Using Nearest Neighbor Search with Ad-Hoc Based Workflow Techniques to Create a Trainable Workflow System

Workflow management systems bear a huge potential for increasing the efficiency and quality of business processes. Unfortunately, the implementation of such systems is very costly and time consuming which often hinders an implementation in the area of small and medium sized companies. These costs mainly consist of the initial licensing fee and the service of an integrator to design and digitalize the processes. While hosted services can help to reduce the licensing cost, this thesis proposes a new way of reducing integration costs by creating a system that can be trained by the customers themselves. During this thesis an artifact has been created which allows users to define workflow paths for each individual instance of a workflow according to the process of the company. The artifact will support this task by giving recommendations for workflow paths according to the previously defined ones. This has been achieved by analyzing the input parameters of new process instances and comparing them with the help of nearest neighbor search to previous workflows. Once a workflow path can be clearly identified by the artifact, an autonomous execution of the workflow without supervision is possible. For evaluation of this theory, the artifact has been implemented on a test environment that includes an invoice and expenses approval process. During the evaluation, the desired functionalities of the artifact have been proven to work as intended. However, if an autonomous execution of workflows is desired, it is important to define the input parameters of a process in a way that allows clear identification of individual workflow paths. The evaluation of the artifact has also shown that users and experts show a very positive response towards the artifact and think that it should be able to cover the processes of their companies.