Imagine you’re running a telecom business. Every time a customer leaves, it’s not just a lost sale — it’s money you spent on marketing, onboarding, and support gone out the window. Worse, it's expensive to bring in someone new.
So what if we could predict who’s likely to leave, understand why, and take action before it’s too late?
That’s exactly what this project aims to do: understand customer churn behavior, build a predictive model, and recommend practical, real-world actions to retain customers and reduce losses.
"How can we predict whether a customer will churn based on their usage patterns, demographics, and contract information — and what can we do about it?"
- Identify the key factors behind churn
- Build a machine learning model that can flag high-risk customers
- Provide business insights to drive smarter retention campaigns
After exploring the data visually using Excel and Power BI:
- Customers with month-to-month contracts churn the most (~45%)
- Higher monthly charges are strongly linked to churn
- New customers (tenure < 12 months) churn far more than older ones
- Gender has no significant effect on churn
I used a Logistic Regression model to predict churn.
- Accuracy: 78.25%
- ROC-AUC: 0.683
- Precision: 0.62
- Recall: 0.47
Here’s what the model says influences churn the most:
| Feature | Impact on Churn |
|---|---|
| Month-to-month Contract | 🔺 Increases churn risk |
| Short Tenure (< 12 months) | 🔺 Strong churn indicator |
| High Monthly Charges | 🔺 Related to dissatisfaction |
| No Tech Support | 🔺 Less sticky, churns more |
| Two-Year Contract | 🔻 Strong retention anchor |
These patterns give us a roadmap for action.
Customers on month-to-month plans are unstable. Offer discounts to switch to 1-2 year contracts — lower churn, higher lifetime value.
Most churn happens within the first year. Run loyalty offers or bonus packages during onboarding.
Churn increases with higher bills. Offer bundled or usage-based pricing to ease billing pressure.
Users without support or security services churn more. Bundle these during signup or as low-cost add-ons.
Your dashboard clearly shows:
- “Churn is highest among month-to-month contract customers.”
- “Customers in their first year have 2x churn risk.”
- “Higher monthly charges correlate with higher churn.”
Two slicers (Contract + Internet Service) allow decision-makers to interact and filter segments quickly.
“This project isn't just about numbers — it's about understanding behavior. We can now tell who’s likely to leave, why, and exactly what we can do to keep them. And we can show it in a dashboard, backed by a model that explains itself.”
“This means fewer lost customers, more targeted retention campaigns, and better customer lifetime value — with data to prove it.”
“The model is better at identifying customers who will stay, but it misses more than half of those likely to churn. It still gives us a useful head start — we can catch around 47% of churners and act on them with precision ~62%.”
- Excel – Data cleaning, summary analysis, power query
- Power BI – Visual dashboard
- Python – pandas, scikit-learn, matplotlib
- Jupyter Notebook – Modeling + feature importance
To enhance this churn analysis and make it more actionable in a real-world scenario, the following ideas could be implemented:
- Merge with Customer Support Logs Understand if churn correlates with service issues, complaint frequency, or resolution delays.
- Integrate with Payment History Identify churn patterns among customers with late payments, billing errors, or missed auto-pay setups.
-
Move beyond basic logistic regression by implementing:
- Random Forest for better handling of feature interactions
- XGBoost for higher accuracy and model interpretability
- SHAP Values to explain model predictions at the customer level
-
Create an interactive web app using:
- Streamlit or Flask for real-time churn prediction
- A dashboard where users can input customer data and receive churn risk scores with explanations
- Integration with Power BI or Tableau for executive-friendly insights