# AI Task Suitability & Human-Review Checklist

_By Gervin Kahunde. Free to adapt._

Answer for each task you are considering automating.

## A. Is AI a good fit for this task?
- [ ] The task is repetitive and high-volume.
- [ ] Inputs are available, lawful to use, and the minimum necessary.
- [ ] A wrong output can be detected by a person before it causes harm.
- [ ] Success can be measured (accuracy, time, rework).
- [ ] A simpler rule or process change would not solve it.

## B. Where must a person decide?
- [ ] Decisions that affect someone's rights, money, record, or access stay with a person.
- [ ] Reviewers see the source next to the AI output.
- [ ] Reviewers can correct, reject, and explain why — and that feedback is captured.
- [ ] There is a clear escalation path for sensitive or unusual cases.

## C. Risk and fairness
- [ ] Sensitive attributes are excluded or justified.
- [ ] Error and outcome rates are compared across groups.
- [ ] There is a stop condition and an owner who can pause the system.

## D. Launch readiness
- [ ] Tested on synthetic or approved data first.
- [ ] Documentation for users and reviewers exists.
- [ ] Operating metrics and review cadence are agreed.
