Skip to main content

Build a Lookalike Segment

When you want to find more people who behave like your best existing customers, you can create a lookalike segment. Building a lookalike segment starts with choosing a first‑party “source segment” that you have built on the Segments page and then identifies people in a larger population who share similar characteristics with that segment.

Depending on your situation, your source segment customers might be high spenders, frequent purchasers, long‑tenured subscribers, highly engaged app users, or recent converters from a key campaign.

You can then distribute the lookalike segment to your desired destinations.

Note

This workflow is in limited availability and is by invitation only.

See the sections below for more information on how to build a lookalike segment and for an overview of lookalike segment building.

Steps to Build a Lookalike Segment

To build a lookalike segment:

  1. From the navigation menu, select Data Management → Segments to navigate to the Segments page.

  2. In the row for the source segment you want to use, click the More Actions menu (the three dots) and then select Build Lookalike Segment.

    Note

    Lookalike segments can only be built from first-party data source segments that have a status of “Ready” and that contain at least 500 members. To simplify the process, you can use the Ownership and Segment Status filters to display only your first-party segments ("My Data") that have a status of “Ready”.

    C-Build_Lookalike_Segment-Build_Lookalike_menu_selection.png

  3. If you’ve already built a lookalike segment from this source segment, click Build New Lookalike Segment to open the builder.

    Note

    If you have not already built a lookalike segment from this source segment, the builder will already be open on the tab.

    C-Build_Lookalike_Segment-Build_New_Lookalikes_button.png

  4. Enter a name for the lookalike segment.

    C-Build_Lookalike_Segment-Name_field.png

    Note

    • Segment names must start with a letter and can contain letters, numbers and underscores. Segment names cannot contain spaces or special characters.

    • Segment names cannot be longer than 90 characters.

    • By default, segment names are preceded with “Lookalike Segment >” to make it easier to identify them..

  5. Select a predefined lookalike segment size or enter a custom size (larger segments reach more people but have lower similarity, while smaller segments have higher similarity but lower reach).

    Note

    • The predefined sizes are displayed in descending order, with the most similar to source size at the top and the least siilar to source size at the bottom.

    • The lookalike segment size cannot be smaller than the size of the source segment.

    • The lookalike segment size cannot be larger than 50% of the modeling universe.

    • After the lookalike segment has been built, a similarity rating will be displayed (for more information, see the "Using Similarity Ratings" section below).

    C-Build_Lookalike_Segment-Select_size.png

  6. Choose how to handle members of the source segment:

    C-Build_Lookalike_Segment-Select_source_options.png

    • To Include members of the source segment, slide the toggle to the right. This will Include all members of the source segment in addition to the highest‑scoring people from the modeling universe until the segment reaches the specified size.

    • To include only optimized members of the source segment, slide the toggle to the right and then select Only Optimized Members. This will Include any source members who score high enough to be selected, but allow low‑scoring source members to be excluded so the segment is built from the highest‑scoring people overall.

  7. Click Build Lookalike Segment.

    C-Build_Lookalike_Segment-Build_button.png

Connect builds the lookalike segment from the modeling universe based on the characteristics of the source segment. This usually takes up to 24 hours for the first lookalike segment built from the source segment.

The lookalike segment now appears on the Lookalikes tab for the source segment and on the Segments page as a new segment. From the Lookalikes tab, you can view the details for the lookalike segment by selecting it.

You can now distribute the lookalike segment to your desired destinations.

Overview of Lookalike Segment Building

At a high level, lookalike segment building works in the following way:

  1. Select a source segment:

    1. You choose an appropriate first‑party segment that represents the consumers you want to replicate (for example, high‑value purchasers). This segment will become the “source segment” for the lookalike model process.

    2. You choose a target size for the lookalike segment and build the lookalike segment. The highest‑scoring people in the modeling universe are selected until the segment reaches that size.

  2. A “model” is built:

    1. The modeling service analyzes that source segment and learns which attributes in the modeling universe (essentially a population dataset) are predictive of membership in the source (for example, certain demographic or behavioral traits).

    2. Each person in the modeling universe is scored based on how closely they match the learned “profile” of the source segment.

  3. The lookalike segment is built: The resulting lookalike segment is displayed as a normal segment on the Segments page and can be activated to your destinations. It is also displayed on the Lookalikes tab for the source segment, along with a similarity rating.

