Create a Look-Alike Model

Last update: 2021-08-19

Describes the required and optional steps that let you create a Look-Alike Model.

Model Builder Section

Model Builder consists of the Basic Information and Configuration sections. To create a model, complete the required fields in these two sections. Save your model to start the algorithm. Audience Manager sends you an automated notification after the first data run completes. After you receive the email, you can go to Trait Builder and create algorithmic traits.

NOTE
  • The modeling process runs only once if you create a model and do not build any traits with it.
  • Build models from data sources that contain a meaningful amount of information. Models with insufficient data will run, but they will not return results.
  • Do not create models with other algorithmic traits or segments.
  • The automated email notification is sent one time only (after the first data run).

Build the Model

Follow the steps below to build a Look-Alike Model:

  1. Go to Audience Data > Models and click Add New in the Look-Alike Modeling section.
    look-alike-add
  2. In the Basic Information section
    • Name the model.
    • (Optional) Provide a brief description about the model.
    • Set the status for the model to Active or Inactive. Inactive models will not run and will not produce any data.
      look-alike-basic
  3. In the Configuration section:
    • Click Browse All Traits or Browse All Segments to select a trait or segment you want to model against. Search for traits by name, ID, description or data source. Click on a folder while searching to limit results to that folder and its subfolders. You can also filter traits by trait type (Folder Trait, Rule-based, Onboarded, and Algorithmic) or population type (Device ID and Cross-Device ID).
      browse-traits
    • Choose a 30, 60, or 90 day look-back period. This sets a time range for the model.
    • The TraitWeight algorithm is selected by default.
    • Select a data source from the Available Data list.
    • Click Save when done.
      look-alike-configuration

Watch the video below for a detailed look at how cross-device metrics work.

Basic Information for Algorithmic Models

In Model Builder, the Basic Information settings let you create new or edit existing models. To create a new model, provide a name and move on to the Configuration settings. The description field is optional.

Field Description
Name Give your model a short, logical name that describes its function or purpose. Avoid abbreviations, special characters, and accent marks.
Description A field for additional descriptive information about the model.
Status Activates or deactivates the model (active by default).

Configuration

In Model Builder, the Configuration section lets you add traits or segments to the model. In this section, select a baseline trait or segment, a look-back period, and data from your first and third-party data sources.

Prerequisites

Complete the required fields in the Basic Information section first.

Field Description

Select a Baseline Trait or Segment (1)

Click the trait or segment button to see a list of all your traits or segments. Your selected segment or trait becomes the baseline that the system algorithms use for modeling.

Note: Select an onboarded trait, a rule-based trait, or a segment as baseline. Otherwise, your models will not run.

Select Look Back Period (2)

Sets a time range for the model. Based on your selection, the algorithm includes and evaluates data from the previous 30, 60, or 90 days.

Select Algorithm (3)

At this time, Model Builder works with our proprietary Trait Weight algorithm only. Audience Manager may add other algorithmic functions in subsequent releases.

Select Model Data from Data Source (4)

Lets you select the first and third-party data sources you want to use in the model.

Exclusions (5)

You can exclude traits from the data sources you selected for modeling. Use the Exclusions list and read Algorithmic Models: Trait Exclusion to learn more.

Watch the video below to learn how to create a first party look-alike model, so that you can find more of your own visitors who look like your converters.

Related Articles
Previous page
Next page

On this page