A/B test sample size calculator
To know how many visitors your A/B test needs, enter your current conversion rate and the smallest lift worth detecting: the calculator gives you visitors per variant at 95% confidence and 80% power, the usual values.
By Fernando Blanco Dosil · Updated November 2, 2026
Work out your test
For example, 3 if 3% of visits convert
Relative: 20 means going from 3% to 3.6%
- Visitors per version
- 13,914
- Visitors in total
- 27,828
How to read the result
The number is per variant. If your test has the original and one variant, you need twice that in total. With three versions, three times.
What moves the result most is the minimum lift. Detecting +20% needs about four times less traffic than detecting +10%. That is why, if your visits are limited, it pays to test big changes rather than colour tweaks.
Some reference values
Visitors per variant, at 95% confidence and 80% power.
| Current conversion | Lift to detect +10% | Lift to detect +20% | Lift to detect +30% |
|---|---|---|---|
| 1% | 163,100 | 42,700 | 19,800 |
| 2% | 80,700 | 21,100 | 9,800 |
| 5% | 31,200 | 8,200 | 3,800 |
| 10% | 14,800 | 3,800 | 1,800 |
Rounded to the nearest hundred. Computed with the same formula as the calculator.
What confidence and power mean
Confidence is how sure you want to be of not accepting a lift that does not exist. Power is the chance of finding the lift if it really exists. At 95% and 80%, out of every five real lifts you test you will find four, and you will accept a false one around one time in twenty.
How Bixlow does it
This calculator uses the classic two-proportion formula, the same as almost every calculator. Bixlow reads results with Bayesian statistics and does not name a winner until there is minimum sample and duration and at least a 95% probability that one version is best. The order of magnitude of traffic you need is the same, so this number works for planning.