ATT opt-in rates are irrelevant

Mobile attribution company Appsflyer last week published optimistic data regarding App Tracking Transparency opt-in rates: the average opt-in rate the company witnessed across a subset of its clients’ apps was 41%, with the Utilities category seeing the highest average rate of opt-in at 45% and the Social Casino category sitting at the very low end of the range at 21%.

These opt-in rates are much higher than consensus estimates. A caveat on the data and methodology: Appsflyer witnessed more than 13MM ATT prompts in the observation period across 300 apps, so the sample is fairly limited. Also: the 41% opt-in number is global for all opt-in prompt exposures, while the per-app average is just 28%. Regardless, the metrics reported by Appsflyer are higher than what many developers expected: a poll I conducted on the Mobile Dev Memo Slack in June of last year revealed that most developers predicted a global opt-in rate of less than 20%.

Appsflyer isn’t the only company reporting observations of encouraging opt-in rates. But I believe that the industry generally misunderstands how opt-in rates will impact mobile advertising efficiency. There are a few factors that render opt-in rates — and especially the opt-in rate of any individual app — mostly insignificant in ameliorating the degradation of efficiency that ATT will visit on mobile advertising.

Firstly, the ATT opt-in metrics reported now are sampled from biased data: many developers are currently running ATT prompt tests to establish opt-in baselines and to optimize pre-prompt designs. It’s not clear whether a user is more (or less) inclined to opt into tracking when they see the ATT prompt for the first time, ever: what happens when the ATT prompt is ubiquitous and users grow accustomed to it — will users reflexively opt out when they begin seeing the ATT prompt regularly? There are several other countervailing forces of opt-in rate perception that muddy their interpretation:

  • Perhaps the companies experimenting with ATT opt-in rates now, ahead of ATT’s rollout, are the most conscientious and the most likely to see generally high opt-in rates as a result? This would imply a lower global average opt-in rate once the ATT opt-in prompt is made mandatory;
  • But perhaps the opposite is true: the apps dedicating the most energy to opt-in text and pre-prompt experimentation are those most worried about low opt-in rates, which would imply the opposite effect (note that Facebook began testing an ATT prompt in February).

I simply think it’s impossible to make sense of ATT opt-in rates as they are measured now, and I think it’s foolhardy to extrapolate those numbers to the post-ATT environment. But a more structural reality prevents a sanguine reading of ATT opt-in rates: because the process of serving an ad requires both a buyer (advertiser) and a seller (publisher), the use of the IDFA for measurement and targeting will be governed by the opt-in rates of both, but mainly constrained by publishers. since it is the publisher that anchors the IDFA to install attribution. And Facebook and Google are the most important publishers for many advertisers.

As I detailed in this article, the IDFA facilitates two things with respect to mobile advertising: targeting and measurement. Targeting is accomplished by ad platforms through using posted-back conversion events from a campaign to understand which users are most relevant for that advertiser (and, consequentially, for other advertisers). And measurement is accomplished by ad platforms and advertisers by attributing users to their source acquisition campaign (“stamping” a user with the ad campaign that delivered them to the product). Unless the publisher has access to a user’s IDFA when it exposes an ad to them, neither of those buckets of functionality will be feasible, regardless of whether the advertiser is able to capture that user’s IDFA.

The diagram below lays out the availability of the functionality that delivers measurement (user-level campaign attribution) and targeting (IDFA-indexed event postbacks) in the various opt-in scenarios. In order for the efficacy of measurement and targeting to be retained, both the publisher and advertiser must receive an opt-in from any given user (likewise: the advertiser must receive opt-in to provide user-level campaign attribution if it serves as a publisher).

I’ve seen this relationship — the need for opt-in for both publisher and advertiser — be described mathematically as the product of each party’s opt-in rate (ie. the global availability of IDFA is captured by Global Publisher Opt-In * Global Advertiser Opt-In). But this assumes publisher opt-in rates are independent of advertiser opt-in rates: that no context can be assumed from the fact that a user opted into any given app, especially an app like Facebook.

I think a better way to describe this relationship is found using Bayes’ Theorem, and this is where the irrelevance of any given app’s opt-in rate is underscored. Imagine an app, App X, with a global opt-in rate of 60% — quite impressive! But now imagine that this app only acquires users from Facebook, and Facebook’s global opt-in rate is low at 10%. Now we assume that 100% of users that opt into tracking on Facebook will also opt into tracking on App X.

It seems reasonable to assume that if Facebook passes the threshold for tracking for some user, almost every app will. According to Business Insider, Facebook experiences the lowest consumer trust of any other social network: if a user has no problem giving Facebook access to their data, they probably aren’t concerned with opting into tracking generally. Given these assumptions, we find that the conditional probability that a user has opted into Facebook given that they have opted into your app is just 17%.

The nature of ATT and the mechanics of advertising means that ads targeting and measurement are constrained by the opt-in rates of publishers, the biggest of which are apps that are likely to experience very low opt-in rates (note that web advertisers are wholly dependent on publisher opt-in rates as the ATT opt-in prompt is not exposed in the web browser). While developers should strive to improve their opt-in rates through pre-prompt experimentation and optimization, a high opt-in rate doesn’t guarantee a similarly high rate of IDFA availability. ATT opt-in rates are mostly irrelevant.

Photo by Nick Fewings on Unsplash