Synthetic Personas vs focus groups: complete 2026 comparison

The choice between Synthetic Personas and focus groups is not ideological — it is financial and operational. Focus groups cost between R$15,000 and R$80,000 per round and take 4 to 8 weeks. Synthetic Personas deliver thousands of simulated answers in hours. The real debate is: which research question was each method built to answer?

For decades, the focus group was the gold standard of consumer research in Brazil. Companies such as Nestlé, Itaú, and L’Oréal built product, communication, and positioning decisions on what twelve people said in a room with a one-way mirror in Pinheiros. That model worked — until costs rose, deadlines tightened, and scale became impossible.

In 2026, the question is no longer “focus group or quantitative research.” It is: when does it make sense to use AI-generated Synthetic Personas, and when is the focus group still the right instrument?

This article answers that question with data, without methodological romanticism.

What each method is

Focus groups: the methodology that defined generations of research

The focus group is a qualitative technique created in the 1940s by Robert Merton to evaluate the effectiveness of radio programs during World War II. In the consumer context, it is a moderated discussion with 8 to 12 participants recruited according to specific demographic and behavioral criteria.

The method matured over decades and has real value: it allows researchers to observe spontaneous interactions between consumers, capture organic language, identify unspoken tensions, and explore unknown territories without a prior hypothesis. When research does not yet know what it is looking for, the focus group offers richness that no structured questionnaire can.

The typical process involves profile definition and recruitment, execution of sessions, analysis, and reporting. The cost includes participant incentives, physical or remote space, moderator fees, and qualitative analysis. In Brazil, specialized providers charge between R$15,000 and R$50,000 for a round of two groups.

Synthetic Personas: what changed with LLMs

Synthetic Personas are computational simulations of consumer segments, built on large language models fine-tuned with real behavioral data. They answer research questions, evaluate creative stimuli, and simulate purchase decisions — like real respondents, but at industrial scale and on demand.

The difference from traditional marketing personas — those cards with stock photos and generic bullet points — is fundamental. Modern Synthetic Personas are trained with real purchase behavior, browsing history, transactions, previous survey answers, and anonymized social data. The result is not a consumer description. It is a simulation capable of answering new questions with statistically validated coherence.

Platforms such as Galaxies’ NEXUS create synthetic replicas of specific audiences — for example, an Itaú Private client base or premium skincare consumers aged 28 to 42 in Brazilian capitals — and allow research teams to run hundreds of stimuli in parallel, in hours, without recruitment, moderators, or incentives.

The most important dimension: time and cost

A focus-group project consumes between R$15,000 and R$80,000 and 4 to 8 weeks of calendar time. An equivalent project with Synthetic Personas costs a fraction of that and delivers results in 24 to 72 hours. This difference is not marginal — it changes what research can do.

Let us break it down. The cycle of a focus group in Brazil in 2025 works like this:

• Weeks 1–2: briefing, script definition, internal approval, and start of recruitment.

• Week 3: active recruitment, finding people who match the profile and agree to participate.

• Week 4: execution of sessions, usually two groups of 90 minutes each.

• Weeks 5–6: transcription, qualitative analysis, and report.

If the result raises questions — which frequently happens — the second round starts from zero. Another 4 to 6 weeks. Another R$20,000 to R$40,000.

According to Qualibest data from 2024, 68% of qualitative research projects in Brazil face participant-recruitment delays. A niche consumer profile — high income, user of a specific product, with a specific behavior — can take weeks to recruit twelve people.

With Synthetic Personas, the cycle is different by nature. The first stimulus can be tested on the same day as the brief. Iteration is instant. If the result suggests the message does not work with ages 35–44 but works with 25–34, that hypothesis can be tested across eighteen more variations on the same day.

The cost per insight drops drastically. And the cost of testing a wrong hypothesis — which in a focus group means weeks lost — falls to practically zero.

The scale dimension

Focus groups work with 8 to 12 people. Online quantitative research reaches hundreds. Synthetic Personas operate with thousands of simultaneous simulations — and the statistical difference between these volumes is not trivial.

Twelve people are not a sample. In statistics, twelve qualitative observations have real exploratory value, but they do not allow safe generalizations about heterogeneous populations. The focus group never promised that — it was designed for depth, not representativeness.

The problem appears when companies make product or communication decisions based on focus groups without enough scale for validation. A Nielsen study from 2023 on product-launch errors in Brazil identified that 41% of cases involved qualitative research without a subsequent quantitative step — the decision was made based on what twelve people said.

Synthetic Personas operate under a different logic. A thousand simulated answers with controlled variations in demographic profile, region, income class, and purchase behavior allow segmentation analyses a focus group could never deliver. For example, it is possible to test whether receptivity to a new package varies between consumers in the North and Southeast, or between convenience buyers and brand-loyal buyers — all in the same project, without extra cost per respondent.

The statistical implication is direct: a higher simulation volume reduces estimate variance and increases confidence in findings. The important caveat is that this statistical validity depends on synthetic-model quality — personas trained with insufficient or outdated data produce systemic bias, not random noise.

