Predictive analytics uses statistics and machine learning to estimate the likelihood of a future outcome from past data. The question is not what happened but what will probably happen. The output is usually a probability or a range rather than a definite answer.
It works like any other modelling job. A training set is built from historical records, a target variable is defined, the model is trained and then tested against a separate period. The time axis is the critical part: the model may only use information genuinely available at the moment of prediction. Let a field that is populated later slip into training and the score looks strong while production collapses.
Common uses on the marketing and analytics side are churn prediction, purchase propensity scoring, lifetime value estimation, demand planning and inventory management. The predictive audiences in GA4 are the same logic turned into a product feature.
A concrete case: a subscription product computes a monthly churn probability for every customer. Customers scoring high who also carry high revenue land on a separate list, and the team runs specific outreach to it. The value of the prediction is measured by how much better that list performs than a random one.
A prediction is not a guarantee. The model assumes the past pattern continues, and when conditions shift the scores go stale quickly and the model has to be retrained.
