ADVANCED ANALYTICS · PREDICTIVE MODELING
AI-Powered Customer Retention System for Company A
Company A, a telecommunications company operating in the United States, was experiencing a decline in its customer and subscriber base. This project used historical subscriber data and machine learning to identify customers at risk of churn, understand the factors influencing their likelihood of leaving, and support targeted retention decisions.
The Challenge
Company A is a telecommunications company operating in the United States. The company was experiencing a decline in its customer and subscriber base. It needed a way to identify subscribers at risk of leaving, understand the factors contributing to customer churn, and take targeted retention action before those customers were lost.
Our Approach
The project developed an AI-powered customer retention system using historical subscriber data and machine learning. The system was designed to predict churn risk, identify the factors influencing customer departures, and support proactive, targeted retention strategies.
ORIGINAL PROJECT PRESENTATION
Complete 14-slide presentation
Slide 1 of 14
Project Summary
Historical subscriber data was analyzed to identify patterns associated with customer churn. The project progressed from exploratory churn analysis and feature engineering to an XGBoost classification model designed to identify subscribers with elevated churn risk.
The model was evaluated on a 15,000-customer test set using a decision threshold of 0.330. The presentation reports a ROC-AUC of 0.695, recall of 91.4%, precision of 55.2%, and an F1 score of 0.688.
The resulting system was designed to support proactive retention by identifying high-risk subscribers, explaining the factors contributing to their churn risk, and helping prioritize targeted retention actions.
PROJECT CONTEXT
What the Analysis Demonstrates
The analysis moves from the customer churn problem and exploratory findings to a predictive early-warning system for identifying subscribers at risk of leaving. It describes feature engineering, an XGBoost model, a 15,000-customer test set, a 0.330 decision threshold, customer-level explanations, and suggested retention actions tied to identified risk factors.
The presentation also contains its own references and a commercial quotation. Those slides are retained in the viewer in their original order and wording.
The model was evaluated on its ability to distinguish churn risk and identify at-risk subscribers for retention intervention.