CFOs have learned to be skeptical of AI business cases. Most are written by the team wanting budget, so they optimistically assume full adoption, perfect model performance, and no integration cost. Then reality shows up.
Here’s the framework we use when we help a client present to finance.
Four numbers. No more.
A business case that opens with ten benefits loses. One that opens with four clean numbers, each defended, wins. The four we always include:
1. The baseline (what it costs today)
Concrete, measured, attributable. “We spend 4,200 hours/year on X. Loaded cost: €38/hour. Total: €159,600/year.” Not “around €200k” or “significant time”.
If you can’t measure the baseline, don’t run the project yet. Measure the baseline first.
2. The expected saving — with a confidence band
Not one number. A range. “Conservative: 30% reduction = €47,880/yr. Realistic: 50% = €79,800/yr. Optimistic: 70% = €111,720/yr.”
The CFO will push back on the optimistic. Show that the business case works at conservative too. If it doesn’t, the project isn’t strong enough.
3. The all-in cost — including change management
Typical AI business cases under-count by 2–3×. What they miss:
- Data audit and cleanup (often 30% of total cost)
- Integration with existing systems
- Internal change management (training, process redesign)
- Year-one support + model monitoring
- Opportunity cost of your people pulled into the project
Be the team that includes all of this. Finance people notice.
4. Payback period and break-even month
Not IRR, not NPV at first. A one-line answer to “when does this pay itself back?”. If the answer is over 18 months, either the scope is too big or the ROI isn’t real yet. Start smaller.
The three objections you’ll get — and the answers that work
“How do we know this will actually adopt?” Show the change-management plan. Name the owner. Commit to a usage metric as part of project success, not just a performance metric.
“What if the model performs worse than expected?” Include a go/no-go checkpoint mid-project where the contract can be unwound. Makes the downside bounded and knowable.
“What about compliance/security?” Bring the security/legal teams into the pre-approval. A business case that arrives at the approval meeting with their sign-off already attached moves through twice as fast.
What we’ve seen work
Business cases that get approved usually share three traits: they pick one workflow (not “AI for the company”), they commit to a measurable outcome with a date, and they treat change management as a first-class cost item. The model is almost a footnote.
The CFO isn’t saying no to AI. They’re saying no to vague, because vague has burned them before.