Build audiences with AI

Use the Agent in Customer Studio to build, refine, and understand audiences with natural language. Describe the audience you want, ask questions about an existing audience, or request changes. The Agent translates your input into audience filters, summaries, or analysis that you review before saving.

What you'll learn

In this article, you'll learn how to:

• Build audiences using natural language prompts
• Explore available fields, traits, and events in your schema
• Summarize, analyze, and explore existing audiences
• Refine audience definitions through follow-up prompts
• Debug audience size changes and unexpected results
• Improve results with more specific prompts and schema metadata
• Understand how Hightouch handles data when you use the Agent

Overview

The Agent is available when you build or edit an audience in Customer Studio. Select Agent to open a chat panel alongside the visual builder.

From the Agent panel, you can:

• Build audience definitions from natural language prompts
• Ask questions about an existing audience
• Get summaries and analysis of audience composition
• Refine audience filters through follow-up prompts
• Explore available events, traits, columns, and properties

Get started

Open the Agent

1. Go to Customer Studio → Audiences.
2. Create a new audience from scratch or from a template, or open an existing audience.
3. Open the Definition tab.
4. Select Agent.

Quick actions

The Agent shows different quick actions depending on whether you're working with a new or existing audience. Select a quick action to start, or type your own prompt.

Build audiences with natural language

Describe the audience you want to build, and the Agent drafts an audience definition with matching filters and conditions.

1. Open the Agent on a new or existing audience.

2. Enter a prompt that describes the audience you want to create. You can also attach images or files for additional context.

   

   For example:

   Create an audience of customers who added items to their cart in the last 30 days but did not complete an order.

3. Review the audience definition in the visual builder.

4. Check that the filters and conditions match your intent.

5. If the definition needs changes, send a follow-up prompt. The Agent keeps context from the conversation, so you can refine without starting over.

Prompt patterns for common tasks

These patterns show how to phrase requests for common audience-building workflows. Include specific field names, time windows, and thresholds when you know them.

Create an audience from criteria

Modify an existing audience

Refine iteratively

The Agent keeps context from your conversation, so you can refine step by step instead of writing one complex prompt.

Explore your schema

If you're not sure which fields, traits, events, or properties are available for filtering, ask the Agent to explore your schema. This is especially useful when working with a data model you didn't set up yourself.

Find the right field

Narrow your search

If the Agent returns too many fields or you can't find what you need, narrow your request:

• By topic: What fields relate to email engagement?
• By data type: Which columns contain dates or timestamps?
• By event: What properties does the Page Viewed event have?

When the Agent can't find a field

If the Agent says a field doesn't exist, check the following:

• The field might exist under a different name. Ask the Agent: List all columns that contain the word "email".
• The field might live on a related model rather than the parent model. Ask: What related models are available, and what columns do they have?
• The field might not be exposed in your schema. Column descriptions and event labels are set during schema configuration. If a field is missing, ask your data team to check whether it's included in the parent model or related models.

Understand and analyze audiences

Use the Agent to get plain-language explanations and analysis of an audience without writing queries or leaving the builder.

Select a quick action like Summarize audience or Generate insights, or enter your own prompt:

Summarize this audience in 30 words or less for a business user.

Which conditions are excluding the most members from this audience?

Break down this audience by most recent order value.

Analysis prompts

Refine and improve audiences

Use the Agent to update an existing audience definition through follow-up prompts. Describe the change you want, then review the updated definition in the visual builder.

You can also send a specific filter to the Agent from the visual builder. Select the more options menu on any filter and choose Add filter to chat.

Apply or undo changes

When the Agent suggests changes to your audience definition:

1. Select Apply these changes in the Agent panel to update the visual builder.
2. Review the updated filters and conditions in the visual builder.
3. Select Save audience to keep the changes, or Discard changes to revert to the original definition.

If you've already applied changes, select Undo in the banner at the top of the builder to revert before saving.

Troubleshoot audience issues

Use these patterns when an audience isn't behaving the way you expect. The Agent can help you diagnose size changes, understand exclusions, and identify configuration issues.

Audience is smaller than expected

Ask the Agent to break down which conditions are filtering out the most users:

Which filter is removing the most members from this audience?

How many users match each individual condition?

Common causes of unexpectedly small audiences:

• Overlapping conditions — Two filters that individually match many users can combine to match very few. Ask the Agent: How many users match both the purchase filter and the email filter together?
• Time windows too narrow — A 7-day window captures fewer users than a 30-day window. Ask: What if I change the time window to 30 days?
• Missing data — If a column contains null values, users with null values are excluded. Ask: How many users have a null value for this field?

Audience size changed unexpectedly

If an audience grew or shrank and you're not sure why:

When did this audience last change in size, and by how much?

What could explain a drop in this audience over the last week?

Common causes:

• Upstream data changes — The source data in your warehouse changed. Check whether recent data loads added, removed, or updated rows.
• Schema or relationship changes — A related model or trait definition changed, which affected filter evaluation.
• Time-based filters — Rolling time windows like "last 30 days" naturally shift which users qualify.

Agent gives an error or can't complete the request

If the Agent returns an error or says it can't complete your request:

• Unsupported data type — Some column types can't be used in filters directly. Ask: What data type is the [column name] column? If the column type is unsupported, you may need to create a computed trait or ask your data team to transform the column upstream.
• Cross-model limitations — The Agent works within the context of the current audience's parent model and its related models. It can't query across separate parent models or access data from other audiences.
• Query timeouts — If analysis queries take too long, try narrowing the scope: Analyze the first three conditions only.

When to use the Agent vs. the standalone Agents chat

The audience Agent works within the context of a single audience: building, editing, analyzing, or debugging one audience definition using the schema already loaded.

For broader tasks — campaign strategy, cross-audience comparisons, reporting, or questions that go beyond the current audience's schema — use Agents.

Tips for better results

Enrich your schema metadata

The Agent uses your schema metadata — model names, column descriptions, and event labels — to understand your data. Clear, descriptive metadata improves the accuracy of Agent responses more than any other factor.

To review and update your schema metadata, go to Customer Studio → Schema. Add or update:

• Column descriptions — Describe what each column contains in plain language.
• Event labels — Give events descriptive names that match how your team talks about them.
• Related model names — Name related models after the real-world objects they represent.

Be specific

Instead of Find my best customers, try:

Find customers who completed at least 3 orders in the last 90 days with an average order value over $100.

Include time windows, thresholds, and the specific events or filters you want. If you don't want the Agent to modify an existing filter, say so explicitly.

Ask about what's available before building

If you're new to a data model, start by exploring what's available before building the audience:

What events and properties are available for audience filtering?

What traits can I use?

This helps you write more precise prompts and avoids building on fields that don't exist.

Review before saving

Always review the audience definition in the visual builder before saving. The Agent can make mistakes, especially with complex logic, vague prompts, or schema fields with unclear names.

AI and data privacy

How data is processed

When you use the Agent, Hightouch sends relevant audience context to the AI model so it can generate a response. This context can include attributes, behavior signals, and schema metadata.

This data is:

• Not retained after the response is generated
• Not used to train or fine-tune models made available to other customers
• Governed by Hightouch's AI Terms of Service Addendum

Chat privacy

Your conversations with the Agent are private to you. Other team members can't see your audience-specific chats.