Eighty-seven percent of finance leaders say they need to be able to tie AI spending to business results within the next year. Only 22 percent can do it today. That gap, documented in a CloudZero survey of 260 senior finance professionals released on June 24, captures the central tension inside every CFO’s AI conversation right now: the investment is real, the budgets are live, and the proof still is not there.
Boards are losing patience
The pressure is coming from the top. According to the CloudZero survey, 66 percent of corporate boards now condition further AI funding on demonstrated return on investment. Forty-three percent of finance leaders are already being asked to produce a number they cannot produce. This is a new dynamic. A year ago, the dominant question was how fast to adopt AI. Now the dominant question is how to justify what has already been spent.
Adoption itself has accelerated sharply. Fifty-six percent of finance leaders now use AI in some form, double the rate seen in 2023. But depth of use tells a different story. Forty-five percent of finance teams are still running limited pilots. Only 17 percent have integrated AI into their core workflows. The gap between experimentation and production deployment is where most organizations are stuck, and it is exactly the gap that makes ROI hard to measure.
What runaway spending actually looks like
Uber provided the clearest cautionary example of the month. The company burned through its entire 2026 AI budget in four months after encouraging engineers to adopt agentic coding tools without usage limits. It has since capped monthly token spending at $1,500 per employee per tool. The speed of that budget exhaustion is striking, and Uber is not alone. Average reasoning token consumption per enterprise organization climbed roughly 320 times over the past 12 months. Finance teams built their AI budgets before that curve was visible.
The broader pattern is one that the CFO Connect State of AI in Finance 2026 report calls a “cost discovery phase” — organizations are realizing that running AI at scale is structurally different from running traditional software. There are no fixed seat licenses. Costs scale with usage in ways that are difficult to predict and, without governance, nearly impossible to control. ChatGPT leads finance team adoption at 35 percent usage, but specialized finance AI tools still lag, partly because the business cases are harder to build when measurement frameworks do not yet exist.
Where the opportunity is clearest
Gartner predicts that by the end of 2026, 40 percent of business software will include AI capable of completing end-to-end tasks independently. For finance functions, the most concrete applications are fraud detection, loan processing, customer onboarding, and automated reporting. These are high-volume, rules-adjacent workflows where AI can reduce error rates and cycle times in ways that are actually measurable — which is precisely why they are where ROI demonstrations tend to succeed first.
A new professional category is also emerging. The “R-Quant,” or Reasoning-Quant, is a finance professional who specializes in orchestrating AI systems rather than running models directly. It is a sign that the industry is beginning to absorb AI not as a tool to bolt onto existing roles but as something that reshapes what the roles themselves require.
Google made a different kind of move this month, launching a new Google Finance app for Android with AI-powered research tools and “key moments” that explain in plain language why a stock moved. It is consumer-facing rather than institutional, but it points at the same shift: the expectation that financial analysis should be explainable in real time is becoming standard.
The harder problem
The measurement challenge is not purely technical. Part of it is that AI’s most significant near-term impact in finance — faster synthesis of large document sets, fewer manual errors in reporting, quicker scenario modelling — produces value that is genuinely difficult to isolate in a traditional P&L. The organizations getting ahead of this are building AI-specific performance metrics from the start, rather than trying to retrofit existing KPIs after deployment.
The 78 percent of executives who cannot yet tie AI spending to outcomes are not necessarily doing anything wrong. They are, in many cases, running the right experiments. The problem is that boards are asking for accountability before the field has agreed on how accountability should even be measured. That conversation is now unavoidable. For more coverage of AI in finance, visit Mylistingo.
