Predicting Individual-Level Call Arrival from Online Account Customer Activity

The data collected from a firm's online account enables enterprises to understand their consumers better and accordingly adjust their business processes. The development of effective customer relationship management strategies and the enhancement of consumers' experience with customer service centers require accurate prediction of future customer telephone queries. In this paper, we leverage the collected big data from customers' activities at a firm's online account to estimate the likelihood that an individual customer will phone the firm's contact centers within the next thirty days. Our predictive modeling approach has two distinguished characteristics: (i) predicting calls at an individual customer level, and (ii) incorporating the big data from online account activities, in addition to the customer's past telephone queries. The individual-level data used in this study is from contact centers of a major U. S. insurance firm. Various classes of features specifying the customer segment, recency and frequency of customer interactions are considered. Different neural network architectures are investigated to achieve the best prediction accuracy. Out-of-sample performance analyses evince the capability of the developed model to accurately predict future policyholders' calls at both the individual and aggregate levels.