Call centers have become an integral part of modern business to facilitate interactions between customers and companies. To determine a proper staffing level is crucial for successful operations of a call center. This workforce management process aims at enhancing the customer service quality and reducing operational cost. In this thesis, two related research questions are studied from different aspects.

The first question concentrates on forecasting call center workload. Our research proposes a new doubly stochastic Poisson process model for call center arrivals to capture both the fixed arrival pattern and the over-dispersed randomness of call arrival process. This model treats the uncertainty of the arrival rate as dynamic while many existing models consider it static. A Bay...[ Read more ]

Call centers have become an integral part of modern business to facilitate interactions between customers and companies. To determine a proper staffing level is crucial for successful operations of a call center. This workforce management process aims at enhancing the customer service quality and reducing operational cost. In this thesis, two related research questions are studied from different aspects.

The first question concentrates on forecasting call center workload. Our research proposes a new doubly stochastic Poisson process model for call center arrivals to capture both the fixed arrival pattern and the over-dispersed randomness of call arrival process. This model treats the uncertainty of the arrival rate as dynamic while many existing models consider it static. A Bayesian approach via the Markov chain Monte Carlo method is developed for the parameter estimation as well as forecasting. In addition, we consider intra-day updating of the arrival forecasts, a challenge faced by call center managers in real time. A particle filtering method is proposed to update the posterior probability distribution of the arrival rate in a sequential manner given the new observed arrival counts. Numerical results demonstrate that our forecasting approach is accurate and outperforms existing models.

The second part investigates the calibration problem of call center performance measures. In our work, a semi-parametric framework with shape constraints is proposed to calibrate a theoretical queueing model to match empirical performance measures. This calibration framework integrates the implicit domain knowledge of call center operations in the empirical dataset and the theoretical queueing system models. With the additive form of our framework, the calibration functions reveal the difference between the strict queueing theory assumptions and the empirical operations. The calibrated service performance fit both the simulated and empirical data well and is easy to interpret. In addition, a staffing level can be uniquely determined after the calibration to meet the target service quality. The service quality of a simulated call center operations with calibrated staffing level is close to the pre-set target. Finally, we simulate call center operations with real arrival data and find that our staffing policy delivers a system performance that is much closer to the target service quality than many commonly used staffing policies.