Is my ad result real?
Enter what each variant got. You'll get the probability each one is truly best, a credible interval on the uplift, and a straight call: pick a winner, keep running, or there's no real difference yet. Built on Bayesian inference, so every look is valid — peeking won't fool it the way it fools p-values.
Your variants
Conversions = the thing you want (clicks, sales, signups). Trials = how many saw it (impressions, visitors). The first row is your control.
How each variant did
Rate is what you observed. "Chance it's best" and the interval account for how little data you have — small samples get wide intervals, as they should.
| Variant | Observed rate | Chance it's best | Likely true rate (95%) | Uplift vs. control |
|---|
How much data do I need?
The question nobody answers before they start. Tell me your typical rate and the smallest improvement worth chasing; I'll estimate the traffic per variant needed to call it.
Why you can trust this (and peek freely)
Classic A/B calculators report a p-value, which quietly assumes you fixed your sample size in advance and looked exactly once. Real people peek daily — and every peek inflates the false-positive rate, so "significant!" often isn't. This tool reports posterior probabilities instead: "given the data so far, the chance B is actually better is X%." That statement is honest no matter how often you look. The math is open and unit-tested — see the self-tests.