Probabilistic modeling
Probabilistic modeling is a statistical method used to estimate outcomes under conditions of uncertainty. Rather than relying on exact matches or deterministic data points, it uses statistical inference to estimate the most likely explanation for an observed result.
What is probabilistic modeling
In mobile marketing, probabilistic modeling (sometimes called probabilistic attribution) is used to estimate aggregate campaign performance when deterministic identifiers – such as device-level advertising IDs – are not available. This became increasingly relevant after Apple introduced the App Tracking Transparency (ATT) framework, which requires apps to obtain user consent before accessing the device advertising identifier. With most users opting out, app developers needed new methods to understand whether their marketing campaigns were generating results – without identifying individual devices or users.
Where deterministic attribution relies on matching a known device identifier to a specific install, probabilistic modeling takes a different approach: it produces aggregate, campaign-level estimates of performance. The question it answers is “did this campaign work,” not “which device installed the app.”
Probabilistic modeling at AppsFlyer
AppsFlyer uses probabilistic modeling as a privacy-preserving statistical method to report on how app developers’ marketing campaigns are performing. It produces aggregate campaign-level performance reports for app developers. It estimates how many installs or engagements followed a developer’s own marketing campaigns – ads, website links, emails, QR codes, SMS, or other owned media – for that developer’s own app. The approach is built on data minimization: it uses only the signals necessary to produce an aggregate report. The output is a campaign performance report, not an identification of any individual device or user.
All measurement is scoped to a single app developer’s own account. No data is shared or linked across developers. The model does not construct device fingerprints, device graphs, user profiles, or any permanent identifier derived from device signals.
Probabilistic modeling is one component of AppsFlyer’s multi-method measurement approach, which also includes SKAdNetwork / AdAttributionKit integration, Aggregated Advanced Privacy (AAP), and deterministic attribution for users who grant ATT consent.
