Sample Size Calculator
Plan your A/B test before launching it. Estimate how many users you need to detect a meaningful change in conversion rate—and how long collecting them might take.
Two-sided Z-testBinary conversion metricEducational toolExperiment assumptions
Enter your expected baseline and the smallest improvement worth detecting.
The current conversion rate of your Control group.
Smallest relative change you want the test to detect. Example: 12% baseline + 10% relative = 13.2% Treatment CVR.
Controls the false-positive rate under the null.
Probability of detecting the assumed effect.
Total daily users across both groups.
Share of users assigned to Treatment.
Required sample
Estimated for an independent, randomized two-group experiment.
Planning estimate using a normal approximation with an unpooled variance under the alternative and pooled variance under the null. Actual achieved power can differ due to rounding, traffic patterns, and test design.
Learn the concepts
Understand the assumptions behind the calculator before using the result.
What is sample size—and why does it matter?
Sample size is the number of eligible, independently observed users needed in each group. Too few users can make a meaningful effect difficult to detect. More users generally improve sensitivity.
What is the Minimum Detectable Effect (MDE)?
The MDE is the smallest change you want to have a reasonable chance of detecting. Here it is a relative percentage, not percentage points. Smaller MDEs require much larger samples.
What do significance and power mean?
A 5% significance level sets the nominal Type I error probability under the null hypothesis. An 80% power target means that, if the assumed effect is real and model assumptions hold, the planned test has approximately an 80% chance of rejecting the null.
Why does a 50/50 split usually need fewer total users?
For two groups with similar variability, equal allocation is generally the most sample-efficient design. An uneven split typically increases the total sample required.
Does the duration estimate guarantee a valid experiment?
No. It assumes stable eligible daily traffic and independent assignment. Run through complete business cycles when relevant, avoid repeatedly peeking and stopping based on significance, and account for exclusions, missing events, clustering, multiple comparisons, and guardrail metrics.
Which statistical method is used?
This calculator plans a two-sided two-proportion Z-test using a common normal-approximation sample-size formula. For extremely low conversion rates or rare outcomes, use a dedicated exact or simulation-based power analysis.