As I said three months ago, beyond the surprising announcement of the new GenAI models developed by Anthropic (i.e Claude) in Azure AI Foundry, one of the true highlights of #MSIgnite 2025 was Microsoft Fabric IQ.

Since the announcement by Zia Mansoor, Arun Ulag and Yitzhak Kesselman, customer interest has skyrocketed. CIOs, CDOs, Chief Transformation Officers, and business unit leaders across industries are asking the same question:

“How can our organization implement Fabric IQ — and is it really worth the hype?”

It’s a fair question. Every year, the market is flooded with AI announcements, frameworks, architectures, “game‑changing” platforms, and new acronyms that barely survive a fiscal cycle.

So why is Fabric IQ different? Why are business leaders — not just technical teams — pushing hard to understand it and adopt it?

The short answer: Fabric IQ finally solves the hardest and most expensive barrier in enterprise AI — the lack of shared business meaning.

Microsoft Fabric IQ (currently in preview) is a core workload within Microsoft Fabric that establishes a shared semantic foundation across analytical data, operational systems, and AI agents. Its primary value is not storage or visualization, but business understanding at scale—ensuring that analytics, dashboards, and AI agents reason over data using the same business definitions and relationships.

When combined with Foundry IQ—Microsoft’s enterprise knowledge retrieval and reasoning layer—organizations gain something entirely new: a unified intelligence system that understands the business context, retains institutional knowledge, and can act on the organization’s behalf via AI Agents.

In this article, I share what I’ve learned over the past three months from studying and testing Fabric IQ firsthand. This isn’t hype. From my experience, it marks the beginning of the first practical, operationally viable enterprise intelligence stack. Let’s get started!

Why Business Leaders Want Fabric IQ So Badly

Business leaders don’t care about ontologies, graph structures, or agentic retrieval engines. They care about outcomes:

• Why can’t we make decisions faster?

• Why do teams present conflicting numbers?

• Why does every AI proof‑of‑concept get stuck at “prototype”?

• Why do we have thousands of dashboards but still no unified view of the business?

• Why can’t AI understand our organization the way people do?

Fabric IQ is the first solution from a major vendor that addresses these issues at the root, not at the analytics layer, not at the Business Intelligence (BI) layer, and not at the Large Language Model (LLM) layer.

It introduces something organizations have needed for decades: A consistent, dynamic, governed business vocabulary — shared across data, analytics, AI, and operations.

This is what business leaders have been asking for (often without the technical terminology) for years.

What Exactly Is an Ontology Model?

To understand Fabric IQ it's crucial to get familiar with the concept of ontology. It may sound academic, but in practice it is one of the most powerful business tools in the intelligent data stack.

What an Ontology Really Is?

An ontology is a formal, governed map of business meaning — a machine‑understandable representation of:

• Concepts (Customer, Asset, Contract, Order, Risk, Invoice, Shipment)

• Relationships (Customer places Order, Shipment affected_by DelayEvent)

• Rules & constraints (e.g., SLAs, thresholds, compliance definitions)

Where data models describe data, ontologies describe the business.

Why Ontologies Matter in the Enterprise?

Ontologies solve problems organizations have struggled with for decades:

• KPIs mean different things across departments

• Definitions drift as teams scale

• Critical business logic gets buried in SQL, Excel, BI models, and code

• AI systems hallucinate because the business context is unclear

• Operational decisions rely on undocumented tribal knowledge

With an ontology:

• Meaning is defined once

• Consistency is enforced everywhere

• Business rules become explicit

• AI systems can reason with context

• Humans and AI speak the same business language

An ontology is not a technical artifact — it is a strategic asset that modernizes how an enterprise thinks, aligns, and makes decisions.

Understanding the Difference Between Semantic Models and Ontologies

Recently, in The Data Massagist Newsletter, I wrote about the importance of semantic models in Microsoft Fabric and how their design directly impacts the accuracy of AI-based agents. When semantic models are intentionally prepared and optimized for AI consumption, the quality, consistency, and reliability of the agents’ responses improve significantly. You can read more here.

So, why do we need Fabric IQ if we already have sematic models in Microsoft Fabric? Here’s the distinction:

Semantic Models (Analytics-Focused)

Semantic models define tables, columns, measures, DAX logic and analytic relationships. They ensure consistent calculations and analytics. They power dashboards, KPIs, and BI experiences.

Here is a typical question it answers “What was revenue last quarter by region and product?”. It defines how data is calculated and analyzed.

But Semantic models do not define:

• Business meaning

• Real-world relationships

• Business rules

• Operational context

• Cross-domain reasoning

Fabric IQ Ontologies (Business-Focused)

Fabric IQ is a semantic intelligence workload that operates above and across semantic models, unifying business meaning, relationships, rules, and actions.

