Capgemini’s survey of 500 executives: half say AI spending is approved without adequate visibility into the full cost. Ask what one live AI tool costs to run
Half the executives in a new Capgemini survey say AI spending is approved without adequate visibility into the full cost. If that holds in your organisation, the AI you already run has a cost nobody has shown you. Ask for it this week. What does one live AI tool cost to run each month, including the people who check its work?
On 12 November 2026 the Capgemini Research Institute published a survey of 500 senior executives at organisations with annual revenue above $1 billion. Capgemini sells AI and consulting services, so it has an interest in this finding. The figures are executives’ own reports. They describe AI spending in general, not measured results after a go-live, and they name no organisation.
That is a sound basis for one question. The answer will come from your own records.
Start with one tool, not the whole portfolio. One answer, worked through properly, shows whether your programme can cost an AI tool at all. Then the same 3 figures can be asked of the next one.
Half of 500 executives say AI spending is approved without adequate visibility into the full cost
Capgemini’s release of 12 November reports these findings from its survey:
- Half of executives say AI spending is approved without adequate visibility into the full cost.
- In the past 12 months, 59% say their organisation’s AI-related expenditure was significantly higher than expected or budgeted.
- Two thirds of organisations that overshot their AI budget faced “negative financial consequences”. The release names lower free cash flow and a cut in discretionary spending.
With 59% reporting spending well above plan, asking for the full cost is ordinary control.
Capgemini ran the survey in September 2026. It describes the people who answered as “CXO-level senior executives”. That means chief officers: the chief executive, the chief financial officer and their peers.
The gap that matters to a sponsor is what was in front of the approver
A finance sponsor already knows that technology spending can overrun. The overspend is not the new part of this release.
The new part is the approval. Half of these executives say the spending is approved without adequate visibility into the full cost. So an overrun may not be a forecast that went wrong. It may be a cost that was never in the forecast at all.
We think that gap widens after go-live. Before go-live, the cost of an AI tool sits in one place: the programme’s budget. After it, the cost splits.
The supplier’s charge arrives on an invoice. Someone in finance sees it each month.
The rest does not arrive on an invoice. A person reads what the tool produced before it is used. Someone puts right what it got wrong. A manager reviews a sample each month. Those hours sit in the budgets of the teams that do the work. No single line shows them as a cost of the AI.
Nobody is hiding these hours. They are spread across teams, in budgets set before the tool arrived, so no report adds them up.
A sponsor who asks “what does this tool cost?” may be shown the invoice. The invoice is the part that was easy to see.
An invented example shows where the hours hide
Every figure here is invented, to show the shape of the sum. Use your own.
An AI tool drafts the monthly commentary on budget variances for 120 cost centres. The supplier charges £1,500 each month. That is the figure the approver saw.
An analyst checks each draft before it goes to the budget holder, at 15 minutes each. That is 30 hours. About 1 draft in 4 needs rewriting, at 20 more minutes each: another 10 hours. The financial controller reviews a sample each month: 4 hours.
That is 44 hours each month. At an invented £50 an hour, the hours come to £2,200. The full running cost is £3,700 each month. The approver saw about 40% of it.
The example does not show the tool is poor value. The commentary may be faster and better than before. It shows that the approval was made on the smaller part of the cost.
Two thirds of those who overshot faced negative financial consequences
The release ties the overspend to money a finance sponsor watches. Of the organisations that overshot their AI budget, two thirds faced negative financial consequences. Capgemini’s examples are lower free cash flow and a cut in discretionary spending.
Those are the sponsor’s own levers. A cut in discretionary spending is a decision someone in finance has to make, and explain. Free cash flow is a number the board reads.
We think an AI tool whose full running cost was never shown at approval is exposed to the same cuts. The cut may land on something unrelated, such as a hire that waits or a project that slips a quarter.
That is why the question belongs with the sponsor, not only with the programme. The programme sees the tool. The sponsor sees what the tool’s cost displaced.
Ask your programme lead for one tool’s monthly cost, people included
Pick one AI tool that is already live in your programme. If none is live yet, pick the first one due to go live.
Ask your programme lead for 3 figures for that tool.
- The supplier’s charge for one month, from the latest invoice or the contract.
- The hours each month that people spend checking, correcting and reviewing the tool’s work, by role.
- Those hours priced at your organisation’s own hourly cost.
Ask for the figures in writing, with the source of each. The charge comes from a document. The hours will be an estimate, and the sheet should say whose.
Then ask one more thing. Was that total in front of whoever approved the tool? If not, what figure did they see?
If you approved it, the question is about what you were shown. Ask it of yourself first.
A rough answer is fine. A timed sample of a week’s checking, multiplied up, is better than no figure. If the programme lead cannot find the hours, that is a finding in itself: nobody holds them today.
If the tool was bought by a supplier or an integrator on your behalf, ask the programme lead to get the charge from them. The hours are still your own people’s.
If the total is close to what was approved, the approval had the cost right. Whether the tool is worth that cost is a separate question. If it is well above, you know before the next budget round, while there is still time to act on it.
Done when: one sheet shows the tool, its monthly charge, the monthly hours and their cost, the total, and the figure the approver saw.
One question puts the running cost on the record
Take this to your next meeting with the programme lead:
What does this AI tool cost us each month, including the people who check its work? And was that figure in front of whoever approved it?
Ask it of one tool now. Then ask it of each new tool before it goes live, so the approver sees the full figure first time.
What comes next
This issue’s long read follows the cost past go-live. It asks who owns it as use grows, and what evidence they bring before the next tranche is funded.
BusinessGPS’s Baseline report scores how an organisation runs change, with a red, amber or green grade at each step. The BusinessGPS Baseline report shows one, run on BusinessGPS itself.
Sources
- Capgemini Research Institute, press release on the economics of AI, Paris, 12 November 2026, read on 15 November 2026. capgemini.com