Lightning fast
Two weeks of research, done in seconds. An AI moderator designs and runs the study end to end, so you stop waiting and start deciding.
Synthetic Humans
Ask anything, answers in seconds
Sunzu turns your customer data into synthetic humans your team can question. Run interviews, concept tests, usability and ad testing in seconds, not weeks.
Used by
Researchers
Product managers
Designers
Research leaders
at
Backed by
Microsoft for Startups
NatWest Accelerator
Watch
See Sunzu in 30 seconds
From your data to synthetic humans you can trust.
Start here
What is synthetic user research?
Synthetic user research means putting your questions to AI models of your customers instead of waiting weeks to recruit real ones. Every answer traces back to a source.
| Dimension | Sunzu synthetic humans | Traditional research | Synthetic users with LLM |
|---|---|---|---|
| Time to answers | Seconds | Weeks | Seconds |
| Grounded in real data | Your data, enriched with our corpus | Yes, real participants | No, generic training data |
| Traceable and auditable | Every answer cited | Partial, manual | None |
| Cost per study | Low | High: recruit and incentivise | Low |
| Best used for | Fast, grounded first-pass research | High-stakes, definitive decisions | Nothing you need to trust |
Read the method: Engineered, not prompted and 9/10 teams make the same mistakes.
Customer-centric teams
Why Sunzu
Anyone on your team can ask anything and get an answer in seconds, from your browser or from Claude, ChatGPT, Slack and Teams.
Actual research
When research is rationed, teams guess. Sunzu is built for everyone on the team, so research stops being a bottleneck and becomes a habit.
Robust
Traditional panels are gamed: professional respondents, cheated screeners, bots posing as humans. Sunzu builds your audience from your own customer data. One study is worth a panel of 6 to 30+ recruited participants.
Realistic
Prompting an LLM to roleplay a user is theatre: generic AI averages toward an agreeable person who does not exist. Sunzu synthetic humans satisfice, contradict themselves, and hesitate where your users hesitate.
From the field
Testimonials
In just a couple of months, over 170 of our colleagues have run over 500 studies.
We’re now directly integrating synthetic humans into many of our research, product and marketing workflows.
In a market flooded with AI-generated noise, this was the first platform that genuinely earned our trust.
By grounding its models in research principles, behavioural science, and real customer data, it delivers insights that feel credible, transparent, and actionable.
It changed our perspective on what’s possible with AI in customer research.
How it works
Your synthetic audience
Turn the customer data you already have into an audience your team can question.
-
Your insight
Bring what you have: CRM exports, surveys, transcripts, NPS, analytics or past research. Sunzu maps each segment across motivations, objections and decision drivers.
-
Our data and model
We enrich it against our own behavioural corpus, then map the audience through our cognitive architecture, so each synthetic human reasons like the customer it is modelled on.
-
Embed in your workflows
Run every study type in one workspace, connected to Figma, Jira, Confluence and Miro. Or launch straight from Claude, ChatGPT, Slack and Teams over MCP.
Every study type
Use cases
Everything in your research stack.
Discover
Customer interviews
Depth interviews that surface the why behind your numbers.
Test
Concept test
Test rough ideas before a line of code is written.
UX
Usability test
Walk synthetic humans through your flows and see where they hesitate.
Creative
Ad testing
Pressure-test copy and creative on the segment it must move.
Run your first exploration with us.
Book a call
Science and validation
How we model human behaviour
Built on computational frameworks from cognitive science and behavioural economics, not AI roleplay. Validated against real human behavioural data, so our models avoid what generative simulators get wrong: homogenised personas, tunnel vision, and optimism bias that erases friction.
The evidence we bring to your data.
Your data anchors the model, but never covers everything a person is. Where it stops, our corpus of 57M+ records from 4,038 sources carries the persona, so gaps are filled with evidence rather than invention.
- 1m+
- Individual population records
- 34m+
- Normalised trait behaviours: demographics, attitudes, values, beliefs, life events
- 8m+
- Consumption and spending records
- 628k+
- UX interaction events, including full shopping sessions
Cognitive Architecture
We model how people think, not just what they say. We simulate perception, memory, attention and goal-setting, so personas reason with both deliberate analysis and fast intuition.
Decision Model
People do not optimise perfectly. They satisfice, lean on heuristics and choose badly under pressure. We model those constraints.
Behavioural Model
Behaviour holds across situations. Personas remember past interactions and keep preferences over time.
Primitives
Complex behaviour decomposes into building blocks. Over 50 decision traits are mapped per synthetic respondent, preserving the edge cases generic AI averages away.
One Sunzu study does the work of a research panel.
Scored blind against a human study run with hundreds of people. This is how Sunzu studies compare to humans.
- Depth interviews
- 6+
- recruited people
- for the same themes covered
- Concept tests
- 30+
- recruited people
- for the same insights found
- Usability tests
- 5+
- UX experts
- for the same defects detected
- Ad testing
- 20+
- recruited people
- for the same insights found
Traceable by default
Every answer is inspectable: why a respondent said it, and what evidence informed it.
Designed for human validation
Sunzu front-loads exploration so your recruited studies are sharper, not skipped.
Common questions
FAQ
No. Sunzu is human-in-the-loop: it helps you explore faster and decide which questions genuinely need recruited participants.
Whatever you already have: CRM exports, surveys, transcripts, NPS, support themes, analytics or past research. You do not need everything. Our own corpus grounds the dimensions your data is silent on.
Roleplay has no provenance and drifts agreeable. Sunzu grounds every respondent in your customer data and makes every output traceable.
Days, not months. Bring the customer data you already have and Sunzu builds the audience from there. The 30-minute call is the fastest way to scope it.
Sunzu is operated by Alpha Base OS Ltd (ICO registration C1103931). Workspace data is processed solely for the contracted research purpose, never for model training or advertising. Full detail is in our Privacy Policy, linked in the footer.
A human study with hundreds of people sets the answer key; one Sunzu study runs it blind and is scored against it. One Sunzu study is worth a panel of 6 to 30+ recruited participants, depending on the study type. Only 4% of the defects it flagged were not real.
The people behind Sunzu
Meet the team
Alex
Head of Commercial
Human-centred research and product. Makes synthetic audiences that enterprise teams trust.
Yoann
Head of Product
Product strategy and engineering. Turns hard questions into answers in seconds.
Ludo
Head of Technology
Enterprise AI engineering. Grounds synthetic humans in real behavioural data.
Bring the customer problem you want to solve.
Let us crack it together.
Book a call
30 minutes. No deck, no commitment.