What's the difference between Google's 'modeled conversions' and Branch's Predictive Modeling?

Even though Google's 'modeled conversions' and Branch's Predictive Modeling (PREM) might sound similar, they're totally different under the surface.

Branch's PREM system watches events (ad clicks + app installs) to try to identify specific matches for the purpose of attribution. We need event metadata (i.e., IP address, user agent) to do that, which requires user opt-in under Apple's ATT policy when used for the purpose of ad attribution.

Google's modeling doesn't need any of that data. They don't discuss how this works, but they're probably doing something like looking at the click-to-installs rate based only on data they already have access to (completely outside of Apple's view), and then simply extrapolating the rate into their 'modeled' numbers.