Utility Optimized Anti Money Laundering Detection Using ISO 20022 Trade Graphs Conformal Graph Neural Networks and SHAP

Digital finance is accelerating the volume, velocity, and complexity of global money movement through ISO 20022 migration, instant cross-border payments, and the digitization of trade documentation such as electronic invoices and bills of lading. Yet most anti–money laundering (AML) systems continue to rely on fixed thresholds that generate excessive false positives and impose high operational costs, with downstream consequences for financial inclusion and cross-border corridor access. We introduce a regulatory-utility framework that formulates AML triage as an economic optimization problem, maximizing expected suspicious-activity-report (SAR) yield under review-capacity and cost constraints. The system fuses ISO 20022 payment messages, Know-Your-Customer (KYC) attributes, and documentary-trade records into a single heterogeneous graph processed by a message-aware graph neural network (GNN), and applies calibrated probabilities with conformal prediction for auditable uncertainty control. SHAP-based subgraph rationales provide transparent, investigator-ready explanations. Experiments on privacy-preserving synthetic datasets reflecting cross-border payments and trade-based money-laundering typologies show that utility-optimized triage improves Utility@K by 18–35% and SAR@K by 12–22%, while reducing false positives per 1,000 investigations by 23–41 compared with rules-based or tabular baselines at equal workload. Linking payments to trade documents substantially increases TBML-ring discovery and maintains stable conformal coverage under distribution drift. A corridor-level welfare simulation shows that the framework also lowers compliance cost per benign digital transaction and reduces de-risking pressure, enabling broader participation in high-risk payment corridors. We release deployment artifacts, including model cards, threshold-governance frontiers, and challenger logs, to support transparent, reproducible, and digitally native AML across financial institutions and supervisory agencies.