=Paper=
{{Paper
|id=Vol-2973/paper_140
|storemode=property
|title=Data-driven Management of Interconnected Business Processes - Contributions to Predictive and Prescriptive Mining (Extended Abstract)
|pdfUrl=https://ceur-ws.org/Vol-2973/paper_140.pdf
|volume=Vol-2973
|authors=Wolfgang Kratsch
|dblpUrl=https://dblp.org/rec/conf/bpm/Kratsch21
}}
==Data-driven Management of Interconnected Business Processes - Contributions to Predictive and Prescriptive Mining (Extended Abstract)==
https://ceur-ws.org/Vol-2973/paper_140.pdf
Data-driven Management of Interconnected Business Processes
Contributions to Predictive and Prescriptive Mining
Wolfgang Kratsch 1,2
1
University of Applied Sciences, Augsburg, Germany
2
Project Group Business & Information Systems Engineering of the Fraunhofer FIT,Bayreuth, Germany
1
Business process management (BPM) is an accepted paradigm of organizational design and a source
of corporate performance [1]. Due to substantial progress in process identification, analysis,
implementation, and improvement [2, 3], BPM receives constant attention from industry [4]. In times
of market consolidation and increasing competition, operational excellence (i.e., continuously
optimizing an organization’s processes in terms of effectiveness and efficiency) is key to staying
competitive. While traditional research in BPM focused on process models and model-based
information systems (e.g., workflow management systems), recently, the focus has shifted to data-
driven methods such as process mining [5]. In contrast to model-driven BPM, process mining uses
execution data in the form of events arising during process enactment, which may be exploited in
several ways [6]. Process mining strives to discover, monitor, and improve processes by extracting
knowledge from event logs available in information systems [7]. The most commonly applied use case
in process mining is discovering as-is process models that also serve as a starting point for more detailed
analysis [8]. Based on the mined as-is-process, the use case of conformance checking helps to point out
deviations from normative, predefined process models and actual process enactments (e.g., unintended
handover of tasks, skipped activities, missed performance goals). As process mining analyzes
information on an event-level, it also helps evaluate the actual process performance (e.g., measuring
cycle times, interruptions, exceptions). In sum, process mining can help ensure process hygiene,
constituting a fundamental requirement to achieve operational excellence [8].
As process mining is one of the most active streams in BPM, numerous approaches have been
proposed in the last decade, and various commercial vendors transferred these methods into practice,
substantially facilitating event data analysis [9]. At the tip of the iceberg, Celonis expanded in only
seven years from start-up to a unicorn, indicating the enormous cross-industry business potential of
process mining [10]. By 2023, Markets and Markets predicts a market potential of 1.42 billion US$ for
process mining technologies [11]. However, there are still numerous unsolved challenges that hinder
the further adoption and usage of process mining at the enterprise level [12]. First, finding, extracting,
and preprocessing relevant event data is still challenging and requires a significant amount of time in a
process mining project and, thus, remains a bottleneck without providing appropriate support [13].
Second, most process mining approaches operate on a single-process level, but organizations are
confronted with a process network covering hundreds of interdependent processes [12]. Third, process
managers strongly require forward-directed operational support, but most process mining approaches
provide only descriptive ex-post insights, e.g., discovered models or performance analysis of a past
period [8]. Since these challenges mainly drive this doctoral thesis, they will be discussed in detail
below.
First, finding, extracting, and preprocessing relevant event data is still challenging. This is most
frequently due to the lack of domain knowledge about the process, the distributed storage of required
data in different databases and tables, and the requirement of advanced data engineering skills [13].
Most recent process mining approaches assume high-quality event logs without describing how such
logs can be extracted from process-aware (PAIS) and particularly non-process-aware information
Proceedings of the Demonstration & Resources Track, Best BPM Dissertation Award, and Doctoral Consortium at BPM 2021 co-located with
the 19th International Conference on Business Process Management, BPM 2021, Rome, Italy, September 6-10, 2021
EMAIL: wolfgang.kratsch@fim-rc.de (W. Kratsch)
ORCID: 0000-0001-9815-0653 (W. Kratsch)
©️ 2021 Copyright for this paper by its authors.
Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
CEUR Workshop Proceedings (CEUR-WS.org)
systems (non-PAIS). In case of solely relying on process-aware information systems (PAIS) that
directly output minable event logs, the risks of neglecting process-relevant information arise, and so-
called blind spots can occur. For instance, if processes contain activities enacted by physical resources
or software bots that are not directly connected to PAIS, details of these enactments cannot be explored
using classical PAIS-based event logs. Due to increasingly digitized organizations, a growing part of
the available data is highly unstructured (e.g., text, video, or audio files) and requires the application of
novel concepts [8]. To sum up, although process mining approaches significantly matured in the last
decade, the step of data extraction is still too weekly supported and often results in bottlenecks that
negatively affect the quality of process mining analysis.
Second, most process mining approaches operate on a single-process level. However, process
mining currently evolves from project-based single-process analysis to an enterprise-wide ongoing task
[8]. Thus, methods for scaling process mining approaches on an enterprise level are needed [12]. One
of the most challenging topics relies on applying process mining methods that operate primarily on a
single-process perspective to enterprise-wide process networks, frequently covering hundreds of highly
interconnected processes. Typically, process mining initiatives consume substantial resources, such as
computing resources, but also expensive experts such as process owners or business analysts. Event-
data-driven process prioritization approaches considering process interdependencies can be the missing
part of the puzzle to ensure allocating these scarce resources to the most critical and central processes.
Third, process managers strongly require forward-directed operational support. Traditionally,
process mining approaches focused on historical data for backward-looking, descriptive process mining
(e.g., discovering process models). Descriptive process mining is an excellent starting point to improve
processes. However, process managers need operational in their forward-looking day-to-day business
[8]. As exemplary forward-looking predictive process mining use cases, predicting the behavior,
performance, and outcomes of process instances help organizations act proactively in fast-changing
environments. By combining process predictions with the decision area from normative process data
(e.g., performance thresholds), prescriptive process mining approaches are able to trigger actions
autonomously, e.g., by scheduling improvement projects [8]. The increasing volume of data (i.e., event
records and event properties) offers new opportunities and poses significant challenges for predictive
monitoring methods.
Visualized in Figure 1, BPM strives for connecting the real-world – physical in nature – with the
digital world enabling value co-creation between human beings and machines (i.e., physical machines
or software systems). The physical world consists of actors interacting with physical resources.
Commonly, actors and resources are orchestrated through processes that relate to PAIS, creating a
digital footprint (i.e., events) of each performed process activity. As the digital world’s central element,
the event log can be seen as a digital twin of the actual processes. Physical actors might also interact
with non-PAIS or perform manual activities that are not connected with the digital world and,
consequently, are not covered by PAIS-generated logs. Inspired by the three challenges introduced
above, this cumulative doctoral thesis consists of six research papers that are assigned to the field of
BPM and process mining, as indicated in Figure 1.
To lower the barriers for non-data-engineers to extract appropriate event logs, research paper #1 [14]
presents RDB2Log, a semi-automated, quality-informed approach to event log generation from
relational databases. RDB2Log takes a relational database as input and assesses its data quality based
on standard data quality dimensions. Thus, RDB2Log supports mapping data columns to event log
attributes and generating an appropriate event log. The artifact proposed in this research paper #1 is
envisioned as a step towards a process data quality lifecycle: systematic detection, repair, and tracking
of data quality issues. By providing a graphical interface that helps users extract high-quality event logs,
research paper #1 also strives to improve usability for non-experts.
Physical World Process-aware Information
Systems (PAIS)
Non-Process-aware
Information Systems (non-PAIS)
Process Network
Software Bot
Unstructured Data
Digital World
Relational Data
RP#3 Human Actors
RP#1 RP#2
Physical Resources
Event
Log
Video Camera
RP#6 RP#4
RP#5
Descriptive Predictive Prescriptive
Process Mining Process Mining Process Mining
E.g. Process Discovery E.g. Predictive Process E.g. Predictive Process
Monitoring Prioritization
Figure 1: Assignment of individual Research Papers to forward-directed Process Mining
Research paper #2 [15] proposes an approach enabling integrated analysis using bot and process logs
that provides new insights into bot-human interaction. An integrated analysis of bot and process data
can also show the effects of bots on business processes and explore how exceptions are handled. Joint
data analysis of bot and process data might also benefit the redesign of bots used in business processes.
