Blog: Mind your business process? Mine your data!

Vinci-researcher Laura Maruster

Data is collected nowadays about anything, at any time, and at any place. The explosion of data is changing the way organizations capture data, analyze information, make decisions, and create value. But how to deal with so much data, and often, messy data? How to transform structured data, but especially unstructured “raw” data (data not organized according to predefined format, for example text) into business value? According to Fortune, “data is the new oil”, which should be appropriately exploited.

My research aims to demonstrate how process-related data facilitates a better understanding of business processes, and how to improve these processes.

For example, it is common to request changes to an existing product design, which can be accepted or rejected. The Engineering Change process is a challenging process: it can become very complex and inefficient, and subsequently it can negatively impact planning, scheduling and project costs. Irrespective of the company or product, the Engineering Change process is seen and logged as a standard process, involving relatively similar phases. However there seem to be significant differences between the standard Engineering Change process - the “official” documented process -, and the “actual” executed process. To eliminate process inefficiencies and redesign it, a good understanding is required; to this end, process mining can be used.

 Process mining [1] is a relatively new business management approach that enables the automatic analysis of business processes based on event logs, in order to improve quality and make these processes lean. Process mining provides insights into Key Performance Indicators such as overall process / activity throughput time, bottlenecks, rework, frequent or exceptional process flows, but also into organizational aspects such as work transfer between departments. Since 2001, the process mining paradigm evolved into a mature framework used by many researchers and practitioners (www.processmining.org ), which also led also to new commercial ventures (www.fluxicon.com ).

Process logs are rarely complete or “clean”, and usually contain noise (e.g. mistakes, or very infrequent process executions) and other characteristics that make their analysis a difficult task. In my work I develop solutions which automatically determine the process model from noisy data.

An additional challenge emerges from unstructured data, which is data not organized according to a predefined format, such as free text. For example, an Engineering Change Notice may contain a lot of unstructured information, truncated words and technical terms, which do not follow the rules of natural language. In such cases, the process cannot be identified anymore by applying existing process mining techniques, because they require data in a predefined format. A solution is to employ information extraction and text mining techniques to transform unstructured into structured information.

Process mining can be used also to identify the navigational patterns of users of web applications. Web applications can target undifferentiated types of users, but also specific users. An example considered in my work is the usage of a decision support system designed for Dutch agricultural farmers. Although perceived as a useful tool, farmers did not use the system as it has been designed. My research showed that the system should be personalized according to the farmer type. For instance, “advanced” farmers used the system to its full decision-making potential, while “normal” farmers used the system mostly for information purposes. Analyzing navigational patterns enables the redesign of IT systems to incorporate the characteristics of specific user groups.

 Such research offers firms valuable insights in how to continuously improve and innovate their business processes.  

 To know more, do not hesitate to contact me at l.maruster@rug.nl.

[1] van der Aalst, W., Weijters, T., Maruster, L. (2004) - Workflow mining: Discovering process models from event logs, IEEE Transactions on Knowledge and Data Engineering 16 (9), 1128-1142.