be avoided entirely. Instead, the regulation is
centered on disincentivizing irresponsible and risky AI
deployment. Going forward, sav v y organizations may
prioritize AI explainabilityŠthe ability to understand the
decision-making process and outputs of an AI modelŠ
both for compliance reas ons and to build customer
trust .
Nearly all companies across s ectors are taking safety and s ecurity
s eriously, with a remarkable 98% of respondents s aying they are
willing to forego ˚rst-mover advantage if it ensures AI is delivered
s afely and s ecurely.
06

06

17
06

06
Conclusion
O
ver the last year, organizations have come
to understand the power and potential of
AI. This year, those organizations are
s eeking to shift from small pilots to
enterpris e-wide deployment of AI at s cale.
This report aims to assist with the next steps of that
journey, o˛ering the following principles as guidance.
This is a critical year for building AI foundations.
As
organizations strive to meet ambitious short-term AI
goals, they are dis covering that having the right data
foundations is critical. Organizations are responding
by doubling down on investments in data quality, data
liquidity, and IT infrastructure. fiThere™s no question in
my mind,fl says McLarty, fithat companies that have
good hygiene and rigor around their data are going to
be way better positioned for the AI landsc ape.fl
The highest-leverage AI use cases drive t argeted,
busines s-speci˜c result s.
Although general-purpos e
generative AI use c as es are increasingly simple to
implement , they are equally available to competitors and
customers. The most valuable use c as es are those that
create unique competitive advantage for the business.
fi You c an deploy AI c apabilities, but if they don™t provide
true business value, you™re just wasting money to s ay
you did s omething cool, with no real reason. That
undermines trust in future projects,fl warns Machado.
Financial considerations and partnerships are key.
This year, the costs of AIŠfrom GP Us to skilled talent
to energy consumptionŠmust be accounted, and
a realistic approach to measuring AI™s ROI must be
developed. Bec aus e few organizations will go it alone
when it comes to AI, making decisions about the right
partners, vendors, and tools will be essential to most
companies™ AI futures.
Speed of AI adoption is being moderated by a
realistic as sessment of its risks .
Organizations are
rightly c autious about the risks pos ed by injudicious
us e of AI, and essentially all agree that caution is
preferable to ˚rst-mover advantage when sc aling AI.
An emerging regulatory environment and a better
understanding of how the risks of AI c an be mitigated
should serve to speed adoption.
Thes e principles, it should be noted, are not deeply
technological. fi The biggest factor holding back AI
implementation,fl s ays McLarty, fiis people not knowing
where to start. People need to understand that you
don™t actually have to be an expert on how to create
generative AI in order to get value from generative AI.fl
fi The biggest factor holding back AI implementation is people not
knowing where to start . People need to understand that you don™t
actually have to be an expert on how to create generative AI in order
to get value from generative AI.fl
Matt McLarty
, Chief Technology O˜cer, B oomi

18

Endtnotes
1.

fiSizing the prize: What™s the real value of AI for your business and how c an you c
apitalise?fl P wC, 2017, https://preview.thenewsmarket .com/Previews/P WC/
DocumentAss ets/476830.pdf.
2.

fi40% of time spent on mundane chores could be automated within 10 yearsŠAI

experts,fl University of Oxford, February 22, 2023, https://w w w.ox.ac.uk
/news/2023-02-22-
40-time-spent-mundane-chores-could-be-automated-within-10-years-ai-experts -0.
3.

fiAI investment forec ast to approach $200 billion globally by 2025,fl Goldman Sachs,
August 1, 2023, https://w w w.goldmansachs.com/intelligence/pages/ai-investment-
forec ast-to-approach-200-billion-globally-by-2025.html.
4.

