Search and Learning at a Daily Deals Website

• Published in: Marketing science (Providence, R.I.) R.I.), 2019-07, Vol.38 (4), p.609-642
• Main Authors: Hu, Mantian (Mandy), Dang, Chu (Ivy), Chintagunta, Pradeep K.
• Format: Article
• Language: English
• Subjects: Consumer behavior; Consumers; deep learning; Dirichlet updating; dynamic search model; Fatigue; Learning; learning model; Probability; Quality; Tenure; Uncertainty; Websites
• ISSN: 0732-2399; 1526-548X
• DOI: 10.1287/mksc.2019.1156

We formulate a dynamic model of search and Dirichlet learning to explain consumer behavior at a daily deal website. We study consumers’ purchase behavior on daily deal websites (e.g., Groupon promotions) using individual clickstream data on the browsing history of new subscribers to Groupon between January and March 2011. We observe two patterns in the data. First, the probability that a given consumer clicks on a merchant in the emailed newsletter declines over time, which seems to be consistent with the notion of consumer “fatigue”—a phenomenon highlighted by the popular press. Second, the probability that the consumer makes a purchase conditional on clicking increases over time, which seems contrary to the notion of “fatigue.” To reconcile these two observations, we propose a model that rationalizes these patterns and then use it to provide insights for companies in the daily deal industry. When consumers first subscribe to a daily deal website, they are unlikely to be fully informed about the quality of the deals offered on that site. The daily newsletter provides only the price and some limited information about that day’s featured deal. To learn more about quality, consumers need to click on the emailed newsletter; this takes them to the deal’s website, where they invest time and effort to learn about the deal’s quality. Such a search for information is costly. Furthermore, consumers do not know about the quality of deals they may receive in the future. Given the cost of searching and the uncertainty about the quality of future deals, consumers are more likely to search early on (i.e., click on the newsletter) in their tenure. As they learn about the distribution of the quality of deals on Groupon, they require less searching, resulting in a decline in clicks over time. As learning accumulates, consumers are better at recognizing the position of a deal in the quality distribution of Groupon deals and are therefore more likely to purchase the clicked deals. This results in an increase in the conditional probability of purchasing. We formulate a dynamic model of search and Dirichlet learning based on the above characterization of consumer behavior. We show that the model is able to replicate patterns in the data. Next, we estimate the parameters of the model and provide insights for managers of daily deal websites based on our findings and policy simulations.