Explainable Fully Automated Business Process Redesign

Abstract

In business process management, business process redesign (BPR) aims to improve business processes. In the past, BPR was mainly a manual task, with little computational power and typically high labor and time intensity. The increasing amount of stored process data and great advancements in generative machine learning (GML) and other analytical approaches have paved the way for automated BPR. However, existing BPR approaches are mainly designed for ofline applications and are therefore restricted to computing historical data samples of business processes. In this dissertation, we argue that performing BPR in runtime and leveraging the prediction capabilities of GML to achieves a high degree of BPR automation is possible and can allow organizations to improve their processes proactively and trustworthily. We contribute to information systems research by designing a GML-based technique for automated BPR and investigating its potential in practice. We also expect our findings to help practitioners manage process redesign and operation automatically and thereby put new AI systems into productive use.

Keywords

• eol>Business process redesign
• business process management
• generative artificial intelligence
• machine learning
• decision support

1. Introduction

Organizations operate in a volatile economic environment [ 1 ], characterized by political instability [e.g., 2] or rising customer expectations [e.g., 3]. At the same time, business processes are the organizational backbone for value creation [ 4 ]. Consequently, business processes need to be flexible and organizations are forced to steadily change their business processes to tackle influences emerging in the ever-changing environment [ 1 ]. BPR has been established in the domain of business process management (BPM) to improve business processes [ 4 ]. While the general idea of BPM is to improve business processes incrementally and cyclically [ 5 ], BPR is generally considered the most value-adding stage is BPM [e.g., 6]. BPR aims to re-organize business processes to improve their performance. In doing so, it can considerably increase the time, cost, quality, or flexibility of business processes [ 7 ], and consequently revenue and customer satisfaction.

Most of the current BPR initiatives are still done completely manually. For example, idea generation techniques are commonly used for manual BPR [ 8 ]. As these approaches lack computational support and do not provide the possibility to automatically gain insights from process data [ 9 ], these BPR approaches are typically labor- and time-intensive [ 8 ]. Recent advances in machine learning (ML) ofer new opportunities for improving this situation by analyzing process execution data, predicting redesign outcomes, and generating alternative process configurations [ 10 ]. BPR approaches of this type generally focus on the application of static redesign patterns [e.g., 6] and pure optimization algorithms [e.g., 11], but also on the use of GML algorithms [e.g., 9], that aim to learn the distribution of the underlying data to infer new artificial data samples.

While existing approaches for automated BPR provide computational capabilities to perform BPR in an automated way, they are usually designed to compute historical data samples of business processes ofline. Apart from automated BPR, predictive business process monitoring (PBPM) and prescriptive business process monitoring (PrBPM), other areas in BPM [ 12 ], aim to provide proactive decision support to process users or other relevant stakeholders in running business processes by predicting aspects like next activities [e.g., 13], or prescribing aspects like next best actions [e.g., 14]. In this work, we not only consider historical data samples of a business process from a conceptual point of view, but the entire data stream of event data produced by a business process. Given that, existing automated BPR approaches consider the dynamics of business processes to a certain extent. Additionally, they are limited to ofering guidance to process redesigners or other relevant stakeholders to reduce manual efort when identifying improvements in process models. These process models are typically derived from historical data samples of business processes [ 15 ].

Against this background, there is a high need for automated BPR approaches that are designed for running business processes and leverage process data produced over time, thereby achieving a higher level of BPR automation [ 16 ]. Besides contributing to research in BPM and BPR, this research also contributes to the research in graph machine learning and explainable AI. Furthermore, following recent research arguing that BPM initiatives should consider the dynamics of the digital age [e.g., 17, 18] most projects in this dissertation use data analysis to emphasize the importance of taking a dynamic perspective on process changes.

2. Background and Related Work

BPM is “a body of methods, techniques[,] and tools to discover, analyze, redesign, execute[,] and monitor business processes” [ 4 ] to improve business processes and assure consistent outcomes. Business processes are the most important subject of BPM, and activities to improve these are structured into phases along the business process lifecycle [ 4 ]. Within the BPM lifecycle, BPR is the phase dedicated to structural transformation, often aiming at substantial performance improvements in time, cost, quality, or flexibility [ 8, 19 ]. BPR activities include process modeling and simulation, process automation, optimization, and structural changes to business processes with the aim of improving these performance dimensions [ 8 ]. Moreover, BPR can also help organizations adapt to changing market conditions and customer needs, and drive continuous improvement and innovation [ 1, 2 ]. Unlike process optimization, which fine-tunes existing workflows, BPR seeks to reconfigure processes at a more fundamental level, ranging from incremental improvements to radical innovations [ 16 ]. In a very broad interpretation of the term, any change to an existing business process qualifies as BPR.

