Large Language Models in Enterprise Resource Planning: A Comprehensive Review of Applications, Challenges, and Future Directions

Abstract

Large language models (LLMs), built on the Transformer architecture and trained on large text corpora, are increasingly proposed as a natural language interface and content-generation layer for enterprise resource planning (ERP) systems, extending capabilities that have historically depended on structured transaction codes and purpose-built reporting tools. This paper presents a comprehensive review of LLM applications within ERP systems, organized around a four-part application taxonomy: natural language query and reporting over enterprise data; automated content and document generation; augmentation of intelligent process automation; and augmentation of embedded analytics and user experience. The review grounds this taxonomy in recent SAP ECC research demonstrating robotic process automation (RPA) combined with machine learning and SAP Fiori-based embedded analytics, both of which represent natural extension points for LLM capability within an existing ERP landscape, and further examines behavioral analytics for fraud detection as an adjacent enterprise analytics capability relevant to LLM-augmented risk narration and explanation. Because LLM-enabled ERP functionality depends on reliable access to live enterprise data across an increasingly distributed application landscape, the review also examines distributed service discovery and communication architecture and intelligent cloud resource scheduling as infrastructural preconditions for deploying LLM capability reliably and cost-effectively at enterprise scale. Drawing on the foundational Transformer, large-scale pre-training, and retrieval-augmented generation literature, the review develops a comparative analysis of the four LLM application types across their grounding requirements, autonomy level, and infrastructural demands, and identifies open challenges in grounding LLM outputs in accurate enterprise data, governance of generated content, and integration with existing SAP business logic, concluding with directions for future research and practice.