"I'm sure these are the three best models we could have," Eric said after they presented their scenarios. "And there's no way for me to know who's right. I'm not going to dig into the individual numbers, because I can't add any value there. So we're just going to take the middle one." It was a quick decision and strategically sound, but a gamble nonetheless.
AOL, meanwhile, nibbled around the edges, dragging out talks so that the cost of losing the deal would increase. "Oh, we just caught this and we need to change the language," Alan's counterpart would tell him, referring to wording that had been settled a dozen drafts earlier. Alan knew how to deal with middle managers trying to score points at the last minute to impress their bosses.
"Fine," he replied. "If you care that much about it, you call Dave Colburn tonight and have him call Eric Schmidt." Alan knew no one wanted to disturb Colburn for something insignificant and be lambasted as an idiot. Suddenly the language was no longer an issue.
As negotiations dragged on, Overture continued to jab at Google. The patent lawsuit was followed by a deal with Hewlett-Packard that made Overture the default "search the Internet" link on new HP Pavilion computers. If HP thought Overture was good enough for Internet search, who needed Google? And Overture pushed Yahoo to renew their short-term advertising contract for a longer period. Omid had lost out to Overture with Yahoo once before. Now he burned to bring that business to Google. He would excuse himself from the negotiating table at AOL to entreat Yahoo on his cell phone to switch to Google's AdWords Select product. The Yahoo contract with Overture would expire in June, so he knew time was short.
On April 24, 2002, Overture issued a press update on AOL. The existing agreement, which had been set to expire, was being extended through May 1, 2002.
The next day, Overture announced they had renewed their deal with Yahoo. Two months early. The term was not the five months of the original agreement, but three years. It was a stunning surprise, and it rocked our world. Rumor had it that Overture had taken Yahoo's execs golfing and, over the course of eighteen holes, scored a major victory. Larry and Sergey had been blindsided by Yahoo again. They took it well, considering. No one died that day.
Overture's Yahoo deal sent Wall Street's analysts into paroxysms of euphoria. Safa Rashtchy nailed the Google coffin shut. Investors had been waiting for the AOL deal and instead got the much more important Yahoo deal! Google clearly had no game. AOL was certain to sign with Overture now.
As May 1 approached, there was some nervous speculation about why AOL had not yet issued an announcement of a deal with Overture. "AOL must be raking Overture over the coals," read one online post.
On April 30, Omid's phone rang. It was Safa Rashtchy. Overture, he said, had announced they were not renewing their AOL contract. Did that mean Google had won it?
"Let me call you back," Omid replied. The deal still wasn't final.
Finally, Omid's fax started humming. Seventy pages chunked out, one text-heavy sheet after another. Miriam checked the document to make sure there had been no last-minute changes and then approved it. Joan signed it. It was done. But not over.
Overture's contract expired on Wednesday, May 1, 2002. AOL and Google were prepared to handle the transition throughout the day so that at midnight AdWords ads would begin filling the holes left by Overture's departure from AOL's pages.
But Overture threw one last wild punch. The contract would end on May 1, they agreed, but at exactly 12:01 a.m. Wednesday morning, not Wednesday night—twenty-three hours and fifty-nine minutes earlier than AOL had expected. "Overture were sore losers," Alan told me. "They said, 'No, no, no. Not at the end of Wednesday—at the beginning of Wednesday. We hate you guys. You're dead.'"
The Google team couldn't believe the depths of spiteful stupidity being displayed. Overture not only gave up a day's worth of revenue but pissed off AOL.
"You guys picked the right horse," Alan assured AOL. "We would never have done that to you." Overture's pettiness had AOL jumping through hoops to advance the schedule by a full day, just in the hope that Google would panic, drop the ball, and send AOL rushing back to their jilted partner. It didn't happen. Google's search results and syndicated ads launched at 12:01 a.m., May 1, 2002, and the world took note.
