Modelling and managing customers in a multichannel setting
Customers are increasingly shopping across multiple channels of the same retailer. While many retailers recognize that these multichannel shoppers are the most profitable, little is understood about how shoppers' channel preferences evolve over time, especially when they are new to the firm, or how these preferences should inform the firm's direct marketing policy. In this dissertation, I examine how the channel purchasing behavior of new buyers develops over time and in response to the impact of direct marketing. The dissertation is structured in two parts. In the first, I build a rich descriptive model of the consumer that captures the evolution of customer channel purchasing over time, and includes all relevant behavior: whether to purchase, which channel to purchase from and how much to purchase. In the second part, I use recent numerical developments in dynamic programming to find the optimal resource allocation using the parameter estimates from the consumer model. I find that marketing has different effects over the stages of customer evolution. I find evidence in support of 2 dynamic states in the population - a migration state that starts offline and gradually moves online, and a hardcore offline loyal state. This provides robust evidence that indeed channel migration is taking place, but only for one group of customers. I find that emails and both emails and catalogs sent simultaneously are effective, only in this first state, while catalogs are effective in the second state. For the second part, I find that the optimal policy increases discounted expected profits by 15 %, with a range of 5-35 %. Overall I find, across a set of representative profiles, that the optimal policy is to send either emails or both when the likelihood of being in the migratory, online segment is high. Conversely when the posterior probability is low, and the customer is likely to be in the hard core offline segment, the optimal policy is either to send nothing or to send only catalogs.
