Fill in these first
- Product and the change being tested: [e.g. showing delivery time on the product card in a grocery app]
- Why we think it will work: [evidence, such as interviews, support tickets, a competitor pattern]
- Primary metric and its current value: [e.g. add-to-cart rate, 11 percent]
- Metrics that must not get worse: [e.g. order value, refund rate, page load time]
- Traffic available per week: [e.g. 40,000 users on that screen]
- Smallest change worth shipping: [e.g. plus 0.5 percentage points]
Write the brief
- Hypothesis: "If we [change], then [metric] will [move] because [reason]."
- Variants: control and each test version, described so a developer can build them.
- Audience: who is included, who is excluded, and how users are split.
- Duration: whole weeks only, long enough to reach the needed sample, and the date to stop.
- Primary metric and how it is counted.
- Guardrail metrics, each with the threshold that stops the test.
- Decision rule written now:
| Result | Decision | | Primary up by at least the smallest worthwhile change and guardrails hold | Ship | | Primary flat or unclear | Iterate or drop, with which one | | Any guardrail breached | Stop and roll back |
- Risks: novelty effects, seasonal traffic such as a festive sale week, and other tests running on the same screen.
Guardrails
- Do not peek and stop early on a good day; name the stop date.
- If the traffic cannot reach the sample needed in four weeks, say so and suggest a bigger change or a different method.






