Survey builds synthetic respondent panels from population data and uses them to forecast response distributions for draft questions. Before fieldwork begins, teams can test wording, compare alternatives, and identify potential acceptability risks.
Real fieldwork is expensive and takes time, and it often leaves little room to test multiple versions of a questionnaire. Survey lets you test the questionnaire first. Synthetic respondents drawn from population data answer draft questions, so you see the likely response distribution, wording risks, and questions that may need revision before fieldwork begins.
It complements fieldwork rather than replacing it. It helps teams sharpen questions, compare alternatives, and flag acceptability risks, then measures its forecasts against the completed survey.
Draw synthetic respondents from population data, then weight them using raking/IPF to align the panel with benchmark distributions for age, gender, and region, so it reflects the observed demographic structure of the target population.
Upload a questionnaire (CSV, XLSX, or JSON). Each synthetic respondent returns a probability distribution across answer choices rather than a single fixed response, which reduces over-concentration on one answer and preserves variation at the panel level.
Get predicted response distributions, subgroup breakdowns by age, gender, and region, and per-question risk flags for ambiguous, leading, or sensitive wording before fieldwork.
Once the real survey is complete, score the forecast against it. Distribution fidelity and top-choice agreement indicate where the panel performs reliably and where further calibration may be needed.
Synthetic response distributions are compared with population-weighted survey results across demographic groups. Below are results from our own end-to-end benchmark — reported with clear limits, including areas that still require calibration.
Respondents are synthetic and are not tied to identifiable individuals. We evaluate aggregate distributions across demographic groups rather than matching synthetic respondents to individual survey participants.
The panel performs better on questions with clear consensus and broad population-level attitudes. Perception-based, sensitive, and high-neutral-response items still need calibration, and each report identifies which item types require additional caution.
Every report includes confidence intervals, a fidelity score, and documented limits on reproducibility. Each one states the same caveat: this simulates a survey; it does not replace one.
The panel can be configured to model a city, a country, a customer segment, or a workforce. Forecasts can then be checked against subsequent fieldwork or an established benchmark.