Strategy & Corporate Finance Practice

Gen AI: A guide for CFOs

How should CFOs approach generative AI—enterprise-wide and
in the finance function—and what can they do right now to rapidly climb
the learning curve?
by Ankur Agrawal, Ben Ellencweig, Rohit Sood, and Michele Tam

November 2023
 Technology changes every business, often Gen AI and enterprise-level
radically, and the pace of change is getting faster. value creation
Now, generative AI (gen AI) is beginning to The most important action that CFOs should take is
show its disruptive potential (see sidebar “Gen AI: to identify the largest opportunities for value
A primer”). The technology won’t affect all creation—and then make sure that they receive the
businesses equally, and certainly not at the same money and other resources that they need. Gen AI
time. Yet across industries and geographies, holds the potential to be a revolutionary technology,
gen AI could present substantial opportunities for but it doesn’t change foundational principles of
significant value creation. finance and economics: a company must generate a
return above its cost of capital.
But value doesn’t create itself. Instead, it’s the
CFO’s role to allocate resources at the enterprise Moreover, company capital (or access to more
level—rapidly, boldly, and disproportionately— capital) is finite, and projects compete with one
to the projects that create the most value, regardless another. For CFOs to maximize value creation, they
of whether they are driven by gen AI. Similarly, must rank the company’s 20 to 30 most value-
in leading the finance function, the CFO can’t accretive projects regardless of whether they are
implement gen AI for everyone, everywhere, all at AI-related. The Pareto principle always applies;
once. CFOs should select a very small number of usually a very small number of opportunities will
use cases that could have the most meaningful deliver most of the company’s cash flows over
impact for the function. In this article, we’ll discuss the next decade. The CFO cannot let the highest-
how CFOs can most effectively approach gen AI value initiatives wither on the vine merely because
company-wide, prioritize specific use cases within a competing project has “gen AI” attached to it.
the finance function, and rapidly climb the gen AI Sooner or later, shareholders have to pay for
learning curve. everything, and none of them should be on the hook
for a gen AI premium.

Gen AI: A primer

Generative AI (gen AI) is a predictive analytical AI. It can be adapted to generate with new innovations in robotics and
language model that produces new (hence the name) content that seems automation, to make human lives better,
unstructured content such as text, images, human, such as written documents, audio more creative, and more self-fulfilling.
and audio. Traditional, or analytical, AI, conversations, software programming,
by contrast, is used to solve analytical tasks charts, and visual images. But it doesn’t For more about gen AI, see “The state of AI
such as classifying, predicting, clustering, create the way a human does: it predicts in 2023: Generative AI’s breakout year,”
analyzing, and presenting structured data. what a human would enjoy or find useful. “The economic potential of generative AI:
And unlike traditional, analytical AI, The next productivity frontier,” “The
Gen AI technology is powered by artificial gen AI doesn’t calculate or do math. The organization of the future: Enabled by gen
intelligence models called foundation technology, therefore, won’t displace AI, driven by people,” and visit our featured
models, which are trained on a broad set traditional AI. Instead, the ideal is that each insights page “Insights on Artificial
of data, including the outputs from will complement and enable the other, Intelligence,” all on McKinsey.com.

2 Gen AI: A guide for CFOs

The best CFOs are at the vanguard of
innovation, constantly learning more
about new technologies and ensuring
that businesses are prepared as
applications rapidly evolve.

