Programmatic iOS Retargeting: A Guide to ATT, IDFA & Probabilistic Matching

iOS retargeting changed dramatically with the introduction of Apple’s App Tracking Transparency (ATT) framework. While ATT reduced access to device identifiers for many users, it didn’t eliminate mobile retargeting. Instead, it shifted the industry toward a combination of deterministic and probabilistic approaches.

In this article

1. What Changed After iOS 14.5
2. How Does Probabilistic Retargeting Work on iOS?
3. AdAttributionKit: Apple’s Native Re-engagement Attribution
4. Creative Best Practices in iOS Retargeting
5. FAQ

What Changed After iOS 14.5

Before iOS 14.5, advertisers could rely on Apple’s Identifier for Advertisers (IDFA) to recognize returning users across apps. ATT changed that by requiring users to explicitly opt in to tracking. Users who decline permission no longer make their IDFA available, limiting traditional deterministic retargeting.

Roughly 30% of users initially granted tracking permission. Opt-in rates have climbed since: Adjust’s 2025 data shows an industry average of ~35%, with wide variance by vertical (sports games ~50%, hyper-casual ~43%) and geography. Everyone else requires privacy-first approaches based on probabilistic matching.

How Does Probabilistic Retargeting Work on iOS?

This created two distinct retargeting environments. For users who opt in to ATT, deterministic matching continues to work much as it did before, enabling precise user recognition and highly personalized campaigns.

For the opt-out majority, programmatic re-engagement relies on probabilistic matching. Rather than identifying a single user directly, DSPs analyze contextual signals – including publisher characteristics, available device information, app category, and behavioral patterns – to estimate the likelihood that an ad opportunity belongs to a particular audience. This is modelled bidding on contextual and behavioral signals from the ad opportunity, not identifying or tracking a device. This approach shifted the focus from individual profiling to granular behavioral context.

Probabilistic Matching: MMPs vs. DSPs

Both Mobile Measurement Partners (MMPs) and DSPs use probabilistic techniques, but they work with different data sets. MMPs typically rely on limited attribution signals such as IP address and user agent, which have become less detailed on modern iOS devices.

DSPs, by contrast, operate inside the real-time programmatic auction and can analyze additional contextual signals available during the bidding process. Combining these signals with machine learning helps improve confidence when estimating whether an ad opportunity belongs to a desired audience.

Because probabilistic matching is based on statistical confidence rather than deterministic identifiers, DSPs continuously evaluate signal quality. If confidence falls below an acceptable threshold, the platform simply chooses not to bid.

AdAttributionKit: Apple’s Native Re-engagement Attribution

SKAdNetwork, Apple’s original privacy-preserving attribution framework, only ever measured installs – re-engagement was invisible to it. AdAttributionKit (AAK), introduced in iOS 17.4 and expanded through iOS 18.4, changes that: ad networks can now flag impressions as eligible-for-re-engagement, and when a user who already has the app installed taps an ad and returns, Apple itself sends a signed re-engagement postback – no IDFA required.

This matters for three reasons. First, it gives advertisers a deterministic, Apple-endorsed measurement path for iOS retargeting that works for opted-out users, complementing MMP probabilistic methods. Second, features like overlapping conversion windows and configurable attribution cooldowns (added in iOS 18.4) let marketers separate UA and re-engagement credit – reducing the cannibalization blind spots discussed in the Measurement entry. Third, it signals Apple’s long-term direction: retargeting is a recognized, supported activity in the privacy-first iOS ecosystem, not a workaround. For deeper technical comparisons, see Adjust’s AAK explainer and Singular’s analysis.

Adoption is still ramping – AAK postbacks are aggregate and delayed like SKAN’s, and don’t replace user-level MMP reporting – but any team running iOS retargeting in 2026 should have AAK re-engagement postbacks configured alongside their MMP setup.

“In Singular, we see that AdAttributionKit can provide good deterministic indicators of re-engagement volumes, so we recommend enabling it in any media channel that supports it. However, since it has only limited granularity, and delayed results, it’s important to use side-by-side with other methods such as deeplink-based attribution, Advertising IDs where possible and probabilistic/modeled signals.”

Eran Friedman
CTO & Co-Founder, Singular

Creative Best Practices in iOS Retargeting

Creative strategy also changes in a privacy-first environment. Because many iOS users have explicitly declined tracking, overly personalized messaging can feel intrusive and reduce campaign effectiveness.
Instead, focus on value-driven creatives that highlight new features, content updates, seasonal events, or limited-time offers. The goal is to remind users why they enjoyed the app – not to make them feel like they’re being individually tracked.

“The most effective retargeting creative understands the reason why someone has lapsed
(or has yet to complete a desired action) and gives them a reason to return that really
resonates, almost becoming unavoidable.

Many advertisers make retargeting assets in all creative dimensions, however, the smaller
the asset dimensions, the harder it becomes to create a connection with the intended
audience. Focusing on ‘larger’ creative (ad unit to screen size ratio) will be a stronger bridge
back into your app.

In terms of prospecting creative, from a design perspective, this means strong visual hooks,
clear hierarchy and simple storytelling. The creative needs to communicate quickly, be
relevant and give the user a clear reason to pay attention.”

Rob Maclean-Eltham
Senior Account Manager (UA), Yodel Mobile