They define:

• Real-world business concepts

• How entities interact across domains

• Business rules and constraints

• Connected meaning across operational and analytical sources

Ontologies allow AI and agents to understand:

• Why things happen

• How concepts relate

• What actions are valid

• How insights across systems fit together

Therefore, a typical question Fabric IQ answers could be “Why did margin drop, which customers were affected, what operational factors contributed, and what action should we take?”. It defines what data means in the business and how it connects to decisions.

Currently, Fabric IQ provides two methods to set up an ontology item:

1. Generate from a semantic model You can leverage an existing semantic model to automatically generate an ontology, which you can then refine and extend. This is my recommended approach, as most organizations already have a well-structured semantic model that accurately represents their business domain.

2. Build directly from OneLake As with other data engines in Fabric, you can create the ontology manually by binding properties directly from OneLake. This option is ideal if you do not yet have a semantic model or if you prefer full control over the ontology design from the beginning.

A third method that is coming is a Copilot for Fabric IQ to help business user to create or improve an Ontology Model. More information is coming soon.

As said, Fabric IQ allows you to generate an ontology directly from a semantic model. It automatically generates a structured semantic layer that includes core ontology elements such as result tables and entities, along with their corresponding entity types. It defines columns as properties, establishes relationships between entities, specifies relationship types, and identifies keys and entity identifiers. This automatically generated ontology provides the foundational structure that enables consistent reasoning, governed querying, and accurate AI-driven responses.

There is no minimum Microsoft Fabric capacity required to use Fabric IQ — I tested it successfully on my F8 capacity environment. The key requirement I’ve observed is ensuring that Power BI is enabled in the Fabric tenant settings; otherwise, creating a new ontology item may fail.

This point is critical—and often misunderstood.

• DAX measures are not converted into ontology logic.

• Calculated columns are not treated as business rules.

• Time-series bindings are not created automatically.

In other words, measures and calculated columns are not supported in ontology bindings. The reason is simple but fundamental: ontology is about meaning, structure, and relationships—not mathematical computation. KPIs and metrics remain authoritative within the semantic model itself. The ontology can reference analytical constructs, but it does not replace them. Semantic models continue to serve as the system of record for metrics.

What traditional semantic models cannot do — and what Fabric IQ enables — is the automatic materialization of a graph derived from the ontology model. This capability unlocks true impact analysis, dependency traversal, and the ability to answer questions such as: “What is affected if X happens?”

For example, a conceptual query like “Find all customers affected by delayed shipments due to weather events” is not feasible within a pure semantic model, because it requires traversing multi-hop relationships across business entities and their contextual dependencies.

Since Fabric IQ is in Public Preview, there are some limitations we need to know when generating ontology models:

• Semantic‑model–based generation constraints: When generating an ontology from a semantic model, it is subject to Power BI semantic model limitations, including model size limits and XMLA endpoint constraints.

• Lakehouse restrictions: Ontology generation only supports managed Lakehouse tables located in the same OneLake directory; external tables or tables pointing to other locations (ie. Shortcuts) aren't supported yet.

• Data binding restrictions: Ontology data binding isn't supported yet for semantic models in Import mode. In the case of Direct Lake semantic models, we can't bind it yet if they the backing Lakehouse has OneLake security enabled.

In summary, Ontology and Graph operate together to deliver both a visual graph representation and an advanced query experience grounded in your business concepts — moving beyond dimensional aggregation into relationship-aware reasoning.

How They Work Together

• The semantic model calculates revenue.

• The ontology defines what revenue means, how it relates to products, customers, regions, risks, delays, or contracts.

• Agents then reason over both.

This pairing is what makes enterprise AI finally viable.

In Fabric IQ, the ontology model is not a conceptual layer — it is operational. It integrates directly with analytics and BI tools, including Power BI, which can use the ontology as a governed data source. Datasets can be automatically created or updated based on the ontology’s entities and relationships, ensuring consistent definitions of metrics and dimensions across reports and dashboards.

At the same time, analysts and data scientists can query data using business entities instead of raw table structures, aligning technical analytics with enterprise language and governance.

The Role of the Fabric Data Agent and Operations Agent

Fabric IQ becomes even more powerful when combined with Fabric’s two agent types:

Fabric Data Agent — The “Explain and Explore” Agent

The Data Agent is built for reasoning and insight generation.

It excels at:

• Answering natural-language questions using enterprise context

• Explaining why performance changed

• Surfacing patterns, trends, and relationships

• Combining ontology knowledge + semantic model metrics

• Connecting analytical insights with real business meaning

Think of it as the organization’s analytical co‑pilot — grounded in context.

We can connect a Fabric Data Agent to both an Ontology (Fabric IQ) and a Power BI Semantic Model at the same time. This is not only supported, but also an intended capability of Data Agents as they support multiple data sources simultaneously. This means a single Data Agent can have several sources at once:

• Lakehouses

• Warehouses

• KQL databases

• Ontologies

• Semantic Models

…of course, all connected concurrently.

