Appendix D. AI Project Canvas and 90-Day Plan

Canvas element	Questions to answer
Problem and purpose	What specific problem are we solving? Who experiences it? Why is AI appropriate?
Users and affected people	Who uses the system and who may be affected by its outputs?
Baseline	How does the current process perform in time, quality, cost, and risk?
Data	What data are needed, who owns them, and are they accurate, lawful, and suitable?
System and provider	Which tool or model is proposed? What are its limits, terms, and dependencies?
Risk and controls	What can go wrong? What safeguards, human reviews, and fallback options are required?
Success measures	Which measurable outcomes determine success? What are the stop criteria?
Ownership	Who approves, operates, monitors, and responds to incidents?
Costs	What are the full implementation, training, monitoring, and exit costs?
Decision	What evidence is needed to stop, revise, or scale the pilot?

Period	Main actions	Evidence / decision
Days 1–30	Define problem, baseline, owner, users, data rules, risk tier, and approved tool.	Approved project canvas and pilot scope.
Days 31–60	Configure, train, test representative cases, record failures, and collect feedback.	Test log, user feedback, risk controls, and interim results.
Days 61–90	Compare outcomes with baseline; review costs, errors, equity, security, and user experience.	Written go / revise / stop decision and monitoring plan.