An exemplary process management server disclosed herein comprises a machine-learning model that may be trained to expose processes from a message stream in response to a training table. In one embodiment, one or more performance metrics of the exposed processes may be monitored to identify a process anomaly or other change in process performance, and to dynamically modify one or more process components in response thereto. Such an arrangement improves system performance, efficiency and resource utilization. For example, system performance may be improved by re-ordering process steps to minimize resource overlap. Efficiency and resource utilization may be improved by re-ordering process steps to maximize parallel processing and reduce lag times and bottlenecks.