Probabilistic Modeling

What is probabilistic modeling?

Probabilistic modeling is a quantitative modeling method that predicts the potential occurrence of future outcomes or forecasts the possibility of a future event by taking into account random occurrences or actions.

When performing app install attribution or deferred deep linking, it is necessary to match a user in a mobile web browser to the user on the app post-install. Since there are no common identifiers between mobile web browsers and mobile apps, attribution and deep linking companies must rely on probabilistic modeling.

At its core, probabilistic modeling is any form of attempting to connect a browser user to an app user without having 100% certainty that they are the same user.

Historically, MMPs have also used the term fingerprinting as a name for what is more accurately described as “point-in-time probabilistic modeling.”

As of iOS 14, probabilistic modeling can only be used for device-level ad attribution on iOS if users explicitly consent to device-level ad tracking via Apple’s App Tracking Transparency framework. If users do not opt in, SKAdNetwork, Apple’s attribution framework that uses aggregate data only, is the only acceptable attribution alternative.

Retargeting in a Post-privacy World

Considering the incredibly successful — and effective — world of ad retargeting, it seems like privacy and retargeting couldn’t co-exist. To help marketers understand, Loïc Anton, Chief Product Officer from the mobile-app retargeting company Adikteev, shares the current state and the future potential of retargeting in mobile advertising.