What is an ontology?
An ontology is a formal, machine-readable model of a domain: the classes of things that exist in it, the properties each class has, the relationships between classes, and the rules that constrain them. In a manufacturing ontology, “Work Order” is a class with properties such as due date and status, it is related to “Asset” through a “performed on” relationship, and a constraint says every work order must reference exactly one asset. Ontologies are usually written in standard languages such as OWL or expressed as schemas for a knowledge graph.
An ontology differs from a taxonomy, which is only a hierarchy of categories, and from a database schema, which describes how one system stores data. An ontology describes meaning independent of any system, so several applications can map to it and agree on what their records mean. It is a central tool of knowledge representation.
Why an ontology matters
Companies accumulate systems that use the same words for different things and different words for the same things. A “customer” in the billing system may be a legal entity, while in the CRM it is a contact person. Reporting across them produces totals no one trusts. An ontology forces those definitions into the open, settles them once, and gives integration, analytics, and AI projects a stable reference. It is also the backbone of a knowledge graph, which stores actual instances (this work order, that asset) according to the ontology’s classes and relationships.
For AI, an ontology turns vague questions into precise ones. A language model asked for “late orders” can be grounded in the ontology’s definition of an order, its promised date, and what “late” means in the company’s terms.
Ontology in practice
A regional healthcare system adopts standard clinical ontologies and extends them with its own classes for facilities, care teams, and payer contracts. Patient records from three electronic health record systems acquired through mergers are mapped to the shared model. Analysts can now report readmissions consistently across all facilities, and a retrieval-augmented assistant for care managers answers questions using the same definitions the reports use.
What to watch for
- Scope the ontology to the decisions and integrations that need it, since modeling an entire enterprise up front rarely finishes.
- Build on published industry ontologies where they exist and extend them rather than inventing parallel terms.
- Give business owners authority over definitions and record the rationale for each decision.
- Treat the ontology as a living artifact with versions, tests, and a change process.