Audience Agent Overview

The mParticle Audience Agent provides a conversational way to build audiences. Instead of building your audience membership criteria by hand, you describe your target audience or the goals you’re trying to achieve in plain language. The agent queries data catalog metadata, aggregate metrics, saved audience context, and user data (limited to what your own user permissions allow), finds the right signals for what you described, and proposes an audience definition for you to review and refine.

The difference from the manual Audience Builder is that you don’t need to know the exact audience criteria in advance. You can describe an outcome, like “users who lapsed after their first purchase” or “people likely to upgrade.” If a relevant prediction or audience-ready signal exists, the agent surfaces the best signals to use, explains what it chose and why, and, when available, shows an estimated audience size so you can make adjustments before saving.

When to use the Audience Agent

Use the Audience Agent when you want to:

• Build a real-time, single-node audience: Describe a goal and let the agent find the relevant events and attributes in your workspace, instead of hunting for the right events yourself.
• Move from idea to definition quickly: Turn a campaign goal into a reviewable audience in a single conversation, without filing a request to your data team for standard segments.
• Check your data before you build: See what’s available and how large an audience may be before committing to a definition, so you’re not guessing at what your workspace can support.

When to use the manual builder instead

Use the manual Audience Builder when you need to create a multi-node audience (an audience group with multiple parent/child relationships) or when you are editing an existing audience. The Audience Agent is designed to help you create new, single-node, real-time audiences only.

How it works

The Audience Agent works as an assisted drafting tool. When you describe what you want, the agent interprets your intent, queries your workspace’s data catalog to identify candidate events and attributes, and presents an audience definition in plain language alongside the underlying criteria. You review every proposal made by the agent, and only after you confirm the final definition and click Proceed to Save is the audience saved. From there, you connect and activate it through the same flow you use for any audience.

Because the agent reasons over your real schema, it only uses events and attributes that actually exist in your workspace. When matching specific attribute values, it uses the data available to it. When it can’t access the right value or the value is unclear, it asks you to clarify rather than guessing.

The Audience Agent operates within your existing mParticle permissions and doesn’t expose personally identifiable information (PII) in the conversation. It uses a large language model from OpenAI to interpret your requests, and under mParticle’s enterprise terms, your prompts and data aren’t used to train OpenAI’s models. For the full picture of what the agent can access, how your data is secured, and where it’s processed, see Data and Privacy.

What you can do with the agent

The agent is more than a replacement for manually building audience definitions. Because you work with it conversationally, it can help you decide what to build in the first place. In one session, it can:

• Explore available signals: Ask what events and attributes exist for a given concept, and which look most useful.
• Answer data questions: Check things like event volume, identity coverage, audience sizes, or where audiences are being forwarded.
• Review existing audiences: Summarize audiences you already have, explain their logic, and avoid building duplicates.
• Refine an idea: Compare candidate signals, weigh precision against reach, and tighten a vague goal into a clear definition.
• Build a new audience: Turn a goal and findings about your data into a net-new, real-time, single-node audience.

What’s not supported in this release

The agent only supports real-time, single-node audiences. It does not currently support:

• Creating multi-node audiences (audience groups with parent/child relationships) or composable audiences.
• Editing, modifying, or deleting existing audiences.
• Creating calculated or predictive attributes, even as part of an audience definition. If your request depends on a calculated or predictive attribute that doesn’t exist (for example, “average order value”), the agent explains the dependency and points you to where to create it, but won’t create it for you.
• Building audiences that use audience membership as criteria.
• Activating audiences or connecting them to destinations. You must manually activate and connect audiences through the normal flow after saving.

If your request is outside the agent’s supported scope, the agent explains what it can and can’t do, then offers the closest useful next step, such as refining the audience logic or pointing you to relevant documentation.

Frequently asked questions

Does the agent create audiences automatically?

No. The agent proposes an audience definition for you to review. Nothing is saved until you confirm the final definition and click Proceed to Save.

Can the agent edit existing audiences?

No. The agent only creates new audiences. It does not edit, modify, or delete existing ones.

What types of audiences does this release support?

Single-node, real-time audiences only: single-level audiences with no parent/child or grouped structure. Multi-node structures, composable audiences, and audiences that reference other audiences are not supported in this release.

What happens if the agent doesn’t understand my request?

It tells you what it could and couldn’t resolve, then asks a focused question or offers the closest useful next step. It won’t guess silently or invent event or attribute names.

What if my request needs a calculated or predictive attribute?

The Audience Agent can create audiences that use pre-existing calculated and predictive attributes. If your request depends on a calculated or predictive attribute that doesn’t exist (for example, “average order value over $100”), the agent explains the dependency and points you to where to create it. It does not create calculated or predictive attributes inline. See Calculated Attributes and Predictions for more information.

Does the agent support Match Boost, Household Reach, and Audience Expansion?

The Audience Agent can’t apply these features while building an audience, but you can apply them through the normal audience flow after saving.

Does the agent use my real data?

Yes. It reasons from your real workspace schema (events, attributes, and their metadata) and never fabricates schema entities. See Data and Privacy.

Is my data used to train models?

No. Under mParticle’s enterprise terms with OpenAI, your prompts and data are not used to train or improve OpenAI’s models. mParticle may review agent conversations to improve the product, but it doesn’t use your proprietary data to improve the agent for other customers.

Where is my data processed?

By default, the Audience Agent processes data in the United States. See Data and Privacy.

Can the agent activate audiences to destinations?

No. The agent displays the audience definition and you confirm and save it. Activation to downstream destinations works the same way it does today. See Connect an Audience to learn more about connecting audiences.

Next steps

• To build your first audience with the agent, see Building Audiences with the Agent.
• For what the agent can access and how it handles your data, see Data and Privacy.