Growth Engineering Playbook

Why your A/B test is inconclusive

Paste your test numbers and get an honest verdict — statistical significance, power, the effect you can actually detect, and how long you'd really need to run it. No backend, nothing leaves your browser.

Try an example:

Numbers reflect the fictional Northstar Outfitters store used across this playbook's shared-data.

Control (A)

Conversion rate: —

Variant (B)

Conversion rate: —

Assumptions

Enter your numbers

Results update as you type. This tool errs toward honesty: an underpowered test is reported as "not enough evidence", not a win.

p-value (two-sided)
Observed absolute lift
95% CI for the difference
Statistical power (observed)
Required sample / variant
Time to significance

Peeking simulator

Why does "just checking" hurt? This runs many A/A experiments where both variants have the same true rate — so every "win" is a false positive. Testing once at the end holds the false positive rate near your threshold. Peeking after every checkpoint and stopping at the first significant result inflates it.