2020-11-14

On Tim Hwang's book: Subprime Attention Crisis

Johannes Ernst

https://reb00ted.org/tech/20201114-notes-on-hwang-subprime-attention-crisis/

My friend Doc Searls has been talking about this book repeatedly in recent months, as have many others interested in rolling back surveillance capitalism, improving privacy and user agency, and cleaning up the unholy mess that on-line advertising has become. Finally I have read the book, and here are a few notes.

Tim Hwang makes three core points:

  1. Programmatic, on-line advertising is fundamentally, irredeamably broken.
  2. It’s not a matter of whether it will implode, but just when.
  3. Apply the lessons from the 2018 subprime mortgage crisis: advertising inventory is a different asset class, but the situation is fundamentally the same: eroding fundamentals in the face of an opaque, overhyped market, which will lead to a crash with similarly major consequences when it occurs.

I buy his first point. I mostly buy his second, but there are too many important differences with the market for collateralized mortgages in 2008 for me to buy his third. Ultimately that parallel isn’t that important, however: if he’s right that programmatic on-line advertising is headed for something dramatic, whether it’s like 2008 subprime mortgages or some other crash doesn’t matter in the end.

Why would anybody say programmatic, on-line advertising is broken? He has many examples, go read the book, but let me mention my personal favorite from personal experience: ads, to me, on Spotify:

  • Spotify, for a long time, advertised joining the Marine Corps to me. I should be flattered how young, vigorous, and gung-ho they consider me, but hmm, I don’t think so. This must be because they have some wrong data about me, and while Spotify got the Marine Corps' money all the same, the Marine Corps totally wasted their spend.

    While this example is particularly egregious, Hwang has many other examples, which argue that this is a major and pervasive problem.

  • I recently downloaded the personal data Spotify have about me, as I can because we have the CCPA in California. Looking at the advertising subjects they have tagged me with, guess what?

It was worse than I was afraid of. I loaded the tags into a spreadsheet, and categorized them into three groups:

  • Interests I definitely have. Example: “Computers and software high spender”. Guilty as charged.

  • Interests I definitely do not have. Example: “March Madness Basketball Fan”. What? Never watched basketball in my life. I don’t actually know what “March Madness” might even be and I’m disinclined to look it up.

  • Interests that I might or might not have, Meh so to speak. Example: “Vitamin C category purchasers”. Maybe I bought some one day. I don’t remember.

How do you think these categories break down? The majority (30/66, almost half) of tags Spotify has about me is in the Meh category. Will I buy more Vitamin C if they advertise it to me? Maybe, but quite unlikely. Consider the ad spend money in this category mostly wasted on me.

But this is the kicker: 24 of the remaining tags were “definitely not” and only 12 were “definitely yes”. Twice as many categories about me were absolutely wrong as were correct!!

Only 18% of the total categories were clearly correct, and worth spending ad money on to target me.

Eighteen.

From the name of the tags in the Spotify export, I guess most of them were purchased from third parties. (Makes sense: how would Spotify know I’m interested in Vitamin C, or not?) In other words, 18% of the data they purchased about me was correct, 36% incorrect, and the rest more or less random. No wonder Hwang immediately thinks of junk mortgage bonds with numbers like these.

But as he points out, advertisers keep spending money, however. Why? I suggest the answer is very simple: because of a lack of alternatives.

If you stop advertising on-line, what are you going to do instead? As long as there isn’t a better alternative, it’s a better plan to pinch your nose and go to your CEO and say, yes, I know that today, not just half but a full 82% of our advertising money is wasted, but it’s better to waste all that money than not to advertise at all. I can understand that. Terrible, but reality.

So, for me, the more interesting question is: “How can we do better?” And I think the times are getting ripe for doing something better… stay tuned :-)