Jason Duke, Founder, Kronaxis
Tag: Guide
A synthetic consumer panel is a group of AI generated consumer personas, each with a unique personality profile, demographics, and life history, that can respond to research stimuli as if they were real participants.
That is the one sentence definition. The rest of this guide covers how synthetic panels work, what makes a good one, where they outperform traditional methods, and where they do not. If you are evaluating whether synthetic panels belong in your research toolkit, this is the place to start.
How It Differs From What You Already Use
Surveys give you one question, one answer, no follow up. You can ask a thousand people whether they prefer option A or B, but you cannot ask the 340 who chose A to explain their reasoning and then challenge that reasoning with a counterargument. The interaction is dead on arrival.
Focus groups solve the follow up problem but create three new ones: cost (£5,000 to £15,000 per session), speed (four to eight weeks from recruitment to report), and sample size (eight to twelve people per group, which is too few for any statistical confidence across subgroups).
Conjoint analysis decomposes attribute values from choice data, but every respondent is a statistical abstraction. You get aggregate willingness to pay for feature X versus feature Y. You do not get an explanation of why persona 247 valued API access over collaboration, or why persona 88 chose the cheapest option despite having a high income.
A/B testing measures outcomes. It tells you that variant B converted 12% better than variant A. It does not tell you why, and it cannot tell you whether the effect will hold in a different segment, at a different price point, or in a different emotional context. You get the "what" without the "why."
Synthetic panels combine the depth of focus groups, the scale of surveys, the analytical precision of conjoint, and the speed of none of the above. A 500 persona panel returns results in minutes, not weeks. Every response comes with a reasoning trace that explains the decision mechanics behind the answer.
How It Works, Step by Step
Building and running a synthetic panel follows a consistent sequence.
First, define your target audience. Country, age range, income bands, gender distribution, regional spread, any segment criteria that matter for your research question.
Second, generate census weighted personas. Each persona is built from real population distributions: age, gender, ethnicity, education, income, occupation, region, household composition. For the UK, this draws from ONS Census 2021 data. For the US, from census bureau distributions. Twenty countries are currently supported.
Third, assign personality profiles. Every persona receives a DYNAMICS-8 profile: eight continuous dimensions scored from 0 to 1. Discipline, Yielding, Novelty, Acuity, Mercuriality, Impulsivity, Candour, and Sociability. These dimensions are not random. They are assigned in patterns that match known population distributions and correlate with the persona's demographics, life history, and economic circumstances.
Fourth, run your stimulus. This can be a question, an image, a product description, a pricing scenario, a policy proposal, an advertisement, a concept sketch, or any combination. Each persona processes the stimulus through its personality profile, life experience, emotional state, and economic context.
Fifth, collect responses with reasoning traces. Every response includes the persona's answer alongside a structured trace showing which personality dimensions were active, what life experiences influenced the decision, and how emotional and economic context modulated the response.
Sixth, export your data. JSON, CSV, Parquet, or JSONL. Slice by any demographic or personality dimension. Run your own statistical analysis or feed the data into your existing tools.
What Makes a Good Synthetic Panel
Not all synthetic panels are equal. The quality depends on four properties.
Census weighting is the foundation. If your panel does not match the real population distribution of your target market, every result is biased from the start. A panel that overrepresents university educated urban millennials will tell you what university educated urban millennials think. That is not market research. That is an echo chamber.
Personality grounding separates genuine behavioural simulation from fancy random number generators. If a persona's responses are not driven by a coherent psychological model, the responses are noise. DYNAMICS-8 provides eight continuous dimensions that shape every decision a persona makes, creating internal consistency across hundreds of interactions.
Multiturn capability matters because research is a conversation, not a quiz. The ability to follow up, challenge, probe, and redirect is what separates useful insight from surface level opinion. A synthetic panel that can only process single turn stimuli is a survey with extra steps.
Causal reasoning traces are what make synthetic panels genuinely useful rather than merely fast. Knowing that 62% of your panel preferred option B is demographic data. Knowing that high Discipline personas preferred it because the value proposition was clearly quantified while high Novelty personas preferred it because the design felt distinctive: that is insight you can act on.
Common Use Cases
Pricing research is the highest value application. Synthetic panels reveal willingness to pay segmented by personality type, not just demographics. Two people with identical income and age can have opposite price sensitivities because one is high Discipline (comparison shops methodically) and the other is high Impulsivity (buys on emotional response).
Brand perception testing runs a close second. Campaign concepts that resonate with high Sociability personas may fall flat with high Acuity personas who find the messaging shallow. Traces show exactly where the disconnect happens.
Concept validation lets product teams screen twenty ideas in a morning instead of testing three over a quarter. The speed advantage changes the shape of the innovation process.
Regulatory prechecking is an emerging use case. Run proposed advertising copy or product claims past a diverse panel before submission and identify which segments find the claims misleading or confusing.
Academic research benefits from the reproducibility. Run the same study twice with the same panel and get the same results. Try that with a human sample.
Limitations, Stated Honestly
Synthetic personas are not real people. They cannot taste your product, feel your packaging, or experience the frustration of a broken checkout flow on a Tuesday evening when they are tired.
Validation against real populations is ongoing. Panel Studio panels produce results that align with known patterns in published research and with traditional methods run in parallel, but the field is young and independent validation studies are still accumulating.
Synthetic panels are best understood as hypothesis generation and rapid screening tools. They narrow the field, surface the personality dynamics driving preferences, and identify where deeper investigation is warranted. They are not a replacement for final stage validation with real humans on high stakes decisions.
Rare population segments may be underrepresented. Census weighted generation matches known distributions, which means that a persona combination that occurs in 0.01% of the real population will appear roughly once in a 10,000 persona panel and not at all in a 500 persona panel.
The Market
Several companies offer synthetic panel capabilities. Evidenza provides AI generated respondents for market research. Aaru offers synthetic audiences for consumer insight. Qualtrics Edge Audiences integrates synthetic respondents into their survey platform.
What differentiates Kronaxis Panel Studio: an open personality framework (DYNAMICS-8 is published under Creative Commons, not locked behind a proprietary model), reasoning traces on every response (not just answers but the structured mechanics of why), a selfhosted option (run the entire system on your own hardware with no cloud dependency), and source available code under the Business Source Licence.
Getting Started
Panel Studio offers a free tier. No credit card required. Ten personas, ten interactions per month. Enough to build your first panel, run a stimulus, and read the reasoning traces.
Install the Python SDK with `pip install kronaxis`, or use the web interface directly. Census weighted personas across twenty countries. Seventeen stimulus templates. Four export formats.