Although the variation of business processes have been recognized very large, its nature and effects have been rarely investigated due to the difficulties in measurement and analysis. In this thesis, we analyzed the probabilistic behaviors of business process times, and an effective BPM methods which considers the time variations are suggested.
To acquire the evidence of business process variations, empirical data of business process execution times from various domains such as finance, service, S/W, and manufacturing industries were collected and analyzed. We found heavy tails which imply that very small parts of business process operation can take extremely long time durations. Based on the literatures which concerned the nature of human actions, possible factors and effects of heavy tails are described. In addition, we found self-similarity which the pattern of processes are similar regardless of observation scale using R/S statistics and Variance-time plots.
If actual execution times of business processes and activities are heavy tailed, the traditional BPM methods can be ineffective. We proposed a workflow engine based simulation which enacts a workflow virtually under predefined conditions for business process planning. Because heavy tailed activities raise difficulty in prediction, simulation may be useful with its realistic and flexible features. In operation phase, we developed am agent-based monitoring system for process variations. Customized software agents were used to automate process data gathering, processing, preservation, and diagnosis. Lastly, we evaluated the empirical time data using process capability indices, and the results represent that normal process indices overestimate than those of non-normal process indices because of skewness of time data.
