A/B test sample size calculator
Before you launch an experiment, find out how many visitors each variant needs — and how long that will take — for a statistically trustworthy result.
Experiment setup
Two variants: control (A) and challenger (B), split 50/50.
The control's current rate, e.g. 3 for 3%.
Relative lift you care about — 15 means detecting 3% → 3.45%.
Chance the detected lift is not a false positive. 95% is the norm.
Chance of detecting the lift if it really exists. 80% is the norm.
Total traffic entering the test (both variants). Used to estimate duration.
Saved on this device. Your numbers never leave your browser.
Required sample
What that sample size buys you
Where the two conversion rates would land at that sample size, if the real lift matches your effect. The less they overlap, the easier the call.
- A · control
- B · variant
- Overlap
Assumes a two-sided z-test on proportions with an even 50/50 split.
Stopping the test early, before the sample is reached, inflates false positives.