Requirements towards Effective Process Mining
Matthias Lohrmann and Alexander Riedel
KPMG AG Wirtschaftspr¨ufungsgesellschaft,
IT Advisory,
{mlohrmann,ariedel}@kpmg.com
Abstract. Process mining is prominent contemporary research topic.
This paper describes requirements to be fulfilled for effective practical
adoption on the basis of application scenarios and a sample project.
1 Introduction
Business process management (BPM) methods have achieved broad acceptance
to support cooperation requirements within and between organizations [1]. In
this context, process mining has been a prominent research issue in recent years
[2, 3]. From our perspective, its purpose lies in complementing or even substitut-
ing a-priori design models with ex-post enactment models of business processes.
In terms of practical application, this is very promising considering the following
issues:
–Design models are subject to an “enactment bias”, i.e. deviations between to-be
model and as-is instantiations, or not available at all.
–As opposed to design models, enactment models can be enriched with actual en-
actment statistics covering, e.g., the prevalence of certain enactment patterns or
process variants, cycle times, or actually contributing individuals or roles. Note
that this leads to a “wider” view of process mining than what is adopted in parts
of related literature – respective tools, however, generally integrate these aspects.
This position paper aims at illustrating potentials of and requirements for
process mining techniques and tools from the perspective of an audit and ad-
visory firm. Its contribution lies in highlighting topics of practical relevance for
future research. In the following sections, we thus describe general application
scenarios (cf. Section 2) and more specific findings from a real-world pilot project
executed with an industrial client (cf. Section 3). Both lead to requirements for
further development as discussed in Section 4.
Note that our affiliation with an audit firm requires us to refrain from citing
concrete tools and suppliers in this paper.
2 Application Scenarios
This section describes application scenarios for process mining enabling to deduct
functional requirements. The scenarios have been selected on the basis of the
topics that have emerged as the most relevant from discussions with our clients.
2 Matthias Lohrmann and Alexander Riedel
Application Scenario 1 (Process Optimization in Functionally Structured
Organizations). In most organizations, function-oriented organizational structures
(e.g., the procurement, logistics, and accounting functions) are still prevalent in com-
parison to process-oriented structures (e.g., purchase-to-pay). In this context, end-
to-end (E2E) process optimization is a very important, but challenging issue. As an
example, consider the capturing of supplier order data. This is often seen as tedious
work by the procurement function, which, however, leads to huge additional workload
in accounting. Process mining can make these issues transparent by analyzing process
patterns independently of organizational borders. Thus, continuous end-to-end process
improvement can be fostered.
Application Scenario 2 (Compliance Management and Identification of De-
ficiencies). For many real-world processes, compliance deficiencies cannot be rigor-
ously precluded through appropriate process design. Moreover, the capability to clearly
retrace process execution often constitutes a compliance requirement in itself (e.g.,
§238 of the German Handelsgesetzbuch). Process mining techniques allow to analyze
execution logs with respect to non-compliant characteristics (e.g., occurrence of non-
approved payment instances). Here, their particular advantage lies in the automated
analysis of large numbers of process instances. Compared to the “traditional” risk-
oriented audit approach, they allow to execute continuous or even full-scope audits [4].
In some cases, even the systematic circumvention of compliance requirements through
certain process patterns can be uncovered (cf. Section 3).
Application Scenario 3 (Internal Benchmarking). Benchmarking refers to the
comparison of good practices and performance indicator values (e.g., cycle times, unit
costs, and transaction volumes per unit of capacity) between peer organizations [5].
From our experience, the so-called internal benchmarking, i.e. the comparison within
one group, for instance between business units or production plants, has emerged as
particularly useful since it reduces the issue of (perceived or real) limitations to the
comparability of peers. In this context, process mining can play an important role.
Typical tools combine the assessment of practices (i.e., enactment patterns or process
variants) with performance indicators. It is particularly suited for internal benchmark-
ing because in that case the probability of equal process-aware information systems
(PAISs) is higher. As an additional advantage with regard to implementation, it makes
use of readily available process log data instead of additional surveys, thus reducing
the cost of benchmarking.
Application Scenario 4 (Capacity Management). Process mining allows to iden-
tify “bottlenecks” in transactional processing. This is achieved by identifying roles or
users with particular high transactional volumes or cycle times in terms of time lag be-
tween “receipt” and “completion” events. The respective capacities can then be subject
to specific management.
3 Pilot Application and Findings
This section shortly discusses findings from a pilot project conducted with a
client. The objectives of the project primarily corresponded to Application Sce-
nario 1 as described above, although aspects of other scenarios were addressed
as well. Example 1 describes the sample process.
