Subprime Attention Crisis

by Tim Hwang

  The company’s “pivot to video” encouraged publishers to invest heavily in the creation of more video content, often at the expense of existing staff. The effort also fed substantial industry hype about the value of video advertising on Facebook.28

  But it turned out that Facebook overstated the level of attention being directed to its platform on the order of 60 to 80 percent. By undercounting the viewers of videos on Facebook, the platform overstated the average time users spent watching videos.29 Facebook later admitted that several other key video advertising metrics were overstated.30 Creative accounting has dogged other marketing metrics that Facebook has promoted. One analyst noted that Facebook claimed to be able to reach 25 million more eighteen- to thirty-four-year-olds in the United States than should exist according to the U.S. census.31 Another report cast doubt on Facebook’s claims about the performance of “Watch,” its stand-alone video platform.32

  These inconsistencies have led some to claim that Facebook deliberately misled the advertising industry, a claim that Facebook has denied.33 Plaintiffs in a lawsuit against Facebook say that, in some cases, the company inflated its numbers by as much as 900 percent.34 Whatever the reasons for these errors in measurement, the “pivot to video” is a sharp illustration of how the modern advertising marketplace can leave buyers and sellers beholden to dominant platform decisions about what data to make available.

  As this book goes to press, the controversies sparked by these and other disclosures have made Facebook and Google more willing to voluntarily accept third-party auditing of their metrics.35 Under these new arrangements, the Media Rating Council—an independent nonprofit founded in 1964 that provides third-party accreditation of media—is granted internal access by the platforms to confirm that the specific metrics they report to advertisers are legitimate.36 This development marks the end of a protracted period in which Google and Facebook resisted suggestions that they accept more verification and accountability of the metrics they distribute.37

  I’m skeptical that the platforms’ newfound tolerance of oversight represents a structural change in the marketplace. These accountability measures leave the fundamental balance of power between the intermediaries and the markets they facilitate largely unchanged. Indeed, both Google and Facebook retain the ultimate power to decide precisely which third-party partners are permitted to verify and measure advertisements distributed on these platforms. This gives them tremendous practical leverage over the degree of scrutiny that they face. Even with the ongoing audits, advertisers’ access to granular data about the overall programmatic advertising marketplace will remain limited.

  Public pressure and the threat of losing major advertisers have forced the platforms to accept only a limited level of third-party verification. Such reluctance might not have been possible in a more competitive marketplace, where media buyers could easily shift to platforms that offered more transparency. Instead, advertisers face the unenviable choice of buying ads on the unverifiable assurances of the major platforms or giving up the huge audiences these companies command. In this sense, the highly concentrated market of online advertising itself increases opacity in the marketplace.

  Why Opacity Matters

  Opacity in the programmatic advertising markets can seem very much like an industry insider’s problem. Why should we care about the degree to which advertisers can get data about advertising inventory? If anything, shouldn’t we be glad that the market is perhaps more opaque than advertisers would like? Opacity limits the intrusiveness of advertising and perhaps in some marginal way helps to protect a modicum of privacy.

  We should care because opacity is a key precondition for market failure. Opaque markets are ones in which participants cannot accurately assess the value of things being bought and sold. This can allow irrational exuberance to push the prices of things far above their real value. It can also allow the value of things to erode significantly without it being widely known. In both examples, expectations about the market diverge from reality. When these expectations come crashing down—whether in the value of online advertising or in the stability of mortgage-backed securities—panic sets in.

  Granted, the infrastructure of online advertising provides more clarity in certain areas, especially when compared with earlier channels of advertising. There is an unprecedented level of information available about the performance of advertising once it has been delivered. How many times and when an ad has been seen, for instance, are standard, widely available metrics that would have been very expensive or impossible to acquire in the past.

  But murkiness persists in critical areas. Algorithmic trading makes it challenging to assess where ads end up and who sees them. Dark pools obscure the real price of ads, allowing selected buyers and sellers access to more information and better opportunities to place ads unavailable to the rest of the market. Market concentration has made buyers and sellers beholden to data provided by a handful of major platforms, raising the barriers to acquiring verified information about ad quality. It is unclear whether the prices emerging from real-time bidding accurately reflect the value of advertising inventory across the entire market. It is unclear why the bidding algorithms behave the way they do. It is easy to get granular information about a given ad that you have placed, but far harder to get a sense of the overall marketplace.

  We should care about opacity in the programmatic advertising marketplace because it has a big role to play in determining the future of these marketplaces and the future of an internet that is dependent on ads as a primary source of revenue.

  On its own, opacity doesn’t create a financial crisis. Pre-internet advertising was highly limited in the metrics that were available about performance, but it grew substantially as a market over the course of the twentieth century. People wanted to buy ads, even in an information-poor environment. It is still challenging for the buyers of attention to assess the quality and effectiveness of the assets they are buying, even with today’s wealth of attention-tracking data. But if the underlying attention purchased online remains high quality nonetheless, the market might remain sustainable.

