Overview
Predictive Modeling (PREM) attributes app conversions to the marketing activity that likely drove them when deterministic identifiers aren't available, and in contexts where its use is permitted by applicable platform policies, customer configuration, and user consent status. It compares a small set of signals from a link click and a later conversion event, and estimates whether they came from the same device. These signals are not used to create a stable or reusable identifier.
PREM is only used where platform privacy policies allow it, and only for measurement and attribution. PREM is not appropriate for authentication, fraud prevention, account recovery, creating persistent identifiers, or other critical user-level use cases.
What PREM supports
With PREM, you can:
Attribute conversions when deterministic values aren’t available.
Measure campaign performance.
Power attribution models such as multi-touch attribution (MTA), including Funnel.
Benefit from timely attribution signals for campaign optimization.
How does PREM work?
PREM records limited signals necessary for attribution. PREM evaluates permitted marketing interaction and conversion context to estimate campaign performance.
PREM uses statistical methods to support campaign performance reporting where permitted by applicable platform policies, customer configuration, and user consent status.
Where PREM runs by default
iOS
On iOS, Branch prioritizes Apple-supported and consent-aware attribution methods, including SKAdNetwork, AdAttributionKit, self-attributing network frameworks, Apple Ads attribution, and ATT-permissioned attribution where applicable.
PREM on iOS:
Is enabled for paid channels only when the user has granted permission through Apple's App Tracking Transparency (ATT) prompt.
Is enabled by default for non-paid channels.
Android
On Android, PREM is enabled by default across all channels.
Where deterministic methods are available, Branch prioritizes those methods before applying PREM.
Limitations
PREM is a measurement tool and is not designed for authentication or any decision that has user-facing consequences. Reporting precision varies with the available measurement context, customer configuration, and applicable platform policies.