Lookalike (LAL) Modelling uses machine learning to find new people with features and characteristics similar to those in the seed segment chosen. Similarity and reach metrics allow you to find a model with the highest similarity at the reach required for your campaign.

You need:

  • Access to the Permutive dashboard
  • A Permutive segment already set up, for which you'd like to create a LAL Model

In the Permutive Dashboard:

Click on the segment you'd like to use for the model

At the bottom of the segment builder, select 'Build lookalike model...'

Creating a Lookalike Model

Select which of your first-party segments you want the LAL model to be trained on.

All: Create your look-alike model based on all existing segments.

All With Exclusions: Create your look-alike model based on all existing segments minus ones selected.

Fixed Set: Create your look-alike model based only on selected existing segments.

The LAL Model will take up to 24 hours to build. During this time, the segment will display that it's provisioning a lookalike model and the time this work started.

A Reach/Similarity graph is returned with a distribution of users you can create a LAL model from. Select the most relevant audience by clicking on one of the dots on the graph. Any number of lookalike segments can be deployed from one lookalike model by clicking on the segment you wish to create.

Weights

  • A look-alike model with high positive weight indicates that users in the segment are very similar to users in the seed segment.
  • A high negative weight means users in the segment are likely not to belong to the seed segment. 
  • Weightings closer to zero indicate that these segments don't significantly influence the model on their own.


The sorting by absolute value is by design. It effectively sorts the features by segments that influence the model the most (whether it's a high positive weight or a high negative weight).

A Const of 0 indicates a 50% probability a randomly selected user is in the seed segment. As it becomes negative that probability goes down.

Creating a Lookalike Segment

Once you have chosen a reach and similarity, the deployed lookalike segment will appear in the list of segments as a separate segment with an automated name
e.g. Sports Enthusiasts — [Lookalike: 28.4% similarity].

You can edit or delete a lookalike model at any time by going to the seed segment you used to create it, and clicking either Edit or Delete above the graph. 

For FAQs on Lookalike modelling, please click here.

Did this answer your question?