Causal Inference and AI for churn reduction

About this session

Causal inference is emerging as a key tool to overcome the limitations of correlation-based analyses typical of traditional machine learning. This study proposes a methodology that integrates causal techniques and LLMs to evaluate the impact of non-core services on customer retention in the telecom sector. Using causal discovery algorithms, we build a causal graph and then robustly estimate ATE and CATE from observational data, isolating confounders. LLMs automate the clustering of high-dimensional features and suggest possible relationships between nodes, improving the interpretability of causal graphs. The case study on TIM Enterprise customers analyzes the effect of a TV service on churn reduction. By combining classical methods (matching, IPW) with advanced techniques (double ML) and prior knowledge, we gain valuable insights for retention strategies. The results highlight the potential of this scalable pipeline in competitive and evolving environments. The relevance of using Google Cloud Platform is also emphasized.

Speakers

Alessandro Taglieri

Lead Data Scientist, TIM



Nicola Vizioli

Strategy Analyst & Data Scientist, TIM, TIM