Probabilistic Attribution FAQ

Learn about Singular's methods for mobile app attribution when device-level IDs are not available. For general background, see  Understanding Singular Mobile App Attribution.

Note: Starting with iOS 14.5, the use of probabilistic attribution methods on iOS is limited to Apple-allowed scenarios.


What is probabilistic attribution?

Probabilistic attribution (also known as fingerprinting) is a mobile attribution method in which Singular collects device information passed through the tracking link when a user clicks on an ad. When a user downloads and opens a mobile app, the Singular SDK collects the same data from the device and sends it to the Singular servers, enabling Singular to search for a matching device in its database.

How does probabilistic attribution compare to using device IDs?

Probabilistic attribution is not as reliable as matching devices by their unique identifier. Singular uses probabilistic attribution as a fallback if no device identifiers are available (and neither is the Google Install Referrer, for Android devices).

Probabilistic attribution also uses a different attribution lookback window. The data used in this attribution method, such as the IP address, changes much more often than device identifiers (which users can reset but rarely do so). The accuracy of the device match decays exponentially after 24 hours. For this reason, Singular limits the lookback window to 24 hours.

When is probabilistic attribution typically used?

Since device identifiers are only available in mobile app environments, any advertising campaigns that run on a mobile web environment have to rely on probabilistic attribution when the Google Install Referrer isn't available.

For example:

  • Email campaigns
  • Tracking organic downloads from your mobile web landing page
  • Campaigns with ad networks working with mobile web inventory

Probabilistic Attribution in iOS 14

Starting with iOS 14.5, Apple's user privacy and data use guidelines restrict probabilistic attribution to limited scenarios. See Singular's attribution workflows for iOS.

What data does Singular collect for probabilistic attribution?

Singular relies on data points such as:

  • Platform
  • IP Address
  • OS Name
  • OS Version
  • User Agent
  • Timestamp

Note that this is publicly available information available in the HTTP headers.

Was this article helpful?