Probabilistic Attribution
Probabilistic attribution assigns conversion credit to ad exposures based on statistical likelihood rather than deterministic device identifiers.
Definition
Probabilistic attribution uses statistical modeling, based on signals like IP address, device type, OS version, timestamp, and behavioral patterns, to infer which ad a user likely saw before installing an app, when a deterministic identifier (like IDFA) is unavailable.
Prior to ATT, user-level attribution was largely deterministic: a unique device ID matched the click to the install with certainty. After ATT reduced IDFA availability, probabilistic methods became essential for attributing the majority of iOS installs from opted-out users.
Probabilistic attribution is inherently less accurate than deterministic attribution, it operates on probability rather than fact. However, at scale, well-calibrated probabilistic models can provide reasonable accuracy for campaign-level budget allocation decisions even when user-level precision is unavailable.
Why it matters
As deterministic attribution coverage shrinks on iOS, probabilistic modeling fills the measurement gap for campaigns that can't rely on IDFA. Understanding the accuracy limitations of probabilistic attribution, and not treating it as equivalent to deterministic data, is essential for correctly weighting mixed iOS measurement signals.
Frequently asked questions
Is probabilistic attribution still allowed after ATT?
Yes, probabilistic methods based on non-identifying signals (IP, device parameters) are generally permitted. Fingerprinting using device hardware attributes was explicitly banned by Apple in its ATT guidelines, creating a distinction between compliant probabilistic modeling and non-compliant fingerprinting.