by Tim Hwang
Opacity isn’t dangerous only because it can cause errors in valuation. It also allows the active inflation of a market despite the fundamental shakiness of the thing being bought and sold. This sometimes results from irrational levels of market confidence, a regular feature of financial crises going back hundreds of years. Reinhart and Rogoff refer to the “this time is different” syndrome, in which “financial professionals and, all too often, government leaders explain that we are doing things better than before, we are smarter, and we have learned from past mistakes.”52
But market inflation can also result from deliberate recklessness. Perverse incentives can emerge in a market, encouraging players to continue pushing the bright horizons for a marketplace despite knowing that major structural problems exist. Later reporting about the mortgage crisis illustrates the extent to which major financial actors were aware of the dangers but continued to participate anyway. In the infamous words of Chuck Prince, then the head of Citigroup, “When the music stops … things will be complicated. But as long as the music is playing, you’ve got to get up and dance. We’re still dancing.”53
Opaque markets, in short, allow expectations about a market to diverge from reality. On the one hand, things can get a whole lot worse without anyone knowing. On the other, conflicted players can overheat the market without being checked. This divergence between rosy outlooks and structural vulnerabilities is kindling for crises of confidence. When a hot, overpriced commodity is discovered to be effectively worthless, panic can set in, causing the market to implode.
Closely examining the state of the programmatic advertising marketplace makes it clear that these dynamics are very much in play in the economy of the web. The online attention marketplaces that underwrite the internet might have inherited not only the structure of modern financial markets but the pathologies of those markets as well. The seemingly endless growth of the modern-day advertising giants of the web may rest on shakier foundations than might initially appear.
3
Opacity
The advertising industry has long struggled with a simply stated but deeply complex question: Does advertising work? In other words, how does one really know that the advertising messages being broadcast actually influence the great mass of browsers, readers, and listeners out in the world? Traditionally, this struggle has been summed up by a pithy adage often attributed to John Wanamaker, an early advertising industry pioneer, who noted that “half the money I spend on advertising is wasted; the trouble is I don’t know which half.”1
To the outsider, this long-standing institutional ignorance can seem a bit puzzling. After all, one would think that making reasoned judgments about what types of advertising work and don’t work is the primary job of the marketer, and the marketing industry as a whole.
It turns out that this is a much more difficult task than it might initially seem. Imagine you have to promote a new brand of cereal. You want lots of people to know about this new cereal, so you pay to put up a billboard right next to a busy highway. You might know roughly how many people pass by that billboard based on traffic statistics, but that doesn’t tell you anything about who noticed the advertising, who was influenced by it, and whether it did anything for the cereal. You might find that cereal sales increased after the billboard went up, but how can you ever know whether the sales were due to that billboard? A person who saw the billboard and then later bought the cereal might have picked it up on a whim at the supermarket, or been persuaded to try it by a friend. There’s simply no way to know whether the billboard played a role in the purchasing decision.
In contrast, online ads are traceable, trackable, and thoroughly quantified. This has been a selling point for digital advertising since the very beginning. Google’s early nemesis GoTo—later renamed Overture—claimed in the early days of programmatic advertising that “[online advertising] is one of the few truly measurable media.”2 Google’s then CEO, Eric Schmidt, would claim in 2008 as the AdWords product was taking off that the company’s growing profits were attributable to its “system [being] better and more measurable than other forms of advertising.”3 Wanamaker’s law appeared to have met its match.
The popular press covering the internet often echoed this opinion. In 1998, Wired magazine opined that “the Net is accountable. It is knowable. It is the highway leading marketers to their Holy Grail: single-sourcing technology that can definitively tie the information consumers perceive to the purchases they make.”4 This claim was repeated almost a decade later in 2006, when The Economist declared an end to the “Wanamaker era” of advertising, arguing that the internet was giving rise to a world where “advertisers pay only for real and measurable actions by consumers, such as clicking on a web link, sharing a video, placing a call, printing a coupon or buying something.”5
That the internet affords a quantity of information inconceivable to earlier generations of advertisers is indisputable. A complex ecosystem of data brokers, tracking cookies, and surveillance now enables advertising to be precisely targeted, and the results of that advertising to be monitored. Today, an advertisement can be easily targeted to a specific audience defined by an incredibly granular set of characteristics, from race and gender to income and location. Ads delivered online can track how long the ad is viewed, when it is seen, and how soon people make a purchase after viewing it.
This monitoring infrastructure allows advertisers to accomplish feats of marketing that would have been prohibitively expensive or otherwise impossible in a pre-internet era. “Retargeting”—which identifies when a specific user has seen a product online, and delivers that user ads for the same product again and again as she browses the web—is possible because of the level of tracking that online advertising provides.6 Computer security experts have shown that it is even possible to identify the location of a single, specific person using only the geotargeting infrastructure of commonplace programmatic advertising tools.7 Our cereal billboard campaign is positively Stone Age, from a measurability standpoint.
