Android attribution reporting after Sandbox

Chrome's Privacy Sandbox retirement started rolling through with the Chrome 144 deprecation and left Android measurement in a different place to where most buyers assumed it would land two years ago. This piece is a status check. It covers where Android attribution reporting actually sits for a mobile game buyer today, not the shutdown itself, which Privacy Sandbox Retirement: What Buyers Actually Lose covered when it was still news.

What survived and what did not

The Privacy Sandbox APIs built specifically for the browser, the ones designed to replace third-party cookies for web advertising, are the part that is genuinely gone from Chrome's default configuration. That was always a web-advertising story more than a mobile-app story. Its retirement matters most to buyers running web-to-app funnels or browser-based re-engagement, covered separately in earlier pieces on web shop attribution.

The mobile-specific side of Android's privacy programme is a separate track. The on-device Attribution Reporting API and the Protected Audience mechanisms aimed at in-app advertising have not gone anywhere; Google has not retired them, and they remain its stated direction for Android measurement. What changed is less about that API's existence and more about pace: with Chrome's parallel effort abandoned, Google has faced less external pressure to force a fast Android cutover, and GAID has stayed meaningfully available for longer than the earlier timeline implied when GAID Deprecation: The Next Android Attribution Shock first flagged the risk. The warning in that piece still stands.

What GAID still gives a buyer

The Google Advertising ID remains accessible on the large majority of Android devices in market today, which means device-level attribution, the kind that lets an MMP join a click directly to an install without modelling, is still the default experience for most Android campaigns. That is the single biggest practical difference from iOS, where for years SKAN and AAK have been the only game. Android buyers have had a longer runway on deterministic measurement than the 2025-era warnings suggested, and that runway is the main reason Android budgets haven't needed the same conversion-value schema discipline that iOS forced.

GAID isn't permanent, though. Every MMP is quietly building the modelled fallback in parallel rather than waiting for a hard cutover to force the issue, so read GAID's continued availability as a delayed transition, not a cancelled one.

How MMPs are blending the two signals

AppsFlyer, Adjust, Singular: each has built a modelling layer that sits alongside GAID-based deterministic attribution rather than replacing it outright, using the deterministic share of traffic that still has a GAID to calibrate and validate the modelled estimates applied to the shrinking share that does not. In practice this means an Android dashboard today mixes two kinds of number in the same report. Some installs carry a device identifier. Others are estimates, produced through statistical modelling calibrated against that same deterministic base.

The operational risk is that these two categories look identical in a standard dashboard view. A buyer reading blended Android numbers without separating deterministic from modelled installs is trusting a number that is, for a growing minority of traffic, already an estimate rather than an observation, and that minority will keep growing as device-level identifiers erode further.

What to check now, and why the pace matters

The slower Android transition has a practical upside worth naming directly: it gives buyers more time to build the conversion-value discipline that iOS forced years earlier, without a hard deadline attached. Studios that treat this extra runway as permission to defer that work will find themselves doing the same schema design and validation exercise under time pressure once the modelled share of Android traffic crosses a threshold that actually changes bidding outcomes. Studios that use the runway instead, checking modelled shares by network now rather than waiting for a forced cutover, will make that later transition close to uneventful.

Ask your MMP directly what share of your Android installs in the last reporting period were deterministic versus modelled. Ask for that split by network too, since it won't be even; networks buying heavily into inventory with privacy-conscious device configurations, or markets with higher adoption of GAID-limiting settings, will show a higher modelled share earlier than the rest of your Android portfolio.

Build the habit now of treating Android ROAS and CPI the way buyers have treated iOS numbers since SKAN arrived: as a number with a stated confidence level attached, not a precise fact. The studios that adapted fastest to SKAN's constraints were the ones that had already learned to ask what a number actually measured before acting on it. Android is asking for the same discipline now, on a slower timeline than most buyers expected in 2025.

Related archive reading

These articles provide related context and remain subject to their stated review status.

• GAID Deprecation: The Next Android Attribution Shock
• AdAttributionKit Re-Engagement Testing After iOS 26