Data modeling

Data modeling is the practice of defining how data is organized: what entities exist, how they relate to one another, and what structure tables or documents take so that data can be queried accurately and efficiently. It happens at different levels, from a conceptual model of business entities down to the physical tables in a data warehouse.

For analytics, the most common approach is dimensional modeling, of which the star schema is the standard pattern, separating measurable facts from descriptive dimensions to support fast OLAP-style aggregation. This differs from the normalized modeling used in transactional systems, which optimizes for consistent writes rather than fast analytical reads. Tools like dbt have made modeling an explicit, version-controlled step in the transformation layer rather than something buried in reports.

Good data modeling matters because it is the foundation everything else, dashboards, a semantic layer, machine learning features, is built on; a poorly modeled warehouse forces every downstream consumer to redo the same joins and fixes. Common pitfalls include modeling too early around a single report's needs instead of general business entities, inconsistent naming and grain across tables, and skipping documentation, which makes an otherwise good model unusable by anyone but its author, hurting overall data quality.