When Ontologies Meet BI — Ontology-Driven Semantic Layers

Stardog, Anzo, EKG fabric — when the ontology IS the semantic layer.

0/1 done

Overview

Two worlds, same word

'Semantic layer' means two adjacent things in industry today:

  • The modern-data-stack semantic layer (dbt SL, Cube, LookML) — a YAML/JS/LookML catalogue of metrics and dimensions on top of warehouse marts. Pragmatic, BI-first, tabular.
  • The ontology-driven semantic layer (Stardog, Anzo / Cambridge Semantics, TopBraid, EKG vendors) — an OWL/SKOS ontology is the model, and queries are answered by rewriting SPARQL or GraphQL into SQL on the fly.

What ontology-driven adds

  • Multi-domain navigation — a question like 'show me all Pharma products whose adverse-event signal correlates with manufacturing batch X' crosses six domains. A flat YAML catalogue collapses under the cardinality; an ontology stays graph-shaped.
  • Semantic reasoningDrug ⊑ ChemicalCompound is declared once in OWL; a query for 'all chemical compounds in shipment X' transparently includes drugs without the analyst remembering the subclass relationship.
  • Enterprise vocabulary — finance, regulatory and operations teams share canonical entities (Customer, Counterparty, Trade, Position) defined once in the ontology, not five times in five YAML repos.

Where it makes sense

  • Pharma R&D, financial-services regulatory, oil & gas operations, government data fabrics — domains with many interlinked specialised vocabularies that already publish ontologies (FIBO, IDMP, OSDU).
  • SaaS startups with one product and ten dashboards: overkill — go modern-data-stack.

Cross-link: this track's Ontology Engineering and Semantic Web Foundations tracks teach the OWL/SHACL toolset that the ontology-driven layer uses underneath.

Neighbourhood directory vs Dewey Decimal

A YAML semantic layer is a single neighbourhood directory: a flat list of streets and shops, perfect for one town. An ontology-driven semantic layer is the Dewey Decimal classification used across every library in the country: every branch holds different physical books (rows) but inherits the same conceptual shelving, so cross-branch search ('find every 19th-century European chemist') just works. The price of that power is the discipline of maintaining the classification — the ontology — itself.

Reflect

The signal that an ontology-driven layer is the right tool: when stakeholder questions routinely cross five or more business domains, when the same conceptual entity lives in three warehouses with different column names, and when the dominant question pattern is 'show me X and everything related to X' — the kind of traversal that a flat metric catalogue cannot answer naturally.

  • Does your domain have a published industry ontology (FIBO, IDMP, OSDU, schema.org, …)? Is anyone using it?
  • What's the most multi-domain question your business asks today — and how brutal is the SQL that answers it?

Reading in progress · 0 of 1 activity done