プロセスマイニングの基本機能である「プロセス発見」は、当初、ペトリネットがベースになっていたが、より現実に近いフローチャートを再現するために、様々なアルゴリズムが開発されてきている。ただ、業界有識者の話によれば、現在実用化されているプロセスマイニングツールのほとんどは、ファジーマイナーと呼ばれるアルゴリズムに基づいたもの（各社独自の改善は行っていると思われる）であると言われている。 同アルゴリズムは、一般にＤFGs(Directly-follows Graphs)と呼ばれる。ペトリネットや、また業務手順をフローチャートとして記述するための世界標準であるBPMN（Business Process Modeling and Notation）と異なり、ノードとノードが直接（Directly）結びつけられたフローチャートがDFGsである。すなわち、分岐ノードが描かれないため、このアルゴリズムでは、どこでどのような分岐が発生しているのか、具体的には、排他的（OR）なのか、並行的（AND）なのか、といったことが把握できない。このため、現状のプロセスを自動的に再現するとはいっても、分岐が明確でない不完全なものになるというのが現実である。もちろん、これについては、BPMN形式のフローチャートへの自動変換や、前述したビジネスルールマイニングの採用などの機能改善が行われてきている。
図１ Petri net、BPMN、Fuzzy Minerのフロー図例 上図でわかるように、DFGsであるFuzzy Minerには、Petri netやBPMNのような分岐ノードが存在しないため、同じプロセスの表現でありながら、Fuzzy Minerでは分岐のルールを判別することができない。
このConvergence/Divergence問題は、プロセスマイニングの分析品質を左右する最大の課題と言える。そこで、近年では、プロセスマイニングのゴッドファーザー、Wil van der Aalst教授が率いる研究者たちが「Object-Centric Process Mining」(1)と称する独自の方法論により当課題の解決に取り組んでいる。 また、myInvenioには、マルチレベルマイニングという機能が実装されており、一つのプロセスについて複数の案件IDを設定することで、プロセスの集約・拡散の状況を加味したフローの再現を実現している。
なお、マトリックスには記載しておりますが、プロセスマイニングの対象とはならない、ビジネスモデル層については、ビジネスモデルキャンバス（BMG：Business Model Canvas）、プロセスモデルキャンバス（PMG：Process Model Canvas）といったツールが活用できます。
Reference Matrix for Process Mining Tool Selection
In recent years, process mining has been further recognized and understood as a useful solution for promoting and establishing DX. In addition, as shown by the recent acquisition of myInvenio by IBM and Signavio by SAP, there is no doubt that process mining will become increasingly important as an indispensable component of corporate IT system development and operation as it is incorporated into the solutions of major IT companies.
Needless to say, the adoption of process mining tools is essential for companies to improve their business processes and to renovate and develop their systems using process mining, and the selection of the best suited tool for your company is a major key to success.
In this article, I will explain a matrix that will help you determine what functions of process mining tools are particularly necessary for your company.
●Horizontal axis of the matrix: Time
From the perspective of time, there are three dimensions which are completed processes in the past, processes in progress at the moment, and future processes to be executed in the future.
In general, data analysis is done on completed historical data. The same is true for process mining analysis. By analyzing completed event log data with process mining tools, we can automatically model current processes and analyze them from various perspectives (basic analysis).
For example, the following basic analysis is available.
Performance analysis (analysis from the perspective of time required and cost)
Conformance checking (comparative analysis of current process and ideal process)
There are some process mining tools which can do continuous monitoring and if problems such as deviations are detected, alerts are sent to the relevant parties by importing ongoing event log data to the process mining tool at a frequency close to real time.
For future processes that will take place in the future, the following functions will be supported.
Simulation (What-IF Analysis)
Simulate how much improvement (throughput reduction and cost reduction, etc.) can be obtained if the current process is improved in some way.
Model the flow of the ideal process to be implemented in BPMN format.
predict how in-process projects will be processed in the future and how much time will be required by using AI.
Based on the results of the above predictions, the tool proposes the best measures to prevent problems from occurring and prolonging the processing time.
Automated Process Improvement (AutoPI)
A process mining tool automatically executes measures for process improvement under certain conditions to achieve a quick remedy.
●Vertical axis of the matrix: Business layer
The business layer is a factorization into more detailed components from a process perspective. Administratively, the higher the layer, the more “strategic” it is, and the lower the layer, the more “tactical” it is, and the more “operational” (day-to-day on-site management) it needs to be.
At the top is the business model. From there, the granularity becomes finer, including the value chain that grasps the processes of the entire company from end-to-end, and the individual processes that make up the value chain.
