2026 Report

Finding the ROI of AI: The Finance Perspective

A new CloudZero survey of 260 finance executives (135 of them CFOs) captures a marked shift in how AI gets funded. The blank-check era is coming to an end, and finance must now prove the ROI of AI or face consequences.

The top takeaway: 87% say they need to tie AI spend to business outcomes within the year. Only 22% can do that today.

Table Of Contents

Executive summary
Part 1: Seeing the spend
Part 2: Measuring the return
Conclusion: The common thread, and what to do

Executive summary

A new survey of 260 senior finance leaders, 135 of them CFOs, asked the same question their boards are asking them: can you prove your AI spend is paying off?

For most, the honest answer is no, and that “no” leaves the CFO caught in a trap.

We call this trap the AI paradox. Like it or not, companies must spend more on AI to stay competitive, but the spend itself has become ungovernable. Finance can see the activity, tokens consumed, models called, the monthly bill, but can’t connect it to the customers, products, and transactions that money was intended to serve.

In other words, the AI era isn’t punishing high (or even low) spend. It punishes blind spend. And right now, most finance teams are spending blind.

First, almost no one can prove the return. Only about one in five (22%) can tie AI spend to outcomes. The other 78% are working with a gap, and for more than half of all finance leaders (54%), it’s serious.

Three in five admit they’re already spending more than they can justify. The capability they want most is the one almost none of them have: 64% say tying spend to outcomes would fundamentally change how they invest, and 87% say they have to close that gap within the year.

Second, funding now depends on proof of return, and the CFO is the one who has to produce it. The blank-check era is ending. Only 26% of boards still say “invest now and sort out the returns later,” while two-thirds (66%) condition further AI funding on proof of return. 22% will approve nothing new until ROI shows, and 43% of finance leaders are already being asked for a number they can’t give. When the board asks whether the money was worth it, finance is alone in the hot seat.

Third, and this is the part that keeps CFOs up at night: the gap is compounding into a competitive threat. When the outcomes can’t be proven, the money stops. Among finance leaders who can’t show the ROI of their AI, three in four (75%) have held back investment and one in three (35%) have killed or paused an initiative. For those who can prove it, those numbers drop to 38% and 11%.

The companies that solve the AI ROI equation are funding what works while everyone else stalls. That gap is also widening every quarter. The stakes are higher than a CFO’s job. The company’s own competitiveness is on the line.

The way out is to prove what the spend returns. The finance leaders who can will invest with confidence. Those who can’t are one board meeting from being exposed. Every quarter of delay means faster rivals pull further out of reach.

Now is the time to move, not after the next board meeting. Every quarter of delay lets faster rivals pull further out of reach.

Key takeaways:

Few can prove AI’s return. Only 22% of finance leaders can tie AI spend to outcomes. The other 78% are flying blind.
Blind spend is the real exposure. 60% of finance leaders admit they’re already spending more on AI than they can justify.
The blank check is expiring. Just 26% of boards still say “spend now, sort it out later.” 66% now condition AI funding on proof.
87% need to prove AI ROI within the year. Plus, the board is asking for this.
Spend discipline is overrated, speed of visibility is underrated. Same-day teams invest aggressively at 2x the rate; among over-budget teams, seeing spend within a day cuts serious consequences from 97% to 64%.
Lack of AI ROI measurement is a real business blocker. Three quarters (75%) of teams that can’t measure AI outcomes have held back investment, and 35% have killed an initiative outright. That drops to 38% and 11% for teams that can.
Finance answers for spend it doesn’t control. AI is driven outside finance in 74% of companies, but finance owns the bill in 60%.

Part 1

Seeing the spend

AI raises two questions for finance: what are we spending, and is it paying off? This part is about the first. Knowing what you’re spending sounds like the easy one, and it’s the simpler of the two, but most teams still can’t see it clearly or fast enough to act. Before you can prove what AI is worth, you first have to see what it costs.

Finding 1

Going over budget isn’t the problem. Justifying the spend is.

When the board comes asking, finance is often the one in the hot seat having to answer for AI spend. But we found that whether a team blew its AI budget says almost nothing about whether the CFO gets burned for it.

Let’s start with the thing everyone assumes is the problem: the overrun. Companies that blew their AI budget hit a serious consequence (board pushback, a spend cap, a cancelled initiative) 80% of the time. Companies that stayed on or under budget: 78%. So, hitting the number (or not) makes little difference.

If anything, coming in under budget made things worse: those teams faced more board ROI pressure than the ones who overspent, 58% to 48%.

Underspending reads as falling behind, and overspending reads as recklessness, so the board isn’t really asking whether you spent too much. It’s asking whether you spent the right amount, and whether you can prove it. And over- and under-spenders alike can’t.

Either way, the lesson is clear: finance is currently unable to answer what the spend bought. That silence can be damaging when it’s a board question.

