Finance chiefs tighten grip on AI spending as ROI remains elusive.
AI budgets overshoot, confidence stays high.
A new quarterly survey of 2,000 finance chiefs highlights a growing mismatch between AI spending and its visible payoffs. In the Q3 Pigment CFO Index, which tracks CFO sentiment in the US and Europe, 83% of companies report going over budget on AI projects this year, with spending overruns typically in the 11% to 25% range ([1]). Notably, however, those same CFOs remain bullish that the investments will eventually deliver value: on average, they rate their return-on-investment “confidence” at 8.2 out of 10 ([2]). In other words, most finance leaders are willingly tolerating cost overruns for now in the belief that the payoff will come.
This paradox is forcing CFOs to pay closer attention to the economics of AI in their organisations. Bloomberg reports that many CFOs are increasingly alarmed by “tokenised” AI usage fees piling up each month ([3]), especially as employees experiment with generative AI tools. The need to curb what one analyst dubbed “tokenmaxxing”, racking up large cloud AI bills without commensurate results, is turning finance chiefs into enforcers of AI cost discipline. The current priority, CFOs say, is to get a tighter grip on where AI spend is going and ensure it translates into real-world improvements.
ROI or bust: proof demanded for AI projects.
Facing investor pressure and uncertain returns, CFOs are increasingly unwilling to fund artificial intelligence for its own sake. Deloitte data indicates that a strong majority of finance chiefs will no longer approve new AI investments without hard evidence of tangible returns, and many are re-evaluating existing projects that failed to deliver clear results. In practice, this often means renegotiating or winding down initiatives that cannot demonstrate measurable business impact. According to KPMG’s latest Global AI Pulse survey, nearly half (49%) of senior executives have already paused or scaled back deployments of AI “agents” because their operating costs outweighed the benefits ([1]).
The turn from enthusiasm to accountability is a sign of AI’s maturation, and a recognition that genuine ROI is still the exception rather than the rule. As one KPMG analysis notes, companies are shifting from a focus on launching AI pilots to a focus on “AI economics and value”, yet “while many organisations are realising benefits, established ROI remains limited” ([2]). CFOs now routinely ask their teams to define success metrics up front and to monitor outcomes closely. Projects that once had “blank cheque” funding are being asked to justify their existence with data. The message from the finance suite is clear: in 2026, AI initiatives must prove how they move the needle on revenue, costs or efficiency, or risk being shelved.
Where returns are finally showing up.
There are signs that disciplined execution can unlock real financial gains from AI. A new global study by IBM and Oxford Economics, published at the end of September, finds that companies with “AI-first” CFOs, those who take the lead in driving AI strategy and governance, achieved revenue growth 23% higher than their peers from 2022 to 2024 ([1]). These AI-engaged finance chiefs are not only committing budget to technology, but also ensuring that each initiative has a clear business case and measurable targets. In short, the CFO’s active involvement appears to be a key differentiator in realising AI’s promised returns.
Concrete examples of ROI do exist, especially where a single business owner has pursued a well-defined efficiency goal. For instance, a global insurance CFO recently reported that AI-driven automation in their call centres reduced call volumes by 20% and cut annual costs by over $80 million ([2]). Such success stories remain relatively rare, often coming from targeted use cases in operations, customer service or supply chain, but they illustrate how the returns on AI are attainable. The common thread is a rigorous approach: setting a baseline for performance, focusing AI on a specific metric, and holding someone accountable for delivering the outcome. As more companies adopt this disciplined model, the examples of real ROI from AI are likely to multiply.