Therefore, Data Agents officially support Ontologies as a data source. In fact, I was able to try it myself when I did the Ontology Tutorial at Microsoft Learn (Yes, Carlos Pardo , I keep following your advice). I do recommend you take a look to the Tutorial part 4: Create data agent using an ontology item.

Operations Agent — The “Monitor and Act” Agent

Operations Agents automate the observe → analyze → decide → act loop. They:

• Monitor real-time operational signals

• Detect anomalies, risks, or SLA violations

• Propose corrective actions

• Trigger automated workflows (with governance)

• Learn from outcomes and improve over time

Operations agents take Fabric IQ’s semantic context and apply it to live operations, becoming digital teammates that:

• Reduce operational latency

• Improve compliance

• Respond to issues faster than humans can

Think of it as the organization’s real-time, context‑aware operator.

Let’s see this through an example in the Oil and Gas industry.

In this scenario, an organization defines its business ontology in Fabric IQ, modeling entities such as Pipeline, Segment, Pump Station, Flow Meter, Weather Event, Environmental Risk Zone, Maintenance Order, and Operator Team.

At the same time, real-time telemetry—pressure, flow rate, vibration, and temperature—streams into Fabric through Eventhouse and KQL. An Operations Agent continuously monitors these signals and detects an abnormal situation:

• Pipeline Segment A-197 shows a sudden 18% drop in pressure,

• increasing vibration on Pump 4, and

• weather data indicates freezing risk along that segment.

Using the relationships defined in the ontology, the agent correlates these signals and identifies a potential issue. It then proposes corrective actions within a human-in-the-loop framework, such as reducing pump throughput, diverting flow to an alternate segment, notifying the regional control center, or triggering a diagnostic inspection. Once the root cause is confirmed—whether a minor leak, ice blockage, or compressor malfunction—the agent incorporates that outcome into the ontology by updating risk attributes or contextual relationships. The Operations agent also could generate real-time notification to the right people to take action. Over time, this feedback loop makes the entire operational system smarter, more resilient, and continuously improving.

Below you can see the architecture associated to this scenario:

How They Work Together with Fabric IQ + Foundry IQ

As Judson Althoff said last week during his keynote at the Microsoft AI Tour in Mexico:

Fabric IQ understands how your business works. Foundry IQ understands what your business knows.

For sure, that distinction is super powerful.

Where Fabric IQ brings unified semantics across structured, analytical, time-series, and operational data, Foundry IQ delivers enterprise knowledge retrieval across::

• Policies

• Procedures

• Contracts

• Emails

• Documents

• Images

• OneLake files

• SharePoint / M365 content

Together, Fabric IQ + Foundry IQ unlock capabilities no AI silo can match:

• AI that knows the relationships between your data.

• AI that cites official sources before responding.

• AI that recognizes business concepts consistently across departments.

• AI that learns from operational outcomes and improves organizational intelligence.

Combined, they form a closed-loop enterprise intelligence system — where knowledge, meaning, insight, and action flow seamlessly.

Let’s examine another scenario — this time in the telecommunications industry.

• Fabric IQ can unify network operations, telemetry streams, and business data into a shared semantic model and enterprise ontology. By harmonizing real-time signals with historical databases and network KPIs, it exposes meaningful patterns such as emerging outage risks, tower battery degradation, performance anomalies, and potential SLA breaches. Instead of isolated metrics and disconnected dashboards, the organization gains a coherent, business-aware view of network health — where operational signals are interpreted in context, not in silos.

• On the other hand, Foundry IQ builds on the semantic foundation established by Fabric IQ, consuming structured insights and transforming them into context-aware, evidence-backed recommendations. Rather than delivering generic guidance, it enables telco operators to receive precise, situation-specific corrective actions — grounded not only in operational telemetry, but also in enterprise knowledge such as policies, procedures, and historical incident documentation.

• The result is a seamless flow from understanding to execution: Fabric IQ delivers a business-aware view of the network by contextualizing operational and performance data within a unified semantic model, and Foundry IQ transforms that contextual understanding into actionable, evidence-based guidance that enables informed, timely decisions.

Together, they enable telco teams to move from reactive troubleshooting to proactive, intelligence-driven operations — closing the loop between data, knowledge, and execution.

The Bottom Line

Fabric IQ and Foundry IQ are gaining traction because they solve the most expensive, most persistent, and most critical problem in enterprise AI:

Organizations don’t need more dashboards or models — they need shared meaning. They need AI that understands their business and acts within it responsibly.

Fabric IQ builds the semantic backbone. Foundry IQ builds the reasoning engine. Together, they create the first enterprise intelligence layer capable of powering the next decade of AI‑driven transformation.