What is a knowledge graph for business data?

Summary

A knowledge graph for business data is a unified repository that connects metrics, definitions, tables, and relationships in a governed semantic layer. It helps AI agents and analytics tools understand which data to trust, which definitions apply to specific business contexts, and how data elements relate to each other. Knowledge graphs reduce the time agents spend exploring and guessing, while ensuring consistency across questions. Databricks Genie Ontology implements this as the business understanding layer for enterprise AI, combining governed semantics from Unity Catalog with automatically learned context from dashboards, queries, and organizational work.

What is a knowledge graph for business data?
A knowledge graph for business data is a semantic layer that represents the business meaning, relationships, and governance rules embedded in an organization's data. Instead of treating all data sources as equally valid, a knowledge graph distinguishes which tables are authoritative, which definitions apply in different contexts, and how metrics relate to each other. This shared understanding helps AI agents answer business questions accurately and consistently without spending time and tokens exploring the data landscape.

Why knowledge graphs matter for AI

Modern AI agents can query any database or table, but they lack context about which one to trust. Without a knowledge graph, an agent might calculate revenue using three different definitions for three different questions, or select a stale table over a freshly maintained one. A knowledge graph provides this context up front: which revenue definition does Finance use? Which table has today's data? Who is the expert on this metric? By answering these questions before the agent begins reasoning, a knowledge graph reduces exploration, speeds up answers, and improves reliability.

Key capabilities of business knowledge graphs

Capability	Why it matters
Semantic coverage	Maps critical business metrics, definitions, calculations, and relationships so agents can identify the correct data sources and definitions for each question.
Authority and trust signals	Ranks context by certification, usage patterns, freshness, and expert verification so agents use the most authoritative version of contested metrics (revenue, customer count, churn rate).
Permission enforcement	Ensures agents respect underlying access policies so a user never sees data they're not authorized to access, and consistency is maintained without compromising security.
Automatic learning	Extracts and maintains business context from existing work (queries, dashboards, notebooks) rather than requiring manual documentation of every relationship and rule.
Live data reasoning	Enables agents to calculate answers from current data rather than retrieving stale documents, ensuring results reflect today's business state.
Integration with AI systems	Makes the same business understanding available to multiple agents, copilots, and applications through standard interfaces, preventing duplicated context engineering.

How Databricks approaches knowledge graphs

Databricks Genie Ontology is a knowledge graph designed as the business understanding layer for enterprise AI. It combines explicitly governed semantics from Unity Catalog (including metrics, domains, glossaries, certifications, and access policies) with context automatically learned and refreshed from the organization's data assets (queries, dashboards, notebooks, pipelines, and analytical work). Genie Ontology uses an authority-aware ranking system (OntoRank) to evaluate signals such as certification, usage patterns, freshness, authorship, and domain expertise, identifying the most relevant and authoritative context for each question while enforcing the permissions of underlying Unity Catalog assets.
When a Genie agent (or any agent accessing Genie Ontology through Genie MCP) receives a business question, Genie Ontology retrieves and injects relevant context before extensive exploration begins. This dramatically reduces the number of exploratory queries, tool calls, and tokens needed to reach an accurate answer.

Common use cases

Finance and reporting: Agents answer questions about revenue, costs, margins, and cash flow using the certified definitions Finance has established, reducing manual reconciliation and report review.
Customer analytics: Sales and marketing agents quickly identify the correct customer count, lifetime value, or churn definition for the region or product line in question, improving consistency across reports.
Operational decisions: Operations and supply-chain agents use current, authorized operational definitions and calculations to recommend inventory levels or staffing, reducing delays from manual verification.
Cross-functional analysis: Data teams expose key metrics and relationships through a shared knowledge graph, allowing analysts and agents across functions to work from the same business definitions.

FAQs

How is a knowledge graph different from a data catalog?

A data catalog documents what data exists and where it lives; a knowledge graph documents what the data means and how to use it. Catalogs answer "where is the revenue table?" while knowledge graphs answer "which revenue definition should I use and why?" Catalogs are descriptive; knowledge graphs are semantic and reasoning-enabled.

Do we need to manually build a knowledge graph, or can it be learned automatically?

Both. A knowledge graph benefits from explicitly governed semantics (certified metrics, validated definitions, access policies) that a team maintains intentionally. But a mature knowledge graph also learns from the organization's existing work (queries, dashboards, notebooks) to capture the long tail of definitions and relationships that teams actually use, reducing the burden of manual documentation.

Can a knowledge graph enforce security and compliance?

Yes. By enforcing the same access policies as the underlying data sources, a knowledge graph ensures agents and users only see data they're authorized for. This allows organizations to build a unified semantic layer without creating new governance risks.

What happens when business definitions change?

Automatically refreshed knowledge graphs detect when definitions, calculations, or trusted sources change (by monitoring usage patterns, certifications, and freshness signals) and update the context agents receive. This helps answers reflect current business conditions rather than stale semantic models.