Geo test sample size calculator

The most common reason geo holdout tests fail is that they were never feasible: too few conversions, too small a holdout, too short a window. Check feasibility before you design anything. Enter your weekly conversion volume across the regions you would test, the share of volume you would hold out, and the lift you realistically expect.

Conversions per week (all test regions)

Holdout share (% of regions' volume)

Lift you expect to detect (%)

Test duration	Smallest lift you can detect (95% confidence)	Verdict
2 weeks	±21.9%	Too noisy
4 weeks	±15.5%	Too noisy
6 weeks	±12.7%	Feasible
8 weeks	±11.0%	Feasible
12 weeks	±8.9%	Feasible

Recommendation: run your test for at least 6 weeks. With 250 conversions per week and a 20% holdout, a 15% lift becomes statistically detectable around that point.

Planning approximation based on a Poisson model of weekly conversions at 95% confidence. It tells you whether a test is worth designing, not the final analysis.

How to read the result

The detectable lift shrinks as your test runs longer and collects more conversions. If the lift you expect (say 15% for a prospecting channel) is smaller than what is detectable, the test will end inconclusive no matter how carefully you run it. Rules of thumb: hold out 10 to 30% of volume, run 4 to 8 weeks, and pick channels big enough to matter. New to geo testing? Start with our free guide to geo experiments.