Richard Vanderford, fiGenerative AI Isn™t Ubiquitous in the Business WorldŠat Least

Not Yet ,fl The Wall Street Journal, April 7, 2024, https://w w w.wsj.com/articles/
generative-ai-isnt-ubiquitous -in-the-business-worldat-least-not-yet-07c07f01?
mod=tech_lead_pos3.
5.
fiAWS Announces AWS HealthScribe,fl AWS, July 26, 2023,
https://press.aboutamazon.com/2023/7/aws-announces -aws -healthscribe-a-new-
generative-ai-powered-
s ervice-that-automatic ally-creates -clinical-documentation.
6.

fiCanvas + USG Partnership Aims to Transform Dry wall Industry,fl Canvas, May 11,
2023,
https://w w w.c anvas.build/c anvas -usg-partnership-aims-to-transform-dry wall-
industry
/.
7.

Aaron Tilley and Miles Kruppa, fiApple Restricts Employee Us e of ChatGPT, Joining
Other Companies Wary of Leaks,fl The Wall Street Journal, May 18, 2023,

https://w w w.wsj.com/articles/apple-restricts-us e-of-chatgpt-joining-other-

companies -wary-of-leaks -d44d7d34.
8.

Siladitya Ray, fiSamsung Bans ChatGPT Among Employees After Sensitive Code Leak ,fl
Forbes, May 2, 2023, https://w w w.forbes.com/sites/siladityaray/2023/05/02/
s amsung-bans-chatgpt-and-other-chatbots -for-employees-after-s ensitive-code-
leak/?sh=34b316360781.
9.

Jennifer Elias, fiWall Street wants to know how Google™s going to pro˚t from AI,fl
CNBC, October 24, 2023, https://w w w.cnbc.com/2023/10/24/wall-street-wants -to-
know-
how-googles-going-to-pro˚t-from-ai.html.
10.

fiArti˚cial Intelligence Index Report 2024,fl Stanford University, 2024,

https://aiindex.stanford.edu/wp-content/uploads/2024/04/HAI_AI-Index-Report-
2024.pdf.
11.

Ryan Peeler, fi The Hidden Costs of Implementing AI In Enterprise,fl Forbes, August

31, 2023, https://w w w.forbes.com/sites/forbestechcouncil/2023/08/31/the-hidden-
costs -of-implementing-ai-in-enterprise/?sh=932cbac4d5c1.
12.

fi2023 Global Trends in AI Report ,fl S&P Global, 2023, https://w w w.weka.io/res

ources/analyst-report/2023-global-trends -in-ai/.
13.

fiCCS Insight™s Predictions for 2024 and B eyond Revealed,fl CCS Insight , October
10, 2023, https://ww w.cc sinsight .com/company-news/cc s -insights -predictions -
for-2024-
and-beyond-revealed/.
14.

fiGartner Places Generative AI on the Peak of In˝ated Expectations on the 2023 Hype
Cycle for Emerging Technologies,fl Gartner, August 16, 2023,
https://w w w.gartner.
com/en/newsroom/press -releas es/2023-08 -16-gartner-places-generative-ai-on-the-
peak-of-in˝ated-expectations-on-the-2023-hype-cycle-for-emerging-technologies
.
15.
Kristin Broughton and Mark Maurer, fiCFOs Tackle Thorny Calculus on Gen AI: What™s
the Return on Investment ?fl March 24, 2024, https://w ww.wsj.com/articles/cfos -
tackle-thorny-calculus -on-gen-ai-whats -the-return-on-investment-24ebf435?mod=s
eries_proaiimplementation.
16.

Richard Vanderford, fiGenerative AI Isn™t Ubiquitous in the Business WorldŠat Least

Not Yet ,fl The Wall Street Journal, April 7, 2024,

https://ww w.wsj.com/articles/generative-ai-isnt-ubiquitous -in-the-business-

worldat-least-not-yet-07c07f01?mod=tech_lead_pos3.
17.

fiHow to Attract , Develop, and Retain AI Talent ,fl B CG, 2023, https://ww

w.bcg.com/publications/2023/how-to-attract-develop-retain-ai-talent .
18.

