First AI Workflow Selector

Compare small tasks, then choose one you can review.

Score each candidate

Give each question 0 = no, 1 = partly, 2 = yes. This is a planning aid, not a validated scoring model.

1. Does the task repeat often enough to be worth improving?

2. Do I have permission to use the required inputs, and can I access them?

3. Can I review the result before it causes an external action?

4. Would an error be easy to spot and undo?

5. Can I measure whether it helped?

Task: ___ Scores: ___ / ___ / ___ / ___ / ___ Total: ___ / 10

If input permission or reviewability scores 0, redesign the task before building, regardless of the total.

Worked example · fictional weekly status

Draft a Friday update from an approved tracker: repetition 2, inputs 2, review 2, reversible 2, measurable 2. Total 10. The lead still checks and sends it. Measure drafting plus correction time against the manual baseline and count factual fixes. A high score does not prove it will work.

Selection prompt

Compare [tasks] using the five questions above. Ask for missing information. Explain each score and flag permission, review and error risks. Recommend the smallest pilot, its human reviewer, success measure and failure alert. Do not build yet.

One-week pilot

Baseline minutes: ___ Pilot minutes including corrections: ___ Errors: ___ Missed runs: ___ Owner: ___ Continue/change/stop: ___

Start with a draft-only version using permitted information. Keep external sending and consequential decisions under review.