Estimation and Analysis of Callers' Behavior in Call Centers

This thesis studies callers' abandonment behavior in call centers. I propose a new method to estimate callers' delay sensitivity from the data, and show how we can use this information to predict the impact of changes in the priority policy of a call center or the delay information conveyed to callers.;In the first chapter, I model callers' decision making process in call centers as an optimal stopping problem. The utility of a caller is modeled as a function of her waiting cost and reward for service. I estimate the cost and reward parameters of the callers using the data of individual calls made to an Israeli call center. I also conduct a series of counterfactual analyses that explore the effects of changes in service discipline on resulting waiting times and abandonment rates.;In the second chapter, I undertake an empirical study of the impact of delay announcements on callers' abandonment behavior and the performance of a call center with two priority classes. Given callers' parameters estimated from the data, I predict callers' behavior in settings with announcement patterns different from that in the data, and analyze the impact on the system performance. Using a Markovian approximation to simplify the queueing analysis, I develop a methodological framework to find the steady-state equilibrium by combining the estimation of callers reward and cost parameters, the model of callers' abandonment behavior, and the queuing analysis that incorporates this behavior.;The third chapter describes how the Stochastic Kriging approach can be used to find the impact of delay announcements in multi-class call centers. I use this approach to derive the waiting time distributions and the steady state probabilities as functions of the abandonment time distributions. This derivation was done in Chapter 2 using a Markovian approximation. However, the Markovian approximation presented in Chapter 2 is sensitive to the priority policy of that particular two-class call center. In addition, developing the same approximation for multi-class class call centers is cumbersome. In contrast, the Stochastic Kriging approach can be used to study call centers with any priority policy and a larger number of classes.