Anand Rao, fiSolving AI™s ROI problem. It™s not that easy.fl P wC, July 20, 2021,
https://w w w.pwc.com/us/en/tech-e˛ect/ai-analytic s/arti˚cial-intelligence-
roi.html.
19.

Maria Yagoda, fiAirline held liable for its chatbot giving passenger bad adviceŠwhat
this means for travellers,fl BBC, February 23, 2024,

https://ww w.bbc.com/travel/article/20240222-air-c anada-chatbot-misinformation-

what-travellers-should-know.
20.

Jacob Fox, fiPrompt Injection Attacks: A New Frontier in Cybersecurity,fl Cobalt, May
31, 2023, https://ww w.cobalt .io/blog/prompt-injection-attacks.
21.

fiCis co AI Readiness Index,fl Cisco, 2023, https://w w w.cisco.com/c /dam/m/en_us/s

olutions/ai/readiness -index/documents/cis co-global-ai-readiness -index.pdf.
22.

fiArti˚cial Intelligence Index Report 2023,fl Stanford University, 2023,

https://aiindex.stanford.edu/wp-content/uploads/2023/04/HAI_AI-Index-
Report_2023.pdf.
23.

fi The AI Act of the European Union,fl TaylorWessing, https://w w

w.taylorwessing.com/en/insights -and-events/insights/arti˚cial-intelligence-act .
24.

fi What Is The EU Arti˚cial Intelligence Act ?fl European Movement Ireland, September

18, 2023, https://w w w.europeanmovement .ie/what-is-the-eu-ai-act/.
25.

fiFACT SHEET: President Biden Issues Executive Order on Safe, Secure, and Trust
worthy Arti˚cial Intelligence,fl The White Hous e, October 30, 2023,
https://w w w.
whitehous e.gov/brie˚ng-room/statements-releas es/2023/10/30/fact-sheet-president-
biden-issues-executive-order-on-s afe-secure-and-trust worthy-arti˚cial-
intelligence
.
26.
fiIntroduction to AI assurance,fl UK Government , February 12, 2024, https://w w
w.gov.uk /government/public ations/introduction-to-ai-assurance/introduction-to-ai-
assurance.
27.

fiAI management systems: What business es need to know,fl International Organization

for Standardization, https://ww w.is o.org /arti˚cial-intelligence/ai-management-
systems.

19
About MIT Technology Review Insights
MIT Technology Review Insights is the custom publishing division of
MIT Technology Review,
the world™s
longest-running technology magazine, backed by the world™s foremost technology
institutionŠproducing

live events and res earch on the leading technology and business challenges of the
day. Insights conducts
qualitative and quantitative res earch and analysis in the U S and abroad and
publishes a wide variety of content ,
including articles, reports, infographic s, videos, and podc asts. And through its
growing MIT Technology Review
Global Insights Panel
, Insights has unparalleled access to s enior-level executives, innovators, and
entrepreneurs worldwide for surveys and in-depth interviews.
While ever y e˜ort has been t aken to verif y the accuracy of this information, MIT
Technology Review Insight s cannot accept any responsibilit y or liabilit y for
reliance by any per son
in this report or any of the information, opinions, or conclusions set out in this
report .
© Copyright MIT Technology Review Insights, 2024. All rights res erved.
About B oomi
Boomi
, the intelligent integration and automation leader, helps organizations around the
world automate and
streamline critic al process es to achieve business outcomes faster. Harnessing
advanced AI c apabilities, the
B oomi Enterpris e Platform s eamlessly connects systems and manages data ˝ows with
API management ,
integration, data management , and AI orchestration in one comprehensive s olution.
With a customer bas e
exceeding 20,000 companies globally and a rapidly expanding net work of 800+
partners, B oomi is
revolutionizing the way enterpris es of all sizes achieve business agility and
operational excellence.
Dis cover more at
boomi.com
.
Illustrations
Illustra tions as sembled by Tim Huxford wi th elements from Shutterstock and Adobe
Stock.

MIT Technology Review Insight s

w w w.technologyreview.com
insights@technologyreview.com