Chapter 4: Eureka

The dawn of programmatic

The Ad Exchange

A few days after we returned from Israel, Elad Efraim called me and said “we’re in!” While we negotiated a contract with Oridian to migrate their entire business to Yield Manager, we started a frenzied effort to make our prototype into a production system. Matt Philips found a European datacenter so we could handle Oridian’s traffic and bought $100,000 of servers on our personal credit cards. Ed Kozek and his team accelerated efforts to build a user interface where ad networks could set up their campaigns and publishers. I spent my days and evenings furiously writing code, waiting for our quality assurance team in Moscow to wake up so I could tell them what I needed them to work on while I slept.

In March 2005 we started moving traffic from Oridian’s ad server to ours, and the bug reports started coming in. Not only was the Yield Manager platform new and inherently buggy, the Oridian business was multinational and much more complex than Right Media’s network. Many use cases only existed in certain markets or certain types of publishers.

To deal with the onslaught of problems, I would wake up in the morning in my West Village apartment, walk to City Bakery on 18th street to get a platter of pastries, and head to the Oridian office on 17th. The Oridian US team would be in a conference room with the Israel team on the phone and they would proceed to list out all of the problems we needed to fix, only slightly mollified by the croissants and muffins. After an hour absorbing their frustration, I would walk to the Right Media office on 26th and hand out post-it notes to each of the team members with all of the bugs and features that we needed to address. We were completely focused on making Oridian successful as we approached our deadline for a full cutover.

On April 1 we launched four networks on the Yield Manager platform: the two that had tested with us the prior August, Right Media, and Oridian. The first few hours were a total disaster as a reporting bug made all of the numbers completely wrong for all of the participants. Servers were crashing, causing outages across the thousands of websites with ads from the four networks. Our phones were ringing wildly as frustrated clients and publishers threatened to remove us from their pages. It was chaos.

By the next week, things were calming down. The servers stayed up, reporting was working properly, and clients were back to filing bug reports and feature requests.

Eureka

I began analyzing the reporting, trying to see the impact of moving the Right Media network to the new platform. There were three metrics that mattered. First, were our advertisers getting good results? Second, were our publishers getting paid well? Third, were we making money? Usually these metrics were in conflict: for a network to make money, either the publisher gets paid less or advertisers pay more.

What I saw was so shocking that I thought there was another reporting bug. Advertisers were paying less for clicks and sales than before. Publishers were getting paid more. And we were making more money as a network. How was this possible?

I called Elad and asked him how Oridian’s numbers looked compared to before the flip. “It looks fantastic! We are making more money and performance metrics are solid.”

I walked into Mike Walrath’s office. I knew he had been smashing F5 for days, watching our numbers obsessively. “Are you seeing what I’m seeing?”

Mike grinned. “Yes.” We stared at each other. What was going on? We started looking at reporting for one web site to try to isolate the situation. Before the flip, this publisher was running ads for 20 of our advertisers, driving decent performance. After the flip, this publisher was running ads for 50 advertisers, including 30 from our partner ad networks.

We knew that more ads meant more choices for the algorithm to match to users. We also knew that more ads meant that users would see the same ad less frequently, a strong corollary to performance. And we knew that more bids in an auction will generally increase the final clearing price.

We also looked at revenue from different countries, and here we saw a huge win. Twenty percent of traffic to this website came from Europe. Right Media had a couple of global campaigns which performed very poorly overseas. But with Oridian’s large base of advertisers in Europe, publishers were seeing massive increases in revenue for their cross-border traffic.

These publisher revenue increases made sense: even though our partners would take a share of the revenue from their ads running on our network, a larger and more diverse set of advertisers would generally drive up publisher revenue.

Next, we looked at an advertiser. Before the flip, there were 30 websites that drove the majority of their clicks and sales. After the flip, there were 60 sites driving performance. Instead of buying just Right Media’s inventory, campaigns could bid on the aggregated inventory across four networks, including the cross-border traffic from Oridian’s publishers. Having more inventory to choose from meant the bidding algorithm could be more selective. Also, since there was less competition on the cross-border traffic, prices were generally lower than domestic websites.

Advertiser performance increases made sense too. More diverse inventory to choose from, even with the partner network taking a share of the bid, meant advertisers could be more selective and get more clicks and sales for their money.

I went back to my office and stared at my screen. Advertisers were seeing better performance. Publishers were making more money. Networks were making more money. I thought about what I knew of economics, of financial markets, of auction theory. We were connecting pools of liquidity into a unified market. It was textbook. It was obvious. And yet… nobody had done it before.

As it sunk in, I whispered “holy shit” then yelled HOLY FUCKING SHIT and ran out of my office. Pandora’s box was open. And for that one moment, before all of the horrible consequences, before we named it the ad exchange, before programmatic and privacy and fraud and politics and betrayal and getting fired and divorced, before we cracked the code on monetizing social media and opened the box the whole damn way, for just that moment: it was glorious.