Perhaps you're asking about "borrowing" in hierarchical Bayesian MNL models (which we often use to analyze MaxDiff and CBC data)? If so, "borrowing" is how folks sometimes explain what goes on in HB, but it is an inaccurate simplification of what goes on - this is unfortunate because "borrowing" makes you think there's some imputation going on, like a "hot deck" correction for missing data, say.
What we really mean is that the dat for the population give us information about the covariance matrix (how the utilities are correlated with one another in the population). HB then blends information about the population's preferences/utilities (inferred from the covariance matrix) and information about an individual respondent's preferences/utilities (from her observed choices).
When we have a lot of information from the respondent then we use very little of the information from the population to temper that respondent's utilities, but when information at the respondent level is sparse, we rely a bit more on the population information (which we're borrowing in this specific sense only).