A Composite AI Framework for Dynamic Operational Risk Management in Banking

Operational risk management (ORM) is central to banking stability, particularly given the rapid growth of digital transactions, open banking, and stringent regulatory mandates like Basel III and IV. Traditional ORM relies on historic loss databases, periodic risk and control self-assessments (RCSA), and static key risk indicators (KRIs). These retrospective approaches struggle to identify complex process workarounds, internal collusion, and emergent systemic failures in real time. This paper proposes a comprehensive, five-layer Composite AI framework that integrates unsupervised anomaly detection (Isolation Forest and Deep Autoencoders), Natural Language Processing (BERT), Graph Neural Networks (GCN), and process mining. This architecture enables simultaneous monitoring of structured and unstructured data, employee-customer interaction networks, and transactional workflow deviations. The framework also incorporates an Explainable AI (XAI) layer using SHAP and LIME to address the "black-box" dilemma of deep learning models, smoothing the path for regulatory compliance. This paper demonstrates that the transition from passive sampling to a proactive, continuous, and explainable AI-driven ORM is a strategic imperative for resilient banking operations. The framework provides banking professionals with practical, actionable tools to modernize their risk management practices without requiring deep technical expertise in artificial intelligence

کلیدواژه ها:

Operational Risk Management ، Composite Artificial Intelligence ، Process Mining ، Anomaly Detection ، Graph Neural Networks ، Basel Accords