Automated Business Process Analysis: An LLM-Based Approach to Value Assessment [Extended version]

William De Michele, Abel Armas Cervantes, Lea Frermann · 2025

Read Automated Business Process Analysis: An LLM-Based Approach to Value Assessment [Extended version] by William De Michele, Abel Armas Cervantes, Lea Frermann (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-06600.txt.

Introduction

William De Michele, Abel Armas Cervantes, and Lea Frermann automate a tedious corner of business analysis: judging which steps in a process actually add value. Value-added analysis follows lean principles, but doing it by hand is slow and subjective. Language models offer semantic understanding at scale, if they can be structured properly.

Their method runs in two phases. High-level activities are first decomposed into detailed steps, then each step is classified by lean value criteria. Development used fifty business process models with hand-labeled ground truth, which the authors publish. Structured prompts consistently beat zero-shot baselines on both phases. The authors discuss how models might augment, rather than replace, human process analysts.

Catalog record

CREATORWilliam De Michele, Abel Armas Cervantes, Lea Frermann
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE31 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2504-06600

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Automated Business Process Analysis: An LLM-Based Approach to Value Assessment [Extended version] William De Michele, Abel Armas Cervantes, Lea Frermann Abstract Business processes are fundamental to organizational operations, yet their optimization remains challenging due to the time-consuming nature of manual process analysis. Our paper harnesses Large Language Models (LLMs) to automate value-added analysis, a qualitative process analysis technique that aims to identify steps in the process that do not deliver value. To date, this technique is predominantly manual, time-consuming, and subjective. Our method offers a more principled approach which operates in two phases: first, decomposing high-level activities into detailed steps to enable granular analysis, and second, performing a value-added analysis to classify each step according to Lean principles. This approach enables systematic identification of waste while maintaining the semantic understanding necessary for qualitative analysis. We develop our approach using 50 business process models, for which we collect and publish manual ground-truth labels. Our evaluation, comparing zero-shot baselines with more structured prompts reveals (a) a consistent benefit of structured prompting and (b) promising performance for both tasks. We discuss the potential for LLMs to augment human expertise in qualitative process analysis while reducing the time and subjectivity inherent in manual approaches. 1 1 1 Code, data, prompts are made available at https://github.com/wdemichele/WIPE . Keywords: Business Process Management, Waste Identification, Large Language Models, Value-Added Analysis, Process Analysis 1 Introduction Businesses are under constant pressure to optimise their operations, reduce costs, and enhance customer satisfaction [ 30 ] . Business Process Management (BPM) is a discipline that enables organisations to model, analyse, and improve their operational processes [ 16 , 4 ] . A key phase in the BPM lifecycle is process analysis, where inefficiencies and problematic areas are identified in the as-is process to inform process redesign. Central to this analysis is the identification of waste – activities or steps in the process that do not add value from the customer’s perspective or are unnecessary for the business to function effectively [ 20 ] . Traditional approaches to waste identification rely on manual expert analysis, which can be time-consuming, costly, and prone to subjectivity [ 24 ] . As organisations scale and processes become more complex, there is a growing need for automated and systematic methodologies that can efficiently analyse business processes and accurately identify areas of waste. Value-added analysis is a qualitative analysis technique that consists of identifying activities or steps in the process that do not deliver value to the customer nor to the business, and which can be potentially removed during process redesign. Figure 1: Overview of our LLM-based value-added analysis illustrated on an excerpt of a rental equipment process (top; [ 4 ] ). The first two activities (left) are broken down into multiple steps (center), which are then labeled as value adding (VA), business value adding (BVA) or non-value adding (NVA) (right). This paper introduces a novel approach that leverages Large Language Models (LLMs) to automate the identification of unnecessary steps in business processes (Fig. 1 ). LLMs have demonstrated remarkable capabilities in natural language understanding and generation [ 1 , 10 ] , making them suitable for semantic analysis of business processes [ 2 , 29 ] . Our approach operates in two phases: first, high-level activities are decomposed into fine-grained steps, and second, a value-added analysis [ 5 ] is performed over the steps to classify them as Value Adding (VA), Business Value Adding (BVA), or Non-Value Adding (NVA) based on Lean principles [ 34 ] . To evaluate our approach, we created a dataset of 50 business processes with BPMN models spanning various industries, including banking, healthcare, customer service, and technology. We manually annotated these processes with activity-to-step breakdown, and step-level value-added analysis. Although we presented our approach using process models as a starting point, the approach can be applied to event logs by focusing on the activities represented. The remainder of this paper is organized as follows: Section 2 reviews related work on LLMs in business process management, Section 3 presents our methodology, Section 4 details our evaluation, Section 5 discusses the implications of our findings, and Section 6 presents conclusions and future work. 2 Related Work The natural language generation capabilities of recent large language models (LLMs) offer new opportunities for automating complex tasks, including in the field of business process management [ 29 ] . For example, LLMs have been used for discovering process models from text, or vice-versa [ 2 , 3 , 7 ] . In this section we review relevant research on the application of LLMs for business process analysis. Large Language Models and Prompt Engineering. LLMs combine transformer architectures with attention mechanisms [ 28 ] with self-supervised pre-training on massive text corpora followed by instruction tuning. This optimises the models’ abilities to respond to natural language inputs (prompts). This strategy has been shown to lead to remarkable capabilities in natural language understanding, generation, and reasoning. The quality of LLM responses largely depends on prompt engineering [ 21 ] – the practice of designing input instructions that elicit desired model behaviors. Techniques range from simple prompting (zero-shot; [ 35 ] ) over extending prompts with input-output examples (few-shot; [ 31 ] ), to more sophisticated approaches like chain-of-thought reasoning, which guides models through step-by-step reasoning processes [ 32 ] . Structured prompting techniques [ 9 ] incorporating role descriptions [ 12 ] , task specifications, and output formats have proven effective for specialised tasks [ 36 ] . Recent research has explored systematic prompt optimization through automatic generation and iterative refinement, with the aim of identifying optimal prompt structures for specific applications [ 37 , 11 ] . These advanced prompting techniques form the foundation for our approach to activity breakdown and value-added analysis. Waste Analysis with LLMs. Lashkevich et al. [ 14 ] present an LLM-based approach for the analysis and redesign of business processes to optimize waiting time. The approach uses five types of waiting time that can be identified on an event log [ 15 ] – batching, resource contention, prioritisation, resource unavailability, and extraneous factors. The approach combines process mining outputs with structured prompting to generate both technical and strategic recommendations for process redesign. It is shown structured knowledge bases significantly outperform unstructured documentation for LLM prompting, with their enhanced method achieving 68% pattern utilization compared to 15% for unstructured approaches. Their findings also reveal a key insight about LLM-based process analysis: different prompting approaches serve distinct purposes, with enhanced prompting generating specific, actionable recommendations for analysts, while baseline prompting produces strategic insights for management decision-making. While [ 14 ] demonstrates the potential of combining LLMs with process mining for waste identification through temporal analysis, they address only one aspect of process waste. Our work on Activity Breakdown and Value-added Analysis complements these efforts by focusing on qualitative waste identification through semantic understanding of activities. The combination of our approach with others, such as [ 14 ] , enables a more comprehensive understanding of process waste and improvement opportunities. While quantitative analysis has dominated much of the existing research of LLM applications in BPM (e.g., [ 3 , 14 , 8 ] ), focusing on metrics such as process performance, execution time, and resource

Read the complete text