When Is More Merrier? A Cloud-Based Architecture to Procure Impressions from Multiple Ad Exchanges

We consider an ad firm that acts on behalf of advertisers to execute mobile, in-app, ad campaigns. The firm commits to provide an advertiser a specified number of ad placements (impressions) on mobile apps, usually in a specified location, and within a specified time horizon. The supply for ad space arrives, in real time, in the form of bid requests from one or more mobile ad exchanges. The ad firm needs to bid on each impression in such a way that the goals of several ongoing campaigns are met at minimum cost. The ad firm needs to execute multiple campaigns simultaneously and get its supply (for ad space, or impressions) from multiple mobile ad exchanges. By working with more than one ad exchange, the direct cost of procuring the necessary impressions can be lowered. However, this lower cost needs to be balanced with the cost of the additional computing resources needed to work with multiple mobile ad exchanges and the (possible) extra cost of meeting the minimum spend (or participation fee) imposed by each ad exchange. Here, there are two key decisions that the firm needs to make. First, it needs to select the set of mobile ad exchanges to obtain its supply; each mobile ad exchange is characterized by specific supply uncertainties, location dependent bid curves, and a participation fee. Second, for each ad exchange and location, the ad firm needs to determine its bidding policy, that is, how much to bid for each bid request. We show that the proposed near-optimal bidding strategy, the strategy to bid at each exchange-location combination, is state independent. We first solve a general problem of selecting among multiple nonidentical ad exchanges. We next analyze the special case with identical mobile ad exchanges and show that, depending on the particular parameter setting, the near-optimal number of ad exchanges and the near-optimal bid amount can be weak complements or substitutes. Finally, we propose a cloud-based architecture to procure impressions where the ad firm uses a selective bidding strategy that can further lower procurement costs. The ideas of this paper are applied to a real problem and the savings from our approach (about 33% lower cost) are demonstrated.