Day 6: Funny running into you in a bathroom in DC

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 6: Funny running into you in a bathroom in DC

YouTube CEO Neal Mohan testified today, focused on the early days of the Ad Exchange and on the AdMeld acquisition. I used to work closely with Neal but haven’t seen him in a couple of years, until today when we had to take a leak at the same time. Thus today’s newsletters’ title.

Here’s what happened.

Stuff we learned

• Prebig.org only exists because Google pressured the IAB Tech Lab to not adopt it (according to BOK)

• Google’s scale allows it to conduct A/B tests 15-30 times faster than other exchanges.

• Google did consider allowing other ad servers to work with dynamic allocation, but didn’t follow-through.

It’s all about me

Plaintiff’s witness 17: Neal Mohan, CEO of YouTube

Neal was the head of product and strategy for DoubleClick before the acquisition and led Google’s entire display effort until 2015, when he departed to YouTube. Neal’s testimony focused around three things: The display strategy overall (the “Three Pillars”), the company’s approach to “yield management” that eventually led to the AdMeld acquisition, and the effort required to build the modern AdX stack.

The Three Pillars

If you’ve been reading this newsletter you could probably guess what the pillars are, but if you need help they are, paraphrased from an email Neal wrote (CAPS are from original):

• The platform to ACCESS inventory.

• The Ad Exchange to AGGREGATE the inventory.

• The Google Content Network [GDN] to MONETIZE the inventory.

Not surprisingly, each of these pillars has now become an accusation from the plaintiff’s about the alleged monopolistic behavior.

Neal made sure to emphasize that at the time of the DoubleClick acquisition and the development of this strategy the market was quite competitive with Yahoo and Microsoft in particular jockeying for position with publishers. An email summed it up:

The threat of Yield Management

Before real time bidding, during the age of the waterfall, vendors like Pubmatic, Rubicon and AdMeld were not known as “exchanges” but rather as “yield managers.” This terminology can be confusing because the same term is used for the actions publishers take to optimize their yield. Nevertheless, these companies all made a name for themselves by optimizing indirect, non real-time spend, by changing the priorities within the waterfall.

Publishers were adopting yield management companies rapidly, since the tech resulted in increased revenue. And if there’s one thing to know about publishers it is that they will do almost anything for increased revenue. But yield management caused a bunch of internal conversations and disagreements inside Google.

Neal testified that he thought of yield management as a hopelessly outdated approach. He used the analogy on the stand that “RTB is like streaming music, while yield management is like a CD collection”. Jonathan Bellack more or less agreed:

But being dismissive of this technology wasn’t helping the fact that the yield management companies were pretty rapidly gaining share among publishers, so conversations turned more practical, about the threat and what to do about it:

In a presentation titled “YM Product Plan” it was noted that “YM’s are disintermediating our access to inventory, inhibiting our overall display strategy”. Which soon led to an email with this quote:

Obviously the word “parking” is a bit problematic. Neal insisted this meant keeping it going while we rebuild into our own stack. This conversation led eventually to “Project 17” wherein Google evaluated buying AdMeld or Pubmatic. Within the deal deck it was estimated that the price for either would be $300-$380 million plus retention. The deck laid out the rationale for the deal as follows:

1. Pressing need for publishers.

2. Deliver TM to the market quickly.

3. Fast growing business.

4. Reduce risk of disintermediation.

5. Team with experience.

Surprisingly, this leaves out the joy of having Ben Barokas on staff.

Only once in the long conversation about AdMeld and yield management were economics mentioned, which was a head-scratcher to me. Testimony indicated that AdMeld’s take rate was only 7%, compared to AdX’s 20%. With that in mind, it seems like the questions being asked by Neal and the Google team were either self-serving, or just not seeing the big picture. The Google team assumed that all indirect demand would move to “real time,” and thought optimizing waterfall demand was a dead end. But there was no way that indirect demand was going to agree to a 20% tariff for access to inventory — that was the real issue. This is an almost exact parallel to what happened five years later with header bidding vs open bidding, when Google tried to move the demand to its own channel but without the consideration of how the channel partners would feel about an additional margin shave.

Epilogue: AdMeld was acquired, it was shut down in late 2013, and you can find Ben Barokas on a yacht in Cannes.

