Status: Draft / In review / Agreed • Owner: [ ] • Version/date: [ ]
1. Decision and outcome
We are deciding [ ]. For [user/segment] doing [task], we want to improve [observable outcome] from [measured baseline or unknown] to [target and rationale].
2. Evidence
| Source ID | Date and segment | Observation | Interpretation | Limitation |
|---|---|---|---|---|
| [ ] | [ ] | [ ] | [ ] | [ ] |
3. Options and rationale
Chosen approach: [ ]
Simpler alternative: [ ]
Do-nothing consequence: [ ]
Why this choice now: [ ]
What could reverse it: [ ]
4. Scope
In scope: [ ]
Explicit exclusions: [ ]
Dependencies: [ ]
Unresolved decisions: [ ]
5. Journey and requirements
| Step | Starting state | Action | Expected result | Failure/recovery | Status |
|---|---|---|---|---|---|
| [ ] | [ ] | [ ] | [ ] | [ ] | Agreed / Proposed / Needs decision |
6. AI behavior, if applicable
Allowed evidence and source permissions: [ ]
When to ask a question: [ ]
When to decline or escalate: [ ]
Allowed actions and approval requirements: [ ]
Output editing and undo behavior: [ ]
Evaluation set and rubric: [link]
Latency/cost expectations: [ ]
7. Acceptance and measurement
Given [state], when [action], then [observable result].
Primary metric with denominator/window: [ ]
Quality and safety guardrails: [ ]
Instrumentation owner: [ ]
Pilot population and rollout gate: [ ]
Pause/rollback owner and trigger: [ ]
8. Review record
Design: [reviewer/date/questions]
Engineering: [reviewer/date/questions]
Research/analytics/domain expert as needed: [ ]
Decision log: [date, decision, rationale, owner]
Do not let AI convert blank fields into invented commitments. Keep “unknown” visible until a responsible person resolves it.