Retrieval-Augmented Generation for Automated ERP Decision Support

Enterprise Resource Planning (ERP) systems manage extensive volumes of structured business data related to inventory, finance, sales, and operational activities. Although these systems provide reliable and centralized information, accessing actionable insights often depends on predefined reports or technical query mechanisms, which limits usability for non-technical users. Recent progress in Large Language Models (LLMs) has enabled natural language interaction with data; however, their direct use in ERP environments raises concerns related to hallucinated outputs, data security, and insufficient domain awareness. This study examines the application of Retrieval-Augmented Generation (RAG) as a controlled mechanism for supporting natural language-based decision-making in ERP systems. An ERP-aware RAG architecture is presented that integrates intent-driven query processing, structured data retrieval from ERP databases, and grounded response generation. The proposed approach emphasizes reliability, explainability, and role-based access enforcement to ensure trustworthy interaction with enterprise data. The findings indicate that RAG can enhance ERP usability while preserving governance, accuracy, and data integrity.