A Robust and Intelligently Adaptive KPI Evaluation Method for Customer Call Center

The customer call center is specialized in handling customer phone calls. Like other businesses, a key cornerstone for customer call center development is the ability to effectively and consistently evaluate call center performance, to determine areas for improvement and overall growth. Executives commonly establish key performance indicators (KPIs), which put a quantitative value on crucial business objectives, to evaluate the performance of an organization. After many years of practice and development, standard KPIs have been established for customer call centers. Those standard KPIs ideally will work. However, some issues are encountered in real applications. First, data sample sizes can be small, which will make KPI estimation very unstable and inaccurate. Second, customer call centers will cover a wide range of questions. As a result, some customer calls may be very challenging to handle, while some may be easy. Agents may be grouped and assigned based on different cell types, which imply different difficulties. It is unfair to ignore such task-specific effects when evaluating an agent. One strategy is to assign different weights to calls with different difficulties. Such weights are generally static or must be adjusted manually along time, which can be inefficient. To overcome these issues, we proposed an empirical Bayes hierarchical model. It can provide a robust estimate of KPIs even when the sample size is small. The call weights will be intelligently learned and automatically adjusted from data. Its superior performance was illustrated by an application to the real data generated from the customer call center at Guangxi Power Grid Corp.