Resource provision for services that have timevarying demands has raised a great concern to service providers aiming at high-standard service quality. We propose a new resource provision approach using service simulation and arrival rate estimation that integrates unsupervised clustering and statistics techniques. We first cluster days that have similar arrival patterns together, where from each cluster we can reveal and separate days having different reasons for time-varying demands of the service. We then adopt the two layer business factor model to estimate multi-interval Poisson arrival distributions on daily bases for simulating stochastic processes. Applying simulation on queuing models with multi-interval Poisson arrival processes, we can observe stochastic changes of customer waiting time, queuing lengths and number of workers under different service strategies. We conduct a case study on an electricity service call center in real industries, showing how to build adequate resource provision and estimation against history data in past years and how the performance improved compared to their previous heuristics in real life operations.
the 13th IEEE International Conference on Services Computing, IEEE Service Society