Although the Collections and Disbursements (FS-CD) module in SAP S/4HANA is essential to insurance finance operations, it is still limited by rule-based decision support, which cannot handle complex payment patterns or utilise unstructured text. To automate invoice prioritisation, payment-matching exception handling, and dunning-letter creation on SAP systems, this paper proposes a modular, event-driven framework that combines three AI-powered components: an LLM narrative generator, an XGBoost-based predictive risk-scoring model, and an Isolation Forest-based anomaly-detection engine. We simulate the core FS-CD tables and add policy PDFs and email threads using the open-source SAP-samples repository. With automation rates surpassing 75%, the AI-augmented pipeline demonstrated superior ROC-AUC, precision, and expert coherence in empirical evaluations compared to rule-based and manual baselines across representative workloads. These results indicate that our approach elevates predictive accuracy and narrative quality and drives scalable, real-time efficiency in SAP FS-CD workflows.