Creating Predictive Audiences

Once a Lookalike Model is built and users are scored (make sure the Model Training Only option is unchecked or on the Model Summary page press Activate in the top right), you can create Predictive Audiences with different prediction decision thresholds for the model or percentiles based on the predictions.

From the Lookalike Models list view, click the model you'd like to use to build a Predictive Audience. Then find the Model Usage section and click Create Predictive Audience.

create-predictive-audience

This opens the Audience Builder. All Lookalike Models are keys under the user field Lookalike Model Predictions. The values are the model prediction, interpreted as probabilities on a scale of 0-1. Users closer to 0 represent a low likelihood to look like users in the target audience and vice versa for users closer to 1.

lookalike-model-predictions

By default, this rule is populated with the model's Decision Threshold, computed as the equilibrium prediction score for both Accuracy and Reach (or maximizing both Accuracy and Reach). However you can adjust this threshold as you like or add additional rules before saving the audience. See Improving Lookalike Models for tips on adjusting the Decision Threshold. Any audiences built using the audience prediction score for your model will display in the model usage module.

Using Lookalike Model Percentiles

Another option to build a Predictive Audience is by using the Lookalike Model Percentiles field. Similar to the Lookalike Model Predictions field, the Lookalike Models are keys for the Lookalike Model Percentiles field.

lookalike-model-percentiles

The percentile for a model represents the value at which a percentage of the predictions fall below. For example, the 80th percentile represents the prediction score at which 80% of all other scores fall below, or more simply put; the top 20% of users. Percentiles help account for the shape of a model's prediction distribution, as it can sometimes be hard to determine who the best users are based solely on the prediction scores, if the distribution is skewed is any direction.


Whatโ€™s Next