The complete guide to Synthetic Personas: what they are, how they work, and why they change market research

Synthetic Personas are consumer profiles generated by Artificial Intelligence from real research data. They simulate behaviors, opinions, and responses with 91% accuracy compared with human respondents, delivering insights in 48 hours at a cost up to 93% lower than traditional qualitative research.

When market research stopped making sense

Imagine a company about to launch a new product. The product team is confident. Marketing has already prepared the campaigns. Leadership approved the budget. Only one thing is missing: knowing whether the consumer will buy.

The traditional solution would be to hire market research: recruit respondents, run focus groups, conduct in-depth interviews, process the data, and build the report. All of that would take two to six months. It would cost between R$50,000 and R$500,000. And in the end, the company would have answers from a small group of people who may not represent the real market.

There is a different path. And that is exactly what this guide is about.

Synthetic Personas are the answer the Brazilian market needed: a way to obtain deep insights into consumer behavior in 48 hours, with unlimited scale and radically lower cost. In this guide, you will understand what they are, how they work, what differentiates them from any other research solution, and why companies such as Bradesco, Nestlé, and Banco do Brasil already use them to make strategic decisions.

What are Synthetic Personas?

A Synthetic Persona is an AI-generated consumer profile that replicates, with high fidelity, the behavioral patterns, attitudes, preferences, and responses of a real group of people.

The key word here is “real.” Synthetic Personas are not invented. They are built from research data collected from real human respondents: questionnaires, interviews, and behavioral data. What AI does is amplify those data, from dozens of respondents to hundreds or thousands of personas, while maintaining the statistical fidelity of the original group.

It is the difference between taking a high-resolution photo and zooming into a small detail: the quality of the original image determines the quality of the final result.

Where do the data come from?

The Galaxies platform works exclusively with data provided by the client, collected by Galaxies in its own research, or sourced from partners authorized by the client. No internet data is used.

This choice is deliberate and directly affects result quality. Internet data are noisy, decontextualized, and often outdated. Research data are structured, relevant, and represent the real consumer behavior that matters to the client.

The result is a persona that reflects the most frequent responses of a real group, not a statistical invention, but a synthesis of verifiable behaviors.

How do Synthetic Personas work in practice?

The process has five stages. Each has a specific purpose and is executed with quality controls that ensure the generated persona is reliable.

Stage 1: Collection and validation of input data

The starting point is real data. The client uploads quantitative research, such as spreadsheets or CSV files, qualitative interview transcripts, or CRM data. Before any processing, the platform performs Pre-Model Statistical Validation: a certification of data integrity and quality. If the input base is poor, the model will know and signal the problem before generating incorrect results.

Stage 2: Clustering with Machine Learning

Real respondents are grouped into homogeneous clusters based on demographic, attitudinal, behavioral, and lifestyle characteristics. The algorithm is calibrated to ensure representativeness, meaning minority groups are also represented, not only the dominant profiles.

This care is fundamental. Research that represents only the average consumer is not enough for market decisions. Market reality is segmented, and personas need to reflect that diversity.

Stage 3: Persona generation with LLM

With clusters defined, the large language model generates personas using the most frequent answers from each group as an anchor. This ensures that each persona has a coherent and contextualized identity, not a random one.

From this moment, the persona can answer questions in natural language, as a human respondent would in an interview. The difference is that it is available at any time, for any volume of questions, with no recruitment and no additional cost per interaction.

Stage 4: Post-Model Statistical Validation

Before releasing personas for use, the platform runs a second round of tests. The DeepEval framework evaluates five dimensions: answer relevance, absence of bias, hallucination control, fidelity to persona information, and toxicity.

Published results show 97.81% approval in Persona Aware Hallucination, meaning the model does not invent information outside the base; 95.08% in Persona Faithfulness, meaning consistency with original data; and 100% in Toxicity, meaning absence of inappropriate language.

Stage 5: Active persona available 24/7

With validation completed, personas are ready to use. In Nexus: Galaxies Lab, the user talks directly to personas, asks open questions, simulates hypothetical scenarios such as “what if we launch the product at R$199 instead of R$299?”, and tests creatives or messages in real time.

Why do Synthetic Personas change market research?

Market research has existed for decades. It worked, and it still works in many contexts. So why do Synthetic Personas represent a real change, not just another technology promise?

The answer lies in three structural limitations that traditional research never fully overcame:

Limitation 1: Time

A complete market study, from brief to presentation, rarely takes less than 60 days. In markets that change week by week, insights generated two months ago arrive too late. Synthetic Personas deliver results in 48 hours, not as a quality compromise, but as the result of an automated and validated process.

