Skip to content
- Clear hypothesis, target metric, and success criteria.
- Event taxonomy implemented with QA.
- Experiment framework or traffic splitter available.
- Scope: define variant(s), surfaces, and guardrails; document pre-analysis plan.
- Implement: add flags/buckets, instrument events, and ship behind safe rollout.
- QA: verify bucketing, event firing, and visual parity across devices.
- Run: maintain duration to reach power; avoid mid-test changes.
- Analyze: compute lift, confidence, and segment cuts; review side effects.
- Decide: ship, iterate, or revert; document learnings.
- Sample ratio match and even exposure; no tracking gaps; assumptions met.
- SRM detected: check bucketing logic and filters.
- Metric noise: lengthen test or use higher-signal proxy.
- 1–2 weeks typical. Improves decision quality and UX outcomes.
