Natural Language Audience Builder
Describe the audience you want in plain English and Zeotap Agent proposes the filter conditions, which you review and apply into the visual builder.
How to Use It
- Navigate to Audiences in the sidebar and click Create Audience
- Pick the Parent model the audience is built on, then continue to Build filters
- The agent panel opens by itself at this step — describe the audience you want in the chat
- The agent replies with a Suggested audience filters card: the conditions it proposes, why, and roughly how many members match
- Click Apply to builder to drop them into the filter builder, or Dismiss to ignore them
- Adjust anything you want by hand, then continue to Name & save
The panel stays open while you work, so you can ask for a change — “also exclude anyone who unsubscribed” — and apply the next suggestion on top of what you have.
Applying a suggestion replaces the filter tree in the builder rather than merging with it. If you have hand-built conditions you want to keep, ask the agent to include them in its next suggestion instead of applying over them.
What the Agent Works From
Before suggesting anything the agent looks at your workspace: the columns on the parent model, the relationships it participates in, the computed attributes defined on it, and your existing audiences. Conditions come back referencing your real schema, and it can reuse an existing audience as a building block when your description implies one.
It can also estimate as it goes, which is where the member count on the card comes from — the estimate runs against your warehouse without saving anything.
Examples
| What you describe | What you get back |
|---|---|
| ”Customers who spent over $500 in the last 90 days” | A condition on your lifetime-value computed attribute, plus a purchase-recency condition |
| ”Users who signed up this month but haven’t purchased” | A signup-date condition ANDed with an order-count condition of zero |
| ”High-value customers in the US or Canada” | A value threshold ANDed with a country condition matching either code |
| ”Everyone in my Black Friday audience who has since churned” | A reference to the existing audience, ANDed with an inactivity condition |
The exact columns depend on your schema — the agent uses what your models actually expose, not a fixed vocabulary.
Tips for Better Results
- Be specific about numbers. “Spent over $500” gives the agent a threshold to work with; “high spenders” makes it guess one.
- Say the time frame. “In the last 90 days” is the difference between a date filter and none.
- Name the computed attribute if you know it exists — the agent will use it rather than rebuilding the same logic inline.
- Iterate in the panel. Start broad, look at the estimate, then narrow. Each round is cheaper than rebuilding by hand.
- Give the workspace context it needs once. Business background and naming conventions belong under Settings → General, where they apply to every session — see AI Agent.
Next Steps
- AI Agent — everything else the agent can do
- Audiences — the filter builder the suggestions land in
- Guardrails — limits on what the agent can create on your behalf