If You Cannot Read the AI Bill, You Have No Position
KPMG fielded its Global AI Pulse for Q2 2026 across 2,145 C-suite and business leaders in 20 countries. Twenty-nine percent struggle to understand and control operating costs as they scale enterprise AI. A third named limited understanding of AI costs and economics as a challenge to deploying agents. Nearly half of organizations have already rephased AI work when cost outweighed expected value. Anthropic, OpenAI, and GitHub have moved some services off a flat seat onto a meter.
That’s the problem.
This is not a token recap, and it is not a lock-in column. The invoice is no longer a seat. If the people who sign the budget cannot read it, you do not have a license position. You have invoices. Traditional FinOps was built for reserved instances and tag hygiene. It was not built for token meters.
What KPMG actually measured
Treat the percentages as KPMG’s own Pulse. This is not our book of work. The named source is the Q2 2026 Global AI Pulse, published with a 24 June 2026 press release.
KPMG’s line: “As usage-based pricing models become more common, many organizations are still building the capabilities required to forecast, monitor, and manage AI spending effectively.”
The survey did not say your SAM team failed. Self-reported. Not an audited license position.
The rephase number is the commercial event. Nearly half of organizations have already restaged AI work when cost exceeded expected value. KPMG’s read is not that confidence collapsed. “These actions do not signal reduced confidence in AI. Rather, they suggest a growing willingness to evaluate where AI creates meaningful value and where it does not.” Access to lower-cost, high-fidelity models is the fastest-rising influence on AI strategy, up 7 percentage points from Q1.
Governance sits next to the invoice. Organizations need clear rules for when employees can intervene, who owns AI-related costs, how outputs are reviewed, and what happens when systems fail. Most report some mechanisms. Relatively few describe them as fully embedded. The Pulse also found that organizations with stronger cost visibility are more likely to have those controls in place.
Do not invent a unit rate and call it a forecast. KPMG did not publish yours. Your number is your history against the meters you are already on.
If the C-suite cannot read AI bills, you do not have a license position
If that feels familiar, the problem is not your team’s maturity. You have been asked to brief operating cost with the tools that already work for cloud showback. Those tools assume a unit you can reserve and a month that looks like the last one. A token meter does not. Model mix moves. Agent loops compound. A seat floor can look clean while overflow is already the bill.
This is not a criticism of the teams building those reports. Traditional SAM tools collect comprehensively and answer slowly. Traditional FinOps collects spend and answers in cloud units. Neither was designed to recast a usage-based AI invoice into something a CFO can defend in fifteen minutes. The gap is interpretation work.
Picture the room. It is a mid-year spend review. Your CIO wants to know why AI opex jumped. Finance wants a range, not a dump. The vendor will say usage-based is the new normal. If SAM and FinOps need three weeks to turn a token export, a seat file, and three contracts into one page the C-suite can read, you have already lost the framing of the meeting.
An analysis that arrives after the salesperson has named “usage” is a document. One that arrives during the meeting is a decision input.
The license question is narrower than the Pulse. Can you say, this week, which AI contracts are still flat and which will move with usage, who owns the cost, and which deployments you already rephased. If the C-suite cannot answer those from the pack you have today, you do not have an AI license position. You have a reporting layer.
Rephasing is not a failure of nerve. It is the decision the Pulse says nearly half of organizations have already taken: from a pack, not from a surprise invoice.
What to put in the pack the C-suite can actually read
Do this in the next thirty days, as a decision pack rather than a project. One page the CFO can hold. Not a token library.
| What goes in the pack | Why the C-suite can read it |
|---|---|
| Which AI contracts are still flat, and which are already metered | So they know which bills can stay a line item and which will move with usage. Name Anthropic, OpenAI, GitHub, and anyone else on the paper: as contract status, not as a product recap. |
| Operating cost this period versus last, by product | A token export is a library. Period-on-period by product is a decision. |
| Who owns AI-related cost, in writing | KPMG asked who owns the cost. If that owner is “the platform team, unless Finance asks,” you do not have a control. |
| Where cost already exceeded value, and what you rephased | Nearly half of organizations, as reported, have already done this. Name yours. |
| A twelve-month range on your own history | Floor, expected, spike. No invented unit price. Use the rates on your current invoices. |
| What the current paper still entitles you to | Seat floor, included pool, or nothing left in the bundle. FinOps will skip this half if you only send tokens. |
Can a non-specialist brief the CIO in fifteen minutes from the pack you have today? If the answer is no, AI operating cost will keep landing as a surprise. That is not a communication problem. It is a missing position.
The decision layer, not another invoice view
You already have seat files, usage exports, and a FinOps view of cloud. The gap is not another inventory. The gap is turning that estate into a decision the C-suite can take into a spend review: what is still flat, what is metered, who owns the cost, and what you would rephase before the next invoice lands. LICENSEWARE sits on the inventory and ITSM tools you already run. It is not a rip-and-replace SAM suite. It is a decision layer: what matters, why it matters now, what should happen next. That is cost optimization work: which contracts are still flat, which will move with usage, and what you would rephase. NEO turns the token dump into a pack a non-specialist can brief, and when the model is the commodity, the license question is how Copilot sits as a digital workforce, not which lab trained the weights.
If you are heading into an AI spend review and your current tools still need three weeks to turn a token dump into a page the CFO can read, book a Software Intelligence Review. You can also start on the free plan and run the analysis on your own data.
The question to walk in with
Do not let the vendor frame this as “usage-based is how AI works now.” The right question is: can the C-suite read this bill, who owns the cost, which contracts will move with usage, and which deployments we would rephase if next month is the spike. That is a data question, not a sales question.
The vendor will walk in with a meter and a capacity story. The only question is whether you have a position first: current, defensible, and tied to the contracts in front of you.
FAQ
What did KPMG find about AI operating cost? The Q2 2026 Pulse of 2,145 senior leaders across 20 countries found that 29 percent struggle to understand and control operating costs as they scale enterprise AI. Treat that as a self-reported figure.
Why isn’t traditional FinOps enough for AI bills? FinOps was built for units you can reserve and tag. Token meters move with model mix, agent loops, and overflow after a seat floor.
If the C-suite cannot read the bill, do we still have a license position? No. A position is something a non-specialist can defend in the meeting. If AI operating cost baffles the C-suite, the pack is not finished.
