App Tracking Transparency (ATT), the privacy framework that will imminently be enforced on iOS, is the most sweeping and tectonic cause célèbre that the mobile ecosystem has faced. This policy change was not unexpected — I predicted in February 2020 that Apple would announce the deprecation of the IDFA at this year’s WWDC. But most large ad platforms were seemingly caught flat-footed by ATT, and Facebook, perhaps more than any other large digital advertising platform, has been scrutinized exhaustively regarding the potential revenue impact delivered to it by ATT. This scrutiny is a reaction to the prodigious success that Facebook has achieved — and the immense value it has delivered to advertisers — by leveraging its vast wealth of user behavioral data through event-optimized and value-optimized products for both mobile app and web advertising campaigns.
That Facebook is portrayed as the exclusive casualty of Apple’s privacy changes is misguided, since Google and other ad platforms face nearly equivalent exposure, as I discuss in the Twitter thread below.
Nonetheless, these changes are problematic for Facebook.
In this post, I attempt to quantify, with analytical support, the impact on Facebook’s revenue of ATT. Below is the model that I use to evaluate Facebook’s revenue loss, broken out into three cases: Best Case, Base Case, and Worst Case. Note that the revenue impact from ATT is projected to begin in Q2 2021 and is modeled through Q1 2022.
Blended mobile loss of efficiency from pre-IDFA
Facebook has stated that it believes its ability to personalize targeting on the Facebook Audience Network (FAN) contributes to about 50% of that ad channel’s CPM prices. This number is really a proxy for efficiency: the degree to which Facebook can pair a user with the most relevant possible ad, given what Facebook knows about the user’s historical ads interaction, in-product engagement, and in-product monetization tendencies.
It’s important here to acknowledge the data that Facebook uses to target ads to users. This QuantMar thread provides a detailed explanation of how Facebook’s ad targeting mechanic works, but in short: Facebook’s SDK is present in almost all apps, and its tracking pixel is present in almost all e-commerce websites. When a Facebook user clicks on an ad in the Facebook Blue app or in Instagram, that user’s journey on the advertised property is observed by Facebook: events documenting the actions the user takes on that property are transmitted back to Facebook and indexed against that user’s Facebook ID. This is what’s known as an “events stream.” This events stream provides Facebook with very valuable insight into the degree to which any user monetizes or engages with various third-party properties, and it helps Facebook to target ads to those people on the basis of their recorded preferences.
The events stream for both app and web advertisers is being broken with the ATT prompt for users that opt out of “tracking.” This means that Facebook will no longer have full transparency into what users do in apps or on websites once they click on an ad in Facebook Blue or in Instagram: instead, Facebook will receive a very limited amount of interaction data from the advertised property. More background on SKAdNetwork, Apple’s app-specific attribution mechanism that delivers in-app context from app ad clicks to advertising networks, can be found here. And this article explains the logic by which events can be sent back to Facebook from mobile web campaigns.
If Facebook believes that its event stream and IDFA-indexed off-property behavioral profiles are worth 50% of CPM prices in FAN, what is the events stream worth to Facebook Blue and Instagram? Can Facebook recover some of that efficiency loss by investing in better contextual targeting mechanisms? Note that I group app ads and mobile web ads together in the model since they are both impacted by the loss of direct attribution incurred via ATT.
In the model, I have laid out what I believe to be the Best Case, Base Case, and Worst-Case scenarios for both ads efficiency degradation and Facebook’s ability to recover that efficiency in the quarters after Q2 2021, which is roughly when ATT is rumored to become mandatory. Even in the Worst Case scenario, I don’t believe that Facebook sees a 50% loss of efficiency for its Blue App and Instagram; Facebook has other proprietary first-party data that it can use to ameliorate the loss of off-platform data. But the consequences of the loss of this data are still substantial, and even in a best-case scenario, I believe ads efficiency decreases by a double-digit percentage.
