Predicting subscriber dissatisfaction and improving retention in the wireless telecommunications industry

Area of Science:

• Statistical Machine Learning
• Telecommunications Analytics

Background:

• Intense competition in the wireless industry necessitates subscriber retention strategies.
• Customer churn, the loss of subscribers, directly impacts carrier profitability.

Purpose of the Study:

• To apply statistical machine learning techniques for predicting customer churn.
• To determine optimal incentive strategies for subscriber retention and carrier profitability.

Main Methods:

• Utilized logit regression, decision trees, neural networks, and boosting algorithms.
• Analyzed a dataset of approximately 47,000 U.S. domestic subscribers' usage, billing, credit, application, and complaint history.
• Emphasized the importance of domain-expert-crafted data representation.

Main Results:

• Predictive churn modeling and targeted incentives yield significant carrier savings.
• Effectiveness demonstrated across various cost-of-intervention and retention-rate scenarios.
• Real-world testing validated the simulation experiments' findings.

Conclusions:

• Machine learning-driven churn prediction and incentive programs are effective for wireless carriers.
• Strategic data representation is crucial for successful implementation.
• The approach offers a viable method to enhance subscriber retention and maximize profitability.