Survival Models for Web-Scale E-Commerce Repurchase Prediction
August 31, 2026
Replacing binary classification with survival models allows for direct prediction of time-to-repurchase in grocery e-commerce. Testing on millions of customers shows that a Log-Normal distribution achieves an R-squared of 0.998 for marginal fit.
HOW THIS AFFECTS YOU
●
builderYou can optimize recommendation engines by predicting specific repurchase windows rather than simple binary outcomes.
●
founderUsing time-to-event modeling can provide more granular customer lifecycle insights than standard classification.