The AI experimentation era ended this year — and it ended more quietly than anyone predicted. Back in February, 71% of CIOs said their AI budget would be cut or frozen if it hadn't proved value by mid-2026. Mid-2026 came and went, and the anticipated bloodletting never arrived: there's no credible evidence of CIOs being dismissed en masse over AI returns.
But something quieter did happen. Nearly half of organisations have delayed, stopped, or scaled back AI projects on budget grounds — killing poor performers, shrinking pilot programmes, and cutting the number of vendors they use. The cull was of projects, not people. And the pressure didn't ease; it became structural: 98% of CIOs report that board pressure to demonstrate measurable AI ROI has risen since 2024, and 85% now have their own compensation tied to measurable AI outcomes.
Meanwhile the money keeps growing — global IT spend is up 14.2% to $6.37 trillion — but it is chasing proof, not promise. After two years of pilots and "we're exploring AI," the question from the board has hardened: show me the number. Teams that can't will quietly lose their mandate and their budget.
This is healthy. The mandate to prove ROI kills theatre and rewards exactly the unglamorous work that pays — the finance, operations, and compliance processes that move the P&L rather than the demo that impresses a steering committee.
Why so much AI spend can't prove itself
Most AI initiatives that fail the ROI test don't fail because the technology didn't work. They fail because no one measured. Three habits are usually to blame:
- No baseline. The pilot launched without capturing the "before" — the hours, the cost, the cycle time. "It feels faster" is not ROI, and you can't prove a saving you never measured.
- Aiming at novelty. Effort went to the impressive and visible, not the repetitive and expensive. The 2026 returns are in the back office, not the showcase.
- Tool sprawl. Overlapping AI subscriptions accumulate like shelfware. 34% of CIOs now cite legacy and redundant systems — tech debt — as the number-one barrier to AI adoption.
How to make AI spend defensible
The fix is a discipline, and it's the same one that makes any operational investment defensible:
- Capture a baseline before you build. Measure the process as it is today. This one step is the difference between "we think it helped" and "it saved £X." (Our AI cost calculator is a fast way to put a conservative number on a manual process before you invest a penny.)
- Aim at cost and cycle time. Target processes where the saving is measurable — finance close, reconciliation, ticket handling, document processing — not the flashiest use case.
- Apply FinOps to AI. Track spend against value, and cut what doesn't land. This is the core of the FinOps of AI: treat AI cost like any other cost centre with a return attached.
- Deal with the tech debt underneath. You can't automate on a foundation you can't integrate. Framing modernisation as tech equity rather than tech debt is what unblocks the AI roadmap and satisfies the CFO.
- Report value per use case. Adoption counts and licence numbers flatter; measured outcomes convince. Report the ones that work, cut the ones that don't, and you keep the mandate.
The board question that finds the savings
Stop asking "are we doing AI?" — the answer is always yes, and it tells you nothing. Ask instead: "which AI spend can you defend with a number this quarter, and which can't?" The second half of that question is where the waste — and therefore the opportunity — is hiding. It reframes AI from a status game into a P&L line, which is exactly where it belongs.
The bottom line
The money is still growing, but the tolerance for unmeasured AI is gone: prove the return or lose the budget. The winners will be the teams that baseline before they build, aim at the processes that move the P&L, apply FinOps discipline, and report value honestly. Turning AI investment into a defensible, board-ready number — tied to EBITDA, not enthusiasm — is precisely what our Executive Advisory work exists to do.
Need to prove — or find — the ROI in your AI spend? Let's put a number on it.