Requirements towards Effective Process Mining 3
Example 1 (Sample Process: Purchase-to-Pay). The purchase-to-pay process covers
process steps that are briefly summarized in Business Process Model and Notation
(BPMN, [6]) in Figure 1. The model has been simplified for reasons of space, but
amended with reference numbers we use to allocate findings. The total mining sample
covered 28,345 process instances with ca. 400 event types incurred at a foreign sub-
sidiary in a timeframe of three months after data cleansing. Data cleansing involved
removing all instances that were not started and completed (with defined start and
end events) within the given timeframe.1
Enter
purchase
requisition
Create
purchase
order + +
Receive
invoice
Receive
goods
Scan
invoice
Enter
goods
receipt
Match
invoice X X
Escalate
deviation
Post
invoice
12
3 4
65
7
8
9
Fig. 1. Sample Process: Purchase-to-Pay
Due to its transactional character, its well-understood semantic content with
a high degree of standardization, its occurrence in basically all industries, and its
broad support in PAISs, the purchase-to-pay process is particularly well-suited
to illustrate the potentials of process mining. The application of two exemplary
process mining tools enabled comparing a broader range of functionality. Major
findings described in the following illustrate which topics should be covered (i.e.,
discoverable) by effective process mining techniques and tools:
1. Up to two thirds of total project effort can be incurred in data integration and
cleansing.
2. 72% of invoices are below the threshold value for mandatory purchase order cre-
ation, leading to missing purchasing data.
3. Missing purchasing data (Ref. 2 in Figure 1) leads to an 88% failure rate in auto-
mated invoice matching (Ref. 7).
4. 86% of postings (Ref. 9) occur on five out of a total of 116 accounts.
5. Transactional volume shifts significantly between months, which leads to issues
with capacity management.
6. There have been more than 60,000 segregation of duty violations (e.g. one user
entering both purchase order and payment approval).
7. In ca. 3% of cases, purchase orders have been created (Ref. 2) immediately before
or after an invoice became overdue, which might point to a compliance violation.
1Note that this constitutes an inherent weakness since issues leading to pending in-
stances (e.g., “deadlocks”) are systematically hidden from further analysis.
4 Matthias Lohrmann and Alexander Riedel
4 Conclusion
Based on the application scenarios and the sample case discussed, we conclude
that the following requirements are of particular relevance to the effective prac-
tical application of process mining:
Requirement 1 (Data Integration). To enable broad application, the integration
of (process log) data from various sources is of critical importance. This issue is cur-
rently cumbersome with most tools and includes recurring and automated data staging
as well as cleansing (e.g. Finding 1).
Requirement 2 (Compliance Rules Modeling). Tools should include a modeling
facility for compliance rules such as the segregation of duties [7]. Automated analysis
with respect to violations should be feasible (e.g. Application Scenario 2, Findings 6/7).
Requirement 3 (Pattern Analysis). The analysis of process variants including
their cumulative frequency has emerged as the most useful aspect for management
discussions. Tools should support this analysis including enrichment of patterns with
performance indicators (e.g. Application Scenario 1, Finding 2).
Requirement 4 (Approximation of Manual Effort). Manual effort incurred in
process activities cannot be measured by mining execution events, but approximation
facilities should be provided by tools (e.g. Application Scenario 3, Finding 5).
Requirement 5 (Automated Regression Analysis). Statistical regression analy-
sis, i.e. the analysis of relations between dependent and independent variables should
be automated. This would, for instance, be useful to identify bottlenecks or for lower
level internal benchmarking (e.g. between cost centers, Finding 4).
Requirement 6 (Sample Delineation). A methodical solution for delineating a
sample of process instances is still missing. For instance, delineation on the basis of
completed instances can result in blanking out precisely critical cases (e.g. Finding 1).
Requirement 7 (Visualization of Variants). For management discussions, process
variants should be visualized effectively (e.g. regarding the difference between two
patterns) using innovative solutions (e.g. Application Scenarios 1-4).
References
1. van der Aalst, W.M.P., ter Hofstede, A.H.M., Weske, M.: Business process man-
agement: A survey. In: Proc. 1st BPM. LNCS 2678 (2003)
2. van der Aalst, W.M.P., Weijters, A.J.M.M.: Process mining: a research agenda.
Computers in Industry 53(3) (April 2004) 231–244
3. Li, C., Reichert, M., Wombacher, A.: Mining business process variants: Challenges,
scenarios, algorithms. Data & Knowledge Engineering 70(5) (2011) 409–434
4. Rezaee, Z., Sharbatoghlie, A., Elam, R., McMickle, P.L.: Continuous auditing:
Building automated auditing capability. Auditing 21(1) (March 2002) 147–163
5. Camp, R.C.: Benchmarking: The search for industry best practices that lead to
superior performance. Quality Press (1989)
6. The Object Management Group: Business Process Model and Notation: Version 2.0
(2011) http://www.omg.org/spec/BPMN/2.0.
7. Ly, L.T., Knuplesch, D., Rinderle-Ma, S., G¨oser, K., Pfeifer, H., Reichert, M.,
Dadam, P.: Seaflows toolset – compliance verification made easy for process-aware
information systems. In: Proc. CAiSE’10 Forum. LNBIP 72 (2010) 76–91
