Personalized Sales Targets with Customer Choices

Abstract

Advertisement, as a major revenue source of e-commerce platforms, is an important online marketing tool for sellers thereof. In this paper, we explore the dynamic ad allocation with limited slots upon each customer arrival for e-commerce platforms when the advertisers specify, for each ad, the budget constraint, the time periods when the ad should be displayed, and the click-through (lower-limit) constraint for certain customer segments. The goal of the platform is to maximize its payoff over the entire horizon. We propose a two-stage stochastic program framework in which the platform first decides the click-through goals for each ad/customer-type pair, and then devises the ad allocation policy to satisfy these goals in the second stage. We show that the optimal click-through goals can be achieved efficiently by solving a convex program, which can further reduce to a scalable linear program if the customer click-through behavior follows the multinomial logit model. Moreover, we provide a family of debt-weighted algorithms to achieve the optimal click-through goals, and prove that they are asymptotically optimal when the problem size scales to infinity. Compared to choice-based linear programming and its variant, our approach has better scalability and can deplete the ad budgets more smoothly throughout the horizon, which is very much desirable for the online advertising business in practice.