An educational estimate, not a lab-grade power analysis. The number below comes from one standard formula with the four inputs you choose. Real sample size depends on design too — one-sided versus two-sided tests, multiple metrics or segments, uneven traffic splits, seasonality and peeking at results all change what you need. Experiment design assumptions matter.
Chance of a false positive if there is no real difference.
Chance of detecting a real difference of this size.
Current rate of the control version.
Smallest change worth detecting.
Total across both variations, for the timeline.
Sample per variation
8,155
Total sample
16,310
Variant rate needed
6.00%
Estimated days
17 days
To detect a move from 5.00% to 6.00% with 80% power at α = 0.05, each variation needs 8,155 visitors — 16,310 in total. At 1,000 daily visitors that is about 17 days, assuming traffic is split evenly between the two versions.
Formula: n per variation = (z-significance + z-power)² × [p₁(1−p₁) + p₂(1−p₂)] ÷ (p₁ − p₂)², rounded up. p₁ is the baseline rate, p₂ the variant rate after your effect size, and the z values are standard normal quantiles (1.960 for α = 0.05, 0.842 for 80% power). No continuity correction is applied. Everything runs in your browser on your inputs — nothing is uploaded or saved.