"How much will this AI project make us?" is the question every executive asks and few teams answer honestly. The instinct is to point at a flashy demo and hope. A better approach is boring, and it works: model the value, model the full cost, and be ruthless about the gap between a pilot that impresses and a system that pays. Here is a framework you can run on a napkin before committing real money.
#Start with the value, in one of three shapes
Nearly every AI business case creates value in one of three ways. Name yours before you build anything.
- Cost reduction — the AI does work people currently do. Value = hours saved × loaded hourly cost × how much of the task the AI actually covers.
- Revenue lift — the AI helps you sell or retain more: better recommendations, faster response, higher conversion. Value = incremental revenue × margin.
- Risk reduction — the AI catches fraud, errors, or compliance issues earlier. Value = probability of loss avoided × cost of that loss.
#Now the costs — all of them
The model cost is the tip of the iceberg. The failures we see almost always come from costs that never made it into the spreadsheet:
- Build — engineering time to design, integrate, and test the system. Usually the largest line item.
- Data — the unglamorous work of collecting, cleaning, labeling, and maintaining the data the model depends on.
- Inference — per-request model costs, which scale with usage and can surprise you at volume.
- Integration — connecting to your existing systems, which is where timelines quietly double.
- Oversight — human review, monitoring, and the cost of the exceptions the AI can't handle.
- Maintenance — models drift, sources change, and dependencies update. Budget for the second year, not just the launch.
#The formula
Put it together and ROI is unglamorous arithmetic: annual value minus annual running cost, divided by the total investment to build and run it. Payback period is the build cost divided by the monthly net gain. If you can't estimate each input within a believable range, you don't yet understand the project well enough to fund it.
#Pressure-test before you commit
- 1Run the numbers at pessimistic assumptions — half the value, 1.5x the cost. Does it still clear the bar?
- 2Ask what has to be true for this to fail, and whether you'd notice in time to stop.
- 3Confirm the value is capturable — hours 'saved' only count if you redeploy them or don't backfill.
- 4Set a kill criterion up front, so a struggling pilot ends cleanly instead of drifting.
“The point of an ROI model isn't to produce an impressive number. It's to force the honest conversation about value, cost, and risk that separates the projects that ship from the ones that quietly die.”
A credible model won't tell you an AI project is guaranteed to pay off — nothing can. But it will tell you which projects are worth the bet, which need reshaping, and which look exciting in a demo and lose money in production. That clarity, before you spend a dollar, is the whole point.
