A/B Testing Sample Size Calculator
Estimate the number of visitors each variation needs to detect a meaningful change in conversion rate with your chosen confidence and statistical power.
Overview
Use this A/B testing sample size calculator to estimate how many eligible visitors you need before comparing two conversion-rate variations. Enter your current conversion rate, the smallest relative uplift worth detecting, your confidence level, statistical power, and average daily eligible traffic.
How it works
The calculator estimates sample size for a two-variation conversion test. It first converts your baseline conversion rate and target relative uplift into expected control and variant conversion probabilities. It then uses the difference between those rates, together with the selected confidence level and power, to estimate the visitors needed in each group. Smaller changes, lower baseline conversion rates, higher confidence, and higher power generally require more visitors.
How to use this calculator
- 1Enter the current conversion rate for your control page or experience.
- 2Set the smallest relative uplift that would be meaningful for your decision.
- 3Choose a confidence level and statistical power for the test.
- 4Enter the number of eligible visitors you receive each day.
- 5Review the required visitors per variation and estimated test duration.
Example Calculation
Current conversion rate
10%
Minimum detectable uplift
20%
Confidence level
1.96
Statistical power
0.84
Daily eligible visitors
1000
Visitors needed per variation
3,837 visitors
With a 10% baseline conversion rate, a target 20% relative uplift, 95% confidence, and 80% power, the test needs about 3,835 visitors per variation, or about 7,670 visitors in total. At 1,000 eligible visitors per day, that is roughly 8 days.
Frequently asked questions
What is sample size in A/B testing?
Sample size is the number of eligible visitors or observations needed in each variation to detect a target difference with a chosen level of confidence and statistical power.
What is a minimum detectable effect?
The minimum detectable effect, or MDE, is the smallest change you want the test to reliably identify. In this calculator, it is a relative uplift from the baseline conversion rate.
Why does a smaller target uplift need more traffic?
Small changes are harder to distinguish from normal random variation. Detecting them reliably requires more observations in each variation.
What confidence level should I use for an A/B test?
A 95% confidence level is a common default. A higher confidence level reduces the chance of a false positive but increases the required sample size.
What statistical power should I choose?
An 80% power level is commonly used for planning. Higher power makes it more likely that a real target effect is detected, but it also increases the sample size required.
Does this calculator work for revenue or average order value tests?
This version is designed for binary conversion outcomes, such as sign-ups or purchases. Revenue and average order value typically need a different sample-size method because they are continuous and often more variable.
Why might my test need to run longer than this estimate?
Actual eligible traffic can vary, traffic may not split exactly evenly, and conversion rates can change over time. Running through complete business cycles can also be important for representative results.
Explore Related Calculators
Assumptions and warnings
Assumptions
- The test compares two independent variations with visitors split evenly between them.
- The calculation uses a standard two-sided comparison of conversion proportions.
- Traffic quality, conversion behavior, and the baseline conversion rate are assumed to remain broadly stable during the test.
- The selected uplift is relative to the current conversion rate, not a percentage-point increase.
- Results are planning estimates and do not account for multiple variants, repeated result checking, or segmentation.
Warnings
- This calculator provides a statistical planning estimate, not a guarantee of test results.
- Avoid ending a test early solely because an interim result appears significant.
- Use a target uplift that keeps the implied variant conversion rate below 100%.