Predictive Audience

A predictive audience is a GA4 audience built on predictive metrics like purchase probability or churn probability rather than on actions users have already taken.

GA4 models each user's likelihood to buy or churn from their behavior, then lets you define audiences off those scores: likely 7-day purchasers, likely 7-day churning users, predicted top spenders. You use them the same way as any other audience, in reports or, when linked, in Google Ads. The eligibility depends on your property having enough qualifying purchase and churn events to train the model.

You create a "likely 7-day churning purchasers" audience, link it to Google Ads, and run a win-back campaign at users the model flags as about to lapse, before they actually go quiet.

No model, no audience. Predictive audiences only become available once GA4 has enough data to generate the underlying metric: roughly 1,000 returning users who triggered the relevant event, and enough who did not, inside a seven-day window, with the volume sustained. Small or low-conversion properties often never qualify.

The model has to be fed before it can guess.