SYNTHETIC RESEARCH · METHOD

How a study works.

Positioning, pricing, and concept research answered in days by synthetic respondent panels. Here is the pipeline, end to end — including the parts most vendors don't show you.

Frame the decision, not the opinion.

We don't ask synthetic respondents what they think — we force them to choose. Every study is built as forced trade-offs around the actual decision: this option or that one, everything else held identical. Opinions flatter; trade-offs measure.

FROM A LIVE INSTRUMENTTwo hospitals identical in every way — same doctors, distance, wait, network — differing only in the name on the building. Which do you walk into?

Build the panel from real people's data.

Personas are sampled from US Census microdata on official person weights — income, age, region, household structure drawn from real distributions, never conjured from a prompt. The panel's marginals are checked against census targets before a single question is asked.

FROM A LIVE PANEL CHECK800 personas · ACS microdata, official weights · panel insured-rate 95.0% vs census ~94.5% — marginals on target

Calibrate against ground truth — and publish the misses.

Before trusting the panel on your question, we test it on questions with known answers. Anchors that land build confidence; anchors that miss are documented and carried as caveats, not hidden. A fictional-brand control arm runs alongside real names to measure what the models "already believe."

FROM A CALIBRATION LOGPublic-figure recognition: 1.0pt off ground truth · insurance coverage: 0.4pt off · care-engagement: ~~over-reports ~10pts — documented, treated as directional~~

Run at scale, with a second model watching.

Choices are order-randomized; known model biases are measured and absorbed in estimation. An independent second model shadows the full run on identical designs — findings that don't replicate across both models don't get published as findings.

FROM A RUN LEDGER32,975 forced trade-off responses · zero parse failures · cross-model shadow run: every published ordering holds

Label every claim by its evidence layer.

Every output states what kind of knowledge it is. Orderings, gaps, and segment differences are the claims we stand behind; absolute percentages are directional. No number in a report pretends to be harder than its source.

SAMPLED — from census microdata CALIBRATED — checked against ground truth MODELED — the instrument's inference

Validate against reality. Track quarterly.

Instrument accuracy is validated against real-world outcomes on public-domain questions — the hardest test available. And a study isn't a one-off deliverable: the same panel re-runs quarterly, same instrument, same segments, reporting the delta on every baseline your plan is built on.

THE OPERATING PRINCIPLEResearch that doesn't cascade into a KPI is trivia. Every finding pairs with a measured baseline and the mechanism that tracks it.

Bring the question you'd take to fieldwork.

We'll tell you honestly whether a synthetic panel can answer it — and where it should hand off to real respondents.

REQUEST A BRIEFING