How do you measure the return on investment of an AI project?

The return on investment of an AI project is often calculated on the automated task: so many minutes saved per case, multiplied by the number of cases. The figure is attractive, and often misleading.

The time saved here does not always show up at the end. It can be absorbed by extra checks, passed on to the next department, which receives more work, or change nothing in the final lead time if the bottleneck is elsewhere. Measuring an AI project means measuring the whole flow, from start to finish, before and after.

• Time savings do not convert by themselves.An hour freed up only creates value if you have decided what it will be used for.
• Indicators are often set after the event.Without a starting measurement, all you can note is an impression of improvement.
• Running costs are underestimated.Supervision, corrections, updates and subscriptions weigh over time, long after the launch.

1. 01Before launching a project, measure the flow concerned as it is: lead times, volumes, errors, rework.
2. 02Choose with the business two or three indicators that matter to it, not just to the project.
3. 03Budget for the running cost over three years, not just the cost of setting it up.

When a gain measured in one department is paid for elsewhere, we take the measurement again across the whole flow.

• How do we choose our first use cases?→
• Is AI mainly about cutting costs?→
• Can automating one department unbalance another?→
• Why do our pilots never scale up?→