FAQ: Impact of Apple (iOS) Limit Ad Tracking on attribution

Introduction

The purpose of iOS Limit Ad Tracking (LAT)  is to improve user privacy.

Before iOS 10, when users actively selected LAT, the OS would send a “flag” indicating the user’s wish. However, Apple’s IDFA identifier was still there and not all companies honored the requests (this is still the case in Android). Apple also allowed companies to use the ID for “frequency capping, attribution, conversion events, estimating the number of unique users, advertising fraud detection, and debugging.”

In iOS 10 Apple decided to toughen up its LAT stance, sending a string of zeroes in place of the user’s IDFA when this feature was activated. Therefore for iOS 10+, LAT users’ IDFA can not be used for any purpose, including attribution.

More than 99% of iOS users have iOS 10 or above, out of which 25% of users have LAT activated. Therefore, this article refers to 1 out of 4 iOS users, who are LAT users.

LAT_distribution_2019-2020_en-us.png.png

As the chart shows, based on AppsFlyer analysis, the percentage of iOS LAT users has climbed from 16% at the start of 2019 to 25% in the middle of 2020. We assume the rate of this climb to decrease with the introduction of ATT in iOS 14.

The effect of LAT on attribution

To understand the real effect of LAT on attribution and media sources let's look at three main cases:

  1. Media sources that can be attributed with Probabilistic modeling
  2. Media sources that don't support Probabilistic modeling attribution
  3. Apple Search Ads

1. Measuring installs with Probabilistic modeling

99% of media sources and 100% of custom attribution links for owned media use attribution links. LAT obviously affects attribution companies’ ability to record installs using ID matching. The loss of 25% of attributed installs could be very troubling for app owners using these attribution links.

Luckily, a fallback mechanism for ID matching can be used - Probabilistic modeling attribution. It uses publicly available parameters (i.e. device name, device type, OS version, platform, IP address, carrier, to name just a few), to form a digital probabilistic ID that statistically matches specific device attributes. As such, Probabilistic modeling is less accurate than ID matching, and therefore the attribution window is short, usually 24 hours. In mobile app install campaigns, the vast majority of the first app opens occur within 1-2 hours, in which case the Probabilistic modeling method is extremely accurate. However, some loss of attributed installs is unavoidable.

Assuming 25% LAT and 90% Probabilistic modeling accuracy, the best case entails a loss of 2.5% attributions, so the effect on these media sources is not dramatic.

Advertisers actually gain somewhat from this.

When advertisers run on a cost per install (CPI) pricing model, and not on exposure (CPM) or clicks (CPC), they receive ~2.5% free organic installs!

2. Measuring installs without Probabilistic modeling

What happens when a LAT user clicks on an ad served by a media source that relies solely on ID matching - among them top sources such as Facebook Ads, Google Ads, and Snapchat?

When that user installs and launches the app, the mobile attribution company’s SDK cannot pull its IDFA, and so the media source is unable to perform ID matching and identify the user as their own. This results in a failed attribution and a free organic user for the app owner.

This is true for CPI and CPA campaigns only, as media sources need to carve out their installs quota out of the 75% non-LAT users only, but still may serve ads to the 25% LAT users for free. This is a potential increase of about 33% (100/75) free installs for advertisers.

It makes sense that these giants of mobile app advertising would soon reach a solution to minimize the effects of LAT and ATT on them. For example, such a solution could be to simply stop serving LAT users with mobile app install ads, or by implementing some Probabilistic modeling API.

In conclusion, on CPI or CPA campaigns, media sources that do not support Probabilistic modeling may contribute up to 33% free installs to app owners.

3. Apple Search Ads

Apple Search Ads is a powerful source for paid installs for iOS app owners. In contrast with all other mobile media companies, Apple knows exactly who its mobile users are, even if they’re LAT users since it can easily use the iTunes account ID.

Apple Search Ads' attribution is therefore unharmed by LAT users, while AppsFlyer attributes them as organic users (or to another engagement, if it occurred). However, audience-targeted Apple Search ads campaigns don’t serve ads to LAT users, so the problem is confined to non-targeted campaigns only.

In conclusion, non-targeted campaigns on Apple Search Ads may bring a high percentage of non-attributed users.

 Note

Apple users aged 18 and younger are always LAT.

Summary

The surprising conclusion from this analysis is that the introduction of LAT on iOS 10 has actually increased the percentage of non-organic installs app owners receive! In AppsFlyer data, the added installs are unattributed and appear as organic installs.

Please note the following important tips:

  • The app owner MUST buy CPI (or CPA) campaigns to enjoy this increase; CPM and CPC campaigns move the beneficial effect from the app owner to the media companies
  • Media sources that do not use Probabilistic modeling (mainly SRNs) may contribute up to 33% free "organic" users, but don't count on it.
  • Targeting in Apple Search Ads installs campaigns is very important as, without it, many paid users are not attributed.

 Caution

As demonstrated, although Apple’s LAT is good for app owners, it has the potential to harm ad networks’ bottom line by forcing them to expose mobile app ads to more users than previously needed. The good news is that when mobile installs are recorded by an attribution company with a good Probabilistic modeling solution, the effect of LAT is rather marginal.

 Note

Android has a similar limitation called "Opting out of ad personalization". However, currently, the LAT percentage in Android devices is marginal, with less than 2% of users.

Was this article helpful?