Exploration and Analysis of Undocumented Processes Using Heterogeneous and Unstructured Business Data

Abstract

The business world has become more dynamic than ever before. Global competition and today's rapid pace of development in many fields has led to shorter time-to-market intervals, as well as more complex products and services. These developments do often imply impromptu changes to existing business processes. These dynamics are aggravated when unforeseen paths have to be taken like it is often the case when problems are solved in customer support situations. This leads to undocumented business processes which pose a serious problem for management. In order to cope with this problem the discipline of Process Mining has emerged. In classical Process Mining, event logs generated for example by workflow management systems are used to create a process model. In order for classical Process Mining to work, the process therefore has to be implemented in such a system, it just lacks documentation. The above mentioned impromptu changes and impromptu processes do, however, lack any such documentation. In many cases event logs do not exist, at least not in the strict sense of the definition. Instead, traces left by a process might include unstructured data, such as emails or notes in a human readable format. In this paper we will demonstrate how it is possible to search and locate processes that exist in a company, but that are neither documented, nor implemented in any business process management system. The idea is to use all data stores in a company to find a trace of a process instance and to reconstruct and visualize it. The trace of this single instance is then generalized to a process template that covers all instances of that process. This generalization step generates a description that can manually be adapted in order to fit all process instances. While retrieving instances from structured data can be described by simple queries, retrieving process steps from unstructured data often requires more elaborate approaches. Hence, we have modified a search-engine to combine a simple word-search with ad-hoc ontologies that allow for defining synonym relations on a query-by-query basis.