Quantitative analysis of probabilistic BPMN workflows

We present a framework for modelling and analysis of realworld business workflows. We present a formalised core subset of the Business Process Modelling and Notation (BPMN) and then proceed to extend this language with probabilistic nondeterministic branching and general-purpose reward annotations. We present an algorithm for the translation of such models into Markov Decision processes expressed in the syntax of the PRISM model checker. This enables analysis of business processes for the following properties: transient and steadystate probabilities, the timing, occurrence and ordering of events, reward-based properties and best- and worst- case scenarios. We develop a simple example of medical workflow and demonstrate the utility of this analysis in accurate provisioning of drug stocks. Finally, we suggest a path to building upon these techniques to cover the entire BPMN language, allow for more complex annotations and ultimately to automatically synthesise workflows by composing predefined sub-processes, in order to achieve a configuration that is optimal for parameters of interest.