Unattributed installs

Definition

An install is unattributed when no signal linking it to a prior ad click or link tap survived the journey through the app store, which can mean the install was organic or that the signal existed and was lost.

Those two cases look identical in a dashboard and have completely different implications, so the useful work is separating them. Some of the gap is fixable by you today. Some of it is a platform limit that no measurement vendor can close, whatever their marketing says — and a vendor reporting a suspiciously small unattributed bucket is usually guessing rather than measuring.

What can actually be known

Attribution is not one mechanism. It is a set of signals with very different guarantees, and which ones exist depends on the platform and how the user reached the store.

Gaps you can close

Roughly in order of how much volume they typically account for, and all of them are your side of the line.

1. The link was never tagged. The most common fixable cause by a distance. A store link posted in a newsletter, a bio, a QR code, or a partner page with no campaign parameters produces an install that is genuinely indistinguishable from organic. It was a campaign, it just did not say so.
2. The store URL lost its parameters. A redirect, a shortener, or a hand-edited link that dropped referrer. Worse, an unencoded & inside the referrer value truncates it, producing partial data that looks like a successful read.
3. The referrer was requested once and given up on. SERVICE_UNAVAILABLE from the Play referrer client is transient, and the value stays available for 90 days. Integrations that treat the first failure as "no referrer" discard recoverable installs permanently.
4. The SDK initialised after the first screen. The payload arrives, the app has already decided what to show, the association is dropped. Attribution data exists and is thrown away.
5. Attribution runs only on cold launch. An install that first opens via a push notification or a warm resume can miss the code path entirely.
6. The match window is too short for the actual journey. A user who taps an ad, gets interrupted, and installs the next evening is real behaviour, not an edge case.

# Android — the referrer must be ONE url-encoded parameter.
# Decoded: utm_source=newsletter&utm_medium=email&utm_campaign=spring
https://play.google.com/store/apps/details?id=com.example.shop&referrer=utm_source%3Dnewsletter%26utm_medium%3Demail%26utm_campaign%3Dspring

# The failure that looks like success: an unencoded & ends the parameter,
# so everything after utm_source is silently dropped.
...&referrer=utm_source=newsletter&utm_medium=email   # BROKEN

# iOS — App Store links accept campaign parameters, but they feed
# App Store Connect analytics, not your app. Nothing is handed to the
# binary on first launch. There is no iOS install referrer.
https://apps.apple.com/app/id123456789?pt=12345&ct=spring&mt=8

Gaps nobody can close

These are platform and policy limits. They are worth stating plainly because the alternative is treating them as a vendor selection problem and switching tools to solve something no tool solves.

Telling organic from lost

Since both land in the same bucket, use signals around the install rather than the install record itself.

• Compare the platform split. If Android's unattributed share is much lower than iOS's, the difference is roughly the platform limit — that part is structural. If Android's is *also* high, that part is fixable and is probably untagged links.
• Watch unattributed volume against spend. Genuinely organic installs do not track ad spend. A bucket that rises and falls with campaigns is measurement loss wearing an organic label.
• Check store listing traffic sources. Play Console and App Store Connect report browse versus search versus referral independently of your SDK. High referral traffic alongside high unattributed installs means links are arriving untagged.
• Look at the timing distribution. Organic installs spread across the day. A cluster minutes after a campaign send is not organic.
• Audit one campaign end to end. Take a live link, tap it on a clean device, install, and watch what your integration receives. One pass finds untagged links and truncated referrers faster than any dashboard analysis.

A stable unattributed share that does not respond to spend, with Android meaningfully lower than iOS, is close to the floor for honest measurement. Chasing it below that means buying inference and calling it data.