AI spend is up. Proof of AI ROI is not.
In 2026, the question about AI changed from “should we invest?” to “what did we get?”, and most companies cannot answer it. AI ROI has become the accountability test that CFOs and CIOs now face together, at every budget review.
The gap is stark. KPMG’s Global AI Pulse for Q2 2026, which surveyed 2,145 senior leaders across 20 countries, found AI remains a top investment priority for 79% of them, yet only 7% report established ROI (KPMG). Nearly one in four face direct investor pressure to prove value.
Finance leaders feel it most. A June 2026 CloudZero survey of 260 finance executives found only 22% can tie AI spend to business outcomes today, while 87% say they need that ability within a year (CloudZero).
This is not a story about AI failing. It is a story about measurement failing. The companies pulling ahead are not spending less. They simply know what their spending buys.
Why is AI ROI so hard to prove?
AI ROI is hard to prove because most organizations can measure neither side of the equation well: they lack clean cost data and they lack agreed value metrics. Four problems stack on top of each other.
The cost side is blurry. KPMG found 42% of leaders have only partial visibility into AI spending, and a third struggle to understand AI cost structures such as tokens. Usage-based pricing means spend accumulates across many teams, tools, and invoices.
Spend and ownership are split. CloudZero’s survey shows finance drives AI spending in only 18% of companies, yet owns spend tracking in 60% (CloudZero). Finance is asked to justify decisions it did not make.
Value is measured as activity, not outcomes. Adoption rates, prompts sent, and hours “saved” are easy to count but rarely reach the income statement. Uber’s COO described the difficulty of drawing a line between AI usage and product outcomes after the company used up its 2026 AI budget by April.
Pilots never earn their keep. Gartner found at least half of generative AI projects were abandoned after proof of concept by the end of 2025, citing poor data, weak risk controls, rising costs, and unclear value (Beri). Spend that never reaches production can never show a return.
How AI funding decisions are changing in 2026
AI money now flows through the CFO, and it increasingly lands in business unit budgets rather than a central innovation pool. Three shifts define the new funding model.
The CFO becomes the gatekeeper. Forrester expected enterprises to defer a quarter of planned AI spend into 2027 as financial scrutiny tightens, with CEOs routing AI investment decisions through the CFO (CloudZero). CFOs now want value that shows up in the P&L, such as savings, avoided spend, or revenue, not just time saved (Kyriba).
Central budgets give way to chargebacks. Early movers are pushing production AI costs back to the business units that use them. When a sales team sees its own AI bill in its monthly review, the conversation turns from access to efficiency. Forbes reported that Uber introduced a monthly token cap per user, and that Amazon, Walmart, Cisco, and Meta also moved to rein in AI tool use (Beri).
CIOs are on the clock. A Dataiku survey cited by CIO.com found 71% of IT leaders believed they had until mid-2026 to prove AI value or face budget or job consequences (CIO.com). The pressure is reshaping spend more than cutting it. The money is leaving AI that nobody is accountable for, not AI in general.
What AI investments survive CFO scrutiny?
The AI investments that survive have two things in common, a clear accountable owner and visible costs tied to a specific outcome. KPMG’s data shows how large the difference is.
| Condition (KPMG Global AI Pulse, Q2 2026) | Established ROI with it | Established ROI without it |
|---|---|---|
| Strong visibility into AI costs | 15% | 3% |
| CEO accountable for AI-informed decisions | 14% | 4% |
Companies with clear CEO accountability also report meaningful business value far more often, 57% versus 21% (KPMG). Visibility and ownership are not overhead. They are the strongest predictors of return in the data.
At the project level, three profiles hold up in budget reviews.
- Workflows with a cost per outcome. Contract review, invoice processing, or ticket resolution where you can state the cost per unit before and after AI.
- Tools with granular usage reporting. Platforms that show who uses what, at what cost, and with what result are easy to defend. Tools that offer only broad productivity claims are the first to be cut.
- Agents with defined scope and spend limits. Agents with cost caps, monitoring, and clear permissions get funded. Open-ended agents that can loop and retry without limits do not.
How to measure AI ROI: a six-step framework
Measuring AI ROI means comparing the fully loaded cost of each use case with a value metric agreed before launch, then reviewing both every quarter.
\text{AI ROI} = \frac{\text{Measured value} - \text{Fully loaded AI cost}}{\text{Fully loaded AI cost}}- Sort investments into portfolio buckets. Efficiency plays should pay back within a year. Growth plays may take longer. Strategic bets are funded as options with a capped budget. Each bucket gets its own payback expectation, so a long-horizon bet is not judged by quarterly savings.
- Set a baseline before launch. Record today’s cost per unit, cycle time, error rate, or conversion rate. Without a baseline, any improvement is an opinion.
- Allocate the full cost to an owner. Include tokens, licenses, cloud infrastructure, integration, and human review, then assign them to a team, product, or agent. Open standards such as FOCUS are adding AI-specific fields that make this allocation consistent across vendors (FOCUS).
- Pick one unit value metric per use case. Examples include cost per resolved ticket, cost per reviewed contract, or revenue per qualified lead. Convert hours saved into money only when they are redeployed or avoid a hire.
- Measure net value monthly. Track spend and value side by side, so a rising bill is always read against what it produced.
- Decide every quarter to scale, fix, or stop. Fund what proves its value, redesign what is close, and retire what cannot connect to an outcome.
The roles split cleanly. The CIO supplies usage data by team, workflow, and model. The CFO sets the thresholds, owns the baseline, and makes the scale or stop call. A practical rule from the field is to require a cost per outcome metric at approval for any AI project above a set annual threshold, such as $250,000 (Beri).
Frequently asked questions about AI ROI
What percentage of companies can prove AI ROI? Very few. KPMG’s Q2 2026 Global AI Pulse found only 7% of leaders report established AI ROI, and CloudZero found 22% of finance leaders can tie AI spend to business outcomes.
How do you calculate AI ROI? Subtract the fully loaded cost of an AI use case from the measured value it produced, then divide by that cost. Fully loaded cost includes tokens, licenses, infrastructure, integration, and human oversight.
Why do AI projects fail to show ROI? Most lack a baseline, a defined value metric, or a clear owner. Many also never move past the pilot stage, so their costs never produce production value.
Who is responsible for AI ROI, the CFO or the CIO? Both. The CIO provides usage and cost data, while the CFO sets thresholds and makes funding decisions. KPMG data shows clear executive accountability strongly improves results.
How long should AI take to show ROI? It depends on the type of investment. Efficiency use cases should pay back within about a year, while growth and strategic bets need longer horizons and capped budgets.
What is AI chargeback? AI chargeback assigns the cost of AI usage to the business units that consume it. It creates accountability and encourages teams to use AI where it adds measurable value.
The bottom line for CFOs and CIOs
The AI era does not punish high spending. It punishes spending nobody can explain. The data is consistent across KPMG, CloudZero, and Gartner, and it shows that visibility and accountability, not budget size, separate the companies that prove AI ROI from those that ration AI.
Before the next budget cycle, every CFO and CIO should be able to answer three questions for each major AI investment. What did we spend? Who owns it? What did it produce? If those answers exist, AI budgets will grow with confidence. If they do not, finance will make the decision for you.