# Sunzu

> Sunzu is a synthetic humans research platform. It turns your customer data into "synthetic humans", AI models grounded in your real evidence, so product, design and marketing teams can run interviews, concept tests, usability tests and ad tests in seconds instead of weeks. Every answer is traceable to a source.

Sunzu is a service and brand of Alpha Base OS Ltd (United Kingdom), registered with the ICO (registration C1103931). Contact: hello@sunzu.com.

## What synthetic user research is

Synthetic user research uses AI synthetic users, grounded in a company's own customer data (CRM exports, interview transcripts, surveys, NPS, support themes, product analytics, prior research), to answer research questions quickly. Unlike asking a generic LLM to "roleplay a user", Sunzu grounds every synthetic respondent in your data and makes every output traceable for methodological review. It complements human research; it does not replace it.

## Key facts

- Category: synthetic user research / AI-driven user research.
- Speed: answers in seconds, versus weeks for traditional recruiting.
- Grounding: synthetic humans are built from your customer data, not generic training data.
- Traceability: every answer is cited to a source, or an explicit abstention is recorded.
- Validation (internal back-testing): about 10% deviation in quantitative concept testing; in a matched-study benchmark Sunzu independently recovered 83% of the findings of a six-person human study and 64% of the findings of a 100+ person study, versus 28% and 22% for a general-purpose LLM (about 3x more); Sunzu produced no findings that contradicted the human study, while the LLM produced two; back-tested against human interviews.- Methods supported: depth interviews, concept tests, usability tests, ad and creative testing.
- Access: web app; directly from Claude, Slack or Microsoft Teams; MCP endpoint (https://app.sunzu.ai/mcp) to launch depth interviews and concept tests from Claude, ChatGPT or any MCP client.

## MCP

Sunzu exposes a Model Context Protocol server at https://app.sunzu.ai/mcp (streamable HTTP; OAuth for Claude and ChatGPT, or API keys). Client tools let an agent list audiences, launch depth interviews (a goal plus 3-5 hypotheses) and concept tests (1-3 variants with image, PDF or video stimulus), poll study status, and fetch reports (executive summary, key themes, hypothesis verdicts, recommendations, winning variant). Results are synthetic: directional signal from research-grounded personas, each answer traceable to a source. Usability and ad testing run in the web app.

## White papers

- [Engineered, not prompted: how we build synthetic humans](https://sunzu.com/papers/2/): the two-stage method (data enrichment, then behavioural modelling), grounding, confidence-rating, and two validation gates before a synthetic human answers.
- [9/10 teams make the same mistakes](https://sunzu.com/papers/1/): the common pitfalls in AI user research and how to avoid them without faking it.
- [All white papers](https://sunzu.com/papers/)

## Key pages

- [Home](https://sunzu.com/)
- [What is synthetic user research (with comparison table)](https://sunzu.com/#what-is)
- [How it works](https://sunzu.com/#workflow)
- [Use cases](https://sunzu.com/#usecases)
- [Trust and validation](https://sunzu.com/#trust)
- [FAQ](https://sunzu.com/#faq)
- [Book a call](https://meetings-eu1.hubspot.com/sunzu/intro)

## Founders

- Alex Street (https://www.linkedin.com/in/alexstreet-london/)
- Yoann Lemain (https://www.linkedin.com/in/yoannlemain/)
- Ludovic Lafole (https://www.linkedin.com/in/ludovic-lafole/)