The pitch for AI in accounting has always been about time. Automate the grind, get your week back, spend the hours you save on work that matters. A new IDC white paper, sponsored by Sage and based on a survey of more than 2,000 senior finance leaders, puts an uncomfortable number on how much of that week you actually get back. Nearly half say they spend more than 15 hours a week verifying what the AI produced. One in five spend more than 30, most of a working week, just checking the machine. Sage has a name for it: the verification tax.

The uncomfortable part is how much of the promised gain the tax eats. More than a quarter of the finance leaders surveyed said that reverse-engineering AI output, so they can explain and defend it to someone else, swallows over a quarter of the productivity the tool was supposed to deliver. About one in five said it eats more than half. That is the productivity story turned inside out: the time AI saves on the doing is handed back, in large part, on the checking. The work did not vanish. It moved.

None of this should surprise anyone who watched the last month, when KPMG and EY each pulled reports that AI had helped write and then filled with citations that did not exist, and Deloitte refunded a government client over the same problem. When the output can be confidently, fluently wrong, a competent professional checks it, and checking is not free.

Two things are true at once, and the honest read holds both. Part of this is a maturity effect. New tools, deployed without much governance, get checked the hardest, and the ratio should improve as controls tighten and trust is earned. It already varies enormously between a firm that rolled AI out with rules and one that just switched it on. But part of the tax is permanent. Work that carries your signature will always need a human who can catch what the model missed, and that floor does not fall to zero because a vendor wishes it would.

Notice, too, who is putting this number in front of you. Sage sells accounting software, AI features included, and it commissioned this research and is promoting it anyway. When the company with every incentive to advertise effortless productivity instead publishes data showing the savings are half-eaten by checking, that is not marketing. It is closer to a confession, and it deserves to be taken at face value.

So the move is to stop booking AI as pure time saved and start booking the verification alongside it. If your team is spending 15 hours a week checking AI output, that is not embarrassing overhead to bury. It is the judgment work that has quietly become the job, and as we argued when AI came for the billable hour, it is exactly what a firm should be pricing and staffing for rather than cutting. The market already sees it: the roles firms are opening now lean toward risk, governance, and judgment over raw technical speed. A firm that treats the reviewer as a cost to automate away is deleting the one person keeping its AI honest.

AI does save time in accounting. It just hands you a bill for the trust you used to get for free. Read the tax, plan for it, and charge for it. The hours your people spend checking the machine are not the machine failing. They are the part of the job the machine cannot do.

— Footnote

Footnote is an independent publication, with no affiliate links and no vendor paying for placement. It is not professional accounting, tax, or legal advice. Survey figures are from an IDC white paper commissioned by Sage, “The Verification Tax: The Emerging Economics of AI in Finance” (a survey of 2,275 senior finance decision-makers across North America and EMEA), as reported by Accounting Today, diginomica, and other coverage; Sage is an accounting-software vendor, noted in the text. Details are current as of July 2026.