Checking AI output takes 26% of the finance work week in a Datarails survey of 270 CFOs

For the tools to come out ahead on labour alone, they have to save each finance worker more than 1.3 days a week of manual work ^[19]. That is what 26% of a five-day week comes to. The Datarails survey reports CFOs' own time separately: 96% spend at least a tenth of their hours verifying or correcting finance AI output, and 8% spend more than half ^[3][4].

About 86 of the 270 respondents overshot their AI budgets by 10% or more in the past year ^[20], and the plan for the next 12 months adds licenses while headcount holds ^[6][12]. "Fears of job losses have been largely allayed, but the challenge of AI output verification is critical," said Didi Gurfinkel, chief executive and co-founder of Datarails ^[13]. Finance chiefs are buying before the checking burden falls. And 13% of respondents report no clear owner for the AI budget at all ^[17].

Where the 26% goes depends on what causes it. If the errors come from data, the share should fall as firms consolidate. Only 4% have a single source of truth, and 23% still rely on disconnected systems and manual reconciliation ^[14]. Among teams whose main problems are manual reporting and consolidation, 86% report confident answers built on wrong data, against 65% of all respondents ^[15]. A model problem would keep the share high: 56% of CFOs have seen team members get materially different outputs from the same prompt and the same data ^[8]. In the third case, pressure outruns readiness and spending climbs whatever happens to the checking. On that, 76% of CFOs report high or very high pressure to implement AI, and 7% call their function fully ready ^[16].

The company behind the survey sells the first remedy. Datarails describes itself as an AI finance operating system for the office of the CFO ^[18]. Its report says successive LLM responses frequently differ unless the model is restricted to governed, consolidated, contextualized data ^[9]. Its survey also finds 32% of finance leaders prioritizing a finance operating system, second only to planning and FP&A tools at 42% ^[11]. "Finance teams have stopped asking 'will we use AI?' and started asking 'how am I going to most effectively check its work?'" Gurfinkel said ^[7].

I'd expect the checking share to stay high on board reports and the close whichever cause dominates, because almost nobody will accept those without review. Only 5% of respondents trust AI to produce board-ready reports without human review, and 4%, about 11 people, trust it with the month-end close ^[10][21]. The counter-case is the 86-against-65 gap. If bad data drives that much of the error, consolidation should let firms check less even there. A follow-up survey in which the verification share falls while license spend keeps rising would break this view, and a flat or higher share among firms that consolidated would support it.