Abstract / Description
Business analytics increasingly relies on analytical performance models to support real-time, data-driven decision-making in multi-stage service processes. Discrete-event simulation and process mining offer valuable insights but face scalability and explainability limits when closed-form performance indicators are needed for optimization. This paper develops a systematic procedure for constructing BCMP queueing-network models directly from the structural properties of a business process formalized in BPMN. The framework introduces a six-level state-space hierarchy (V0–V5) that ranges from a full path encoding to a single system-level counter, and three allocation strategies (D1–D3) that govern how the actions of an organizational unit are distributed among its workers. A product-form stationary distribution is derived for the resulting open network with FCFS and infinite-server nodes, yielding closed-form expressions for the mean sojourn time at a node and for the mean process flow time. We validate the framework on the university admissions campaign at RUDN University (2023 cycle, more than 85,000 applications). Numerical experiments show that the baseline mean flow time of about 174 min can be reduced to 111 min—a 33% improvement—by only a 3% increase in total staffing, allocated by greedy coordinate descent. The closed-form model agrees with discrete-event simulation under exponential service times to within 1%, and its analytical tractability enables real-time staffing recommendations that simulation alone cannot deliver.