Composable Audiences Overview

Composable Audiences is mParticle’s zero-copy solution for segmentation. With Composable Audiences, you can create dynamic user segments using data stored directly in your cloud data warehouse. This approach lets you take advantage of rich, custom attributes and events that already live in your own systems, without needing to copy data into mParticle. By querying your warehouse directly, audiences are always built on the most up-to-date and complete picture of your users.

When to use Composable Audiences

Composable Audiences are the right choice when you care about a customer’s accumulated history, attributes, or modeled data—things that live in your warehouse. These audiences are ideal for planned, lifecycle-driven campaigns where hours or days of latency are acceptable. For example:

• High-value or loyal customers: Segment users with a lifetime value above $1,000 and more than 5 purchases in the past 12 months.
• Dormant customers: Identify users who haven’t made a purchase in 300+ days but are still opted into marketing emails.
• At-risk subscribers: Target users with a churn risk score greater than 0.7, calculated by your data science team.

Because these audiences run on a scheduled basis against your warehouse, they are well-suited for strategic campaigns like loyalty promotions, cross-sell recommendations, or reactivation offers.

When to use Real-Time Audiences

Real-time audiences are best for reactive, intent-driven triggers where seconds or minutes matter. These audiences use streaming data ingested directly into mParticle and update in real time. For example:

• Cart abandonment: Reach out within 60 minutes of a user abandoning their cart.
• Flight delay notifications: Send travel alerts when an itinerary changes.
• Streaming engagement: Promote more content immediately after a subscriber finishes a show.

If the value of your engagement depends on acting while intent is fresh, real-time audiences are the right fit. See the Audiences Overview to get started with Real-Time Audiences.

Setting up Composable Audiences

Before you can start building audiences, there are two parts to the setup process:

1. Warehouse setup: This step is typically done by a data engineer or database administrator. It includes:

   • Connecting mParticle to your cloud data warehouse (Snowflake, BigQuery, Redshift, or Databricks).
   • Creating data models that define which warehouse tables and fields will be available in the Audience Builder.

2. Audience setup: Once the connection and models are in place, a marketer or analyst can:

   • Use the Audience Builder to define membership criteria based on the data models.
   • Connect the audience to downstream platforms like Google Ads, Facebook, or TikTok.

Together, these steps give you a zero-copy way to create audiences: your data stays in your warehouse, while mParticle queries it securely on your schedule to keep audience membership up to date.

To get started, go to the User Guide.

How is data from your warehouse processed?

When a composable audience initializes, mParticle translates your audience definition (the criteria that determine which users should be included in the audience) into a SQL query. This query is then executed directly within your data warehouse using your warehouse’s compute resources.

The SQL query returns a list of users and (optional) attributes that make up the audience at that particular point in time. This data is stored in a temporary table in your data warehouse that is provisioned for internal processing only. These datasets are not part of your production tables and are only used by mParticle to track audience membership over time. These additional datasets are automatically managed and cleaned up by mParticle.

mParticle pulls the users and any optional attribute data from your warehouse before forwarding them to your connected downstream integration.

When you create a composable audience, you specify the cadence at which it refreshes. Note that composable audiences can’t be refreshed in real-time.

Composable Audiences operates independently of mParticle’s identity resolution framework (IDSync). No user profile enrichment or merging takes place. Audience membership is based solely on the logic provided in the SQL query and the structure of the source data. This makes the feature highly transparent and deterministic, giving you full control over how your audiences are built and updated. It also means that the records in your data warehouse must include some form of user identifier (like a user ID or customer ID) that can be used when determining audience membership.

What data is stored in mParticle?

mParticle only stores data about your warehouse that’s needed to create and activate composable audiences, including:

• Information needed to connect with, and authenticate to, your warehouse provider.
• Metadata about your warehouse database and tables, like column names.
• If explicitly enabled, mParticle caches up to 1000 of the most frequent values from each column in your table for up to 7 days to power the autocomplete feature within the audience builder. You must enable autocomplete for each column included in your data model.
• A future release of Composable Audiences will allow you to create “hybrid composable audiences” in which composable audience membership can be optionally noted on user profiles in Customer 360 or the User Activity View.

No other data is stored in mParticle beyond these items.

How are your user profiles modified?

Composable Audiences does not modify your user profiles in any way.

A future release will add the ability to optionally have composable audience membership reflected on user profiles.

How are composable audiences refreshed?

To improve performance, mParticle only tracks changes in audience membership (specifically, the users who have been added to or removed from the audience) rather than recomputing the full list each time. The first time the audience is executed, the entire set of qualifying users is returned. For subsequent runs, mParticle compares the new audience snapshot with the previous one to determine which users have been added or dropped.

This “diff-ing” logic happens entirely within the data warehouse environment. Two snapshots of the audience (previous and current) are compared using SQL, and only the incremental changes are passed back into mParticle for forwarding to downstream integrations. This approach ensures both scalability and efficiency in audience refresh cycles.

Are composable audiences compatible with Real-Time mParticle Audiences?

While support for this functionality is planned for a future release, you can’t currently clone audiences comprising data stored in the mParticle CDP from a composable audience, nor can you reference audiences built with mParticle data from a composable audience using the isMemberOf() function (or vice versa).