Balancing Results and Revenue

At Google and other search sites, advertisers bid for search terms in auctions run by the search engine; higher bids typically result in more prominent placement for the accompanying ads. But because advertisers pay the search engine only when users actually click on these ads, the Googles of the world have a strong incentive to analyze which ads are successful and, when deciding the winning bidder, consider that expected performance along with the amount of the bid. At the same time, any search engine must take its own users’ interests into account when displaying sponsored search ads, which may not always line up with the interests of its advertisers.

The dynamics of this incredibly lucrative marketplace intrigued Song Yao, an assistant professor of marketing at the Kellogg School of Management, who wanted to know how the interactions between search engines, users, and advertisers affect profits and consumer satisfaction. “As a consumer and search engine user myself, I noticed these sponsored ads, and I wanted to learn about the processes behind them,” he says. But since almost no empirical data about these outcomes existed, Yao decided to design a model that could be used as a foundation for exploring financial outcomes of keyword markets for search engines and their advertisers, as well as the welfare of search engine users.

Two-sided Markets

Sponsored search ad auctions are an example of two-sided markets: the search engine connects two separate groups (advertisers and users) who hope to benefit from each other. Credit card companies, HMOs, and ad-supported social networks like Facebook and Twitter also engage in two-sided markets. In an ideal two-sided market, all three actors in the system benefit from the interaction. In the case of sponsored search ads, an ideally balanced system would generate healthy profits for the search engine, supply advertisers with a sustainable customer conversion rate for their keyword bids, and display relevant ads to the right customers without degrading their searching experience on the site.

Maintaining this tricky balance is key to any search engine’s success, and not surprisingly, Google closely guards the details of its keyword-auction algorithm. In order to test his sponsored search advertising model, Yao obtained real-world data on user behavior, advertiser bidding, and keyword auction pricing from a smaller search engine (which, for the purposes of Yao’s research, remained anonymous). “We wanted to build a general framework in which we can try to calibrate the parameters that are guiding the behavior of these three parties,” Yao says.

Yao and his co-author, Carl F. Mela, a professor at Duke University, investigated the financial and user-satisfaction impact of three different policies that the anonymous search engine might implement with regard to its sponsored search ads. The first was allowing users to sort and filter search results. “Intuitively, this should increase customer satisfaction—for example, if I go to Expedia and search only for four-star hotels, I’m going to get the results I want much easier than if I saw a list including three- and two-star hotels,” Yao explains. “The problem with filtering from the advertiser’s perspective is that if I win the keyword auction, the result may be filtered out by the user, so my incentive for participating in the auction may be lower.”

When applied to the data, Yao and Mela’s model bore out these assumptions: a policy allowing sorting and filtering increased consumer welfare by 3.8 percent, and the negative effects on advertisers reduced the attendant search engine profits by 2.6 percent. However, the model also showed that in the long run, this better user experience enhanced overall profits for the search engine by 2.9 percent. The change would attract more searchers, which would increase exposure for ads that do not get filtered out. “There’s a loss on the ad side, but the net effect of offering sorting and filtering features is positive,” Yao says.

Different Approaches

Another policy that Yao and Mela considered was market segmenting and targeting. When a search engine allows targeted and segmented markets in its keyword auctions, it reduces competition among bids because the advertisers may bid in different markets, potentially lowering the search engine’s revenue. But it also allows advertisers to bid on the customers they actually want, which results in high-quality ads that users find useful. Yao’s model proved both assumptions correct. Plus, market segmenting and targeting resulted in a 1 percent increase in revenue for the search engine. “Because both parties in the market [the advertisers and users] are happy, the search engine benefits too,” Yao says. “It’s win-win-win.”

The last search engine policy was implementing second-price auctions in keyword markets. Unlike first-price auctions, in which the winning bidder must pay the amount that they bid on the item, second-price auctions (such as those run by Google’s keyword markets) allow the winning bidder to pay the amount bid by the “highest loser” instead. If the winning bid on a keyword is $10, and the next highest bid was $9, the winner pays $9 instead of $10.

“Economic theory says that this will make advertisers change their bidding behavior,” Yao says. “If I’m in a first-price auction, I’ll shave or reduce my bid a bit because I know I’ll actually have to pay that amount if I win, and I don’t want to overbid. But if I’m in a second-price auction, I know that if I win I’ll pay less than what I bid anyway, so I’ll just bid an amount that actually matches how I value the item.”

Yao and Mela’s model empirically validated the economic theory that second-price auctions encourage this “truth-telling” behavior from bidders: second-price keyword bids averaged 98 percent of their valuations, whereas first-price bids averaged only 70 percent. However, the model also showed that this second-price bidding behavior had a negligible effect on search engine revenues, which is consistent with existing theories on auctions.

Beyond Search

The value of his model comes from precisely this kind of empirical validation of “things we already expect,” Yao says. “Without doing this kind of research, we’ll never know whether these things we intuit or theorize about economic behavior are actually true, and we’ll never know the magnitude of the effects.” Yao is working on integrating his model “to apply to more contexts besides search engines,” such as the online microlending markets at Kiva.org. He admits that any search engines that implement his model are unlikely to acknowledge his contribution, because “these things become trade secrets.”

But he says that advertisers can also benefit from modeling the dynamics of the sponsored search markets they participate in. “Advertisers call me all the time, because these markets are like a black box to them,” Yao says. “But by using this model, they may be able to start to see inside.”

Related reading on Kellogg Insight

Increasing Revenue from Online Auctions: Buyer-seller interactions affect customer value in two-sided markets

Why Markets Tip to One Platform (or Not): Uncovering the secrets of market tipping

Not All VCs Are Created Equally: High-quality venture capitalists can improve financial management and reporting