An Empirical Study On Deciding Stuffing Numbers In A Call Center

With the high-speed explosion of mobile telephone business, it has been a state-or-art operations strategy to build telecommunication call centers and provide 24-hour hotline customer service. Service level of a typical inbound call center is measured by TSF (Telephone service factor) and ASA (Average speed of answer), which is under the setting of queuing theory. The dissertation describes a telecommunication call center providing free customer service and dealing with customer complaints via telephone. The caller has to wait before getting service if there is no agent idle. The longer the caller waits in the tele-queue, the higher probability that the customer would abandon the queue and turn to the competitor. Hence, to assure the high proportion of customers being served within 20s has been the primary goal of this call center.As a random quality-efficiency-driven service system, the call center manager has to decide the number of agents arranged in each time interval within a day to match the randomly varying arriving calls and retain the high service level. Erlang C queuing model is commonly used in deciding the number of agents needed. However, several assumptions which this queuing model is based on cannot hold in real call center application. Based on the analysis of three-month call center operational data, this dissertation addresses the difference between ErlangC and ErlangA queuing model and proposes ErlangA as the appropriate model in estimating the number of agents needed. The three key parameters of ErlangA queuing model is further discussed based on the study of time-varying call arriving process and customers'abandonment in real queuing and the statistical distribution of call duration. The number of agents calculated through ErlangA model is not sufficient in a high utilization call center. To reduce the random variation effects of high utilization system, the dissertation proposes an additional safe stuffing numbers through square-root stuffing rules to guarantee the expected service level.The innovation of this dissertation is the proposal of a new call center model of calculating the number of agents in different time intervals within a day. The lognormal distribution of call duration is verified and the estimation of lognormal mean and confidence interval addressed. Finally the number of agents needed in high utilization system is calculated based on ErlangA queuing model.