Fabric ontology: when to use it with a semantic model

Key insights

• A semantic model can supply the starting structure for a Fabric ontology.
• Ontology entities and relationships can be renamed to use business terminology.
• The demo's WinBackCandidate rule uses order count, days since last order and lifetime margin.
• An operational agent can use the ontology as a knowledge base and trigger a pipeline.
• A Data Activator approach can also automate an alert, but its logic may be distributed across separate activators.

Microsoft Fabric ontology adds a business-meaning and rules layer above a semantic model, helping teams use data to drive actions as well as reports. It is most relevant to Fabric builders and BI or business teams considering automated actions from data. Reza Rad (RADACAD) [MVP] demonstrates an ontology-led customer winback workflow in a YouTube video.

What's new

The practical distinction is between describing data for analysis and using its meaning to guide an operational process. A semantic model already provides tables, relationships, measures and calculations that support reporting. In the demonstrated scenario, an ontology builds on that model with entity-level business terms, relationship definitions and rules that an operational agent can use.

That does not mean every organization with a semantic model needs an ontology. If the goal is to understand data through reports or dashboards, the model may be enough. The case for an ontology grows when a team wants to identify a situation from data and have an agent help initiate a response.

The example starts with customer, product and sales information, including calculations such as order count and days since last order. The ontology turns technical names into terms that make sense to the business, describes how entities relate, and holds a winback rule. An operational agent then uses that context to identify potential customers and prepare an email for approval.

Who it is for and what you need

This approach is aimed at Fabric builders and Power BI developers who already work with semantic models, as well as data and BI leaders exploring automated processes. The demonstration begins with a sales semantic model and uses customer retention information. It does not give a required license, plan, region, or minimum permission level.

The transcript says that several of the items demonstrated were in preview at the time of recording. It does not identify each preview item individually or establish current availability, so teams should verify the status in their own Fabric environment before planning a production workflow. The video also does not provide a detailed security or governance checklist.

You can generate an ontology from a semantic model or start one from scratch in the Fabric portal, according to the demonstration. The model can in turn contain data brought together through sources and processes such as a Lakehouse, Warehouse, pipeline, notebook or dataflow. These are examples in the demonstration, not a complete list of supported sources.

How it works

The example moves from an existing sales model to an ontology and then to an operational agent. The model provides the initial entities and relationships; the ontology adds business-facing terminology and rules. The agent uses the ontology as its knowledge base and connects to an action that runs a pipeline and notebook.

1. Open a sales semantic model in the Fabric environment and create an ontology from it, or begin from scratch. The generated structure reflects the model's tables and relationships.
2. Review the entities and rename technical labels. In the demo, DimProduct becomes Product, DimCustomer becomes Customer, and FactInternetSales becomes SalesOrderLine.
3. Give relationships business-readable names. For example, the relationship is described as “Customer places SalesOrderLine,” and the sales order line “contains Product.”
4. Add a rule associated with the Customer entity. The demo's WinBackCandidate rule checks whether OrderCount is more than 2, DaysSinceLastOrder is between 90 and 180 days, and LifetimeMargin is more than $2,000.
5. Create or configure an operational agent, add the ontology as its knowledge base, and connect an action. The example uses a Fabric pipeline that runs a notebook to generate an email and sends it for approval in Teams.

The approval step matters: the demonstrated flow prepares a message and waits for approval before sending it to a customer. The presenter notes that the approval step could be removed, but the example keeps it in place. The transcript describes a sample result from a previously created agent; it does not provide a repeatable production deployment guide.

Limitations and open questions

An ontology is not presented as a replacement for a semantic model. The semantic model remains the source structure in the example, while the ontology adds business definitions and rules for an action-oriented workflow. If your users only need reporting, the demonstration does not make a case for adding this layer.

The presenter also acknowledges an alternative: a team could use a semantic model with Data Activator to raise an alert and run a pipeline. His argument for ontology is that separate activators can leave business logic in multiple places, while an ontology can centralize rules and definitions. That is a design rationale from the demonstration, not a claim that Data Activator cannot support the workflow.

Important implementation details remain unspecified, including licensing, exact permissions, deployment controls, and current preview or general availability status for individual components. The example uses a sales dataset and a particular Fabric workflow; it does not establish that every environment has the same items or configuration.

Sources

• Reza Rad’s Fabric ontology demonstration

Frequently asked questions

Do I need a Fabric ontology if I already have a semantic model?

Not necessarily. The demonstration positions a semantic model as sufficient for reporting, while an ontology is useful when you want business rules and context to support actions based on data.

Can I create a Fabric ontology from a semantic model?

Yes. The demonstration shows generating an ontology from a semantic model or starting from scratch in the Fabric portal.

How does an ontology help an operational agent?

In the example, the ontology supplies business terminology and a customer winback rule. An operational agent uses it as a knowledge base and triggers a pipeline that prepares an email for approval.

What license do I need for Fabric ontology?

The source material does not state a required license or plan.