Home Insights How to Build an AI Business Case Your CFO Will Accept
Pilot to production
How to build an AI business case your CFO will accept
Last reviewed: 29 September 2026
Key takeaways
- Finance rejects AI business cases that rest on generic benchmarks and single-point estimates.
- Measure a baseline first, express benefits as ranges, and state every assumption.
- Agree up front how savings will be banked — otherwise they will never appear in the P&L.
An AI business case is a financial argument that a specific AI use case will change a specific cost, revenue or risk measure by enough to justify its full cost. Most AI business cases fail finance review for the same reasons: they use industry statistics instead of the company’s own numbers, promise precise savings nobody believes, and say nothing about how the savings will be realised.
Here is a structure that holds up.
1. Start with one process and a measured baseline
Pick one process. Measure it for two to four weeks: volume, handling time per item, error and rework rates, cycle time, and the cost of the people and systems involved. Use system data where possible and time sampling where not.
A baseline changes the conversation. "We handle 6,200 supplier invoices a month at an average of 11 minutes each, with 7{92aa103281e6849eb31fdffc4c8cf6f7eb32fdab66a5738451c3768cacd59970} needing rework" is a fact finance can check. "Companies typically save 30{92aa103281e6849eb31fdffc4c8cf6f7eb32fdab66a5738451c3768cacd59970}" is not.
2. Express benefits as ranges with explicit assumptions
For each benefit, state the mechanism, the assumption and the range:
| Benefit | Mechanism | Assumption | Range |
|---|
Ranges are more credible than points, and they make the conversation about assumptions rather than about trust.
3. Count all the costs
Include licences and model usage, build and integration, data preparation, change management and training, governance work, and ongoing operation — monitoring, evaluation, support and improvements. Finance will add them if you don’t, and your credibility will suffer.
4. Separate capacity from cash
Time saved is capacity, not cash. It becomes cash only when you avoid hiring, reduce overtime or contractors, redeploy people to revenue-generating work, or reduce headcount through natural attrition. Say which one applies, and when. This single step does more for a business case’s credibility than anything else.
5. Price the risk
Include the risk side: what could go wrong, what controls address it, and the cost of those controls. Where the use case reduces risk — fewer compliance errors, better audit trails — include that too, even if you express it qualitatively.
6. Agree how value will be tracked
Define the KPIs, the reporting frequency and who signs off realised value. A monthly one-page value report, agreed with finance before build, turns the business case into a management tool rather than a funding document.
A one-page template
1
Process and owner
2
Baseline
(with measurement method and period)
3
Proposed change
(what AI does, what people do)
4
Benefits
(ranges and assumptions)
5
Costs
(one-off and recurring)
6
How capacity becomes cash
(and when)
7
Risks and controls
8
KPIs, reporting and sign-off
9
Payback range and decision requested
The payoff
A business case built this way does more than secure funding. It sets up the measurement that proves the result — and that proof funds the next wave.
Estimate the value of AI in one process in two minutes.
Frequently asked questions
How long should baselining take?
Two to four weeks is usually enough for high-volume processes. Longer is rarely needed; shorter risks missing monthly patterns.
Should we include benefits we can’t quantify?
Yes, listed separately and described qualitatively. Don’t let them carry the financial argument.
What payback period do CFOs expect?
It depends on the organisation, but AI process use cases are often expected to pay back within 12 to 18 months. Ask your finance team for their threshold before you build the case.