EvidenceField

01 The gap

Research assumes a world that isn’t there.

The modern research stack — Zoom automation, web panels, transcription pipelines — quietly assumes a digitally fluent participant. In Vietnam, Indonesia, Brazil, and Mexico, people often won’t, or can’t, complete digital onboarding or sit for a remote study.

We learned this running a marketplace across 130+ countries: the method breaks long before the insight does.

The respondent the stack imagines: digitally fluent, always reachable, a screen away.

02 The evidence

Synthetic respondents fail exactly where research is hardest.

r = −.70

GPT–human response similarity falls with cultural distance from the United States.Atari et al. 2023, World Values Survey

107

countries where the pattern replicated, across five GPT generations. The model is most confident where it is least valid.Tao et al. 2024, PNAS Nexus

WEIRD in, WEIRD out.

Western, Educated, Industrialized, Rich, Democratic — the samples the models learned from.

The evidence wall — the real human material the argument turns on.

03 The model

So we built the model the science supports.

Synthetic panels

Where data is abundant — fast, cheap, exploratory.

5,441
responses in one synthetic pilot: $0.70
total cost, inside an hour

Perfect for piloting and questionnaire debugging.

Best for: piloting, questionnaire debugging, cheap first passes.

Human fieldwork

Real people, met where they are.

Synthetic data augments human research, it does not replace it.

— Qualtrics

AI does the transcription, translation, and synthesis it’s genuinely great at. It supplements the human; it doesn’t stand in for them.

Best for: data-thin markets, lived context, the truth on the ground.

Try the router

Synthetic panel, human-confirmed

72% synthetic28% human

Synthetic panels reproduce US aggregates closely enough to screen concepts fast — screening, not truth. 0.0004 KL-divergence vs. the real US vote split — Sun et al. 2024

Ask a real question — see how we’d answer it, in twenty seconds or read how the routing works

04 The product

Ask a question. Approve a Brief. Decide with receipts.

You ask, in plain language.

See a delivered study

05 The expert layer

Experts route every study. Every insight traces to real humans.

A research-design layer sits on top and routes each study to the right mode — synthetic pretest, in-person fieldwork, or both — so the method always matches the market. Every recommendation carries its receipts: the field interviews, synthetic pilots, and sources behind it.

Built by an operator who ran a $2B+ earner marketplace across 130+ countries — global research that scales, honestly.

Field interviews
Synthetic pilots
Sources & citations

The creed

Synthetic where data is rich.Human where it isn’t.Expert judgment deciding which.

Start a conversation Building with us? Read the plan.