A major tenet of the service-oriented computing paradigm is the just-in-time approach to composing web services. While the majority of related research projects use nonfunctional evaluators to choose the best, cross-organizational composition scheme from many options, few projects leverage the real, domain-specific functional data. In this paper we describe a data-centric modelling approach that leverages the real operational data of any web service workflow. This model is a core aspect for the control of adaptive collaboration software that integrates web services that span distributed enterprises. Using Bayesian inference, we assess the message flow of long-running service workflows, and, in this paper, we introduce an operational model that represents the service workflow behavior. We believe that our operational model can be used for off-line or real-time adaptation of operational service workflows.