Similar Customer Predictions Overview

Similar customer predictions help you find more users who resemble a reference segment you already understand. They are useful when you have a valuable audience, such as high-value customers or users with a declared preference, but many other user profiles are missing the attribute you would normally use to include them.

Unlike future behavior predictions, which estimate whether a user is likely to take a future action, similar customer predictions identify users who look like a known group. A Similar Customer prediction appears as a user attribute on a user profile in mParticle, so you can build and activate audiences based on how closely users align with that group.

Data requirements

Similar Customer Predictions have specific data and environment requirements. Reviewing these before you create a prediction helps you avoid failures and get reliable results.

Audience type

Your reference audience must be defined only with user attributes, such as preferences, profile fields, or other known traits. Audiences that include events, predictive attributes, calculated attributes, or group attributes aren’t eligible for Similar Customer Predictions. Use Future Behavior predictions when you want to predict who is likely to perform an event or behavior.

Environment

Data must be in the Production environment. Similar Customer Predictions won’t run on data from the Development environment. If your audience was built using development data, you’ll need to recreate it in production before you can use this feature.

User data requirements

Similar Customer Predictions work by comparing users who match your reference segment with users who have a different known value. mParticle then makes a prediction for users where the relevant attribute is blank, helping you decide which of those unknown users are similar enough to target.

Your data must include three groups:

• Matching users: users who match the audience you want to expand; for example, subscription_tier = Premium. This is your reference group. At least 500 matching users are required.
• Known non-matching users: users who have a different known value for the same attribute(s); for example, subscription_tier = Free. Ideally you will have at least 500 known non-matching users to improve the prediction strength.
• Unknown users: users where the attribute(s) are blank or missing. These are the users mParticle scores for similarity. At least 500 unknown users are required.

The matching group should be smaller than the total population, with enough known non-matching users to compare against and enough unknown users to expand into. For the clearest setup, start with a single user attribute that has at least one known value for the audience you want to expand, at least one different known value for comparison, and a large population where that attribute is missing.

Volume guidance

The 500-user minimum per group is required for the prediction to run at all. For reliable results, aim for:

• 2,000 or more matching users
• 3,000 or more non-matching users
• 10,000 or more unknown users

Example

If your audience is subscription_tier = Premium, a well-suited dataset might look like:

• 2,500 users where subscription_tier = Premium (matching)
• 4,000 users where subscription_tier = Free (known non-matching)
• 12,000 users where subscription_tier is blank (unknown users to predict)

If requirements aren’t met

If any of the required groups are below 500 users, the prediction can fail or produce results that are not useful for targeting. If the prediction does complete but the dataset is small or uneven, it may have Weak strength, which means results should not be used for campaign targeting.

Historical data

Similar Customer Predictions require at least 45 days of historical user data. For reliable results, 90 or more days of historical data is recommended.

Use case examples

Expand an existing audience

Use similar customer predictions when you already have an audience you trust, and you want to reach additional users who resemble that audience.

Step 1: Define a reference segment

Start with your existing audience. For example, select an audience like “dietary preferences = Vegetarians” as your reference segment. When you create the prediction, this audience is used as the basis for the prediction.

Step 2: Identify eligible users

mParticle evaluates users who are missing the attribute value used to define the reference segment and identifies which of those users resemble your reference segment.

Step 3: Create and activate a similar customer audience

Create the prediction, then choose a targeting range such as Most similar customers for higher precision or All similar customers for broader reach. Save the range as a new audience, choose Predictive Audience (only predicted users) or Expanded Audience (predicted users and your reference segment users), then connect the audience to an output for activation.

Scaling a high-value customer audience for cross-sell

Use a similar customer prediction when you have a segment you trust, such as high-value customers, and you want to find more users who match the same patterns.

Step 1: Define a reference segment

Select an existing audience that represents your high-value customers, or define the segment using profile criteria that capture the traits you care about.

Step 2: Identify eligible users

mParticle evaluates users who are missing the attribute value used to define the segment and identifies which users resemble the reference segment.

Step 3: Create and activate a similar customer audience

Create the prediction, choose a targeting range that balances reach and precision, then save it as a new audience. Use the resulting audience for upsell campaigns or personalized experiences.

Finding users similar to a survey or preference panel segment

Use a similar customer prediction when only some users have completed a survey or provided explicit preference signals, but you want to reach more users who match the same patterns.

Step 1: Define a reference segment

Start with users who are members of an audience built from survey responses or panel membership, or build a reference segment from those known attribute values.

Step 2: Identify eligible users

mParticle evaluates users missing those survey-derived attributes and identifies which of them resemble your reference segment.

Step 3: Create and activate a similar customer audience

Create the prediction, then save a “most similar” percentile range as a new audience. Activate that audience to extend the same messaging, recommendations, or offers to users who align with the segment but did not complete the survey.

Audience expansion

Audience Expansion is the guided workflow you start from the audience builder. It is not a separate prediction type. When you open an existing audience and click Expand with AI, mParticle automatically determines which prediction to create based on how that audience is defined:

• User-attribute-only audiences routes to a Similar Customer Prediction. mParticle uses the audience as the reference segment automatically.
• Event-based or mixed audiences routes to a Future Behavior Prediction. You will be prompted to define the future behavior you want to predict and the users you want to evaluate.

Audience expansion eligibility

The Expand with AI button appears on audiences that meet the eligibility requirements for at least one prediction type and are not currently calculating. Audiences that include predictive attributes, calculated attributes, or group attributes are not eligible for Audience Expansion.

For audiences defined using only user-attribute criteria, the same data requirements apply as when creating a Similar Customer Prediction directly: the Production environment must be used, and your data must include the three required groups. See the data requirements above.

Audience expansion and existing downstream connections

Audience expansion always creates a new, separate audience. If you expand an audience that is already connected to downstream outputs or campaigns, those connections remain on the original audience. You will need to connect the new expanded audience to your downstream destinations separately.

As a workaround, you can add the expanded audience as a membership criterion in your original audience definition, so that users identified by the prediction are included without disrupting existing connections.