Day 13: The end is nigh!

The Monopoly Report is going daily during the trial. Our intrepid founder is in a windowless room without a phone or laptop to get you the good stuff.

Day 13: The end is nigh!

The biggest news today happened before any witnesses took the stand, when Judge Brinkema laid out the timeline for the rest of the trial. Spoiler alert, she wants to issue a final opinion by year-end! More below…

Stuff we learned

• There are 1,175 US customers that Google considers “large”, meaning more than $25 million/year in total spend.

• Hair BnB is a thing.

• The judge definitely doesn’t know what an LLM is

Corrections department

I’ve been slinging out these newsletters as fast as possible each day and as a result, there have been some errors (as well as many typos).

Yesterday I praised some insightful prose from Matt Stoller’s The Big Newsletter but it turned out the eloquent exploration of the defense’s Obama-era vibe was written by Tom Blakely, not Matt. Nice job, Tom.

Last week when Google’s former head of publisher products, Jonathan Bellack, testified, I quoted him as shown below.

This quote includes an error due to my poor handwriting. The correct quote is “I don’t think offering open RTB to everyone would end the efforts to play games with ad selection,” which changes the meaning to make it clear Mr. Bellack was referring to outside efforts, rather than potentially Google’s options. We apologize for the error.

The schedule

I am a huge Stan of Judge Brinkema. To get through this intensely complicated trial in three weeks is an amazing accomplishment. Here’s the amazingly tight schedule moving forward:

• Tomorrow: Google will rest its case tomorrow (Thursday). Their last remaining important witness is an expert who will testify about market definition.

• Friday: If the DOJ wants to rebut, they will start at 10AM on Friday. I can’t imagine it goes more than a day, in which case we’re done on Friday.

• November 4th: Both sides have until November 4th to submit their revised findings of fact. These are detailed written documents that summarize their respective cases.

• End of November: Oral closing arguments will be scheduled for either the last week of November or the first week of December. The judge specifically said she would likely have a draft of her findings by then.

• End of the year: The judge said “My hope is to have this resolved in December.”

Witness; Professor Judith Chevalier, Expert Witness, Yale

Professor Chevalier served as a rebuttal witness to the plaintiff’s expert witness Professor Simcoe. Professor Simcoe testified on days 8 and 9, and estimated that AdX’s 19.8% take rate should actually be 16.4% without monopolistic ties. Professor Chevalier spent a couple of hours testifying that Simcoe’s analysis was entirely wrong.

“Full Stack Revenue Share”

Professor Chevalier claims that looking at AdX revenue share is the wrong way to do things, and instead we should look at the total amount taken by the buy- and sell- side, which she referred to as the “Full Stack Revenue Share”. As an aside, before I get to the details, I just have to put in a meme:

OK, I got that out of my system. The witness claimed the following revenue shares were supported by the data:

Let’s examine this data in a couple of ways.

First, on cross-examination, the 36.1% rate for alternative paths was analyzed. This number was calculated as the weighted average of all DSP take rates, plus the weighted average of all SSP take rates, with weighting by volume. The DOJ pointed out that 7 out of the 8 highest take rate paths included a single DSP, “company 5”, that had very high take rates. While the identity of “company 5” was not revealed, the smart money says it is Criteo, which isn’t technically a DSP and takes very high, non-transparent amounts. So the 36.1% is significantly too high.

Second, the witness acknowledged that Google Ads → Other Exchanges was not a path that had available data to analyze. And, for the first time in this godforsaken trial, the DOJ mentioned that Ads’ take rates on non-AdX exchanges is much higher than on AdX.

Finally, if you assume that AdX’s take rate is 19.8% — a fact that has been attested to many times in this trial — then you can back out the supposed buy-side rates for the Ads and DV360 paths.

