Show-to-Purchase CR — Confidence Intervals across significance levels

Зелёные ячейки — интервалы НЕ пересекаются (тест значим на этом уровне). Формула: CR ± z·√(CR/N).

ARPU — Confidence Intervals across significance levels

Формула: ARPU ± z·SE, где SE = STDEV(daily ARPU) / √n_days.

Method Comparison — Verdict Triangulation

Method 1 — analyst's CI overlap (formula: CR ± z·√(CR/N), ARPU ± z·SE_daily). Conservative: CIs don't overlap → significant. Method 2 — Wilson CI + two-proportion z-test (industry standard for proportions, p < 0.05 = significant). Method 3 — Bayesian Beta-Binomial for CR + Bootstrap for ARPU (Adapty/RevenueCat-style). P(B > A) directly answers "is B better".
Methods agree → high confidence. Methods disagree → ambiguous, need more data or look at segments.

Banner Efficiency — Per-day per-product breakdown

Sources