Limitation 2: Cost

The cost of qualitative research with 30 in-depth interviews ranges from R$80,000 to R$300,000 depending on segment and audience profile. This places quality research out of reach for smaller companies and limits how often large companies can research. Galaxies reduces this cost by up to 93%: the Bradesco Seguros case recorded a cost of R$1.20 per synthetic respondent, compared with R$150 to R$300 per traditional interview.

Limitation 3: Scale

Qualitative research with 30 respondents is, by definition, small. It is enough to explore hypotheses, but rarely enough to validate decisions with statistical confidence. Synthetic Personas allow a base of 30 real respondents to be expanded to 600, 1,000, or more personas, while maintaining representativeness and precision.

Who uses Synthetic Personas and for what?

Synthetic Personas are already in use in some of the most demanding sectors of the Brazilian market. The following cases illustrate how different types of companies apply the technology to solve concrete problems.

Financial sector: fast validation of complex products

Bradesco Seguros used Synthetic Personas to validate a new product before launch. The traditional process would have taken months of focus groups and extensive questionnaires. With Synthetic Personas, 600 synthetic respondents were generated in 48 hours. The result: a launch 10.5 times faster and a 93% lower cost per respondent. In the operation director’s words, the company obtained “60 times more qualitative answers in a quarter of the expected time, with one tenth of the cost per respondent.”

Banco do Brasil is also a Galaxies client and uses the platform for market research with synthetic data, addressing both insight speed and the regulatory demands of the financial sector.

CPG and consumer goods: more launches, less risk

Consumer-goods companies such as Nestlé and O Boticário face the challenge of launching dozens of products per year in highly competitive markets. Synthetic Personas allow positioning, price, and message to be tested for each new SKU before any investment in production or distribution.

Marketing and communication: creative validation before going live

Agencies and marketing teams use Synthetic Personas to test creatives, copy, and campaigns before media. One of the most documented cases is Mahta Bio, which reduced customer-acquisition cost by 35% after validating its campaign with Synthetic Personas and identifying the highest-converting creative approach before investing in paid media.

Innovation and product development

Product teams use personas to validate features, simulate reactions to price changes, test onboarding flows, and quickly explore strategic profile combinations, such as active customer versus former customer, or frequent user versus occasional visitor.

What about LGPD? Are Synthetic Personas legal in Brazil?

Yes. The Galaxies approach was designed to operate in full compliance with Brazil’s General Data Protection Law, the LGPD. The reason is technical: Synthetic Personas are not individual personal data; they are statistical representations of groups. No persona corresponds to an identifiable real person.

In addition, the platform does not use internet data to generate personas. All input data are provided by the client or collected with appropriate consent. This eliminates the risks of improper personal-data processing that affect other AI-based research approaches.

Galaxies presents 100% compliance with the LGPD and with the highest market standards for data-security validation.

Frequently asked questions about Synthetic Personas

What are Synthetic Personas?

Synthetic Personas are consumer profiles generated by Artificial Intelligence from real research data. Using Machine Learning algorithms and language models, they replicate behaviors and responses from real groups with 91% accuracy, delivering insights in 48 hours at a cost up to 93% lower than traditional research.

How are Synthetic Personas created?

The process has five stages: collection and validation of real data, clustering with Machine Learning, persona generation with LLM, post-model statistical validation, and activation for use. Only data provided by the client or collected with consent are used, never internet data.

What is the difference between traditional research and Synthetic Personas?

Traditional research takes two to six months and costs between R$80,000 and R$500,000. Synthetic Personas deliver results in 48 hours with cost up to 93% lower, unlimited scale, and 24/7 availability. Quality is validated by double statistical certification.

Are Synthetic Personas compliant with LGPD?

Yes. The Galaxies methodology uses aggregated data and statistical group representations, not identifiable personal data. The platform operates with 100% LGPD compliance and international data-security standards.

How accurate are the Synthetic Personas generated by Galaxies?

They show 91% accuracy compared with real respondents, validated by independent tests. The DeepEval framework evaluates relevance, bias, hallucination, fidelity, and toxicity, with approval above 95% in critical metrics.

Market research is not over, it evolved

Synthetic Personas did not come to replace every form of research. They came to solve a real problem: the gap between the speed the market demands and the time traditional research requires.

For most business questions — validating a product, testing a campaign, understanding consumer reaction to a price change, mapping sales objections — Synthetic Personas deliver faster, cheaper, and equally reliable answers.

For cases where exploration of a new phenomenon requires in-person human contact, traditional research still has a place. The smartest approach is to combine both: a small and fast qualitative phase to generate hypotheses, followed by synthetic validation at scale.

Galaxies built the platform that makes this new research model possible for companies of all sizes, from startups to large corporations that need to decide quickly in markets that do not wait.

Schedule a demonstration of the Galaxies platform and see Synthetic Personas in action!

Explore use cases by sector at Galaxies