5 ways CFOs can own the financial side of AI strategy
The following is a guest post from Eric Czervionke, partner at Oliver Wyman. He also leads the firm’s CFO Agenda team. Opinions are the author’s own.
Spend first, ask questions later. That has been the approach to artificial intelligence investing by most large companies thus far.
But now, in year four of the generative AI era, the bill is coming due as corporate boards, along with shareholders, employees and other stakeholders, increase their focus on what these large investments might amount to someday.
The burden for answering such questions is likely to fall not on the technology leaders overseeing AI deployment, but rather on chief financial officers since they are best positioned to defend their organization’s AI spending.
Most haven’t fully embraced their role in governing these investments. Three-quarters say they plan to raise AI spending this year, according to a survey of 494 CFOs worldwide conducted by the New York Stock Exchange and the Oliver Wyman Forum. But only 6% believe that steering those AI investments is the best way they can create enterprise value for their companies.
In my work advising large-company CFOs, I am seeing this contradiction firsthand. CFOs are generally very optimistic about the technology itself; four out of five CFOs rank data, automation, AI and digital capability among their top three priorities for transforming finance, according to the survey. But most are still figuring out how to help their organizations deploy AI at scale: Nearly three-quarters (74%) said their companies are still exploring AI or are only piloting it.
Yet a small group of CFOs is beginning to show what financial accountability for AI looks like. Roughly 14% of the CFOs surveyed work at companies that qualify as AI leaders, meaning they are deploying at scale in at least two different use cases. Across those companies, a picture is emerging of how CFOs can own the financial side of AI strategy.
CFOs who master the following five disciplines will build sustainable competitive advantages through their growing AI investments. Over time, that edge will widen the gap over slower-moving rivals.
1. Separate the builder from the scorekeeper. The CFO governs investment discipline enterprise-wide and is best positioned to judge whether an AI program has created value. An independent finance or benefits-assurance function should validate the baseline, account for full costs and test whether benefits are attributable and durable.
CFOs can also boost their own credibility by submitting their own finance-function AI programs to that same process.
2. Redesign end-to-end processes. One reason AI investments fail to produce meaningful financial returns is that companies automate individual tasks without changing the process around them. A tool might make one employee or function more productive while leaving the economics of the broader workflow largely unchanged. CFOs should push teams to start with the end-to-end business outcome — say, faster order-to-cash, lower service cost, better customer retention or reduced working capital — and redesign the process around that outcome. The full value of AI often requires changing workflows, systems and roles well beyond the technology itself.
3. Build AI capabilities that compound. A pilot proves a model can work once, often on data an analyst assembled by hand. But pilots do not create durable value unless they leave behind infrastructure the next use case can reuse. Data has to arrive on its own, correctly, every morning. Someone has to own it, a control has to catch it when it drifts and a workflow has to actually use it rather than just display it.
The same principle applies to how the company understands value creation. CFOs can’t measure the return from AI if they can’t trace how an operational change ultimately affects the financial results. Companies therefore need a shared map of how the business works: what a change in price does to churn, what churn does to cash and how long that effect takes to show up in the numbers. That map, along with reusable data pipelines, controls and workflows, becomes part of the asset created by the investment. The test for an AI program should not only be whether the first use case pays off, but also whether it makes the second one cheaper, faster and easier to govern.
4. Make sure the company owns and controls the assets it has built. Pilots often move fastest by leaning heavily on a model provider or implementation partner, which can make sense early on but turn costly later. When a company’s data structures, process logic, controls and institutional learning are all embedded in one proprietary stack, switching providers can mean rebuilding much of the transformation from scratch.
CFOs should press the vendor on one question above the rest: If we walked away in two years, what would we take with us? The honest answer usually reveals whether the data, the business rules and the audit trail actually belong to the company or to the platform.
5. Rearchitect roles as you transform. Saving 10% of the time across hundreds of roles carries no guarantee of a 10% cost reduction; if the jobs themselves remain unchanged, the released capacity simply gets absorbed. Leaders need to redesign roles and staffing models around the new way of working — and be honest about where the savings are supposed to show up. In their governance role, CFOs ensure this job-architecture work, done with human resources partnership, is embedded into major AI transformations.
This redesign also has to account for what’s lost as routine junior work changes. Sixty-four percent of CFOs surveyed by NYSE and the Oliver Wyman Forum expect to shift away from junior roles, while 70% plan to intensify succession planning. That means deliberately rebuilding the apprenticeship that routine work used to provide. Without it, AI deployments might produce efficient outputs but fail to develop the leaders needed to interpret results and recognize when they’re wrong.