Today’s advertisers are flooded with data about consumers. If anything, they are struggling to figure out how to manage the tidal wave of data now available to them. It is technically feasible to know that this ad was viewed by this individual, at this time, in this location. It’s even possible to know when and whether the individual purchased the advertised good or service.
Deeper context, however, remains elusive. Consumer behavior is exposed in fine detail, but the overall market stays opaque. The wealth of tracking data doesn’t help advertisers determine a fair going price for reaching a particular kind of person, and buyers in the programmatic marketplace can have limited knowledge about where or how their ads will appear.
The measurability of the online ad economy is an inch wide and a mile deep. As such, the tidal wave of data that has accompanied the development of online advertising provides only an illusion of greater transparency.
This matters because opacity permits market bubbles to form. In the mortgage crisis, the complexity of the market in exotic financial instruments made it challenging to see that the underlying mortgages were extremely likely to default. If buyers had known that AAA-rated mortgage-backed securities contained so much bad debt, it would have been challenging to get anyone to buy the assets.8
Modern online advertising remains deeply opaque on three fronts. First is the ever-increasing automation of the marketplace. Second is the creation of dark pools of liquidity where advertising inventory is bought and sold outside the public eye. Third is the dominance of platforms, like Facebook and Google, that have frequently introduced new layers of opacity into the advertising marketplace.
I. Algorithmic Trading
Programmatic advertising delivers the vast number of ads that you see online today. In order to function, the entire system depends on a real-time auction between advertisers that takes place while a web page is loading. Making this real-time bidding (RTB) process seamless requires extreme speed.
Ad exchanges
have achieved this speed through automation. Advertisers and publishers alike program systems that autonomously offer and price ad inventory in the marketplace. Advertisers can precisely define the kind of attention they would like to bid for, specifying audience parameters, maximum bid prices, timing, frequency, and certain kinds of inventory.9 In turn, publishers can set minimum prices, rules for advertiser eligibility, and other preferences on the sale of their inventory.
This highly automated arrangement has allowed online advertising exchanges to deliver mind-bogglingly large quantities of inventory night and day, unceasingly. As in the financial markets, algorithmic trading has become the rule, rather than the exception.10
This blend of speed, automation, and scale has introduced new opacity into the marketplace even as it has allowed those marketplaces to substantially expand. The vast number of transactions taking place can make it impossible for anyone to monitor precisely where an ad ends up and why. This is a problem even for the professional buyers and sellers of ad inventory themselves. One New York Times investigation quoted Alex Treadway, then chief operating officer of the conservative website The Daily Caller, saying, “There is so much junk between us and the companies that buy ad space on our pages it will blow your mind … It would take us weeks of research to figure out which ad network provided [an] ad.”11
Consider the chronic challenge posed by “brand safety”—the industry term for avoiding having advertisements show up next to objectionable content. YouTube has been the subject of multiple embarrassing news reports highlighting that brands advertising on the platform occasionally find their ads alongside extremist videos. In 2018, CNN reported that ads from more than three hundred companies on YouTube were placed next to channels that promoted “white nationalists, Nazis, pedophilia, conspiracy theories and North Korean propaganda.”12 This included major brands like Adidas, Hilton, and Hershey. This is not an isolated incident. The issue has been a long-standing concern among advertisers as the programmatic advertising ecosystem has matured over the past two decades.13
The persistence of this problem is not due to a lack of interest in resolving it. A survey in 2018 showed that brand safety was one of the top concerns among professionals within the industry, next to worries about “transparency” and the European Union’s General Data Protection Regulation.14 A 2017 survey of its members by the Association of National Advertisers found that 78 percent of respondents were “either concerned or very concerned about brand safety issues in programmatic buying.”15 The fact that brand safety remains an ongoing challenge even among platforms run by the most well-resourced companies in the world suggests how structural this opacity problem is.
To date, the primary approach has been to use human review in an effort to weed out these errors. The scale of the modern online advertising ecosystem makes this at best a partial solution. Mark Zuckerberg, commenting on Facebook’s decision to hire more content moderators in 2017, declared, “No matter how many people we have on the team, we’ll never be able to look at everything.”16 There is hope that recent advances in artificial intelligence might be able to ensure greater brand safety, but it is unclear whether these systems will ever have the context sufficient to address these issues in a comprehensive way. AI systems are able to recognize specific, concrete targets within images but remain extremely limited in understanding broader context. This makes AI a promising tool for dealing with problematic content where the boundaries of what is prohibited are clear, but a highly limited tool when dealing with more general categories of objectionable content.17 AI will not serve as a cure-all to brand safety writ large.
In the programmatic advertising marketplace, it can be challenging to know where an ad will end up and why. This risk to brand safety pressures well-positioned buyers and sellers to look for safer ways to place ads online. This pressure, ironically, has introduced another compounding layer of opacity in the marketplace for online advertising.
