Soft-launch sample size: days and installs before a kill call

Kill calls often happen before statistical humility. Apple’s product page optimization flow even estimates whether a desired conversion lift can reach 90% confidence within 90 days given impressions — a reminder that sample size is a first-class input. This lab does not invent a universal soft-launch install target.

Define the calculation (labelled synthetic)

For a binary retention rate r = retains / installs, a simple Wilson interval at 95% confidence is:

• centre ≈ (p + z²/2n) / (1+z²/n) with p=r, z=1.96
• half-width grows quickly when n is small

Decision rule for this teaching sheet: do not call a kill on D1 retention alone while the 95% Wilson half-width still exceeds a reader-set max_halfwidth (synthetic default 0.05).

Worked synthetic example

Width readiness is not the same as product readiness — payer quality and creative leakage still matter. Numbers are synthetic.

Limits

No AppsFlyer account export was used. No industry “median kill day” is claimed.

Download the artifact.

Research checked 20 September 2026. Local draft; human editorial review pending.