Desarrollo de agente de inteligencia artificial para la automatización de procesos en un sistema ERP

Abstract

In modern ERP systems, traditional graphical user interfaces (GUIs) can be complex, particularly for inexperienced users. This project develops an AI-driven conversational agent that enables natural language queries and the execution of selected ERP operations. The primary objective is to flatten the learning curve and facilitate access to key functionalities without requiring advanced technical expertise. The implemented architecture integrates Large Language Models (LLMs), Retrieval-Augmented Generation (RAG), and the Model Context Protocol (MCP) for the controlled execution of ERP tools, supported by an orchestration service and a vector database for embedding storage. Evaluations involving both experienced and inexperienced ERP users demonstrated that the agent primarily benefits new users by improving process comprehension and discovery. In contrast, experienced users utilized the agent as a support tool for general inquiries. However, findings indicated that response latency during this pilot phase could hinder its effectiveness for specific tasks. In conclusion, the project demonstrates that a conversational agent is a viable support layer for enhancing human–ERP interaction. Keywords: Conversational agent, ERP systems, automation, natural language, MCP