When I've done affinity walls in the past, it's basically been analyzing a process where all of the interviewees/data comes from people working relatively closely together or at least from the same organization. If all of your interviewees come from different organizations, do you think that affinity walls would still be useful? Or would you use a different method for this?
The more you want to segment your data, the more raw data you need. But segmenting data is always a good thing if you have a statistical valid population. This could also be useful if you have several distinct target audiences to your site/application. Then you could analyze different behaviors on different groups, which would make it possible to address these problems. However, on a public web site with anonymous users, segmenting may not be of value since you can't identify the anonymous users’ audience belonging.