About
A research operating system, not a persona toy.
Evidence is the positioning. Field is the moat.
Origin
We didn’t believe the pitch, so we tried to break it.
EvidenceField started in July 2026 as a study of the “synthetic humans” pitch — AI personas sold as a replacement for real research participants. Instead of taking the pitch at face value, we ran an independent deep-research pass: 104 agents, 22 sources, 110 claims extracted, and the top 25 claims run through adversarial, independent skeptic panels. All 25 survived.
The finding: the pure-synthetic pitch is overclaimed at its core, but genuinely sound in a narrow niche — and the exact shape of where it fails is the opening for this company. Evidence-first isn’t a tagline here. It’s how the company was founded.
Why we exist
Every research stack assumes a user who looks like Silicon Valley.
Remote panels, digital recruiting, and English-first research tooling all assume a participant who is online, urban, and culturally close to the United States. Most of the next billion users aren’t. And the industry’s newest fix — LLM synthetic respondents standing in for real people — turns out to be least reliable exactly where research is hardest: similarity between AI-generated and real human responses falls as a country’s cultural distance from the US grows, at r = −.70 (Atari et al. 2023).
The gap nobody covers is the combination: emerging-market fieldwork operations, AI-assisted synthesis, and an expert layer that knows which mode to trust where. We built EvidenceField to close it.
What we believe
Synthetic where data is rich. Human where it isn’t.
We believe in market-adaptive research: synthetic AI panels for early screening and questionnaire debugging where digital, English-adjacent data is abundant; AI-augmented human fieldwork where it isn’t; and an expert layer that routes every study to the mode it actually needs.
Human data is the binding constraint. Even the best statistical hybrid — calibrating 100,000 LLM responses against 10,000 real ones — adds only ~13% effective sample size (Broska et al. 2025). That isn’t a gap model upgrades close. It’s the scarce asset our model is built around.
What to expect
Bring a question. Leave with a decision you can defend.
Working with EvidenceField starts with a conversation, not a scoping call: you ask a business question in plain language, and our research director sharpens it into a study you can read in one minute — the method, who we’ll talk to, the timeline, the price, and an honest account of what synthetic evidence can and cannot tell you for this question.
Approve it, and the study runs in the mode it actually needs — automated panels where they’re valid, real conversations in real markets where they’re not. What comes back is not a deck: it’s the answer, its confidence, a recommended action, and a finding-by-finding evidence trail where every claim links to the words of a real source.
If an insight can’t be traced to its evidence, we don’t ship it.
How we work
Every claim on this site traces back to a source.
We hold our own positioning to the standard we’re selling: every public claim traces to peer-reviewed or preprint research, cited inline. We’re honest about the limits of the model we’re building — and about the fact that competitors and incumbents alike have converged on the same conclusion: synthetic data supplements human research, it does not replace it.
That same rigor applies to how we talk about the markets and the people we work with. Care in language isn’t a compliance checkbox — it’s a research discipline.