Day 6: What's Up with the Open Sourced Final Auction?

This is the coverage of the remedies trial in US v. Google. I’ll be in Virginia writing every day on the courtroom activities.

Day 6: What's the Open Sourced “Final Auction Logic”?

Today was Google’s first day of witnesses and the testimony focused on poking holes in the DOJ’s remedies, especially the AdX spin-out and the open sourced final auction logic. I thought it would be worthwhile to spend a couple of minutes on the latter, since testimony about this remedy has a blind-man-describing-an-elephant feel to it and I’m not sure we’re all on the same page.

What is the DOJ asking for?

The DOJ wants Google to isolate what it calls the “Final Auction Logic” from DFP and open source it. Sounds easy. Here’s a description of how they define it (note all the details here):

The requirement would be to identify this code, find someone to administer it as an open source project, then hot swap the live code in DFP with this new open source code within 30 days of publication:

As you might guess, Google doesn’t like this

Google has been poking holes in the concept of “Final Auction Logic” throughout the first six days of the trial. They’ve argued that this phrase is not an industry standard, and is poorly defined. They’ve also argued that the definitions above are too broad and cover lots of stuff that isn’t in the process of selecting the winning bid.

Today’s witnesses, Director of Product Management George Levitte, and Director of Engineering Glenn Berntson went into some depth about the problems with the DOJ definition. In George’s testimony he went through the DOJ’s definition item by item and described why they were not part of the “Final Auction Logic”:

• Enhanced Dynamic Allocation sets bids based on delivery, it is not part of the bid selection (final auction) logic.

• Billable Event Rate Adjustments are also part of bid determination, and essentially estimate a conversion from performance CPAs to biddable CPMs.

• Reserve Price Optimization sets floor prices and is not part of bid selection.

• Multi-Slot Auctions are an ad server function, not part of the final auction.

You may note how often I’m putting “Final Auction Logic” in quotes. I’m doing to acknowledge Google’s argument that it isn’t really a defined term. While we’re having this argument in court about what should and should not be included, its worth pointing our that we also don’t even know what we’re trying to accomplish with this whole requirement!

I believe the point of open sourcing this logic is to prevent Google from manipulating the auction with “last look” or another hidden finger on the auction scale. But that’s really an audit requirement, not a licensing one.

Further, if you look at the bullets above describing what Google says should be excluded from the final auction, you’ll notice they are mostly about setting bids. This implies a very big difference in implementation. Google thinks they should be able to set bids, select eligible ads, and otherwise prep their server to decide a winner, and only then should the auditable/open portion of the stack take hold. The DOJ, in contrast, has thrown a bunch of stuff into their definition that contribute to the bid pricing, which dramatically increases the technical scope.

Imagine, for example, that Google gets its way and the open source component is limited to bid selection, not pricing. And big, bad, Google tells one of its SWEs that her OKR is to help increase the AdX win rate in DFP. She might devise a new feature in DFP, call it the Internal Combined Bid Multiplier or ICBM. This ICBM would, for the sake of stupid simplicity, just increase all bids from AdX by 20% before going to the open source part of the stack. Boom.

Of course, this ICBM example is so stupid that customers would be able to suss it out in log files almost instantly. And if you require all AdX bids to go through prebid, then manipulating pricing becomes more difficult. But you get my point.

What is the use case, exactly?

Open sourcing has multiple benefits. It allows the community to use the code, contribute to the code, and importantly, audit the code. These use cases are being smashed up in these remedies hearings without clear priority.

I think we can all agree that there’s merit to the audit use case. Given DFP’s importance to the ecosystem and Google’s previous illegal monopolistic behavior, this seems warranted.

But who exactly would contribute or use this Final Auction Logic?

To start, DFP itself would have to, as shown in the screenshot above. Practically, this means that DFP would have to rip out its real-time engine and replace it with a similar, but slightly different, version. This would be done with zero additional functionality, likely reduced performance, high risk, and all at 8.5 million QPS. Put me in the skeptical corner.

Technically, this is where the definitions of what it is, exactly, are critical. Because in a larger, optimized, high performance system like DFP the ways the different components interoperate and are orchestrated have enormous performance impacts. If the court found, for example, that the Enhanced Dynamic Allocation logic had to be included in the open source project, that would mean a vastly different shape of the API and scope of the underlying compute needed to run the system. These details matter.

The second use case would be for a competitive ad server to adopt or emulate the DFP logic to ease customer switching costs. This use case has been mentioned zero times in the trial, so maybe I’m the only one who thought of this? Also, I’m not convinced that having identical auction logic is on any buyers’ list of key reasons to switch.

The final use case is for publishers, or vendors serving those customers, to host the Final Auction Logic themselves. This seems to be the assumption of all the witnesses, and it naturally brings up a bunch of questions. Why? Who? Where? Latency?

Having built a custom bidder framework at Beeswax I have some experience in this area. It certainly is possible to make a network call-out for simple bids or data augmentation in 10ms or so. But I’m wracking my brain to think of a good use case for doing this in a publisher ad server. The whole point of a publisher ad server is to deeply control the priorities and targeting and data of your ad stack — what additional do you need to do in real time that’s worth it?

Further, this mythical use case is fodder for the Google defense. The witnesses keep talking about how “small publishers will have a burden to host the Final Auction Logic.” This is like saying your mom owning a Toyota Corolla is going to spend her days changing the break pads. It’s a use case that either a) doesn’t exist; or b) would instantly be outsourced.

Question: Who’s idea is this anyway?

One interaction today on the stand gave me pause as to why we’re talking about this in the first place. Regular readers may remember that last week we learned of secret Google projects to evaluate selling AdX and open sourcing the bidder. Today, this exchange took place:

Question: “Was this an idea from the team?”

George Levitte: No.

So Google was undertaking the analysis of open sourcing the ad serving auction logic even though the PM team that runs it wasn’t the originator of the idea. I think clearly this whole thing was started with a settlement demand from the EU. Then, the same idea, with additional fleshing out, found its way into the DOJ remedies. I wonder if there’s some technical person the regulators are relying upon who put this in their ear.

Tea: Who would host it?

This is not important at all, but you know I’m a basic bitch who likes tea. Throughout the trial witnesses have been asked who could possibly host and administer the open source project. Every witness gave the same two options: The IAB Tech Lab, or Prebid.org. When Michael Racic of Prebid was on the stand and was asked the same question he said Prebid was the only possible choice. Then when George from Google was asked the same question, he said it would have to be the IAB.

Note, as I describe in my book, Google has a long history of being against Prebid and was instrumental in assuring that the standard did not get adopted by the IAB. Google is also not currently a member of Prebid.