Analytics for labor planning in systems with load-dependent service times

This paper presents a generalized framework for labor planning in systems with load-dependent service times. Our approach integrates customer arrival forecasting, service time estimation, and staffing into an end to end process. More specifically, (i) we propose a hybrid model to forecast customer arrivals, that combines a traditional time-series technique with a state-of-the-art machine learning algorithm, thereby allowing to incorporate a rich set of predictors; (ii) we develop a methodology to estimate the distribution of load-dependent service times from past transactions data; and (iii) we present a stochastic programming formulation to determine staffing levels under quality-of-service constraints. Finally, we develop a heuristic solution algorithm that utilizes an embedded discrete event simulation to evaluate system performance.