Skip to content

Field guide

Run a 14-day AI pilot

Two weeks is enough to learn whether a narrow workflow deserves a larger test.

Days 1 to 3: define

Choose the workflow, baseline, owner, approved data, expected output, review rule, and stop condition. Save five representative fictional examples.

Days 4 to 10: operate

Run the same job repeatedly. Record time spent, edits required, errors, skipped cases, and reviewer confidence. Do not expand scope during the test.

Days 11 to 14: decide

Compare results with the baseline. Continue only when the owner can name the benefit, recurring cost, risk control, and next evidence target.