Building Audiences with the Agent

This guide explains how to build an audience with the Audience Agent, from describing the audience you want or your business goal, to reviewing and iterating on the audience definitions the agent suggests, to saving a final audience proposal.

Step 1: Open the Audience Agent

1. Go to Segmentation > Audiences using the left-hand navigation.
2. Click + New Audience.
3. Under Build with AI assistance, select Build with mP Agent.

The agent panel opens with a message box and a set of 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 take the form of a specific question about your data, a specific audience (“users who purchased in the last 7 days”) you want to build, or a broad business goal you’re trying to reach (“re-engage customers who lapsed after one purchase”). You don’t need to know your exact event or attribute names.

If you ask about your data, the agent replies with summaries, tables, and sometimes graphs about the data in your workspace instead of an audience definition, such as:

• The events and attributes available, and which look most useful
• What a particular signal means, how it’s currently used, and its recent volume and data sources
• A summary of the audiences you already have

Use what you learn to decide what to build. When you describe an audience or goal, the agent works toward a definition:

• If your request is specific, the agent proposes a definition directly.
• If your request is open-ended, the agent proposes a sensible default and tells you what it assumed, so you can review the response and adjust accordingly.
• If the agent needs more detail to proceed, it asks follow-up questions.

Step 3: Review the agent’s definitions

The agent works in two stages: first it prepares one or more audience definitions (up to three in a single message) for you to review and refine. When you’re happy with one of the audience definitions, you turn it into a single audience proposal that you can save as an audience.

Each definition appears as a card labeled Definition A, Definition B, and so on. Above the card, the agent summarizes the suggested name, the membership criteria, the pattern it drew on, and the refinement it thinks would help most.

Read the summary and the following details carefully to confirm the agent interpreted your intent correctly before moving on.

A definition card includes:

Name and size

The header shows the suggested audience name and an estimated size. While the agent calculates, the size shows a loading indicator. If the population is too small to estimate reliably, a warning icon appears instead of a number. Hover your cursor over the warning icon to see the reason, for example: “Population is too small to accurately estimate.”

Membership criteria

Each row states a rule 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 to show they were resolved against real entities in your workspace.

Data sources

Each row shows how many platforms it draws from, such as “From 3 platforms.” You can hover your cursor over this to open the data sources popover, which lists the platforms and feeds included. Click Update platform(s) or feed selection in that popover to adjust them.

If you work in more than one workspace, you can change the workspace a definition draws from by hovering over “From … workspace” and clicking Update workspace(s) selection.

Create Audience Proposal

When you’re happy with an audience definition, click Create Audience Proposal to turn the definition into an audience proposal (see Step 5).

Step 4: Refine the definition

If a definition is too broad, too narrow, or doesn’t capture the exact user segment you want to target, reply conversationally with a follow-up like “make it the last 14 days” or “only users who purchased more than 3 times.” The agent revises the definition and adds a new one.

Each definition is tracked in the audience definition bar at the top of the panel, where every definition (Definition A, Definition B, and so on) shows its own size estimate. Click a definition to jump back to it in the conversation, so you can compare definitions before deciding which one to turn into a proposal.

When you’re satisfied with a definition, click Create Audience Proposal on its card to move to the proposal stage.

Step 5: Review and save the proposal

After you click Create Audience Proposal, the agent presents the audience proposal card with a label like Audience Proposal #1 and a note on the definition it’s built from, such as “uses Definition B.” Above the card, the agent states the estimated size and its confidence, and reminds you that you can save the proposal as-is or keep refining.

The proposal card summarizes the finalized audience:

• Audience Name: the exact name that will be given the audience when it’s created.
• Environment: the data environment the audience is built on (either Development or Production).
• Refresh Frequency: how frequently the audience will be updated. This is currently limited to real-time.
• Audience Definition: the membership criteria used to select audience members.

Nothing is saved until you click Proceed to Save. To return to drafting, click Continue Refining.

When you click Continue Refining, you can keep adjusting the definition and create additional proposals. When a conversation has more than one, each proposal is numbered and tracked in the proposals bar at the top of the panel, alongside the audience definition bar. Click a proposal to jump back to it in the conversation, the same way you move between definitions, so you can compare proposals before deciding which one to save. Remember that you can save only one proposal per conversation.

After saving, the audience appears in your Audiences list and behaves like any other audience. To connect and activate it, follow Connect an Audience. The agent does not connect audiences to destinations.

Conversation controls

Several controls are available in the agent panel itself to help you manage your conversation:

• Copy chat (top right of the panel): copies the entire conversation. Use it to keep a reference for future conversations or share a transcript with a teammate.
• Copy message (below any agent response): copies that response only.
• Rate response (the thumbs-up and thumbs-down icons below a response): sends feedback about that answer to mParticle. Feedback like this is one of the ways mParticle improves the agent.

Troubleshooting

The size of my audience shows a warning instead of a number.

Hover the warning for the reason. “Population is too small to accurately estimate” means too few users currently match. Change the definition (for example, broaden it using a longer time window), and the size will recalculate.

The agent says it can’t find a signal for what I asked.

The agent only uses signals that exist in your workspace and won’t substitute a weak proxy. If your request needs a metric that isn’t instrumented, it explains the dependency.

The agent says my request isn’t supported.

Multi-node structures, editing existing audiences, scheduled audiences, and audience activation are out of scope for this release. The agent explains the limitation and offers the closest useful next step. For manual control, use the Audience Builder.