Day 6: Google Begins Its Defense With an Ode To Its Benevolence

The intractable issue of “Final Auction Logic” suddenly starts to move, to everyone’s surprise.

Yesterday, the DOJ rested its case, though it will call two late witnesses tomorrow, and Google began its defense. The main topic was the Final Auction Logic of Google’s publisher software, but the trial day began on a different note, with a discussion of a Google motion to dismiss the morning’s first witness, Arnaud Créput, testifying by video from France. It was a quick conversation before the bench, and made mostly of references to Google’s complicated legal history in France.

DOJ opened the conversation by saying up front it did not intend to ask the witness about Google’s compliance (or not) with French court orders. This was a reference to the day three “recidivism” issue I wrote about last week: recall Judge Brinkema was chiding DOJ for the presumption of bad faith from Google that “permeated” its witnesses’ testimony. As US v Google (a website covering this trial) put it in their coverage that day, when Judge Brinkema asked if there is any history of Google being enjoined and then violating the injunction (i.e. recidivism), “DOJ jumps in to point to Google’s 2021 adtech antitrust fine in France, which they note that Arnaud Creput would be speaking to.”

This is the background to Google’s motion to dismiss Créput’s testimony. So, DOJ opened with a promise not to raise the issue, as Judge Brinkema immediately pointed out that without testimony from the French Competition Authority, it would be hearsay that Google had violated its order. At least, that was my understanding of the exchange. In any case, she denied the motion except with regard to hearsay, allowing Créput to testify but only about his own experience.

Arnaud Créput is CEO of Equativ, and ad tech server company. Despite the fiery introduction to his testimony—or perhaps because it constrained him—that testimony was pretty tame. He is another witness from the business community of Google’s competitors, basically the French equivalent of James Avery of Kevel, from whom we heard on day two. He reiterated the main points: the lack of transparency from Google, the need for divestiture of AdX (at least), the importance of preventing Google from replicating the tying by bidding directly from its demand side into DFP, etc. But we already know from Judge Brinkema last week that she considers this kind of testimony “window dressing,” and thinks the technical experts are the heart of the matter.

The mechanics of Créput’s testimony provided some moments of humor. He spoke pretty good, heavily accented English. So, he answered in English, but preferred to hear the questions in French, via a translator. The feed from overseas froze occasionally for two or three seconds at a time, which just enough time for interrupted answers to be incomprehensible, and for yes-or-no answers to disappear entirely, leaving all parties in silence as Créput thought he had answered while we had heard nothing.

My favorite moment came from the French tendency to adopt English words. When Créput was asked if his company’s ad server and exchange were in the same cloud, he replied that they were in separate instances. On being asked what he meant by the word “instances,” the translator asked him, in French, what he meant by the word “platform” using the English word “platform” as her translation into French of “instances.” The witness looked confused at being asked to explain an English word to the Americans. Okay, maybe you had to be there. But I wondered if Judge Brinkema, who told us last week that her French is decent, caught the humor.

After Créput, we heard the briefest of testimony from last Friday’s final witness, Goranka Bjedov, who had not officially completed her testimony. She was sworn in, Google passed on further cross-examination, DOJ asked a couple of quick questions on redirect, and then Julia Wood formally announced that plaintiffs rest their case.

Google Begins Its Case

Google’s first witness was Scott Sheffer, Google’s Vice President of Global Partnerships, Sell-Side Monetization. Most of Sheffer’s testimony was what we have come to expect from last week’s cross-examinations: migration of Google’s large systems will inevitably hurt publishers; the product will slow, its quality degrade; and so on. He spent some time on the Final Auction Logic. He considered DOJ’s definition of it excessively broad and vague, and argued that their definition contained all of DFP’s decision-making logic, not just “final.”

Sheffer described at some length the parts DOJ included in the Final Auction Logic. Ari Paparo shows DOJ’s whole complicated definition and briefly describes several terms in his day six coverage at The Market Report, but I’m going to turn to another part of Sheffer’s testimony. I’ll come back to the Final Auction Logic with the next witnesses, as it ended up dominating the day.

Sheffer described the future as he saw it of advertising technology. He described the opportunities artificial intelligence is bringing for integrating advertising into our consumption of digital content. An interactive avatar named Luzia was one example now in use in Europe. Another was a Washington Post test of technology that reads news stories by AI and integrates ads into the reading. Now, it is easy to be a little creeped out by Google’s ambitions, but I heard something else, too, in Sheffer’s testimony. Call it “enthusiasm” or even “wonder.” He seemed genuinely excited by the possibilities for the future he was seeing. It is easy to see the greed, the dishonesty, and the arrogance in Google. So, it is nice to see also a joy in technology and pride in high-level work. There are a lot of caveats, and we’ll get to those.

