To be honest with you, Process Mining is really yesterday's news. The first approaches by Cook and Wolf can already be found in the literature from as far back as 1998. Though they did not have anything to say about process visualization, they laid the foundations for this technology. The following two blog articles are intended for those of you who are about to take a decision for or against process mining. First, we will present a simplified example of what a process mining project involves before setting out the top 5 challenges in the next blog article.
Business Process Mining allows companies to identify, analyze, and improve their business processes. Event logs, which can be generated from data from the process-supporting systems, such as the ERP or CRM system, are used as the basis for doing this. This information can be evaluated by extracting the data from these systems and processing it into event logs. Process mining methods and techniques can be used to extract process information about the corresponding processes from the event logs. Based on the process information, a process model can be automatically derived, which can be used for the analysis and improvement of existing processes. For example, a resource bottleneck during the inspection of delivered goods in the incoming goods process could lead to long waiting times, which would increase the costs for the freight forwarding service. Process mining visualizes this bottleneck in an appropriate way, enabling process weaknesses and costs to be reduced efficiently. But what obstacles can lie in wait on the path to the finished process models? Let's take a look:
Process Mining Project
The process for a process mining project can be divided into six phases, which are summarized in the following diagram.
The first step is to plan the implementation of the project. In this phase, we define which business process is to be analyzed and to what extent. Along with a number of different stakeholders, the planning phase can essentially be broken down into the following three steps:
Furthermore, the business question, which is to be answered by the use of process mining, is formulated (e.g. "Which process steps significantly influence the processing time of order-to-cash processes?").
In the second phase, Process Mining first extracts the raw data needed to generate the event logs from the system. This process can be associated with considerable difficulties. For example, the required information is often only available in an unstructured form and is often distributed across many databases. Furthermore, Process Mining often extracts very large amounts of data from the systems, which requires efficient extraction procedures that are tailored to the volume of data.
The extracted information must then be prepared and transformed into the event log structure. This supposedly simple step, however, always turns out to pose a major problem in reality and therefore requires separate planning.
Once the planning has been completed, the third phase, Data Processing, can commence. In this phase, the data is filtered, for example, to prevent the process model from being falsified by incorrect event logs or process flows that extend beyond the extraction period. Through further transformations and processing steps, the extracted raw data is then used to create event logs specific to process mining, which form the basis for the actual process mining. Each event can be assigned to an activity or a precisely defined step in a process. In addition to information about the execution time, so-called timestamps, these event logs can also contain other information such as the executing user or the device or system that generated the event log.
In the fourth phase, the actual mining is carried out based on the event logs.
The heart of mining is the mining algorithm, which determines the correlations from the event logs. There are many different algorithms. Among the most important groups are deterministic, heuristic, and genetic mining algorithms. Whereas deterministic algorithms only produce defined and reproducible process models, heuristic algorithms also consider the frequency of the individual process steps. The non-deterministic approach of genetic mining algorithms, on the other hand, produces a large number of process models which, in a manner similar to natural evolution, are then combined by mutation and adaptation to form a final process model.
The resulting process model can then be used for various analytical activities. Three application areas can be identified: Discovery, Conformance, and Enhancement. During discovery, a process model is automatically created from the event logs, which visually represents the process flows. Based on these process models, the models created can then be compared with defined target processes as part of the conformance test. In the third application area, enhancement, the process model can be optimized to make business processes more efficient.
The insights gained in the previous step usually require further evaluation, which takes place in the subsequent fifth phase.
Up to this point, the processes have been identified by process analysts and examined for deviations and improvement potentials without specific knowledge of the technical process. In this phase, the resulting results are linked to the business context by the business experts.
Finally, in the final phase, it is a matter of using the knowledge gained and developing concrete improvement measures.
With the help of various methods such as Six Sigma, the real business processes will be optimized in this step by developing improvement measures in cooperation with the respective departments, business experts and key users.
We will go into more detail about the challenges await you in each of the individual phases outlined above in the next blog article. If you would like to study the topic in more depth in the meantime, you can find further reading here:
- Process Mining for Dummies Part 1 (Event Logs, Mining)
- Process Mining for Dummies Part 2 (Algorithms)
- Process Mining for Dummies Part 3 (Discovery, Conformance, Compliance Checking)
This blog post is based on a student project in cooperation with NORDAKADEMIE University, edited by Marcus Müller, Hendrik Peters and Alexander Vogt.
Once again, many thanks for the good cooperation.