Large Language Models can accomplish Business Process Management Tasks

“In this paper, we illustrate how LLMs can be utilized for three BPM tasks that require textual documents as input. For all tasks, we follow the same approach, illustrated in Fig. 1. We start by assembling a prompt with the following parts:

1. A general description of the BPM task that is to be accomplished.

2. A specification of a particular output format that the LLM should adhere to. This ensures that the generated text output has a certain level of consistency and that results are sufficiently standardized so that they can be further processed by, for example, parsing algorithms.

3. The natural language text that we want to abstract information from, e.g., a textual process description 4. Optionally, if suitable for a given task, few input-output pairs as example.”

“In this paper, we developed and applied an approach that utilizes the LLM GPT4 for diverse BPM tasks. The approach itself is simple and leverages the capabilities of GPT4 by instructing it to accomplish the task at hand. We selected three BPM tasks to illustrate that GPT4 is indeed able to accomplish them: mining imperative process models from the textual description, mining declarative process models from the textual description, and assessing RPA suitability of process tasks from textual descriptions. For all three tasks, GPT4 performs similarly to or better than the benchmark, i.e., specific applications for the respective task. We analyzed the input and output robustness of the approach and found that the output is relatively insensitive to different executions of the same prompt, even if different authors formulated them. Further, we found that some prompts should include examples to help the LLM.”