You are here:
Build a Segment by Describing It
Describe the audience you want in plain language, and the Data 360 Marketing Agent works out the filter criteria and creates a segment for you. The agent builds standard and nested segments, and filters can include calculated insights.
Required Editions
| Available in: All Editions supported by Data 360. See Data 360 edition availability. |
| Permission Sets Needed | |
|---|---|
| To create a segment: | Any one of these permission sets:
|
Use this task when you aren’t on the segment canvas. To add filters to a segment you already have open, see Add Filters to a Segment.
-
Go to the Segments page and click the Agentforce icon in the
global header.
If the panel opens on a different agent, select Data 360 Marketing Agent from the menu next to the agent name. Stay on this page for the rest of the conversation. Don’t open a segment on the canvas in the middle of the request.
-
Describe the audience you want, including the conditions, how they combine, and anyone
you want to leave out.
For example: Create a segment of loyalty members with at least three orders in the past year, but exclude anyone who unsubscribed.You don’t need to name objects or fields, but naming the object directly saves the agent a question. Include any counts or totals you have in mind. If you say at least three orders, the agent uses three rather than asking.
-
Choose the segment type and publish schedule.
The agent asks this first, before anything about your audience. It proposes a segment that you publish manually, and offers two scheduled alternatives: a schedule of every 12 or 24 hours, or rapid refresh of every 1 or 4 hours.
-
Answer the agent’s questions as they come up.
When the agent needs to resolve several attributes or values, it asks about up to three at a time, as a numbered list, so you can answer them all in one reply.
- If a phrase in your request could match more than one attribute, the agent lists the candidates and asks which one you mean, because your data model has several attributes that fit the description. Answer with something like: Use the loyalty tier field, not the membership status one. To learn more about an attribute, ask the Data 360 Data Q&A Agent.
- If none of the attributes the agent offers is the one you meant, describe the attribute you actually want. The agent needs the right attribute before it can continue, and it rebuilds from your answer. Answer with something like: Neither of those, I mean their lifetime purchase total.
- If more than one object could be your Segment On object, the agent lists the candidates and asks you to choose, because the population depends on which object the segment is built on. Answer with the object name, such as: Individual.
- If your request could be grouped more than one way, the agent shows the possible groupings and asks you to confirm, because how the conditions group changes the population. Answer with the grouping you want.
- If a value is missing or described loosely and can’t be inferred from your data, the agent asks for it, because guessing would silently change your population. Counts and totals for aggregations are always asked. Answer with an exact date range. For example: December 12, 2025, to August 20, 2026.
- If an attribute lives on a related object, the agent asks which related records to count, because the condition has to be expressed against the object you’re segmenting on. Answer with something like: Count orders.
Each option the agent offers names the attribute, its object, and where the object sits relative to your Segment On object. For example, Gender Identity (Individual), directly on the Individual record is a direct attribute, and Gender (AccountContact), on the related Account Contact record is a related attribute. See Segment Canvas.
If an attribute you name can’t be reached from your Segment On object, the agent tells you and offers a different attribute. Whether a path exists depends on the derived relationships your match and reconciliation rules generate.
If you correct an attribute after the agent has already used it, the agent rebuilds from your new answer. You don’t need to start the conversation over.
- Review the criteria in the segment canvas before you publish or activate.
The agent generates a name and description for the segment, keeping the name to 60 characters or fewer. In testing, a number was appended to keep the name unique, so the saved name won’t always match the name the agent quotes in the conversation. You can also name the segment yourself, as in Save this as a new segment called High Value West Coast.
Open the segment on the segment canvas and check the criteria before you publish or activate.
Check these in particular:
- Every condition you asked for is present. A request can be matched to an existing calculated insight that satisfies part of it, leaving the rest out. For example, a request for customers who spent over $500 last quarter can produce a spend filter with no date filter at all. The agent can also skip data model objects (DMOs) or calculated insights that aren’t mapped or that don’t appear in the attribute library, and continue with the filters it can create. Confirm that nothing you asked for is missing.
- Values the agent inferred. The agent fills in values you left out or described loosely. Check the date range it resolved a phrase like last quarter to.
- The operator on each condition. The agent assigns an operator and a value without asking. See Segmentation Operators in Data 360.
- Exclusions. Anything you asked to leave out should appear in the Exclude tab, not as a regular filter in Include.
Click Preview to check the audience itself, and Calculate population to get counts for individual filters and containers. See Segment Canvas.
Keep going in the same conversation. Ask for a different value, a different attribute, or a different audience, and the agent rebuilds from your new request. Refinements update the same segment. Asking for a different segment creates a new one.
The agent adds and builds criteria. It can’t remove or edit existing filters; delete those in the segment canvas. Don't combine an addition and a removal in one request: the agent attempts both, adds the new criteria, and leaves the filter you asked it to remove in place.
