How to calculate AI ROI before you spend anything.

Nearly every AI business case is built backwards: pick the tool, then find the justification. Do it the other way and the decision usually makes itself in twenty minutes.

Step one: price the work as it is today

People who touch the workflow × hours per week × loaded hourly cost × 52. Loaded means salary plus employment costs, not the headline wage. This is your baseline, and it needs to exist on paper before anyone demos anything — because after the demo you will be anchored.

Step two: estimate the realistic share removed

Not 100%. Be deliberately pessimistic: assume the system handles the clear-cut majority and humans keep the rest. For well-chosen document and enquiry workflows, 40–70% is a defensible planning assumption. Use the bottom of that range in the business case and let reality beat it.

Step three: add the costs nobody quotes

Build plus running cost plus maintenance at 15–20% of build annually, plus your own people's time during setup — which is real and usually a week or two of someone's attention.

Step four: check the payback period, not the percentage

ROI percentages are easy to inflate. Payback period is not. Divide total first-year cost by monthly saving. Under twelve months is good. Under six is excellent. Over eighteen means the workflow was probably the wrong choice, not that AI was.

The two numbers to ask any vendor for

“What is your accuracy on our actual cases?” Not benchmark accuracy — yours, measured against an evaluation set built from your real historical work, before you commit. If they cannot produce that, the ROI model has no foundation.

“What happens if we miss the target?” A vendor with no answer has priced in your failure.

You can run the whole calculation with your own volumes on our ROI calculator — it shows its assumptions, so you can argue with them.