But to that same point of maximizing shareholder as well. Morgan Stanley’s Wealth Management
value, a CFO must recognize existential threats division, for one, has shown remarkable progress in
to a company’s businesses and be clear about the developing an internal-facing service that uses
most important levers for generating and sustaining OpenAI technology and Morgan Stanley’s proprietary
higher cash flows. When an opportunity squarely data to provide its financial advisers with relevant
addresses or significantly relies on gen AI, CFOs content and insights in seconds.
should not shunt it aside because they don’t
understand the technology or lack imagination A world-class CFO ensures that these and other
to recognize the value it could create. gen AI initiatives aren’t starved of capital. Indeed,
one of the biggest misconceptions we find is the
Often, a choice about capital allocation won’t be belief that it’s the job of the CFO to wait and see—or,
either/or: an important business or value lever worse, be the organization’s naysayer. Capital
can have an even greater impact by incorporating shouldn’t sit; it should be aggressively moved to
gen AI. That applies whether the most important fund profitable growth. The best CFOs are at
drivers are revenue generators (such as creating an the vanguard of innovation, constantly learning
interface that will attract more customers or more about new technologies and ensuring
encourage more cross-selling), margin expanders that businesses are prepared as applications rapidly
(for example, reducing manufacturing, procurement, evolve. Of course, that doesn’t mean CFOs should
or distribution costs), or a factor that spans throw caution to the wind. Instead, they should
revenues and costs (such as helping to attract, relentlessly seek information about opportunities
retain, and motivate employees by freeing them and threats, and as they allocate resources, they
for more creative work). should continually work with senior colleagues to
clarify the risk appetite across the organization
Microsoft, for example, has been far ahead of and establish clear risk guardrails for using gen AI
the curve in investing in gen AI to build competitive well ahead of the test-and-learn stage of a project
advantage in key core businesses, such as by (see sidebar “New technology, new risks”).
creating the Microsoft 365 tool Copilot, which
provides real-time suggestions to improve For some CFOs, it may feel orthogonal as a
documents, presentations, and spreadsheets. “numbers person” to champion visionary innovation.
While demonstrated commercial success has But they’ve got to do it: market-beating growth
largely come from digital natives, some traditional, won’t come from incremental change. Behind the
nontechnology companies are moving aggressively scenes, CFOs can take advantage of their

Gen AI: A guide for CFOs 3

New technology, new risks

The CFO is often a company’s de facto what appeared to be a convincing legal particularly applicable for the finance
chief risk officer, and even when a company brief—except that its citations were fantasy, function for internal use cases—company
already has a separate risk team (as including court cases and quotations data is often proprietary.) Other risks
is the case, for example, with financial supposedly made by judges but in fact include privacy breaches, such as exposing
institutions), CFOs remain a key partner in conjured by the model.1 Gen AI models can confidential or even market-moving
helping to identify and mitigate risks. also produce wildly incorrect financial information to third-party models, model
reports; the product appears flawless, but bias, and tail event errors that could
Generative AI (gen AI) brings a slew of the line items don’t apply to the company result from an absence of having a human
them. In fact, the old phrase that “to err is and the math looks like it should sum but being stress test what the solution
human; to really foul things up requires doesn’t. What seems like a real 10-K form creates.2 An overreliance on gen AI and
a computer” applies now more than ever. on the first flip through may be wholly lack of understanding underlying analyses
To start with, even the most cutting- untethered from reality. or data can also reduce the preparedness
edge gen AI tools can make egregious of finance teams to gut check “reasonable­
mistakes. Since gen AI can’t do math Beyond hallucinations, other important ness” of outputs. It’s critical to bear in mind
and can’t “create” out of thin air—instead, concerns include legal issues stemming that gen AI is designed to enhance the
it’s constantly solving for a what a from the intellectual property used as the productivity of people, not to replace them.
human would want—it can “hallucinate,” source of gen AI models, not just in terms While it can boost efficiency tremendously,
presenting what seems to be a convincing of the rights to present the information real people must always be involved.
output but what is actually a nonsense but also to process the information to
result. Such was the case, for example, teach the solution as it learns. (This is a
when one leading gen AI platform wrote major reason why gen AI can be

1
Dan Mangan, “Judge sanctions lawyers for brief written by AI with fake citations,” CNBC, June 22, 2023.
2
See Roger Burkhardt, Nicolas Hohn, and Chris Wigley, “Leading your organization to responsible A.I.,” McKinsey, May 2, 20129; and Benjamin Cheatham, Kia Javanmardian,
and Hamid Samandari, “Confronting the risks of artificial intelligence,” McKinsey Quarterly, April 26, 2019.

