Describe the situations where an normalised approach to data modeling

for MongoDB would be preferable to a de$-$normalised $($embedded$)$

approach.