KPMG's Quarterly AI Pulse Survey for Q3 2026, released September 24 2026, reports spend governance spreading fast. 74% of organizations now include cost reviews in AI approval, up from 61% a quarter earlier, and 43% have usage or token budgets. The same release reports that nearly 6 in 10 leaders see measurable business value, but it does not say how value was judged. A cost review records what AI costs. It does not record whether anyone's work changed, and that is the evidence a CFO needs before the next budget cycle.
Two things are true at once. Finance is right to review AI spend before it is approved, and that discipline was overdue. The value claims sitting beside it are leader-reported, and a leader's report is a view, not a baseline.
What did KPMG's Q3 2026 AI Pulse find about AI cost governance?
KPMG surveyed 314 US C-suite and business leaders at organizations with annual revenue of $1 billion or more, between July 24 and August 25 2026.
The figures KPMG reports
| Measure, as reported by KPMG | Counted as | Q3 2026 | Earlier figure in the release |
|---|---|---|---|
| Include cost reviews in AI approval | Organizations | 74% | 61% last quarter |
| Use AI monitoring dashboards | Organizations | 70% | Not given |
| Have usage or token budgets | Organizations | 43% | Not given |
| Confident in their AI governance and capabilities | Leaders | 73% | 57% last quarter |
| Report significant workforce adoption | Organizations | 44% | 23% last quarter, 10% a year ago |
| Building, deploying or developing AI agents | Organizations | 62% | 53% last quarter |
| Cite productivity gains | Leaders | 55% | Not given |
| Cite stronger financial performance | Leaders | 37% | Not given |
What the release leaves out
The press release does not define "measurable business value" or "significant workforce adoption," and beyond describing "visibility into AI spending and performance" it does not say what the monitoring dashboards track. It does not say whether the same organizations answered each quarter, so the moves from 61% to 74%, 57% to 73% and 23% to 44% are changes in survey results, not proof that any one company changed. We read the release only. The downloadable report did not render for us, so any definition it contains is not reflected here.
Is an AI cost review the same as measuring AI value?
No. A cost review is an input control. It asks, at approval, whether the spend is justified. A usage or token budget caps consumption, and a dashboard shows whatever it was built to show. None of these, on their own, records whether a team's output, cycle time or quality changed after AI arrived.
We made the same point about seat counts in AI license utilization: a license records what finance approved and cannot detect whether work changed. Counting agents has the same limit, which we covered in Counting Agents Is Not a Strategy.
The risk is that a mature spend process produces the feeling of measurement. KPMG's confidence figure rose 16 points in a quarter, from 57% to 73%, and confidence is also something leaders report, not something the survey observed. A different population paints a more cautious picture of oversight: we covered Deloitte and the CAQ's finding that 82% of audit committees call their AI oversight emerging or limited.
Why can reported AI value run ahead of captured value?
Deloitte's 2026 Enterprise Cost Transformation Survey, released September 14 2026, polled 500 senior leaders worldwide in April 2026, with more than half at publicly traded organizations above $5 billion in revenue. 67% expect to meet or exceed their overall transformation targets this year, and only 14% fully captured the value targeted by their highest-priority initiative last year. Only 24% account for value leakage when setting targets.
The limits matter. This is cost transformation in general, not AI, and it is a between-organization survey, so it does not show that KPMG's respondents face the same gap. Deloitte's release also states the 14% in two ways: full capture of the highest-priority initiative's targeted value last year (key takeaways), and achieving or exceeding program savings targets in 2025 (body). We use the first. The 67% is an expectation for this year, so the two figures are not a same-period comparison. What it shows is that in a closely related finance setting, a forward expectation and a backward result sat far apart when leaders reported on their own programs. That is a reason to treat reported AI value as a hypothesis to test. Deloitte also lists skills gaps and workforce adoption rates among the leading obstacles to AI-enabled savings, which is consistent with looking at how people work, not only at the spend line.
How long does AI take to deliver value in finance?
Gartner surveyed 160 senior finance function leaders between January and April 2026. Data extraction, accounts payable and receivable automation and report creation generally deliver returns within 9 to 10 months, while more complex tasks such as data management, insight generation and forecasting typically need a longer development period (Gartner press release, September 24 2026). A value claim without a date attached is hard to read, because the same figure means different things at month 3 and month 12. Gartner's portfolio framing is covered in AI portfolio management: four practices.