relationships with functional and business Gen AI and the finance function
unit leaders to prod them about exploring gen AI For many finance functions, gen AI will be table
opportunities, and repeatedly follow up in stakes—one among several of the essential tools
subsequent interactions. They should upskill and that every effective, forward-looking finance
empower their own team members to build function will use. The technology has the potential
important relationships across the organization to save meaningful amounts of time and resources.
and better understand the assumptions That in itself is a reason to move forward—and
underpinning innovation projects. And they should why most, if not all, finance functions in large
be “always on” when it comes to innovation— enterprises will likely be using gen AI in significant
not just in periodic reviews or when closer scrutiny ways within the next three to five years. In fact,
is needed for struggling projects. one way to conceptualize gen AI is to consider it as

4 Gen AI: A guide for CFOs

digital’s “third wave” (Exhibit 1). The first wave is to bold CFOs put their finance team in the best
establish a digital foundation; in our biennial survey position to learn to work with these tools as
of global CFOs completed in late 2023, about two- the technology gains momentum.
thirds of respondents reported that their functions
were digitally connected and using data for the Getting started in the finance function
basics such as visualization in dashboards.1 CFOs typically aren’t software engineers, let alone
practiced experts in predictive language models.
The second wave, clearly under way, is analytics But they don’t have to be. Their first step should be
empowerment; about half of the CFOs reported to try out the technology to get a feel for what
that their functions were already using advanced it can do—and where its limits are at the moment.
analytics for discrete use cases such as cost Solutions such as OpenAI’s ChatGPT are available
analysis, budgeting, and predictive modeling. The online, and other applications (including McKinsey’s
third wave will make extensive use of robotics and Lilli) are already in use.
AI. Very few companies are at the third wave yet. But

Exhibit 1

Generative AI is part of the ‘third wave’ of digitization—and leading finance

functions are already using it.
Financial performance, by wave level, 2023,1 % (illustrative)

WAVE 1 WAVE 2 WAVE 3

Digital Analytics Intelligent automation
foundation empowerment revolution

Top-performing
organizations 65 60 36

1.1× 1.6× 1.4×

All other
organizations 62 37 26

Data Advanced Robotics

visualization analytics
and connectivity for finance
1
Self-assessed financial performance vs competition.
Source: McKinsey biennial global survey of CFOs

McKinsey & Company

1
See “CFOs’ balancing act: Juggling priorities to build resilience,” McKinsey, August 31, 2023.

Gen AI: A guide for CFOs 5

 Try experimenting by uploading publicly available Gen AI doesn’t create like a human does or have a
earnings calls transcripts from your competitors and eureka moment. It doesn’t even do math (that’s
asking the AI tool to produce the five most-asked the remit of traditional, or analytical, AI). Gen AI is
questions—and to suggest answers. Or upload your a predictive language model—a translator that
company’s and its competitors’ financials, and sits above existing unstructured data and seeks
ask the gen AI solution to take the perspective of to generate content that a human would find
an activist investor: What elements of your pleasing. The data sets themselves first need to
company’s performance would an activist home be rigorously processed and curated, just as data
in on? Depending upon the sophistication of the gen scientists prepare data lakes for advanced analytics
AI solution, CFOs can also upload invoice and and analytical AI.
payments data and ask it to create charts that
visualize the information—including a request for Identifying use cases
the one, most important chart. We find that We believe that gen AI can have an impact on
when CFOs experience the technology firsthand, finance functions in three major ways. First, through
they not only better understand what gen AI is automation—performing tedious tasks (such as
but also more rapidly grasp near- and immediate- creating first drafts of presentations). Second, by
term opportunities. augmentation—enhancing human productivity
to do work more efficiently (such as by gathering
We advise CFOs to budget a nominal amount at the and synthesizing multiple pieces of information
learning stage, not for purposes of deploying AI at into a coherent narrative). Third, through
scale but rather to improve the learning experience acceleration—extracting and indexing knowledge
for themselves and their team members. Again, to shorten financial reporting cycles, and
though, the goal is not to let a thousand flowers speeding up innovation. Gen AI can greatly enhance
bloom. Instead, CFOs should select a handful of use CFOs’ ability to manage performance proactively
cases—ideally two to three—that could have and support business decisions. A high-performing
the greatest impact on their function, focus more on finance function understands the use cases that
effectiveness than efficiency alone, and get going. could most significantly and feasibly improve their
function (Exhibit 2).
One point that quickly becomes apparent when
moving forward is that gen AI is not plug and play; For example—and by no means as an exhaustive
companies can’t simply set the models on existing list—a few multinational enterprises have already
sources of information and let them have at it. begun to implement the following:

