Leveraging Oracle Integration Cloud as an Agentic Tool Layer for Generative AI in Enterprise ERP Workflows: Architecture, Implementation, and Evaluation

The rapid advancement of Generative Artificial Intelligence (GenAI) presents a transformative opportunity for enterprise resource planning (ERP) systems. However, the seamless integration of GenAI capabilities into production ERP workflows remains a complex, underexplored challenge. This paper proposes and evaluates a formal architecture that positions Oracle Integration Cloud (OIC) as an agentic tool layer, enabling large language model (LLM)-based agents hosted on Oracle Cloud Infrastructure (OCI) to interact with Oracle Cloud ERP modules through structured integration patterns. We focus on a high-impact use case — AI-assisted Purchase Order (PO) automation — encompassing PO creation, exception handling, and supplier communication. The proposed architecture leverages OIC's native REST/SOAP adapters, FBDI file-based ingestion, and error-handling frameworks alongside OCI GenAI services. Through implementation and experimental evaluation, we demonstrate improvements in PO processing throughput, exception detection accuracy, and developer effort reduction compared to traditional manual and rule-based approaches. Results indicate a 62% reduction in manual processing time, a 74% improvement in exception detection rate, and a 41% reduction in integration development effort. We further discuss security, governance, and data privacy considerations critical to enterprise GenAI deployments. This work provides a reusable reference architecture and empirical insights for practitioners and researchers integrating GenAI into Oracle Cloud ERP environments.