Data Lake or Data Warehouse? Why the Question Is Usually Wrong

“Should we build a data lake or a data warehouse?” is one of the most common questions we get early in a platform conversation, and it’s usually the wrong question. It treats two complementary layers as if they were competing options.

What each one is actually for

A data warehouse holds modeled, structured data, organized around the business logic that reports and dashboards depend on. It’s built for fast, predictable queries against data that’s already been cleaned and shaped. It’s the layer that should back your BI tools.

A data lake holds raw and semi-structured data at scale, often before anyone has fully decided how it will be used. It’s cheaper to store data in, more flexible about format, and built for the ingestion side of the pipeline — landing data first, deciding what to do with it after.

The two aren’t in competition. In most platforms we design, raw data lands in a lake, gets cleaned and modeled through a series of transformation layers, and the modeled result lands in a warehouse (or a warehouse-like layer within the same lake) for reporting to run against. The lake is where data arrives. The warehouse is where it’s ready to be used.

Where the “or” question comes from

The instinct to frame it as a choice usually comes from one of two places: budget pressure (build one thing, not two) or a bad experience with one of them in the past — a warehouse that became too rigid, or a lake that turned into an unstructured dumping ground nobody trusted.

Both of those are architecture and governance problems, not a reason to avoid one layer entirely. A lake without ingestion discipline becomes what people call a “data swamp.” A warehouse without a lake in front of it just pushes all the raw-data problems into the modeling layer, where they’re harder to fix.

What we actually recommend

Match the layer to the job: lake for raw ingestion and flexibility, warehouse (or a modeled layer on top of the lake) for the queries the business actually runs against. Get the ingestion and layering discipline right — raw, cleaned, modeled — and the “lake vs. warehouse” question mostly answers itself, because you’re no longer choosing one, you’re designing how they work together.