Global Average Opt-in Rate
One misconception about the ATT opt-in rate is that any given app’s opt-in rate exclusively governs that app’s ability to serve targeted ads. But it’s important to note that any successfully-served ad requires the participation of two parties: the publisher (who shows the impression) and the advertiser (who buys the impression). If a publisher receives ATT opt-in from a user and is thus able to provide the IDFA for an impression to serve to that user, only advertisers that also have received permission from that user for access to their IDFA will be able to target them.
For mobile app ads, the events stream is dependent on both Facebook’s and the advertiser’s access to a user’s IDFA, which is governed by ATT opt-in. In other words: if Facebook does receive opt-in but any given advertiser’s app doesn’t, Facebook will not receive an events stream for that advertiser’s app and thus its visibility into what users do in that app will be totally restricted.
In speaking to developers while conducting research for this article, I was given a wide range of opt-in rates that various apps have experienced in ongoing ATT trials: from a low of 5% to a high of 50%. But again, for mobile app advertising, Facebook’s access to the events stream is dependent on receiving opt-in for both its own app and for a particular advertiser’s app from any given user. For that reason, and because I believe Facebook is generally viewed with a jaundiced eye by the public with respect to user privacy, I expect the global average Facebook opt-in rate to be fairly low: from 30% in a best-case scenario to 10% in a worst-case scenario.
Note that Bumble, the ubiquitous dating app that frequently occupies a Top 10 Grossing position on iOS in the US, stated in its recently-filed S-1 that it expects its ATT opt-in rate to be no more than 20%. While this is just one benchmark, I think it serves as a useful comparison to Facebook and sets a realistic standard for what popular apps might expect for ATT opt-in.
iOS Advertising Revenue Share
To approximate the percentage of Facebook’s revenue represented by iOS, I used SensorTower’s 2019 mobile consumer spending report, which estimates 2019 iOS app revenues of $54.2BN and Android app revenues of $29.3BN, giving iOS a 65% share of app spending. I used consumer spend as a proxy for ad spend in the model.
Dispelling myths and misconceptions about ATT
The assumptions above are plainly conjecture. That said, misapprehensions about and misinterpretations of Apple’s ATT guidelines abound. I attempt to clarify what I see as four of the most common distortions of both Apple’s impending iOS privacy changes and Facebook’s ability to maneuver around them.
Facebook’s ad targeting capabilities are primarily driven by first-party data collected from its Owned and Operated channels.
The data that carries the most signal in determining if a user is a good candidate to be exposed to a given ad is that user’s recency of interaction with other such ads, and the degree to which that user engages and monetizes in products similar to what the ad represents. The data that Facebook collects from the usage of its Owned and Operated properties does not contribute to its knowledge of the ad interaction, engagement, or monetization proclivities of users.
The idea that Facebook’s O&O data contributes the majority of insight into ad targeting can be dismissed out of hand because Facebook shut down its Audience Network for the mobile web in response to Safari blocking third-party cookies (and Google announcing that it would do so), and because Facebook announced that it might have to shut down its Audience Network for iOS traffic when Apple announced the deprecation of the IDFA. If Facebook’s O&O properties supplied the lion’s share of data used in ad targeting, neither of these ecosystem shocks would impact performance on Audience Network for web or for mobile app campaigns.
Facebook’s ad targeting and optimization is driven primarily by the event stream it receives from apps and websites, and when that event stream is broken on iOS, ad targeting will be hindered considerably. That said: the data it collects from its O&O properties does provide some value, which is why even in the worst-case scenario above, ad efficiency doesn’t drop by the full 50% of price contribution that Facebook estimates can be attributed to CPMs on FAN.
Facebook has already collected all of the behavioral data it needs from users in order to capably target ads to them going forward.
Recency plays an incredibly important role in the signal parsed from the data identified above. Not only must an advertiser know that a user has historically engaged and monetized with products following an ad click, but they must know that a user has done so recently: the older the data is, the less helpful it is in targeting ads to that user. Facebook’s ad targeting models require a constant supply of data — the off-property events stream — in order to retain efficacy. Facebook’s ads targeting judgment will languish as its IDFA-indexed behavioral profiles are starved of new data and become stale.
