The DoJ sues Google: exploring the accusations

Last week, the US Department of Justice, along with a group of eight US states, sued Google, alleging that it exercised monopoly power to suppress competition in the markets of publisher ad servers, advertiser ad networks, and ad exchanges operating within the commercial realm of open web display advertising.

In this piece, I’ll attempt to clarify the 10 specific accusations of alleged misconduct with which the Department of Justice charges Google. The purpose of this article is not to editorialize the Department of Justice’s case or to offer prognostications as to how it will be resolved. What I aim to achieve with this piece is to elucidate and make plain the accusations against Google, deciphering the inscrutable advertising jargon of the case into language that can be generally interpreted.

In its complaint, the Department of Justice accuses Google of utilizing M&A transactions and various commercial tactics to prevent meaningful competition from taking root in these markets, thus allowing Google to charge supra-competitive rates on its services, which, the Department of Justice claims, harmed advertisers, publishers, and, ultimately, consumers. From the complaint:

Google abuses its monopoly power to disadvantage website publishers and advertisers who dare to use competing ad tech products in a search for higher quality, or lower cost, matches. Google uses its dominion over digital advertising technology to funnel more transactions to its own ad tech products where it extracts inflated fees to line its own pockets at the expense of the advertisers and publishers it purportedly serves…The harm is clear: website creators earn less, and advertisers pay more, than they would in a market where unfettered competitive pressure could discipline prices and lead to more innovative ad tech tools that would ultimately result in higher quality and lower cost transactions for market participants.

The complaint is lengthy at 155 pages, and it is dense, making for tedious reading. This is the fifth suit that the Department of Justice has filed against Google since October 2020. I covered the first of these suits, which focused primarily on Google’s search product and was generally seen as feeble, in a podcast here. The focus of the Department of Justice’s most recent suit is Google’s display advertising business — particularly, its control of an entire “ad tech stack” — spanning its demand aggregation tools (Google’s DSP, DV360, and its ad network, Google Ads), its ad exchange (AdX), and its publisher ad server (Doubleclick For Publishers/Google Ad Manager). The Department of Justice argues in its complaint that Google’s ownership of tools that facilitate the entire, end-to-end process of serving ads across the two-sided advertising marketplace allows it to employ anti-competitive tactics that prevent other firms from gaining traction with alternative point solutions.

The Department of Justice alleges that Google’s unique ability to aggregate demand for open web display ads was buttressed by its ability to bundle that function with demand for its proprietary ad inventory across Search, YouTube, the Google Play Store, etc. This unique access to demand allowed Google to lock publishers into its publisher ad server, and then to benefit its own ad exchange, AdX, with privileged access to publisher inventory through first-look and last-look rights and with opaque bid manipulation tactics through its demand aggregation tools. The Department of Justice asserts that these tactics allowed Google to charge supra-competitive rates with its ad exchange, AdX, given that demand was artificially and unfairly (that is: unbounded by commercial terms available to all market participants) routed through it.

The Department of Justice accuses Google of using monopoly power to suppress competition through these 10 points of conduct:

  1. Google’s acquisition of DoubleClick to obtain not only a dominant publisher ad server, DFP, but also a nascent ad exchange, AdX, in order to pursue its goal of dominance across the entire ad tech stack;
  2. Google’s restriction of Google Ad’s advertiser demand exclusively to AdX;
  3. Google’s restriction of effective real-time access to AdX exclusively to DFP;
  4. Google’s limitation of dynamic allocation bidding techniques exclusively to AdX;
  5. Google’s providing AdX with a “last look” auction advantage over rival exchanges;
  6. Google’s acquisition of AdMeld to stop its yield management technology from promoting multi-homing across ad exchanges;
  7. Google’s use of Project Bell, which lowered, without advertisers’ permission, bids to publishers who dared partner with Google’s competitors;
  8. Google’s deployment of sell-side Dynamic Revenue Share to manipulate auction bids—again, without publishers’ knowledge—to advantage AdX;
  9. Google’s use of Project Poirot to thwart the competitive threat of header bidding by secretly and artificially manipulating DV360’s advertiser bids on rival ad exchanges using header bidding in order to ensure transactions were won by Google’s AdX;
  10. Google’s veiled introduction of so-called Unified Pricing Rules that took away publishers’ power to transact with rival ad exchanges at certain prices.

