AI Enabled ERP-CRM Integration for Predictive Customer Analytics

Abstract

Enterprise Resource Planning (ERP) and Customer Relationship Management (CRM) systems provide complimentary insights on customer‘s transactions, interaction and engagements. However, fragmented enterprise data, lack of cross-system relationship learning, limited behavior modeling over time and weak interpretability of prediction hinder reliable predictive customer analytics. To overcome these obstacles, this research introduce a proposed technique called Cross-System Temporal Customer Intelligence Engine (CSTCIE), an integrative solution aimed at enhancing customer ‘s churn prediction based on unified enterprise data. The CSTCIE method uses probabilistic entity resolution for alignment of customer records, Dual Multilayer Perceptron (MLP) encoders for ERP and CRM feature representation and cross-attention to learn relations between operational and engagement data. The Transformer learns customer's behavior dynamics, Light Gradient Boosting Machine (LightGBM) predicts churn probability and SHapley Additive exPlanations (SHAP) finds influential features in prediction. Accuracy of 96.40% and Receiver Operating Characteristic–Area under the Curve (ROC-AUC) of 91.30% from experiments performed as part of the manuscript were found, indicating promising classification and churn discriminant performance. Recall and precision-recall performance, however, should be further improved. In conclusion, CSTCIE represents a promising integrative and interpretable approach to customers churn analysis.