Accelerate DMAIC using Process Mining

Frank van Geffen1,

Rudi Niks2

1
Rabobank Nederland,
2
O&i Management Consultants
{[email protected], [email protected]}

Abstract. Process mining is an accessible technique to visualize and analyze

process variation and yield improvements. We experienced that Process Mining
can help to overcome some of the barriers of the Six Sigma DMAIC cycle in
improvement projects. This results in a significant acceleration to complete
such a cycle.

Keywords: Process Mining, Lean, Six Sigma, DMAIC, Process Improvement,

Rabobank, O&i Management Consultants

1 Introduction

Recently we started using Process Mining as an additional technique to increase

efficiency (cost side) and effectiveness (benefit side). In order to accomplish this,
Lean Six Sigma was embraced as a philosophy to continuously improve and achieve
permanent benefits. While reading the Volvo case we found a lot of similarities with
our own experience when optimizing the IT Service Desk of the Rabobank. Sharing
our knowledge and experience could help Volvo to use Process Mining as a tool to
accelerate their improvements.

In our daily work we regularly encounter the limitations and difficulties in applying
the DMAIC cycle to achieve actual improvements. DMAIC is a data-driven
improvement cycle used for improving, optimizing and stabilizing business processes
and designs. The DMAIC improvement cycle is one of the tools that drives Six Sigma
projects. DMAIC is a structured and step-by-step framework, with the strength of
underpinning root causes based on facts. However when applying this framework in
practice one could encounter a number of pitfalls. First of all, the success of this
framework depends on being able to collect the appropriate data. Secondly, measuring
the data (for example a stopwatch or manual forms) is precious and expensive. This
way of collecting data results in irritation and discussion about data quality, when
drawing conclusions. Thirdly, special skills are required to perform the statistical
analysis to identify the process variations. If root causes are identified based on these
statistical analysis techniques, the translation to practical solutions tends to be
challenging. Main cause is the gap between the statistical analysis and the practice.
This could result in difficulties when translating the statistical problem to a practical
solution. We experienced that Process Mining aids to overcome some of these
barriers. Still it remains important to carefully define and scope your improvement
ambition. Collection of data can still be a complex undertaking. If certain
preconditions are met and best practices are followed one can benefit from Process
Mining by accelerating the implementation of improvements.

2 DMAIC with Process Mining

In this chapter we propose an approach based on the DMAIC framework combined

with Process Mining tools and techniques. Figure 1 shows the DMAIC cycle
composed of the phases: Define, Measure, Analyze, Improve and Control. Without
the application of Process Mining it took 9-12 weeks on average to complete the
DMAIC cycle. Application of Process Mining has reduced the cycle time to 4-6
weeks on average. In the following paragraphs we discuss our proposed approach for
each phase in more detail.

2.1 Define

Whether or not Process Mining is applied in the following phases, it is important to

define a clear and precise objective and scope. Setting an unrealistic objective or
broad scope could result in lack of commitment and a diffuse focus. Based on the
objective the improvement strategy can be determined, which is in line with the
scope.
For example, one may choose to pick the low-hanging fruit or put the emphasis on
resolving fundamental issues in the process or organization.
For the Volvo case we advise to investigate if incidents and problems can be
prevented. Preventing incidents and problems is more cost effective than resolving
them afterwards. However not all incidents or problems can be prevented, such that
improving this process will be of value for the customer. The first question is: “Do we
need to focus on the incident or problem management process?”. Both processes
could be candidate for improvement, but they also impact each other. For example,
it’s expected that improving the incident process will result in less problems. We
advise to keep a clear focus and propose to focus on the incident management process
first. The second question is: “What value is provided to the customer when resolving
incidents?”. We think customer satisfaction depends on two important aspects: quality
of the solutions (first time right) and the lead time (time in which the solution was
delivered). Due to absence of data supporting the customer quality of the solution we
primarily focus on the lead time as the objective to create customer value. Third
questions is: “What impacts the lead time?” We assume that the lead time is
negatively impacted by a low 1st line resolution and the ping pong behavior between
teams. Additionally one could focus on impacts of norms, such as priority and
urgency. To illustrate the approach we kept the example simple as presented in figure
2.

2.2 Measure

Collecting the data is a challenging steps in the DMAIC cycle. Previously data was
collected by performing measurements with stopwatches and manual forms. This
resulted into a lot of discussion about the meaning and quality of the data. This is a
distraction to draw actual conclusions from the data. One of the advantages of using
Process Mining is that in many cases the data is already available in the information
systems. People involved in the process understand the meaning of this data and
quality of the data is part of their responsibility. This results in a better understanding
of the meaning of the data and less discussion about the data quality. However certain
preconditions have to be satisfied in order to take advantage of collecting data this
way:

• Availability (to answer the questions)

• Access
• Format
• Quality
• Meaning
We assume that for the Volvo case these requirements where satisfied. However if
one of these precondition is not satisfied it tends to be more difficult to get a useful
dataset for applying Process Mining. The following actions could help to resolve
some of these issues:
• adjusting the scope (back to the define phase)
• alternatives to collect data (drive system changes to collect transaction data)
• involvement and collaboration with: database specialist, data transformation
specialist, database administrators, data owners, domain experts, etc.
• …

2.3 Analyze

During the analysis phase the general performance of the defined measure is
evaluated. The collected data is used to answer the questions and bottlenecks are
identified. Detailed analysis on the bottlenecks are performed to determine the actual
root causes of the bottleneck.