I unpack each of these specific accusations below.

Google’s acquisition of DoubleClick

In 2008, Google acquired DoubleClick, a developer of advertising technology tools, for $3.1BN — nearly double what it had paid for YouTube one year earlier. DoubleClick’s publisher advertising tool, DoubleClick for Publishers (DFP, which was later renamed to Google Ad Manager, or GAM), allows publishers to sell their inventory to advertisers programmatically; the Department of Justice’s complaint refers to DFP throughout as a “publisher ad server.” Through the acquisition, Google also took possession of DoubleClick’s fledgling ad exchange, AdX. In 2010, Google acquired Invite Media (the Department of Justice’s complaint puts the acquisition price at $81MM), which operated a demand-side platform (DSP), and renamed it to DoubleClick Bid Manager (and later, Display & Video 360, or DV360).

Google outbid Microsoft for the acquisition of DoubleClick; Microsoft protested the acquisition, claiming that it would give Google too much control over the nascent digital advertising market. In December 2007, the Federal Trade Committee approved Google’s acquisition of DoubleClick in a 4-1 vote following an eight-month investigation; the European Commission similarly approved the acquisition in March 2008, and the transaction closed that day.

The Department of Justice’s complaint alleges that Google had, prior to the acquisition, attempted to launch a publisher ad server with minimal success and that DoubleClick’s market share stood at 60% when Google purchased it. It’s important to note that Google launched its first advertising product, Google AdWords, in 2000: AdWords originally provided for advertising to be inserted alongside search results. In acquiring DoubleClick, Google allowed its advertiser clients to place their ads on the “open web” — across websites that Google didn’t own. From the complaint (page 6):

The only problem with Google’s plan was that Google’s publisher ad server failed to gain traction in the industry. So, Google pivoted to acquiring the market-leading publisher ad server from an ad tech firm called DoubleClick. In early 2008, Google closed its acquisition of DoubleClick for over $3 billion. Through the transaction, Google acquired a publisher ad server (“DoubleClick for Publishers” or “DFP”), which had a 60% market share at the time. It also acquired a nascent ad exchange (“AdX”) through which digital advertising space could be auctioned. The DoubleClick acquisition vaulted Google into a commanding position over the tools publishers use to sell advertising opportunities, complementing Google’s existing tool for advertisers, Google Ads, and set the stage for Google’s later exclusionary conduct across the ad tech industry.

The complaint includes a helpful diagram that is utilized throughout:

In the diagram, Google Ads represents Google’s ad network, which offers a self-serve platform for creating ad campaigns that can operate across the inventory aggregated by Google: its owned-and-operated (O&O) inventory such as in YouTube and the Google Play store, as well as inventory available across the open web and in apps via its ad exchange, AdX. Google’s Demand Side Platform (DSP), DV360, allows for the same, but DSPs are designed for more sophisticated advertisers who want additional, granular control over the type of inventory they purchase, and on what commercial terms. DSPs often allow advertisers to upload data for targeting specific sets of people.

Google’s O&O advertising inventory at the time of the DoubleClick acquisition was mostly search-related, given that its acquisition of YouTube had only recently taken place. And growth in Google’s O&O advertising business has outpaced that of its open web display, “Google Network” business since Google’s acquisition of DoubleClick. In 2008, the percentage of advertising revenue that Google generated from Google Network websites was 32%.

By 2015, that percentage had decreased to 22%:

And in 2021, the last year for which full-year results are available, the percentage of total advertising revenues contributed by Google’s network was 15%:

A chart of Google Display Network as a percentage of total Google advertising revenue, from Q3 2004 — Google’s first as a public company — through Q3 2022 is presented below:

Google’s restriction of Google Ad’s advertiser demand exclusively to AdX

The Department of Justice’s complaint alleges that, after acquiring DoubleClick, Google restricted its demand — which, again, represents advertisers seeking to bid on both Google’s O&O inventory as well as inventory aggregated across the open web via DFP — from being routed through any ad exchange other than AdX, which it relaunched as “AdX 2.0” in 2009. In other words: if publishers wanted access to Google’s unique advertising demand, they were required to sell their inventory through AdX. From the complaint (page 37):

The DoubleClick acquisition was a first step in Google’s march to monopoly. After purchasing DoubleClick, Google restricted Google Ads’ purchasing of display inventory to sources controlled by Google (inventory owned by Google or by publishers using Google’s monetization products, including its newly acquired publisher ad server). The goal was ultimately to lock publishers into its ad exchange and publisher ad server, and block competing ad exchanges and publisher ad servers from accessing Google’s valuable pool of advertiser demand. Google implemented this restriction when it launched “AdX 2.0” on September 17, 2009. At the time, Google identified one of AdX’s two differentiators from other ad exchanges as unique “access to AdWords advertisers.”

