A/B Test Sample Size Calculator

How many users — and how many weeks — does your test actually need? Standard two-proportion power analysis, with an honest verdict when the test isn't worth running.

Baseline conversion rate 3.0%

Your current rate for the primary metric (e.g., visitors who purchase). Use the metric closest to the change you're testing.

Minimum detectable effect (relative) 5%

The smallest relative lift worth detecting — a business question, not a stats one. 5% means detecting 3.0% → 3.15%.

Daily users entering the test (all variants)

Unique users per day exposed to the experiment surface, split across the two variants.

Statistical power

The probability of detecting the effect if it truly exists. Significance is fixed at 5%, two-sided.

—

Users per variant

—

Total users needed

—

Recommended duration

The math: standard two-proportion sample size with two-sided α = 5%: n = (z_1−α/2·√(2p̄(1−p̄)) + z_power·√(p₁(1−p₁)+p₂(1−p₂)))² ÷ (p₂−p₁)². Duration is rounded up to full weeks (running partial weeks biases results through weekday/weekend mix). Why peeking early inflates false positives, how CUPED can cut this duration by 30–50%, and the rest of the discipline: The Experimentation Playbook.