Lab · Your own numbers
A/B test calculator
The costliest experiment mistake is running it with too small a sample: you get a result, but it means nothing. Here you can see what is detectable before you start.
For example, 10 means from 3% to 3.3%
Minimum sample per variant
Total across both variants
Enough to detect a
Smallest detectable improvement
Equivalent to reaching a rate of
How this is calculated
Two-proportion test with normal approximation, 5% two-sided type I error and 80% power. This textbook approximation is for planning; analysing the actual experiment result is a separate calculation. Sample sizes are always rounded up.