In The Plex
Veach knew in his heart this was the right way to go, but he had to do a lot of explaining. “Larry and Sergey kept asking me if it wasn’t simpler to have an auction where we just have people pay what they bid,” he says. “And I kept saying, ‘No,’ because then people have this incentive to keep lowering their bids.”
To run its ad operation, Google had hired Sheryl Sandberg, former chief of staff to the secretary of the treasury in the Clinton administration. She’d gotten to know Eric Schmidt when he visited D.C. to argue against Internet taxes. Though she’d never been involved in high tech—besides her Treasury post, her résumé included McKinsey & Company and the World Bank—she’d spent the past few years observing what was happening in Silicon Valley. Part of her job at Google was explaining its innovative auction. She kept staring at the formula, wondering why it seemed so familiar. So she called her former boss, Treasury Secretary Larry Summers.
“Larry, we have this problem,” she said. “I’m trying to explain how our auction works—it seems familiar to me.” She described it to him.
“Oh yeah,” said Summers. “That’s a Vickery second-bid auction!” He explained that not only was this a technique used by the government to sell Federal Reserve bonds but the economist who had devised it had won a Nobel Prize.
Veach had reinvented it from scratch.
One fan of Veach’s system was the top auction theorist, Stanford economist Paul Milgrom. “Overture’s auctions were much less successful,” says Milgrom. “In that world, you bid by the slot. If you wanted to be in third position, you put in a bid for third. If there’s an obvious guy to win the first position, nobody would bid against him, and he’d get it cheap. If you wanted to be in every position, you had to make bids for each of them. But Google simplified the auction. Instead of making eight bids for the eight positions, you made one single bid. The competition for second position will automatically raise the price for the first position. So the simplification thickens the market. The effect is that it guarantees that there’s competition for the top positions.”
Veach and Kamangar’s implementation was so impressive that it changed even Milgrom’s way of thinking. “Once I saw this from Google, I began seeing it everywhere,” he says, citing examples in spectrum auctions, diamond markets, and the competition between Kenyan and Rwandan coffee beans. “I’ve begun to realize that Google somehow or other introduced a level of simplification to ad auctions that was not included before.” And it wasn’t just a theoretical advance. “Google immediately started getting higher prices for advertising than Overture was getting,” he notes.
That wasn’t only because of the auction model; Veach and Kamangar had made other significant advances. One of the biggest was the adoption of the other Overture idea, pay per click. Google’s improved version of the original AdWords, called AdWords Select, would no longer charge per impression, according to how many people saw an ad. Instead, the click-through rate would become the measure of online advertising. The bids advertisers submitted would specify how much they were willing to pay each time a user clicked on the ad and was sent to the landing page on the advertiser’s website.
The longtime joke in the marketing world was that only half of advertising is worth the money—but no one can tell which half. Google was switching the game: using its system, you would pay for ads only when they worked.
That was not all. The ad model that Veach and Kamanger created had yet another major innovation, but this one was exclusively Google’s. It would become the least understood, most controversial, and ultimately most powerful component of AdWords Select: a built-in function to regulate ad quality. The new system instituted financial incentives for better ads. It lowered the price for effective ads and meted out monetary punishment and even an online ad version of the death penalty for bad ads. It also opened Google to the charge that it had created a “black box” in which advertisers could never understand, or trust, the calculations that Google made to place their ads.
Here was the rub: The bids submitted by contenders for the ad slots were only half of what ultimately determined the winners of the auction. The other half was the quality score. This metric would assure that the ads Google showed on its results page were helpful to its users—a high quality score meant that the ad was relevant to the user’s quest. Low quality scores were for ads that were irrelevant, misleading, or even spamlike. In the early version of AdWords, the sole determinant of the quality score was Google’s guess at the percentage of times a user would click on an ad when it appeared on a results page—the click-through rate. Later Google used a more complicated formula to determine quality score by adding factors such as the relevance of the ad to the specific keyword and the quality of the landing page. But the biggest factor remained the predicted click-through rate.
Say that Alice, Juan, and Ted are all bidding for the keyword “hand lotion.” Alice is selling an artisanal form of hand lotion popular in upscale spas. Juan owns a big drugstore that sells hand lotion, among many sundries. Ted has a travel site. He isn’t selling hand lotion at all but wants to expose his ad to the kind of person who buys hand lotion. Alice bids ten cents per click. Juan bids fifteen cents. Ted bids fifty cents. If you think that Ted’s high bid automatically puts him in the top position, you’re wrong. It’s quite possible that Alice, the low bidder, would get the favored spot. Google’s calculations might determine that users who click on her site are more likely to find what they want and thus assign her a very high quality score. Juan’s quality score would be downgraded—as would his effective bid—because users may go to his site and have difficulty finding hand lotion. He may be in position two, paying a little less than Alice. Ted would have an even lower quality score. People looking for hand lotion are unlikely to click on a travel ad. His bid would be downgraded even more. (He may even be required to pay a prohibitively high “minimum bid”—a practice that ultimately engendered a lot of grumbling among certain advertisers.)
