In data work, an ontology names the things a business deals with, such as customers, contracts and shipments, states which properties each one has, and defines how they relate: a contract belongs to one customer, and a shipment fulfils one order line. Standards such as the W3C OWL 2 Web Ontology Language let teams write these definitions in a form software can reason over.
An ontology sits above individual tables. Two systems may both have a column called status, and the ontology records whether the two mean the same thing and which values each one allows.
What an ontology usually defines:
- classes of things the business tracks, such as customer, product or claim
- the properties of each class and the values they may take
- relationships and constraints, such as one active contract per customer for each product
How is an ontology different from a knowledge graph?
- An ontology is the model of meanings: the classes, properties and rules.
- A knowledge graph holds the actual records, connected according to that model.
- A semantic layer maps business metrics to queries for analytics, and often borrows definitions from an ontology.
Where does an ontology break in AI work?
AI agents that read an ontology inherit its gaps. If a definition was written for reporting and never covered exceptions, such as a refund that reopens a closed order, an agent applies the clean rule to the messy case and answers with confidence. An ontology tells an AI system what terms mean. It does not show that the records still follow those meanings, so teams still check the data a specific AI task uses against the definitions, with business rule validation, and recheck when either side changes.