(1-DSP rate) * (1-SSP rate) = (1 - Full Stack Revenue Share)

Google Ads: (1-DSP rate) * 0.802 = 0.704

DV360: (1-DSP rate) * 0.802 = 0.731

This works out to an Ads take rate of 12.3% and a DV take rate of 8.8%. Check my math — now that I’m issuing corrections all bets are off. Anyway, I find the 12.3% implied Google Ads take rate difficult to swallow for two reasons. We have seen evidence in this case that GDN was experiencing 14-15% take rates on the buy side. And second, we know that the whole enterprise of sending Ads demand to AdX involved a huge subsidy to AdX so it could build share. In a sense the finding of a low take rate on this path is incriminating!

Finally, the Google-Google take rates may be subject to dynamic revenue shares, which is another alleged abuse by the company.

Criticisms of Simcoe’s methodology

The professor had a number of criticism’s of Simcoe. First, Simcoe used a “Comparables” approach where he used data from competitive exchanges to determine if AdX was overcharging. Chevalier’s counter points included:

• While AdX was above the average, so were six of the seven exchanges studied. The average was drawn down by one competitor with consistently low rates (likely AppNexus).

• AdX was never outside of one standard deviation of the mean of other exchanges.

• Simcoe used worldwide data, if you only use US the supposed over-charge is closer to 2% vs 3.4%

• If you only use the largest competitor (presumably Magnite) the over-charge is only 1.8%.

• If you only use the largest competitor’s highest annual take rate, the over-charge is only 1.2%.

Simcoe also studied log level data in an “Events” analysis. Chevalier also criticized this study, with the following points:

• UPR effects could not be separated from the effects of the first price auction change.

• Many publishers like UPR, so the assumption that all the UPR effects were an argument for a monopoly don’t hold weight.

• While impressions went up after UPR, revenue to Google was flat.

This last point was bostered with a chart showing the after the first-price/UPR change, volume of won auctions increased for both AdX and other exchanges. This supports an argument that all of this was just growing the pie.

In sum, Chevalier cast some doubt on the 16.4% take rate, but we are still left with a rate that is too high at some level.

Witness: Courtney Caldwell, CEO ShearShare

Courtney was a charismatic witness attesting to how much she loves Google. The main thrust of her testimony was that when she uses Google Ads she doesn’t care about ad formats necessarily, and that she also uses other platforms like TikTok and Meta.

Her business, ShearShare, allows stylists and beauty professionals to do short-term rentals of facilities, which has given her firm the nickname “Hair BnB.”

Witness: Adam Stewart, VP Google

Adam runs a big portion of the US sales effort at Google. There was not a lot of substance to his testimony, with the key points sort of repetitive with previous testimony, including:

• Google is very competitive with Facebook, Amazon and Criteo. The witness later claimed he was not familiar with Criteo’s business.

• 10% of GDN customers are 70% of spend, and separately 70% of spend is from customers who dual platform GDN and Google Ads.

• The “Demand Gen” campaign type in Google Ads was built to compete with social.

There was one interesting document shown in this testimony, an internal competitive review versus The Trade Desk. In this doc, the amount of Google revenue at risk versus the Trade Desk was estimated and aligned to the competitive reasons that spend might move. The reasons were:

1. Identity [2x more than #2]

2. CTV [3x more than #3]

3. ID-based targeting

4. Less restrictive targeting policies (gambling, etc)

5. Service

The document also pointed out that DV360 has a price advantage since it includes data, fraud, and brand safety fees while these are extra on TTD.

Witness: Alejandro Borgia, Google

Alejandro is the Director of Product for ad safety. He follows yesterday’s witness, Per Bjorke who works in a similar, but totally different position. Alejandro’s testimony was uneventful, though the cross was pretty aggressive.

On cross, the DOJ showed documents indicating that of 313 ad quality policies, only about 10% were subject to “active management”, meaning they had dedicated models and metrics, while the rest were enforced using more generalized approaches. There was also a request for funding titled a “step-change investment” that would add 148 additional policies to be actively managed. Out of 28 privacy policies, for example, only one was actively managed.

The witness strongly disagreed with this characterization, saying that active management was only one way to assure compliance, and that with the growth of LLMs and more restrictive onboarding policies they have made their ecosystem safer despite what the document said.

We also learned that while we all love Judge Brinkema, she does not know what an LLM is.

What’s next?

Tomorrow should be Google’s last day and will focus on market definition.