Tuesday, January 21, 2014

An Information Stock Model of Customer Behavior in Multichannel Customer Support Services

Read the working paper
INSEAD Working Paper 2014/06/TOM revised version of 2012/73/TOM

In this paper, we develop a novel information stock-based model to capture patterns in customers’ observed behavior in a multichannel customer support setting (e.g., web and phone) using data of a US based health insurance firm. In case of a query regarding health insurance coverage or claims, customers can either call the firm’s call center to get the desired information from a call center representative, or visit the web portal and get the desired information themselves. We model a customer’s observed behavior, in terms of her query frequency and channel choice for queries, as a stochastic function of her latent information stock. Specifically, we assume that each customer has a latent “information stock” which is a function of customers’ “information needs” (which arise when customers file health insurance claims) and “information gains” (which customers obtain when they contact the firm’s support channels to resolve their queries), and other factors such as seasonal effects (for instance, queries that arise at the time of annual contract renewal). We estimate our model on individual-customer-level data from the firm. Based on the estimates, we find that the average information gain for a customer from a telephone call is three times higher than that from visiting the web portal. In addition, customers prefer the telephone channel for health event-related information needs but, interestingly, prefer the web portal for seasonal information needs that are typically more structured. We also find that customers are very heterogeneous in terms of their propensity to use the web channel, and can be broadly classified into “web avoiders” and “web seekers.” Our model is general enough to be applicable in other similar multichannel contexts and it can be used to aid in call center management and staffing decisions as it provides very good out-of sample predictions for future query volumes on different channels at the aggregate and individual levels, and it can even help to accurately identify customers who are expected to have high telephone call volumes in the near future.