Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.permutive.com/llms.txt

Use this file to discover all available pages before exploring further.

Lookalike models use machine learning to find users who behave similarly to a specific publisher “seed” audience. This guide walks you through the process of creating a new model.

​
Prerequisites

Before you begin, ensure you have:
  • A high-quality publisher seed segment (e.g., a custom cohort of email subscribers or converters).
  • At least 1,000 unique users per day in your seed segment for optimal accuracy.
  • At least 10-20 active custom cohorts in your project to provide enough behavioral signals.
1
Access the Models Tab
2
Log into the Permutive dashboard and navigate to Modeled Cohorts in the left-hand navigation bar. Then, click on the Models tab.
3
Start a New Model
4
Click the + Add Model button. You will be prompted to choose the type of model you want to create. Select Lookalike.
5
Select Your Publisher Seed Segment
6
Choose the publisher cohort you want to use as your seed.
7
Only matched cohorts or custom cohorts can be used as seed segments.
8
Configure Advanced Settings (Optional)
9
You can further refine your model by using the following options:
10
  • Include Segments: Restrict the model to only consider users within specific cohorts.
  • Exclude Segments: Ignore users in specific cohorts during the modeling process.
  • Fixed Set: Manually select the specific cohorts the model should use as behavioral signals (advanced users only).
    • 11
    Build the Model
    12
    Click Build Lookalike Model.

    ​
    What Happens Next?

    Once you click build, Permutive starts the training process.
    • Training Time: It typically takes up to 2 hours for a model to finish training.
    • Status: The model will show as “Updating” or “In Progress” until it is ready.
    • Result: Once complete, you will see a Precision vs. Reach curve, which you can use to create lookalike cohorts.
    If your seed segment was created on the same day as the model, the build may fail or show 0 reach until a full day of data has been processed.