The complaint points out that Google brought its considerable user data to bear in modeling bids on behalf of advertisers. According to the complaint, because Google was able to compile behavioral profiles of its users across their interactions with its proprietary products, as well as through Google’s contextual understanding of the open web gleaned through its website indexing activities related to Search, Google was uniquely positioned to pair advertisers with the most appropriate and valuable inventory. And, per the complaint, when Google did open AdX to demand from sources outside of Google Ads — which the complaint implies was precipitated by requests from the Google Ads team — Google withheld its targeting data from those exchanges, which disadvantaged those rival exchanges from conducting price discovery with the same degree of efficiency as could Google. From the complaint (Page 39):

The advertiser make-up and data advantages of Google Ads lead it to buy large swaths of inventory that otherwise would go unsold. Certain inventory is valuable only to advertisers that use Google Ads exclusively; other inventory is undervalued without the user targeting and contextual data that Google makes available only to Google Ads. Google does not simply limit access to this data to its own advertiser buying tools. It also has exercised its market power to undercut rivals’ ability to compete using the same or similar data. For example, after the DoubleClick acquisition, Google “hashed” (i.e., masked) the user identifiers that publishers previously were able to share with other ad technology providers to improve internet user identification and tracking, impeding their ability to identify the best matches between advertisers and publisher inventory in the same way that Google Ads can. Of course, any purported concern about user privacy was purely pretextual; Google was more than happy to exploit its users’ privacy when it furthered its own economic interests.

And Page 40:

Later internal studies confirmed Google Ads’ stranglehold. A 2014 Google experiment found that more than half of the impressions that publishers offered on its ad exchange would go unsold without the critical Google Ads’ demand. If the Google Ads’ demand was removed from the ad exchange, Google’s publishers would experience a 65% drop in revenue because no advertisers outside of Google Ads were interested in buying the unique impressions available or able to do so in light of the auction restrictions described below.

To be clear: the complaint proposes that some portion of these outsized outcomes for publishers were delivered at the expense of advertisers — that because Google’s demand was only routed through its own supply mechanisms, advertisers had access to less inventory than they would if Google Ads demand was routed through other exchanges. Internal communications at Google, according to the complaint, described this as a “buyside-subsidizes-sellside model.”

Google’s restriction of effective real-time access to AdX exclusively to DFP

The complaint alleges that Google required publishers to use its publisher ad server in order to receive real-time bids from its ad exchange, AdX. Publishers that sold impressions to AdX using another publisher ad server could only sell their impressions to AdX based on historical prices, which might not reflect the specific value of any given impression at the time it became available. According to the complaint, a product manager at Google wrote: “[O]ur goal should be all or nothing – use AdX as your SSP or don’t get access to our demand.”

The complaint presents this tactic as somewhat of a corollary to Google’s restriction of demand to AdX. The direction of exclusivity flowed from Google Ads (Demand) to AdX (SSP / exchange) to DFP/GAM (publisher ad server), but not in the other direction: publishers could sell their inventory to AdX using non-Google ad servers. The complaint asserts that, in order to lock publishers into the use of Google’s publisher technologies, Google made its demand less attractive when filled through other publisher ad servers by only making real-time bids available to its publisher ad server. From page 45 of the complaint:

Google knew that its Google Ads’ advertisers provided a unique, rich source of advertiser demand, and that no other publisher ad server (or ad exchange) could offer similar access to such a lucrative pool of advertiser demand. Many publishers could not afford to use a rival publisher ad server because they could not afford to lose the revenue that Google’s exclusively-linked platforms could provide. In essence, Google dictated publishers’ choice of each key ad tech tool used to sell their inventory: publishers must make their inventory available through Google’s publisher ad server and ad exchange to get the opportunity to sell a portion of it to Google’s extremely valuable Google Ads’ advertisers.