The next witness was George Levitte, Google’s Director of Product Management. The important thing here is that Levitte was one of the authors of the 2023-24 internal report I described on day four as contemplating Google doing “practically the whole DOJ structural remedy.” In testimony, he tried to argue that DOJ’s plans would take many years, that his project had only looked at much simpler plans, and that the “final auction logic” his project considered was only the very last step rather than the substantial decision logic DOJ included. His testimony was not effective. Once those reports were in the court records, that battle was lost. On cross-examination, Levitte’s team was quoted on time estimates similar to DOJ’s on step after step in the divestitures. Much as Levitte insisted those statements were just “brainstorming,” they implied that his team considered those estimates plausible and the whole divestiture feasible.

Even Google’s main argument against making the Final Auction Logic open-source—that it cannot be disentangled from all the substantive decision-making logic in their ad server DFP—seemed to backfire for Levitte. In making this argument, he said that if DOJ took all that they consider “Final Auction Logic” out of DFP, all that would be left (which DOJ calls the DFP Remainder) would be the front end and the reporting. This suddenly gave the issue a be-careful-what-you-wish-for character. Should Google succeed in convincing the court that all of the ad server decision logic is continuous with the final decision, it is within the court’s power—and, given testimony like this, it is also a plausible interpretation of DOJ’s remedy proposal—to transfer all the DFP decision logic to open-source; it may be far easier than the messy division of functions currently being envisioned, and less likely to cause disruptions, too. (Interestingly, Jay Friedman of Goodway Group suggested this on day two: all of DFP should be open-sourced.)

The final witness of the day was Glenn Berntson, Google’s Engineering Director for GAM. (Google Ad Manager—GAM—is what Google now calls the combined AdX and DFP.) Berntson had his fair share of seemingly disingenuous arguments, such as that the Google software engineer’s guide in which Weissman found those coding principles— simplicity, uniformity, flexibility, separation of concerns, and testability—was just for very junior coders. He said that it left out performance and scalability, as if senior coders then abandoned those basic principles (which are recognized throughout the industry and not just in some Google manual).

But something seemed to happen as Berntson’s testimony went on; he seemed to shift gradually from specious rationalizations to trying to truly explain what he thought Google was right about. It started with his admitting that DOJ’s divestiture proposals are technically feasible. He just thought it would slow Google’s processes and negatively affect publishers. He described what a sophisticated product GAM is, and, as with Sheffer, I began to hear pride in the accomplishment and excitement about what technology could do.

Berntson described GAM’s mission as meeting campaign goals for publishers. The essential logic is not an auction at all. Rather, GAM looks at the whole history of bids for each particular ad space, and treats them less as “bids” than as data about the demand for that space and how it fluctuates over time. Then GAM maximizes the revenue that space can command. To do this, GAM runs a daily analysis of all the data it has on all the ad spaces and their bid history, an analysis that takes ten hours every day running on 4,000 computers. Berntson is saying that GAM’s purpose is much bigger than auctioning ad space: GAM is managing the entire universe of advertisements and website ad spaces for the maximum benefit of all participants. So, yes, Google “manipulates” bids in a small, pedantic sense, but only for the greater good.

This Google ad tech trial, then, is an instance of a very old philosophical debate about benevolent dictatorship and democracy. And it is proceeding according to ancient pattern. Even if it stayed benevolent, dictatorship always has some vision of the good life and good future that not everyone is going to share. But the “staying” is always a bigger problem, and the benevolence starts to slip and cheat. Corruption sets in. There may be disagreement about how far it has gone, but Google’s conviction in the liability phase, along with its several other convictions, show the corruption has started. The divested products may well not be quite as terrific as Google’s current products, which a bunch of DOJ witnesses have acknowledged as excellent. But despite the lip service we pay it, democracy was never meant to be supremely competent and efficient; it is meant to be robust, fair, adaptable, free. A fair ad tech market would be a societal choice of competition over Google-managed welfare maximization, or the dream of it, anyway.

About this time in Berntson’s testimony, Judge Brinkema asked him if “Final Logic” wasn’t, then, a better label than “Final Auction Logic,” to which he said, “Yes.” When asked how Google could satisfy the demands for greater transparency, he suggested we get away from the auction idea and provide all the data to any publisher. Asked if that could be done, he acknowledged it probably could be, and that he would probably have to be involved. He would not enjoy that role, he said, as he thought it would worsen not only to GAM but also the open internet.

I promised you a cliff-hanger. On cross-examination, Michael Huppert asked Berntson again about the possibility of Google revealing all the data it had along with all of DFP’s decision logic (now called the “Final Logic”). He repeated that it could be done. Perhaps Huppert anticipated this moment from foreshadowing in the day four testimony of Google’s Tim Craycroft. Here is how The Monopoly Report saw it that day: “I don’t know if this is normal, but when Tim was asked about flaws in both proposals, he essentially started negotiating on the stand.” In any case, Huppert immediately and boldly asked Berntson if he was authorized to speak for Google, to which he replied, “I believe I am, yes.” Perhaps a bit sadly. Judge Brinkema stepped in then to say since this is an important topic, and it’s almost 6:00, we will continue this tomorrow morning at 9:30.