Analyzed customer churn patterns for StreamWorks Media, a UK-based streaming platform competing with Netflix & Prime Video. Using data from 1,500 subscribers across 6 countries, I built predictive models to identify churn drivers and improve retention.
🏆 Key Finding : Low watch hours → 3× higher churn risk.
⚙️ Business Problem : 📉 Churn Rate: 23.4% (351/1500)
✅ Identify why customers leave
✅ Predict who’s likely to churn
✅ Recommend retention actions
| Category | Description |
|---|---|
| Records | 1,500 subscribers |
| Regions | UK, US, Canada, India, France, Germany |
| Period | October 2025 |
| Features | Age, Gender, Plan Type, Monthly Fee, Watch Hours, App Usage %, Promotions, Referrals, Complaints |
| Target | Churn Status (Active / Churned) |
1. Data Cleaning & Preparation
✅ Converted date fields to proper datetime format
✅ Handled missing values using median imputation (10% missing in monthly fees)
✅ Standardized categorical variables for analysis
✅ Created binary encodings for machine learning models
2. Feature Engineering
✅ Created strategic features to enhance predictive power :
Tenure Days : Customer lifetime with the platform
Loyalty Flag : Customers with 6+ months subscription
Watch-to-Fee Ratio : Value perception metric
Heavy Mobile User Flag : Mobile-dominant behavior indicator
Age Groups : Teen, Young Adult, Adult, Mid-age, Senior
Watch Time Segments : Low, Medium, High, Very High
3. Statistical Analysis
T-Tests : Compared numerical variables between churned and active users
Chi-Square Tests : Examined relationships between categorical variables and churn
Correlation Analysis : Identified key variable relationships
4. Predictive Modeling
Logistic Regression : Churn prediction model
Linear Regression : Watch time prediction model
Class Balancing : Addressed dataset imbalance for accurate predictions
a) Churn Drivers (Statistical Significance)
✅ Low Engagement is Critical
✅ Churned users watched significantly fewer hours than active users
✅ Average watch hours is the strongest churn predictor
b) Tenure Matters
✅ Long-term customers (6+ months) have dramatically lower churn rates
✅ New subscribers represent the highest risk segment
c) Premium Users Stay Longer
✅ Premium subscribers churn the least
✅ Basic plan users show highest cancellation rates
d) Promotions Work
✅ Users receiving promotional offers are significantly less likely to churn
✅ Referrals also improve retention rates
e) Price is NOT a Factor
✅ No significant difference in monthly fees between churned and active users
✅ Engagement and satisfaction matter more than price
f) High-Risk User Profile
✅ Low watch time (< 15 hours/month)
✅ Short tenure (< 90 days)
✅ No promotional offers received
✅ Basic subscription tier
✅ High mobile-only usage
| 🔍 Finding | 💬 Action |
|---|---|
| 📉 Low engagement = 3× higher churn | Personalized onboarding, watch reminders |
| ⏱️ Tenure < 90 days = high churn | Early retention campaigns |
| 💰 Price not significant | Focus on engagement over pricing |
| 🎁 Promotions reduce churn by 35% | Target new/basic users |
| 📱 Heavy mobile users churn more | Improve mobile UX & notifications |
-
🎬 Engage low-watch users in first 30 days
-
💎 Reward loyal & premium users (exclusive content)
-
📲 Optimize mobile app experience
-
🎯 Target new/basic users with promo offers
✅ Python 3.8+
✅ pandas - Data manipulation and analysis
✅ numpy - Numerical computations
✅ scikit-learn - Machine learning models
✅ matplotlib & seaborn - Data visualization
✅ scipy - Statistical testing
✅ Jupyter Notebook - Interactive analysis environment
✅ Statistics - T-tests, Chi-square tests, correlation analysis
✅ Clear understanding of churn drivers
✅ Predictive model for at-risk users
✅ Estimated 25–30% churn reduction
✅ £500K+ potential annual savings
Charu Madaan
📧 Email: charumadaan88@gmail.com
💼 LinkedIn: linkedin.com/in/charumadaan
🔗 Portfolio: github.com/charumadaan
Acknowledgments
Project completed as part of UpTrail Data Analytics Programme
Dataset and business case provided by UpTrail