Any business process can be broken down into a number of sub-processes. One more sub-process is composed of finer-grained tasks, and those tasks are composed of multiple activities.
For example, if we consider a sub-process called “invoice processing” in the accounting department, this includes activities such as “receiving invoices,” “checking the contents of invoices,” “registering invoices in the accounting system,” and “processing payments for registered invoices.
Among these activities, in the case of “receive invoice,” each task step is executed one by one, such as “open the email with the PDF invoice attached” and “download the attached PDF invoice.
In addition, these task steps are executed in the smallest units of PC operations, such as clicking on the mail software icon, clicking on open mail, and clicking on the attachment. These are called “atomic activities” because they cannot be decomposed any further.
Process mining basically analyzes the activity layer (or task step layer, as the case may be) from the process layer. transactional data recorded in IT systems are often at the activity level, which is relatively coarse-grained. In many cases, transaction data recorded in IT systems is at a relatively coarse activity level.
Therefore, task mining is used to analyze task steps and atomic activities with finer granularity. Task mining is still in its infancy, and it is still at the stage of trial and error for deeper analysis besides BI-like aggregation. However, by using it together with process mining, it can contribute to process automation, especially with RPA.
Now, in light of your company’s business process issues, which should be the target of analysis: past, present, or future? Also, at what granularity should the process be analyzed as a business layer?
With the person in charge of the tool vendor, let’s look at this matrix together to understand the extent to which these functions can be implemented while recognizing where the company is aware of the issues.
For the business model layer, which is not subject to process mining, tools such as Business Model Canvas (BMG) and Process Model Canvas (PMG) can be used.
Interest in process mining was born in Japan in late 2018. About two years have passed since then, and interest in process mining has grown even more, with an increasing number of companies, especially major corporations, introducing process mining and achieving a certain level of success.
For example, KDDI, a major telecommunications carrier, IHI, a major heavy industry company, Hitachi Transport System, a major logistics company and MISUMI, a major mold trading company, are actively using process mining, and use cases of their efforts are being reported.
However, there are not a few Japanese companies that are still skeptical about process mining, and it can be seen that there are some myths that can be called Japan-specific myths.
This article aims to debunk two myths that Japanese companies tend to hold about process mining.
Myth 1: Japanese companies’ systems are too complex to be analyzed by process mining
It is true that Japanese companies rarely adopt packages like SAP as the default, and they often perform extensive customization to match current business procedures or develop from scratch. In addition, it is not uncommon for the system configuration to be very complex and bizarre, as it has been repeatedly modified in response to changes in business operations.
For this reason, some people assume that the event log is also complex and that process mining analysis will not work. However, this is an illusion.
Many European companies, which are leading the way in the adoption of process mining, are still running legacy systems that have grown in age and complexity. The transaction data extracted from such systems is certainly not of high quality, and it takes a lot of man-hours to format the data into an event log that is suitable for process mining analysis.
However, it is not impossible, and even if it is not possible to draw a beautiful process model, it is possible to get an overview of the complexity of the current system, and, using process mining as a starting point, to design better business processes and define requirements for the supporting business system.
Reality: As long as there is some kind of event log, process mining is feasible for any complex system.
Myth 2: There are many tasks that are not performed in the system, such as manual processing of various paper documents and Excel operations, so process mining is useless.
The fact that much of the work is done outside the system, such as by hand or in Excel, is not limited to Japanese companies; a recent survey by Forrester, an IT research firm, found that a whopping 70 percent of U.S. and European companies do some kind of manual work in their operations.
The manual part, such as processing paper slips, cannot be captured as digital data and cannot be analyzed by process mining, of course. Even operations in office suites, such as Excel, are not automatically recorded as transactional data.
However, manual operations can be digitized by using OCR and other tools, or incorporated into workflows to leave a high percentage of digital footprints. In addition, Excel operations, for example, can be captured as a PC operation log or user interaction log by task mining.
Thus, with the advancement of digitalization, an environment is being created in which process mining analysis can be easily performed day by day.
In addition, even if a business process involves a lot of manual and PC operations, if the business system is used some part of it, it is possible to describe a process model based on milestone activities and discover where inefficiencies and bottlenecks exist.
And the inefficiencies and bottlenecks discovered by process mining are often manual operations or Excel operations, and as improvement measures based on the results of process mining, digitization and digitalization of manual operations and standardization through workflow systems are promoted. This creates a virtuous circle that further increases the effectiveness of process mining analysis.
Reality: Process mining for operations that involve manual operations and Excel can also lead to satisfactory results.
Will Process Mining tool and BI tool be amalgamated?
The answer is yes. The integration has already begun.
