The problem I had was mongo doesn't scale super well when really going to big data-sets and sharding it is very ops heavy
I considered Cassandra but decided against it because it's also very ops heavy and documentation (at the time) was lacking
Hence we went to DynamoDB as it's super scalable, sharded by default and doesn't require ops (managed service). Beware though the data modelling is very different (and very close to Cassandra)