The AdX build out

Part of Google defense is the “duty to deal” argument that Google didn’t have any obligation to work with other parties on their tech stack. During Google’s cross examination of Neal they put forth a bunch of evidence to support this argument, as well as the large effort it took to rebuild both DFP and AdX on the Google stack.

New to me were the documents that discussed actually giving some non-DFP ad servers access to dynamic allocation. In a May, 2013 email Neal and Scott Spencer discussed brining dynamic allocation to in-house ad servers. (Reading between the lines, I’m going to guess this was for one of the larger non-DFP publishers, like The New York Times, or CNET, both of whom eventually moved to DFP.) In the email were concerns about spam detection and inventory quality. The email further discussed potentially moving the integration to client side [oh heavens, what about the latency? Faints…].

Finally, Neal made a statement that he wasn’t aware of requests from other exchanges to integrate into DFP, or something to that effect, to which he was confronted with an email from Pubmatic co-founder Amar Goel asking for exactly that.

Overall, I think Neal’s testimony was very positive for Google’s case as it showed how much work went into making the display business what it is today.

Plaintiff’s witness 16: Brian O’Kelley (continued)

Brian O’Kelley, former CEO of AppNexus, gave video testimony continued from Friday. Not a lot new here, except testimony that in a closed door meeting at the IAB Tech Lab Google representatives “objected vehemently” to the takeover of Prebid.js and this led to the Prebid.org founding.

Plaintiff’s witness 18: Professor Gabriel Weintraub

Professor Weintraub is a professor at Stanford Business School and formerly the head economist for AppNexus. He offered expert testimony about the quantitative effects of Google’s various programs.

His high level findings were that Google’s conduct decreased rivals’ scale and their ability to compete. He further made arguments that there was a network effect whereby reduced scale hurt competitiveness since worse results for publishers would result in worst inventory, which would then result in worse results, etc.

Since there was an extremely robust cross examination, I’m going to present Professor Weintraub’s claims along with the rebuttal, and you can judge for yourself.

Scale effect

Weintraub: Because AdX has so much more volume they can do experiments faster to get statistical significance. If AdX needs to run an experiment to get 0.5% increase in win rates it will only take 1 day, whereas competitors can take 15 or 30 days.

Google rebuttal: These are theoretical results and not reflective of any specific study.

First look (dynamic allocation)

Weintraub: Running a simulation model, if Google had to compete directly with another source of demand before running a waterfall, their share would have only been 34.5% instead of 44.6%.

Google rebuttal: This assumes that Google would have to expend the effort to make this happen. It is just a theoretical.

Last look (dynamic allocation after header bidding)

Weintraub: Assuming each AdX win in last look would have been won by a header bidding participant, third-party exchange impressions were reduced 14.25% and revenue by 8.72%.

Google rebuttal: Using actual data* from June, 2023 these numbers are much lower, only 2.23% for impressions and 1.78% for spend.

*Note, when we say “Actual data” we referring to work Weintraub also calculated in his report

Sell-side dynamic pricing

Weintraub: -2.39% for impressions, -2.74% for spend.

Google rebuttal: Using actual data, only 0.34% for impressions and 0.53% for spend.

Uniform Pricing Rules (“UPR”)

Weintraub: -7.95% for impressions, -2.75% for spend.

Google rebuttal: Using actual data, only 1.18% for impressions and 0.53% for spend. Also, the studies these Weintraub numbers were done before the auction moved to first price, so are not valid.

Poirot (DV360 bid shading)

Weintraub: Version 1 of Poirot reduced monthly spend on third party exchanges by $59.5 million, while version two by $267.9 million.

Google rebuttal: The study being relied upon says “publishers consistently make more revenue”, that advertisers get a “surplus” of 11.48%, and on “clean” exchanges, revenue went up.

Finally, all of the data from Google, and thus from the expert witness, were global in nature. Whereas the judge seems most interested in US-only.

Plaintiff’s witness 19: Rosa Abrantes-Metz

Another expert witness, focused on economic theory. In the half hour I saw her testify no new news was found.

What’s next?

The next three days should open with Google witnesses. I believe we’ll hear from Scott Spencer, Sam Cox, and Jonathan Bellack. 🍿