Part 27 - Google Launches Project Poirot to Manipulate Its Advertisers’ Spend to “Dry Out”

IV. GOOGLE’S SCHEME TO DOMINATE THE AD TECH STACK

D. Google Responds to the Threat of Header Bidding by Further Excluding Rivals and Reinforcing Its Dominance

2. Google Blunts Header Bidding By “Drying Out” the Competition

d) Google Launches Project Poirot to Manipulate Its Advertisers’ Spend to “Dry Out” and Deny Scale to Rival Ad Exchanges That Use Header Bidding

208. By the fall of 2016, Google worried that it needed to take additional steps to stem the competitive threat from header bidding. As such, Google considered options “for mitigating [the] growth of header bidding infrastructure.”

209. The first place Google turned was the substantial volume of advertiser spend that flowed through its large advertiser demand side platform, Display and Video 360 (“DV360”). By 2016-17, Google’s DV360 was already one of the largest demand side platforms in the market with a nearly 30% share of gross digital advertising revenue flowing through demand side platforms. Google was familiar with wielding the power stemming from the advertiser spend locked into Google Ads, but it had previously allowed the larger, more sophisticated advertisers and ad agencies on DV360 greater control over where and how they bought advertising inventory. As a result, these advertisers frequently transacted on rival ad exchanges that provided lower-cost or more valuable inventory. As of early 2017, more than half of total advertising spend by DV360 advertisers flowed through rival ad exchanges.

210. Google became increasingly concerned that spend from DV360 advertisers was driving header bidding’s growth and helping rival ad exchanges compete. As one of Google’s product management leaders for publisher ad products reported to his colleagues, “I think you know this, but I am told regularly that [DV360] is the top buyer on every other ad exchange, so a huge chunk of publisher HB [header bidding] revenue is Google demand going outside our ecosystem and then coming back via 3PE/HB [third-party ad exchanges/header bidding].” DV360 advertisers often represented the largest buyers on rival ad exchanges engaged in header bidding.

211. If header bidding was left unchecked, Google feared its own advertisers’ spend would continue to shift to rival ad exchanges and thereby allow those rival ad exchanges to gain the scale and network effects needed to become serious competitors. Google could not allow that to happen.

212. In October 2016, Google employees responsible for DV360 reached an “overall consensus” that Google did not “want to compete on [header bidding] queries.” To that end, Google engaged in a series of projects designed to reduce the flow of DV360 advertiser spend to rival ad exchanges engaged in header bidding and redirect that spend back to Google’s ad exchange, regardless of the cost to publishers and advertisers. Hence, even though DV360 was supposed to be a tool that enabled advertisers to bid most effectively for the inventory they wanted, in Google’s hands, DV360 was weaponized to stifle competition. Not only was this conduct against the best interests of Google’s own advertisers, it was against the best interests of DV360 itself because it harmed the very quality and profitability of the tool Google had created and promoted to customers.

213. As Google’s Managing Director for Global Publisher Solutions and Innovation explained in response to news that a competing ad exchange had lowered its fees, the overarching goal for Google was not for DV360 advertisers to benefit from reduced fees on other exchanges but for advertiser spend flowing through Google’s tools to “only buy on AdX impressions that are [available both] through AdX and multiple” ad exchanges in order to “dry out HB [header bidding]” ad exchanges.

214. Google’s initial proposed strategy was to go all in—simply block DV360 advertisers from buying on rival ad exchanges any inventory offered by publishers Google believed used header bidding. In late 2016, Google experimented with creating a “white list” of publishers that did not use header bidding, retaliating against publishers that dared embrace innovative technology that could improve publisher revenue. Google’s DV360 advertiser tool would permit advertisers to submit bids on rival ad exchanges only for inventory from such white-listed publishers; it refused to allow its advertisers to submit bids on rival ad exchanges if Google suspected the relevant publisher used header bidding. The intended effect was obvious: “move a lot of rev[enue] to AdX and put pressure on HB [header bidding] infra[structure].” Unfortunately for Google, experiments testing such an extreme strategy showed that the approach would not only harm advertisers and publishers but also Google. The experiments predicted Google’s DV360 buying tool would lose approximately 30% of both impressions and revenue.

215. At a meeting discussing the experiments’ results, however, a Google Product Manager suggested an alternative solution to surgically target the threat of header bidding on rival ad exchanges while minimizing losses to Google: “instead of stop bidding on HB [header bidding] queries, we could bid lower on HB queries.” When combined with the various advantages Google had afforded its ad exchange within Google’s publisher ad server, this approach potentially could achieve all of Google’s goals: inhibit advertisers from transacting through rival ad exchanges engaged in header bidding while allowing Google to redirect revenue and transactions back to Google’s ad exchange, where it could charge supra-competitive fees.