  Opacity merely sets up the circumstances for expectations to diverge from reality. To precipitate a market crisis, other forces must both erode the true underlying value of the assets being traded in the marketplace and wildly inflate their perceived market value. It is to those forces we now turn.

  4

  Subprime Attention

  Opacity in a marketplace creates a smoke screen behind which an economic situation can deteriorate significantly without the broader market’s becoming aware of it. The realization that the thing being bought and sold is in fact worth less than buyers and sellers believe shakes confidence and can produce widespread panic.

  There are good reasons to believe that online advertising inventory is steadily decreasing in value over time. Two forces drive this erosion of value: structural shifts in what people pay attention to, and a massive global economy of fraud in the programmatic advertising marketplace. These trends are hidden by the murkiness of online advertising, as well as by a pattern of bad incentives that encourage ongoing efforts to pump up and hype the market.

  Amid all the industry jargon, one can lose sight of advertising’s ultimate objective: to shape the behavior or perceptions of the viewer in some way. For it to do this, a few factors have to be in play. The message being delivered should be relevant to the recipient. It should be delivered at the right time. Most fundamentally, the ad must capture the recipient’s attention. If the ad is ignored, money spent delivering it is effectively wasted.

  This is a long way of saying that advertising packages attention: it is not the attention itself. All an advertiser wins in an ad exchange auction is the right to display its content on a loading web page. When a demand-side platform (DSP) is programmed to seek out opportunities to reach a demographic like “males 18 to 24 living in the United States,” it tells us whom the advertising will ideally reach, but not whether the people who ac
tually see the ad will be persuaded, or even interested.

  This divergence between the asset being bought—ad inventory—and the asset underlying it that defines its value—attention—directly parallels what happened to collateralized debt obligations (CDOs) during the 2007–2008 crisis.

  CDOs were, in effect, bundles of mortgages. Financial institutions packaged together mortgages of differing risk and then sold the stream of payments coming from these loans as a single asset. But the CDOs were not the mortgages themselves, and each CDO in practice contained different bundles of mortgages taken by different homeowners in different places. One CDO might contain high-quality home mortgages that would reliably pay out over the entire lifetime of the loan, and an identical CDO might be filled with high-risk mortgages likely to default. In financial parlance, both CDOs and online advertising inventory are derivatives—they derive their value from an underlying asset. CDOs draw their value from the mortgages they contain; online ad inventory draws its value from the attention that it represents.

  We can think about any unit of advertising—a banner ad on a website or a billboard on the side of a highway—as a rough proxy for the collection of eyeballs that see it. In the same way that we might peel back a CDO to learn what mortgages it actually contains, we can peel back an ad and assess the quality of the attention that it captures.

  When we do this, a twofold problem emerges. First, the value of the attention “packaged” by online advertising is declining. Online advertising is increasingly ignored—or actively resisted—by the public at large. Second, the “attention” that ads do receive is increasingly garbage—the product of a massive, fraudulent economy designed to extract money from advertisers.

  Attention is subprime. The bottom is falling out even as prices are pushed higher and higher. This is made possible and exacerbated by the pervasive opacity of the marketplace, which allows real value and market value to drift further apart over time.

  Who’s Paying Attention?

  South Park is a pretty neighborhood located in the SoMa district of San Francisco. Set away from the busy, loud urban streets that bound the area, the neighborhood is built around a small oval park with a playground. It’s home to a collection of startups, marketing agencies, venture capital firms, and very expensive apartments. There’s a craft brewery, an artisan coffee shop, and a high-end grilled-cheese restaurant within a block of the park.

  South Park is a significant location in the lore of San Francisco technology. The initial idea for Twitter was reputedly sketched out in the South Park playground, and the neighborhood was home to many leaders of the first dot-com boom.1 But, beyond being a hub for product innovations that would shape the web, South Park is historically important for the role it plays in the broader development of the web as a business.

  It was here in late 1994 that the online advertising economy was born. Two South Park was then home to HotWired, a subsidiary of Wired magazine that owned and operated the publication’s website.2 In partnership with Volvo, AT&T, and other major brands, HotWired launched the web’s first banner ads.

  The early banner ads of the mid-1990s were wildly successful compared with the ads of today. One common ad industry benchmark for success is the click-through rate, which measures the percentage of people viewing the ad who subsequently clicked on the ad. In other words, just how compelling was an ad to given users that they went out of their way to click on that ad to learn more?

