CFOs say AI is paying off. Read the fine print on 'expectations.'

Finance chiefs are pleased with AI so far. In Grant Thornton's Q3 2026 CFO Survey, released September 24, 84% of the nearly 230 U.S. finance leaders surveyed said AI return on investment is meeting or exceeding expectations. Almost two-thirds, 65%, rated the performance and quality of the technology good or excellent. Just 2% rated it poor or very poor.

That's a real data point from the people who sign off on spend. It's also a measure of satisfaction, and satisfaction depends on where expectations were set.

Where the value is showing up

According to the release, most of the value so far has come from productivity. Only 33% of respondents listed higher revenue as an AI benefit. Finance and accounting was named the top function for AI-driven transformation (39%), followed by customer service (36%) and cybersecurity (30%). Supply chain barely registered at 6%.

Sumeet Mahajan, a partner in Grant Thornton's AI, Data & Tech practice, said CFOs are "using a broader lens of what value is," which helps explain the high marks. A broader lens can be reasonable. It also makes it easier to count soft gains as returns.

Two other findings are worth noting. Only 24% of finance leaders said their organization might have layoffs in the next six months, the lowest figure since Grant Thornton started asking in 2022. The survey doesn't tie that figure to AI, but it shows little sign of near-term layoffs in this sample. And the top obstacles respondents named weren't about technology. They were competing priorities (43%) and budget constraints (43%).

How to read "meets expectations"

A satisfaction score answers a different question than an ROI calculation. Before you borrow this 84% for an internal memo, keep three limits in mind:

• It's relative. "Meeting expectations" says nothing about the size of the return, only that it matched what people expected. If expectations were modest or vague, the bar was low.
• The sample isn't described in detail. The release says "nearly 230 U.S. finance leaders" but doesn't give company size or industry mix for this quarter. Check the full report before assuming it reflects a 100-person firm.
• Consider the source. Grant Thornton sells AI and transformation advisory work. The survey is standard in its field, but the firm benefits from the "AI is working" story.

The same day, a vendor-commissioned survey of large-company CIOs from Dataiku found that only 28% consistently measure both operational and business outcomes across their AI agents. The two surveys asked different people different questions. Read together, they fit a familiar pattern: leaders feel good about AI, and far fewer can show the numbers behind the feeling.

The operator version

If you're a CFO or owner at a smaller company, turn "meets expectations" into something you can check. For each AI tool you pay for, write down what you expected before you bought it. That might be hours saved per week, cost per ticket, or days to close the books. Then write down what actually happened. If you can't fill in the first column, that's why your own tools feel like they're meeting expectations.