An Eiv-based Predictive Segmentation Framework for Retail CRM Optimization: Machine Learning Benchmarking and Causal Field Experimentation For Email Conversion Uplift

Abstract

Identifying customers who are most likely to convert following marketing communication remains a central challenge in retail customer relationship management (CRM), particularly in omnichannel environments characterized by fragmented behavioral signals. While machine learning–based propensity models have demonstrated strong offline predictive performance, limited empirical evidence exists regarding their effectiveness when deployed in real-world marketing campaigns. This study proposes an EIV-based predictive segmentation framework that integrates Engagement (email interaction), Intent (website activity), and Value (transaction history) into a unified customer representation, and evaluates its impact through both offline model benchmarking and online causal field experimentation. Using customer-level omnichannel behavioral data from the Indonesian retail sector, five algorithmic families—Logistic Regression, Random Forest, Gradient Boosting Machines, Artificial Neural Networks (ANN), and Deep Neural Networks (DNN)—are trained and evaluated under an identical experimental protocol. Given the highly imbalanced nature of email conversion data, model performance is assessed using precision, recall, F1-score, and Matthews Correlation Coefficient (MCC). The results indicate that the Gradient Boosting Machines model delivers the most stable and robust predictive performance across evaluation metrics and is therefore selected for operational deployment. The selected model is implemented in a large-scale field experiment involving 59 email campaigns across 15 retail e-commerce websites. Customers are randomly assigned to either a predictive segment or an existing operational segmentation, with conversion defined as a completed transaction within seven days of email exposure. The experimental results demonstrate that predictive segmentation consistently yields higher conversion rates and statistically significant uplift across campaigns and SBU. This study contributes to an end-to-end methodological framework that bridges offline predictive modeling and online causal validation, providing empirical evidence that unified, propensity-based segmentation can enhance CRM effectiveness in real-world omnichannel retail settings.