Give Every AI Estimate a Confidence Level and a Reason
When AI turns a plan into a timeline, a date found in the supplied documents can look just as certain as one filled in from common patterns. To expose the difference, assess every estimate separately and record the item and number, its basis in the supplied material, its level, and the one assumption that would in it.
Use HIGH for data or comparable past results, MEDIUM for from incomplete data, LOW for judgment, and UNKNOWN when no basis is given. Never present a LOW- number without its key assumption, and flag any number that looks exact despite having LOW .
The reason matters more than the label. A note that the source email never included a date immediately shows what information is missing and whom to ask for it.
Key points
- Separate dates supported by source material from dates the AI merely inferred.
- Record each estimate, its evidence, level, and key assumption.
- Use HIGH for data or history, MEDIUM for partial evidence, LOW for judgment, and UNKNOWN when no basis is stated.
- Do not show a LOW- number without its assumption, and flag precise-looking numbers with LOW .
- Write a specific reason that points to the missing evidence or the person who can resolve it.