Obviously running ad-hoc queries against unstructured data is a use case for a NoSQL database, and MongoDB is good for that. This would include things like ecommerce shopping carts and other memcache type data. However, once the user clicks the final purchase button, then you want your order entry and fullfillment system running SQL Server.

There are also cases where very large raw files are injested from external sources, but only some columns and a much smaller subset of records are operationally useful and worthy for ETL into a normalized relational database. However, there may still be a need to retain a complete history of all these files in some online queryable fashion for archival or exploratory purposes. The DBA would prefer not to use SQL Server as a dumping ground for that type of thing. So, Hadoop to the rescue.

So, yes, are occasions where it's very useful to dump things on NoSQL. Find space for it on the rack and put it to work.