Google’s limitation of dynamic allocation bidding techniques exclusively to AdX

The Department of Justice’s complaint alleges that Google forced rival exchanges to compete for inventory in Google’s publisher ad server through a “waterfall” methodology. With a waterfall, a publisher ranks exchanges on the basis of their historical average bids. According to the complaint, when an impression became available to Google’s publisher ad server, it queried the exchanges for bids based on their priority position in the waterfall. The first exchange that responded to the bid request with a bid that exceeded the impression’s floor price, or minimum acceptable price, won the impression — even if other exchanges lower in the waterfall would have bid more for that impression. I discuss this waterfall methodology in more detail within the context of in-app mobile ads in In-app header bidding and the road to programmatic.

The complaint also alleges that Google gave its own ad server, AdX, a privileged position within this waterfall by allowing it a “first look” at inventory — meaning AdX was given the opportunity to bid on the impression before the waterfall was instantiated, regardless of where AdX would have ranked in the waterfall on the basis of historical paid CPMs. Google called this system “Dynamic Bidding.”

Further, the complaint claims that, before participating in dynamic bidding, Google provided advertisers with insight into what the average historical CPM was of the exchange highest in the waterfall, which effectively served as a price floor for the dynamic bidding mechanism. And because AdX operated a second-price auction, so long as one bid from AdX exceeded the highest average price from the exchanges in the waterfall, the price paid by AdX was essentially the highest average historical CPM from the waterfall. The diagram below depicts the Dynamic Allocation mechanism.

From the complaint (page 52):

In addition, through dynamic allocation, Google’s ad exchange had the opportunity to win impressions whenever it matched a rival’s average price. This permitted Google’s ad exchange (and its largest buyer, Google Ads) to win more impressions than its rivals, especially higher-value impressions. But for dynamic allocation, a rival ad exchange might have won the impression because it could offer a higher price or better match. Over time, this distortion of the auction process meant that advertisers were more likely to win the impressions they most wanted through Google’s ad exchange as compared to a rival ad exchange.

The diagram below depicts the interplay between the Google Ad Stack and the Non-Google Ad Stack, across Pooled Demand, the SSP / Exchange layer, and the Publisher Ad Server, as characterized by the Department of Justice’s complaint:

Google’s providing AdX with a “last look” auction advantage over rival exchanges

At one point, a practice known as header bidding arose, according to the complaint, as an industry solution to eliminating the inefficiencies of the waterfall-based bidding approach. Instead of offering access to inventory to exchanges through a ranked prioritization of their historical CPMs, header bidding allowed publishers to facilitate a real-time auction for each impression, wherein each participating source of demand could bid on an impression at the same time. Header bidding is implemented through Javascript included in a website’s header; a header bidding tool runs a unified auction before the publisher’s ad server is called. This article provides a helpful overview of the concept of header bidding, and I have also written about its application within mobile apps (for apps, the practice is called in-app bidding, since mobile apps have no “header”).

The Department of Justice’s complaint asserts that header bidding represented a serious threat to Google’s ability to route traffic exclusively through its ad tech stack. Since a header bidding tool conducts a unified auction across demand sources before invoking a publisher ad server (such as DFP/GAM), then the bid that wins that unified auction (which is first-price) represents the highest offer by competitive demand sources that Google’s demand must beat in order to win the impression. From the complaint (Page 73):

Despite these limitations, for the first time, Google’s ad exchange was forced to compete, at least in some fashion, against real-time bids from rival ad exchanges rather than against static, historical average prices from those ad exchanges. In assessing the impact of header bidding, a 2016 Google internal presentation noted “header bidding and header wrappers are BETTER than [Google’s platforms] for buyers and sellers.” Google explained that competition between AdX and buyers using header bidding increased publisher revenues by 30 to 40%, and provided additional transparency to advertisers. In essence, header bidding allowed publishers, advertisers, and Google’s rivals an opportunity to at least partially circumvent Google’s restrictions against real-time competition.

The complaint asserts that Google’s dynamic allocation mechanism conferred an advantage on Google’s ad server despite the unified auction taking place via header bidding because it essentially provided Google with “last look” privileges when Google’s publisher ad server was invoked (per the diagram below).

