Android Privacy Sandbox attribution: what it changes

Most studios spent the last three years bracing for Android to repeat what iOS did. When Apple's App Tracking Transparency arrived, deterministic per-install attribution collapsed and the whole measurement stack had to be rebuilt around aggregated, delayed signals. The expectation was that Google's Privacy Sandbox would do the same thing to Android.

It does change the rules, but not in the same shape. Android's Attribution Reporting API is structurally richer than Apple's SKAdNetwork, and a studio that treats it as "iOS again" will miss what is actually different. The replacement for device-level tracking on Android gives you more to work with than the iOS equivalent does, if you understand how it is built.

What Privacy Sandbox actually replaces

For years, Android attribution ran on the Google Advertising ID — a device-level identifier that let networks tie a specific install to a specific ad with near-perfect accuracy. As that identifier is deprecated, the deterministic, user-level attribution that underpinned most Android user acquisition stops working in its current form.

The Attribution Reporting API is the privacy-preserving replacement. Instead of handing networks a device ID they can match across apps, it provides attribution through aggregated, on-device measurement that limits cross-app identification. The destination is similar to iOS in spirit — no more per-user matching — but the mechanics differ in ways that matter for how you measure.

This is worth saying plainly because the framing has been wrong in a lot of industry conversation. Android is not getting a carbon copy of SKAdNetwork. It is getting a different system that happens to share the same privacy goal.

Three differences that matter

The Android model does several things SKAdNetwork cannot, and each one changes what a studio can learn.

The first is multiple attribution sources per event. SKAdNetwork gives you a single postback per install — one winner takes the credit. Android's Attribution Reporting API supports up to three attribution sources per triggering event. That means a conversion can be credited across more than one touchpoint, which gives you a less brutally simplified picture of which interactions contributed to an install.

The second is native view-through attribution. SKAdNetwork is built around clicks. Android's model supports view-through measurement directly, so impressions that influenced a player without a click are part of the picture rather than invisible. For genres where players see an ad, don't tap, and install later, this is a meaningful gap that Android closes.

The third is app-to-web measurement. The Android model supports both app-to-app and app-to-web attribution. For studios running direct-to-consumer web shops or web-based acquisition flows, this connects journeys that crossed between an app and the web — a path that was effectively unmeasurable under a strictly app-centric model.

None of this restores per-user tracking. It is still aggregated, and you still cannot follow a named individual across apps. But within the aggregated world, Android gives you a wider lens than iOS.

Why this means you can't run one playbook

The practical consequence is that the optimisation playbook you built for iOS does not transfer cleanly to Android, and pretending otherwise leaves signal on the table.

On iOS, you learned to live with a single click-based postback and a delayed conversion value. You built models to compensate for what SKAdNetwork couldn't tell you. If you apply that same compensated, click-only thinking to Android, you ignore the view-through and multi-source data that Android actually provides.

The studios getting this right run platform-specific measurement. They use Android's view-through data where it exists rather than discarding it to match the iOS model. They account for multiple attribution sources instead of forcing a single-winner attribution because that's what iOS trained them to expect. The two platforms are no longer measured the same way, and the reporting that pretends they are is quietly less accurate on Android than it looks.

What still breaks, and what to do about it

Honesty matters here: Android's richer model does not make the underlying problem go away. The data is still aggregated, still subject to thresholds that suppress small cohorts, and still delayed relative to the real-time deterministic past. You cannot optimise at the per-install level on Android any more than on iOS.

This is why first-party behavioural signals remain the timely truth source on both platforms. What a player does in their first three days — session depth, early engagement, time to first purchase — is information you own, it arrives immediately, and it does not depend on any platform's attribution window. When the attributed install-quality signal is delayed or aggregated away, your own early-behaviour data is the fastest read you have on whether a cohort is any good.

The platform attribution systems tell you, roughly, where players came from. Your in-game behavioural data tells you what those players are worth. The second question is the one that decides whether a campaign was profitable, and it is the one you have full control over. To understand how early-behaviour signals are computed and measured, read the docs.

A measurement stack that respects both platforms

The right structure is layered, and it differs by platform.

For Android, use the Attribution Reporting API for what it does well — multi-source and view-through credit at the campaign level — rather than collapsing it into an iOS-shaped single postback. For iOS, use SKAdNetwork and its successor for click-based aggregated credit and accept its narrower view. For both, lean on first-party behavioural analytics as the timely, owned signal that answers the value question the platforms can't.

The mistake to avoid is treating attribution as one number from one source. It never was, and on Android in 2026 it especially isn't. The data comes from several places, each answering a different question, and the studio that knows which layer answers which question makes better decisions than the one staring at a single ROAS figure it doesn't fully trust.

Android did not repeat the iOS disaster, and it did not leave measurement untouched either. It replaced device-level tracking with a privacy-preserving model that, within its aggregated limits, gives you more than SKAdNetwork — more attribution sources, view-through credit, and app-to-web reach.

The studios that benefit are the ones who stop treating Android and iOS as the same measurement problem. Build platform-specific reporting, use the Android signals you're actually given, and anchor both platforms in the first-party behavioural data you own. The attribution era of perfect per-user tracking is gone on both. What replaces it on Android is richer than most studios expected — but only if you measure it on its own terms.

Book a demo to see how first-party behavioural measurement complements platform attribution on both Android and iOS.