Recurrent Neural Network based Automated Workload Forecasting in a Contact Center

Abstract

In a contact center, Customer Service Agents (CSAs) provide product support or valuable information to customers. A key requirement in a contact center is to balance customer satisfaction, by having enough CSAs to support incoming calls, and not too many to reduce costs. This project aims to simplify and improve forecasting and scheduling of CSAs by contact center administrators or operations managers. An approach that helps to forecast demand for required services are described and it assigns relevant services to CSAs without the administrative overhead. Workload forecasting helps to predict the service demand that can help to manage incoming call peaks, utilize CSAs precisely, and minimize the idle time of CSAs in a contact center. Our approach is to look at the previous historical contact center data for an extended period, learn the patterns and trends, and then forecast the overall incoming call count for each service supported by the contact center. It has been found that, neural network techniques namely, traditional Recurrent Neural Network (RNN) and its variants Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and Bi-Directional LSTM (BiLSTM) were capable of forecasting incoming calls volumes and the effects of this forecasting supported accurate CSA scheduling.