A Data-Driven Approach to Predict an Individual Customer's Call Arrival in Multichannel Customer Support Centers

Abstract

The availability of big data collected by multichannel contact centers creates opportunities for businesses to more accurately predict future interactions with their customers. This paper presents a data-driven modeling approach to forecast the likelihood of a call arrival by an individual customer within the next thirty days, based on the multichannel data from contact centers. This model incorporates information related to the past Web activities of an individual customer to predict his future telephone queries. Our study relies on big datasets from contact centers of one of the largest U.S. insurance companies. Various characteristics related to the customer segment, recency and frequency of customer interactions, and cross-class features are considered. We find evidence that some of the recent web activities of a policyholder significantly increases the probability that the policyholder would make a telephone call in the next 30 days. In addition, recency and frequency of contacts impact the probability of the policyholder's call, for a specific set of reasons for past contacts.