Advertisers are living in a new golden age. Online tracking of consumers’ web-browsing history makes targeting potential customers easier than ever. Online publishers are reaping the benefits of tracking, too, since they can now offer advertisers access to more finely tuned segments of consumers. The ever-more granular nature of online advertising is being increasingly facilitated by online ad exchanges, platforms where automated auctions for ad slots enable publishers to match advertisers to their desired audiences in real time.
Historically, large online publishers would manage online ad sales much as they had managed print ad sales, offering advertisers guaranteed contracts that ensure, for example, that an advertiser would have his ad viewed by a given number of viewers with pre-specified characteristics over a pre-defined period of time. Today, ad exchanges offer advertisers an alternative to such contracts, with a notable difference. These exchanges give advertisers the ability to decide in real time (and at near-lightening speed) if they want to buy each individual impression based on specific user information, such as distant and recent browsing history — and even the exact ads or merchandise a reader was looking at just prior to visiting the site.
As a consequence, when a reader clicks on a webpage an invisible but ferocious high-speed battle for her attention takes place before the page finishes loading — in less than 1/3 of a blink of an eye. Publishers put the reader’s profile and impression up for auction, vying with advertisers who enter a bidding war with each other, with the highest bidder winning the impression. Millions of these auctions happen each second, every time a reader clicks to a new page.
The thousands of new options this gives advertisers for how to use their marketing campaign budgets makes online advertising more competitive all the way around. In addition, the growth of the transactions performed though ad exchanges has created the need for publishers to better understand the analytics associated with this new channel.
Professors Omar Besbes and Gabriel Weintraub, together with doctoral student Santiago Balseiro, have proposed a new framework that helps publishers design these auctions and manage this new channel. Exploiting the unique features of the ad exchange market — its millions of daily transactions with many advertisers at any point in time — they use game theory and yield management to explore the structure of the market, and based on this, to optimize the auctions.
“One naïve way to think about analyzing these markets would be to observe how advertisers bid in the past and design the auctions based on the assumption they will continue bidding that way in the future. But this ignores the fundamental fact that any change in the rules of the auctions, volume in the exchange, and the information publishers provide about the users is likely to influence the bid decisions of advertisers, who use more and more sophisticated analytics to run their campaigns,” Besbes says. “Understanding those reactions is key to understanding how to design the auctions, and in the absence of a transparent way to account for such reactions, one is often stuck with the naïve approach. The framework we propose allows one to characterize explicitly advertisers’ reactions, essentially mapping how market and auction parameters and advertiser characteristics translate into bids.”
Traditional tools of auctions and game theory are not well suited for studying the complex interactions between advertisers in ad exchanges. One innovation of Besbes and Weintraub’s research is the development of a tractable theory to analyze this complex market, which is both behaviorally appealing and approximates well the rational behavior of advertisers. The framework and the equilibrium concept underlying it, the Fluid Mean Field Equilibrium, is pared down enough on the computational end to be practical for researchers and practitioners alike, allowing them to quantify the tradeoffs that publishers face when considering changes in the auctions.
Examples of key questions that publishers face when managing this channel include how to set the reserve price — the minimum price a publisher is willing to accept for a given impression, and a decision that is key to maximizing revenues; whether to funnel readers to satisfy impressions for guaranteed contracts or to ad exchanges; or, more unique to ad exchanges, how much information to provide about the user.
“We propose that publishers provide as much information as they can as long as they simultaneously adjust the reserve price,” Weintraub says. “While the increase in information decreases competition for each impression, since it narrows the scope of possible advertisers, the value of each impression increases because any advertisers looking for the more narrow segment are more likely to reach a prime target. By adjusting the reserve price accordingly, publishers can still extract this value even as they attract fewer competitors.”
Advertisers can also use the framework, which offers a simple and intuitive formula for how and when they should shade their bids. Bid shading — underbidding on what they think the actual value of the slot will be — is a common strategy advertisers use to spend down their campaign ad budgets at a constant rate over the length of their campaigns. This strategy can be especially useful for advertisers with smaller budgets: these smaller players may lose many auctions by bidding low, but protect themselves from spending too much too soon, and are making a bet that they will win enough good deals over time.
Because the complexities of ad exchanges can’t be captured in a single model, the researchers’ future work will consider yet-unexplored aspects associated with ad exchanges. For example, how do the actions of intermediaries in the online exchanges impact the actions of advertisers in the auctions, and how do those in turn change the way publishers should design their auctions?
Though their framework was developed with display advertising in mind, it has the potential to be adapted to other industries. “Sponsored search, while distinct, is a close cousin of online display advertising, and has enough in common that some of the core ideas may help us better understand bids one observes in sponsored search auctions,” Besbes notes.
Omar Besbes is assistant professor of decision, risk, and operations at Columbia Business School.
Gabriel Weintraub is associate professor of decision, risk, and operations and a senior scholar at the Jerome A. Chazen Institute of International Business at Columbia Business School.
Read the Research
Omar Besbes is an Associate Professor in the Decision, Risk, and Operations division at the Graduate School of Business, Columbia University.
His primary research interests are in the are of data-driven decision-making with a focus on applications in e-commerce, pricing, and revenue management, online advertising, operations management, and service systems. His research has been recognized by the 2012 INFORMS Revenue Management and Pricing...
Gabriel Weintraub's research covers several subjects that lie in the intersection between operations/management science, and applied economics. He is particularly interested in developing mathematical and computational models for the economic analysis of problems in operations; as well as making contributions to industrial organization and computational economics. Before obtaining his Ph.D. at Stanford, Professor Weintraub was a full-time instructor at the Department...
Read the Research
Santiago Balseiro, Omar Besbes, Gabriel Weintraub
"Repeated Auctions with Budgets in Ad Exchanges: Approximations and Design"