Build an Audience
Updated (Rahul) on 11/03/2026 - pending testing
Persona: Marketer
Overview
In this lab you will use Adobe CX Enterprise Coworker Chat to build an audience of high-value home loan prospects, using only natural-language prompts. Coworker finds the right fields, proposes targeting rules, shows how many profiles qualify as each rule is applied, and builds the audience only after you approve it. The lab has three acts:
Act Act 1: Do it Act 2: Understand it Act 3: Make it yours | What you do Run three scripted prompts and build the audience. See what Coworker did at each step and learn the pattern behind the prompts. Apply the pattern to your own audience and explore what else Coworker can do. |
|---|
Act 1: Do It
Audience Details
Identify existing, contactable customers who are the best candidates for a home loan campaign. They are high-value (by wealth or account value), show signs of home loan intent or propensity, don’t already hold a mortgage, and have given consent to be contacted.
Run the three prompts below in the same chat, in order. Each one ends with a stop so you stay in control.
Step 1: Discover, Reuse, and Propose Logical Rules
Goal: Find what you already have, see which of your data fields help identify high-value home loan prospects, and get a first set of targeting rules.
PROMPT A - copy and paste in a New Coworker Chat
What to look for | In the transcript, a line such as "Loaded skill" followed by what Coworker used to search your data. A list of existing audiences and datasets related to home loans, mortgages, or wealth. A list of fields grouped by purpose (for example value, propensity, consent), and an explicit note of anything missing. A short list of proposed rules, each with a reason |
|---|---|
Checkpoint | You have a list of candidate fields and a proposed rule set, and Coworker has stopped and is waiting for you. |
Your turn | Review the proposed rules. Decide what to keep, change, or remove. For example: raise or lower a value threshold, add an exclusion for customers who already have a mortgage, or require consent for a specific channel. |
Step 2: See where you lose people (waterfall) and check for duplicates
Goal: See how many profiles remain as each rule is applied, find which rule removes the most, and make sure you are not about to create a duplicate audience.
PROMPT B - copy and paste in the same chat as continuation after Prompt A results
What to look for | A waterfall (chart or table) with a profile count after each rule, ending in a final count. A statement of which rule removes the most profiles. A duplicate check result: either "no match" or a similar audience with the differences explained. |
|---|---|
Checkpoint | You know the final audience size, where the biggest drop-off is, and whether something similar already exists |
Your turn | Is the final number big enough to run a campaign? If the biggest rule is Coworker Audience Lab | Page 5 too restrictive, loosen it (for example lower a threshold) and re-run Prompt B with the updated rules. Counts are estimates and can change as profile data updates. |
Step 3: Review the plan, approve, and build
Goal: Create the audience only after you have reviewed Coworker's plan, and name it using the lab convention.
PROMPT C - copy and paste in the same chat as continuation after Prompt A results
What to look for | heck the rules, the audience name, and the evaluation type if it asks (batch or streaming). Then approve. A confirmation that the audience was created, with its name. |
|---|---|
Checkpoint | Your audience exists, and its name is exactly: High Value Home Loan Prospects - Coworker - [Your Initials] |
Your turn | If Coworker asks about permissions, grant them. Then open Real-Time CDP and find the audience to confirm it is there. |
Act 2: Understand It
You just built an audience in three prompts. This act, with no typing, shows what happened and why, so you can write prompts like this yourself.
What you asked | What Coworker does |
|---|---|
"Show me existing audiences and datasets related to home loans" | Searches the current sandbox for audiences and datasets that match the topic, so you can reuse what exists. |
"Tell me which customer details we have data for" | Searches your data model for fields by name, description, or value, and shows where each one is used. |
"Suggest rules... let me adjust them" | Turns fields into audience conditions you can edit. |
"Show how many people remain as each rule is applied" | Breaks the audience definition into individual conditions and shows how each one changes the size |
"Check whether an audience with this same definition already exists" | The audience-creation workflow asks for approval at each phase. |
"Name it ... - Coworker - " | Saves the audience (DRAFT) to Experience Platform using the name you specified. |
Act 3: Make It Yours
Now the guard rails come off. Use the pattern from Act 2 on your own use case, and try following Coworker's suggestions.
Choose any challenges below, in any order.
Challenge 1: Build your own audience Pick a different product or segment (for example, wealth management prospects, auto loan refinancing, or lapsed card holders)
Challenge 2: Follow Coworker's suggestions Coworker suggests follow-up actions after each response. Click one and see where it goes.
Then try the Lost your place? prompt from Act 1 to return to your original task.
Note how well Coworker holds context across a detour.