Geo-test power estimator

How big an effect could a geo experiment detect? Free, runs in your browser, and nothing you enter is stored.

Units that could be split into test and control groups.

50% is the most efficient split.

Use the same KPI you’d judge the test on.

Typical range 5% to 30%. Higher means noisier data.

0 = each day independent. Higher is more cautious.

How much units still differ in how they change over time, even after careful matching.

Fixed for this tool: significance level 5% (two-sided) and desired power 80%.

Illustrative

What this test could detect

View data

This is a planning-grade approximation. It treats units as independent, models day-to-day noise as an AR(1) pattern, and ignores spillovers, seasonality and uneven unit sizes. Matching quality (the remaining difference between matched units) often matters more than test length. A real design should be checked with simulation on your own history.

Planning aid only. Illustrative. Built to help you ask better questions, not to replace a proper analysis of your data. Nothing you enter is stored.

This is a planning aid. It’s built to help you ask better questions, not to replace a proper analysis of your data.

Want a proper answer?

Take what you’ve learned here into a conversation.