The beauty of the ad quality formula, says Sheryl Sandberg, is that “it made the advertiser do the work to be relevant. You paid less if your ads were more relevant. So you had a reason to work on your keyword, your text, your landing page, and generally improve your campaign.” There were some downsides, though. Chief among them was that the system was fairly complicated and risked befuddling an advertiser.
Veach would acknowledge this. “It’s not so much that any of these ideas on their own are complicated,” he says. “But when you put the three together, you can’t easily explain it to advertisers.” On the other hand, he adds, “It actually turns into a fun mathematical problem, which I loved.”
From the start, Page and Brin had an idealistic view that Google would run ads only if users deemed them a useful feature. Using mathematical wizardry, Veach and Kamangar had come up with a mechanism to realize that fantastic aspiration. Google’s original system asked advertisers to pay a fixed rate to expose their ads on a results page triggered by targeted keywords. The new system asked advertisers to participate in an auction that determined how much they would pay every time someone clicked on the targeted ad. What’s more, by rewarding better ads the new system made users happier by increasing the odds that what appeared on the page was relevant to their queries. The system enforced Google’s insistence that advertising shouldn’t be a transaction between publisher and advertiser but a three-way relationship that also included the user.
But would it work? For one thing, executing the system was a huge technological challenge. Every AdWords Select ad would be the winner of a unique auction requiring the execution of a complicated formula. The auction would be conducted in stealth, generated the instant someone typed a keyword into the Google search box, with the result shown in a fraction of a second. “I don’t know the number of auctions that we run per day, but for purposes of argument, use a billion or a hundred million,” says Schmidt. “We run many more auctions than anyone else on the planet because we run them in real time, we run one auction per ad per
page, and that’s multiplied by the numbers of ads per page. It’s a phenomenal number. Technologically, because of latency, you have to do this very, very quickly.” Fortunately for Google, even in 2002, when the new system was completed, the company was fanatically focused on huge computational feats performed at dizzying speed on a platform of thousands of computers, so it was able to leap the technology hurdle.
The dicier challenge was getting skeptical customers of the original AdWords to leave a system they were happy with to try this complicated new one. On January 24, 2002, Google tested AdWords Select by offering it to selected advertisers. In order to lure them to the new program, Google stacked the deck: it placed ads bought under the new AdWords Select system in more favorable positions than the advertisers had actually paid for. “The old AdWords customers would say, ‘How do we get to position one?’ And we’d say, ‘Oh, you sign up to this system over here.’ They were signing up in droves, so it really simplified our lives,” says Veach. Spurred by the initial returns from higher positions, advertisers began spending more money in the new system—and getting clearly better returns. Within a month, Google simply pulled the plug on the old CPM system and sent all its advertisers an email informing them of the change.
From that point on, revenue from the right-hand side of Google’s search results page—which had previously constituted only 10 to 15 percent of Google’s ad take, with the bulk coming from the direct sales of premium ads—began rising. That area of screen real estate, which had previousxly been regarded as the wrong side of the tracks in Googleland, had been transformed as suddenly and dramatically as South Beach after Madonna bought a condo there. It wasn’t just little guys with credit cards buying AdWords Select. National corporations such as Procter & Gamble and Coca-Cola began bidding at figures that exceeded those coming from the corporations that had been occupying slots in the premium program. “There was definitely a bit of a conflict there, because now some of the internal salespeople had to deal with AdWords, almost against their will,” says Veach.
In any case, Google was reaping rewards, and 2002 was its first profitable year. “That’s really satisfying,” Brin said at the time. “Honestly, when we were still in the dot-com boom days, I felt like a schmuck. I had an Internet start-up—so did everybody else. It was unprofitable, like everybody else’s, and how hard is that? But when we became profitable, I felt like we had built a real business.”
Best of all was that Google, against all odds, was making that profit without surrendering its ideals. “Do you know the most common feedback, honestly?” Brin asked. “It’s ‘What ads’? People either haven’t done searches that bring them up or haven’t noticed them. Or the third possibility is that they brought up the ads and they did notice them and they forgot about them, which I think is the most likely scenario.” (This would track with an experiment that Google repeated regularly—the “no ads” test that compared users who saw ads with those served results pages free of sponsored links. Every time the test was run, the outcome was similar: dropping ads did not increase searching. More often than not, the users in the control group who continued to see ads searched more than those with ad-free pages. Google’s relieved conclusion: its ads made people happy.)
From that point on, Brin and Page saw nothing but glory in the bottom line. Google was profitable, and its hiding strategy was successfully masking the extent of its success. Its name was synonymous with search. The Wall Street Journal’s famous tech critic Walt Mossberg called it “the most useful site on the World Wide Web.” Everyone was asking the founders when they would have an IPO, but “it’s not an issue for us,” Page said in 2002. “Every month we make more money than the last one.”
