How EvidenceField works
The right method depends on where you ask.
Research tooling assumes a US-shaped participant. We route each study to the mode the evidence supports — synthetic panels where data is abundant, AI-augmented human fieldwork where it isn't, with expert design deciding which.
The routing instrument
Choose a market. Watch the method change.
The same question routes differently depending on where it's asked. Pick a market to see which mode engages, the evidence behind the call, and what you'd receive.
Market context
Synthetic panel, human-confirmed
Synthetic personas carry early screening; a lean human panel confirms the concepts worth pursuing.
Expert design & routing — the constant beneath every mix
Where a population resembles the model's training data, synthetic panels reproduce aggregates closely enough to screen concepts and debug questionnaires — fast and cheap. Screening, not truth.
— Sun et al. 2024, arXiv:2402.18144
What you getDirectional concept screening in hours, then a small human panel to confirm the handful that matter.
Put the human where the human is superior.
A 106-experiment meta-analysis found human–AI teams win only when the human is the stronger performer, and lose when the AI is — the scientific basis for routing, not blanket automation.
— Nature Human Behaviour, 2024
Under every study
One expert layer runs the whole sequence.
The instrument shows the decision; a researcher makes it. Every study moves through the same four steps, whichever mode it lands on.
Frame
Start from the decision the research has to serve, then name the market and the question underneath it. The mode follows from the market — not the other way around.
Route
A researcher chooses the mode per market and task — the logic of the instrument above, applied by a person. Synthetic where the evidence supports it, human where it doesn't, hybrid where the answer is split.
Field
Run it: synthetic panels, AI-augmented human fieldwork, or both at once. In data-thin markets, AI carries recruitment ops, transcription, translation, and synthesis so human effort goes where only humans can.
Synthesize
Reconcile the signals, flag where synthetic and human disagree, and deliver a graded, traceable read — every finding tied back to the evidence that produced it.
Human data stays the binding constraint: adding 100,000 synthetic responses to 10,000 human ones lifts the effective sample by only about 13%. — Hullman et al. 2026, arXiv:2602.15785
Where are you trying to ask?
Tell us the market and the decision behind it. We'll route the study to the method the evidence supports.
Start with your market