Start with a job people repeat. Describe its input, the steps they take and the finished output. A workflow such as preparing a weekly operating summary is easier to evaluate than a broad ambition to use more AI.
Measure a few examples before changing anything. Record hands-on time, review time and the frequency of exceptions. Ask the people doing the task what makes a result usable. A short baseline is more useful than an impressive savings estimate with no source.
Choose work with an accessible, approved input and an output somebody can check. Document summarisation, draft reporting and a simple spreadsheet transformation can be good candidates when the scope is clear. Tasks that need several inaccessible systems or unattended consequential actions usually need more preparation.
Name the person who will own the pilot. Agree what happens when data is missing, a document is out of date or the result is uncertain. Then build the smallest useful version and compare it with the baseline.
Our workflow scorecard helps a team make that choice together. Its score is a conversation aid, not a prediction of savings.
Use the workflow scorecard ↗