Uber ran out of its entire 2026 AI budget by April. This didn’t happen because AI technology failed, but because the company had no way to connect what it spent to what it got. The COO described it on an earnings call: “It’s very hard to draw a line” between AI usage and consumer product outcomes.
And with that one sentence, we have the CFO problem of 2026.
We asked finance leaders the same question and got a sharper answer. CloudZero surveyed 260 finance leaders (52% CFOs) in June 2026 and found that only 22% can tie AI spend to business outcomes today, even though 87% say they need to within the year. Boards are already losing patience: 66% now condition further AI funding on proof of return, and 43% of finance leaders say they’re being asked for an ROI number they can’t produce.
The cost of that gap is real. Among finance teams that can’t measure AI’s return, 75% have held back investment and 35% have killed or paused an initiative outright, compared with 38% and 11% for teams that can.
Bain found the same pattern industry-wide. Surveying 951 executives at large enterprises, Bain found the technology worked, but the value didn’t materialize. Most expected AI to cut costs by 10% to 20%; instead, 40% saw reductions of 10% or less, and 44% are already budgeting the next round of AI investment on returns that haven’t materialized.
This is, first and foremost, a financial visibility problem, not necessarily a tech problem or vendor problem. It leads into three questions that every CFO running a real AI deployment should be able to answer right now.
What did we spend?
This isn’t about the aggregate spending on AI, the one you get a big bill for at the end of the month, it’s about the bill breakdown. Which teams, which products, which customers, which workflows consumed the budget, and at what cost per unit of output?
When GitHub moved Copilot to token-based billing in June, one developer saw costs jump from $29 a month to $750 a month overnight. Flat software subscriptions are giving way to consumption-based AI billing. As that occurs, the finance function’s job changes. Without fixed line items you can budget, you’re stuck with a variable cost moving with usage: every prompt, agent call, and workflow run. The month-end bill tells you what was bought, but it doesn’t tell you who consumed it, why, or what it produced.
Without that breakdown, CFOs and finance leaders are managing a total number, not a portfolio. The reality is, this means you can’t separate defensible spend from waste.
Report
Finance needs to prove AI’s return: CloudZero report
260 senior finance leaders (more than half CFOs) told us why the speed of seeing AI spend, not the size of it, separates who pulls ahead on AI from who gets burned.
Was it worth it?
Morgan Stanley asked Meta on its Q1 2026 earnings call what signals the company was watching to ensure it would generate a return on AI capital. But no one answered with a number. Meta is not unusual. Accenture’s CFO, asked about internal AI productivity on the same cycle, gave a non-answer on ROI. This is the pattern across the Fortune 500 right now. Leaders are making confident announcements of AI investment increases, followed by silence when the follow-up question is “what return are you projecting, and how will you measure it?”
Answering whether AI was worth it requires connecting spend to outcomes, not to activity. Token counts, model usage, and acceptance rates on code suggestions are inputs. Revenue protected, customers served, transactions processed, and costs avoided are outcomes. Only 14% of CFOs report clear, measurable AI ROI today (RGP, 2025). The gap is not that AI is not producing value. The gap is that most organizations have no instrument to connect what they spent to what it produced.
Where should I invest next?
If you cannot answer the first two, you have to guess at this one. You are, as in Uber’s case, running a leaderboard for usage without knowing whether that usage is generating margin or consuming it.
The CFOs who separate from the field over the next two years are not the ones who spend more on AI or less. They are the ones who can break their AI portfolio into what is working, what is not, and what the marginal return on the next dollar is. That requires per-outcome economics: cost per customer, cost per transaction, cost per product feature, cost per strategic bet.
Boards are starting to ask these questions in terms that CFOs cannot defer. The accountability phase that follows the experimentation phase has arrived, and the companies that built the financial infrastructure to answer it will grow their advantage rapidly. The ones that did not will spend the next year doing what Uber is doing: rationing.
CloudZero was built to answer all three.
