Why not do Agile (EAI/SOA) projects?

The Agile approach is not new anymore, but still I see problems with trying to do projects the Agile way.
What I see:

1. Business plans have to be made with functionality and ROI upfront
   This does not fit the Agile way (Customer Collaboration over contract negotiation). The ROI is sometimes hard to calculate and what you see are naive numbers just to get the project approved. Furthermore it is a long and costly phase. Of course you have to discuss if a functionality is worthwhile.
   
   
2. The business wants to know upfront what the costs are and the needed resources
   This is because resources (people) need to be reserved and hired. They want to know the investments needed in advance because usually budgets are given each (half) year.
   
   
3. A project is divided in several subteams which must be coordinated
   The teams only communicate in the beginning (for the requirements) and at the end (integration testing). They do not communicate during the project and do not integrate during the project. I call this the big bang approach.
   
   
4. Projects do not want to pay for future advantages for the enterprise
   Especially in the beginning architectural decisions, POCs and implementations have to be made. Think of product selection for BPEL, BPM, ESB, etc. The business does not want to pay for all these (needed!) upfront investmentsfor the sake of the enterprise.
   
   
5. Documents are made just to satisfy the QA process
   Technical design, unit test plan for example are written upfront without review or written afterwards. These documents are most of the time outdated and are not maintained.
   

Ad 1
Begin with small projects where the ROI is easy to calculate and small. Take a small part of a business process. Maybe we should even call it a feature/business process plan instead of project plan and keep them small and manageable.

Ad 2
You could reserve some budget for feature (enhancements) and after each iteration/sprint (or whatever method you use) choose to continue or stop. This way you never go over budget.

Ad 3
If you keep the features small a lot of communication may be circumvented and communication is easier. However in lots of cases the places in which the team work are different. Maybe modern tools like wikis, twitter, use groops, google talk etc may help to communicate better and often.

Ad 4
Some budget may be reserved not only for the business but also for enterprise architecture in general. This way such POCs and product selection trajects may be funded. This means management support is really needed, but this is always the case in such trajects.

Ad 5
I like the use of TDD (Test Driven Development) in which automatic tests are made and ran during builds. It takes some effort but it really is very helpfull during refactoring and regression tests. It also makes designs better because it forces you to think about the testability of the component. Also document only the most important design decisions and why some decisions are made. Those decisions can be very valuable in the future, instead of describing attributes of classes in detail for example. These can be (and should be) documented in the code.

I am curious to hear from you in other issues and/or solutions for these issues!