Beginning with manual process redesign for incremental process improvement, there are various collections of process redesign patterns, heuristics, and methods, reducing cognitive efort and guiding process redesigners in process improvement [ 4, 20, 21 ]. In manual process redesign for radical process innovation, there are also methods that provide guidance for creating new processes with new value propositions [22, 23]. However, these BPR approaches do not replace manual eforts with automation. Moreover, approaches for semi-automated process redesign for incremental process improvement were developed that can be positioned between manual and automated approaches. Initial approaches generally guide process improvement in a user-interactive way [ 6 ]. More advanced approaches enable the automated identification of useful process changes [ 9 ] or take an online perspective on business processes to automated BPR [ 10 ]. Earlier foundational work already explored the automation of redesign based on heuristics and optimization: Reijers and Limam Mansar [20] proposed a taxonomy of redesign heuristics and described how algorithms can apply them systematically, while Vergidis et al. [24] developed a multi-objective evolutionary optimization approach to generate performant process alternatives. In contrast to these ofline methods, the AB-BPM methodology [ 25] introduced runtime-oriented improvement through activity-based costing and process mining. While AB-BPM demonstrates that continuous process improvement is feasible, it focuses on task and resource allocation and does not support structural ideation or generative redesign. Therefore, the first foundations for the automation of incremental process improvement exist, both in ofline and runtime contexts, but they do not yet address process innovation, structural generalization, or explainability. In contrast, “the automation of process innovation proves to be an unsolvable problem to date” [ 16 ].

Some prior reviews have also touched on the intersection of ML and BPR. Fehrer et al. [26] developed a taxonomy of Process Improvement and Innovation Systems (PIIS), organizing tools across dimensions such as automation level and input types. While comprehensive, their work remains largely technologyagnostic and does not dissect ML-specific mechanisms. Their work identifies the growing importance of data-driven tools but stops short of analyzing ML paradigms or structural learning capabilities. Weinzierl et al. [ 12 ] ofer a systematic overview of AI-driven BPM methods with a focus on predictive and prescriptive runtime interventions. Their review covers methods that forecast outcomes (e.g., next activity prediction or compliance breaches [27, 28]) or prescribe actions (e.g., next-best-action systems [29]), but largely overlooks the structural reconfiguration of process models, which is at the core of BPR.

3. Research Approach

We adopt the design science research (DSR) methodology to structure this work into two information technology (IT) artifacts. The DSR paradigm enables information systems (IS) researchers to pursue a dual mission in solving a relevant real-world problem while contributing to the scientific knowledge base [30, 31]. In this case, the practical goal is to improve business process redesign using GML, and the theoretical goal is to generate new prescriptive knowledge for process redesign eforts. By focusing on artifacts that address an organizational need, DSR ensures the research is both relevant and rigorous [31]. We follow the well-known six-step DSR process by Pefers et al. [32], beginning with a problem-centered initiation. This provides a high-level framework (Problem Identification, Objectives, Design & Development, Demonstration, Evaluation, Communication) for iterative cycles of building and evaluating our artifacts. Figure 1 illustrates the first iteration of this process. It is important to note that Pefers’ process is a methodological framework, not a single prescriptive method for artifact construction. We therefore complemented it with more specific design and evaluation methods. In particular, we adhered to the principle of iterative build-and-evaluate cycles, where each cycle centers on adding to the body of knowledge through artifact improvement [33].

The artifacts we develop are techniques, which are a subtype of a method artifact in design science terms [34]. In other words, our contributions take the form of procedural solutions (algorithms and guidelines) for process analysis and redesign, rather than physical tools. Each artifact was developed through iterative refinement and evaluated to ensure it efectively addresses the identified problem. We structure this work around three research question with the overarching quesiton being: How can we design a fully automated business process redesign system that can be used in practice?

For the evaluation, we plan to incorporate both formative evaluations (e.g., expert feedback sessions) and summative evaluations (e.g., case studies in real organizations). Ultimately, to demonstrate utility in practice, we plan to conduct in-depth case studies with industrial partners (one in the insurance sector and another in enterprise software). These field studies will inform us on how organizations can implement our techniques in real process management environments and provide feedback.

