Nota bene: The arbitrage that won the 2016 election

Last year, the amount of money spent on digital advertising in the US surpassed the amount spent on TV advertising for the first time. It was an important milestone, and not just in terms of where America’s ad dollars get spent. There are basically two types of advertising, and for the entire post-war era, brand advertising has held the cultural and commercial upper hand. Those days, however, have no come to an end. For the first time in living memory, the heart of the advertising industry is to be found in the world of direct marketing -- or, as it used to be known, junk mail.

The implications of this seismic shift are only now beginning to be understood. Direct marketing has always been home to some of the sleaziest and most unsavory characters in the advertising world, but never until now has it had major geopolitical implications. In its digital form, especially, it’s still highly immature, prone to excesses and glaring inefficiencies. By exploiting one of those inefficiencies, Donald Trump was elected president. And if the inefficiency isn’t fixed before the next big election, the national discourse is going to become even more poisonous and unedifying than it was in 2016.

The arbitrage was recently revealed by Antonio García Martínez, the first product manager for Facebook’s Custom Audiences. Underlying the trade is an inversion of what used to be considered a timeless, universal truth: that direct marketing would always cost more than brand advertising, on a per-person-reached basis. That wasn’t true, in 2016. The result was that Donald Trump got millions and possibly billions of dollars’ worth of brand advertising from Facebook for free, while Hillary Clinton was largely left out in the cold.

Most people intuitively understand the difference between brand advertising and direct marketing. Brand advertising is the Tide spot at the Superbowl, or the fashion billboard on the side of the road, or luxury-car ads in a glossy magazine. The idea behind brand advertising isn’t to get you to go out and buy something; instead, it’s top-of-the-funnel stuff, trying to make sure that the next time you want to buy a toothpaste, or a cola, or bottle of perfume, you’ll think of, and be well disposed towards, Brand X, rather than the competition.

Direct marketing, by contrast, is bottom-of-the-funnel stuff. You get a coupon in the mail, and you use it; you see an ad in your Instagram feed, and you click on it. The advertiser can track those clicks and calculate what they generate, in terms of extra sales, which makes selling direct-response ads online one of the easiest jobs in the world. If the average click makes an advertiser $3, then they’ll buy as many clicks as they can at $1 apiece.

That’s why Facebook and Google now sell more ads than any other companies on the planet, with annual revenues of $41 billion and $110 billion, respectively, in 2017. (Not all of those revenues come from selling advertising, but most of them do.) Advertisers flock to those companies because they sell a product which is quantifiable: No longer do they need to worry that half the money they spend on advertising is wasted. Instead, every dollar spent has an associated return on investment; it can be targeted and calculated and optimized to as many decimal places as you wish.

None of that comes cheap. At Google, if you want an ad to appear at the top of a search for “business services” or “bail bonds”, you’ll have to pay more than $50 per click you get. Similarly, the free Vince gift card I just received in the mail, complete with glossy packaging, probably cost Vince (the brand) and Amex (the company mailing out the card) a couple of bucks per mailing at least. Compare those figures to the world’s most expensive brand advertising, a $5 million, 30-second Superbowl slot. That works out to just about 5 cents per person, if it reaches 100 million people.

Of course, direct-response ads are generally crafted to make the brand look good, which means that a certain amount of value accrues to the brand just by dint of people looking at the ad. But there’s not a lot of straight-up brand advertising on Facebook or Google. That’s because they’re not prestigious sites, the content surrounding the ad is impossible to control, and the way the ad itself is seen is similarly unpredictable, changing according to phone, browser, operating system, and hundreds of other fickle variables. The result is that Facebook, especially, at least outside its Instagram subsidiary, is happy to effectively give away brand-advertising benefits for free, whenever you buy a direct-response ad.

Which created the perfect way for Donald Trump to hack the Facebook algorithm.

All modern presidential elections ultimately come down to branding, and presidential campaigns spend hundreds of millions of dollars on brand advertising. Every lawn sign, every bumper sticker, every #MAGA hashtag is part of trying to get your candidate’s brand out in front of as many people as possible, as often as possible.

But as far as Facebook was concerned, this branding exercise would still get bought and sold using its direct-marketing tools. Which means that instead of paying per impression, the Trump campaign would pay per click, or like, or comment, or share.

The result was an enormous incentive to buy ads featuring outrageous (and quite possibly totally untrue) clickbait, since those ads would get the most engagement. Normal brands -- brands selling stuff for money to consumers -- tend not to go down this road, since it makes them look skeevy and tends to ruin their reputation. So Facebook never felt the need to protect itself against this tactic.

In the political context, however, it worked wonders. If Trump and Clinton both bought an ad at a rate of $1 per click, and the Trump ad got clicked on ten times more often than the Clinton ad, then the Trump ad would make ten times as much money for Facebook than the Clinton ad. The result was that Facebook would have no real incentive to run a Clinton ad unless she paid $10 per click. As Martinez says, “Clinton was paying Manhattan prices for the square footage on your smartphone’s screen, while Trump was paying Detroit prices.”

What did that mean in practice? According to Brad Parscale, who worked as an advisor to the Trump campaign, his candidate often found himself paying less than 1% of what Clinton was paying per impression. Trump’s CPMs -- the amount paid per 1,000 impressions -- were, he says, sometimes as low as “pennies”.

Facebook disputes this, and has put out a spiky chart which purports to show that Trump paid higher CPMs than Clinton. But almost no one understands what the chart is actually showing, and Facebook seems disinclined to provide details. One big question: What is the “M” in CPM? Is it just impressions that the campaigns paid for, or does it also include organic impressions when those ads were shared?

Either way, it seems clear that the more incendiary Trump’s ads became, the less he had to pay to run them, and the more of an audience they could reach. It’s hard to imagine an incentive structure more damaging to democracy and the substantive nationwide debate that presidential elections are supposed to encourage.

The good news is that it’s pretty easy to fix this problem. Facebook could simply insist that political campaigns buy ads on a CPM basis, with the M measuring actually-seen impressions. At the very least, it could insist on a CPM-based floor: a minimum price paid per impression no matter how inflammatory a campaign is. This is urgent. Facebook’s ad-pricing algorithms have failed American democracy once. They cannot do so again.