When Apple introduced App Tracking Transparency in 2021, the mobile advertising industry faced a crossroads: rebuild for iOS, or redirect budget somewhere easier.

Many players redirected.

Android retargeting was still deterministic. Rebuilding that signal strength for iOS meant starting from scratch — custom probabilistic models trained on non-PII signals, cohort logic that didn't map to any existing infrastructure, and a measurement framework that SKAdNetwork explicitly wasn't designed to support. 

The downstream effects of what that crossroads moment created is one of the least competitive programmatic auctions in mobile advertising right now for some of the highest-value users on the planet.

• Revenue per install: $2.28 on iOS vs. $0.89 on Android
• Spend per app: iOS users spend nearly 2.1x more than Android users
• Repeat purchases: iOS users are 35% more likely to buy again

This piece is about why most teams are still paying UA prices to reach their iOS lapsed users and how iOS probabilistic retargeting offers a dedicated, lower-cost pathway to the same audience.

The redownload problem hiding in your UA numbers

Before getting to the opportunity, there's something worth understanding about what iOS UA campaigns are actually doing, because the numbers are more compelling than most growth teams realize.

According to Apple's 2024 App Store Transparency Report, the App Store processes an average of 839 million new downloads per week and 1.9 billion redownloads per week. That means redownloads outnumber new downloads by more than 2 to 1.

This matters because it shows a meaningful share of what iOS UA campaigns report as "installs" are users returning to an app they previously had. Which means your UA budget is already funding reactivation. It's just doing it at UA CPMs, which are materially higher than what a dedicated retargeting pathway would cost for the same user.

Now, there is a structural issue here that I want to note, which is that the standard UA attribution process  doesn't cleanly separate first-time installs from redownloads. So the reactivation keeps happening, the budget keeps flowing, and the blended install number looks fine in the dashboard.

Why 65% of your iOS install base is unreachable by most DSPs

The deeper problem is what ATT did to the reachability of lapsed iOS users.

Adjust's Q2 2025 benchmark data puts the industry-wide ATT opt-in rate at 35%. That number has been climbing slowly — improved prompt design, better onboarding flows — but the ceiling is real. The majority of iOS users are not opted in, and they're not going to be.

What that means practically:

• Deterministic retargeting — targeting based on IDFA — reaches only the opted-in minority
• SKAdNetwork offers no retargeting support at all; it was built exclusively for first-install attribution
• Most DSPs are working with the opted-in 35%, competing for the same fraction of your audience

Make no mistake: the 65% of users who opted out aren't unreachable. They're just unreachable through the methods most of the market rebuilt around. Probabilistic retargeting — using non-PII signals like IP ranges, device model, user agent, and contextual behavior — is how you reach them. The challenge is that building it properly is technically expensive, which is why few players did.

Why the lapsed user pool keeps refilling

The other market reality working in favor of iOS retargeting is the sheer scale of app churn.

Industry benchmarks consistently approximate that 71% of app users churn within 90 days of installing. Some verticals are worse. Most apps lose the majority of their daily active users within the first week.

This means every cohort you've ever acquired is continuously generating lapsed users who:

• Already know your app
• Have demonstrated some intent
• Don't need to be convinced of your core value proposition
• Can be reactivated at a fraction of new-user acquisition cost

The lapsed pool replenishes constantly. And on iOS, most of it is sitting in an auction with less competition.

How probabilistic retargeting works

The mechanics are worth understanding briefly, because "probabilistic" sometimes gets treated as a black box or a compliance risk. It's neither.

Probabilistic retargeting builds user cohorts from signals that are:

• Non-PII — no personally identifiable information collected or stored
• Non-persistent — no device-level profiles that follow users across the web
• Apple-compliant — fully compatible with ATT's framework

The signals used include partial IP address ranges, device model and OS version, user agent strings, and contextual behavior patterns. Combined, they allow a DSP to identify with reasonable confidence that a given impression is likely being served to a user who has previously had your app — even without an IDFA.

The result is incremental reach to an audience your other partners simply can't access, at CPMs that reflect the lower competition in this auction rather than the premium you'd pay in UA.

The proof is in the CPAs

Deblock is a useful example because the vertical makes the point starkly. Deblock is a European crypto and digital assets app, with users that skew technically sophisticated, privacy-conscious, and deeply skeptical of ad tracking. Crypto users, as a whole, opt out of ATT at rates well above the industry average, which means the lapsed audience is largely sitting outside the reach of deterministic retargeting.

After activating iOS probabilistic retargeting, Deblock saw a 5-8x reduction in CPA. That CPA drop reflects what happens when you stop paying UA prices to reach users you already have — same audience, sharper pathway, at a lower cost.

How RZR built for the iOS signal shift

When ATT changed iOS signal, RZR rebuilt its deep neural networks to work without the IDFA.

Our current iOS retargeting stack includes:

• Hashed signal ingestion (SHA-1 + SHA-256) for stronger match rates on opted-in audiences
• Household IP targeting to extend reach beyond individual device signals
• Audience segmentation by behavior and LTV — payers vs. non-payers, lapsed windows from 3 days to 180+ days, event-based targeting tied to live ops or seasonal moments
• In-house incrementality measurement — a clean read on what RZR media is actually driving versus organic, without third party tools
• Dual-path targeting that covers both deterministic (ATT opted-in) and probabilistic (opted-out majority) audiences in a single campaign structure

One of the persistent concerns with retargeting is whether reactivations would have happened anyway. RZR's in-house incrementality testing — via geo holdout, Ghost Bidding, and PSA test — answers that question end-to-end.

The window is real, but not for long

The iOS probabilistic retargeting auction is undercompeted in because building the infrastructure is hard and most players didn't do it. That's a structural advantage but structural advantages in this space tend to compress as more capital and engineering attention flows toward them.

The teams making the move on iOS probabilistic retargeting are doing so into an auction where:

• CPMs reflect lower competition
• The audience is warm, brand-aware, and cheaper to convert than cold acquisition
• The redownload pool replenishes faster than most UA dashboards suggest

If your iOS lapsed audience is something you've been meaning to address — or if you're currently addressing it through UA campaigns without realizing it — it's worth understanding what a dedicated retargeting pathway actually looks like.

Talk to RZR about building an iOS probabilistic retargeting strategy.