Note

All lookalike segments are given a prefix of “Lookalike Segment >” to their segment name to help you identify those segments.

Smaller lookalike segments tend to be more similar to your source (higher precision, lower reach). Larger lookalike segments reach more people but will usually include individuals who are less similar to the source (higher reach, lower precision).

After the first lookalike segment is built from a particular source segment, additional lookalike segments that are built from the same source segment use the same model (the model does not need to be rebuilt).

Lookalike Segment Building Restrictions

The following restrictions help ensure that lookalike segments remain both effective and compliant with LiveRamp’s data usage policies:

  • Lookalike modeling is currently available only for segments containing US data.

  • Source segments must be based on first‑party data and contain at least 500 members. You cannot use Data Marketplace segments themselves as the source for lookalike modeling.

  • The lookalike segment size cannot be smaller than the size of the source segment.

  • The lookalike segment size cannot be larger than 50% of the modeling universe.

Guidelines for Building Lookalike Segments

Use the guidelines listed below when planning and building lookalike segments in Connect.

Choosing a Source Segment

When choosing a source segment to build the lookalike segment from:

  • Use a first‑party source segment that represents the audience you want to expand (for example, repeat purchasers, high spenders, or highly engaged users).

  • Make sure the source segment is large enough for modeling. As a general rule, avoid extremely small sources. A minimum of 5,000 individuals is recommended for optimal performance. Source segments with under 500 members cannot be used.

  • Avoid sources that are too broad (for example, “all customers”) or that combine multiple very different behaviors into one segment, because the model will have difficulty learning clear predictive patterns.

Choosing the Lookalike Segment Size

During lookalike segment building, you choose the desired segment size:

  • Smaller lookalike segments contain people who are, on average, more similar to the source. These provide higher precision but lower reach.

  • Larger lookalike segments reach more people by including lower‑scoring matches. These Increase reach but usually lower the average similarity and uplift over random targeting.

Note

  • The lookalike segment cannot be smaller than the size of the source segment.

  • The lookalike segment cannot be larger than 50%  of the modeling universe size).

Using Similarity Ratings

After you create a lookalike segment, Connect can display a similarity rating that indicates how closely the lookalike segment matches the characteristics of the source (source) segment.

Here’s how similarity ratings are generated:

  • The modeling process learns which attributes are strongly or weakly predictive of being in the source segment (for example, “takes two or more vacations per year” might be strongly predictive, while “has children” might be weakly predictive).

  • Each person in the modeling universe is then scored based on how many predictive attributes they share with typical source members and how strong those attributes are.

  • The similarity rating summarizes this information for the lookalike segment as a whole. Higher similarity means that, on average, people in the lookalike segment are more likely to share key attributes with the source segment than a random person in the population.

You can use similarity ratings to compare different lookalike segments built from the same source or to understand the trade‑off you made between reach and similarity when you selected a particular segment size.

Including Optimized Members

When you build a lookalike segment, you can choose whether and how to include members of the original source segment. One of the available options is to include only optimized members of the source segment.

Including only optimized members can be useful when you want the strongest possible modeled segment and are comfortable with letting the model exclude source members who do not look like the rest of the source.

When you select the “Only optimized members” option, the lookalike segment:

  • Includes any members of the source segment that score high enough against the modeling universe to be included while reaching the specified segment size.

  • Does not guarantee that all source segment members will be included. Source segment members who look very different from the rest of the source (outliers) or who score lower than other candidates in the population may be excluded.

This option lets the model “optimize” which source members to keep based on how well they match the learned profile, so that the final segment is as similar to the source segment as possible for the selected size.

Overview of Model Refreshes

When you build the first lookalike segment for a particular source segment, the initial modeling job usually completes within about 24 hours (depending on system load).

After the initial model has been trained for a given source segment, you can create additional lookalike segments at different sizes and with different similarity settings much more quickly, because those segments re‑use the existing model.

At times, a lookalike model might be refreshed and generate an updated lookalike segment. A refresh occurs when there is a large change in the underlying dataset for the source segment that results in the composition of the source segment changing at least 5%-10%. 

Model refreshes are triggered in the following situations:

  • There is a change to the people in the modeling universe.

  • There is a change to the people in the source segment.

If you significantly change the underlying source segment (for example, by changing the segment rules), you should create a new lookalike configuration so that the model can re‑learn from the updated source.

In this section: