LLM synthetic respondents are known to produce response distributions that are too narrow. Nobody writes down what that costs you. Running a conjoint through a mixed logit and into a market simulator, a panel with correct mean price sensitivity and compressed heterogeneity reproduced the human holdout choice shares almost exactly, then set the revenue-maximizing price 26% too low and sized the premium segment at less than half its true value. Simulated in R with logitr, cross-checked in Python by Gauss-Hermite quadrature.

The 2026 measurement consensus is triangulation: calibrate the MMM with an incrementality test. On simulated data with a known answer, the recommended recipe moved a badly biased ROI estimate about two thirds of the way to the experiment’s answer and then stopped, leaving it 77% above the truth. The prior lost to the likelihood, because a misspecified MMM is confidently wrong. Here is the diagnostic to run before you calibrate anything.