CFOs’ first step should be to try out

the technology to get a feel for
what it can do—and where its limits
are at the moment.

6 Gen AI: A guide for CFOs

Exhibit 2

A high-performing finance function understands the use cases that could

most significantly and feasibly improve it.

Matrix of impact High

Prioritize these
and feasibility use cases
in finance, by use
cases, score 5
10
(illustrative)
15 3
11 6
13 4
7
Impact/
value creation 8
1
12
9

2 14

Low Feasibility/ High

ease of implementation

Cognitive Adaptive Autonomous Data-centric Predictive

decision performance performance investor cost and
making optimization monitoring relations risk control

1 Scenario and 5 Predictive cash flow 8 “One” digital 11 Market value 14 Early fraud
response planning forecasting reporting analytics prevention

2 Augmented revenue 6 Enhanced working- 9 Automated root- 12 Predictive market 15 Cost

forecasting capital management cause analysis sentiment analysis

3 Market and competition 7 Dynamic capital 10 Real-time 13 “Talking point”

monitoring allocation performance creation
tracking
4 Prioritizing growth/
M&A targets

McKinsey & Company

— Synthesis of information, which can create — Digital performance management, which

customizable interactive charts through natural- answers performance-related questions,
language queries. For example, solutions exist synthesizes status and scenarios, identifies
that provide a general Q&A chatbot, a chart drivers and root causes of budget variances, and
creation tool that generates charts seconds suggests resolutions. This solution is typically
after receiving a prompt or description of code, self-serve, business user–friendly (as opposed
and a visualization tool that customizes charts to finance user–friendly), and can lead to more
by using existing code and validating the effective performance management dialogues.
accuracy of the code.

Gen AI: A guide for CFOs 7

 — First drafts of external reporting, which not only The array of gen AI use cases is wide, varied—and
Find more content like this on the
can save weeks of team time in preparing no longer merely theoretical. And while it’s still early
McKinsey Insights App
advanced first drafts of securities filings and days, the rate of adoption is speeding up. Those
stakeholder reports (such as sustainability realities make it even more important for CFOs to
reports) but also runs queries on the current get started in a considered and proactive way.
regulations and standards to help ensure that
the reports meet current standards.

— Working capital management with features Gen AI can be an important tool for value creation.
such as an always-on support bot to help CFOs should strive to be gen AI enablers, not
Scan • Download • Personalize
facilitate collections and payments, and an gatekeepers, and make sure that strategically critical
always-updated customer payment history initiatives rapidly and continually receive necessary
risk assessment, including the capability resources. They should also ensure that they and
to limit customer credit based on real-time their own function quickly climb the gen AI learning
information about customer-specific activity curve. The future is already starting.
and market events.

Ankur Agrawal is a partner in McKinsey’s New York office; Ben Ellencweig is a senior partner in the Stamford,
Connecticut office; Rohit Sood is a senior partner in the Toronto office; and Michele Tam is an expert associate partner
in the Chicago office.

Designed by McKinsey Global Publishing

Copyright © 2023 McKinsey & Company. All rights reserved.

8 Gen AI: A guide for CFOs