Customer Churn & Revenue Risk Analysis
Marketing Analytics Capstone • CSU East Bay
Model accuracy
81%
Logistic regression
AUC score
0.86
Strong classification capability
Dataset size
7K+
Customer records
Baseline churn rate
26%
~1 in 4 customers
Marketing Analytics Capstone · MKTG 693 · Professor Yi He · California State University, East Bay · Grade: A
OVERVIEW
Led a comprehensive churn and revenue risk analysis using the Kaggle Telco Customer Churn dataset, 7,000+ customer records across demographics, services, billing behavior, and contract type. The objective was to build a predictive framework to identify customers at high risk of churn, quantify potential revenue exposure, and deliver actionable retention strategies grounded in statistical modeling and data visualization. This project simulated a real-world retention analytics engagement for executive review.
WHAT I DID
Data preparation & feature engineering
Cleaned and transformed 7,000+ records using Python/Pandas. Addressed missing values, standardized categorical variables, and engineered predictors including tenure bands, contract structure, and billing behavior.
Exploratory data analysis
Analyzed churn distribution across contract type, tenure, internet service, payment method, and monthly charges. Used pivot tables and segmentation to uncover early revenue leakage signals.
Predictive Modeling
Built a Logistic Regression model using Scikit-learn with train/test split validation. Evaluated with ROC Curve and AUC. Achieved AUC = 0.81 after correcting data leakage that was producing false 100% accuracy.
Revenue risk quantifications
Calculated predicted churn probability per customer and estimated revenue at risk by combining churn probability with monthly recurring revenue. Segmented customers into Low, Medium, and High risk tiers.
CUSTOMER RISK TIERS
Low risk
Long tenure · Annual contracts · Auto-pay
Medium risk
Mixed contract · Manual payment · Mid-tenure
High risk
Month-to-month · Electronic check · Early tenure
KEY FINDINGS
-
AUC of 0.86 — strong model reliability; data leakage identified and corrected from initial false 100% accuracy result
-
Contract type, tenure, billing method, and monthly charges identified as primary churn drivers
-
Month-to-month contracts exhibited significantly higher churn probability than annual contracts
-
Electronic check billing correlated with elevated churn risk vs automatic payment methods
-
Developed a structured retention prioritization framework based on projected revenue exposure per customer
RECOMMENDATIONS
-
Incentivize migration from month-to-month to annual contracts — single highest-impact retention lever
-
Deploy targeted retention campaigns for customers within early-tenure risk windows (first 12 months)
-
Implement billing optimization strategies — auto-pay incentives to migrate electronic check customers
-
Align CRM automation workflows with churn probability tiers to prioritize outreach efficiently
TOOLS USED