As a central artifact, research paper #2 proposes an integrated conceptual data model specifying the
relations between bots and business processes. Based on this data model, it is possible to merge bot logs
and process logs, allowing for integrated analysis.
Research paper #3 [16] focuses on analyzing manual processes that are not supported by process-aware
information systems. In the case of solely relying on process-aware information systems (PAIS) that
directly output minable event logs, the risks of neglecting process-relevant information arise, and so-
called blind spots can occur. By providing an initial idea of how video data can be leveraged for process
mining purposes, research paper #3 strives to exploit valuable process-relevant information beyond
structured data sources bearing the potential to broaden the coverage of process mining analysis
substantially.
To decide which processes should be in focus of process mining initiatives, process prioritization can
be applied. Research paper #4 [17] proposes the Data-driven Process Prioritization approach (D2P2),
leveraging performance and dependency data from process logs to determine the risky performance of
all involved processes. Thereby, the D2P2 accounts for structural dependencies (e.g., processes that use
other processes) and stochastic dependencies (e.g., instances that affect other instances of the same
process). Based on the dependency-adjusted risky process performance, the D2P2 predicts when each
process is likely to violate predefined performance thresholds and schedules it for in-depth analysis to
future planning periods. Process analysts can then check whether the process under consideration
requires improvement. Based on event log data, the D2P2’s output is more reliable and detailed than
other process prioritization approaches.
While D2P2 ends up providing a prioritized list of process candidates for an in-depth analysis, research
paper #5 [18] expands the scope of process prioritization to schedule improvement projects providing
even more prescriptive support. To do so, research paper #5 proposes the PMP2 drawing on the main
concepts of D2P2 and extends an economic decision model optimizing the assignment of improvement
project alternatives. By combining Markov reward models and normative analytical modeling, PMP2
helps organizations determine business process improvement roadmaps (i.e., sequential implementation
of improvement projects on business processes), which maximize an organization’s long-term firm
value while catering for process dependencies and interactions among projects. Thereby, PMP2 takes a
multi-period, multi-process, and multi-project perspective. Thus, the PMP2 considers dependencies
between processes and improvement projects and thus schedules improvement projects to optimize an
organization’s long-term firm value.
Research paper #6 [19] explores the third challenge of providing operational, forward-directed support
to process managers by extensively comparing the performance of different ML (i.e., Random Forests
and Support Vector Machines) and DL (i.e., simple feedforward Deep Neural Networks and Long Short
Term Memory Networks) techniques for a diverse set of five publicly available logs in terms of
established evaluation metrics (i.e., Accuracy, F-Score, and ROC AUC). To provide generalizable
results, research paper #6 combines data-to-description and description-to-theory strategies [20]. Also
referred to as Level-1 inference, data-to-description generalization takes empirical data as input,
condensed into higher-level yet still empirical observations or descriptions [20]. In a nutshell, the
observations led to conclude that the application of DL is specifically promising when it comes to
variant-rich processes producing a vast amount of data during runtime.
In sum, the thesis contributes to the existing body of knowledge on data-driven management of
interconnected business processes. Hence, this thesis provides a basis for applying process mining in a
forward-looking view and, thus, supports researchers and practitioners on the journey of converting
project-based and isolated process mining initiatives to an ongoing supplement to the core of traditional
BPM methods.
References
[1] Dumas, M., La Rosa, M., Mendling, J., Reijers, H.A.: Fundamentals of Business Process
Management. Springer, Berlin, Heidelberg (2018)
[2] Recker, J., Mendling, J.: The State of the Art of Business Process Management Research as
Published in the BPM Conference. Business & Information Systems Engineering, vol. 58, 55–72
(2016). doi: 10.1007/s12599-015-0411-3
[3] Vanwersch, R.J.B., Shahzad, K., Vanderfeesten, I., Vanhaecht, K., Grefen, P., Pintelon, L.,
Mendling, J., van Merode, G.G., Reijers, H.A.: A Critical Evaluation and Framework of Business
Process Improvement Methods. Business & Information Systems Engineering, vol. 58, 43–53
(2016).
