An AI business case is an ordinary investment case with four extra traps: benefits depend on adoption, costs include usage-based lines, time saved is not automatically cash, and the technology and prices change during the project.
The six parts
- The problem and the metric: what process, what volume, what baseline (time, cost, errors, cash, accuracy).
- The benefit model: volume × time saved × loaded cost, plus other benefits (errors avoided, early-payment discounts, fraud prevented, working capital). State how each becomes cash or capacity.
- Costs, in three groups:
- one-off: build, integration, data clean-up, change and training
- run: licences or seats, tokens or usage, people to supervise and maintain
- hidden: security reviews, audit work, model updates, vendor management
- Adoption: a curve by year or a ramp in months. Benefits arrive only as people actually use the tool.
- Risk: a haircut on benefits for delivery risk, and a sensitivity on the drivers that matter.
- Decision and tracking plan: hurdle rate, payback, kill criteria, and how value will be reported after go-live.
Common errors
- Counting 100% adoption from month one.
- Treating every hour saved as cash.
- Leaving out run costs that scale with usage.
- No baseline, so value can never be proven.
- No end date or kill criteria for the pilot.
Use the ROI builder
The hub's ROI & business case builder turns these inputs into yearly cash flows, NPV, IRR, payback, a sensitivity tornado and break-even adoption, and exports a one-page case for your investment committee.