Start with a hypothesis, not a variant
Write what you expect to change and why. A test without a hypothesis teaches nothing when it wins.
Size the test before you launch
Sample size and duration come from the baseline rate and the effect worth detecting, not from patience.
No peeking
Calling a test early on a good day is how teams ship losers. The stopping rule is set in advance.
Guardrails on every test
Revenue, refunds and support contacts are watched alongside the primary metric.
Prioritize by revenue impact and effort
The roadmap is ranked by expected value, so the big tests run first.
Keep the log
Every test, its result and its lesson in one place. The log is worth more than any single win.
What we deliver for A/B testing
The concrete work behind these principles, on the Services page.