Realm by Rook
AI product engineeringProduct designAI implementationWorkflow automationSystems integrationDomain strategyAI product engineeringProduct designAI implementationWorkflow automationSystems integrationDomain strategy
SEO & Marketing Tools

A/B Test Sample Size Calculator

Compute sample size + days-to-significance for conversion rate experiments

Your current rate (e.g. 5% = 5 conversions per 100 visitors).
Smallest improvement worth detecting (10% means 5% → 5.5%).
Probability of a false positive (typically 5%).
Probability of detecting a real effect (typically 80%).
Sample per variant
31,234
Total sample
62,468
Days to significance
63
at 1,000 daily visitors
Scenario

At 5% baseline conversion, detecting a 10% relative improvement (i.e. 5.00%5.50%) at 95% significance with 80% power needs 31,234 visitors in each variant, or 62,468 total.

How to use

  1. Enter baseline + MDE
  2. Set significance and power
  3. Review sample size and days

Use cases

  • Growth marketing
  • Product experimentation
  • Conversion optimization

About the A/B Test Sample Size Calculator

Running an A/B test without computing sample size first is how teams end up with 'inconclusive' results that waste months. This calculator uses the exact formula for two-proportion z-tests to tell you upfront: how many visitors per variant, and how long you'll wait. Fix your MDE first, plan the test, then ship.

Features

  • Inputs: baseline rate, minimum detectable effect, significance level (α), power, daily traffic
  • Outputs: per-variant sample, total sample, days to significance
  • Uses inverse normal CDF for exact z-score derivation
  • Plain-English scenario summary

How it works

  1. Enter your current (baseline) conversion rate.
  2. Enter the minimum lift you'd care about (MDE).
  3. Keep default significance (5%) and power (80%) or adjust.
  4. Enter your daily visitors to see days-to-significance.

Frequently asked questions

Why is my sample size so huge?

+

Because detecting small improvements at low baselines is statistically hard. At 1% baseline, detecting a 5% relative lift (1.00% → 1.05%) needs ~60k per variant. The smaller the effect, the exponentially more data you need.

What values should I use for α and power?

+

Industry standard: α=5% (accept a 5% false-positive rate) and power=80% (detect 80% of real effects). More rigorous teams use α=1% and power=90% but sample sizes balloon.

What's a 'minimum detectable effect'?

+

The smallest improvement you'd care about. If a 2% lift isn't worth shipping, set MDE=5% and you'll stop the test much sooner. Honest teams set this BEFORE running, not after seeing results.

Should I use a Bayesian approach instead?

+

Fixed-sample frequentist (what this tool computes) is still industry norm — easy to explain to stakeholders, works with most A/B tools. Bayesian is better if you can stop early based on peeking; requires specialized tooling.

Why does the calculator say 'infinity' days?

+

Your daily traffic is 0 or the required sample exceeds your traffic × reasonable run length. Either increase traffic, relax your MDE, or pick bigger pages to test.