Using Survival Analysis to understand customer retention
summary
Uses Kaplan–Meier and Cox models to turn partial subscriptions into concrete LTV inputs, running with lifelines and scikit-survival, encoding high‑cardinality features (breeds by median survival) and scoring with the concordance index. Claims work on a customer base of 85,000 customers feeding four million meals/month and projects sales around twenty million; email engagement multiplies hazard by ~0.63. Pipeline produces weekly segmented retention estimates that plug into LTV and marketing decisions, and delivers actionable predictive segments rather than vague examples.
AI-written summary · about
chapters (10)
- 0:00Survival analysis intro
- 3:36Lifetime value in subscriptions
- 7:03Binary retention limits
- 8:29Survival analysis purpose
- 13:24Kaplan-Meier estimation
- 15:57Cox proportional hazards
- 18:00Feature engineering
- 21:52Model performance metric
- 24:26Improving retention
- 25:46Q&A
related talks
52mSurvival Analysis of Customer Retention using Postgres, SQL and R
31mHow to Validate you Client Churn Model
36mCustomer Lifetime Value: Models, Metrics and a Multitude of Uses
27mPreventing churn like a bandit
36mImplementing and Training Predictive Customer Lifetime Value Models in Python
01h 24mBayesianism and Survival Analysis
52mSQL Server Predictive Analytics: Customer Stories
45mKnow Your Customer