eBay Classifieds Mexico & South America
Churn Model for Paid Listings
Churn model across listing, traffic, and marketing signals to retain paying users.
Founder’s track record · built at eBay Classifieds (Vivanuncios Mexico & South America). The in-house operator experience Clarivant is built on.
- customer features unified in Databricks
- 40+
- customer features unified in Databricks
The starting point
eBay Classifieds was spending heavily to acquire real-estate agents and developers, but paying customers churned at rates that broke the acquisition math — and nobody could explain why. The business sent the same generic win-back to everyone, masking very different root causes: too little traffic, poor listing quality, or simple seasonality.
The method
I assembled a unified, customer-level feature set in Databricks — 40+ features across listing behavior, traffic, marketing, and support, pulled from Hadoop, Google Analytics, and ProTool — then modeled churn in R, using gradient-boosted trees for accuracy and logistic regression for the ‘why.’ Feature importance showed listing quality, not traffic volume, was the stronger predictor. Tableau scored every paying customer monthly into risk tiers, each with a tailored intervention.
The result
The model lifted retention 15% within six months and changed how the business thought about customer health — tracking listing quality and engagement as leading indicators instead of just counting sign-ups. Monthly refreshes caught early signals, marketing spend got targeted by risk segment, and the framework ran for over two years and ported to the South American markets.