Underpowered tests lie politely
Stopping an A/B test early because variant looks better is how false positives enter product roadmaps. Sample size math forces you to commit to a minimum detectable effect before traffic arrives.
A one-point lift on a 3% baseline needs far more users than intuition suggests. This calculator uses a standard two-proportion approximation — adjust for multiple comparisons if you run many variants.
Pair the number with runtime: divide total sample by daily traffic to see calendar time. QA Studio helps you document the hypothesis while you wait.