From a persona's text answer to a score with a confidence interval — every step is documented and deterministic. This is the math behind Senthiq.
The model doesn't guess the score. The persona answers in text — and the number is derived by deterministic math: an embedding, comparison against scale anchors and a probability distribution.
*Illustrative values. The score is derived deterministically from the meaning of the answer — more stable and more robust to question phrasing than a number guessed by the model.
Every persona carries a weight matching its share of the target population. The panel score is a weighted mean of individual answer distributions — minority segments don't vanish into the majority.
Each result ships with a bootstrap confidence interval and significance of differences. You know whether the gap between variants A and B is real or just noise.
Synthetic respondents tend to be systematically too positive. Senthiq measures this error and actively corrects it — results reflect critical voices too.
Upload real campaign or survey results and Senthiq measures how far the simulations deviate from them. Accuracy is not a promise — it's a metric.
Panel composition comes from a deterministic sampler with a stored seed. The same brief means a reproducible panel — no hidden lottery inside the model.
Every result carries a confidence score derived from answer consistency, calibration quality and domain coverage. Below the threshold, we recommend real-world validation outright.
For entirely new, unprecedented products confidence drops — comparable data is missing and Senthiq admits it in the confidence score.
Simulation does not replace real research for critical decisions — it helps you target it where it delivers the most value.
Accuracy depends on the quality of the audience brief and calibration data. A vague brief means a wider interval, not false certainty.
