Tools
A/B test sample size calculator
How many visitors does your test need? Enter your baseline conversion rate and the lift you want to detect — get the required sample per variant and how long the test will take at your traffic.
Test assumptions
Assumes a 50/50 split between two variants. Runs entirely in your browser — nothing you enter is stored or sent anywhere.
Planning estimate at 95% confidence, 80% power
You need 14,749 visitors per variant
To detect a lift from 10.00% to 11.00% — about 30 days (5 weeks) at your expected traffic.
29,498
total sample (both variants)
Per variant
14,749
visitors in each arm
Total sample
29,498
across both variants
Target rate
11.00%
from 10.00% baseline
Estimated runtime
30 days
at 1,000 visitors/day
This is a planning estimate under explicit assumptions (two-proportion z-test, two-sided α, balanced 50/50 split) — not a measured result. Smaller effects, lower baselines, and higher confidence or power all increase the required sample.
How the estimate works
The calculator uses the standard power analysis for a two-proportion z-test. It asks: to distinguish your baseline rate p₁ from the improved rate p₂ = p₁ × (1 + MDE) with your chosen confidence and power, how many independent observations does each variant need?
n = (z₁₋α/2 + z₁₋β)² · ( p₁(1−p₁) + p₂(1−p₂) ) / (p₂ − p₁)²
Three levers dominate the result: the baseline rate (rarer conversions need more data), the MDE (halving it roughly quadruples the sample), and power (90% instead of 80% adds ~30% more traffic). The runtime estimate simply divides the requirement by your expected daily visitors on the slowest arm.
Sizing the test is only the start: you still have to run it without peeking, check the traffic split held, and judge whether the measured lift is safe to ship. Converise plans MDE and runtime per experiment, then runs the quality checks before every decision. When your test is done, check the result with the significance calculator.
Frequently asked questions
How does this sample size calculator work?
It uses the standard two-proportion z-test power formula: given your baseline conversion rate, the minimum relative lift you want to detect (MDE), a confidence level (two-sided α), and statistical power, it computes the number of visitors each variant needs. Dividing by your expected daily traffic gives the estimated runtime.
What is a minimum detectable effect (MDE)?
The MDE is the smallest lift you want your test to reliably detect, expressed as a relative change — a 10% MDE on a 10% baseline means detecting a move to 11%. Smaller MDEs require dramatically more traffic: halving the MDE roughly quadruples the required sample.
Why does my test need so many visitors?
Conversion differences are usually small compared to the natural noise in visitor behavior. To tell a real 1-point lift from random variation, you need enough observations for the noise to average out. Low baseline rates and small MDEs both push the requirement up sharply.
What does statistical power mean?
Power (1 − β) is the probability that your test detects the effect if it truly exists at the MDE size. At 80% power, a real effect of that size is still missed in 1 out of 5 tests. Use 90% when missing a real winner is costly and you can afford the extra traffic.
Can I stop the test early once it looks significant?
No — peeking at results and stopping at the first significant reading inflates the false-positive rate far above your chosen α. Decide the sample size up front with this calculator, run to completion, then evaluate. If you must monitor continuously, use a sequential testing method designed for it.
Is this calculator free? Where do my numbers go?
It is completely free, with no signup. All calculations run in your browser — nothing you type is stored or sent to any server.
Plan the test. Trust the decision.
Converise estimates MDE and runtime for every experiment you plan, then checks the result for SRM, guardrail declines, and segment conflicts before you ship.
Try Converise free