Predictive Audiences Overview

You can build predictive audiences, which are goal-oriented, dynamic, and adaptive. By contrast, real-time and standard audiences are static and based on historical data.

Create a predictive audience by specifying a user prediction in your regular audience creation workflow in mParticle. The user prediction accesses Cortex machine learning algorithms that analyze data on customer behavior, preferences, and interactions with your brands. You can use predictive audiences to anticipate the needs and desires of your target audiences, and thus deliver more relevant and personalized messaging. Predictive audiences maximize impact by automatically finding the best-fit users for a desired outcome.


In the past, you created a static (real-time) audience that identifies every user who has viewed shoes two or more times in the last week, and sent a coupon to all those users. However, now you can create a goal-oriented, predictive audience that contains all the users most likely to purchase shoes based on an analysis of all the available information, not just one factor (purchased shoes twice before).

Score and percentile values

When you create a predictive audience, you can choose between two types of results:

  • score, the likelihood of an audience member performing a future event
  • percentile, the likelihood of an audience member performing a future event as compared to other users scored by Cortex, mParticle’s machine-learning engine: typically the number of active users in the last 90 days

For example, if you wanted to know how likely it is that a user will purchase shoes in the next 7 days, you could see that likelihood displayed as a score or percentile:

  • With a score of 80%, the user has an 80% chance of purchasing shoes.
  • With a percentile value of .9, the user is more likely to purchase than 90% of users.

How it works

Creating a predictive audience is simple:

  1. Create an audience using the same criteria builder that you use for any audience or journey creation.
  2. In the criteria builder, select User Prediction and fill in the requested information.
  3. When you save the audience, mParticle creates two user attributes for the user prediction, one expressed as a score and the other as a percentile. Choose one.
  4. mParticle populates your audience. After a delay of up to 24 hours, your audience is available for use.
  5. Connect the output to the forwarding destinations as you do for real-time audiences.

For step-by-step instructions, see Using Predictive Audiences.

Predictions are rerun weekly to regenerate fresh predictions.


Predictions created in mParticle have the following limitations:

  • Predictions can only be created for custom event types.
  • Predictions use the future event prediction model in Cortex.

More about Cortex

Cortex is the machine-learning engine available with mParticle’s CDP. To learn more, you can visit the Cortex documentation.

Was this page helpful?