II. Dark Pools
Not knowing where and why ads end up in the places that they do is a big problem. But in the attention marketplace, advertisers sometimes do not even know whose attention is available, much less the price that they should be paying.
“Dark pools” are a major phenomenon in the financial markets, accounting for some 14 percent of all equity volume traded in the United States.18 These are, in effect, private stock markets run by investment banks and other financial institutions about which no public information is available. These structures ostensibly serve to allow buyers and sellers to execute large-scale transactions without triggering reactive fluctuations in price on public exchanges like the Nasdaq.
Dark pools are controversial in the financial sector because the opacity they create provides opportunities for abuse. When the public markets no longer reflect the supply and demand of all players, the prices shown on these exchanges will not reflect the actual value of the security. Dark pools enable practices like “front-running,” in which high-frequency trading firms leverage their insider knowledge of these private marketplaces to profit off traders with slower access to information.19
It turns out that parallel structures have also emerged in the online advertising markets. Platforms increasingly give select buyers and sellers access to private marketplaces (PMPs)—exclusive exchanges for ad inventory. PMPs allow selected advertisers who have negotiated a special deal with a publisher to bid for advertising space, usually of a higher quality and in a less crowded, and therefore less competitive, market. These arrangements are attractive because they offer better transparency to the participants and also allow buyers to keep targeting data and other valuable information away from the public markets.20
PMPs are a growing segment of the transactions taking place in online advertising. In 2018, 45 percent of all the money spent in real-time bidding auctions took place within the confines of a PMP. Of the $19 billion that are projected to enter into the programmatic ad space between 2018 and 2020, the majority will go toward PMPs and other private arrangements.21 By 2021, spending on PMPs is projected to exceed spending in open exchanges by more than $2 billion.22 Ironically, this trend is driven by the opacity of public programmatic marketplaces, where brand safety and an overall lack of transparency make for a more risky advertising environment.
On a systemic level, PMPs introduce a new source of opacity while ameliorating the problem for a select few. If you lack access to a PMP, you cannot see the activity in these premium dark pools of advertising inventory. This introduces many of the same structural issues that accompany the rise of dark pools in the financial markets.
For one, the publicly available price no longer reflects the supply and demand across all the players in the marketplace. That means it is harder to know what the prevailing value of an ad is. Is the price paid to deliver an ad to a given consumer too much or too little relative to the overall market? Is a specific kind of ad receiving more demand over time, or is that demand falling?
Ambiguity about price reintroduces some of the challenges that programmatic advertising was intended to solve. By creating an interconnected set of markets with competitive bidding, programmatic advertising promised to foster price transparency in the marketplace. That is, players in the market would have a good sense of how the rest of the market was sizing up the opportunities to buy and sell attention. The growth of PMPs muddies the waters again.
Programmatic advertising makes it challenging to effectively track whose attention an advertiser is buying. The growing presence of PMPs creates a hazy landscape of pricing and available inventory for the ad buyer. But the opacity problem is not limited to PMPs alone. The highly consolidated nature of the programmatic advertising ecosystem itself contributes significantly to the murkiness of the internet’s attention economy.
III. The New Intermediaries
The development of online programmatic advertising is often cast as a story of disintermediation. This is true to some extent. Traditionally, marketing agencies have played a major role in
facilitating the interaction between advertisers and publishers. Programmatic advertising has supplanted this middleman role, allowing buyers and sellers of ads to transact directly. As a result, the gatekeeping role of agencies has eroded, and marketing agencies have struggled as “self-serve” programmatic marketplaces have grown.23
That is undoubtedly a positive story in some respects. Advertisers have more direct control over how their messages are distributed to the public and can choose how best to achieve their specific goals. Publishers are no longer beholden to marketing agencies, who long served as gatekeepers to major advertising dollars. In economic terms, disintermediation is a happy ending: costs should be lower overall, and supply should be able to meet the demands for advertising faster and more effectively.
This story overlooks one important point: the companies that brought about disintermediation have themselves become intermediaries in the advertising space. Google, Facebook, and major ad exchanges have huge sway over the rules of buying and selling ads because of their relative size and the high level of consolidation in the marketplace.
That consolidation increases market opacity. The availability of data about the state of the global online advertising marketplace is now dependent on the business decisions of a small number of platforms. The failure to disclose relevant information or the publishing of misleading information can severely distort the advertising markets.
One striking illustration is the subject of an ongoing lawsuit around claims that Facebook made in 2015 promoting the attractiveness of video advertising on its platform.24 At the time, the company was touting online video—and the advertising that could be sold alongside it—as the future of the platform, noting that it was “increasingly seeing a shift towards visual content on Facebook.”25 Mark Zuckerberg argued publicly that “the vast majority of the content that people consume online will be video.”26 Facebook backed this push by tweaking its News Feed algorithm to promote video, directing huge amounts of traffic toward this type of content.27
