Measurement & testing
Minimum detectable effect (MDE)
Also known as: MDE · minimum effect size
The smallest lift the test is powered to detect at the declared significance and power levels — determines required sample size: smaller MDE requires larger sample.
Minimum detectable effect (MDE) is the smallest relative lift an A/B test is statistically powered to distinguish from noise. Tight MDE (detect a 2% lift) requires large samples; loose MDE (detect a 20% lift) requires small. The relationship is nonlinear — halving the MDE roughly quadruples required sample size. Set MDE honestly: what lift is actually worth finding? A 2% lift on a program that moves millions in revenue is worth powering to detect. A 2% lift on a small-program where only a 20% lift would merit a roll-out is over-powered; set MDE at 10-15%. Most A/B testing failures come from setting MDE too tight for the volume available — the test runs for weeks, never reaches significance, gets declared inconclusive or cherry-picked. Better discipline: set MDE, compute required sample, run until target hit, then decide.