The complaint posits that, as opposed to Google’s “first look” advantage exercised when Google’s publisher ad server could run the waterfall after Google was allowed to submit a real-time bid, Google’s dynamic allocation methodology became a “last look” advantage when the unified auction was run before Google’s publisher ad server was invoked. From the complaint (Page 72/73):

The highest bid from the header bidding auction was then sent to the publisher’s ad server. Because of the way Google configured DFP, the winning bid from the header bidding auction was then sent to Google’s ad exchange to see if it could beat that price. Critically, through dynamic allocation, Google’s ad exchange always received this “last look” advantage, essentially a right to buy any impression as long as it had at least one advertiser willing to match the competing bid price from the header bidding auction.

Google’s acquisition of AdMeld to stop its yield management technology from promoting multi-homing across ad exchanges

In 2011, Google acquired a company called AdMeld, which operated a Supply Side Platform (SSP) that offered advertising yield management efficiencies to publishers, for roughly $400MM. AdMeld facilitated real-time bids for remnant (programmatically purchasable) inventory from a range of demand sources such that publishers could achieve CPM bids from multiple bidders that were priced for specific pieces of inventory, in real time. This article about the AdMeld acquisition provides helpful context around the market environment at the time. According to the complaint, AdMeld characterized its product as “the largest, independent practitioner of RTB [Real Time Bidding] behind Google…connect[ing] to more than 200 ad networks, & 35 Demands Side Platforms (DSP) and process[ing] more than 11 billion bids daily.” The Department of Justice investigated and ultimately approved Google’s acquisition of AdMeld.

In its complaint, the Department of Justice suggests that Google acquired AdMeld simply to shutter it and eliminate a competitive threat to its own ad tech platform. From the complaint (Page 67):

Shortly after the AdMeld deal closed, Google combined the yield management functionality of AdMeld into DFP and migrated all AdMeld customers to AdX. Critically, it then shut down AdMeld’s nascent real-time bidding technology, quashing a competitive threat that otherwise might have challenged Google’s market position and forced Google to move toward a more open system that allowed publishers to utilize AdMeld’s innovative technology to facilitate real-time competition among non-Google ad exchanges and advertisers.

Google’s use of Project Bell, which lowered bids to publishers who dared partner with Google’s competitors

The Department of Justice alleges that Google engaged in a series of schemes that allowed it to dynamically adjust the fee it applied to advertisers’ winning bids. Through these projects, as depicted by the Department of Justice’s complaint, Google utilized its broad set of historical data as well as its price discovery capabilities to render advertisers’ bids more competitive for highly-sought inventory through lower extracted margins. Note that advertisers’ bids are reduced by the platform margin, or “take rate,” before being passed to SSPs or exchanges; the less margin a platform takes from a bid, the higher the bid that eventually competes for inventory.

The complaint alleges that Google reduced its Google Ads margins — that is, the fee charged to advertisers for filling impressions on their behalf — in cases where doing so increased the probability that the bid would win an impression, but that it padded its margins in less competitive auctions to compensate. The complaint theorizes that Google’s dominant position in the digital advertising ecosystem, accommodated by its ownership of the entire value chain — from pooled supply via its publisher ad server (DFP/GAM), to its exchange (AdX), to its pooled demand, as well as its demand-side technologies, DV360 and Google Ads — allowed it to dynamically manage margin in a way that other participants couldn’t.

The first of these programs was Dynamic Revenue Share, which dynamically adjusted advertiser bids. Rather than taking a 14% fixed margin on each filled impression, Google used Dynamic Revenue Share to decrease its margin in cases where the advertiser’s bid was uncompetitive, giving the advertiser better odds at winning the impression. Dynamic Revenue Share also increased its margin where an advertiser’s bid was higher than necessary to win an auction. According to the complaint, through Dynamic Revenue Share, Google sought to achieve a 14% average margin applied to impressions filled for each publisher over time.

