Type in the results of an A/B test. This tells you whether the difference is likely real, how big it could plausibly be, and whether the traffic split itself looks trustworthy. Everything runs in your browser; nothing is sent anywhere.
1. Analyze a test
2. Plan a test: how many visitors do you need?
3. Why you should not peek
Imagine testing two identical versions. A properly run 95% test falsely announces a winner about 5% of the time. But if you check the results repeatedly and stop the moment it looks significant, you will be fooled far more often. This simulates 1,000 such no-difference tests.
What this does and does not do
It compares two conversion rates with a two-sided two-proportion z-test (pooled for the test, unpooled for the interval). It is a good fit for large samples; with very few conversions per group, use an exact test.
"Significant" means unlikely to be chance. It does not mean the lift is big enough to matter: look at the interval.
The traffic-split check flags tests where assignment or logging may be broken (a very small p-value, below 0.001).
It does not handle multiple variants, revenue or other non-binary metrics, or sequential testing methods.