"The America Online pact now establishes Google as a major competitor in the paid-listings market, which Overture had dominated," wrote the Wall Street Journal. Overture's stock dropped thirty-six percent. Inktomi, whose search results Google had replaced as part of the same deal with AOL, dropped twenty-five percent.
"It represents an attempt to capitalize on Google's search brand," rationalized Overture CEO Ted Meisel. "paid listings just came with the package." AOL switched because of the weakness of Inktomi's search results, according to Meisel's spin. Overture was just collateral damage.
Eric Schmidt's choice of Susan's mid-range revenue estimate proved unduly conservative. By the end of the contract's first year, we were far above the highest projections. Part of that success may have been attributable to a small shift made by an enterprising engineer. The day after the deal went live, John Bauer added code that boldfaced the keyword a user had searched for when it appeared in an ad, making it obvious that the ad was relevant. That single improvement increased clickthrough rates by four hundred percent. One engineer. One change. Four hundred percent.
I had plenty to do in the run-up to AOL's switch to Google. All of our advertiser communications had to be revamped; our new, more stringent editorial policies had to be communicated; and numerous slides showing the power of Google and AOL combined had to be prepared. It was clear that marketing had a role to play, and I plugged in to do my part as well as I could. It wasn't very exciting. Sheryl's support team was closer to the advertisers, and Jonathan's new product-management team did the deep thinking about integrating Google with AOL. I and my group were, as Jonathan described us to his team, "the ad agency"—a service bureau for implementing others' strategic visions. There was pressure, though. Everything had to be ready for the launch, which coincided with deadlines for the GSA team and Google answers, a service that used live researchers to answer complex questions for a fee.
I had no complaints, but after two and a half years, my job was taking the shape of a more traditional communications manager in a mid-sized firm. You wouldn't know it watching from the outside, but things were settling into a groove in my part of the Googleplex. Come in, work out, eat breakfast, answer email, put out fires, eat lunch, clean up messes, eat dinner, answer email, go home, write copy, answer email, go to bed.
Sheryl Sandberg, on the other hand, was sitting on a volcano. Her AdWords support team of four people could not possibly handle the incoming barrage of work AOL required. Google ads now had to meet AOL's editorial policies, and AOL was not as willing to embrace risk as Larry Page. They wanted every ad checked by hand before it ran. We had a hundred thousand ads in our system, and every day they didn't run, we lost money we needed to pay off our AOL guarantee.
Sheryl had been planning to double her department to eight full-time staffers. Now when she did the math, she realized she needed forty-seven people—immediately. She would need fifty more soon after that. The day the deal was announced, Sheryl put out a plea for staff volunteers to work on approving ads twenty hours per week for at least two weeks. The engineers offered some resistance.
"Doesn't everyone working here already have a job to do?" asked one, before pointing out that expanding the human component of the system was a flaw. Scaling by adding staff instead of algorithms and hardware would be a mistake. Salar came to Sheryl's defense and pointed out that there was a long list of tools that needed to be created, but not every task could be automated. Besides, Larry and Sergey had endorsed the idea of staff volunteers. Few were persuaded.
With almost no volunteers, Sheryl moved on to plan B. She called an agency and requested they send over fifty temps. Two days after the contract went into effect, dozens of temporary workers sat at desks in the open area we had used for T
GIF meetings. Alana Karen from AdWords support stood at the front of the room and started walking them step by step through the process of approving an ad. Mass confusion ensued. The temps had no clue about online advertising, no familiarity with our approval process, and very little computer literacy. And the approval-tool software barely functioned. Sheryl watched with increasing frustration until she couldn't take it anymore. She marched upstairs to confront the half-dozen ads engineers.
"Come on," Sheryl said politely but firmly. "Come with me."
"Wait a minute," the group's manager protested. "You can't just take the entire engineering group with you somewhere."
"You come, too," Sheryl told him, using a tone she had developed as chief of staff to the U.S. secretary of the treasury. The engineers followed her downstairs and stood at the back of the training room as Alana tried to teach the temps to use the approval software they had written.