Advertisers can collect first-party data from users in apps and on websites and send conversions back to Facebook using server-to-server methods, such as the Conversion API for e-commerce websites.
This notion is addressed in this article (Apple has stated clearly that emails can’t be used for ad-targeting), this article (fingerprinting is unlikely to be a viable tracking tool post-ATT), and in this Twitter thread. The broader idea is that Apple expects advertisers and developers to conform to the spirit of ATT, which is that any data collected for the purposes of ad tracking — not just the IDFA — is governed by the ATT consent opt-in. It seems unlikely that top tier ad networks will build mechanisms for defying this directive, and advertisers should be very careful with risking their ability to publish their apps to the App Store by exploiting potential loopholes in the ATT guidelines.
Mobile web ad campaigns (app-to-web) are far less impacted by ATT than mobile app campaigns (app-to-app).
The most important revelation to come out of the new guidance that Facebook issued to advertisers last month was that the ATT prompt would govern app-to-web campaigns. This had not previously been the prevailing understanding of ATT: most thought that ATT would only apply to app-to-app campaigns. But what was also uncovered in Facebook’s guidance around app-to-web campaigns was that Facebook would introduce a new measurement methodology called Aggregated Event Measurement which would aggregate advertising performance data at the level of the campaign.
Unpacking Facebook’s guidance further, the treatment they are applying to app-to-web campaigns seems to emulate the SKAdNetwork framework: a limited number of events are available to advertisers, of which only the highest value is transmitted in a postback, and the default attribution window has been reduced to 7-day click-through (from 28-day click-through). I walk through what these changes mean in this Twitter thread, but given the parity between web and app advertising which Facebook has ostensibly struck, it seems unlikely that performance will differ meaningfully between those two classes of campaign types.
One idea I’ve seen circulated regarding any drop in efficiency is that a decrease in CPM will simply activate the market mechanics of the auction: when CPMs drop because some advertisers reduce spend, other advertisers will enter or increase their activity in the market at these lower prices, driving CPMs back up. This may be true to some extent, especially since many advertisers will pull spend back at the onset of ATT simply because they need time to re-calibrate their ad spend ROI models given the fundamental change in targeting.
I have accommodated for this in the Efficiency and Price Recovery assumption, but a general pullback for the purposes of collecting data won’t be the only reason spending will decrease. The dynamics of targeted ads are such that some advertisers can only deploy ad budget when they are able to target specific user segments, and those advertisers may experience a sustained reduction in spend potential.
This group of advertisers — many app publishers, D2C brands, and eCommerce companies — might need to institute a permanent step-change reduction in ad spend when Facebook can no longer target relevant spenders on their behalf (I explore this idea in more detail in this post and in this analysis of which verticals suffer most from ATT).
In other words: a slight drop in ad efficiency will result in these businesses dramatically reducing ad spend. Note that eCommerce was the largest single advertising vertical for Facebook in Q3 2020:
The disproportionality of CPM prices to advertising activity based on ad targeting means that a slight decrease in ad efficiency (which creates a slight decrease in CPM prices) can cause a dramatic decrease in ad spend for certain businesses that rely on that targeting. The advertisers that are already targeting broadly can increase their ad spend commensurate with a CPM decrease, but the advertisers that require granular targeting in order to efficiently deploy budget will drastically reduce spend or leave the market altogether.
That said, the “bull case” for Facebook, even in light of the looming ATT framework (or perhaps because of it), seems to be a rallying cry championed by investors, ad tech executives, and advertisers alike on Twitter and elsewhere. I believe that a long-term bull case does exist for Facebook related to the thesis of my three-part series on the future of mobile content platforms. Facebook is likely to accelerate its strategy of subsuming more consumer interactions into its apps, such as with Instant Games, Instant Articles, Facebook Video, Facebook Shop, and Instagram Shopping, etc.
But the arc of that paradigm shift is long, and this adaptation will progress over a protracted timeline. Very little evidence exists, from my perspective, to support the notion that Facebook’s revenue will not be materially undermined by ATT in the short term. This article serves as an attempt to appraise that impact.
Disclaimer: the author holds long positions in Facebook