How do you build an AI business case your CFO will approve?

Lead with the number your CFO already trusts, not the tool a vendor is selling. Write the success metric and the pre-AI baseline into the case before anyone signs. In aibl’s survey of 755 UK mid-market leaders, that habit separated firms with a proven return from the rest.

Why most AI business cases get rejected

The reality is a weak AI business case doesn’t fail because the CFO dislikes AI. It fails because it asks for money against a promise that nobody will be able to check later. That’s a measurement gap, and you can close it before you ever ask for budget.

Most pitches describe what the tool does. Fewer describe what will be different, by how much, and compared to what.

A CFO’s job is to weigh a claim against a baseline. If your case has no baseline, there’s nothing to weigh. The answer defaults to no, or to a small pilot that nobody follows up on.

The fix isn’t a better pitch deck. It’s measuring AI ROI before you ask for the cheque, not after.

What a CFO-ready business case actually includes

A case that gets approved names one success metric and states where it stands today, before any spend happens. Everything else in the document supports that single line. Without a baseline written down first, there’s nothing to measure the result against a year later.

Start narrow. Pick one function and one task within it, not a company-wide overhaul. Broad promises are the easiest ones to reject.

State the metric in the CFO’s language: hours saved, error rate, conversion, cost per unit. Not “efficiency gains” or “better decisions”. Numbers, not adjectives.

Name an owner. A case with no named owner reads as nobody’s job, and CFOs know what that means for follow-through.

Set a review date. Put a date in the case when you’ll report back against the baseline, whether the number moved or not.

Frame it in peer terms, not vendor terms

The pitch that gets budget approved sounds like a comparison, not a sales pitch. Try this line: peers at our maturity report ROI of X per cent, we’re at Y, and here’s my plan to close the gap. A CFO already trusts a benchmark; a feature list, less so.

That framing works because it leads with a number the CFO already trusts, benchmarked against other businesses. It doesn’t lead with a capability a vendor wants to sell you.

Bring your finance partner in before you write the case, not after you’ve drafted it. A CFO who helped shape the metric rarely rejects the case built around it.

This is also where governance earns its keep. In aibl’s survey of 755 UK mid-market leaders, 22% with no governance showed a return, against 85% of the most mature. Same tools, different discipline. A case built on top of a governance framework is a stronger case. Someone can already show how outcomes get tracked.

Set the right payback expectation

Plan for the return to show in year two, not year one, and write that into the case itself. Most initiatives in aibl’s survey of 755 UK mid-market leaders turned positive in the second year, once the foundations were in place, not in the first flush of a launch.

A case promising first-year payback sets an expectation the project usually can’t meet. That gap becomes the reason the CFO points to next time you ask.

Gartner’s 2026 research backs the patient framing. Firms most satisfied with their AI spent 30% more than the unsatisfied on foundations: data, governance and talent, not just tools.

That spend doesn’t show up as savings in month one. It shows up as the reason year two pays off, and the case should say so plainly.

Plan for the failure that’s coming anyway

Seventy-nine per cent of leaders in aibl’s survey of 755 UK mid-market leaders have already had at least one AI failure. A CFO who has seen a number like that won’t be surprised by it, so don’t hide the risk from the case. Plan for it instead.

Build a checkpoint into the plan instead of a single go/no-go moment at the end. Say what you’ll do if the metric hasn’t moved by the review date.

That single line does more for credibility than any optimistic projection. It tells the CFO you’ve thought past the launch, which is usually the actual test the case is failing.

This connects to the wider AI adoption roadmap: cases that survive scrutiny get built in stages. Each stage has its own checkpoint, not one large bet.

What to do before your next AI purchase

Pick one use case. Write the success metric and the pre-AI baseline before anyone signs anything. Name the owner and the review date in the same document, and set the date you’ll report back against that baseline whether the number moved or not.

That’s the whole habit. It costs an afternoon, and it’s the difference between a case that gets approved and one that dies in committee.

Frequently asked questions

Read the full research

This is one finding from State of UK AI Adoption 2026, aibl’s benchmark of 755 UK mid-market leaders, in partnership with Executive Summary. The full report sets out the governance ladder, the ownership data and the four moves that separate firms with a provable return from the rest.

Read the full State of UK AI Adoption 2026 report →