For example
Instead of a field named, COLUMN_1, that holds an embedded document with fields A, B, and C.
Why not 3 separate fields with names COLUMN_1_A, COLUMN_1_B, COLUMN_1_C.
In mongoDB you prefer to work with documents the way they exist naturaly , afcourse you can create routines that can translate every time and assemble/disasemble the document:
COLUMN_1:{A:X,B:Y,C:Z}
to:
COLUMN_1_A:X,COLUMN_1_B:Y,COLUMN_1_C:Z
But this is additonal work that you dont want to do every time , you are a lazy effective developer that prefer just store your json document the way it will be used unless there is a specific use case that make sence to do it this way ... :)
Also please, note embedding is possible up to 100 levels , but max document size is still limited to 16MB , so in your document model it is not expected that you embed all your database in single document...
When you need performance , you add indices and optimize queries , that way your most searched data stay in memory , there is no big difference if your fields are embeded or in the root if they are not indexed ...