Skip to main content
At the end of a pilot, check whether the work became faster and easier to review without reducing quality. Record the existing baseline and compare it with the pilot using the same tasks and criteria.

​
What to measure

Track time to complete the work, the number and type of corrections, missing evidence, unresolved questions, whether users can follow the process independently, and exceptions sent to a responsible person. Include the time spent reviewing and correcting an Agent’s result; estimated time saved is not the same as realised benefit.

​
Review and decide

Compare several outputs with their original sources. If evidence is missing, adjust the connection or instructions. If access is too broad, fix the scope. If errors could cause material harm, keep a person in control or stop that workflow. When the process has a clear owner, stable inputs, and results that pass review, consider adding one adjacent team, source, or trigger. Recheck access, output quality, and exception handling each time the scope changes.