Case Study
Churn Model for Paid Listings
Predicting churn to retain paying agents & developers.
At eBay Classifieds (Mexico & SA), Arturo built churn model combining listing, traffic, quality, and marketing data to optimize retention strategy.
Key Results
The Transformation
The Challenge
eBay Classifieds (Vivanuncios in Mexico, plus South American markets) was spending heavily on acquisition — paid search, display ads, email campaigns — to bring real estate agents and property developers onto the platform. But paying customers were churning at rates that made the acquisition economics unsustainable. The frustrating part was that nobody could explain why.
Some agents posted a few listings and disappeared. Others were active for months, then suddenly stopped renewing. The business was treating all churn the same — sending generic win-back emails and discount offers — but the patterns suggested very different root causes. Some agents weren't getting enough traffic to justify the subscription cost. Others had listing quality issues that suppressed their visibility in search results, but they blamed the platform rather than their own content. Some were simply seasonal — real estate agents who only listed during peak buying months. Without a predictive model, the retention team couldn't distinguish between these segments or tailor their interventions. They were spending money on campaigns that didn't address the actual reasons customers left.
Our Approach
We built the churn model from the ground up, starting with the hardest part: assembling a unified dataset. Listing data lived in Hadoop, traffic and engagement metrics were in Google Analytics, customer service interactions were in ProTool, and marketing touchpoints were scattered across campaign platforms. We used Databricks to compile these into a single customer-level feature set — over 40 features per account covering listing behavior (volume, quality scores, photo counts, description completeness), traffic patterns (views per listing, contact rates, search ranking positions), marketing engagement (email opens, campaign responses), and support interactions.
The modeling work in R tested multiple approaches — logistic regression for interpretability and gradient-boosted trees for accuracy. The gradient-boosted model won on prediction quality, but we kept the logistic regression outputs because the business needed to understand which features drove churn, not just predict it. Feature importance analysis revealed that listing quality (specifically photo count and description length) was a stronger churn predictor than traffic volume — agents with poor listings blamed the platform for low traffic, when the real issue was their content.
We operationalized the model by building Tableau dashboards that scored every paying customer monthly and segmented them into risk tiers. The retention team could see not just who was likely to churn, but why — and the recommended intervention differed by segment. Low-quality listers got onboarding support and listing optimization tips. Low-traffic agents got promotional boosts. Seasonal agents got re-engagement campaigns timed to their historical activity patterns.
The Outcome
The model delivered a 15% retention lift across the paying customer base within the first 6 months. But the deeper impact was strategic: the churn analysis fundamentally changed how the business thought about customer health. Instead of measuring success by new sign-ups, the team started tracking listing quality scores and engagement metrics as leading indicators.
Marketing spend became more efficient because campaigns were targeted by churn risk segment rather than blasted to the full base. The monthly model refresh meant the team caught emerging churn signals early — if a previously healthy agent's traffic dropped two months in a row, they got proactive outreach before they hit the cancellation page. The framework was later adapted for the South American markets with localized features, proving the approach was portable across geographies. The churn model became a standing analytical asset that ran monthly for over two years, continuously refined as new behavioral patterns emerged and the marketplace evolved. Perhaps most importantly, the model shifted the organization's mindset from reactive churn management to proactive customer health monitoring — a fundamental change in how the business understood and served its most valuable customers.
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.