# Large Language Models can accomplish Business Process Management Tasks

**Authors:** Michael Grohs, Luka Abb, Nourhan Elsayed, Jana-Rebecca Rehse  
**arXiv:** [2307.09923](https://arxiv.org/abs/2307.09923)  
**Date:** 2023-07-19  
**Summary page:** https://www.summarizepaper.com/en/paper/2307.09923v1/

## Key points

- Large Language Models (LLMs) have potential in Business Process Management (BPM)
- Existing BPM solutions using Natural Language Processing are limited
- LLMs offer a general-purpose instrument for multiple BPM tasks
- LLMs can be used for mining imperative and declarative process models from textual descriptions, as well as assessing process task suitability for robotic process automation
- Approach involves assembling a prompt and using GPT4 LLM
- Results show that LLMs perform comparably or better than existing solutions without extensive configuration or prompt engineering
- LLMs have implications for future BPM research and practical usage
- LLMs provide versatile and efficient solution for text-related BPM tasks by comprehending process-related text and integrating it into different applications
- Study demonstrates how LLMs revolutionize BPM by addressing multiple process-related problems through remarkable reasoning capabilities
- LLMs have significant potential in improving organizational activities and outcomes by managing processes using unstructured textual documents.

## Summary

This paper explores the potential of Large Language Models (LLMs) in Business Process Management (BPM). The authors highlight that existing BPM solutions, which extract information from textual documents using Natural Language Processing techniques, are limited to specific tasks and cannot address multiple process-related problems as a general-purpose instrument. However, with the emergence of LLMs with remarkable reasoning capabilities, there is now an opportunity to develop a general-purpose instrument with multiple applications in BPM. The authors illustrate how LLMs can be utilized for three BPM tasks: mining imperative process models from textual descriptions, mining declarative process models from textual descriptions, and assessing the suitability of process tasks for robotic process automation. They propose an approach where a prompt is assembled consisting of a general description of the task, a specification of the desired output format, the natural language text to abstract information from and optionally input-output pairs as examples. This prompt is then entered into a state-of-the-art instruction-following LLM called GPT4. The results show that without extensive configuration or prompt engineering, LLMs like GPT4 perform comparably to or better than existing solutions for these BPM tasks. The authors discuss the implications for future BPM research and practical usage. They emphasize that LLMs have the potential to provide a versatile and efficient solution for various text-related BPM tasks by comprehending process-related text and seamlessly integrating it into different applications. Overall, this study demonstrates how LLMs can revolutionize BPM by offering a general-purpose instrument capable of addressing multiple process-related problems through their remarkable reasoning capabilities. The findings suggest that LLMs have significant potential in improving organizational activities and outcomes by effectively managing underlying processes using unstructured textual documents.