Build Audiences

This guide explains how to use the mParticle Agent to prepare, review, refine, and save a real-time audience.

• Start with a business goal: Describe the customers you want to reach without knowing the exact event or attribute names in advance.
• Evaluate data before building: Check whether the relevant events for your audience are available, what data inputs provide them, and whether there’s enough activity and coverage for them to be useful.
• Review criteria before saving: Check the criteria the agent proposes and refine them conversationally before you save the audience.

For example, you can ask the agent to identify customers who lapsed after an initial purchase, customers showing recent purchase intent, or high-value customers who are actively browsing.

Step 1: Open the mParticle Agent

You can open the mParticle Agent directly from the left-hand navigation by clicking mParticle Agent.

You can also open the mParticle Agent from Segmentation by clicking + New Audience and selecting Build with mP Agent. The agent panel will open with a message box and starter prompts for common ways to begin, such as I have a campaign goal, help me define the right audience and Show me what data I have to work with.

Step 2: Describe your goal or ask about your data

In the message box, ask a question or describe what you’re trying to achieve using plain language. Your query can be a specific question about your data, a specific audience such as users who purchased in the last 7 days, or a broad business goal such as re-engage customers who lapsed after one purchase. You don’t need to know the exact event or attribute names in your workspace.

If you ask about your data, the agent replies with summaries, tables, and sometimes graphs about the data available in your workspace instead of an audience definition. For example, it can describe:

• The events and attributes available and which ones might support your goal.
• What an event or attribute represents when a definition is available, which sources report it, and its recent volume when an aggregate query returns results.
• The saved audiences visible in your workspace.

To learn how best to explore available data using the agent, see Explore Your Data with the mParticle Agent.

Use what you learn to decide exactly what audience to create; when you describe an audience or goal, the agent works toward a definition:

• If your request is specific, the agent proposes a definition right away.
• If your request is open-ended, the agent makes and explains its assumptions so you can review and adjust them if necessary.
• If the agent needs more details to proceed, it asks follow-up questions.

You can also ask the agent to review pre-existing audiences. Available information can include metadata, a summary of the audience’s criteria, a complete saved definition, and whether the audience is eligible for use as an Audience Membership criterion. When a complete definition isn’t available, the agent discusses only the available summary and doesn’t infer missing criteria from the audience name. The agent doesn’t edit the saved original or expose its members.

For example, ask:

• What saved audiences are available in this workspace?
• What does this audience’s available definition do?
• Can I use this audience as an Audience Membership criterion?

Use membership in other audiences as part of your criteria

Ask the mParticle Agent to use an eligible saved audience as an Audience Membership criterion in a new real-time audience. It can use an eligible existing Composable Audience for this purpose. When you combine that membership with live behavioral criteria, the result is a hybrid real-time audience.

If a Composable Audience cannot be used, the mParticle Agent explains the limitation so you can choose another supported source or workflow. For details about Composable Audience membership and identity matching, see Hybrid Audiences.

Step 3: Review the definitions

The agent works in 2 stages. First, it prepares 1 or more audience definitions for you to review and refine. When you’re satisfied with a definition, you turn it into an audience proposal that you can save.

Each definition appears as a card labeled Definition A, Definition B, and so on. Above the card, the agent summarizes the suggested name, membership criteria, the pattern it used, and a possible refinement. Read the summary and definition details carefully to confirm that the agent interpreted your intent correctly before you continue.

Name and size

The card header shows the suggested audience name and an estimated size. While the agent calculates the estimate, the size shows a loading indicator. If the population is too small to estimate reliably, a warning icon appears instead of a number. Hover over the warning icon to see the reason.

Membership criteria

Each row states a criterion in plain language, such as performed page view WHERE url contains “checkout” more than 1 time in the last 180 days. Event and attribute names appear in color when the agent resolves them to entities in your workspace.

Data sources

Each criterion shows the number of platforms or feeds it uses, such as From 3 platforms. Hover over the count to open the data sources popover, which lists the included platforms and feeds. Click Update platform(s) or feed selection in the popover to adjust them.

