From Network Experience to Subscriber Retention: An Explainable AI Framework for Mobile Operators

Explainability matters for real-world AI: The framework generates automated English-language explanations for why individual subscribers are predicted to churn, making the AI system's decisions understandable to business stakeholders rather than treating it as a "black box."

Data quality and label noise are practical challenges: The researchers developed a "truncation heuristic" that excludes about 3% of subscribers with low-confidence predictions to protect against unreliable training data, demonstrating that responsible AI implementation requires acknowledging and addressing data limitations rather than ignoring them.

Business metrics matter more than raw data volume: Despite network data comprising the majority of the 250+ TB dataset, the framework found that subscriber commercial and experience indicators provided "stronger churn signals," suggesting that responsible AI requires prioritizing meaningful features over simply using all available data.

AI can reveal nuanced customer behavior: The analysis uncovered that high-value customers face "gradual value leakage rather than outright attrition," a finding that wouldn't emerge from simpler approaches and could help companies develop more targeted retention strategies.

Responsible AI requires industry-specific design: The framework was built on an "industry reference architecture" and validated on real production data from a major telecom operator, illustrating that responsible AI systems must be thoughtfully designed for specific business contexts rather than applied as generic solutions.