In case of Volvo we propose to perform a product based benchmark as a strategy to

look for improvements with a high benefit potential. A snippet of the benchmark for
lead time and impact of 1st line resolution is presented in the table 1. An example of
the benchmark of ping pong behavior between teams for PROD698 and PROD424 is
presented in figure 2a,b. We found Rapid Miner 1 is a useful tool to create these
benchmarks in a short time.

The benchmarks are used to trend the products, to select the product(s) with the best
benefit potential. First, one needs to decide for which products the norms for the lead
times are not satisfied (see column e. in table 1). The norm is not defined for the
Volvo case, therefore we decided to use the average lead time as the norm. Second,

1 http://rapid-i.com
one could determine if there is evidence that this norm was impacted by a low 1st line
resolution (see column f. of table 1). We assumed a norm of 80% for the 1st line
resolution. Based on this information a selection of the top 3 (gold, silver and bronze)
can be made. In the selection process not only the quantitative, but also the qualitative
aspects should be taken into account (value, time, budget, experience and knowledge
of the teams).

In our experience every dataset has a story to tell. In order to determine the root
causes and formulate measures for the selected top 3 products, the right people need
to be involved to tell this story. The data can in some cases even help to select the
people that need to be involved during the analysis in workshops. During these
workshops the data is given further meaning in terms of root cause analysis. For
example for PROD698 and PROD424 limited evidence could be found that the ping
pong behavior between teams influences the lead time (see figure 2a,b). Process
Mining tools like Disco 2 are able to visualize the process and it’s variation to identify
bottlenecks. This tool allows to investigate other factors that influence the lead time.
For example we could explore the ping pong behavior between members of one team
like within G22 2nd or G96. Figure 3 is a process map presenting the total process map
of interaction between teams for PROD424 (high variation).

During workshops one is able to use the process map to zoom into the bottlenecks to
underpin the root causes. Linking the bottleneck and the actual cases that were part of
this bottleneck help to link the statistical problem to the practice. Questions during a
workshop can directly be answered, such that these insight can be taken into
consideration when measures are formulated. This enables that the members of the
process can be involved in an early stage of analysis to give practical meaning to the
analysis. We experienced that the Process Mining tool Disco also adds entertaining
value, returning the fun by adding the practical value when performing complex
statistical analysis.

2 www.fluxicon.com
2.4 Improve

In the improve phase the root causes identified during analysis are translated into
practical measures. For each measure the impact is determined and a potential benefit
is estimated to underpin the investment.

In our experience there is a thin line between the analyze and improve phase. During
the workshops we experienced that linking the root cause (statistical problem) directly
to a measure (practical solution) is an important driver for the success of the
implementation. Avoid endless discussion about statistical facts and figures, which
are often difficult to link to practical solutions. Instead focus on drawing conclusions
based on the power of visualization and linking this to practical examples. In this way
one can speed up the process of formulating actions that will drive the implementation
of the actual improvements forward.

2.5 Control

The purpose of this step is to sustain the benefits. Replicating the benchmark can be
used to monitor the improvement measures and deep analysis with Disco supports
further analysis to understand the impact of the change. Less effort is required to data
collection, as this is performed in a similar way as described in the measurement
phase. Challenging is how to keep people motivated to sustain the implemented
measures. In our opinion Process Mining can be used in a fun way to perform audits
and continue to challenge people to give their best effort.

3 Summary

Based on our experience with Process Mining in a practical context, we propose an

approach to apply Process Mining in a structured and step-by-step way to improve
business processes. The basis for this approach is the DMAIC framework from the
Six Sigma toolkit. DMAIC is a data-driven improvement cycle used for improving,
optimizing and stabilizing business processes and designs. The following phases are
part of this cycle: define, measure, analyze, improve and control. For each phase we
have presented best practices to use Process Mining as a tool to accelerate the
completion of the DMAIC cycle.
Whether or not Process Mining is applied, defining a clear and precise objective
and scope is an important first step.
Taking advantage of existing data extracted from the information systems is one
of the advantages when applying Process Mining. This speeds up the process of
measurement significant, resulting in less discussion about the quality of the collected
data. However certain preconditions must be satisfied in order to take advantage of
the data from the information systems.
During the analysis stage we proposed to use benchmarks to trend products, to
give direction toward the right causes. Further analysis can be performed in order to
underpin the root causes. Due to the ease of Process Mining tools such as Disco it is
possible to involve the stakeholders in an early stage of analysis.
In our workshops we have experienced that linking the identified root causes
(statistical problem) directly to a measure (practical solution) is an important success
factor for the implementation.
Tools like Disco also add entertaining value, making complex analysis fun and
keen people motivated by challenging them to give their best effort.

Acknowledgements

We like to give special thanks to our Rabobank team members Kees Aarts, Rob
Heuzen, Jacqueline Looise, Theo Reymer, Gea van Rooijen and Nanda Voskuilen for
their valuable input for the BPI Challenge 2013.