What evidence should a CFO require before accepting an AI value claim?
Five checks turn a leader's report into something a finance team can test:
- A baseline with a date. State what the metric was before AI and when the clock started, so the claim can be checked later.
- A comparable group. Compare teams that adopted against comparable teams that have not adopted yet, over the same period, controlling for tenure, role and tool stack. That is a stronger design than a company-wide before-and-after.
- A time-to-value window per use case. Judge each use case against a realistic time-to-value range, such as the 9 to 10 months Gartner's finance leaders report for the less complex tasks, and not against one quarter.
- A leakage line. Net out the cost of getting the value. Deloitte finds only 24% of organizations account for value leakage when they set targets.
- A confidence level. Every result carries a confidence rating and states what it cannot tell you, with team-level views shown only for groups of 5 or more people.
What this still cannot tell you
A comparison of adopting and non-adopting teams is a controlled cohort analysis, not an experiment. Teams that adopt early can differ in ways the controls miss, so a result is a bounded comparison and not full attribution. It is still a stronger basis for a budget decision than a leader's report or a spend dashboard. For the longer version of this method, see how to measure AI workforce ROI without the vendor self-grading problem.
Where does Levos fit?
Levos is a Human Capital Operating System: an intelligence layer above the existing stack. Spend tools, license reports and monitoring dashboards stay where they are and keep doing their job. Levos reads behavioral signals from the tools where work happens and shows, with a confidence score, whether the work changed. AI Impact is one of its six signal families, and the measurement methodology uses controlled cohort analysis with confidence scoring, compares adopting teams with non-adopting teams, and discloses its limits. We do not claim full attribution.
Individual-level data is visible to the person and their direct manager. Above that level, people see team aggregates of at least 5 people, apart from logged escalations of urgent issues; see the privacy model.
KPMG, Gartner and Deloitte each sell advisory or research services, and Levos sells measurement. Each of us has a view on what these figures should prompt, so we cite them as published and ask you to read them with that in mind.
Frequently asked questions
What did KPMG's Q3 2026 AI Pulse find about AI cost governance?
KPMG's Quarterly AI Pulse Survey, released September 24 2026 from 314 US C-suite and business leaders at organizations with $1 billion or more in revenue, found that 74% now include cost reviews in AI approval, up from 61% last quarter. 70% use AI monitoring dashboards and 43% have usage or token budgets. Nearly 6 in 10 leaders report measurable business value, but the release does not say how value was judged.
Is an AI cost review the same as measuring AI value?
No. A cost review is an input control that asks whether spend is justified at approval. A token budget caps consumption and a dashboard shows what it was built to show. None of them, alone, records whether a team's work changed after AI arrived. Value needs a baseline, a comparison with comparable teams, and a stated confidence level.
How long does AI take to deliver returns in finance?
In a Gartner survey of 160 senior finance function leaders, fielded January to April 2026, data extraction, accounts payable and receivable automation and report creation generally returned value within 9 to 10 months. More complex tasks such as data management, insight generation and forecasting typically need longer. The release does not say whether these windows are realized or expected, so treat them as a planning range and not a guarantee.
What evidence should a CFO require before accepting an AI value claim?
Ask for five things. A dated baseline, a comparison with comparable teams that have not adopted yet, a realistic time-to-value window for each use case, a line for value leakage, and a confidence level on every result. Deloitte's 2026 cost transformation survey found that 67% of leaders expect to meet or exceed targets this year, and separately that 14% fully captured the value of their highest-priority initiative last year. Reported value should be treated as a hypothesis until it is tested.
See value, not just spend
If your AI approval process is mature but your value evidence is a leader's report, that evidence can be rebuilt from observed work. Request a Demo and we will walk through what a comparison-based readout contains, or read how the measurement methodology works first.
Levos is accepting design partner applications from US organizations of 150 employees or more with an active AI rollout. Large organizations typically start with one function or division, measured against comparable teams that have not adopted yet. That is a stronger attribution design than a company-wide before-and-after.
