Why is a database schema insufficient?
A schema describes storage, but a column such as “amount” may mean gross price, payment, or refund. A semantic model maps fields to explicit business concepts and explains which values can be aggregated, filtered, and joined. Reports and agents can then query meaning rather than guess from physical names.
Which objects and relationships belong in it?
Entities identify business objects such as orders and customers. Measures are aggregatable values, dimensions group or filter them, and grain specifies what each row represents. Relationships define valid join keys and directions. Metric definitions can add filters, time windows, and formulas. Descriptions, synonyms, versions, and access rules help the semantic layer serve different tools consistently.
For example, joining item-level orders to refund events may duplicate sales amounts. The model should specify each table’s grain, valid join path, and net sales calculation. Defining business semantics in the data engine illustrates this approach.
How does it differ from a data model?
Data models can describe physical, logical, or conceptual structures. A semantic model focuses on business meaning and analytical rules exposed to users. Friendly column names alone are insufficient if join and aggregation boundaries remain ambiguous. Business and data teams should approve definitions and update them when policies change.
FAQ
Is one model enough for every team?
Not always; shared concepts can coexist with scoped team definitions.
Can models be used by AI agents?
Yes, to resolve terms and generate valid queries.
How are changes managed?
Version definitions and test affected queries and reports.