It depends a bit on what kind of models you're running and what your segments are. Assuming your segments are groups of respondents defined by some other variables in the study (e.g. demographics, usage variables) and assuming you're running hierarchical Bayes MNL, you can run traditional "frequentist" stat tests with the respondent level utilities, just like you might do for any other statistical test of any other survey research entity. More appropriate will be to run Bayesian tests. In either case you can contact me for more details if you like: email@example.com.