If you work in multiple workspaces, hover over the workspace count and click Update workspace(s) selection to change the workspace the definition uses.

Step 4: Refine the definition

Reply with the change you want if the definition is too broad, too narrow, or doesn’t capture the customers you want to reach. For example, ask the agent to use the last 14 days or include only customers who purchased more than 3 times.

The agent adds each revised definition to the audience definition bar at the top of the panel. Each entry shows the definition label and its own size estimate. Select a definition to return to it in the conversation, review its criteria, and compare it with other versions before deciding which one to turn into a proposal.

Step 5: Create and review the audience proposal

1. Click Create Audience Proposal when the definition reflects the audience you want to create.
2. Review the proposed name, environment, refresh frequency, and final audience definition on the audience proposal card.

The proposal card identifies the definition it was built from and shows its estimated size and the agent’s confidence when available. It displays the proposed audience’s:

• Audience Name: The name the audience would receive when you save it.
• Environment: Whether the audience would use the Development or Production data environment.
• Refresh Frequency: How frequently mParticle would update the audience. The mParticle Agent can currently only create real-time audiences.
• Audience Definition: The membership criteria used to select audience members.

If you want to change the definition, click Continue Refining. You can adjust the definition and create another proposal. Each proposal is numbered and remains available in the proposals bar alongside the audience definition bar. Select a proposal to return to it in the conversation and compare it with other proposals before deciding which one to save.

Review before saving

Before saving the proposal, confirm:

• Name and environment: The audience has the intended name and uses the correct data environment.
• Membership criteria: Every condition represents the customers you want to include.
• Time windows and exclusions: Lookback periods, thresholds, and exclusions match the campaign goal.
• Data sources: Each condition uses the intended platforms and feeds.
• Estimated size: The estimated audience size is reasonable for the intended use.

Step 6: Confirm and save the audience

Click Proceed to Save. Nothing is saved until you confirm the final proposal.

The agent doesn’t activate the saved audience or connect it to an output. You must manually activate the audience and manually connect it to an output using the mParticle UI.

Saving an audience preserves the proposal you selected, but it doesn’t preserve the other definitions, proposals, or rationale generated throughout the conversation. Use Copy chat before closing the panel if you want to retain that material.

Conversation controls

Use the controls in the agent panel to manage the conversation:

• Copy chat: Use the control at the top right of the panel to copy the complete conversation for later or share it with a teammate.
• Copy message: Use the control below an agent response to copy only that response.
• Rate response: Use the thumbs-up and thumbs-down icons below a response to send feedback about that answer to mParticle.

If you close the panel without saving an audience, the exit dialog confirms that your conversation progress won’t be saved and offers an option to download a copy. After you save an audience, closing the panel doesn’t prompt you to save the conversation.

Troubleshooting audience building issues

A saved audience doesn’t show a complete definition

The information available for a saved audience can vary. When a complete saved definition is available, the agent can explain it. Otherwise, it can discuss only the available summary and won’t infer missing criteria from the audience name.

Open the saved audience in the Audience Builder to review or edit information that isn’t available to the agent.

A composable audience isn’t available as an Audience Membership criterion

Review the Agent’s explanation for why the selected Composable Audience cannot be used. Choose another eligible audience or see Hybrid Audiences for details about how Composable Audience membership becomes available in real-time audiences.

The agent says the request isn’t supported

The mParticle Agent prepares one new individual real-time audience per creation flow. It can’t edit an existing audience, create or restructure an Audience Group, create a Composable Audience, activate an audience, or connect an audience to an output.

If the goal requires an Audience Group or changes to a saved audience, continue in the Audience Builder.

The audience size shows a warning instead of a number

Hover over the warning icon to see why the agent couldn’t provide an estimate. If the population is too small to estimate reliably, broaden the definition and wait for the size to recalculate.

Next steps

• Connect an Audience: learn how to activate your new audience and connect it to an output,
• FAQ: find answers to common questions about the mParticle Agent.