[Paper Review] Large Process Models: A Vision for Business Process Management in the Age of Generative AI

Timotheus Kampik, Christian Warmuth+13 more|arXiv (Cornell University)|Sep 2, 2023

Business Process Modeling and Analysis7 citations

TL;DR

This paper proposes Large Process Models (LPMs) as a hybrid framework that integrates generative AI, particularly Large Language Models (LLMs), with symbolic reasoning and knowledge-based systems to enable context-aware, reliable, and actionable business process intelligence. By fusing LLMs with process mining, knowledge graphs, and automated reasoning, LPMs aim to automate the generation of tailored process models, analytical insights, and improvement recommendations across diverse organizational contexts, significantly reducing transformation effort while enhancing trustworthiness and precision beyond pure statistical models.

ABSTRACT

The continued success of Large Language Models (LLMs) and other generative artificial intelligence approaches highlights the advantages that large information corpora can have over rigidly defined symbolic models, but also serves as a proof-point of the challenges that purely statistics-based approaches have in terms of safety and trustworthiness. As a framework for contextualizing the potential, as well as the limitations of LLMs and other foundation model-based technologies, we propose the concept of a Large Process Model (LPM) that combines the correlation power of LLMs with the analytical precision and reliability of knowledge-based systems and automated reasoning approaches. LPMs are envisioned to directly utilize the wealth of process management experience that experts have accumulated, as well as process performance data of organizations with diverse characteristics, e.g.,\ regarding size, region, or industry. In this vision, the proposed LPM would allow organizations to receive context-specific (tailored) process and other business models, analytical deep-dives, and improvement recommendations. As such, they would allow to substantially decrease the time and effort required for business transformation, while also allowing for deeper, more impactful, and more actionable insights than previously possible. We argue that implementing an LPM is feasible, but also highlight limitations and research challenges that need to be solved to implement particular aspects of the LPM vision.

Motivation & Objective

• To address the limitations of purely statistical, LLM-based approaches in business process management (BPM), which lack reliability, explainability, and safety for critical operational decisions.
• To propose a new conceptual framework—Large Process Models (LPMs)—that combines the correlation power of LLMs with the analytical precision of symbolic systems and automated reasoning.
• To enable organizations to receive tailored, actionable insights, process models, and improvement recommendations based on a vast, heterogeneous corpus of process knowledge and performance data across industries and organizational types.
• To assess the technical feasibility of LPMs while identifying key research challenges and risks related to trust, safety, and scalability in real-world BPM applications.
• To lay the foundation for future research and industry adoption of AI-augmented BPM that balances generative AI’s capabilities with human oversight and formal guarantees.

Proposed method

• LPMs are constructed by integrating fine-tuned general-purpose LLMs with domain-specific knowledge graphs and symbolic process models, enabling context-aware reasoning and inference.
• The framework leverages off-the-shelf LLMs for natural language understanding, query generation, and model augmentation, avoiding costly pre-training from scratch.
• LPMs utilize process execution traces and performance data to train specialized foundation models for prediction and counterfactual simulation, enhancing analytical depth.
• Symbolic reasoning and temporal logic are embedded to ensure correctness, explainability, and verifiability of LLM-generated outputs in process modeling and analysis.
• The system supports end-to-end BPM lifecycle support—from process design and mining to analysis and optimization—by orchestrating LLMs, knowledge graphs, and rule-based reasoning.
• Multimodal generative AI components are explored for processing non-textual inputs such as BPMN diagrams from images or audio from interviews, with potential pre-processing via speech-to-text or image-to-BPMN models.

Experimental results

Research questions

• RQ1 How can generative AI be meaningfully integrated with symbolic and statistical methods in BPM to ensure reliability, explainability, and safety in business-critical decisions?
• RQ2 To what extent can fine-tuned LLMs generate accurate, context-specific process models, queries, and improvement recommendations without compromising correctness?
• RQ3 What are the feasibility and limitations of using foundation models trained on process execution traces for predictive and prescriptive process analytics?
• RQ4 How can multimodal inputs (e.g., images, audio) be effectively processed and transformed into formal process models using generative AI?
• RQ5 What are the key technical, ethical, and organizational challenges that must be resolved before LPMs can be deployed at scale in enterprise environments?

Key findings

• Fine-tuned LLMs can effectively serve as contextualizers, generators, and augmenters of symbolic process models and queries, demonstrating strong feasibility for BPM applications.
• Preliminary evidence from existing works—such as Berti and Qafari (2023) and Klievtsova et al. (2023)—supports the viability of using off-the-shelf LLMs for process mining and modeling tasks.
• The integration of LLMs with knowledge graphs and automated reasoning enables the generation of explainable, reliable, and context-sensitive process insights beyond the capabilities of standalone LLMs.
• While general-purpose LLMs show promise for BPM augmentation, specialized foundation models trained on process execution traces remain nascent and require further research before industrial deployment.
• The use of generative AI for processing non-textual inputs (e.g., images of BPMN diagrams or audio from interviews) is feasible with pre-processing pipelines, though direct multimodal processing remains challenging.
• Significant risks remain in terms of trust, safety, and societal impact, particularly when deploying LLMs in mission-critical business process contexts without rigorous validation and oversight.