About

A comprehensive predictive analytics investigation into Zerve platform user retention, this canvas analyzes 4 months of event logs (Sep–Dec 2025) to identify which first-week behavioral signals predict long-term success. The workflow flows from data loading and preparation through success label definition (composite 5-criterion proxy), feature engineering (16 early-stage behavioral indicators), and dual machine learning models (Gradient Boosting and Logistic Regression) to rank behavioral predictors, ultimately revealing that AI tool diversity (≥3 distinct tools), verification behavior (run_block after agent calls), and entry pattern (AI-first vs. observer-first) are the strongest success signals—corroborated by extensive EDA including session metrics, user archetypes, event co-occurrence, cohort retention, and device/geographic segmentation.