Process Mining Without Perfect Data? Anne Rozinat Says Yes!: A Practitioner’s View on Event Log Quality

PROFILE Process Mining Without Perfect Data? Anne Rozinat Says Yes! A Practitioner’s View on Event Log Quality Adela del Rı ´o Ortega •Iris Beerepoot •Han van der Aa •Joerg Evermann ÓThe Author(s) 2025 Anne Rozinat has been a process mining enthusiast for more than two decades. She holds a PhD degree in process mining from the Eindhoven University of Technology (TU/ e). Together with Christian Gu ¨nther, she is a co-founder of one of the oldest process mining tool vendors in existence: Fluxicon (since 2009). 1 Their Disco tool is used by professionals, and has a long-standing tradition of being used by research groups and teachers all over the world, thanks to their Academic Initiative. Fluxicon’s Flux Capacitor blog 2 and Process Mining Cafe ´ 3 regularly provide insights on the intersection of industry practice and academic research on process mining. Thanks to her wealth of experience on both sides of the process mining world, Anne is a perfect candidate to provide her views on the topic of our special issue related to Exploring the (Mis)- Match Between Real-World Processes and Event Data. BISE: Hello Anne, thank you for taking your time and joining us today. To start off, can you share a bit about your journey in process mining and what sparked your interest in this field? Anne: Thank you for inviting me! I came across the topic as a student in the early 2000s. Prof. Weske offered great BPM lectures and seminars at the HPI in Potsdam, Germany, where I stumbled upon process mining. Back then, it was still called workflow mining and ProM did not yet exist. I loved the idea of turning BPM on its head and automatically, magically, discovering process maps from event data. I was sold. In 2004, I moved to the Netherlands to do my Master thesis with Wil van der Aalst at the TU/e and I have been working on process mining ever since. After our PhDs at the TU/e, my co-founder Christian Gu ¨nther and I started Fluxicon in 2009. With our process A. del Rı ´o Ortega SCORE Lab, University of Seville, ETSII, Avda. Reina Mercedes, s/n, 41012 Sevilla, Spain e-mail: [email protected] I. Beerepoot (&) Department of Information and Computing Sciences, Utrecht University, Princetonplein 5, 3584 CC Utrecht, The Netherlands e-mail: [email protected] H. van der Aa Faculty of Computer Science, University of Vienna, Wa ¨hringer Str. 29, 1090 Vienna, Austria e-mail: [email protected] J. Evermann Faculty of Business Administration, Memorial University of Newfoundland, St. John’s, NL A1B 3X5, Canada e-mail: [email protected] 1 https://fluxicon.com/disco/. 2 https://fluxicon.com/blog/. 3 https://fluxicon.com/cafe/. 123 Bus Inf Syst Eng https://doi.org/10.1007/s12599-025-00938-2 mining software Disco, we enable practitioners to do process mining themselves. BISE: So you have been in the process mining landscape since the early days! Given your extensive experience in practice, how would you describe the importance of having high-quality and representative event data for process mining projects? Anne: Generally, I find that people are too worried about the quality of their data. You don’t have to have perfect data to start a process mining project. You only need to be diligent in properly evaluating the data quality, so that you know which parts of the data you can rely on and which not. Even from imperfect data you can often learn a lot. So, waiting for high-quality and representative event data should not be an excuse to not get started! Therefore, it is better that you start with the data that you have: any insights into data quality problems are the first valuable result of your process mining project. BISE: In our special issue, we focus on understanding the (mis)match between event data and real-world processes. Can you elaborate on the common challenges faced in ensuring that event data accurately represents real processes? Anne: The mismatch exists because most IT systems were not created for process mining. It is natural that the data that they collect focuses on the original purpose of the system. For example, healthcare information systems often focus on the financial accounting of the performed procedures. The precise time of a procedure is not so relevant for reimbursement, although it can be critical from a process mining perspective. Fortunately, with process mining we are usually interested in process milestones on a business level. And this means that the IT system typically stores this information somewhere. For example, if an offer is sent to a prospect in a sales process, then the CRM system will capture the date of when the offer was sent somewhere in the business data. Still, a common challenge is that you have blind spots (not all steps may be captured, or they happen outside the system). You might only have the start or only the completion time and, therefore, not know how long a step takes. Event data can be aggregated by reporting and BI systems, which loses the history that you need for process mining. If timestamps are not captured precisely enough (for example, if you only have a date but no time of the day) then you don’t know the sequence of events that happened on the same day. Etc. etc. BISE: With respect to such challenges, are data privacy and security somehow at odds with event data accurately reflecting real-world processes? Anne: No, not for process mining anyway. To analyze and improve a process, you don’t have to invade anyone’s privacy. The basic data that you need for process mining (case ID, activity name, timestamp) is process-related and can be further anonymized if needed. For example, Tobias Zepter from BVV Pension Management in Germany created a system that allows employees to collect data about process executions with privacy considerations built in. 4 In our Process Mining Cafe ´with Dirk Fahland and Felix Mannhardt 5 we talked about how being a responsible process miner is part of ‘‘doing a good job’’. And Le ´onard Studer from the City of Lausanne shared 6 an ethical charter that helps to put people at ease when you start analyzing their work with process mining. You need to establish an open and trusting culture of collaborative improvement to enable people to speak the truth about their process. So, data privacy, ethics, and security are very important for acceptance and out of responsibility. We have a step called ‘Ethical considerations’ in our 12-step project guide 