After a torturous hour, they stepped outside to talk to Sheryl. "The software doesn't work," they admitted. "We're going to rebuild it for you this weekend." Sheryl told the temps they would be paid for that morning and the next day, but not to come back until Monday.
On Monday all the temps came back, but they still weren't up to the task. Within two weeks, Sheryl had weeded out all but one of them.
On to plan C.
Sheryl cast her net at Stanford, which was stocked with recent graduates about to enter a dot-bombed local economy with few jobs. She promised them temporary positions that could convert to full-time if they worked hard and came up to speed quickly. The temp-to-hire program immediately took off, and dozens of history and sociology and philosophy majors unexpectedly had something to do the day after graduation. They would be AdWords reps.
Meanwhile, Eric Schmidt, the enormous AOL guarantee gnawing at his serenity like an ulcer, hovered about Sheryl's cube asking for updates. "So, how many advertisers do we have?" he'd ask. We needed to add tens of thousands to catch up to Overture.
"We're here," Sheryl would reply, pointing at a number on a spreadsheet.
"How many advertisers do we have now?" he'd ask a few hours later.
"Not many more than the last time you checked," Sheryl would say with the patience of a mom answering her child's query "Are we there yet?" We didn't pass Overture that first day, or even the first week, but it wasn't long before Google's ad network was as large as that of our biggest competitor.
More advertisers generated more ads, which required more AdWords reps to approve them. Sheryl's universe expanded from the AOL big bang until it filled half a building, and still it showed no signs of slowing. By October 2002, we had rented additional space in a facility as large as the one we already occupied. We officially called it the "Saladoplex" because it was on Salado Road, but everyone knew it as the "Honeyplex," because so many of the recent college grads Sheryl hired happened to be women. The outdoor patio lent itself to sunning on the warm California afternoons, and some AdWords staff members exhibited a predilection for clothing that facilitated tanning. Many a male engineer made the ten-minute walk from the Googleplex to enjoy lunch in the Saladoplex café.
Despite the sexist overtones, the Honeyplex name fit, as the building was a hive of nonstop revenue generation that paid for more machines, more engineers, and more AdWords reps to harvest the sweet bounty of our collective labors. The revenue engine driving Google's profits now hummed with a power equal to the search technology behind the growth of its user base.
Advertisers began finding us by word of mouth, so AdWords reps shifted from approving ads (a process that was increasingly automated) to persuading advertisers to spend more with us. Once they were in the system, Sheryl's organization—divided into tiers based on the budgets of the advertisers they served—assigned optimizers to improve performance. AdWords reps suggested more relevant keywords, set up multiple accounts for different product lines, and helped advertisers improve the sites linked to by their ads. Our AdWords reps didn't acquire customers, they just showed the ones we had how to make more money for themselves and for Google. By the time Sheryl left in 2008 (to become chief operating officer at Facebook), the AdWords group had grown from four people to four thousand, with most of the ad approvers stationed at offices in India and Dublin.
By winning AOL, Google had proved beyond a doubt the superiority of our ad-ranking technology and the value of maintaining a clear separation between ads and search results. So what happened next surprised everyone, including me.
We Walk the Line and Make a Misstep
"Is this new?" asked KeyMaster, a poster on the WebmasterWorld bulletin board. "Usually, ads are placed on the top of the search result page or boxed on the right side of the page. Now I see ads mixed in with the other search results."* KeyMaster was correct. For some users, on May 9, 2002, a Google search returned a paid listing embedded in our supposedly objective results. The paid listing was identified with the label "sponsored link," but it was the first time Google had displayed ads directly in line with regular results. It was essentially a form of paid placement, the exact practice Google had railed against so vehemently when it profited others.
By four a.m., a Google engineer using Matt Cutts' nom de plume, "GoogleGuy," had reassured the group: "This is a bug. Our new ad-distribution code had a flaw. It is mostly fixed already. It should be completely gone soon."
