A well-known issue in "intercept" sampling is that it is not a probability sample of the customer population. If you were a grocery store interviewing 1/20 shoppers as they exit the store, for example, a customer who typically visits twice per week is twice as likely to participate in the survey as a customer who typically visits once per week.
Sometimes you don't care, because the transaction is the base you want to track. Grocery stores typically don't make any adjustments because, subject to certain simplifying assumptions, somebody who visits twice per week is exactly twice as important as somebody who visits once per week.
Sometimes you do care, because your image and how people talk about you is based on your entire population, not just your most frequent users. If you want the base for the analysis to be the individual user/customer, you would have to under-weight respondents by their frequency of visits or change your sampling procedure.
I'm trying to decide whether I should care or not. I'm updating a satisfaction survey for a website that currently has a certain percentage chance of appearing when the user logs out. Some users visit daily or more than daily; some visit less than once/week. More frequent users are more valuable, but it's not linear; a daily visitor is <7x as valuable as a weekly visitor.
If you were building a satisfaction survey for a website with regular users who vary in their frequency of use, would you base the analysis on the users or the visits?