Skip to content

Framework · Measurement

Sample size, minimum detectable effect and weeks to read

Per arm, n is about 16 x p(1 - p) / d^2 at alpha 0.05 and 80% power, where d is the absolute change to detect. Weeks to read add enrollment, metric maturation and any app-release lag. A step is Ready at 6 weeks or less, Slow at 6-10 and Underpowered beyond 10, where bigger bundles, proxies, triggered analysis or the low-traffic loop take over.

ValueDiscoverStartReturnStayPaySpread

Plain label: can this step be tested in timeAlso known as: MDE, power analysis, weeks to read, detectability, CUPED

Worked examples (alpha 0.05, 80% power, relative lifts). A 5% baseline needs roughly 30,000 users per arm to detect a 10% lift and over 120,000 for a 5% lift. A 40% baseline needs about 600 per arm for a 20% lift. Halving the lift you want to detect roughly quadruples the sample.

Maturation. A 7-day activation metric adds a week; a 60-day repeat-purchase metric adds about 8.6 weeks.

Small products. Design for 10-20% MDEs on single changes, or a larger one for a bundle on an observed broken step; use higher-baseline metrics earlier in the funnel; analyze only users who reach the changed step; reduce variance with pre-period data (CUPED cut variance about 50% at Bing Evidence grade A. Robust evidence: replicated, peer-reviewed research.).

High stakes. Pricing, paywall and billing tests use alpha 0.01.

Sources

Source IDs: F26

Book content is paraphrased, never quoted at length. Last verified October 24, 2026.

Your product

Find the stage your product is stuck in

Gamify It reads your product, picks the one move that matters at its stage, and sizes the test to your traffic. Connecting is free; a Pro subscription starts it.