216. Google’s product and engineering teams quickly turned to implementing this plan of attack under the code name Project Poirot. [21] The purpose of the project was straightforward: Google would shift transactions away from ad exchanges using header bidding and to Google’s AdX by artificially manipulating the bids sent to rival ad exchanges so that Google’s AdX could win those transactions more often (even if that meant harming Google’s own advertisers). Or, as Google put it: “for HB [header bidding] we should win back more on AdX.” By July 2017, Google changed the settings of DV360 so that by default all advertising campaigns were opted into Project Poirot; only 1% opted out.

217. 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.

218. In important ways, the Project Poirot advertiser bid manipulation scheme was both more insidious and more profitable for Google than Google’s initial proposal of not bidding at all into rival ad exchanges using header bidding. First, by allowing its DV360 advertisers to bid on all inventory on rival exchanges, albeit at substantially lower levels that won fewer transactions, advertisers could still buy the limited set of inventory that publishers chose not to make available at all to Google’s ad exchange, as well as inventory that only appeared particularly valuable in light of targeting data available through a rival ad exchange. However, reduced bids designed to win fewer transactions went against the interests of DV360 advertisers and DV360 itself as a platform. Both would have benefited from winning more inventory on rival ad exchanges at prices advertisers were willing to pay, but for Google surreptitiously lowering their bids.

219. Second, for inventory available to DV360 on both a rival ad exchange and Google’s ad exchange, Poirot and dynamic allocation worked together to ensure that Google’s ad exchange often won the transaction. To do this, Google used its control over the bids of advertisers using its DV360 product. While these large, sophisticated advertisers set general parameters for bidding, especially after the launch of Poirot, Google alone determined the particular bid made on behalf of the advertiser on each of the millions or billions of pieces of inventory an advertiser bid on within each ad exchange; Google opted advertisers into the program while affording them no meaningful visibility into this level of bidding. 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. Google has only been able to implement this scheme by virtue of its control throughout the ad tech stack: advertisers and ad agencies using DV360, the AdX ad exchange, and the publisher ad server.

220. A hypothetical illustrates Project Poirot in action. An advertiser using DV360 might configure an ad campaign to pay a maximum price (for example, $1 CPM) for a particular type of advertising impression. Under Poirot, Google would lower that maximum bid when bidding on rival ad exchanges that used header bidding by applying an ad exchange-level multiplier, for example, bidding $0.38, $0.42, and $0.40 on three rival ad exchanges. In this example, the $0.42 bid is the highest and wins the header bidding auction; that bid is then passed to the publisher’s ad server after the ad exchange deducts its revenue share fee (assumed here to be 15%, reducing the net bid to $0.36). Next, the ad server sends that price along with a request for a bid to Google’s ad exchange via dynamic allocation. The $0.36 serves as a price floor and is shared with bidders on the ad exchange. On behalf of the same advertiser as before, DV360 now bids the advertiser’s maximum bid ($1) and wins the impression. Because Google’s ad exchange ran a second-price auction, however, the publisher receives only the floor price, $0.36. Google charges the advertiser this price plus the applicable ad exchange revenue share (20%), which translates to $0.45. [22] Ultimately, this means the advertiser pays more for the impression than it would have paid bidding via a rival ad exchange, Google is able to profit by extracting its revenue share fee at the ad exchange level, and the rival ad exchange that otherwise would have won (because it would have charged a smaller revenue share fee than Google’s ad exchange from the same DV360 bid) is denied the transaction.

221. Google’s prelaunch experiments found that Project Poirot would reduce publisher display revenue from DV360 by over 10%. By contrast, the predicted “surplus”—additional revenue shared between Google and advertisers—was only about 1%. The total number of impressions DV360 purchased would drop by almost 5%. In particular, Google recognized that some advertisers would no longer be able to buy certain impressions only offered on rival ad exchanges, because even if those advertisers were willing to pay the minimum price set by the publisher, Poirot reduced their bids below the minimum. Poirot prevented these advertisers from spending their full advertising budgets and resulted in some ad inventory going unfilled—a loss for the display advertising market as a whole.

222. Project Poirot formally launched in July 2017 under the name “Optimized Fixed CPM Bidding.” Google did not afford its advertisers a meaningful opportunity to choose whether Google could systematically lower their bids on rival ad exchanges. Instead, Google imposed these changes while providing virtually no information to its advertisers on the nature or extent of the program. Over 99% of advertising campaigns were subject to Poirot: all “automated bidding” campaigns on DV360 incorporated Poirot automatically, and Google opted in by default all “fixed CPM bidding” campaigns.

223. Internally, Google sometimes justified Poirot as benefiting advertisers, but such justifications were pretextual. Google designed Poirot to lower DV360’s bids into third-party ad exchanges ostensibly to account for “non-second-price auctions,” such as where a publisher uses soft floors that are set above the second-highest bid or uses floors that change based on bidding history. But Poirot did not apply the same rules to Google’s ad exchange that it applied to its competitors. Poirot did not adjust DV360’s bids into AdX even though AdX was not a true second-price auction: AdX itself used features such as Reserve Price Optimization (“RPO”) that—like those used in purported non-second-price auctions—similarly adjusted floors based on what bids Google expected to receive. As one document explained, RPO went live in 2015 (before Poirot) and “move[d] Google away from a 2nd price auction.” Google employees discussed the interaction between RPO and Poirot and noted that RPO was designed to go undetected by Poirot, and they concluded that if AdX continued to develop more aggressive versions of RPO then DV360 would adjust Poirot to avoid detecting this auction dynamic within AdX. Poirot intentionally did not target all non-second-price auctions; it gave the auctions Google conducted a pass.