Finding 2

The people spending on AI aren’t the ones tracking it.

AI is driven by engineering, the CEO, and the business; finance is left to account for the bill. This has two consequences.

The overrun isn’t what hurts you, but it does reveal where the problem might start. AI spend is run up by people who don’t report to finance, and the overrun pattern makes that distance visible.

Across all companies, 32% blow their AI budget by more than 20%. When the CFO drives AI, that falls to 15%. When the CEO drives it, it more than doubles to 37%, and when engineering drives it, 41%.

The further AI sits from finance, the worse the miss. The overrun is a distance gauge; it measures how far AI has drifted out of finance’s line of sight.

And the second consequence: those running up spend usually aren’t the ones tracking it. Finance is the primary driver of AI spend in just 18% of companies, yet owns the spend tracking in 60%. In roughly three in four, a function other than finance leads the spending, so the people generating it have no native reason to report it on finance’s clock. Finance ends up tracking a bill someone else is running up.

But the overrun was never the real problem; the lag behind it is. When the spender and the tracker work from different data, that lag is built in, and as Finding 3 shows, speed is what separates who gets burned. Closing that lag isn’t a reorg, but rather, it’s giving everyone who touches AI spend (finance, engineering, and the business alike) the same live view into what’s being spent and what it’s producing.

Finding 3

Same-day spend data means finance is more aggressive.

The speed of spend visibility marks the line between who gets burned and who keeps investing in AI. Fast movers, particularly, bet bigger when they can see their spend data in real time.

Let’s start with what fast data prevents: freezing. Teams whose AI spend data reaches them within a day are far less likely to hold back AI investment for want of data, 37% vs. 57%. It holds across company sizes and AI-intensity levels, and it has nothing to do with whether they’re over budget; latency and overruns move independently.

And speed doesn’t just keep them investing; it keeps them out of trouble. Take the teams that went over budget: those who could see the spend within a day hit a serious consequence 64% of the time, against 97% for those waiting on the bill. The overruns were the same; what differed was whether finance saw the spend quickly enough to act rather than reading it off an invoice later.

Plus, the same-day group is more than 2x as likely to hold an “invest aggressively” posture (37% vs. 15%) and 4x as likely to plan 50%+ AI spend growth (28% vs. 7%).

Two things explain it. They trust their numbers: 78% rate their spend data decision-grade, against 48% of slower teams. And they spend boldly with eyes open: the finance leaders most confident in their data are the most likely to be over budget (63% vs. 40%), not because they’ve lost control, but because they can see the spend as it happens and choose to push. It’s easier to spend more when you can watch exactly what you’re spending.

The result is a widening split. The same-day group can floor it, while the slower ones slow down or freeze.

Part 2

Measuring the return

Part 1 was about seeing what you spend. Part 2 is about proving what it’s worth, and that is where almost everyone is still stuck. Knowing what you’re spending is not the same as knowing what it returns, and the second is the harder, mostly unsolved problem. Proving the return, not just the spend, is what now opens the next round of investment.

The teams leaning in hardest can already see their spend in real time. Among the “invest aggressively” boards, 71% have same-day spend data, well above the 50% average.

But real-time spend data answers only one question: what am I spending right now? It says nothing about what that spend is worth, or where it’s headed next quarter.

In other words, they can see what they’re spending, but can’t yet answer what it’s earning.

Finding 4

Finance wants outcome-based ROI but mostly can’t measure it.

Finance wants this more than almost any capability in the survey, and almost no one has it. 64% say tying AI spend to the outcomes the business is accountable for would fundamentally change how they invest, and only 16% disagree.

Nearly nine out of 10 (87%) need outcome measurement within the year. Just 4% call it “not a priority”. Yet only 22% can do it cleanly today; the other 78% are working with a gap in that capability.

That gap stops real decisions cold. Two out of five (40%) finance leaders say the ability to measure outcomes would change how they invest, yet lack that ability today. This means investment is put on hold while waiting on a capability they don’t have.

Finance knows the bet it wants to make, but can’t get the number it needs to justify making it.

Finding 5

The measurement gap is everywhere, and finance wants many lenses.

It isn’t one missing capability. The survey tested three, and all three are about equally broken today: unit-level spend allocation (49% meaningful-or-worse gap), forecasting (48%), and outcome measurement (54%). Put together, 64% have a meaningful-or-worse gap in at least one of the three, and 35% are weak in all three.

And teams know exactly where they’re behind. Asked where they need to be in 12 months, the urgency mirrors the gap almost one-for-one: unit-level allocation 50% urgent or critical, forecasting 53%, outcome measurement 55%, with 80% to 87% rating each one needed or more. The capability they’re weakest at today, outcome measurement, is the one they most want to fix.