The only slight regret? They never got those PhDs.
“I’ve been meaning to,” said Sergey.
“Maybe someday,” said Larry.
“My mom keeps asking,” said Sergey.
Larry frowned. “My mom doesn’t ask me anymore.”
Originally, Google’s goal in providing its search engine results to portals such as Yahoo and Excite was to collect licensing fees in exchange for providing a higher quality of search. Now that Google search came bundled with ads that brought cash with every click, the business model changed. Google could offer a portal not only an effective search feature but a nice share of the revenue that came from those clicks. Google’s business plan, with revenues split in thirds among syndication of search, customized search for businesses, and advertising, was delegated to the delete bin. Hereafter, ads would dominate.
Google’s main competition for the portal deals was the company that had invented ad auctions, Overture. “For a long time they were ahead of us,” says Susan Wojcicki, who began leading the ad team in 2002. “But now we had a more targeted ad system that could generate better results for our advertisers and more revenue for our publishers.” Google’s first breakthrough was a deal with the Internet service provider EarthLink. On the day the arrangement was announced, Overture’s total stock value dropped by $800 million. But the big whale was AOL, the dominant portal on the Internet, with hundreds of millions of daily visitors. Its contract with Overture was due to expire in 2002. “It was one of the biggest profit centers at AOL, making them hundreds of millions,” says Bill Gross. “We would put our paid results at the top of the list for AOL search queries. Once they got hooked on that heroin, there was no changing.”
Nonetheless, AOL was eager to have Google and Overture compete for the new contract, and its huge audience allowed the online service to dictate onerous terms, including a huge guarantee, requiring the winner to pay AOL a giant nonrefundable advance on sales. Google’s executives were split on whether to meet its demands. “There was real risk,” says Wojcicki. “We could make $40 million on the deal, or we could lose $40 million. We only had $10 million in the bank. So it really mattered who was right.”
Eric Schmidt, the CEO for only a year and not yet unconditionally trusted by Brin and Page, thought it too risky. “I was the conservative, everyone else was a liberal,” he says. But the founders were gung ho. To make sure that not every minute of their interaction was spent arguing about AOL, Schmidt suggested that they limit the discussion to a daily bout at 4 p.m. “We would haul everybody in and just argue the numbers,” he says. Ultimately, Schmidt would take his case to the board—and find that in this case the VCs were willing to back Brin and Page. “The board said that in the worst possible case, they would come up with the $50 million, so it wouldn’t bankrupt the company,” says Schmidt.
As negotiations progressed, Omid Kordestani became a familiar figure in Dulles airport, near AOL’s Vienna, Virginia, headquarters, trying to convince AOL that this not-quite-ripe company would be able to satisfy all its requirements in a big ad deal. AOL wanted to know the difference between Google’s auction and Overture’s. One of those requirements was that the winning company have a broad sales force. “There was a perception that there weren’t a lot of people working for Google,” says former Google ad exec Jeff Levick. And, he admits, the perception was accurate. When AOL did an onsite visit to New York, “we had to physically marshal in people to make it look like we were a real company,” he says.
Google was better positioned for the deal than Overture was. First, its search technology was better. Also, adding AOL users to its search traffic would increase the value of Google ads, even those served at www.google.com, because it would have a larger inventory for search ads, and more spirited bidding. As a result, Google could afford to give up a much higher share of the revenue from the ads clicked on by AOL users. At least that was Google’s best judgment; later on, even Brin would acknowledge that if that assessment had been overly optimistic, the $50 million guarantee would have bankrupted Google.
Ultimately, AOL became convinced that Google could make it more money than Overture and so gave the Googlers the contract.
Now Google had to handle a deluge. The difficulties didn’t come in the raw traffic; Google had been quietly
building its infrastructure for years and was confident it could handle more users. But Google had assured AOL that no ad would appear on its service that violated AOL’s standards. It was unprepared to implement that promise.
Google had already given some thought to the matter of ad approval. Originally, there had been a consensus that screening ads was a good idea. The lone dissenter was Larry Page, who believed that letting customers see their ads appear almost instantly would be intoxicating. Not to mention that skipping a labor-intensive step could pacify Google’s god of scale. The AdWords business team, who had actual experience selling ads in traditional media, worried that if you didn’t screen ads, users’ screens would be plagued by neo-Nazi and sex ads. But Page argued that if something distasteful showed up, Google could address it afterward. Fixing the small percentage of bad ads after the fact was much more efficient than building a bureaucracy to prevent any from appearing in the first place.
But AOL wanted a system guaranteed to ensure that no objectionable phrases would appear for even a second. “The only way to meet the policy was manually reviewing the ads,” says Sandberg. She was taken aback when Kordestani came by one day to say, “We’re going live at midnight—how many approved ads do we have?” The answer was none.