Customer Churn & Retention Analytics
RFM customer segmentation, churn prediction with Gradient Boosting (87% AUC-ROC), cohort retention analysis, and revenue-at-risk calculation with an interactive Streamlit dashboard.
Simulation Note: This project uses a synthetic/sample dataset. Financial impact, savings, ROI, and SLA-cost estimates are modeled scenarios for analytical demonstration and do not represent actual NTC financial figures.
Modeled Estimate (synthetic data)
Interactive Dashboard
RFM Customer Segments
Revenue Share by Segment
Churn Rate by Tier
Feature Importance for Churn Prediction
Business Problem
A telecom or retail company is losing high-value customers. Leadership wants to know who is at risk, why, and what retention actions will protect the most revenue.
Methodology
Data Preparation
Loaded customer transaction data with RFM attributes, membership tiers, purchase history, and churn labels
RFM Segmentation
Calculated Recency, Frequency, and Monetary scores; segmented customers into Champions, At Risk, Can't Lose Them, and other groups
Predictive Modeling
Built and compared Logistic Regression, Random Forest, and Gradient Boosting classifiers; Gradient Boosting achieved 87% AUC-ROC
Dashboard Deployment
Deployed multi-page Streamlit dashboard with RFM analysis, churn prediction, cohort retention, and revenue-at-risk views
Key Insights
Champion Segment Value
Champions (8% of customers) generate 35% of revenue --- highest retention priority. Platinum tier customers have the lowest churn rate (8%) but highest CLV ($15,000+).
At-Risk Revenue Exposure
At Risk segment contains 15% of customers with 20% of revenue at stake. Can't Lose Them segment has the highest average spend but 60% churn rate.
High-Accuracy Churn Prediction
Gradient Boosting model achieves 87% AUC-ROC for churn prediction. Top predictive features: days since last purchase (30%), total spend (25%), frequency (20%).
Revenue at Risk
$2.8M+ annual revenue at risk from high-churn-probability customers. Basic tier has 35% churn rate --- opportunity for tier upgrades.
RFM-Churn Correlation
RFM score shows -0.65 correlation with churn (higher score = lower churn). Mobile app users show 25% lower churn than online users.
Business Recommendations
Modeled Estimate (synthetic data)Targeted Win-Back Campaigns
Focus on At Risk + Cant Lose Them segments. $150K investment for 3,000 customers. Expected recovery: $420K (28% retention rate).
Tier Upgrade Program
Focus on Basic to Silver upgrades. $100K investment for 2,000 customers. Expected impact: Increase CLV by 40%.
Champion Retention Program
VIP rewards for Champions segment. $50K investment for 400 customers. Expected impact: Maintain 95% retention rate.
Channel Optimization
Invest in mobile experience and push notifications. Mobile app users show 25% lower churn than online users.
Technical Highlights
RFM Segmentation
- ▸Recency, Frequency, Monetary scoring
- ▸Customer lifecycle classification
- ▸Revenue concentration analysis
Machine Learning
- ▸Gradient Boosting (87% AUC-ROC)
- ▸Feature importance ranking
- ▸Churn probability scoring
Statistical Testing
- ▸Chi-Square for tier vs churn
- ▸Spearman correlation for RFM
- ▸Mann-Whitney U for spend comparison
Dataset
Customer transaction dataset with RFM attributes, membership tiers, purchase history, and churn labels.
Project Outcome
Modeled Estimate (synthetic data)Delivered actionable customer segmentation, a high-accuracy churn prediction model, and retention strategies projected to recover $420K+ annually through targeted win-back campaigns, tier upgrades, and champion retention programs.
Limitations & Assumptions
Synthetic churn labels - modeled performance
The 87% AUC-ROC was measured against synthetic churn labels. No per-class precision/recall breakdown was computed, and class balance in the generated data is favorable; performance on real customer data would likely differ.
Revenue figures are modeled
Revenue at risk ($2.8M+), projected recovery ($420K+) and all ROI estimates are modeled from synthetic transaction data and assumed campaign parameters, not observed revenue outcomes.
Segments reflect the synthetic snapshot
RFM segments, tier churn rates and CLV ($15K+ Platinum) describe this one synthetic dataset. Segments were not validated against external sources or tested via live campaigns.
No post-launch validation
Retention strategies are recommendations only. No A/B test or before/after measurement exists to confirm the projected recovery rates.