Multiple Digital Sellers and Consumer Click Stream Behavior: Lead Generation in A.I. Driven Democratized Online Platforms
In electronic market places, optimal number of digital sellers can improve sales lead generation and consumer choices by minimizing costly decision making efforts. The ensuing buyer-seller relationship can improve chance of consumer clicks (or lead generation) on implicit recommender systems leveraging collaborative filtering that forecasts consumer’s priorities based on those of resembling consumers. Despite these obvious payoffs, extant marketing literature appears to have ignored the influence of multiple digital sellers on consumer’s click stream behavior and subsequent lead generation. Using close to 18 million observations from a well-known A.I. driven e-commerce platform, I monitor over 800 digital sellers’ attempts to recommend some 2.3 million offers to over half a million customers for over 270 product categories through four weeks. In a first in marketing literature, I first parametrically sample from posterior distributions suited for a high-dimensional Bayesian model, using Markov chain Monte Carlo, and then use a centered Dirichlet prior mixture to non-parametrically model both endogeneity and heterogeneity. The author uses this clickstream data to also model contingencies (number of users, number of offers and numbers of categories) related to the effect of multiple digital sellers on consumer clicks. For a ten percent increase in number of sellers, at the overall platform level, I see a 5.2% drop (linear effect) and a 0.1% increase (nonlinear effect) in the odds of a consumer’s click; however, for the nine product-categories studied, the linear effect ranges from a 25% drop (mobile accessories) to a 0.4% increase (cellphones), while the nonlinear effect ranges from a 1.4% drop (building material) to a 6.2% increase (mobile accessories). This research has implications that can enhance the demand-supply interface in electronic markets.
Keywords: digital sales, two-sided markets, recommender systems, implicit feedback, click-stream data, network effects, hierarchical bayesian model, endogeneity, heterogeneity