Integrating Business Intelligence and CRM Systems With a Machine Learning Approach for Predictive Customer Retention in E-Commerce.

In the rapidly evolving e-commerce landscape, retaining existing customers has become more cost-effective and strategically important than acquiring new ones. This study proposes a data-driven framework that integrates business intelligence (BI) tools, machine learning, and customer relationship management (CRM) decision support to improve predictive customer retention. The framework was developed using the publicly available Brazilian E-Commerce Public Dataset (Olist), which contains more than 100,000 orders and includes transactional, payment, delivery, product, and customer-review information. After SQL-based integration and feature engineering, customer segmentation was performed using K-means clustering on recency, frequency, monetary (RFM) variables, identifying three behavioral groups: loyal, at-risk, and occasional customers. For churn prediction, Random Forest and XGBoost classifiers were trained on customer-level behavioral, satisfaction, and service-related features. XGBoost achieved the best overall performance, with accuracy = 0.81, precision = 0.79, recall = 0.83, F1 - score = 0.81, and AUC = 0.85, outperforming Random Forest (accuracy = 0.76, precision = 0.74, recall = 0.71, F1 - score = 0.72, and AUC = 0.76). The resulting segmentation and churn scores were then exposed through Power BI dashboards and mapped into a proof-of-concept CRM decision framework for retention planning. Unlike studies that treat BI, machine learning, or CRM in isolation, this research presents an end-to-end analytical pipeline that links data preparation, predictive modeling, dashboard-based decision support, and scenario-level CRM action design. The framework provides a reproducible basis for e-commerce retention analytics and a practical foundation for future live deployment and A/B-tested CRM validation.