Part of the Maths & Statistics suite · 235 calculators

A/B Test Calculator

Whether variant B really beat variant A — the two-proportion test on conversion rates, with the lift, its confidence interval, and an honest note on peeking.

Two conversion rates are compared with a two-proportion z-test: the difference divided by the standard error of that difference under a pooled rate.

Results update as you type
Results
p-value (two-tailed)
0.02829487
A — conversion rate
B — conversion rate
Relative lift
Absolute difference
z statistic
CI for the difference — lower
CI for the difference — upper
Decision
Visitors per variant to detect this lift reliably
Reviewed September 2026. Pure mathematics: the result does not depend on where you are. Terminology follows the Australian Curriculum (maths, brackets, decimal point).
No account required · Google Analytics off unless allowedCalculator arithmetic runs in your browserResults update as you type
All calculations run 100% in your browser. The calculator code does not submit your figures to GlobalCalc to obtain a result.
About a/b test

How the a/b test calculator works

Two conversion rates are compared with a two-proportion z-test: the difference divided by the standard error of that difference under a pooled rate. The p-value is the chance of a gap this large if the two variants were identical.

The statistics are the easy part. The hard part is that stopping the moment the result looks significant inflates the false-positive rate several times over — so the sample size must be fixed before the test starts.

Formula: z = (p₁ − p₂) / √(p̄(1−p̄)(1/n₁ + 1/n₂))

Worked examples

Inputsp-value (two-tailed)Note
5% against 6% on 5,000 each0.02829487p = 0.031 — B wins
Identical rates1p = 1
The same lift on 500 visitors0.48796934nowhere near significant

Frequently asked questions

How long should I run an A/B test?

To a sample size fixed in advance, and for whole weeks — weekday and weekend traffic behave differently.

Why is peeking a problem?

Because checking repeatedly and stopping at the first significant result can push the false-positive rate from 5% to over 25%. If you must monitor continuously, use a sequential testing method designed for it.

What is a realistic lift?

Most real tests move conversion by a few percent relative. Detecting a 5% relative lift on a 5% base rate takes tens of thousands of visitors per variant.

Should I report relative or absolute lift?

Both. "20% lift" and "one percentage point" describe the same result and give very different impressions.

What about testing more than two variants?

Each extra comparison adds its own false-positive chance. Correct for multiplicity, or the winner is often just the luckiest arm.

Where these figures come from

Last checked: September 2026. Formulas are fixed by mathematics and do not change with tax years or regulations.