In terms of specific developments, a process mining tool called “PAFnow” is available as an add-on for Power BI. Similarly, “MEHRWERK ProcessMining” is offered as an add-on for Qlik.
On the other hand, process mining tools have also been enriching their “dashboard features” in addition to the standard features of process mining, such as “process discovery” which automatically creates a process model from the event log, but this dashboard feature is now close to the level of functionality provided by BI tools.
By the way, both process mining tools and BI tools are the same in that they take in various data related to corporate and organizational management, calculate numbers from various angles, and present the results visually in tables and graphs.
The decisive difference between a process mining tool and a BI tool is in how the calculation results are interpreted and utilized.
Concretely, we can explain as follows.
Calculation results presented by process mining tools
Process mining tools mainly look to performance of activities (processes) that create value = causals. In other words, process mining tools mainly cover Key Performance Indicators (KPIs).
For example, in the case of an insurance company’s claims processing process (from insurance claim to payment), process mining tools can analyze the number of cases for each activity in the process, the total time required for processing (throughput), processing cost, and the number of people in charge, and so on. In addition, the process discovery function can automatically draw a flowchart of business procedures to identify problems such as bottlenecks and inefficient repetitive tasks.
In this way, by analyzing activities that create value, i.e., causal data analysis, it is possible to link them to business process improvement measures to further increase value or reduce costs.
Calculation results presented by BI tools
BI tools mainly look at The size of value (sales, profit, etc.) generated = outcomes. In other words, BI tools cover KGI (Key Goal Indicator).
BI tool basically calculates sales, profit, market share, etc. as a result of corporate activities, and enables multifaceted analysis in various dimensions such as by division, area, and product.
BI tools can make judgments about which business units or areas are producing superior (or inferior) results, but they cannot infer the causes of why results are superior (or inferior). This is because it does not analyze causal data in the first place.
As explained above, to summarize the differences between them, BI tools are like a report book at the end of the term, and they are used to make final evaluations and to set new goals for theKGI in the next term. On the other hand, process mining tools are used to analyze performance in detail during the period and consider how to improve it in order to achieve the goals of KGI.
There is one more difference in the way data is analyzed that has recently emerged.
While BI tools only calculate a snapshot figure of historical data for the entire analysis period, process mining tools are now adding the ability to perform real-time monitoring that sequentially analyzes the data of the cases in the processing.
In order to continuously look back on the status of corporate and organizational operations, and to improve what needs to be improved, ensuring the achievement of goals, it is essential to combine KGI evaluation using BI tools and KPI evaluation using process mining tools.
Currently, more and more companies are using a combination of both tools, but as mentioned at the beginning of this article, the boundary between process mining tools and BI tools is blurring, and in the future, they will be provided as a combined tool.
Process Mining can find 5 problems being called, Muda, Muri, Mura, Mo-re, Miss in the target process.
Quality management as typified by Total Quality Control (TQC) is basically about solving various problems in the execution of business operations and aiming for reform and improvement.
In particular, TQC has been actively used in manufacturing plants, but it has also been applied to a variety of industries and corporate activities, including logistics, service, purchasing and sales. In recent years, the terms TQC and quality control have become less common, but the concepts and methods are universal and still valid today.
In the field of quality control, issues and problems to be improved are grouped into three main categories, which are called “darari” (muda, mura, and muri) or “3Ms”. In addition to these three items, More(mo-re) and Miss -Mistake are added to the 3Ms and called “5Ms”
The 5Ms is, of course, a framework that can be used for process mining analysis based on event logs. In fact, it is a familiar and simple term that provides an excellent starting point for smooth analysis implementation.
So, in this article, we will outline the 5Ms and tell you how they relate to process mining analysis methods.
What is 5M?
First, let’s discuss each of the 5M’s.
Muda can be translated in English as “Inefficient”.
It is a task that takes too long to complete because there are too many steps to accomplish the objective, or because it is too complex, or because it continues to be a formality that does not need to be done in the first place. It is an activity that does not generate much or no value. As these activities are literally “wasteful,” they need to be reduced or eliminated.
Muri is “over-burdened” in English.
Even if the number of cases to be processed is very high, or even if the number of cases is not so high, the number of cases waiting to be processed is rapidly overflowing due to the low number of staff assigned to the case. The result is stagnation and backlog of business. A “bottleneck. The number of projects and processing capacity are not in balance, and an excessive load is being placed on them.
Mura is “inconsistent” in English.
Inconsistency is, simply put, Too many variations in work procedures. If manuals don’t exist, or even if they do exist, they are not being used, leaving a large part of the process to the discretion of the individual, the procedure will vary from person to person. As a result, the throughput varies widely and the quality of the output varies.