Phase 1 Problem Identification Need for automated business process redesign approaches, that are designed for running processes and can be leveraged for process data produced over time, achieving a higher level of automation finished

DePfhinaisteio2n of DePshigasnea3nd Objectives Development Enable automatic 1) Event log business process transformation redesign in running 2) Process model business processes predictor training using graph machine 3) Process model learning prediction 4) Change propositions determination and ranking 5) Update models

Phase 4 Demonstration Instantiation of the technique and application of it to real-life event logs

Phase 5 Evaluation

Phase 6

Communication Comparison of Presentation and predictive quality for discussion of the next-as process preliminary results at model prediction with the conference to baseline models for receive feedback for real-life event logs to the subsequent receive preliminary iteration cycles results ongoing

RQ1: How can we design a technique that makes process mining analyses explainable for practitioners?

The ongoing implementation of information systems in organisations, along with the subsequently enhanced availability of event log data, have enabled process analysts to discover as-is models of processes with process mining with relative ease [35]. However, the crucial challenge lies in identifying potential areas for process improvements (i.e., process analysis) with respect to a strategic goal [36]; this requires analytical capabilities such as Pareto or root cause analysis [ 15 ]. Process model-based analysis - that is, process analysis based on the discovered process model - is able to make users aware of the business processes behind the data and can subsequently guide process analysts as they improve these processes [37]. To facilitate analysis beyond the simple discovery of a process, the process model must provide information suitable for the improvement initiative.

Therefore, we aim to answer this research question by designing an artifact that guides practitioners in the analysis of processes using DSR [32] and expert interviews [38]. For our preliminary results, we designed a technique to determine relevance scores of process activities with respect to performance measures extracted from event log data to aid in goal-directed process analysis and also conducted a case study with a German industrial company [27, 28].

RQ2: How to design a machine learning-based technique for automated business process redesign?

To address the limitations of manual, labor- and time-intensive BPR initiatives [ 9 ], as well as the focus on static patterns or pure optimization approaches [ 6, 11 ], we aim to contribute to research by combining three data-driven BPM research streams: PBPM, PrBPM, and data-driven BPR. The combination of these streams forms a foundation for a new type of technique than enables automated BPR in runtime, which consequently achieves a higher degree of redesign automation.

We aim to answer this research question by applying DSR [32]. As preliminary result, we propose the technique outlined in Figure 3 to tackel automated BPR in runtime using GML. This technique consists of an ofline phase and an online phase. In the ofline phase, we first load an event log and transform it into a dynamic graph of as-is process models (showing how the process was actually performed). Second, we create a graph-based autoencoder model, a form of GML model, for predicting the next as-is process model in a dynamic graph and train it with the entire dynamic graph of as-is process models. In the online phase, we first apply our predictor to an as-is process model discovered from current process instances and extract the deviations between the current as-is process model and the predicted as-is process model. These deviations represent our candidate change proposals, which can be ranked using a chosen approach like a key performance indicator (KPI)-based heuristic. Lastly, the candidate change proposals are integrated into the to-be process model (shows how the process should be performed) and the prediction model is fine-tuned with the current as-is process model.

Further, the technique addresses both a predictive and a prescriptive task to realize automatic BPR. While the first is addressed via the prediction of as-is process models to determine and rank change propositions, the latter is addressed by the provision of the updated to-be process model with relevant change propositions to process redesigners or other relevant stakeholders, to realize performance improvements in the execution of the business process.

Offline phase

Online phase

RQ3: How can organizations redesign their processes in runtime?

Our technique described in the previous chapter is right at the border between incremental and radical process redesign. Hence, with this question, we try to answer how organizations can redesign their processes in runtime and what degree of radical improvement is possible.

We aim to conduct an in-depth case studies in two organizations that apply our technique in conjunction to their own process redesign systems. This case study research allows us to investigate how a real-world organization will make use of automated business process redesign systems. Within this research question, we also aim to evaluate the applicability of the technique suggested in RQ2 for predicting concept drifts and its value for process practitioners.

4. Outlook

This dissertation demonstrates how process analysts can confidently adopt automated BPR systems using GML by presenting and evaluating prototype GML-based artifacts for process redesign and analysis within organizations. For practitioners, we show that BPR and process analysis can be automated beyond prior expectations, with a case study illustrating concrete implementations. For researchers, we integrate three data-driven BPM streams: PBPM and PrBPM, and data-driven BPR.

