Here's where I've never been smart enough to use this model... help me explain how you deal with scenarios like this...
1) There's an incident and we are trying to replicate the issue, but our database is just shapeless blobs and we can't track down the data that is pertinent to resolving the issue.
2) Product management wants to better understand our users and what they are doing so that they can design a new feature. But they can't figure out how to query our database or how to make sense of the unstructured data.
3) QA wants to be able to build tests and set up test data sets that match those tests, but rather than a sql script, they've got to roll all this json that doesn't have any constraints.
These are just some examples, but maybe you get the gist of why I don't quite understand how this model scales "outside" the database.