7 that asks people to think about these topics early on. BISE: Can you share a specific example from your experience where data quality issues significantly impacted a process mining project? How did you address these issues? Anne: A key example that comes to mind is when we worked with an insurance company in the Netherlands and found that for one of their core processes people did not use the IT system as intended. The employees worked on the dossiers outside of the system and documented their steps later with a long delay and all at once.Therefore, the timestamps collected by the system did not reflect the actual time of the work at all. As a consequence, process mining did not make sense and they stopped the project. Instead, the company first had to focus on improving their IT system, since it apparently did not support employees in their work, and was rather just seen as an additional burden of documentation. A take-away from this is that the data gets better the more IT systems support workers to do their jobs. Think of a tablet that a doctor is using while visiting patients in the ward compared with the same doctor sitting in their office at the end of the day to document what they have done throughout the day. The more the IT systems are actually used during organizational operations, the better the quality will be for process mining. 4 https://fluxicon.com/blog/2023/11/process-mining-cafe-26-record ing/. 5 https://fluxicon.com/blog/2021/09/process-mining-cafe-privacysecurity-ethics/. 6 https://fluxicon.com/blog/2021/04/process-mining-cafe-freestyledata-transformation/. 7 https://fluxicon.com/blog/2023/11/project-guide-2/. 123 A. del Rı ´o Ortega et al.: Process Mining Without Perfect Data?, Bus Inf Syst Eng BISE: Could you share some best practices for addressing data quality issues in process mining? Anne: We have a checklist for finding data quality problems that are common in process mining data. But even more important than following the checklist is to make sure you perform a data validation session with a domain expert. The goal of the data validation session is not yet to analyze the process; the goal is just to get a sense of how good the data is and where it might have problems. Together, you look at every aspect of the process. For everything that seems strange (strange paths, strange variants, strange times, etc.), you then inspect specific examples. Ideally, you can look up these exact cases in the operational system to see if this is what really happened in the process or if you are dealing with a data quality issue. The data validation session is important for two reasons: (1) Not every data quality problem can be found with a checklist but a domain expert who knows the process reality will quickly see if something does not make sense. (2) The process of inspecting the data together builds trust in the validity of the data. So, you get better support for the analysis results once you present them later. BISE: Process mining traditionally uses data from sources such as ERP systems or databases. Have you observed any innovative uses of additional data sources in process mining? Anne: There are so many innovative uses of process mining. I particularly like an older example of Philips. 8 They analyzed the usage profiles of their MRI scanners in the field to create additional system tests. This allowed them to increase the reliability of future generation machines based on insights from real-life workflows. A recent example is the Urban Mobility project by Andrea Delgado and Daniel Caligari Garcı ´a from the Universidad de la Repu ´blica in Uruguay. 9 They discovered process flows for the bus lines in Montevideo using open data from the city. They then enhanced the data by using location information about the bus stops to display the discovered processes in a spatial city map, yielding very interesting insights. BISE: Looking towards the future, what role do you see emerging technologies, like machine learning and AI, playing in improving the quality and analysis of event data in process mining? Anne: There are many different types of AI and they are suitable for different things. The progress in the field is at the same time amazing and surprising, as well as overestimated. What is complex for AI is not the same as what is complex for humans, and vice versa. While the data validation session with a domain expert is not something that you can replace by AI, people will find innovative use cases in many areas. I liked the video mining example by Wolfgang Kratsch et al., where they created event logs from unstructured video data. 10 Another common data quality problem area is the creation of meaningful activity names for free-textfields. For example, when Le ´onard Studer tried to use text mining to address this issue in 2015, 11 the results were not good enough to truly reduce the 50 K ?different free-text fields into usable activity names. However, today, this has probably become feasible, although I have not tried it. BISE: Finally, in your view, what are some of the main research gaps in the context of this special issue topic that you would like to see the academic community tackle? Anne: Good question! I personally like terminology. The scientific rigor can help sharpen the thinking by clearly defining the language. For example, in a recent Process Mining Cafe ´we talked with Xixi Lu about data preprocessing categories, 12 implicit biases, outliers vs. noise, and whether something like free-text data is actually a data quality problem or rather an issue of data suitability. But overall, all data quality research is useful. I appreciate any research into practical topics! If you ever need a reality check for something you are thinking about, we are always happy to provide feedback—just get in touch. And keep up the good work, all of you! Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons. org/licenses/by/4.0/. 8 https://fluxicon.com/blog/2018/09/usage-profiles-for-systemrequirements-in-the-context-of-philips-mr/ and https://fluxicon.com/ camp/2019/5. 9 https://fluxicon.com/blog/2024/03/how-to-bring-location-into-pro cess-maps/ and https://fluxicon.com/blog/2024/03/process-miningcafe-30-recording/. 10 https://fluxicon.com/blog/2022/05/performing-process-mining-onvideo-data/. 11 https://fluxicon.com/camp/2015/1. 12 https://fluxicon.com/blog/2024/11/process-mining-cafe-36-record ing/. 123 A. del Rı ´o Ortega et al.: Process Mining Without Perfect Data?, Bus Inf Syst Eng