Ad Tech’s Magic Act Could Blow Up the Internet

Former Google employee Tim Hwang warns that today’s ad-driven internet bubble is an elaborate fiction — and a ticking time bomb.

Sam Island

In Googled: The End of the World as We Know It, Ken Auletta identifies Viacom CEO Mel Karmazin as one of the first old-media tycoons to recognize the existential threat that Google posed to the ad industry establishment. When he visited the company’s headquarters in 2003, founders Larry Page and Sergey Brin “extolled the value of being able to measure everything, including the effectiveness of advertising.”

Alarmed, Karmazin countered that ambiguity was the industry’s bedrock. “Advertisers don’t know what works and what doesn’t,” he remarked. “That’s a great model.”

He departed with a dry warning: “You’re fucking with the magic!”

Karmazin’s blunt wit evokes a bygone era of admen hatching schemes in smoke-filled rooms. In the decades since, the game has changed in fundamental ways. “Today, the marketplace is effectively algorithms buying and selling attention online,” said Tim Hwang, author of Subprime Attention Crisis: Advertising and the Time Bomb at the Heart of the Internet, in a recent interview.

Instead of advertisers buying ad space from publishers, programmatic advertising allows them to bid directly on units of human attention — bundled into demographic markers to target particular users — as commodities via automated ad exchanges, which in turn distribute ads across innumerable websites. Because the whole system is heavily automated, the details of how it sustains itself are surprisingly opaque.

“Even engineers that work at Google aren’t particularly knowledgeable about how the company makes money,” said Hwang, himself a former policy lead for artificial intelligence at Google. “That’s a really interesting aspect of our everyday experience of the internet. There’s this enormous financial infrastructure that powers the whole thing, but its actual mechanics are a little obscure.”

Subprime Attention Crisis is Hwang’s attempt to demystify that infrastructure. In it, he folds a tidy history of programmatic advertising into something like a manifesto, arguing that much of the internet’s attention-fueled economy is built on an unstable fiction. The book draws an extended comparison between the pre-2008 subprime mortgage crisis and the business models of internet titans like Google and Facebook, arguing that we may be living through the swell of a second dot-com bubble. If Hwang is right, it’s hard to forecast how much of the digital economy could burst along with it.

Advertising accounts for as much as 90 percent of Google’s revenue, boosting the market capitalization of its parent company, Alphabet, to roughly the size of Mexico’s GDP. Facebook — Google’s biggest competitor in digital marketing — launched soon after Karmazin’s visit, and within a decade had claimed the biggest IPO in history on the back of a business model that was entirely focused on fucking with the magic. When you factor in Amazon’s significant growth in the digital ad space in recent years, three of the highest-valued companies in the world control about two-thirds of the market.

Much of today’s internet is built around the incentives of digital advertising. The ubiquitous convenience of paying for services like music distribution, video streaming, social media, pornography, free-to-play games, and journalism with attention and data has created a generation of consumers who are prone to reject alternatives. For the time being, venture capitalists also appear to be willing to subsidize platforms like Twitter, which took over a decade to turn a profit, simply because they command hordes of potentially monetizable eyeballs. Meanwhile, the attention-maximizing incentives of these platforms have proved vulnerable to manipulation by political actors, conspiracy theorists, and opportunists of all stripes.

We need to take the system down, but we don’t want it to explode in an unexpected way.

In Hwang’s telling, the trouble started when Google officiated the once-odd marriage of two familiar quants: Silicon Valley’s street-dumb engineers and Wall Street’s shrewd opportunists, who came together to build AdWords (now Google Ads), an ad marketplace and distribution platform that launched in 2000. Advertising was only expected to account for 10 to 15 percent of the company’s revenue at the time, but efficiently leasing any amount of ad space alongside search results — amorphous streams of algorithmically-generated content — presented a logistical nightmare. “The reason they invented this programmatic advertising thing,” Hwang told me, “is just that they needed to deliver a lot of ads.”

AdWords tacked an automated auction system onto a relevance prediction algorithm, laying the basic foundations of programmatic advertising and setting the course of the online economy to come. “When I request to load a web page, there is an automated auction where buyers will bid for my attention,” explained Hwang. “Based on the winner of that auction, an ad is delivered instantaneously. That all happens in the time between when I click on a link and when the website actually loads.”

This model was initially seen as an engineering solution to a technical problem, but Hwang chronicles a process by which its machinery was swiftly handed off to a new class of financial managers — people like Hal Varian, Google’s chief economist and “godfather to the advertising effort” at the company, who oversaw a project to create a keyword indexing system designed as the “search-word advertising equivalent of the stock market.” With software engineers on the gas and financial planners at the wheel, the Google of the early aughts consolidated the constituent bits of what Nicholson Baker would later describe as “an empire built on tiny grains of keyword-searchable sand.”

Transforming the ad industry into a financial market demanded an extreme degree of commodification, segmenting human behavior into measurable and standardized units to be bought and sold like futures on a stock exchange. This reality is the basis for the emergent cliché that data is “the new oil” of the digital economy, with Google and Facebook its great refiners.

“What is different about the present-day online advertising system,” said Hwang, “is the extent to which it has enabled the bundling of a multitude of tiny moments of attention into discrete, liquid assets that can then be bought and sold frictionlessly in a global marketplace.”

Google developed programmatic advertising, but Facebook perfected the art of constraining behavior to suit its needs, delivering “innovations” that invite users to boil their identities and relationships down to a standard menu of form responses. In our conversation, Hwang singled out Facebook’s infamous 2015 “push to video” as a prime example of the digital ad market’s opacity.

