- Target users — Who is this for, specifically? (e.g. first-year undergraduates, parents of school-age children)
- Core use case — What single task should this solve well, before anything else is added?
- Data sources — What content or data does it need access to, and where does that data currently live?
- Guardrails — What should it explicitly refuse or defer to a human on?
- Success metric — How will you know it's actually helping, beyond usage numbers? (e.g. task completion, reduced repeat enquiries)
- Pilot plan — Who tests it first, for how long, and what would make you delay a wider rollout?
- Feedback mechanism — How do users flag something wrong, and who reviews that feedback?
- Review cadence — When will this be reassessed after launch — one month, one term?
How to use it: Fill this in before writing a line of code or briefing a vendor. Most AI project problems trace back to skipping questions 3 and 4.