How the calculator works
The calculator uses the standard formula for comparing two conversion rates, the normal approximation for two proportions. It answers one planning question: how many visitors does each variant need so that a real lift of the size you entered would show up as a significant result?
The formula in words
Start from the baseline rate and the target rate, which is the baseline plus the minimum detectable effect. A relative effect multiplies the baseline (10% plus 10% relative is 11%), and an absolute effect adds percentage points (10% plus 1 point is 11%).
- Take the normal quantile for your confidence level, split across both tails for a two-sided test, and multiply it by the spread of a conversion rate at the average of the two rates.
- Take the normal quantile for your statistical power, and multiply it by the combined spread of the baseline and target rates.
- Add the two, square the sum, and divide by the square of the gap between the two rates. That is the sample size per variant, rounded up to a whole visitor.
The total sample is the per-variant figure times the number of variants, control included. The duration is the total divided by your daily visitors, rounded up to a whole day, and the calculator then rounds it up again to whole weeks. The table under the result repeats the sum at half, one and a half and twice your effect, so you can see what a bigger or smaller target costs in days.
What the estimate assumes
Traffic is split evenly between the variants, every visitor counts once, and the conversion rate holds steady while the test runs. With three or more variants the calculator makes no adjustment for comparing several variants at once. Treat the result as a plan, not a promise: if your traffic or conversion rate moves, run the numbers again.
How this relates to Bayesian testing
This calculator is a frequentist planning estimate. Split Test Pro reads its results the Bayesian way: it gives the probability that each variant is the best, updated as the data arrives, so a test needs no fixed sample size before it can show a result. A planned length still helps. It tells you whether a test is worth running on your traffic at all, and it keeps you from stopping a test early because of a lucky first few days.
Whichever way you read the result, the same habits hold. Run every test for at least 7 days. Treat 300 to 500 visitors per variant as a starting point, and wait for 1,000 or more on high-stakes decisions. Split Test Pro also waits for ten conversions in every variation, control included, before it names a winner.
Once your test has data, check it with the Bayesian A/B test calculator. For the full explanation of the model, read our guide to Bayesian A/B testing.