Case Study
Churn Model for Paid Listings
Predicting churn to retain paying agents & developers.
At eBay Classifieds (Vivanuncios in Mexico, plus South America), our founder built a churn prediction model for paid real estate agents and car dealers. By compiling data from Hadoop, Databricks, Google Analytics, and ProTool into a unified modeling dataset, he identified the behavioral patterns preceding cancellation — enabling the marketing team to intervene before churn happened. The model delivered a 15% retention lift and was operationalized with monthly reruns and Tableau dashboards for GMs.
Key Results
The Transformation
The Challenge
eBay Classifieds' paid customer base — real estate agents and car dealers who purchased premium listing packages — was churning at rates that undermined marketing spend:
- Paying customers were canceling subscriptions despite heavy acquisition ad spend, making the unit economics of paid listings unsustainable
- No visibility into why customers churned — the team couldn't distinguish between customers who left due to poor results versus those who simply weren't engaged
- Customer data was scattered across four systems: listing activity in Hadoop, processing in Databricks, web behavior in Google Analytics, and internal quality metrics in ProTool
- Marketing was running retention campaigns with no targeting — same message to every at-risk customer regardless of their specific churn signals
Our Approach
**Data Compilation & Feature Engineering:**
- Built a unified modeling dataset combining listing activity (Hadoop), data processing pipelines (Databricks), web traffic and engagement (Google Analytics), and listing quality scores (ProTool)
- Engineered features capturing the behavioral signals that preceded churn: declining listing views, reduced login frequency, dropping listing quality scores, and marketing campaign response patterns
**Churn Model Development:**
- Developed predictive churn model in R, testing multiple approaches to identify which combination of features best predicted cancellation 30-60 days in advance
- Validated model performance against historical churn data to establish reliable prediction thresholds
**Operationalization & Reporting:**
- Deployed model on a monthly rerun cycle, generating refreshed churn risk scores for every paid customer
- Built Tableau dashboards for the GM and marketing team showing at-risk accounts segmented by churn driver — enabling targeted retention campaigns instead of blanket outreach
The Outcome
**Retention & Revenue:**
- 15% retention lift among paid customers — the model identified at-risk accounts early enough for marketing to intervene with targeted offers
- Optimized retention campaign spend by focusing resources on customers with the highest probability and value of save
**Operational Intelligence:**
- Marketing team shifted from blanket retention campaigns to segmented outreach based on specific churn drivers (quality issues, engagement decline, poor listing performance)
- Monthly model reruns gave GMs across Mexico and South America a consistent view of subscriber health
- Listing quality improvements followed as the model revealed which quality factors most strongly predicted churn
Want similar results for your organization?
Get in touchReady to turn data into decisions?
Let's discuss how Clarivant can help you achieve measurable ROI in months.