Forecasting and Dynamic Updating of Uncertain Arrival Rates to A Call Center

Abstract

Motivated by queueing models and recent empirical studies of call centers, we model call arrival processes as inhomogeneous Poisson processes. Our primary interest lies on forecasting the unobserved intraday call rate profile using the historical call volume data. We develop methods for both interday forecasting and dynamic intraday updating of call arrival rates. Such forecasts are of great importance for effective call center workforce management. Our methods combine the data-driven approach in Shen and Huang (2007) [9] with the model-driven approach in Weinberg et al. (2007) [10]. A Poisson factor model is first formulated to achieve dimension reduction. We then describe how the estimated model can be used to provide interday forecasting as well as intraday updating. Our methods show very promising results in an application to real call center data.