MWM
Contact salesMWM AI
Attribution & Measurement

Probabilistic Attribution

Also known asProbabilistic MatchingFingerprint AttributionModeled Attribution

Attribution that statistically estimates which ad click drove an install using non-unique signals (IP address, user agent, timestamp) when a deterministic device ID is unavailable.

Key takeaways

  1. 01Probabilistic attribution estimates the most likely ad source for an install using signals like IP, user agent, and timestamp — no unique device ID required.
  2. 02It's the fallback when deterministic matching (IDFA/GAID) isn't available, which post-ATT is the majority of iOS traffic.
  3. 03It's a statistical best-guess, not a 1:1 certainty — accuracy is typically modeled around 85-90% over short windows and degrades over time.
  4. 04Apple discourages fingerprint-based probabilistic matching on iOS, pushing advertisers toward SKAdNetwork instead.

Probabilistic attribution estimates which ad click most likely caused an install using a combination of non-unique signals — typically IP address, [[user-agent]] string, and timestamp — rather than a unique [[device-id]]. It answers "which click probably drove this install?" with a statistical model instead of a definitive match.

Deterministic vs probabilistic

Deterministic vs probabilistic attribution

DeterministicProbabilistic
BasisUnique device ID (IDFA/GAID)IP + user agent + timestamp
Certainty1:1 matchStatistical best-guess
AvailabilityNeeds ATT opt-in / GAIDWorks without a device ID
Typical accuracyNear-exactModeled ~85-90% short-window

Probabilistic matching fills the gap deterministic attribution can't cover post-ATT — but as a model, it's less precise and decays as the matching window widens.

Why it matters now. Since [[att]] gated the IDFA, most iOS installs lack a deterministic ID, so probabilistic methods fill the gap — but Apple's guidelines discourage fingerprinting, steering advertisers to [[skadnetwork]] for privacy-preserving aggregated measurement. The practical stack post-ATT blends SKAdNetwork, the opted-in deterministic slice, and probabilistic modeling for the rest.

Quick answers

What is probabilistic attribution?

Probabilistic attribution estimates which ad click drove an install using non-unique signals — IP address, user agent, and timestamp — instead of a unique device ID. It produces a statistical best-guess rather than a 1:1 match, and it's the fallback method when deterministic identifiers like the IDFA aren't available.

How is probabilistic different from deterministic attribution?

**Deterministic** uses a unique device ID (IDFA/GAID) for an exact 1:1 match between click and install. **Probabilistic** infers the match statistically from IP + user agent + timestamp when no device ID is available. Deterministic is near-exact but needs consent; probabilistic works without an ID but is a model, typically cited around 85-90% accuracy over short windows.

Is probabilistic attribution allowed on iOS?

Apple's guidelines discourage fingerprinting-based probabilistic matching and steer advertisers to SKAdNetwork for privacy-preserving measurement. In practice, probabilistic modeling still gets used in parts of the ecosystem, but the sanctioned iOS path is SKAdNetwork plus the opted-in deterministic slice — so reliance on probabilistic matching is decreasing on iOS.

Next reads

Back to glossary