224. 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. This spend was subjected to Google’s 20% ad exchange revenue share fee—one of the highest in the industry—resulting in an additional $40 million in profit for Google.

225. This substantial shift in advertising spend by the largest demand side platform in the market, combined with the systematic drop in bid price, had real consequences for competition between ad exchanges. DV360’s lowered bids reduced the competitiveness of header bidding auctions, which in turn lowered the win rates of ad exchanges relying on header bidding. The win rate on Google’s ad exchange increased even though Google had made no improvements to its ad exchange, offered no additional benefits to publishers, and reduced advertisers’ reach (without their knowledge). Because publishers consider win rates before investing the time and resources necessary to integrate with ad exchanges, Poirot’s hit to rival ad exchanges made it even less likely publishers would continue to offer inventory to these ad exchanges and less likely that new publishers would integrate with these ad exchanges through header bidding.

226. 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. And because Poirot generally reduced the clearing price of header bidding auctions, Google’s ad exchange could win the impression at a lower price, thereby decreasing the revenues ultimately paid to publishers. Without the systemic advantages that Google’s publisher ad server forced publishers to afford to Google’s ad exchange, and without Google’s dominant position among advertiser buying tools, Poirot could not have been nearly as successful in halting the potential rise of rival ad exchanges.

227. Seeing Poirot’s initial success, Google doubled down on the strategy. Google was unconcerned about any potential blowback from publishers (who had no meaningful alternative to DFP) or advertisers (who had no insight into the changes). In September 2018, Google launched “Poirot 2.0.” Under this new version of the program, Google reduced DV360 advertiser bids further, by as much as 90% to some ad exchanges.

228. Google’s prelaunch experiments indicated that Poirot 2.0 would significantly affect rival ad exchanges, even more than the initial iteration. Google anticipated Poirot 2.0 would further decrease DV360 advertiser spend on rival ad exchanges by another 20%, lower publisher payouts from auctions DV360 won by 20%, and lower rival ad exchanges’ win rates by 10%. The estimated impact of Poirot 2.0 on the most vocal proponents of header bidding technology was even more pronounced: AppNexus/Xandr would lose 31% of DV360 advertiser spend, Rubicon would lose 22%, OpenX would lose 42%, and Pubmatic would lose 26%. For Google’s ad exchange, on the other hand, Poirot increased revenue, publisher payouts, and win rates, through a sleight of hand forcing a shift in advertiser spend. As Google’s Director of Engineering—and chief architect of Poirot—explained to colleagues, Poirot had largely achieved in practice Google’s earlier plan to boycott header bidding auctions: “lowering bids may have a similar effect” to stopping all bidding on rival ad exchanges. Poirot 2.0 also accomplished Google’s strategic goal to “dry out” header bidding without the need for Google to take the significant hit to its revenues and profits that initial experiments suggested might be necessary.

229. Rival ad exchanges lost significant transaction volume from Poirot, undercutting efforts to gain scale. Immediately after the launch, OpenX experienced a 30% year-over-year decline in DV360 advertiser spend, and Google internally identified Poirot as the “biggest culprit.” As a result of the loss, OpenX was forced to lay off approximately 100 employees. Other ad exchanges also felt significant drops in DV360 spend and complained to Google. Google’s internal discussions confirmed that Poirot 2.0 was the cause.

230. Poirot’s success enabled Google to maintain the 20% revenue share fee it has charged on its ad exchange since 2009. Before Poirot, Google employees believed such a high fee was no longer sustainable, as header bidding risked commoditizing ad exchange services; if header bidding continued, Google’s employees expected the fee to drop to 5%. By 2019, after Poirot 2.0 was fully implemented, Google’s Americas Partnership Finance Lead noted that on Google’s ad exchange fee, “we should continue to hold the line, esp. given currently healthy growth levels, since project Poirot.” Poirot, which continues in some form today, shifted spend to Google’s ad exchange with its higher revenue share fee; deprived competing ad exchanges of scale, auction pressure, and higher win rates; lowered publisher payouts; and limited advertisers’ ability to fully and effectively spend their budgets—all without any improvements to Google’s own ad exchange.

[21] Project Poirot was named after Agatha Christie’s iconic master detective character, Hercule Poirot. Project Poirot was designed to identify and respond effectively to ad exchanges that had adopted header bidding technology.

[22] Google also charges an additional fee for use of its DV360 service (on average 9%), which is the same regardless of the ad exchange where the inventory is purchased.