Resources

Measuring the value of an AI project: a calculation framework

An estimate supports a decision when observations are separated from assumptions. Use this framework to prepare a pilot, then check whether it remains worthwhile after human review, exceptions and recurring costs are included.

Updated: 12 October 2026

Define a comparable unit of work

Choose one unit: a handled request, prepared file or checked document. Measure active work separately from waiting time. Distinguish straightforward and complex cases. A different mix between the baseline and the pilot would distort the comparison.

Calculate net time, including exceptions

For each category, calculate: eligible volume × actual usage rate × (baseline time − preparation − review − corrections). Convert minutes to hours at the end. A negative result means additional work. Do not count savings from the same step twice.

  • An hour freed up is not automatically a cash saving.
  • Specify how freed capacity could be reassigned.

Build three cost scenarios

Prepare cautious, central and favourable scenarios using the same cost categories: setup, tool access, consumption, support and training. Vary useful volume, adoption and correction time. Keep the date, owner and rationale for every assumption so it can be revised.

Set a decision rule before testing

Agree acceptable quality, errors that stop the trial and a maximum budget. After the pilot, replace assumptions with available observations and identify what remains unknown. Present the decision with its uncertainty rather than treating a return percentage as certain.

Your checklist

Your time log

Enter three comparable observations. The download contains your entries without transmitting them.

Further reading

One process. One next step.

Start with your daily work to decide what is worth checking together.