Multivariate statistical process control approach for performance evaluation in call centers

Abstract

In this thesis, a multivariate performance assessment model for the inbound department of a call center was proposed. The model is treated with principal component analysis (PCA). The main reasons for the employment of PCA are its simplified signal decomposition ability, adaptability in recursive approaches and autocorrelated structure of the collected data. Due to the time-varying dynamics of the call center processes, conventional static PCA algorithms are not suitable for such an application. For this reason, a new recursive PCA algorithm which is capable of tracing time-varying nature of call center dynamics, was constructed as an alternative to the previously introduced recursive algorithms in the literature. In the proposed model, six performance indicators were defined for the inbound call center process. While these indicators represent the variables on the control charts, the days stand for observations. Consequently, a signaling point on any statistical control chart designates to a day at which the performance of the call center process has significantly drifted. Two years of data were processed through the model and the results were analyzed. Additionally, to make our performance assessment model easier to understand and more practical, a decision support table involving the mostly encountered types of signaling days and their practical interpretations was given at the end of the study. Also, the study showed that the PCA method which has been mostly used for industrial processes, may efficiently be adapted into a service system performance evaluation works.