Some workloads fit MongoDB 1,815 badly enough that no amount of modeling skill rescues them.
Money that must balance. Double-entry accounting, inventory reservation and anything with a legal audit requirement wants constraints the engine enforces on every write — foreign keys, CHECK, composite uniqueness, and a transaction touching several tables as one unit. MongoDB has multi-document transactions, but they are a tool you reach for deliberately, not the default state of a write.
Reporting nobody wrote the schema for. When analysts join five entities in a way no one anticipated, SQL is the interface they and every BI tool already know. Aggregation pipelines are expressive, but you write them yourself, forever.
Highly relational data read from many directions. If the same entity is read per user, per order, per region and per month, no embedding choice serves all four. You end up with $lookup everywhere — a relational database with extra steps.
Small, stable, single-server applications. A blog, an internal tool, a booking form: PostgreSQL 1,289 or SQLite 4,756 does this with less operational surface, and MongoDB's replica-set requirement for transactions pushes even a modest correctness need into a three-node deployment.
Data whose shape you have not thought about. "Schemaless" is often heard as "no decisions required." The decisions still exist; they move into application code, where nothing enforces them.
What MongoDB is genuinely good at is the inverse of that list: entities read as a whole, shapes that vary by tenant or source, write volumes that eventually need sharding, and teams that iterate on structure faster than a migration pipeline can follow. Those cases are common and well served.
Run this comparison against your own workload before the first commit, not after the first outage. Write down the five queries you run most, the one correctness rule you cannot violate, and the largest record you expect. If the queries are per-entity, the rule lives inside one document, and the record fits comfortably under 16 MB, MongoDB will serve you for years. If any answer points elsewhere, say so early: switching databases costs weeks at the start of a project and quarters once live.