The Fragility of Digital Advertising and the Future of the Attention Economy: Part I

 By Brad Honigberg

At the turn of the 20th century, American marketing pioneer John Wanamaker lamented that “half the money I spend on advertising is wasted; the trouble is I don’t know which half.” Over a hundred years later, digital programmatic advertising claims to have solved this problem.
The modern Internet economy is premised on the belief that algorithmically targeted programmatic advertising — driven by trillions of data-points about consumers behavior fed through sophisticated artificial intelligence software — works better than the Mad Men-style marketing of the pre-digital era. From mom-and-pop coffee shops seeking new customers to Russia’s Internet Research Agency hoping to sway elections, everyone online is using the same persuasion tools.
Today, Google earns more than 80% of its revenue from advertising and Facebook around 99%. Advertising also makes up a fast-growing share of Amazon’s revenue. Those three companies alone account for nearly 10% of the U.S. stock market’s total value, making them inextricably linked to the global economy. With the help of these Tech Giants and other advertising middlemen, brands, governments, and political campaigns invest in ad inventory to persuade audiences through targeting, analytics, and performance optimization. By directing ads across multiple channels (social, search, display, and mobile) and analyzing their performance to optimize for future campaigns, advertisers can achieve what has been termed “personalized mass persuasion.”
But what if digital advertising — the dirty fuel that has powered the rise of the modern Internet — isn’t as effective as we believe it to be? Increasing opaqueness, outright fraud, and perverse incentives give the digital advertising marketplace the features of a dangerous shell-game. As Tim Hwang asserts in his provocative book Subprime Attention Crisis, “the whole edifice of online advertising is, in short, bunk.”
Digital advertising is essentially an AI-driven global bazaar of human attention. Within this automated marketplace, attention vendors sell advertising inventory to the highest bidder among a pool of buyers through real-time bidding. On the supply-side, buyers specify the kind of attention they would like to bid for, audience parameters, maximum bid prices, timing, frequency, and certain kinds of inventory. On the demand-side, sellers set minimum prices, rules for advertiser eligibility and other preferences on the sale of their inventory. The guts of the programmatic advertising ecosystem mirror the architecture of global stock market. This design attempts to standardize and commodify human attention — an incredibly abstract concept — turning it into a liquid asset that can be bought and sold.
The speed, automation, and ever-increasing scale of this marketplace introduces elements of opacity. The nearly unlimited supply of granular user 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. Like in financial markets, ad exchanges lack transparency which enables insider trading without triggering reactive fluctuations in prices. Online advertising platforms increasingly give select buyers and sellers access to private marketplaces (or dark pools), allowing them to bid for advertising space that is often higher quality and in a less competitive market. This dynamic inflates the value of digital ads so that they no longer reflect the true nature of the marketplace.
Each year, the attention digital ads receive grows increasingly undeserved. The purported benefits of relying on digital ads are called into question by rising click fraud. “Click farms” rely on automated scripts or armies of paid humans to drive false engagement. Domain spoofing, when ad inventory is falsely made to look like space on a high-value website, is on the rise. According to some estimates, $35 billion are lost annually to such scams. There is zero guarantee that an ad will be placed on a prominent location on a website. More Internet users are using ad-blocking software. Furthermore, there is a growing movement to restrict invasive data tracking: New laws in Europe and California are helping people opt out of data sharing; Google plans to block third-party cookies that track user browsing; and Apple just rolled out its app tracking transparency (ATT) framework which allows users to decide whether their personal data can be tracked by app publishers.
This market for “fake attention” flourishes for the simple reason that it is highly lucrative and, in the short and medium term, it further inflates the perceived value of digital advertising. Marketing agencies and Big Tech ad exchanges are driven by perverse incentives to measure their own effectiveness by correlation-based conversion rates rather than by lift — the effect of someone seeing the ad on their chances of buying the product. Agencies engage in arbitrage, buying ad inventory at a discount and selling it at a markup to their own clients. A recent two-year study by PwC found that half of online ad spending goes to industry middlemen and almost a third of those ad-placing costs were completely untraceable.
As hype around targeted digital advertising continues unabated, an emerging body of research suggests that ads often hit people that are already the most likely to engage in the behaviors they are trying to encourage and that a lot of the personal data that fuels targeted advertising is useless. For example, in large study comprising 500 million user-experiment observations and 1.6 billion ad impressions, researchers found that Facebook ads had been overestimated by up to 4000% and that, for half of the fifteen experiments, lift had been overestimated by a factor of three or more. When analyzing six different advertising platforms ability to reach Australian men between the ages of 25 and 44, another study found that targeting performed slightly worse than random guessing.
A growing number of companies are scaling back their digital advertising budgets. For example P&G cut its budget by $200 million in 2018 and reported a 7.5% increase in organic growth the following year. More recently, Uber discovered they’d been defrauded out of $100 million — two-thirds of its total online ad budget — and found basically no change in the advertising effectiveness after significantly downsizing their spending.
To be fair, some digital advertising campaigns work, especially those powered by Google and Facebook which have their own specialized tools separate from the global ad exchange. Massive psychographic profiling campaigns can improve the persuasive power of digital advertising. As MIT data scientist Sinal Aral describes in The Hype Machine, micro-targeted advertising works best on a broad scale, especially when the campaign informs consumers of the characteristics, location, and prices of products and services that they may otherwise be ignorant about. Essentially, the grander the effort, the easier it is to optimize. Still, campaigns like this are the exception to the rule.
To conclude, there is reason to believe that the financial foundations of the Internet are not as firm as we believe them to be. In Part II, I will examine what the “subprime attention crisis” would look like and what can be done to deflate the digital advertising bubble before it bursts.

Brad Honigberg is pursuing a masters degree in Security Studies with a concentration in technology at the Georgetown University Walsh School of Foreign Service. He previously served as Social Media and Outreach Coordinator at the Center for Strategic and International Studies.

The Strategic Technologies Blog is produced by the Strategic Technologies Program at the Center for Strategic and International Studies (CSIS), a private, tax-exempt institution focusing on international public policy issues. Its research is nonpartisan and nonproprietary. CSIS does not take specific policy positions. Accordingly, all views, positions, and conclusions expressed in this publication should be understood to be solely those of the author(s).