Organisations store analytical data in three broad kinds of system.
Data Warehouse
Stores structured, cleaned data in tables optimised for SQL queries and reporting.
Strengths: fast queries, strong consistency, governance and access control, familiar SQL. Limits: less suited to unstructured data (images, text files) and raw data of uncertain use.
Data Lake
Stores large volumes of raw data of any type — CSV, JSON, Parquet, images, logs — cheaply in object storage.
Strengths: flexible, low-cost, keeps raw data for future uses including machine learning. Limits: without discipline it becomes a "data swamp": undocumented, inconsistent and hard to query reliably.
Lakehouse
Adds warehouse-like features on top of lake storage using open table formats (such as Apache Iceberg, Delta Lake and Apache Hudi): transactions, schema enforcement, versioning and efficient SQL queries.
Strengths: one copy of data serving both BI and machine learning, open formats, time travel. Limits: more components to manage; maturity varies by platform.
Choosing
- Mostly structured reporting and analytics → a warehouse is often simplest.
- Large, varied raw data and machine learning → lake or lakehouse.
- Many organisations use a combination.
Whatever You Choose
Invest in data modelling, documentation, quality checks and access control. Architecture alone doesn't make data usable.