[justaaron]

Mongo itself, per-se? Never. (single write master = scaling bottleneck)

A document database? Whenever I wanted to lump certain "tables" worth of data together literally ALWAYS (as in modeling many real-world entities) in my CRUD operations. Literally the opposite use case of what the person below me outlined with "whenever the data structure may change too frequently" as I see relational-dbs directly applicable to aggregations (data analysis/metrics) and frequently-changing amorphous data structures in which various concretely and atomically known entities have their interactions and composition fleshed-out.

When you need to take that and run and run hard, documents make sense, then you can look at turning that into primitives (key:value mapping like ohm...?)

Incidentally, there are decent commercial MongoDB SAAS vendors, but running your own replica-set is somewhat cumbersome, and I've personally experienced data-loss suddenly on a self-hosted system after 1 year of flawless operation and 5 nines of uptime...