Quick Answer
Google Cloud's ROI of AI 2026 report found that 84% of 2,403 executives see AI returns rising, while CloudZero's survey of 252 finance leaders found that 61% can't justify what they already spend on AI. The gap is allocation. Most finance teams can't break AI spend below a single bill line, so they can't show which part of it is paying off.
Google Cloud is back in inboxes this month just as FY2027 budgets lock, with its third annual ROI of AI report, first published in July.
In it, 84% of the 2,403 executives surveyed say financial returns from AI are increasing. 86% say AI is driving cost-efficient growth, scaling revenue or output without a matching rise in operating costs.
Now run those numbers past your own close.
CloudZero ran that test in August and September. It asked 252 finance leaders, 79% of them CFOs, a narrower question: can you justify what you already spend on AI? 61% said no.
Both surveys can be right. Google’s respondents span CXOs, VPs, and heads of business units, the people who launch AI programs and watch the work speed up. Our own respondents are the people who get the invoice. Different chair, different answer.
A quick note on methodology, since the two surveys asked different people: Google Cloud and National Research Group surveyed 2,403 senior executives across industries and regions. CloudZero commissioned Centiment to survey 252 US finance leaders who own or influence the AI budget, none of them CloudZero customers. Both surveys are self-reported.
Everyone is searching for ROI. They’re finding efficiency.
Google’s own data shows where attention has gone. Searches for “AI ROI” rose 1,400% in the 12 months to August 2025, and searches for “token efficiency” climbed 335% in the first half of 2026. Everyone wants the cheaper token.
They’re getting it. In CloudZero’s inference telemetry, the average effective AI token cost fell from $1.03 per million tokens in July to $0.69 in October, down a third in three months. The median company’s AI spend grew every single month over the same stretch, according to the November AI Economics Pulse.
Call it the Jevons paradox, applied to AI. A cheaper token becomes headroom, and headroom becomes usage. Efficiency is a unit-price story. ROI is a total-cost story. Right now the two are drifting apart, and your FY2027 plan has to pick which story to follow.
2026 State of AI Spend Report
Why Finance Caps What It Can’t See
Learn how finance leaders are managing AI spend, and how they’re approaching FY2027
Agents pay, and agents spend
That usage has to come from somewhere, and more and more of it comes from agents. Google’s headline finding on agentic AI ROI is strong: 94% of respondents say AI agents contribute to both cost savings and revenue. That’s fair. Agents are also the hungriest token consumers you’ll ever put to work, and they can run around the clock even without your overt permission.
You can see that spend piling up in the AI Economics Pulse, which tracks 430 CloudZero customer companies. Of those companies, 9.8% now spend at least a quarter of their cloud bill on AI, more than four times the share a year ago. And at the median company, 82% of classified spend runs production workloads. Production spend is cost of revenue, so when an agent graduates from pilot to product, its bill lands in gross margin.
“Cost-efficient growth” is a claim about margin, so it’s worth checking against the margin. Google’s own respondents rank margin expansion last of nine outcomes AI is measurably advancing, at 31%.
The missing denominator in AI ROI
Google’s 26% of what it calls “AI ROI Leaders” share some sensible habits, starting with clear ownership of AI decisions and AI built into core business processes. That’s solid advice. None of it requires knowing what the AI cost. Neither does the way Google’s respondents keep score. Only 21% measure AI success primarily by direct financial outcomes, and just 12% use a formal value-tracking framework.
A return is a ratio, and most companies are reporting only the value side of it:
The denominator, what the AI cost, is where finance gets stuck. In CloudZero’s survey, 54% of finance teams can see AI spend only as a single bill line or a rough category split. That’s an allocation problem. Allocation is the ability to break AI spend down by team, product, or customer. And the office that owns the AI budget most often sees it least clearly: 65% of CFO-led AI programs sit in the two weakest allocation tiers, against 43% of engineering-led ones. Finance can own the number, but it can’t tag a workload it doesn’t run.
Self-reported ROI on its own is another weak spot. In the same CloudZero data, companies that claim board-defensible ROI but can’t allocate the spend behind it report the same board-proof gap as companies claiming no ROI at all, 56% against 54%.
A single ROI number won’t satisfy a board that wants to know which part of the AI investment is paying off. That question only becomes more crucial as AI takes a bigger slice of the budget.
When the number can’t be defended, finance adds friction
You might expect finance to pull the plug on AI. It isn’t. 87% of the finance leaders in CloudZero’s survey are raising FY2027 AI budgets, in line with the 97% of Google’s executives who plan to invest more. Finance has a real say in how that money gets spent: in Deloitte’s third-quarter CFO Signals survey, nearly 80% of CFOs said finance has a formal role in AI decisions, and 41% are required approvers. What changes is how hard that money is to draw down.
In CloudZero’s survey, 27% name new approval gates as their primary FY2027 budgeting change, ahead of the 18% tying funding to demonstrated ROI. Gates are popular because they work without visibility. It’s easy to cap spend you can’t see.
But the cost shows up elsewhere. Gate-led teams killed or paused AI initiatives at 20%, against 7% for teams funding on ROI, and delayed hiring at 33% against 19% of everyone else.
Two kinds of leaders
23% of CloudZero’s sample can allocate AI spend systematically and prove what it returned. They aren’t spending less: 41% plan to grow AI spend by more than 20% in FY2027, against 38% of everyone else.
The difference shows up when that 23% asks for money. Only 3% of that cohort failed to secure an AI investment, against 19% of everyone else, and they absorbed roughly half the consequences.
Google found its leaders by asking who sees returns accelerating. We found ours by asking who can show their work. You want to be in both groups, and only the second is within finance’s control.
Before the plan goes to the board
Budgets lock in the next few weeks. A few questions worth putting to your team first:
- Can we break the AI line down by team, product, or customer, or only by vendor?
- Did the plan book the falling token price as savings, or did it model usage?
- Which AI spend runs in production, and does the margin forecast show it?
- If we’re adding approval gates, what would let us remove them in FY2028?
Google is right that ROI is now the question. Search volume says so, and so does your board. The companies that answer it well will be the ones who can point at a line on the bill and say what it bought.
See how your AI allocation compares with 252 finance leaders using the AI Allocation Benchmark, or download the full report, Why Finance Caps What It Can’t See.
Frequently asked questions about AI ROI
