How to calculate the real ROI of an AI project before you build it

Most AI projects fail the business case long before they fail technically. Teams get excited about a capability, build a prototype, and only then ask what it is worth. Flip that order.

Start with the decision, not the model

Write down the single decision or task the AI will improve. Who makes it today, how often, how long does it take, and what does a wrong answer cost? Those four numbers give you a baseline.

Estimate three costs

Build cost (data prep, modelling, integration), run cost (infrastructure, API usage, monitoring) and change cost (training, process redesign). The last one is the one most estimates forget.

Model a range, not a number

Estimate a pessimistic, expected and optimistic improvement. If the project only pays back in the optimistic case, it is not a first project. Pick the use-case that works in the pessimistic case and learn from it.