End of precise attribution: what replaces it in 2026

For most of the last decade, mobile games could answer a precise question: this specific player came from this specific ad, and here's exactly what they did afterward. The attribution chain was deterministic. The optimization loop was fast. The data was granular enough that creative testing, audience segmentation, and ROAS reporting all worked at the per-install level.

That chain is broken now, and not by an incident — by structural changes the major platforms have rolled out deliberately over the last three years. Apple's SKAdNetwork has reached version 5.0. Google's Privacy Sandbox is in full rollout on Android. The deterministic per-user attribution that earlier mobile UA depended on has been replaced by aggregated, probabilistic, delayed signals.

For studios that built their measurement around the old model, this is a real disruption. The reporting they look at is less precise than it appears. The cohort comparisons they trust may be apples to slightly different apples. The optimization decisions they're making with the same speed as before are being made with worse data.

The good news is that the new model is functional, once you understand what it actually provides and how to make decisions with it. The bad news is that no amount of additional tooling restores what was lost; the underlying signal really has changed.

What the platforms actually do now

The two ecosystems handle this differently, but the destination is similar.

On iOS, SKAdNetwork is Apple's privacy-preserving attribution framework. It returns aggregated, delayed signals about which installs came from which campaigns. The granularity varies by version. SKAdNetwork 4.0 introduced multiple postbacks and conversion windows. SKAdNetwork 5.0 has refined this further with more flexible measurement windows and broader signal types. What it doesn't provide is per-user identity that crosses the install boundary.

On Android, Privacy Sandbox is the broader framework that includes Attribution Reporting, Topics, and Protected Audience APIs. The implementation in 2026 represents a meaningful departure from the previous GAID-based attribution model. The signals are aggregated, the timing is delayed, and the per-user precision that GAID enabled is no longer available in the same form.

The common pattern across both: a campaign report tells you "this campaign produced these conversions in this aggregate window" — not "user X clicked ad Y and converted Z minutes later." The shift is from event-level signals to cohort-level signals.

What this breaks about the old optimization loop

Mobile UA used to work like this: a studio ran multiple ad variants, observed which ones produced installs that converted, attributed the conversions back to the specific creative, and shifted budget toward winning variants. The feedback loop was fast (hours to days), granular (per creative, per audience, per device), and trusted (the attribution was deterministic).

The current loop is structurally different in three ways.

The feedback is slower. Aggregated conversion windows mean the signal arrives delayed — sometimes by days, sometimes by weeks for conversion events that matter. The fast iteration that worked when feedback was immediate doesn't work when feedback is delayed.

The signal is coarser. A studio can see which campaigns produced installs that converted, but the specific creative variant that drove the result is often obscured. A/B testing creative at the variant level produces results that may not separate cleanly.

The reliability is harder to verify. Aggregated reports have built-in privacy noise. A small lift between two variants may be real or may be within the noise floor of the reporting itself. The threshold of detectable effect has moved up.

The net effect is that the old optimization loop runs slower, on noisier data, with less confidence in any specific outcome. Studios who haven't adjusted their loop speed and confidence thresholds to match are making decisions on signal that isn't as strong as they think.

What replaces it

Several measurement patterns have emerged to work with the new signals. They're not direct replacements — they're different ways of producing actionable information from the data that's actually available.

Cohort-level retention curves are more reliable than per-user attribution chains. A studio can see how the cohort of installs from a specific period behaves over weeks, even if individual attributions within the cohort are noisy. The cohort itself is real; what each member of it did precisely is not.

Incremental measurement via holdout becomes more important. Instead of trying to attribute specific conversions to specific ads, the studio compares treated populations to held-out populations and measures the difference. This works at the aggregate level the new signals natively provide.

Media mix modeling — a statistical approach that estimates each channel's contribution to overall outcomes from aggregated data — has come back into favor. It's an older technique, designed originally for traditional media, that handles aggregated signals natively. The trade-off is that it produces channel-level insights rather than creative-level ones.

Predictive LTV from early in-game signals reduces the dependency on attribution accuracy. If a studio can predict a player's lifetime value from their behavior in the first three days, attribution precision matters less — the studio can value installs based on what the players actually do, not on what the attribution claims they did.

These patterns share a property: they treat attribution as an estimate, not a measurement. The studios using them well are explicit about the uncertainty in their reporting and don't make precision claims the underlying data doesn't support.

What needs to change architecturally

For studios building or rebuilding their measurement stack in 2026, a few architectural choices matter more than they used to.

Capture rich first-party in-game data. Anything you measure inside your own product is still under your control. Session depth, progression, monetization moments — these are signals you can use to value players that don't depend on platform attribution. The studios that have invested in their own analytics infrastructure are less affected by the attribution shift than those who depended on the platform's signals.

Build for cohort comparisons, not per-install ROAS. A reporting stack that wants to answer "which install paid back" is asking a question the platform won't reliably answer anymore. A reporting stack that asks "how is the cohort from this campaign performing relative to baseline" is asking a question the platform can answer well.

Plan for measurement delay. Decisions about which campaigns to run get made on data that's days or weeks old. Operations that assume yesterday's report reflects yesterday's reality need to adjust to operating on a lagged signal.

Accept noise. The signals you get have privacy noise built in. A 2% lift may or may not be real. The threshold for "this is working" has moved up. Studios who chase small lifts in the new environment are often optimizing for noise.

What it means for the marginal install

The most important practical consequence is that the per-install value calculation gets fuzzier. The previous question — "is this specific install profitable" — is increasingly answered "we don't know precisely." The newer question — "is this cohort profitable on average" — is the one studios can actually answer.

This makes the marginal install riskier. A studio that wants to grow by buying installs at the platform-average CPI is making a bet that the average install in that pool will pay back. With weaker per-install attribution, the bet has wider error bars than it used to.

The studios who adapt to this typically buy more conservatively at the margin and invest more aggressively in product depth — because product depth produces durable per-player value that doesn't depend on attribution precision to measure. The studios who keep buying as if the old signal was still available are paying for installs whose actual value they're estimating with worse data than they realize.

See how we measure value when attribution is noisy →

The end of precise attribution is one of those transitions that doesn't have a single dramatic moment — it just keeps shifting under everyone's feet. The studios who notice and adapt their reporting and decision-making stay aligned with the actual signal. The studios who keep operating on the old assumptions are increasingly making decisions on data that no longer matches the world it describes.

Talk to us about per-player value measurement without precise attribution →