The CPM math doesn’t work

One theory being circulated related to Apple’s soon-to-be-implemented ATT privacy policy is that the CPM compression caused by degraded ad targeting efficiency will allow many advertisers to spend more money, given that ad inventory will become cheaper. The second-order effect of this being that competition for impressions will increase as these advertisers scale ad spend, and the supply / demand equilibrium for impressions will settle at the pre-ATT CPM price. In essence, ATT will change the composition of ad spend, but the volume of dollars being spent won’t decline.

This logic is flawed. To understand why, it’s important to discern how pricing works in digital advertising. This article by Facebook provides helpful detail on the topic, and so does this QuantMar thread. But conceiving of CPM price as a determination rather than a measurement is the culprit of the faulty logic described above: CPM prices describe the competitive demand for impressions. Advertising is ultimately served according to an auction, and on Facebook, that auction is a second-price (Vickrey) auction. I describe a generalized second-price auction in Freemium Economics with:

The purpose of the GSP, which is a form of what is known as a Vickrey auction, is to give participants the incentive to place bids based on what the object’s value is to them. Since no participant knows the bids of the other participants, they cannot game the auction system by strategically outbidding the others.

I also describe the second-price auction, especially in comparison to a first-price auction, in What is a first-price auction?. Essentially, a Vickrey auction allows the advertiser to bid the amount of money for an impression that they believe the impression will deliver to them, because the price they’ll pay if they win the auction is less than that.

The details related to second-price auctions, which are unpacked in detail in the above-linked article, aren’t important here except to say: direct response advertisers are already bidding as much for Facebook inventory as they think it is worth to them. What ATT will do is effectively make it impossible for some advertisers to bid anything. As I describe in IDFA deprecation: winners and losers, there are certain categories of products that wouldn’t exist but for Facebook’s (and Google’s) targeting precision: without being able to target users on the basis of behavioral monetization, they simply wouldn’t be able to scale their businesses. I describe this idea in Is it healthy for start-ups to spend so much money on user acquisition?

The price paid by an advertiser for a block of impressions is not the only constituent metric in the unit economics of digital products. Revenue is the numerator in the ROI calculation: an influx of cheap traffic doesn’t help a company if that traffic is unqualified and doesn’t monetize. In How does IDFA deprecation impact ad prices?, I illustrate the relationship between ad targeting efficiency and product monetization:

The idea here is that if top-of-funnel marketing metrics decrease even slightly, the CPM decrease needed to compensate for the lack of conversion is disproportionately large. CTR and IR don’t exist in isolation: if ads are less personalized, not only will fewer users survive the marketing funnel, but those that do will likely monetize to a lesser degree as a result of the loss of precision in targeting.

The consequence of this dynamic is that, as ad targeting becomes less precise with the imposition of ATT, some portion of ad spend will simply evaporate. This is why CPMs will decrease: inventory will become less competitive because it will be targeted more broadly, and so it will be worth less. Some advertisers will cut spending as a result.

(Note that in the above diagram, I use the word “bid” somewhat misleadingly. The bid for a conversion outcome — which could be an impression being served, or an install, or some down-funnel event like making a purchase — is just one input that Facebook uses to calculate the expected value of serving that ad, and it is ultimately expected values that are used to rank ads in the auction. More info here and in the QuantMar thread linked above.)

The theory I describe at the onset of this article posits that some subset of advertisers will increase spend when CPMs decrease, causing an increase in CPM prices. But this CPM math doesn’t work. Because of the nature of a second-price auction, advertisers are already bidding for traffic at a level that is economical to them. Assuming that the degradation of granular targeting precision doesn’t cause a cascading weakening of down-funnel metrics for some products that have broad, universal appeal — and there are products for which this is certainly true — those products can’t simply summon better product monetization into existence to spend on advertising.

Imagine a case where the four advertisers above all experience a perfect, 100% rate of conversion from an impression: their bid on an impression is the exact amount of revenue they’ll generate if their ad is shown to the user. In the pre-ATT scenario, the winning bidder pays $7.01 in the individual auction above, and in the post-ATT scenario, the winning bidder pays $1.01.

The below diagram considers what happens if that same auction happens 1,000 times without the underlying conversion dynamics changing: in the pre-ATT scenario, the winning bidder wins each auction and pays $7,010 for those 1,000 conversions. In the post-ATT scenario, the new winning bidder pays $1,010 for those 1,000 conversions.

So while the new winning bidder ends up spending more on advertising in the post-ATT scenario simply because they are now winning auctions they don’t win in the pre-ATT scenario, overall ad spend for the same 1,000 auctions decreases from $7,010 to $1,010.

Put another way: CPM is just the amount of money spent by advertisers divided by the number of impressions served (divided by 1000). The number of impressions that Facebook or any other ad network serves is mostly stable in the short term. Some of the companies bidding the most for ad impressions now will remove a portion of their spend from the ecosystem, but the advertisers that are left can’t increase their current bids — by definition they can’t, because if they could, they’d be winning auctions now.

And so while the advertisers left in the post-ATT environment will spend more than they are now, overall ad spend will shrink without the participation of advertisers that can bid aggressively for precisely-targeted traffic.

Photo by Clayton Robbins on Unsplash