In the world of UI design there is a technique called Personas. It’s a technique in which you sketch a stereotype user for every stereotype you can find in the group of users which is using your UI. These sketches will live throughout the creation of the UI for aiding developers and analysts in keeping in mind what users are going to use the UI. It helps in making those small decisions in the UI. See http://www.naarvoren.nl/artikel/personas/ for an article about it.
But what if you have the whole world interacting with your system through webservices. Would this same distinction between stereotypes of persons be helpful when stereotyping systems? At the current place we have a lot of different systems integrating. Some systems are desktop applications which aren’t capable of receiving messages because the PC on which they are installed are simply turned off outside working hours. This stereotypes the possibilities you have when integrating with a desktop application. And what if a website is using your system? Should we also deliver a JSON or (plain) XML interface?
So instead of researching which actors a system has an analyst should also research the stereotypes inside a single actor. This would be of great help when designing the system. But maybe this is only helpful when creating middleware.