Google’s second project along this competitive vector, according to the complaint, was called Project Bernanke, which extended Dynamic Revenue Share by actively subsidizing bids. With Project Bernanke, Google increased advertisers’ bids above what they had submitted, thereby creating losses for Google in the cases where those bids won. To compensate for these losses, according to the complaint, Google applied higher margins to bids for impressions that Google’s advertisers were likely to win. From the complaint (Page 62):

Second, later in 2013, Google implemented Project Bernanke, which doubled down on Dynamic Revenue Share by subsidizing bids (i.e., bidding above the advertiser’s willingness to pay) on competitive impressions, thereby sacrificing any profit on the transaction. Of course, Google ensured that its own margins would be maintained. Google offset any loss on a given transaction by charging much higher fees (i.e., 50% or more) on impressions where Google Ads faced no competition—the majority of impressions Google Ads had already been winning. In doing so, Google Ads and AdX were able to win more impressions over their respective rivals, increasing Google Ads’ spend by 20% and profits by 30%, and increasing overall ad exchange revenue by 8%. A Google simulation of the program confirmed that advertisers using non-Google buying tools won fewer of the coveted high-value impressions, decreasing their relatively smaller spend on AdX by 14%.

Finally, Google modified Project Bernanke by adjusting bids such that the average margin achieved across the exchange settled at 14%. Google called this approach Project Global Bernanke; according to the complaint, this tactic systematically decreased the margin for publishers that saw a high degree of competition for their inventory and were thus at risk of leaving Google’s publisher ad server for a rival’s publisher ad server. The complaint alleges that Google subsidized this decreased margin by increasing its margin for publishers who saw less competition for their inventory and were thus unlikely to abandon Google’s publisher ad server. The goal of Project Global Bernanke, per the complaint, was to dynamically manage margin such that each publisher might generate different levels of margin for Google but that the margin delivered by AdX, broadly, would settle at some global target value. From the complaint (Page 62):

Finally, in 2014, Google implemented Project “Global Bernanke” which changed the method by which Google calculated the Google Ads’ take rate (sometimes referred to as “margin”). Instead of applying the same take rate to each publisher’s ad inventory, Google took an average take rate at the ad exchange level. Google took a higher cut of advertiser spend for some publishers while taking less for others. The effect was to further shift the publisher benefits of Google Ads’ two-bid system to the most important publishers and away from “noncompetitive” publishers (i.e., publishers whom Google believed were unlikely to risk switching to a rival ad server). Google candidly acknowledged that by 2014 it was not worried it might lose “non-competitive publishers.” As one document explained, it is “unlikely they can do better on another network (which doesn’t have any [Google Ads] demand).”

The diagram below, which is taken from the complaint but originated from an internal Google presentation, illustrates the margin adjustments implemented by Project Bernanke:

I’ve attempted to diagram the conceptual mechanics of Dynamic Revenue Share, Project Bernanke, and Project Global Bernanke to the best of my ability, based on the information available from the complaint, below:

Project Bell, as depicted by the complaint, functioned similarly to the above three tactics but worked in the opposite direction. Through Project Bell, Google decreased bids from its demand for impressions from publishers that bestowed first-look privileges on other, non-Google demand sources. From the complaint (Page 72):

Beginning in 2014, Google recalibrated the earlier Project Bernanke to decrease Google Ads’ bids on AdX for publishers that allowed rivals an opportunity to buy inventory ahead of AdX. Google called this “Project Bell.” Where publishers tried to partner with rival sources of advertising demand for “first look” access to inventory, Google reduced bids— without any input from or awareness of the underlying advertiser—by about 20%. According to Google’s documents, it explicitly warned publishers that utilizing innovative “first-call” technology from rivals would cause publisher yield to drop 20 to 30%. Of course, only a company like Google with substantial market power across the entire ad tech stack would have the incentive or ability to implement such a program. Project Bell both insulated Google’s ad exchange from this new form of competition and preserved preferential access for buyers on Google’s ad exchange, including Google Ads.

Google’s deployment of sell-side Dynamic Revenue Share to manipulate auction bids to advantage AdX

It’s important to note that Dynamic Revenue Share, Project Bernanke, and Project Global Bernanke all utilized bid adjustment techniques that impacted the margin paid by advertisers for their use of Google’s ad network demand tool, Google Ads. This article explains how Google’s demand-side fees are applied.