As we move forward, our current approach has some limitations: it relies on event log data, which precludes greenfield design, and its focus has so far been on incremental rather than radical redesign. Additionally, while our current technique does not yet incorporate mechanisms such as AB-BPM, which enables automated simulation and A/B testing for validation, it presents a promising extension point. At the doctoral consortium, we are looking forward to feedback on the industrial application of our redesign technique, its relation to ongoing research on process concept drift, and the design of our industry case study.

Declaration on Generative AI

During the preparation of this work, the authors used ChatGPT-4 in order to: check grammar and spelling, draft content, paraphrase and reword, as well as simulate peer reviews. After using this tool, the authors reviewed and edited the content as needed and take full responsibility for the publication’s content. process management capabilities in view of digitalization: results from a delphi study, Business & Information Systems Engineering 63 (2021) 83–96. [19] T. H. Davenport, Process innovation: reengineering work through information technology, Harvard Business Press, 1993. [20] H. A. Reijers, S. L. Mansar, Best practices in business process redesign: an overview and qualitative evaluation of successful redesign heuristics, Omega 33 (2005) 283–306. [21] J. Becker, M. Kugeler, M. Rosemann, Process management: A guide for the design of business processes, Springer, 2003. [22] T. Grisold, S. Groß, K. Stelzl, J. vom Brocke, J. Mendling, M. Röglinger, M. Rosemann, The five diamond method for explorative business process Management, Business & Information Systems Engineering 64 (2022) 149–166. doi:10.1007/s12599-021-00703-1. [23] S. Groß, T. Grisold, J. Mendling, J. Haase, Idea generation in exploitative and explorative business process redesign techniques, Information Systems and e-Business Management (2024) 1–29. [24] K. Vergidis, A. Tiwari, B. Majeed, R. Roy, Optimisation of business process designs: An algorithmic approach with multiple objectives, International Journal of Production Economics 109 (2007) 105–121. [25] S. Satyal, I. Weber, H.-y. Paik, C. Di Ciccio, J. Mendling, Business process improvement with the ab-bpm methodology, Information Systems 84 (2019) 283–298. [26] T. Fehrer, L. Moder, M. Röglinger, A taxonomy for process improvement and innovation systems,

Business & Information Systems Engineering (2025) 1–21. [27] M. Harl, S. Weinzierl, M. Stierle, M. Matzner, Explainable predictive business process monitoring using gated graph neural networks, Journal of Decision Systems 29 (2020) 312–327. [28] M. Stierle, S. Weinzierl, M. Harl, M. Matzner, A technique for determining relevance scores of process activities using graph-based neural networks, Decision Support Systems 144 (2021) 113511. [29] S. Weinzierl, S. Zilker, M. Stierle, M. Matzner, G. Park, From predictive to prescriptive process monitoring: Recommending the next best actions instead of calculating the next most likely events., in: Wirtschaftsinformatik (Zentrale Tracks), 2020, pp. 364–368. [30] S. T. March, G. F. Smith, Design and natural science research on information technology, Decision

Support Systems 15 (1995) 251–266. doi:doi.org/10.1016/0167-9236(94)00041-2. [31] R. L. Baskerville, A. Baiyere, S. Gregor, A. Hevner, M. Rossi, Design science research contributions: Finding a balance between artifact and theory, Journal of the Association for Information Systems 19 (2018) 358–376. doi:10.17705/1jais.00495. [32] K. Pefers, T. Tuunanen, M. A. Rothenberger, S. Chatterjee, A design science research methodology for information systems research, Journal of Management Information Systems 24 (2007) 45–77. doi:10.2753/MIS0742-1222240302. [33] J. Mendling, H. Leopold, H. Meyerhenke, B. Depaire, Methodology of algorithm engineering, arXiv preprint arXiv:2310.18979 (2023). [34] S. T. March, G. F. Smith, Design and natural science research on information technology, Decision support systems 15 (1995) 251–266. [35] W. M. P. van der Aalst, Process discovery from event data: Relating models and logs through abstractions, Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery 8 (2018) e1244. doi:10.1002/widm.1244. [36] D. Beverungen, J. Buijs, J. Becker, C. Di Ciccio, W. M. P. van der Aalst, C. Bartelheimer, . . . , V. Wolf, Seven Paradoxes of BPM in a Hyper-Connected World, BISE 18 (2020). doi:10.1007/ s12599-020-00646-z. [37] M. L. van Eck, N. Sidorova, W. M. P. van der Aalst, Guided Interaction Exploration and Performance Analysis in Artifact-Centric Process Models, BISE 61 (2019) 649–663. doi:10.1007/ s12599-018-0546-0. [38] M. D. Myers, M. Newman, The qualitative interview in is research: Examining the craft, Information and organization 17 (2007) 2–26.