Not everyone was convinced. A poster named 4crests replied, "There is nothing like this anywhere else that it could have been confused with ... was it a bug or a test run? Or is it something that is going to be served up to another search engine soon?"
Several other posters accepted the idea that Google was testing an interface for AOL and somehow it had slipped out onto Google itself. They weren't far off, but it wasn't AOL Google was conducting experiments for. It was Yahoo.
While our primary battle at AOL had been against Overture to provide advertising, we had also unseated Inktomi to provide AOL's actual search results. Inktomi wasn't ready to cede us the entire search market, however. Before taking AOL from them we had taken Yahoo, and they intended to win Yahoo back when their contract came up for renewal in June.
To Sergey's deepening frustration, I had been unsuccessful in finding third-party metrics to prove that adding Google search helped grow a partner's traffic. Without data to refute Inktomi's claim that we were stealing Yahoo's users, we had to look for other ways to show our value. Perhaps we could provide unique new sources of revenue? Inktomi didn't offer ads.* Overture didn't offer objective search results. Google could deliver both in potentially lucrative ways. The Overture contract prohibited Yahoo from running Google's ads where Overture's ads appeared, but left an opening for our ads to appear as paid placements in the search results. Setting aside the irony, we planned to show Yahoo how that could work. We would test pay-for-placement AdWords on sites that used our "free-search" service.
We let free-search sites put a Google search box on their pages, but they couldn't run ads around the Google results. Only we could do that. We provided them with search service for free, and in exchange we kept the revenue generated by their users. Hardly controversial, but rolling out the new pay-for-placement test might be, so Larry and Sergey decided to show the ads to one percent of users for half an hour in the middle of the night—just long enough to gather some data on how much revenue they might generate.
We had made no commitment to actually implement such a system, and clearly there was no way in hell we'd ever actually do this on our own Google.com results pages. Marissa had a bad feeling, though, and expressed her concern that our test would be noticed and brought to the attention of the press. The only way that would be likely to happen was if the pay-for-placement ads ran on Google.com itself, which was not the plan. But things didn't go according to plan.
The paid-placement ads began showing up in Google's own search results. There was a bug in GWS, Google's web server, and it caused the ads to spread outside the limited test zone. Webmasters whose livelihoods increasingly depe
nded on reading the entrails of Google's ranking system couldn't miss such a significant change. We stuck by our explanation that a bug had unintentionally caused the test of a partner interface to appear on our own results pages.
Engineer Howard Gobioff later let me know that there were other instances when engineering was instructed to code ads into search results. I never saw them because, according to Howard, "someone always cared enough to make noise." He laid the blame at the feet of new PMs and business-development folks who argued that what we did on partner sites didn't matter and that, besides, the ads would still be marked as paid placements.
Howard said the engineers required to write the code buried some editorial commentary in its internal documentation. "This is evil but they made me do it," one engineer wrote. In Howard's opinion, the idea finally died because Sergey decided that it crossed into an ethically gray area and wouldn't play well in the press. We were willing to walk up to the edge of evil to get a closer look, but ultimately, Larry and Sergey were unwilling to cross certain lines. "Don't be evil" is not the same as "Don't consider, test, and evaluate evil."
Yippee! Yahoo!
On Friday, May 10, we celebrated the AOL deal with another company-wide luau catered by Charlie. Our executive team showed up in grass skirts. AOL sent over a giant blue lava lamp with their logo on it. Devin Ivester created a commemorative t-shirt with a traditional Hawaiian motif, but Larry made him change one element in the design: he insisted Devin remove the date. Larry didn't want anyone walking around wearing a reminder of when our AOL contract would be up for renewal.
Two months later, we took Ask Jeeves away from Overture. Overture CEO Ted Meisel remarked to the Associated Press, "We are still winning more deals than we are losing and I think we are winning all the right ones."* Now Overture had backhanded both AOL and Ask Jeeves. Not the way to make or keep friends.
I'm Feeling Lucky: The Confessions of Google Employee Number 59 Page 36