I know this is basically quoting the CPM, but the only factor to consider is the user(s)' view. The user may be a person or another system/application, but the boundary is defined by the users' view. If you're struggling to define the application boundary it can help to look at the transactional functions first. Since the transactional functions must cross the boundary, you can for example look at a report to see where the data is coming from, and where it's going. If it's going to the printer/screen/email, then you have the "user" side so now you just look at what database the data came from. Though some of the databases may be EIFs and external to the boundary, there are most always elements coming from ILFs. Find the ILFs and you are now looking inside the application boundary.
Thank you very much for your detailed answer. The other question would be, whether an application system can include other application system. For example: A typical web application (e.g. Online Shopping) consists of more than one tiers. One of them is the Enterprise Service Bus. This tier does not matter for the web application “Online Shopping”. But the tier “Enterprise Service Bus” is from the project perspective an application with its boundary. Would you also count the tier “Enterprise Service Bus” as application system with its boundary or not?
I think I understand what you're talking about. I just had a count with something similar. I needed to include the functionality of this "middleware" piece because we were looking at internal productivity metrics on the project, not just functionality delivered to the user (in which case we wouldn't count the "Bus"). For middleware you basically treat it as a separate application and do a count as you would for the online shopping. Two applications combine for a single project FP count.