A/B Test
An A/B test compares two variants using a defined method to understand their effect on an indicator. Assignment, volume, duration, and analysis criteria must be established. A difference observed in a small sample may be due to chance. The decision must take into account uncertainty and side effects.
A concrete example
Two randomly assigned groups receive different email subject lines, with the same content and at the same time. The decision indicator is fixed before sending.
A point to consider
A difference observed in a small volume may be due to chance. Avoid multiplying the variants and stop as soon as a result seems favorable.
Write the decision before launching the test
Specify the population, the unit of distribution, the primary indicator, the expected duration, and the adverse effects to monitor. Keep a control variant and avoid changing multiple elements if you want to understand which one is having an effect. Document exclusions before the analysis. An uncertain result calls for a cautious conclusion; it does not become positive by retrospectively selecting the segment that appears most favorable.
Questions to ask your team
- Does the test cover the time required for the measured action?
- Have stopping and analysis rules been defined before reviewing results?


