U-CAI: A Secure Conversational AI Architecture for Natural Language Interaction with Enterprise ERP Systems

Abstract Enterprise Resource Planning (ERP) systems serve as the backbone of modern organizations by integrating core business processes such as finance, procurement, supply chain management, and human resource management into unified digital platforms. Despite their benefits, ERP systems are frequently criticized for complex graphical user interfaces and rigid workflow navigation that impose high cognitive demands on users. Recent advances in Large Language Models (LLMs) enable conversational interaction with digital systems, offering new opportunities to simplify enterprise software usability. However, integrating conversational AI with enterprise ERP systems introduces challenges related to deterministic workflow execution, security enforcement, and reliable translation of natural language commands into structured system operations. This paper proposes U-CAI (Unified Conversational Artificial Intelligence), a secure conversational architecture designed to enable natural language interaction with enterprise ERP platforms. The proposed framework integrates an LLM reasoning engine, an adaptive dialogue manager, a deterministic ERP connector layer, and a security module implementing role-based access control and policy validation. The architecture translates natural language instructions into structured ERP transactions while ensuring workflow integrity and enterprise security compliance. A prototype implementation was developed using the Odoo ERP platform and evaluated through controlled experiments involving ERP users performing operational tasks across multiple modules. Experimental results demonstrate that the conversational architecture reduces task completion time by 61% and improves operational accuracy by 28% compared with traditional ERP graphical interfaces. The system achieved a System Usability Scale score of 82.4, indicating high user acceptance. The findings demonstrate that conversational ERP interaction can significantly enhance system usability, operational efficiency, and accessibility while maintaining enterprise-grade security and workflow reliability.