CFOs Say AI Is Paying Off, but Only 48% Can Point to Financial Gains

Finance leaders are telling surveyors that AI is working. They are also telling a different set of surveyors that they cannot prove it. Both statements come from data published within days of each other in late September, and both can be true at once, because the word "return" is carrying two very different meanings. Hours saved are easy to count and easy to believe. Dollars that reach the income statement are harder to trace, and that is the kind of evidence a board or an audit committee eventually asks for. The gap between the two is now the most useful number in finance AI.

The Confidence Is Real, and So Is the Caveat

Grant Thornton's Q3 2026 CFO Survey, which polled nearly 230 U.S. finance leaders and was published September 24, found 84% saying AI return on investment is meeting or exceeding expectations. Another 65% rated AI technology performance as good or excellent, and only 2% rated it poor or very poor. The same survey found 80% of leaders expect net profits to grow over the next 12 months, an all-time high across the 18 quarters the question has been asked. The firm's release quotes Sumeet Mahajan as saying CFOs are using a broader lens of what value is, so they are seeing those returns.

That broader lens shows up in the detail. Finance and accounting led the list of areas being transformed by AI at 39%, ahead of customer service at 36% and cybersecurity at 30%. But only 33% of respondents reported enhanced revenue as an AI benefit. Most of the value being claimed is efficiency, which is real and worth having, yet it is a different claim from saying AI is growing the business. A CFO who reports that AI is meeting expectations may simply mean that expectations were set at the level of time saved.

Productivity Is Easy to Count. Cash Is Not.

The sharper evidence comes from Esker's 2026 Global Finance AI Trust Index, released the same day. The company surveyed 338 finance leaders in August, 137 of them CFOs, across the U.S., U.K., E.U., Australia and Canada. Among them, 72% exceeded their AI spending plans. Yet 58% report productivity improvements and only 48% cite direct financial gains. More than half of respondents are spending past budget on a technology that fewer than half can connect to money earned or saved.

CFOs are the most candid about it. The Esker data shows 65% of CFOs struggle to connect AI usage to outcomes, against 40% of other finance leaders. The barriers they name are unglamorous: 48% cite data quality, and 42% report insufficient system integration. Among CFOs specifically, 58% point to integration, against 31% of other leaders. Nobody is saying the models fail. They are saying the plumbing underneath makes results impossible to attribute cleanly.

There is also a governance shadow. Esker found 66% of finance leaders know or suspect that unapproved AI tools are in use. Spending that exceeds plan, outcomes that cannot be traced and tools nobody sanctioned make a difficult combination to defend. "Finance leaders are wired to ask what could go wrong, what the return will be, and what controls need to be in place before making an investment," said Scott McDermott, CFO at Esker. The survey suggests many are now asking those questions after the money is spent.

Trust Is Being Rationed, and the C-Suite Is Not Aligned

Finance is not handing AI the keys. In the Esker survey, 32% use AI to recommend actions that teams then carry out manually, and 29% let AI act with human approval. On the highest-stakes decisions, CFOs draw a hard line: 78% say AI should not set revenue targets, compared with 11% of other finance leaders, and 74% oppose AI controlling hiring budgets, compared with 17%. The people accountable for the numbers are far more cautious than the people who use them.

A smaller study points to the same tension from the top of the organization. The Open Future Forum's CFO AI Leverage Report, based on 290 respondents between March and August, found 70% of CEO-seat respondents expect AI to pay back within six months, against 42% of the 52 finance-seat respondents. That is a 28-point optimism gap, and the finance sample is small enough that it should be read as directional. Still, it matches the pattern: the person who funds AI is more hopeful than the person who has to reconcile it.

The Playbook: Turn Claimed Returns Into Booked Ones

• Define the return before the rollout: For every AI project, name the line item it should move, whether cost of close, days sales outstanding or audit hours, and record a baseline before launch so the result can be verified.
• Separate time saved from money saved: Track productivity gains honestly, then decide what happens to the hours. Redeployed capacity, avoided hires and reduced overtime are each a different kind of return.
• Fix the data before funding the next tool: With 48% citing data quality and 42% citing integration, a cleaner source layer will do more for measurable returns than another license.
• Inventory the AI you already have: If two in three leaders suspect unapproved tools, start with a register of what is running, who approved it and what data it touches.
• Set decision rights by stakes: Let AI recommend broadly, act with approval on routine work, and keep targets and budgets with people until auditability is proven.
• Align the CEO and CFO on payback: Agree on the timeline and the evidence standard together, so a six-month promise does not meet a twelve-month reality at the board table.