The typical Bandit MaxDiff strategy assumes your main interest is to identify the most preferred items for the overall population. Rather, if your goal is to identify the most preferred items for targetable segments of the population, then you should use a segmented Bandit MaxDiff approach.
Let’s imagine that previous research had identified six known and targetable segments (such as by geography, income, age, risk tolerance, brand preference, or some combination of such traits). Upfront, we could ask a few questions needed to assign respondents into one of the six segments. Six identical Bandit MaxDiff exercises could be programmed within the Lighthouse Studio questionnaire and respondents would be skipped into the exercise prepared for their segment. With Lighthouse Studio’s implementation of Bandit MaxDiff, means and variances are estimated only using the data within each MaxDiff exercise, so the Thompson sampling procedure would be customized for each respondent group. The data later could be combined for analysis, for example, using HB analysis with the six-group variable as a covariate (assuming the data were not very sparse). Or, if the data were particularly sparse, aggregate logit could be run separately within each respondent group.