Enterprise Context Layer
Does our AI understand how our business works?
Models know the world, not your company. Shared definitions, relationships and rules turn generic answers into decisions you can act on.
Maturity
1ExploringExample
Context lives in prompts and in people's heads.
2Establishing
Semantic models for BI. Document retrieval built per project.
3Scaling
Shared business definitions reused by analytics and agents.
4Compounding
An enterprise ontology agents reason over, run as a product.
Key decisions
- Extend BI semantic models or start an ontologyReuse what exists versus model the business properly
- Document retrieval or structured knowledgeSpeed versus precision
- Central or domain-owned definitionsConsistency versus ownership close to the work
Anti-patterns
- Every agent team re-inventing what a customer is
- Retrieval over documents as the entire strategy
- Context with no owner and no versioning
Measure progress
- Share of agents using shared definitions
- Answer accuracy on business-critical questions
- Time to connect a new agent to enterprise context
Builds on
- Semantic models sit on a central data platform.AI-Ready Data Estate at Establishing, needed for Establishing
- Shared definitions need owned data products underneath.AI-Ready Data Estate at Scaling, needed for Scaling
- Definitions need a forum that can sign them off.Governance & Trust at Establishing, needed for Scaling
- An enterprise ontology is kept alive by the domains that own it.Organization, Skills & Operating Model at Scaling, needed for Compounding
Plays
Microsoft-specific guidance, as of September 2026.
Ratings are illustrative and pending author review.
Capabilities
| Capability | Role in this pillar | Rating |
|---|---|---|
| Power BI semantic models | Reusable business definitions already in use | Adopt |
| Fabric IQ | Business ontology on top of OneLake | Assess |
| Foundry IQ | Knowledge retrieval for agents | Trial |
| Work IQ | Context from how work happens in Microsoft 365 | Assess |
Adopt
Proven. Use it by default for this purpose.
Trial
Production-ready. Use it with a clear scope.
Assess
Promising. Explore with a small team.
Hold
Not recommended for new work.
Known gaps
Ontology tooling is young. The hard part is modeling discipline and ownership, not the product.