Landing Layer vs. Bronze Layer — When to Use Which?

[VTiw25]

I’ve seen two common approaches in data architectures:

1. Landing → Bronze: Raw files are first dumped into a landing layer, archived, and then loaded into Delta tables as Bronze.
2. Raw files = Bronze: The raw file dump itself is treated as the Bronze layer.

Both have their pros and cons, but I haven’t found a clear rule for when to choose one over the other.

What factors should drive this decision? For example, replay/reprocessing needs, retention, governance, data volume, cost, or source-system characteristics.

Would love to hear how others make this decision in real-world data platforms.

[balajij8]

@VTiw25 

It hinges on the source reliability and reprocessing needs. You can use a separate landing layer when ingesting from unreliable sources (APIs, third-party SFTP, partner feeds) where data may arrive corrupted, incomplete or need to be rejected before bronze. The landing zone acts as an immutable archive for replaying the exact original state without re-requesting from source. Choose landing when regulatory compliance or audit requirements mandate preserving exact raw files indefinitely, when source systems cannot re deliver historical data or when you need to decouple ingestion cadence from processing. You can skip the landing layer for high volume streaming sources (Kafka), or cases where the source system can replay data on demand. Direct to bronze with Auto Loader or streaming ingestion reduces storage costs and latency.

If schema evolution happens frequently requiring file level reprocessing or governance demands separating data received from data validated, the landing layer is a must. You can implement archival policies (move to cold storage after 30 days etc) to control costs.