Before he was a research fellow at Georgetown University, Tim Hwang used to work on machine learning technologies at Google. His first book, “Subprime Attention Crisis”, will be out tomorrow, and Gilad Edelman of Wired reviewed it:
If the financial market of the aughts was dangerously opaque, so, too, is modern internet advertising. In the early days of online ads, a brand would strike a deal with a website owner to host a paid banner. The onscreen space for that image, known as the ad inventory, would be sold by the publisher directly. (The magazine you’re reading right now made the first such transaction, back in 1994.) Today, the process has grown far more complicated, and humans are barely involved. “As they do in modern-day capital markets, machines dominate the modern-day ecosystem of advertising on the web,” Hwang writes. Now, whenever you load a website, scroll on social media, or hit Enter on a Google search, hundreds or thousands of companies compete in a cascade of auctions to show you their ad. The process, known as “programmatic” advertising, occurs in milliseconds, tens of billions of times each day. Only automated software can manage it.
Similar conditions were in place when mortgage-backed securities flooded the market in the early 2000s. These financial instruments traded at prices far above their true value, because the average trader had no idea they were backed by toxic assets. Once the truth came out, the bubble burst.
Hwang thinks online ads are heading in the same direction, since no one really grasps their worthlessness. There are piles of research papers in support of this idea, showing that companies’ returns on investment in digital marketing are generally anemic and often negative. One recent study found that ad tech middlemen take as much as a 50 percent cut of all online ad spending. Brands pay that premium for the promise of automated microtargeting, but a study by Nico Neumann, Catherine E. Tucker, and Timothy Whitfield found that the accuracy of that targeting is often extremely poor. In one experiment, they used six different advertising platforms in an effort to reach Australian men between the ages of 25 and 44. Their targeting performed slightly worse than random guessing. Such research indicates that, despite the extent of surveillance tech, a lot of the data that fuels ad targeting is garbage.
I am intrigued by the premise that automated advertising is comparable to subprime mortgages and CDOs. It is an argument that squares with my belief that, similar to the 2008 financial crisis, the laissez-faire regulatory environment and industry domination of the United States is bound to have catastrophic worldwide consequences. This book is going on my reading list.