Semantic Models

In data management, we’ve often focused on physical structure – how to store, index, query, and process data efficiently. But as enterprises grow and data becomes distributed across cloud services, domains, and platforms, the harder challenge isn’t storage, it’s meaning. How does one business unit’s “client” relate to another’s “customer”? What does “exposure” mean in a trading system versus a risk model?

This is where semantic models step in. They sit above raw schemas and pipelines to capture meaning, context, and relationships, making data not just available but understandable.

A semantic model is an abstraction that defines the business meaning of data elements and their relationships, independent of underlying technical storage. Where a 3NF relational model concerns itself with tables, keys, and constraints, a semantic model describes concepts; entities like “Customer,” “Contract,” or “Product” and the semantics of how they interact.

Normal Forms

You can think of it as a bridge between business language and machine-readable metadata. A semantic model enables both people and systems to interpret data consistently, whether for reporting, machine learning, or compliance.

A simple example:

Database column: cust_id
Semantic model: “Unique identifier for a customer, defined as a legal or natural person who purchases goods or services”

The semantics here go beyond datatype, they encode business rules, ontological categories, and contextual meaning.

Why Semantic Models Matter

Disambiguation of Business Terms
In large organizations, different teams use different terms for the same thing—or worse, the same term for different things. A semantic layer clarifies meaning and prevents costly misinterpretations
Interoperability Across Systems
Modern enterprises run Snowflake, Databricks, SAP, Salesforce, and hundreds of SaaS products. A semantic model provides a unifying layer, allowing federated queries and data virtualization without redefining everything per system
Self-Service BI and Analytics
Tools like Power BI and Looker rely on semantic layers to let business users explore data without writing SQL. The semantic model defines measures (e.g., “Gross Margin”) and hierarchies (e.g., Product → Category → Division)
AI and Knowledge Graphs
Semantic models are foundational for AI-driven systems. Ontologies and graph databases like Neo4j or Stardog rely on semantic modeling to power recommendations, reasoning, and natural language interfaces
Governance and Compliance
Regulations (GDPR, BCBS 239, HIPAA) require clear data definitions and lineage. A semantic model provides traceability from “reported risk exposure” back to raw data elements

Historical Roots

Semantic modeling has its roots in the 1970s, alongside Peter Chen’s Entity-Relationship (ER) model. Where Chen formalized structural relationships, semantic modeling evolved toward meaning-based relationships.

Peter Chen’s ERD

Notable milestones:

1980s–1990s: The rise of conceptual modeling languages like NIAM and ORM, focused on natural-language-like semantics
2000s: Emergence of the semantic web and RDF/OWL standards, championed by Tim Berners-Lee and W3C
2010s: Graph-based approaches, with enterprises adopting ontologies for master data management and integration
Today: Cloud-native BI tools (Looker’s LookML, dbt Semantic Layer) embed semantic models directly into pipelines

Anatomy of a Semantic Model

A modern semantic model typically includes:

Entities / Concepts
Business objects such as Customer, Order, Account
Attributes / Properties
Descriptive fields (Customer Name, Account Balance)
Relationships
How entities connect (Customer places Order, Account belongs to Customer)
Business Rules
Cardinality, constraints, definitions (e.g., “Active Customer = customer with ≥1 order in the last 12 months”)
Hierarchies
Useful for drill-down and reporting (e.g., Region → Country → City)
Measures and Metrics
KPIs and calculated fields (Net Revenue, Gross Margin)
Semantics / Ontologies
Higher-level meaning captured using taxonomies, controlled vocabularies, or formal ontologies (e.g., FIBO in financial services

Semantic Models vs Other Models

Relational Models: Define data structures (tables, keys, constraints)
Dimensional Models: Optimize for reporting (facts, dimensions, star schemas)
Semantic Models: Define meaning, business rules, and domain context

They don’t replace one another but complement: a semantic model maps onto relational or dimensional schemas while providing a business-facing layer.

Modern Implementations

BI Tools
- Looker (LookML): Declarative model for business metrics and dimensions
- Power BI Tabular Models: Semantic layer defining measures, hierarchies, and relationships
- Tableau Data Model: Joins and relationships wrapped in semantic definitions
Data Platforms
- dbt Semantic Layer: Centralized definition of metrics for consistent analytics
- AtScale: Semantic virtualization platform across multiple data warehouses
Knowledge Graphs
- Neo4j, Stardog, Ontotext GraphDB: Ontology-driven semantic layers enabling reasoning and discovery
- RDF/OWL Standards: Formal semantics for linked data, used in healthcare, life sciences, and finance

Challenges with Semantic Models

Complexity: Designing an ontology requires domain experts and governance
Adoption: Business users may resist new layers unless integrated seamlessly into tools
Performance: Semantic virtualization adds abstraction that must be tuned for large-scale queries
Evolution: As business definitions change, the semantic model must evolve without breaking downstream use cases

Semantic Models in AI & LLM Era

Large language models (LLMs) like GPT are powerful but statistical, not semantic. They can generate coherent answers without truly “understanding.” Combining LLMs with semantic models, often called neuro-symbolic AI, promises more trustworthy, explainable systems.

Examples:

Using an ontology to constrain LLM outputs to valid product codes
Enabling chatbots to answer questions with definitions from the semantic model, not just surface patterns

This hybrid approach is gaining traction in enterprises wanting both flexibility of LLMs and rigor of semantic governance.

Case Studies

Financial Services
The FIBO ontology provides a semantic model of financial instruments, enabling interoperability across risk, compliance, and trading systems. Banks use it to reconcile data from multiple counterparties. Good luck trying to use it mind, it’s vast and I’ve personally never seen it add real value
Healthcare
SNOMED CT is a semantic model of clinical terms, ensuring consistent definitions across EHRs and medical research
Retail
Semantic product hierarchies allow retailers to unify online and in-store catalogs, driving consistent reporting and personalization

Best Practices for Building Semantic Models

Start Small, Deliver Value
Don’t model the entire enterprise upfront. Begin with a domain (Customer, Product) and prove value in reporting or compliance
Leverage Standards
Use industry ontologies (FIBO, SNOMED, schema.org) instead of reinventing
Embed into Workflows
Connect semantic models to BI, AI, and governance platforms to avoid being “shelfware”
Iterate and Govern
Treat the semantic model as a living artifact with stewardship, versioning, and review processes
Bridge Business and Tech
Involve both data architects and business SMEs—semantics are meaningless without shared agreement

The Future of Semantic Models

As data ecosystems become more fragmented, semantic models will be central to:

Data Mesh Architectures: Each domain owning a semantic model for its products, federated into an enterprise knowledge graph
Explainable AI: Grounding LLMs and ML predictions in ontological meaning
Automated Governance: Tools like Collibra, Alation, and Informatica embedding semantic layers for lineage, policy, and compliance
Query by Natural Language: Semantic models powering “SQL-less” analytics, where users ask questions and systems resolve them against meaning-rich metadata

Data Mesh

Conclusion

A semantic model is not just a layer on top of data, it’s the language of the business made explicit, computable, and shareable. It transforms raw tables into concepts, joins into relationships, and metrics into meaning.

In a world where data volumes keep growing, what we lack isn’t more pipelines, but more understanding. Semantic models bridge from storage to knowledge.