I can imagine that you set up such an architecture when you MUST do it: you cannot do another way. Because the performances are not as good as a standard application (certainly by far, as we got with H OLAP architectures in the past : part of the data was in the cubes, part was in the relational databases)
I would say you will use it:
- for real time data: like production process for example (you need to fetch the most up to date data)
- when you do not have the right to perform a partial load (access right)
- we can imagine also that the volume of data is too big. 90% of the business requests will be done on 20% of data stored in QV: you accept to fetch some data "outside" for a particular question in a particular sheet
I am interested much by the answers by other members.