“Facebook basically said, ‘According to our data, if you want to get in front of consumers on Facebook, you need to invest in video production,’ ” recalled Hwang, describing the move as an attempt to siphon money from the lucrative television advertising market. “It later turned out that they had overstated the amount of time that people spent watching video on Facebook, on the order of 60 to 80 percent. Whether or not you think there is malfeasance here, it goes to show how opaque the marketplace is, because there is no way to fact-check Facebook on these claims.”

As with the pre-2008 housing market, programmatic advertising has dodged oversight in part by hiding behind a veil of complexity.

Programmatic advertising promises to deliver human qualities as fixed commodities at inhuman scale, but any argument for its efficiency rests on the accuracy of the data that fuels it. There are plenty of reasons to be skeptical.

“There’s a tendency to start with a very basic question,” said Hwang. “Do online advertisements succeed at altering people’s behavior? People have to realize that there are so many layers at which ads can fail before we can even get to that.”

Take impressions, for example. The basic unit of digital ad performance, impressions measure how many times an advertisement is loaded on a web page. But even with a metric this simple, we need to remember that all impressions are not created equal. A 2014 Google study estimated that more than half of all impressions are actually not viewed by people. Two years later, Forrester Research concluded that a similar percentage of all display ad dollars, totaling $7.4 billion, were being wasted on fraudulent or unviewable inventory. The rising popularity of ad blockers means that the proportion of impressions that are seen by real people has likely dropped even further.

Click-throughs, the gold standard for assessing ad impact, are perhaps even more vulnerable to manipulation. Programmatic advertising makes fraud simple: A scammer sells ad space through an exchange, then uses automated scripts or paid humans (usually exploited workers in the global south) to drive up their value and deliver useless clicks that register as genuine. Some firms even generate fraudulent clicks on competitors’ ads in order to burn their budgets on wasted ad spends.

When advertisements reach their intended targets, engagement has been trending down for a long time. The first banner ad, hosted on the web version of Wired in 1994, saw an absurdly high click-through rate of 44 percent; in DNO the SEO Revolution, Robert McAnderson notes that in the 1990s, “it was not uncommon [for banner ads] to have rates above five percent.” In 2018, the average click-through rate of banner ads hovered around 0.46 percent, with around 60 percent of organic mobile clicks believed to be accidental. Even if someone clicks an advertisement and then makes a purchase, it’s hard to know whether the ad actually influenced the sale, because demographic micro-targeting exacerbates selection bias: If an advertiser’s content is only shown to people who are already considered likely to buy the advertised product, it’s possible that the ad will merely capture sales that would have happened anyway.

Meanwhile, there are incentives to exaggerate reality at almost every rung of the supply chain. Publishers that lease ad space and the exchanges that fill it all make more money if metrics indicate success, regardless of actual impact. Marketing teams and comms strategists may even have reason to turn a blind eye, as artificially boosted metrics can help convince employers to fund their work. As anti-ad fraud hawk Augustine Fou put it in Forbes: “Everyone up and down the supply chain wants the fraud to continue because they make more money while it persists.”

Programmatic advertising was touted as a solution for uncertainty, but its metrics, for all their placating appeal, are drifting further and further from reality. Mel Karmazin thought Larry Page and Sergey Brin were fucking with the magic; what they really did was improve the pitch.

Hwang makes a convincing case that programmatic advertising is at the heart of an enormous speculative bubble, but what could replace it is far less clear. 

“The fact that advertising has been a way to pay for content online for so long makes it difficult for alternatives to form. People have become so unused to paying a certain amount of money for a certain type of content,” he told me, pointing out that many tech companies have thrived simply by commodifying already-existing tech solutions in order to normalize exchanges that might otherwise seem unnatural. Amazon, for example, standardized a practice of paying book-like prices for what are essentially text documents that have been made artificially harder to edit, share, and move around. The programmatic advertising model has become so universal, and so thoroughly naturalized, that it’s difficult to imagine how alternatives can compete their way into existence. 

“I have a couple friends who I would call ‘ad anarchists,’ who are basically like, look, if your argument is correct, and the marketplace has this time bomb built into it, let’s just let it go off,” Hwang told me. “That is a position you can hold, but the system is so interwoven that it’s not just a question of whether Mark Zuckerberg will make another billion dollars. Journalism, for example, is deeply bound up in it. We need to take the system down, but we don’t want it to explode in an unexpected way.”

Hwang instead advocates for a “controlled demolition” of programmatic advertising, with room for both legislative action and bare-knuckled activism. The former, he imagines, could look like the federal response to the Great Depression: market transparency regulations in the mold of the Securities Act of 1933. As for the latter, he speculates that hacktivists could draw attention to the ad market’s structural weaknesses by spreading bad data and stuffing sand in its gears.

As with the pre-2008 housing market, programmatic advertising has dodged oversight in part by hiding behind a veil of complexity. “If we had a broader public understanding of how these markets work, I think that would be a really powerful tool,” Hwang said. With renewed scrutiny facing Google, Facebook, and Amazon, the idea of using antitrust law to go after tech companies has recently gained serious traction in congress, in legal circles, and in the media.

But it’s unclear whether Hwang’s preferred regulations would slowly kill programmatic advertising or simply save it from destroying itself. And are the slow-moving, partisan bodies of American government really even capable of keeping pace with — let alone effectively addressing — such a diffuse and rapidly evolving technical problem? Building a sustainable alternative to today’s flimsy digital economy will likely require exploring radically different models of ownership, decision-making, and engaging with our collective humanity online — and it will almost certainly demand that we abandon the perverse incentives of today’s digital ad market.

Artwork By

Sam Island

Contact Us

Have an idea for a story or illustration? Interested in discussing partnerships? We want to hear from you. Send us a note at info(at)thereboot(dot)com.

Recommended Reading