References

[1] D. Beverungen , J. C. Buijs , J. Becker , C. Di Ciccio , W. M. P. van der Aalst , C. Bartelheimer, J. vom Brocke, M. Comuzzi , K. Kraume , H. Leopold , et al., Seven Paradoxes of Business Process Management in a Hyper-Connected World, Business & Information Systems Engineering 63 ( 2021 ) 145 - 156 . doi: 10 .1007/s12599-020-00646-z.

[2] F. Carmignani , Political Instability, Uncertainty and Economics, Journal of Economic Surveys 17 ( 2003 ) 1 - 54 . doi: 10 .1111/ 1467 - 6419 . 00187 .

[3] T. Kreuzer , M. Röglinger , L. Rupprecht, Customer-Centric Prioritization of Process Improvement Projects, Decision Support Systems 133 ( 2020 ) 113286 . doi: 10 .1016/j.dss. 2020 . 113286 .

[4] M. Dumas , M. La Rosa , J. Mendling , H. A. Reijers , et al., Fundamentals of business process management , 2 ed., Springer, 2018 .

[5] W. M. P. van der Aalst , M. La Rosa , F. M. Santoro , Business process management: Don't forget to improve the process! , Business & Information Systems Engineering 58 ( 2016 ) 1 - 6 . doi: 10 .1007/ s12599-015-0409-x.

[6] T. Fehrer , D. A. Fischer , S. J. Leemans , M. Röglinger , M. T. Wynn , An assisted approach to business process redesign, Decision Support Systems 156 ( 2022 ) 113749 . doi:doi.org/10.1016/j.dss. 2022 . 113749 .

[7] N. Brand , H. Van der Kolk , Workflow analysis and design , 1995 .

[8] S. Gross , K. Stelzl , T. Grisold , J. Mendling , M. Röglinger , J. vom Brocke , The business process design space for exploring process redesign Alternatives , Business Process Management Journal 27 ( 2021 ) 25 - 56 . doi: 10 .1108/BPMJ-03-2020-0116.

[9] C. van Dun , L. Moder , W. Kratsch , M. Röglinger, ProcessGAN: Supporting the creation of business process improvement ideas through generative machine learning , Decision Support Systems 165 ( 2023 ) 113880 . doi: 10 .1016/j.dss. 2022 . 113880 .

[10] M. V. Harl , S. Zilker , S. Weinzierl , Towards automated business process redesign in runtime using generative machine learning ( 2024 ).

[11] P. Aflerbach , M. Hohendorf , J. Manderscheid , Design it like darwin - A value-based application of evolutionary algorithms for proper and unambiguous business process redesign , Information Systems Frontiers 19 ( 2017 ) 1101 - 1121 . doi: 10 .1007/s10796-016-9715-1.

[12] S. Weinzierl , S. Zilker , S. Dunzer , M. Matzner , Machine learning in business process management: A systematic literature review , Expert Systems with Applications ( 2024 ) 124181 .

[13] D. Breuker , M. Matzner , P. Delfmann , J. Becker , Comprehensible predictive models for business processes , MIS Quarterly 40 ( 2016 ) 1009 - 1034 .

[14] S. Weinzierl , S. Dunzer , S. Zilker , M. Matzner , Prescriptive business process monitoring for recommending next best actions , in: Proceeings of the BPM 2020 Forum , Springer, 2020 , pp. 193 - 209 .

[15] M. Rosemann , The service portfolio of a BPM CoE , in: J. vom Brocke, M. Rosemann (Eds.), Handbook on BPM , Springer, 2010 , pp. 267 - 284 .

[16] M. Röglinger , C. v. Dun, T. Fehrer, Automated process (re-) design ( 2021 ).

[17] A. Baiyere , H. Salmela , T. Tapanainen, Digital transformation and the new logics of business process management , European journal of information systems 29 ( 2020 ) 238 - 259 .

[18] G. D. Kerpedzhiev , U. M. König , M. Röglinger , M. Rosemann , An exploration into future business