Corporate artificial intelligence has consumed billions in spending over the past two years. According to the people who sign the cheques, most of it has yet to prove itself. A survey of chief financial officers found that 83% say most of their organisations’ AI projects have not produced measurable results — a statistic that captures the growing gap between AI enthusiasm in the boardroom and AI returns in the accounts.
How the money was spent
The spending pattern will be familiar to any finance chief. Licences for copilots were rolled out to thousands of employees at once. Hackathons produced demos that never became products. Data platforms were commissioned to fix the messy foundations everyone agreed were the real problem — and are still being built. In case after case, the benefits cited were anecdotes: a team that felt faster, a draft that seemed better. When CFOs asked for the number — hours saved, errors reduced, revenue gained — the trail went cold. Reporting by outlets such as Reuters has tracked the same tension across earnings calls, where AI investment is announced in bold terms and AI revenue is described, if at all, in vague ones.
Why measurement fails
The CFOs surveyed were not describing useless technology so much as unmeasured technology. Most projects launched without baselines: nobody recorded how long the old process took, so the improvement could never be quantified. Others targeted work whose value is genuinely diffuse — drafting, summarising, searching — where gains are real but scattered across thousands of small moments. And some projects automated steps inside workflows whose bottleneck sat elsewhere, speeding a queue that simply formed downstream. Finance leaders quoted in coverage by the Financial Times describe the same conclusion arriving in firm after firm: AI spending is being reclassified from innovation theatre to capital discipline, with stage gates, owners and payback periods.
What the successful minority do differently
The 17% tell a consistent story. Their projects began with a costed problem — invoice processing, claims handling, customer contact — where volume, cycle time and error rates were already measured. They deployed narrowly, integrated with the systems staff actually use, and assigned a business owner whose budget depended on the outcome. Crucially, they counted results in the units finance respects: full-time-equivalent capacity released, days shaved off the close, loss ratios improved. Our section profiles several such programmes.
The reckoning is healthy
The vendors feel the ground move
The survey’s shockwaves are landing in the software industry too. When 83% of finance chiefs cannot find results, pricing power migrates from vendors to buyers. Enterprise AI companies report sales cycles lengthening, proof-of-concept demands hardening, and procurement teams arriving with measurement frameworks of their own. Some vendors are adapting brilliantly — building instrumentation into their products, publishing customer outcome data, and accepting contracts that tie fees to usage or results. Others are discovering that the demo-driven playbook of 2023 no longer survives contact with a CFO’s spreadsheet. The shake-out will not shrink the total AI market; the same surveys show budgets still growing. It will concentrate it — in fewer tools, bought by people who intend to count what they return.
None of this means enterprise AI is a bubble about to burst; it means it is growing up. Technologies from cloud computing to mobile went through the same cycle of indiscriminate spending followed by financial discipline. The CFO survey is best read as the moment the discipline arrived. Budgets will keep flowing — but increasingly to projects that can answer the oldest question in business: what did we get for the money? Follow the enterprise AI story at our homepage.
