How do you build a synthetic audience of your customers?
TL;DR
- Start with the decision, not the demographics. The traits that matter depend on what you're testing.
- Define the audience with behavior first (what they bought, how often), then demographics, then attitudes.
- Match the mix to real population data. Public sources like the American Community Survey, sent to about 3.5 million addresses a year, give the targets.
- Ask the audience questions you already know the answer to before asking new ones.
- Neon Blue builds the audience from your spec and labels every result as simulated. Run your first study free
Neon Blue builds simulations of human behavior. You build a synthetic audience the same way you would brief a panel recruit: define exactly who counts as your customer, choose the traits that drive the decision you're testing, set the mix so it matches the real population, and check the audience against things you already know before asking it anything new. The difference is that a synthetic audience can be rebuilt or re-cut without a new recruit. Neon Blue builds simulated audiences from that kind of spec.
What goes into a synthetic audience spec?
| Layer | What to define | Example |
|---|---|---|
| Who qualifies | The behavior that makes someone your customer | Bought frozen pizza in the past month |
| Size of groups | How many in each segment you'll read | Heavy and light buyers, at least a few hundred each |
| Demographics | The mix by age, region, income and household | Matched to census targets for the category |
| Behavior | Frequency, brands used, channels | Shops online for groceries at least monthly |
| Attitudes | Beliefs that drive the decision | Prioritizes price over brand |
| Context | The situation they're in when they decide | Weeknight dinner for a family |
A spec built this way also works for a live panel, so the two can be compared later.
How do you make a synthetic audience match the real population?
Set targets from public data, not guesses. In the US, the Census Bureau's American Community Survey samples about 3.5 million housing-unit addresses each year and publishes age, income, household type, region and more. That gives you the mix for the general population. For a category audience, combine it with what you know about your buyers: your CRM, past surveys or loyalty data.
Three common mistakes:
- Building the audience from your best customers only. You'll overstate demand. Include light and lapsed buyers.
- Setting too many traits. Every extra trait you fix narrows the audience and can make it act like a stereotype. Fix the traits that drive the decision, and let the rest vary.
- Forgetting non-buyers. If the decision is about growth, the people who don't buy you yet are the audience that matters most.
How do you check a synthetic audience before using it?
Ask it questions you already know the answers to:
- Which brands in [category] have you bought in the past 3 months? Compare with your share data.
- How often do you buy [category]? Compare with your purchase frequency data.
- Where do you usually buy it? Compare with your channel mix.
- What do you like least about your current brand? Compare with reviews or past open-ended answers.
If the answers are far off on things you know, fix the spec before asking about things you don't. This is the same discipline described in how to judge whether synthetic respondents are good enough.
What does building a synthetic audience look like in practice?
Hypothetical: a frozen pizza brand wants a synthetic audience to test a new thin-crust line and a price change.
- Audience: US adults who bought frozen pizza in the past month, split by heavy buyers (weekly or more) and light buyers, by households with and without children, and by whether they buy a store brand.
- Spec: demographics matched to census targets for the category, behavior layered from the brand's own purchase data, and one attitude trait (price first or taste first).
- Check: brand share, purchase frequency and channel compared with known data before the first study.
Example output layout (illustrative, not a real result)
| Check question | Known figure source | Synthetic audience answer | Within range? |
|---|---|---|---|
| Brand bought most often | Brand's share data | Coded | Yes / No |
| Buys weekly or more | Purchase panel | __% | Yes / No |
| Buys mostly online | Channel data | __% | Yes / No |
| Main complaint | Past open-ended answers | Coded theme | Yes / No |
Once the checks pass, the same audience can run a concept test on the thin-crust line and a price ladder on the change.
How do teams use a synthetic audience once it's built?
- Reuse it. Keep the same audience across studies so results are comparable over time.
- Re-cut it. Need a read on Gen Z buyers in the South? Rebuild that segment rather than recruiting.
- Pair it with real people. For big calls, run live interviews on the finalist and keep the sources apart.
Questions about building a synthetic audience
What is a synthetic audience?
A synthetic audience is a simulated group of people built to match a real target audience, used to field surveys, concept tests and other studies before or alongside live research.
How many people should a synthetic audience include?
Enough to read each segment you care about with room to spare, often a few hundred per segment. Size follows the cuts you need, not a fixed total.
Can you build a synthetic audience of your own customers?
Yes. Start from who your customers are in behavior and demographics, add what you know from your own data, and check the audience against known figures before using it.
How does Neon Blue build a synthetic audience?
Neon Blue builds a simulated audience from your spec, including the population targets and segments you set, and labels every result as simulated. Start a free first study.