Probabilistic Attribution (Fingerprinting)

Probabilistic attribution, also called fingerprinting, is statistical matching that guesses which ad click led to which install using signals like IP address and device characteristics, used when a deterministic device ID is not available.

You have a click and, later, an install, but no shared identifier connecting them. Probabilistic attribution tries to bridge that gap by comparing signals that both events share, things like IP address, device model, OS version, screen size and timing, and estimating the most likely match. It was the fallback the industry reached for once deterministic IDs started disappearing after App Tracking Transparency.

On Android, in narrow cases, you may still see it used. On iOS, you should treat it as off the table. Apple's App Store Review Guideline 5.1.2 prohibits using device signals to fingerprint and track users, and that prohibition applies regardless of whether the user consented to tracking. Enforcement tightened over the years to the point where fingerprinting can get apps rejected or pulled.

Do not quietly build on it. Probabilistic attribution is increasingly disallowed and, even where allowed, it is an estimate rather than a fact, so accuracy degrades as more networks and privacy features scramble the signals it depends on. On iOS specifically, where Apple bans fingerprinting outright, basing your measurement on it puts your app at risk and your numbers on sand. Use it, if at all, as a rough cross-check on Android, never as your system of record, and never as your iOS answer to the IDFA going away.

Whatever you do, do not build your iOS measurement on fingerprinting, because Apple has already closed that door.