Given a workflow system used to process orders for services, the authors assert that a faithful model of the system, loaded to reflect the actual system's state, can then be used to predict performance. Building faithful models of processes with high degrees of uncertainty can be very challenging, especially where this uncertainty exists in terms of processing times, queuing behavior and rework. Most of the literature focuses on predicting system-level performance where the servers in the system exhibit standard queuing. The authors will instead present the theory and methodology for predicting performance for an individual job in an environment where the queuing behavior is not standard. The context that the authors will use to address the aforementioned uncertainty is a multi-tiered workflow system used to accept orders for training services and return proposals against those orders. The authors will specifically explore the use of machine learning and embedded discrete event simulations to analyze and predict individual job due dates.