Overview

Homepage Redesign Experiment — 14 days, 247,892 visitors

Live

Total Visitors

247,892

+12.3%

Total Conversions

10,784

+17.5%

Relative Lift

+17.5%

CI: [12.1%, 23.2%]

Statistical Sig.

p < 0.001

Power: 98.2%

Daily Conversion Rate

Cumulative Lift Over Time

Conversion Funnel

Performance by Channel

Device Breakdown

A

Control Group

50% traffic — Original design

Visitors

123,946

Conversions

4,958

Conversion Rate

4.00%

B

Variant Group

50% traffic — New design

Winner

Visitors

123,946

Conversions

5,826

Conversion Rate

4.70%

+17.5% lift

Bayesian Posterior Distribution

Confidence Interval Over Time

Statistical Summary

# Two-Sample Z-Test for Proportions Control (A): n=123,946 conversions=4,958 rate=4.00% Variant (B): n=123,946 conversions=5,826 rate=4.70% Absolute Lift: +0.70 pp Relative Lift: +17.50% 95% CI (Abs): [0.48pp, 0.92pp] Z-Statistic: 8.42 P-value: < 0.0001 *** Power (1-beta): 98.2% Min. Sample Size: 24,800 per group (achieved: 123,946) # Conclusion: Reject H0. Variant B significantly outperforms Control.
Logistic Regression Incrementality validation: Does the treatment have a genuine causal effect on conversion, controlling for covariates?
# Model: conversion ~ treatment + age_group + device + channel + hour_of_day + is_returning AIC: 142,387 BIC: 142,501 Pseudo R2: 0.0847 N: 247,892

Coefficient Plot (Log-Odds)

Odds Ratios with 95% CI

ROC Curve

Regression Table

VariableCoeffStd ErrzP>|z|OR
treatment (B)0.1820.0218.67<0.0011.200
age: 25-340.3420.02812.21<0.0011.408
age: 35-440.2870.0319.26<0.0011.332
device: mobile-0.1560.024-6.50<0.0010.856
channel: organic0.4210.02616.19<0.0011.524
channel: paid0.1980.0296.83<0.0011.219
is_returning0.5340.02224.27<0.0011.706
hour: peak (12-14)0.1120.0333.390.0011.119
Causal Forest (GRF) Heterogeneous treatment effect estimation using Generalized Random Forest
# grf::causal_forest(X, Y, W, num.trees=2000, honesty=TRUE) ATE (Average Treatment Effect): 0.0070 (SE: 0.0008) # +0.70pp conversion lift CATE Range: [-0.002, 0.031] # strong heterogeneity detected Variable Importance (top 3): is_returning (0.34), age (0.28), channel (0.19)

CATE Distribution

Variable Importance

Uplift Curve (AUUC)

Subgroup Treatment Effects

Treatment Effect by Age

Treatment Effect by Channel

Conversion by Device

Hourly Conversion Pattern

Targeting Recommendations

High Uplift

Returning users, Age 25-34, Organic

CATE: +3.1pp | 18% of traffic | Expected +892 conversions

Medium Uplift

New users, Age 35-44, Paid Search

CATE: +1.2pp | 24% of traffic | Expected +712 conversions

Low / Negative

Mobile, Age 45+, Social

CATE: -0.2pp | 12% of traffic | Consider excluding