doi: 10.1007/s12599-015-0417-x
[4] Harmon, P.: The state of business process management 2020, vol. (2020)
[5] Diba, K., Batoulis, K., Weidlich, M., Weske, M.: Extraction, correlation, and abstraction of event
data for process mining. WIREs Data Mining Knowl Discov, vol. 10 (2020). doi:
10.1002/widm.1346
[6] van der Aalst, W. (ed.): Process Mining. Data Science in Action, vol. . Springer Berlin Heidelberg,
Berlin, Heidelberg (2016). doi: 10.1007/978-3-662-49851-4
[7] van der Aalst, W., Adriansyah, A., Medeiros, A.K.A. de, Arcieri, F., Baier, T., Blickle, T., Bose,
J.C., van den Brand, P., Brandtjen, R., Buijs, J., et al.: Process Mining Manifesto. In: BPM 2011
Workshops Proceedings, vol. 99, pp. 169–194 (2011). doi: 10.1007/978-3-642-28108-2_19
[8] van der Aalst, W.: Academic View: Development of the Process Mining Discipline. In:
Reinkemeyer, L. (ed.) Process mining in action. Principles, use cases and outlook, pp. 181–196.
Springer International Publishing, Cham (2020). doi: 10.1007/978-3-030-40172-6_21
[9] Viner, D., Stierle, M., Matzner, M.: A Process Mining Software Comparison. In: ICPM 2020
Proceedings (2020)
[10] Browne, R.: How three friends turned a college project into a $2.5 billion software unicorn. CNBC,
vol. (2019)
[11] Research and Markets: Process Analytics Market by Process Mining Type (Process Discovery,
Process Conformance & Process Enhancement), Deployment Type, Organization Size,
Application (Business Process, It Process, & Customer Interaction) & Region - Global Forecast to
2023 (2020), https://www.researchandmarkets.com/reports/4576970/process-analytics-market-
by-process-mining-type
[12] vom Brocke, J., Jans, M., Mendling, J., Reijers, H.A.: Process Mining at the Enterprise Level. Bus
Inf Syst Eng, vol. 62, 185–187 (2020).
doi: 10.1007/s12599-020-00630-7
[13] Li, J., Wang, H.J., Bai, X.: An intelligent approach to data extraction and task identification for
process mining. Inf Syst Front, vol. 17, 1195–1208 (2015).
doi: 10.1007/s10796-015-9564-3
[14] Andrews, R., van Dun, C., Wynn, M.T., Kratsch, W., Röglinger, M., ter Hofstede, A.: Quality-
informed semi-automated event log generation for process mining. Decision Support Systems, vol.
132, 113265 (2020).
doi: 10.1016/j.dss.2020.113265
[15] Egger, A., ter Hofstede, A.H.M., Kratsch, W., Leemans, S.J.J., Röglinger, M., Wynn, M.T.: Bot
Log Mining: Using Logs from Robotic Process Automation for Process Mining. In: ER 2020
Proceedings, vol. 12400, pp. 51–61 (2020).
doi: 10.1007/978-3-030-62522-1_4
[16] Kratsch, W., König, F., Röglinger, M.: Shedding Light on Blind Spots: Developing a Reference
Architecture to Leverage Video Data for Process Mining. Working Paper submitted to Information
Systems and currently facing major revisions, preprint available on
https://arxiv.org/pdf/2010.11289 (2020).
[17] Kratsch, W., Manderscheid, J., Reißner, D., Röglinger, M.: Data-driven Process Prioritization in
Process Networks. Decision Support Systems, vol. 100, 27–40 (2017). doi:
10.1016/j.dss.2017.02.011
[18] Bitomsky, L., Huhn, J., Kratsch, W., Röglinger, M.: Process Meets Project Prioritization – A
Decision Model for Developing Process Improvement Roadmaps. In: ECIS 2019 Proceedings
(2019)
[19] Kratsch, W., Manderscheid, J., Röglinger, M., Seyfried, J.: Machine Learning in Business Process
Monitoring: A Comparison of Deep Learning and Classical Approaches Used for Outcome
Prediction. Bus Inf Syst Eng, vol. (2020). doi: 10.1007/s12599-020-00645-0
[20] Yin RK (1994). Case study research: design and methods (2nd ed). Sage, Thousand Oaks, Calif.