Intelligent Rebate Platform Automation in Pharmacy Benefit Management: A Generative AI and Deep Learning Architecture for Healthcare Financial Optimization

Abstract

Pharmacy benefit management rebate processing operates at the intersection of contractual complexity, entity classification ambiguity, and high-stakes regulatory compliance, creating conditions under which manual and rules-based audit approaches consistently produce financial leakage and reconciliation failures. This paper presents the Intelligent Rebate Platform, a hybrid artificial intelligence architecture designed to automate and govern the full rebate lifecycle across three specialised computational layers. The Semantic Contract Engineering Layer applies a fine-tuned large language model augmented with retrieval-augmented generation to extract structured obligation data from heterogeneous rebate contracts with high accuracy and full source traceability. The Geometric Entity Resolution Layer employs a Relational Graph Convolutional Network to classify and reclassify healthcare entities across complex organisational hierarchies, replacing brittle identifier-matching logic with graph-based probabilistic reasoning. The Deep Transactional Auditing Layer utilizes a variational autoencoder to detect any abnormal behavior of claims through unsupervised reconstruction error learning and identifies the leakage channel, which cannot be identified using deterministic rules. In all the three layers, the mechanism for explainability, such as SHAP value, retrieval source, and computational graph logging provides explainability for every decision made automatically. An AIOps-governed operational layer automates event correlation, predictive alerting, and routine remediation, substantially reducing manual monitoring burden while maintaining audit trail continuity. Federated learning extensions enable cross-environment model improvement without centralising protected health information, satisfying applicable data protection obligations. The platform advances a generative AI and deep learning paradigm for healthcare financial optimisation that is simultaneously high-performing, explainable, and governance-ready.