The Department of Justice’s complaint alleges that Google employed a similar program to adjust margins on the sell-side — through Google’s publisher ad server, DFP/GAM — in a program called Sell-Side Dynamic Revenue Share. Through this program, the complaint alleges that Google used its dynamic allocation mechanism to determine what bid its own demand had to beat from competitive demand sources in order to win the auction. The complaint proposes that Google used this information to dynamically reduce its sell-side fee, thereby effectively increasing bids for its own demand. From the complaint (Page 88/89):

Beginning in 2014, Google changed the way it applied AdX’s fee. Instead of taking a 20% cut on every individual transaction, Google allowed its take rate to fluctuate across transactions with the goal of averaging a 20% fee for each publisher over the course of the month, which continues today…Through dynamic allocation, buyers on Google’s ad exchange, including Google Ads, were able to see the highest rival bid before competing. After running its own internal auction, Google’s ad exchange compared its highest bid to the highest rival bid, which set the price floor of the Google ad exchange auction. When comparing bids, Google considered the “net bid” to the publisher, i.e., the amount the publisher would receive after all ad exchange fees were deducted. If Google’s ad exchange would have lost a transaction because Google’s ad exchange fee brought its net bid below the rival bid, Google could adjust its fee for that impression to win the transaction. Depending on the rival bid and the publisher at issue, Google could reduce its ad exchange fee to 0%, essentially boosting its ad exchange’s bid by 20%. If no rival ad exchange’s bid was competitive, Google’s ad exchange charged the full 20% fee, or more.

The complaint also alleges that, like Dynamic Revenue Share, Sell-Side Dynamic Revenue Share compensated for lower sell-side margin on some impressions by increasing its sell-side margin on other, less competitive impressions, and that, like Project Bernanke, Google allowed Sell-Side Dynamic Revenue to produce negative margin (eg. loss) by subsidizing some bids.

Google’s use of Project Poirot to secretly and artificially manipulate DV360’s advertiser bids on rival ad exchanges in order to ensure transactions were won by Google’s AdX

According to the Department of Justice’s complaint, Google devised a strategy, which it named Project Poirot, to stem the growth of header bidding by diverting demand from its DV360 DSP away from rival ad exchanges to AdX. The complaint alleges that Google engineered this by systematically decreasing bids on DV360 to any exchange that ran a first-price auction and therefore was likely utilizing header bidding. The complaint contends that Poirot and Google’s dynamic allocation practice worked in tandem: through dynamic allocation, Google was able to observe the bids made by rival exchanges (which had been reduced through Poirot) and to set the highest of those as the floor price for the dynamic allocation auction run later by AdX.

The complaint asserts that, in the 2016-2017 timeframe, DV360 claimed “nearly 30% share of gross digital advertising revenue,” and that, “as of early 2017, more than half of total advertising spend by DV360 advertisers flowed through rival ad exchanges.” In this timeframe, Google ran a second-price auction on AdX and, according to the complaint, actively sought to suppress the adoption of header bidding, which the complaint theorizes was a technological solution devised and adopted specifically to upset Google’s market power. Per the complaint, Poirot served two purposes: to route more of Google’s demand through AdX instead of rival exchanges and to undermine the benefits of header bidding. From the complaint (Page 94):

Project Poirot worked by systematically lowering all DV360 bids to rival ad exchanges that no longer employed second-price auctions—a proxy for identifying ad exchanges using header bidding. For each ad exchange, Google set a percentage by which it reduced all DV360 bids to that ad exchange. Initially, Google reduced advertiser bids by 10% to 40%; later Google reduced bids for some ad exchanges by as much as 90%. Because Google’s AdX did not participate in header bidding, none of DV360’s bids on Google’s ad exchange were decreased, even where DV360 bid on the same impression on both a rival ad exchange and Google’s ad exchange. This manipulation of advertiser bids virtually ensured that Google’s ad exchange would win the relevant auction by virtue of the deliberately decreased bids supplied to rival ad exchanges for the same impression.

And Page 95:

Through Project Poirot, Google used this power to lower DV360 advertisers’ bids on rival ad exchanges, and in turn, that ad exchange’s winning bid. Through dynamic allocation, the winning bid on the rival ad exchange—now lowered by Poirot—served as the price floor for Google’s ad exchange auction. DV360 could then win the same impression on Google’s ad exchange by matching that price. Working together, Poirot and dynamic allocation has led to reduced price competition for Google’s ad exchange and has ensured that more transactions flow to Google’s ad exchange, even if Google charges higher ad exchange fees.