In addition, although the basic procedure exists, if there are more cases where the flow of the procedure is complicated due to exceptions to the situation in the field, it is also an inconsistent business process.
Basically, the improvement measures for inconsistency are to reduce exception handling and standardize the process.
More is “omission” in English.
It means intentionally or inadvertently skipping a procedure that should be done. For example, in the inspection process of a production line, it is a customary practice to perform only three types of inspections and then skip them even though there are four types of inspections to be performed. Since some of the inspections that should be done are cut back, this may lead to accidents when the purchaser uses the product or to a recall.
In addition, from a compliance perspective, omission of a specific activity is a clear violation of compliance in a business where the work to be done is strictly regulated.
Since a deviated process that should not be allowed, you need to consider improvement plans such as providing training and education to ensure that the process is implemented, giving a systematic framework to prevent omission, and issuing alerts through continuous monitoring.
The word “miss” is “mistake” in English. To begin with, the Japanese word “mistake” is a diversion from this English word. It also includes the meaning of “error”.
A mistake is a human error, that is, a variety of mistakes made by the staff in charge of a task. Mistakes such as entering the wrong operating procedure or inputting the wrong value cannot be left as-is, but must be corrected by returning to the previous process or redoing the same operation.
The more mistakes are made, the more repetitive and wasteful work is required, and the more steps are required, the more “muri” and “mura” are generated. In other words, the cause of waste, wastefulness and unevenness is “mistakes,” so creating a system that prevents mistakes and RPA automation is an effective improvement measure.
Identification of 5Ms in Process Mining Analysis
Next, we will briefly discuss which of the various analytical functions of process mining analysis should be primarily used to identify the 5Ms, i.e., muda, muri, mura, more and miss.
Muda was an observable occurrence event, which was performing work that did not generate value. This is an issue of reduced efficiency and productivity.
In process mining analysis, after creating an as-is process model (flowchart) from the event log data, we first perform a “frequency analysis” to check how many cases are processed and where in the activities that make up the process and how many cases are flowing to the next process. This is because where there is a high volume of processing, there may be wasted steps lurking.
The next step is to look for process patterns that appear to be performing wasteful activities through “variant analysis,” which allows us to compare process variations. In addition, let’s look for wastage by “rework analysis” that discovers repetitions that are occurring.
Unreasonable workloads and improper procedures can create challenges that cause work to stagnate. Therefore, identifying “muri” means identifying the bottleneck in the process.
Therefore, first of all, we check the areas with a large number of processes with “frequency analysis”. This is because the areas with a large number of processes are not only inefficient, but also tend to be stagnant due to a high load. In addition, the “performance (time) analysis” mainly looks at areas with long waiting times (the time between the previous process and the next process). The part with long waiting time is really a bottleneck. At the same time, “social network analysis” is used to understand the business transfer relationships among the participants in the process in question, and a deep dive is made into which participant in the process is most likely to cause a bottleneck.
The mura is that work procedures vary from person to person, and the lack of standardization makes variation in process quality an issue.
This is the first step in a “variant analysis” to see how many processing patterns there are. The more patterns you have, the more you have, the more various procedures are being performed. We also identify which activities are deviating from the standard by means of “conformance checking”, which is a comparative analysis against the standard process (to be process).
As improvement measures, since standardization is the goal, it is effective to prepare manuals and systematic measures that do not allow multiple procedures.
Because More omits or skips some of the required procedures, it is considered to be a deviation from the standard and a business process with issues of non-compliance.
Therefore, we need to conduct a comparative analysis of the standard process (to be process) and the current process (as is process) reproduced from the event log, or in other words, a “conformance checking”, to identify the deviation.
As for improvement measures, as mentioned earlier, the system should be structured so that the procedure cannot be omitted, and compliance training should be provided to raise the awareness of the person in charge.
Miss(Mistakes) are specifically wrong procedures, careless mistakes, and various other errors that result in rework.
Since it is difficult to determine whether a mistake has been made or not through process mining analysis, we can check whether a large number of repetitions have occurred in activities with a large number of processes or in activities with long processing times by conducting “frequency analysis,” “performance (time) analysis,”“rework analysis,” etc., and finally, we can conclude that The process is verified to see if any mistakes have occurred through interviews with the people in charge on site and by understanding the detailed process at the task level through task mining.
It is difficult to reduce the number of mistakes to zero, but with RPA automation, theoretically, the number of mistakes can be reduced to zero. In addition, if the user interface is difficult to use, or in other words, if the usability is low, mistakes are more likely to occur, so the system will need to be modified to improve usability.