01

Look for repeated information work

Good candidates often involve assembling, classifying, retrieving or routing information that teams perform repeatedly. The task should have a clear start, useful inputs and a known result.

02

Check the exception rate

A process with too many ambiguous exceptions may need redesign before automation. A small, bounded workflow with predictable exceptions is often a safer first candidate.

03

Define the measurement

Set a baseline for time, volume, quality, response or rework. Do not describe a workflow as successful merely because a model can produce an output.