Corporate spending on artificial intelligence has reached record levels, yet the people who sign the cheques are increasingly unconvinced. In a widely cited survey of finance chiefs, 83 percent of CFOs said most of their AI projects have not produced measurable results. For a technology that was supposed to transform productivity, that is a sobering verdict from the executives best placed to judge value for money.
The finding does not mean CFOs have turned against AI. Most continue to fund it, and many describe it as essential to their company’s future. What the number captures is a measurement problem: money is flowing out on licences, cloud computing, consultants and data projects, while the financial returns are proving remarkably hard to pin down. As we report regularly at Weekly Business Records, the gap between AI enthusiasm and AI evidence has become the defining finance story of the year.
Why the Returns Are So Hard to See
The first problem is that many AI projects were approved without a baseline. Companies rushed to deploy chatbots, copilots and forecasting tools in 2024 and 2025, often under board pressure not to be left behind. When finance teams later asked what had improved, there was frequently no “before” picture to compare against. Time saved by employees is real, but unless it converts into fewer hires, faster close cycles or higher output, it never reaches the profit and loss account.
The second problem is cost creep. Pilot projects look cheap; production systems are not. Data cleaning, integration with legacy systems, security reviews and model monitoring all add recurring expense. Several CFOs have described discovering that the true cost of an AI tool was three or four times the licence fee once the supporting work was counted. Meanwhile the benefits, where they exist, are scattered across departments in small increments that no single budget line captures.
There is also a timing issue. AI returns tend to arrive in years, while costs arrive in quarters. A finance chief reporting to the board every three months is naturally sceptical of a project whose payback sits permanently on the horizon. Research from firms such as McKinsey consistently shows that only a minority of companies have redesigned workflows around AI, which is precisely the step that turns experimentation into earnings.
What Separates the Successes
The minority of companies reporting clear results share recognisable habits. They started with a small number of use cases tied to a specific financial metric: days sales outstanding, cost per invoice, forecast accuracy, fraud losses. They assigned a business owner, not just a technology owner, and they measured relentlessly from day one. Crucially, they redesigned the process around the tool instead of bolting the tool onto the old process.
Finance functions themselves offer some of the clearest wins. Automated invoice matching, anomaly detection in expenses, and AI-assisted cash-flow forecasting have produced documented savings because the baseline metrics already existed. Where CFOs control the use case directly, the results are far more likely to be measurable. It is the diffuse, enterprise-wide deployments — a copilot for everyone, a chatbot for everything — that struggle to justify their budgets.
How CFOs Are Responding
The response is discipline rather than retreat. Finance chiefs are demanding stage gates: pilots must hit agreed metrics before funding scales. AI spending is being moved out of innovation budgets, where scrutiny was light, and into operating budgets, where it competes with every other priority. Some companies now require every AI proposal to name the cost it removes or the revenue it adds, in writing, before approval.
Vendors are feeling the change. Renewal conversations that were once formalities now involve hard questions about usage rates and realised savings. Tools bought per-seat but used by a fraction of staff are being cut or renegotiated. Analysts at Gartner have noted a similar pattern across enterprise software: the era of buying on promise is giving way to buying on proof.
The Bottom Line
The 83 percent figure should not be read as AI failing. It should be read as finance doing its job. Technologies that genuinely raise productivity eventually show up in the numbers, and CFOs are simply insisting that AI be held to the same standard as any other capital allocation. For readers planning their own budgets, our guide to smart financial moves makes the same point at household level: measure, prioritise, and cut what cannot prove its worth. Companies that apply that discipline to AI now are the ones most likely to report real results when the next survey lands.
