Procurement Policies for Mobile-Promotion Platforms

Abstract

Mobile-Promotion platforms enable advertisers (individual users or businesses) to directly launch their personalized mobile advertising campaigns. These platforms contract with advertisers to provide a certain number of impressions on mobile apps in their desired sets of geographic locations (usually cities or zip codes) within their desired time durations (for example, a month); the execution of each such contract is referred to as a campaign. To fulfill the demands of the campaigns, the platform bids in real-time at an ad exchange to win mobile impressions arising over the desired sets of locations of the campaigns and then allocates the acquired impressions among the ongoing campaigns. The core features that characterize this procurement problem – supply is uncertain, supply cannot be inventoried, and there are demand-side commitments to be met – are applicable to a variety of other business settings as well. Our analysis in this paper offers near-optimal policies for both a static model and a dynamic model of campaign arrivals. The static model represents a subscription-based setting, where customers provide information of their campaigns in advance to the platform. The dynamic model represents a setting where campaigns arrive randomly and the platform reacts to these arrivals in real time; for this model, our rolling-horizon policy periodically re-computes the platform’s procurement (or bidding) and allocation decisions. We establish performance bounds on our policies for both models and demonstrate the attractiveness of these bounds on real data. By isolating important structural features of a given set of campaigns, we discuss practical implementation issues and offer procurement-policy recommendations for a variety of settings based on these features.