Testing campaigns, concepts and messages with the target
Three campaign routes, a new claim, a packaging variant, a price message. Small decisions, made every week, each one reaching the same consumer.
Synthetic personas built from your own research
Explore more of your own data with Synthetic Lab.
A study delivers one result and goes into a drawer. Synthetic Lab brings together the projects you have already run and builds synthetic personas of your target from them. Then you ask again: test a campaign, run a focus group, close a quantitative study. The same day, with the evidence behind every insight.
Illustrative data. The scenes play on their own.
01 The work
The launch gets a study. The headline, the pack and the three campaign routes get a meeting.
Every company has far more decisions than research budget. It also has years of studies sitting in PDF, each one answering only the question it was commissioned for.
Field for the big bets, opinion for the rest
A study takes weeks and a budget, so it is kept for the launch. The tactical idea is decided internally, or pasted into a general chatbot that answers like the average of the internet.
Synthetic is still an unknown
Everyone wants to test it, and everyone is wary of agents. The market also mixes two different things under one name: synthetic data and synthetic personas. Nobody can tell what is behind the answer they received.
Personas of your target, with method
A stratified sample built from census statistics, grounded in your past studies, interviewed and put to debate, read by several models, with every insight traceable to what supports it.
A language model predicts the next word. Give it deep context about one target, and it lands close to where that group had already landed.
02 What it answers
Research questions, answered the same day.
Each block of the platform exists because it answers a question someone asks before approving an idea.
"Which of the three routes does my target prefer, and why?"
Variant testA, B and C
Every variant goes to the same stratified sample. The result is a go or no-go per variant, a heatmap by age group and social class, and the reasons in the personas' own words.
"What would they say in a focus group?"
Mirror Room12 personas
A synthetic focus group. Twelve personas debate the stimulus in real time with a moderator, while you watch from behind the glass, step in with a question and follow the sentiment meter.
"Does the message hold with the skeptical ones?"
Voice variationliteracy, skepticism, region
Personas differ in literacy, in skepticism and in regional tone. The answers do not come back in the voice of one polite model agreeing with everything.
"How do I know this was not made up?"
Evidence trailper insight
Every insight points to the interviews and sources that support it. Several models read the same data, and a final judge settles the points where they disagree.
"Which lens should I read this through?"
28 methodsfive families
Quantitative, qualitative, semiotics, UX and visual, and neuro. From purchase intent and brand recall to tone of voice, readability and cultural reading.
"What do I take to the meeting?"
Deliverablesfour formats
An interactive dashboard, a master report, a presentation and a specialist agent of the project, ready to answer the cut nobody thought of before the meeting.
03 Method
Six steps, with a person approving each gate.
The platform runs in guided mode, with approval gates, or end to end on autopilot. The governance is that of a research firm. The speed is that of software.
- 01
Brief and plan
The briefing is captured in a structured way and becomes a methodological plan in minutes.
You approve the plan before anything runs.
- 02
Build the target
A stratified sample with census statistics, margin of error and confidence level, grounded in your past studies.
Every persona has a profile you can open.
- 03
Write the instrument
A copilot drafts the quantitative questionnaire and the qualitative discussion guide.
Approval gate before collection.
- 04
Collect
Interviews with text, audio, image and video stimuli, and the Mirror Room for the group debate.
Every completed interview is recorded.
- 05
Analyze
Specialist agents, one per method, read the collection with several language models.
No single model decides alone.
- 06
Deliver
Dashboard, report, presentation and the project's specialist agent, from the same flow.
Every insight carries its evidence.
04 What is behind it
Rules of the platform, stated with their scope.
Each number below is a rule of how the product works. None of them is a promise about your result.
28analysis methods, in five families
12personas debating in each Mirror Room
3fidelity levels, chosen per project
30 mintarget for the first insights of a study
What it never does
- Never presents a synthetic answer as a real consumer's answer. Every output is labeled synthetic.
- Never replaces fieldwork in a decision that needs a real sample. It narrows the options before field.
- Never sells synthetic personas as synthetic data. They are different things, and the report says which one you received.
- Never shows an insight without the interviews and sources behind it.
- Never lets a single model decide. Where models disagree, the disagreement is judged and recorded.
Who it is for
- Research firms
- Insights teams
- Agencies and creative teams
- Brand and marketing teams
For tactical questions that would never justify fieldwork, and for narrowing many ideas down to the few worth taking to real consumers.
05 Questions
Questions and answers
What is Synthetic Lab?
Synthetic Lab is a Cassi.ai product for market research with synthetic personas. It builds a stratified sample of personas for one specific target, grounds them in the studies the company already has, runs interviews and a synthetic focus group, and returns a verdict with the evidence behind every insight.
What is the difference between synthetic data and synthetic personas?
Synthetic data is a generated dataset that imitates the statistical shape of a real one, used to extend or protect a database. A synthetic persona is an agent that answers as a member of a target would, in an interview or a group. The market uses the two terms as if they were one. Synthetic Lab works with synthetic personas, and every report states that.
How can a language model answer like my target?
A language model predicts the next word. On its own it answers like the average of what it read. With context engineering and a volume of research that deepens what is known about one target or cluster, it reaches conclusions very close to the ones that group had already reached. That is why the personas are grounded in your own studies.
Does Synthetic Lab replace fieldwork?
No. It serves the tactical projects and ideas that would not justify going to field, and it narrows many options down to the few worth validating with real consumers. When a decision needs a real sample, the study still goes to field, now with fewer and better options.
What is the Mirror Room?
The Mirror Room is a synthetic focus group. Twelve personas of the target debate a stimulus in real time with a moderator. The team follows from behind the glass, can step in with a question, and watches a sentiment meter move as the discussion goes on.
How do I trust an insight from synthetic personas?
Every insight points to the interviews and sources that support it. The same collection is read by several language models, and a final judge settles the points where they disagree, recording the disagreement. The project also chooses one of three fidelity levels, from fast to maximum.
Which analysis methods does it run?
Twenty-eight methods in five families: quantitative, such as brand recall, purchase intent and message fit; qualitative, such as sentiment and cultural reading; semiotics, such as tone of voice and brand consistency; UX and visual, such as readability and heuristics; and neuro.
What do I receive at the end of a study?
An interactive dashboard, a master report, a presentation, and a specialist agent of the project that answers new questions and new cuts over the same collection. All four come out of the same flow, with the evidence trail of each insight.
Bring the studies you already have.
Send the past projects and the target. We set up the personas and show a new study running on top of them, from the plan to the verdict.