Re-Thinking Process Mining in the AI-Based Agents Era

TL;DR

PM-on-LLMs face limitations in complex reasoning for multi-step PM tasks. The authors introduce AI-Based Agents Workflows ($AgWf$) to decompose tasks and fuse deterministic tools with AI-based reasoning, formalized as $AgWf=(F, T, \textrm{tools}, \textrm{selector}, \textrm{prec}, t_1, t_f)$. They implement this idea via the CrewAI framework and illustrate running examples such as fairness and root-cause analyses, along with a taxonomy of AI-based tasks (prompt optimizers, ensembles, routers, evaluators, and output improvers). The work highlights a practical path to higher-quality PM pipelines in the AI-augmented era by leveraging modular tooling and targeted LLM capabilities, while outlining future directions for automatic orchestration and richer tool-support ecosystems.

Abstract

Large Language Models (LLMs) have emerged as powerful conversational interfaces, and their application in process mining (PM) tasks has shown promising results. However, state-of-the-art LLMs struggle with complex scenarios that demand advanced reasoning capabilities. In the literature, two primary approaches have been proposed for implementing PM using LLMs: providing textual insights based on a textual abstraction of the process mining artifact, and generating code executable on the original artifact. This paper proposes utilizing the AI-Based Agents Workflow (AgWf) paradigm to enhance the effectiveness of PM on LLMs. This approach allows for: i) the decomposition of complex tasks into simpler workflows, and ii) the integration of deterministic tools with the domain knowledge of LLMs. We examine various implementations of AgWf and the types of AI-based tasks involved. Additionally, we discuss the CrewAI implementation framework and present examples related to process mining.

Figures (7)

• Figure 1: Example AgWf, reported from Fig. \ref{['fig:agwf1']}, with four tasks aimed to combine the DFG and variants abstraction of an event log to respond to the inquiry of the user, which is preliminarily optimized by another task.
• Figure 2: Example AgWf in which the first task (T1) is optimizing the received inquiry, then two different tasks (T2 and T3) are executed using two different textual abstractions (directly-follows graph and process variants) to retrieve an answer, and eventually the responses are synthesized by an ensemble (T4).
• Figure 3: Example AgWf for bias detection in process mining (single task - multiple tools).
• Figure 4: Example AgWf for bias detection in process mining (multiple tasks - multiple tools).
• Figure 5: Example AgWf for bias detection in process mining (multiple tasks - single tool per task).

Theorems & Definitions (2)

• definition thmcounterdefinition: AI-Based Agents Workflow (AgWf)
• definition thmcounterdefinition: AgWf Execution - Sequential