I have attempted to capture the multi-tiered bidding process engineered with Poiret as described in the Department of Justice’s complaint below:

The complaint alleges that Project Piroit materially reduced the number of bids that were won by rival exchanges — which the complaint asserts was only possible as a result of Google’s “last look” privilege through its publisher ad server. From the complaint (Page 98/99):

As Google’s Director of Product Management for Display and Video Ads noted, Poirot’s initial implementation in 2017 was “quite effective, resulting in [DV360] spending 7% more on AdX and reducing spend on most other ad exchanges.” One employee on Google’s team explained that with Poirot, “spend on 3PEs [third-party ad exchanges] dropped by a whopping 32%.” Poirot shifted approximately $200 million of DV360 advertiser spend away from rival ad exchanges and toward Google’s…Google’s success with Poirot was possible because of Google’s ability to control the auction process run by its monopoly publisher ad server and Google’s last-look advantage stemming from its ad server’s dynamic allocation function. Google was able to lower DV360’s bids into rival ad exchanges without fear of losing impressions—even if DV360’s reduced bid lost in the rival ad exchange’s auction—because for the majority of impressions, it would get another bite at the apple when Google’s ad exchange was later called for a bid.

The complaint claims that, as a result of Poirot’s effectiveness in shifting advertiser spend away from competitive exchanges and into AdX, the company engaged in further bid reductions on DV360, “by as much as 90% to some ad exchanges,” through a project named “Poirot 2.0.”

Google’s veiled introduction of so-called Unified Pricing Rules that took away publishers’ power to transact with rival ad exchanges at certain prices

According to the complaint, Google realized that rival exchanges were winning auctions from DV360 demand — despite the influence of Poirot — because Google’s publisher ad server allowed publishers to set different price floors for different ad exchanges. Some of those publishers had set higher price floors for Google than for rival ad exchanges, which resulted in Google losing auctions in some cases when the bid it submitted was higher than a rival’s. The complaint rationalizes the logic of setting higher price floors for AdX than for other exchanges as follows (Page 103):

One Google Senior Product Manager explained to his colleagues: “The general idea is that the pub[lisher] doesn’t care about maximizing revenue on every individual query – they want to maximize revenue for their overall business, and that might mean sacrificing a few pennies of lost indirect revenue” on a single transaction. Similarly, Google’s Director of Global Partnership and Publisher Solutions explained: “Pub[lishers] are also rational[] when they decide to diversify their source of revenues” using floors given that “[i]t help[s] them to keep Google at bay and put pressure on us (similar to any industry).” By using different price floors, publishers expressed a willingness to occasionally accept a slightly lower price from a rival ad exchange than from Google’s ad exchange for the same inventory.

The complaint asserts that Google had inserted a clause into its publisher ad server contract that required publishers to allow AdX to compete with rival exchanges on “equal footing,” but that this clause was difficult to enforce in practice. To ameliorate the purported practice of publishers setting higher price floors for AdX than for rival exchanges, the complaint charges that Google simply removed the ability to set exchange-level price floors from its publisher ad server, presenting this change as a new feature called Unified Pricing Rules. An internal analysis of the auctions that AdX lost relative to exchanges equipped with header bidding, taken from the complaint, is presented below:

The complaint claims that Google packaged the rollout of Unified Pricing Rules with an announcement that its ad exchange, AdX, would abandon second-price auctions and run on a first-price auction, which was aligned with how its competitors operated. The complaint alleges that Google presented Unified Pricing Rules as a natural adaptation to a first-price auction environment while, behind closed doors, the company recognized that the primary motivation behind these changes was to route more traffic through its ad exchange. From the complaint (Page 107):

Internally, Google acknowledged that getting rid of higher price floors for Google’s ad exchange was the “primary internal objective for the entire launch” of bundled changes and its “key driver.” Internal Google documents explained that the changes “will be a shift in DV360 spend patterns away from [thirdparty ad exchanges].” Not surprisingly, internal Google documents identified the “winner” of the new rules to be AdX, its own ad exchange, and accurately listed rival ad exchanges to be the “losers” under the new rules.

The complaint claims that Google’s publisher clients were “livid” as a result of these changes, viewing them as a means of “preventing publishers from preferencing other ad exchanges and…refusing to allow rival ad exchanges to compete for transactions on any dimension other than per-impression price.” (Page 109) The complaint asserts that these changes resulted in Google’s ad exchange market share increasing by 6% in 2019, while “the average floor price faced by Google Ads’ advertisers on Google’s ad exchange dropped from a little over $3 to about $1.” (Page 109)

Photo by Mitchell Luo on Unsplash