Data Science

Dealer Churn Prediction

Strategizing retention improvement by estimating each subscribed dealer's propensity to churn

Machine LearningPropensity ModelingPredictive ModelingClassificationFeature EngineeringClickstream DataMarketing AnalyticsCustomer AnalyticsDashboardRExcelPowerpointTechnology

Problem Statement

A US-based e-classifieds company was experiencing high churn of dealers subscribed to their platform and wanted to explore ways to improve retention. A churn propensity predictive model was required to identify dealers that are at high-risk of churning and devise intervention measures accordingly.

Challenges

  • Making churn propensity scores and insights easily interpretable for non-technical sales representatives
  • Processing and analyzing massive clickstream data to identify meaningful churn drivers while filtering out noise.
  • Ensuring that predictions are generated early enough for the sales team to take proactive intervention measures before dealers churn.

Solution

Summary

As the lead consultant and client point-of-contact on the project, I managed analytics & predictive modeling workflow in which we designed a supervised machine learning solution to predict dealer churn. Each dealer was assigned a churn propensity score based on the model probabilities and further descriptive analytics was conducted to profile each dealer. The propensity model was deployed on-premises to refresh the propensity scores periodically. A dashboard was deployed on the client's Salesforce platform to give reps access to insights.

Approach

  • Brainstorming and stakeholders interviews to build hypotheses around potential drivers of churn
  • Gathering, cleaning and exploratory analysis of clickstream data including dealer ads, dealer interactions with leads, sales calls and dealer login activity
  • Differentiating dealers by segment and building separate churn propensity models for each segment using Logistic Regression
  • Visualizations and dashboard design for end-user consumption
  • Prioritization framework and identification of areas of highest opportunity
  • System integration of deliverables including model deployment and dashboard deployment on the client's Salesforce CRM

Deliverables

  • Predictive model to estimate churn propensity
  • ML pipeline in production for periodic data refreshes
  • Dashboard deployed using Salesforce Canvas
  • Periodic Insights Reports
Modeling Approach

Modeling Approach

Prioritization Framework

Prioritization Framework

Results & Impact

From 40%+ to <30%

Estimated dealer churn reduction

Have a similar challenge?

Let's discuss how we can develop solutions for your specific use case.

More Case Studies

Dealer Churn Prediction | Case Study