The bias problem in focus groups

Focus groups concentrate distortion vectors that no moderation technique fully removes: social desirability, moderator influence, groupthink, and the dominant-participant effect. Synthetic Personas do not have these problems — but they have their own.

Social-desirability bias is perhaps the most persistent. Focus-group participants tend to report behaviors and preferences they consider socially acceptable, especially on sensitive topics such as food, personal finance, beauty practices, and politics. Research has shown that group settings can underreport stigmatized behaviors significantly.

The groupthink effect is well documented: the first person to take a position on a product or concept anchors the whole discussion. If that person is articulate and confident, the group converges — not because everyone agrees, but because public disagreement has social cost. Experienced moderators mitigate this, but do not eliminate it.

There is also a structural recruitment problem: people who agree to participate in focus groups, usually for an incentive, are not representative of the median consumer. They have availability, willingness to be exposed, and often repeated participation in research — the so-called “professional focus-group participant.”

Synthetic Personas eliminate these specific vectors. There is no group pressure, no social desirability, and no moderator who unconsciously directs the tone of questions. Synthetic bias has another nature: it sits in training data, calibration, and model assumptions. It is a different bias — more controllable, more auditable, and more correctable.

What focus groups still do better

Focus groups remain the most effective instrument for exploratory discovery with no defined hypothesis, for capturing organic consumer language, and for researching completely new behaviors. No synthetic model can simulate what has never happened before.

Honesty about this matters. Synthetic Personas are trained on historical data. They simulate well what consumers have already done or verbalized, and extrapolate with reasonable accuracy inside known territory. But when research involves a radically new behavior, a product with no category, an emerging technology, or a habit change with no precedent, the synthetic model does not have an adequate training base.

A focus group with real consumers in 1998 about smartphone usage would have captured intuitions that no historical data could generate, because the behavior did not yet exist.

Focus groups also capture language organically. The vocabulary consumers spontaneously use to describe a product — the metaphors, comparisons, and terms they invent — has immense value for communication and positioning. This type of linguistic capture is not the strongest point of current Synthetic Personas.

For true exploratory research — when the company does not even know which questions to ask — the focus group remains the right instrument.

The complete comparison table

When to use each one

The decision is not about which method is better in the abstract. It is about which research question is being asked. The most useful criterion: if the question can be formulated with a prior hypothesis, Synthetic Personas deliver faster and cheaper. If the question is still being discovered, the focus group is the right instrument.

Use Synthetic Personas when:

• The question already has defined hypotheses to test, such as which package version performs better or which message resonates with segment X.

• The project requires simultaneous segmentation by profile, region, or behavior.

• Fast iteration is critical, for example a campaign decision in 48 hours.

• The budget does not allow multiple rounds of qualitative research.

• You need to validate qualitative findings at scale before taking them to the board.

Use focus groups when:

• The company is entering completely new territory: a new product, new category, or new audience with no historical data.

• The goal is to capture consumer vocabulary and language for communication briefs.

• The research explores behaviors that have never been mapped in digital data.

• The decision requires evidence from real consumers for stakeholders who do not accept synthetic data as a proxy.

As a reference, Forrester Research identified that 73% of research questions in consumer-goods companies have hypotheses defined enough to be answered with quantitative or synthetic methodology — only 27% genuinely require open qualitative exploration.

The ideal combination

The companies that advanced most in research efficiency in 2025 did not choose between the two methods. They redesigned the flow: synthetic screening for fast hypothesis triage, focus groups for selective depth on the most relevant findings. The total cycle cost falls 60% to 70% without loss of decision quality.

The model works like this in practice:

Phase 1 — Synthetic screening, 1 to 3 days: ten concept variations are tested with Synthetic Personas representative of the target segment. Three concepts stand out on purchase intent, perceived relevance, and differentiation. The other seven are discarded before any investment in production or qualitative research.

Phase 2 — Selective qualitative depth, 2 to 3 weeks: the three finalist concepts go to a focus group. The moderator enters with hypotheses already formed — knowing what Synthetic Personas said, where the data were ambiguous, and which questions need qualitative answers. The session is more focused, more productive, and cheaper.

Phase 3 — Quantitative validation, optional, 1 week: if the decision involves significant investment, quantitative research confirms the finding with a representative sample.

The result: what once consumed 8 weeks and R$60,000 now consumes 3 to 4 weeks and R$25,000 — with more confidence in the final result because hypotheses arrive at the focus group already filtered.

The most important change is not replacing the moderator with AI. It is arriving at the focus group knowing what you are looking for. When you enter the room with hypotheses tested at scale, the conversation changes completely.

Specific use cases by company type

FMCG — Nestlé, Unilever, P&G

Launch pressure in Brazilian retail is relentless. Shelf windows open and close in 90-day cycles. At that pace, waiting 6 weeks for a focus group to validate packaging is not viable.

FMCG companies use Synthetic Personas mainly for:

• Packaging testing: color variations, visual hierarchy, and product claim — dozens of versions tested in parallel with specific segments.