The gap runs wide as well as deep. Asked which unit-cost metric matters most to them, finance leaders didn’t pick just one. They wanted several: cost per dollar of revenue led at 54%, with per employee hour (49%), per customer interaction (46%), per transaction (45%), and per feature per user (40%) close behind.

Only 31% chose a single metric while nearly half (48%) picked three or more metrics. No metric dominates because no business model does. What finance is asking for here is the ability to take one dollar of AI spend and break it out by whatever its business runs on, and then send the next dollar to the unit that’s returning the most.

Finding 6

The teams that can’t measure AI outcomes are hurting the most.

The gap is expensive. Teams with a serious-or-severe outcome-measurement gap are worse off on every count: 75% held back investment (vs. 38%), 71% face board ROI pressure (vs. 34%), 71% call AI spend a top stressor (vs. 38%), and 35% killed or paused an initiative (vs. 11%).

A serious forecasting gap hits just as hard as the outcome gap: 77% held back and 76% face board ROI pressure. And a serious allocation gap is harder still on investment, with 87% holding back.

Again, the same things follow: investment freezes, initiatives get killed, the board turns up the pressure, and the people who run finance carry the stress home.

In Part 1, we traced harm to slow data. Here, it’s whether finance teams can measure what the spend returned. Holding back investment is the clearest overlap: Part 1 tied the freeze to latency (37% vs. 57%); the outcome-measurement gap drives it harder still. Not seeing the spend fast and not being able to prove its return are two routes to the same place.

Finding 7

The biggest AI adopters most want better measurement.

Among the companies furthest down the AI path, the appetite for measurement and the appetite for spending move together. Among companies where AI already makes up more than a third of their total infrastructure spend, the teams calling outcome measurement urgent or critical are 5x more likely to hold an “invest aggressively” posture (66% vs. 13%). Among lighter adopters, wanting it says nothing (3% vs. 10%).

This isn’t just AI intensity: even among companies at the same AI share and size, urgently wanting outcome measurement still tracks with an aggressive posture.

And they want it for a hard-nosed reason: measurement is what lets them keep spending. Without it, even the heaviest adopters get forced to pull back. Among companies where AI is more than a third of their infrastructure spend, those with a measurement gap held back investment at more than twice the rate of those without, 73% to 29%.

And the board’s blank check is already turning to “prove it”. Only 26% of boards still say “invest now and sort out the returns later”, while 66% now condition AI money on proof, 22% will approve nothing new until ROI shows, and 43% are being asked for ROI they can’t provide.

The companies spending most aggressively on AI and the ones most desperate to measure it are the same companies, one move apart: spend hard now, then build the allocation, forecasting, and outcome measurement that keep the funding flowing.

What they want is exactly what Part 1 was about: spend they can see fast and clearly, now aimed at the harder question of what that spend returned.

Conclusion:

What to do when the board asks

It’s clear across every finding: the question for many finance executives is whether they can see what is spent on AI and, especially, the return on that spend. The speed of that insight matters as well; businesses want and need to respond quickly in a fast-moving competitive space.

Seeing the spend and proving the return look like two jobs, but they’re one job viewed from two ends: getting the right number (spend tied to the outcome it produced) to the people accountable for it, in time to do something. Part 1 showed that speed is what keeps teams from freezing. Part 2 showed that connecting spend to outcomes is what keeps the board funding the next round. Neither is about cutting the bill.

The companies pulling ahead have the capability to read AI spend as a live, per-outcome signal; every dollar tied to the value it created. That is what lets them invest with confidence while everyone else waits for the bill.

Three moves follow from the data:

Close the latency gap. The break point is one day. Spend data that reaches finance within a day behaves like a different category than data that arrives several days later or with the bill.
Give the spender and the tracker the same view. The people driving AI spend and the people answerable for it are usually different. A shared, real-time picture closes that gap without reorganizing the company.
Measure spend against outcomes so you can move the money. Finance doesn’t want a smaller number. It wants cost per the thing its business runs on: per customer, per feature, per dollar of revenue, so it can pour more into what’s paying off and pull back from what isn’t. That’s what turns a spend report into a decision and an action.

The urgency is palpable, with nine out of 10 executives telling us they have less than a year to answer the AI outcome question. The good news is that this capability is within reach. Once teams build it, they can see the spend, prove the return, and move the money towards what works.

When they can’t (or won’t), the spend remains opaque, the checks dry up, and most of all, the rivals who solved it move on without them. That reckoning comes on the board’s timeline, not finance’s, and it arrives whether the answer is ready or not.

A note on the data

Every percentage in this report is drawn from 260 completed survey responses. We report a figure only where the group behind it clears a minimum sample threshold of at least 30 respondents; cohorts too small to be reliable are not shown. That over/under rule keeps each number meaningful rather than an artifact of a handful of responses. All figures describe associations in a single point-in-time survey, not proven cause and effect.