About

This project implements an end-to-end analytical system that proactively identifies customers likely to churn, quantifies the financial impact of churn risk, and generates actionable recommendations to retain high-value customers. The core business problem addressed is that many subscription-based businesses (e.g., telecom, SaaS, OTT platforms) lack systems that anticipate churn and guide prioritization of retention efforts before revenue is lost.

The solution uses machine learning combined with a business logic layer to transform raw customer data into decision-ready insights. A Logistic Regression classification model was trained to estimate the probability of churn based on customer behavior, service usage, and billing characteristics. The dataset was preprocessed by handling categorical variables through one-hot encoding and applying an 80/20 train-test split for model validation.

Validation was conducted using holdout testing with standard evaluation metrics including accuracy and F1 score. Although the dataset contained an imbalanced target class, the model demonstrated reasonable predictive performance with approximately 84% accuracy and support for identifying risk patterns.

Beyond model performance, the system computes monthly revenue and revenue at risk for each customer by multiplying churn probability with projected revenue, providing a tangible business metric. A ranking of high-risk customers and an automated recommendation engine (e.g., offer discount, send retention email) were included to guide retention strategies.

The entire workflow was deployed in Zerve as a scheduled production pipeline that automatically runs at defined intervals, recomputing churn risks and updating prioritized customer lists. This satisfies the requirement for a deployed, measurable analytical system rather than an exploratory analysis in a notebook.

This solution demonstrates a complete pipeline from raw data to a production-ready predictive system that not only generates insights but also quantifies their impact and embeds business recommendations, making it directly actionable for operational teams.