Building an AI business case your CFO will sign

Every AI vendor has an ROI slide. Very few of them survive a CFO's questions: compared with what, measured how, and what does it cost to run every month after the project ends?

A business case your CFO will sign is built from your own numbers, includes the cost of running the agents, and shows a payback period finance can test and later check.

The problem, and who has it

Business sponsors want funding. Finance has seen technology programmes promise savings that never reached the P&L. Without a shared method, the conversation becomes belief against scepticism.

The steps

1. Measure the baseline first. For the target workflow, record volume per month, minutes per case, cost per hour, error or rework rate, and cycle time. Use two to three months of real data, not estimates.

2. Separate the value types. Cost removed, capacity released, revenue protected and risk reduced are different, and finance treats them differently. Do not add them into one number.

3. Include the full run cost. Model usage per task, platform and integration costs, monitoring, and the people who approve and review agent work. Run cost is where many business cases quietly fail.

4. Show the sensitivity. What happens if adoption is 60% instead of 90%, or the agent needs approval on twice as many cases? A case that only works in the best scenario will not be signed.

5. Agree how you will check it. Name the metric, the report and the date finance will review actual against plan. This turns the business case from a pitch into a commitment.

Who signs off

The CFO or a finance business partner signs the method and the baseline. The budget owner signs the targets. For anything that changes headcount plans, HR is consulted before the case is presented.

What good looks like

• A baseline measured from real data, signed by finance.
• Run cost per task shown next to cost per task today.
• A payback period with a best, expected and worst case.
• A review date in the finance calendar.