Post-IDFA UA for mobile games: what changed, and what to do about it

What is IDFA?

IDFA stands for identifier for advertisers: a random device identifier that Apple assigns to every iPhone and iPad. From its introduction in 2012 it was the key that made iOS advertising measurable, because an ad click and a later install could be matched on the same device ID. Advertisers used it to attribute installs to campaigns, to build retargeting audiences from past behaviour, and to find high-value players by looking for devices that behaved like ones that had already paid. It was, for mobile games, the equivalent of the tracking cookie.

The history from limit-ad-tracking to App Tracking Transparency, and the mechanics of what replaced it, are on our guide to SKAN and IDFA. This page is about the operating changes that follow.

What changes are happening to IDFA with iOS 14.5?

Since iOS 14.5 in April 2021, the IDFA is no longer available by default. An app must show Apple's App Tracking Transparency prompt, and only a user who taps "allow" exposes the identifier; for everyone else there is nothing on the install side to match against the click. Consent went from implicit with an opt-out to explicit opt-in, and the device-level match that iOS attribution was built on became the exception rather than the rule.

The replacement is SKAdNetwork, Apple's attribution framework, and its successor AdAttributionKit. Apple itself attributes the install to a campaign and sends a postback that carries a coarse conversion value, delayed, and subject to privacy thresholds that null the value when a campaign is too small to keep users anonymous. SKAdNetwork 4 added multiple postback windows and crowd-anonymity tiers; AdAttributionKit adds re-engagement attribution. What none of them return is a user: no device ID, no behaviour, no join between an install and what that player later spent.

The impact and consequences of IDFA changes

For users who do not opt in, five things that iOS UA relied on went at once: user-level attribution; retargeting audiences built from past actions; behavioural and lookalike targeting built from device-level signals; campaign reporting below the privacy thresholds, so small geo or creative splits return nulls; and the user-level join between an install's source and its lifetime value, which is the join every LTV model used to sit on. Publishers were left with less data and coarser data at the same time.

The market did not stop buying installs. Adjust's paid-to-organic ratio for games reached 3.33 in 2025, more than three paid installs for every organic one, and the global gaming CPI rose 30% to $0.56 in the same year (Mobile app trends 2026). Where the change shows is in the gap between the two platforms. Liftoff's 2025 Casual Gaming Apps Report (with Singular; data Feb 2024 to Feb 2025) measured casual installs at $1.41 on iOS against $0.14 on Android, and D30 ROAS at 47% against 15%, so the per-dollar comparison is closer than the headline numbers suggest; Singular's Q4 2025 report puts Android game CPIs roughly three to four times below iOS as a rule of thumb. iOS UA still works. The operation that used to make it work does not, and that is the subject of the rest of this page.

The post-IDFA playbook: six operating changes

Each change below replaces something the IDFA used to do for free. None of them needs a new tool; all of them need the team to stop asking user-level questions of aggregated data.

1. Measure on cohorts, not users

The unit of measurement on iOS is now the cohort: the installs a campaign delivered on a day, read together. That starts with the conversion-value schema in your mobile measurement partner, which decides what the coarse postback can tell you; design it around the metric you pay back on (an early purchase, a purchase count, a revenue band) rather than the default event list. Then read ROAS by cohort day, D3, D7, D14, D30 and onward, and fill the gaps between postbacks with modelled predictions rather than waiting for actuals that arrive weeks late. Our predictive ROAS page covers the ladder and the models.

2. Restructure campaigns for the thresholds

SKAN's privacy thresholds null the conversion value on campaigns too small to keep users anonymous, so the pre-2021 habit of splitting iOS spend into dozens of small geo and creative campaigns now produces dozens of blank postbacks. Consolidate: fewer, larger iOS campaigns that clear the thresholds, with the splits you still need moved into the conversion value where they fit or run on Android where the attribution is deterministic. The granularity you give up on iOS is granularity you were no longer getting anyway.

3. Let creative do the targeting

Behavioural and lookalike targeting selected the audience before the ad ran. Without device-level signals, the creative selects it: the concept and hook decide who taps, and the network's value-based optimisation learns from the conversions you send back. Treat creative tests as the audience tests they now are, read them at the ad level, and keep the pipeline moving. Video took 53.7% of worldwide game ad impressions in 2025 and playables more than doubled to 13.3% while static images fell to 33.0% (Sensor Tower, State of Mobile 2026), which is the mix an iOS plan has to be able to produce. Our A/B testing guide covers the method.

4. Budget for delay, and pay back at a window

With actuals arriving late and coarse, the budget decision has to be made on predictions against an explicit payback window: day 30 to day 90 for casual and hybrid-casual titles, day 180 to day 360 for mid-core and strategy. Hold scaling until a cohort's predicted ROAS clears the window net of fees, scale in steps and watch CPI as you do, and keep a testing reserve so the plan can keep learning. How to set the window and the target is on the predictive ROAS page; how it fits the wider budget is in our mobile game marketing strategy guide.

5. Use Android as the calibration set

Android did not go through the same change. Google retired its Privacy Sandbox initiative in October 2025 without deprecating the GAID, so where the user has consented, Android still attributes deterministically and still joins an install to its lifetime value. Run the same creative and geo on both platforms, read the deterministic curve on Android, and use it to calibrate the iOS predictions for the same cohorts. Android is also, on the figures above, often the cheaper buy per dollar returned, which is a second reason not to treat it as the afterthought it was when iOS was fully measurable.

6. Own the data you can still get

What survives is worth keeping in one place: the MMP's raw event export, store and revenue data, ad-monetisation revenue by cohort, network cost, and the deterministic sample of users who did opt in, which is a check on the models rather than a population to plan on. Land all of it in a warehouse you own, on one set of definitions, so that two consoles that both say "ROAS" stop disagreeing. SuperPlatform does exactly this: 30+ no-SDK integrations into one gaming-native source of truth, deployed in your own BigQuery. You own the data.

How SuperScale can help you get past post-IDFA challenges

The six changes above are how we run UA for the games we manage. SuperPlatform is the data layer: your MMP, every ad network, the app stores and your warehouse unified in your own BigQuery. SuperVYZR is the app on top of it, where cohort- and creative-level ROAS predictions replace the user-level reports that ATT took away.

If you would rather have the campaigns run for you, SuperMedia gives fully-managed access to 35+ ad networks under a single insertion order, and onboarding is a tracking link, not an integration project. If you license the SuperScale Stack, UA Management adds our team to run it. Both add operating capacity without adding headcount. One platform: licensed or managed; the rate card is on /pricing.