• All
  • Country
  • Product
Save as pdf

Lookalike Extension

Lookalike audience extension is a process that takes original UUIDs (unique user identifiers) from the segment, builds a behavioral model for it (in other words, describes behavior that is common for all these UUIDs) and then scans all profiled Adform UUIDs. Those UUIDs with high similarity to original UUID set are added to the segment. In this way, original audience gets extended with new UUIDs.

lookalike_aud_grid

 

Each time only original set is used to extend the audience in order not to lose the accuracy of the segmented data. Audience can be extended via UI or API.

Lookalike audience extension can be enabled for any segment which has more than 3000 and less than 1 000 000 UUIDs.

 

  • How UUID pairs are found?

    Audience Exchange 3Lookalike audience extension relies on cookies' ability to track user's browsing activity and detect what they might be interested in.

    Cookies anonymously store data about audiences' browsing behavior. The DMP Audience Extension uses deep learning technology combined with clustering methods to identify cookies that are similar to those cookies in a segment. Similarity is based on:

    • Browsing habits (Days from the last visit, URLs visited, etc.)
    • Hardware (Device type)
    • Geography (City, Country)


    In order to determine the behavior pattern of the selected audience and create a profile model, we're taking the above-mentioned data for the period of 7 days from the last visit.

    The similarity score (threshold value) is used as a filter, such that a lot of cookies are initially filtered and only the remaining are classified using the predictive model.

    Audience Extension 2

     

    If the profiles do not overlap and behavior pattern differs extremely, audience might be extended poorly even if similarity score value is low. Also, profile model will be not descriptive if the segment is populated with not active UUIDs.

     

  • Accuracy of extension vs segment size

    Your business goals should help you make the right decisions about proportion of accuracy and segment size when using Lookalike audience extension. If accuracy is your goal, please note that audience population can increase or decrease across behavioral model runs. We recommend to set the threshold (similarity level or accuracy) to 70 – this means that all UUIDs with similarity of 0.70 or higher will be added to original set. This should result optimal performance of the functionality – the extension will be reasonable and relevance still strong.

    Segment size changes are the results of the algorithm making decisions during daily processing and evaluation. Sometimes, the algorithm finds more lookalikes based on selected accuracy level and, during other executions, it may find fewer. Results are determined by the baseline data used to create behavioral model. By contrast, when working with reach, the user population count remains constant. It is very hard to predict the amount of UUIDs that will be added during the extension because each audience is unique and has own behavioral model.

    The focus on accuracy or segment audience size depends on what you want to achieve with a particular segment. Having high audience extension accuracy is useful for targeted campaigns when you want to reach a specific audience. As an opposite, having low audience extension accuracy, but high segment size is useful for brand campaigns when you're interested in reaching an audience of a high size.

     

  • How often audiences are extended?

    Once Lookalike audience extension is enabled, it will take time until newly selected segment will be processed and profile model created, however that will be done within 24 hours. Profile model is recalculated and updated each day. Based on the model, process is adding new lookalikes constantly.

    It is very easy to check whether the audience was extended by lookalike or not. In the Audience grid, check the Extensions and UID's Lookalike columns:

    • In Extensions column you will see whether a Lookalike extension is enabled and what is its similarity level.
    • In UID's Lookalike column, two numbers are given – an amount of UID's added via Lookalike extension and its percentage of the total UID's in the audience.

    lookalike_amount

     

    In the Audience by Sources graph (go to audience details page > Insights > Composition), total amount of UIDs and date of the last audience extension is shown.

    aud_sources_graph

     

    Audience extension data can be also retrieved via API.

  • Can Lookalike extension be turned off?

    Yes, it is very easy to turn off Lookalike audience extension for one or multiple selected audiences.

    For a single audience, unselect a Lookalike box in audience settings section:

    lookalike_disable_single

     

    For multiple audiences, open a multi-editing window (1) and edit the setting (2) for all selected audiences:

    lookalike_disable_multiple

     

    After Lookalike audience extension is disabled, the process won’t be adding new lookalikes. However extended audience won’t be removed automatically.

    All the UUIDs from the segment will be removed after TTL set for segment expires (the same way as for original UUIDs).

     

  • Lookalike extension and Audience Builder

    When building audiences in the Audience Builder, you can choose Lookalike audience extension as a data source.

    In the Audience Builder, choose My Audiences and add an extended audience. Open the list of data sources of the added audience (1) and select Lookalike (2). Thus only UID's added via Lookalike extension will be included.

    add_lookalike