• Pre-launch concept testing: naming, positioning, and main message validation before investment in final art and production.

• Price-sensitivity analysis: simulation of how different price ranges affect purchase intent by income segment.

The focus group enters afterward — to capture emotional reactions and vocabulary that no synthetic metric replaces.

Financial — Itaú, Bradesco, XP

Financial products have two research problems: consumers rarely admit real money behaviors in a group, due to extreme social desirability, and regulatory approval cycles do not tolerate research delays.

Synthetic Personas eliminate social-desirability bias in financial research. The most common uses include:

• NPS simulation by product segment before functionality changes.

• Digital onboarding journey testing.

• Evaluation of receptivity to new investment products by risk profile.

Beauty and personal care — L’Oréal, Natura, Boticário

The Brazilian beauty market is the fourth largest in the world and operates in trend cycles that last less than six months. Focus groups do not fit that rhythm.

Typical applications with Synthetic Personas:

• Emerging trend identification: simulate receptivity to ingredients and formats emerging in markets such as Korea and the United States.

• Product reformulation: test reformulated performance claims without exposing real consumers to a version still in development.

• Multicultural communication: evaluate how the same brand message resonates across ethnic and regional segments in Brazil.

Frequently asked questions

What are Synthetic Personas and how do they work?

Synthetic Personas are computational simulations of consumer segments, built on language models trained with real behavioral data. They answer research questions, evaluate stimuli, and simulate purchase decisions at scale — without recruitment, moderators, or participant incentives.

Do Synthetic Personas completely replace focus groups?

Not completely. Synthetic Personas replace most use cases — especially concept testing, hypothesis validation, and research with a defined hypothesis. Focus groups are still superior for genuine exploratory research, organic language capture, and analysis of completely new behaviors with no historical data.

What is the cost of a Synthetic Personas project compared with a focus group?

A full focus group in Brazil costs between R$15,000 and R$80,000 and takes 4 to 8 weeks. An equivalent project with Synthetic Personas costs a fraction of that, with marginal cost per respondent close to zero and delivery in 24 to 72 hours.

Are Synthetic Persona answers reliable?

They depend on the quality of the training model. Personas trained with real, updated, and representative behavioral data from the target segment produce simulations with high predictive accuracy. Synthetic bias exists, but it is auditable and correctable, unlike human focus-group bias.

How many respondents can I simulate with Synthetic Personas?

Scale is practically unlimited. It is possible to run one thousand, five thousand, or more simultaneous simulations with variations in demographic profile, behavior, and consumption context — something operationally impossible with focus groups.

Which method has less bias?

Both have bias of different kinds. Focus groups suffer from social desirability, groupthink, and moderator influence. Synthetic Personas carry training-data bias. Synthetic bias is generally more controllable; human focus-group bias is structural and hard to eliminate completely.

How can the two methods be combined efficiently?

The most effective model is Synthetic Personas for rapid screening of multiple hypotheses in phase one, followed by selective focus groups for qualitative depth on the most relevant findings in phase two. This flow reduces total cycle cost by 60% to 70% without losing decision quality.

Do Synthetic Personas work for consumer research in Brazil?

Yes, as long as the model has been trained or calibrated with Brazilian-market data: purchase behavior, consumption habits, economic context, and regional context. Generic models trained only on American or European data do not adequately capture the heterogeneity of Brazilian consumers.

Conclusion

The choice between Synthetic Personas and focus groups is not a question of modernity versus tradition. It is a question of fit with the type of research question.

For most questions faced by Consumer Insights teams — concept validation, communication testing, response segmentation, price-sensitivity analysis — Synthetic Personas deliver faster, cheaper answers with less structural bias.

For the minority of questions that require genuine discovery, organic language, and exploration without a prior hypothesis, the focus group remains the correct instrument.

The companies that will win in consumer intelligence over the next few years are not the ones that choose one method. They are the ones that redesign the entire flow — using synthetic methods to filter, iterate, and scale, and focus groups to deepen what truly matters.

Discover Galaxies NEXUS

Galaxies is the synthetic consumer-intelligence platform that research teams at companies such as Nestlé, Itaú, Bradesco, L’Oréal, and Natura use to replace the slow and expensive traditional research cycle with scaled simulations in hours.

NEXUS creates synthetic replicas of your real audience and allows you to test concepts, messages, packaging, and pricing strategies before investing in production or field research.

Request a demonstration at galaxies.com.br — we understand your segment, show NEXUS with your audience, and you leave the meeting knowing whether the method makes sense for your context.

References cited in this article: IBOPE Inteligência (2024); ABA — Brazilian Association of Advertisers, research investment report (2024); Qualibest, study on delays in qualitative projects in Brazil (2024); Nielsen, analysis of launch errors in the Brazilian market (2023); Forrester Research, Consumer Insights Methodology Report (2024); Euromonitor International, Beauty & Personal Care Brazil (2024); Tourangeau & Yan, “Sensitive Questions in Surveys,” Journal of Consumer Research (2007).