  The information we have suggests that this early generation of ads captured attention in a way that is virtually unheard-of today. When they launched in 1994, the first banner ads generated a remarkable click-through rate of 44 percent.3 That is, close to half of the people who saw these banners clicked on them. Today, banner ads command far less attention. One data set drawn from Google’s ad network suggests that the average click-through rate for a comparable display ad in 2018 was 0.46 percent. For some industries, that number is as low as 0.39 percent.4 That’s about one in every two hundred people. Recent attempts to measure click-through rates on Facebook ads reveal similar rates of less than 1 percent.5

  These banner-type ads are a hundred times less effective than they were about twenty-five years ago. Even these sub-1-percent click-through rates may overstate the effectiveness of ads on some platforms. On mobile devices, close to 50 percent of all click-throughs are not users signaling interest in an advertisement, but instead accidental “fat finger” clicks—users unintentionally clicking on content while using a touch-screen device.6 Ads may also drive a response among only a small segment of the population. In 2009, one study estimated that 8 percent of internet users were responsible for 85 percent of all advertisement click-throughs online.7

  Public indifference toward online ads is reflected in the surprisingly ambiguous empirical evidence that these ads do anything at all. One large-scale experimental study of online search ads in 2014 concluded that “brand-keyword ads have no measurable short-term benefits.” Ironically, ads generated engagement mostly among “loyal customers or [consumers] otherwise already informed about the company’s product.” The ads, in other words, were an expensive way of attracting users who would have purchased anyway, leading to “average returns that are negative.”8

  This indifference toward advertising is particularly pronounced among younger internet users. In 2013, a controlled experiment on more than a million customers to evaluate the causal effect of online ads concluded that a customer “between ages 20 and 40 experienced little or no effect from the advertising.” This was in spite of this demographic’s proportionally heavier usage of the internet. In contrast, the study found that customers older than sixty-five, despite constituting only 5 percent of the experimental group, were responsible for 40 percent of the total effect observed as a result of the advertising.9 This result suggests that the current effectiveness of advertising may depend on an aging and rapidly disappearing segment of the population.

  Skeptical minds might counter that click-throughs and sales are imperfect proxies for measuring whether ads are effective. Marketers frequently draw a distinction between “direct response” advertising and “brand” advertising. The former encourages an audience to make a purchase—by promoting a discount, for instance, or highlighting the attractive features of a product. “Brand” advertising, in contrast, is less about the immediate purchase and more about shaping the public’s associations with a brand and differentiating it from its competitors. Nike might invest in television advertising that is—at least in the near term—less intended to make consumers go out and buy sneakers and more intended to ensure that Nike maintains its status as a “cool” brand in the public eye. In brand advertising, clicks and sales are secondary to the advertiser’s goal. Even if the ability of advertising to drive clicks and sales is falling over time, there might be enough demand in brand advertising to ensure that the value of online advertising inventory remains sound.

  The vaporous nature of brand advertising means that the goalposts are perpetually moving: the objective of the advertiser may not be to drive any behavioral change that is actually measurable in any sensible time frame. It becomes hard to refute the effectiveness of advertising when the bar is set so low. But one thing is certain even in the brand advertising context: if the ad is never delivered in the first place, there is no way that it can be effective. Industry trends suggest that this is precisely what is happening in the programmatic ecosystem.

  The first problem is the ineffectual placement of ads on pages. The real-time bidding system might successfully load up an ad on a website, but there is no guarantee that the ad will be somewhere that a reader of the website will actually see. Ads might load at the bottom of the page outside the browser view, for instance, or they might be so small that they escape notice entirely. “Ad viewability,” as it is known in the industry, has emerged as a major concern. In 2014, Google released a report suggesting that 56.1 percent of all ads displayed on the internet are never seen by a human.10 One 2017 report by C
omscore found that this problem is particularly pronounced for ads purchased through the programmatic ecosystem.11 A staggering number of those ads are never seen by anyone at all.

  But the problem goes deeper than bad placement. Not only are people paying less attention to ads; they are also taking proactive steps to prevent ads from reaching them in the first place. Ad blocking grows more popular every year. One study by Deloitte from 2017 suggests that fully three-quarters of North Americans engage in “at least one form of regular ad blocking.”12 In 2016, 615 million devices around the world were actively blocking ads.13

  The news for advertisers gets even worse. Trends suggest that the industry is not even able to evade highly blocked markets by delivering ads to less blocked places. Markets outside North America, where ad blocking was historically less prevalent, might have been a potential haven for advertisers. But ad-blocking growth is particularly pronounced in these emerging markets. Driven largely by the growth of ad blocking in Asia-Pacific, global ad blocking grew by about 30 percent year over year from 2015 to 2016.14

  Mobile platforms might also have been a promising new frontier, given their relative growth compared to desktop devices and the lack of norms around ad blocking on mobile devices. But it turns out that mobile is where ad blocking has grown the most in the past few years. Ad blocker use on mobile devices grew by 40 percent to 380 million devices between 2015 and 2016. Fifty-nine percent of the smartphones in India implement ad blocking.15