The Librarian Who Read Every Book: Fabric IQ, the Business Capability That Makes OneLake Decision-Ready

A capability-based view of how Fabric IQ adds the semantic layer that lets people and agents ask business questions of the enterprise data estate.

A Tuesday morning meeting

It’s 9:14 a.m. The VP of Sales leans into her webcam and asks what sounds like a simple question:

“Who are our top 20 customers at risk of churning this quarter?”

Three analysts lean in, eager to help.

Maya in Finance opens the warehouse. “Revenue” there means recognized revenue. Raj in CRM pulls Salesforce. His “customer” includes trial accounts. Priya in Product opens the Real-Time dashboard. Her “churn” is 30-day inactivity. Finance’s is contract non-renewal.

By Thursday they’ll have three thoughtful, well-built answers, and three slightly different lists. None of them is wrong. Each one is true in its own dialect.

Here’s the exciting part: every byte they needed was already sitting in OneLake. Unified. Governed. One open Delta copy, instantly reachable by every engine in Fabric. The foundation is already extraordinary.

Now imagine giving that foundation a shared language.

The next frontier on top of OneLake

You already run on OneLake. One tenant, open Delta, mirrored sources, shortcuts everywhere. Storage is solved.

The next layer up is shared meaning. The questions your business actually asks, like ”who’s at risk?”, ”which products are accelerating?”, or ”where should we invest next quarter?”, don’t live in columns and data types. They live in concepts like customer, revenue, churn, region, and risk.

That’s the layer Fabric IQ adds, and it’s the layer that turns every workload, every Copilot, and every agent in your tenant into something that speaks your business’s language out of the box.

Enter Fabric IQ

Picture a librarian who has read every book in your library. She knows that the “Mrs. Chen” in the 1998 contracts is the “M. Chen” in the 2026 CRM. When you ask a question in plain English, she doesn’t hand you a stack of books. She hands you the answer, with citations.

That librarian is Fabric IQ.

Concretely, it’s a semantic and knowledge layer over OneLake that captures, once and for the whole enterprise, what your business means.

• It knows the things in your business (customer, product, account, opportunity, region).

• It knows what they mean, like the agreed definition of “active customer,” blessed by the business.

• It knows how they connect, which tables join to which across Lakehouse, Warehouse, KQL, and Power BI.

• It knows how to ask, translating natural-language questions into the right query on the right engine with the right filters and security.

• It knows what’s trustworthy, so Purview sensitivity, lineage, certification, and row-level security flow through every answer.

Define it once. Every Copilot, every Foundry agent, every custom app, every Power BI report draws from the same well of meaning.

This is the layer the community has been thoughtfully experimenting with for years, building semantic models, metric stores, glossaries, retrieval pipelines. Fabric IQ brings all of that craft into one first-class home, so the patterns you already love become reusable across the whole stack.

What changes on Tuesday morning

Rewind to 9:14 a.m., this time with Fabric IQ in the room.

The VP types into Copilot:

“Top 20 customers at risk of churning this quarter.”

Fabric IQ does in two seconds what used to take a small team of brilliant analysts two days:

I. Resolves ”customer” to the canonical entity, not trials, not duplicates.

II. Looks up ”at risk of churning” in the certified metric catalog, where the CRO-signed definition is declining usage plus an open escalation plus a renewal within 90 days.

III. Knows which OneLake tables hold usage (Real-Time Intelligence), support (Lakehouse), and renewals (Warehouse). Joins them automatically.

IV. Applies the VP’s row-level security. She sees her regions.

V. Returns the 20 names. With a lineage trail. With the metric definition inline. With a why this customer explanation per row.

Maya, Raj, and Priya now get to spend their Tuesday on the part that needs their judgment, what do we do about these accounts?, instead of on reconciling definitions. That’s where their talent was always meant to go.

Couldn’t we build this ourselves?

Absolutely, and many of us have, in pieces. Most data teams have already built a glossary here, a metric store there, a clever text-to-SQL prototype somewhere else. Those instincts are exactly right.

Fabric IQ is what happens when all of those instincts become one product, maintained for you, plugged into every Copilot and agent automatically:

| Layer | What Fabric IQ handles natively |

| Semantic model across all your tables | Defined once, reused everywhere |

| Entity resolution (cust_id = customerKey = acct) | Built-in, drift-aware |

| Certified metric definitions | One catalog, business-owned |

| Joins & relationships | Inferred and curated |

| NL → SQL/DAX/KQL | First-class, multi-engine |

| Grounding for every agent | Shared, not rebuilt per app |

| Governance flow-through | Purview, lineage, RLS inherited |

| Freshness & schema-change tracking | Continuous |

The win isn’t replacing your team’s craft. It’s freeing it. Your data engineers stop maintaining grounding plumbing and start designing the semantic shape of the business. Your analysts stop reconciling definitions and start asking sharper questions. Your developers stop rebuilding RAG for the fifth time and start shipping agents that actually know your domain.

From a unified lake to a thinking estate

OneLake made your data one. Fabric IQ gives it a shared mind.

The Tuesday after

Six months in, the 9:14 meeting feels different.

The VP asks. Copilot answers. Everyone is working from the same picture. Maya, Raj, and Priya are spending their morning on the conversation that actually matters, what do we do about these customers?, and the rest of the company is asking better questions because better questions are finally easy to ask.

The lake is still a lake. But somewhere inside it, a librarian is awake.

And she’s read every book.

Try it yourself: build your first Fabric IQ ontology

Microsoft has a hands-on tutorial that walks through creating your first ontology (currently in preview), enriching it with live streaming data, and exposing it to a Fabric data agent for natural-language Q&A. It uses a fictional ice-cream retailer called Lakeshore Retail with sales data plus freezer telemetry.

Start here: Ontology (preview) tutorial part 0, introduction and environment setup.

What you’ll need

I. A Fabric workspace on a Fabric-enabled capacity.

II. A Fabric admin to enable these tenant settings:

• Enable Ontology item (preview)

• User can create Graph (preview)

• Users can create and share Data agent item types (preview)

• Users can use Copilot and other features powered by Azure OpenAI

III. The sample CSVs from the fabric-samples GitHub repo (DimStore.csv, DimProducts.csv, FactSales.csv, Freezer.csv, FreezerTelemetry.csv).

The path from zero to “ask it a question”

i. Pick your starting point. Generate the ontology from an existing Power BI semantic model (faster if you already have one) or build it directly from OneLake (full control, no semantic model required).

ii. Stand up the lakehouse. Create a OntologyDataLH lakehouse, upload the four entity CSVs, and load each one to a Delta table.

iii. (Semantic-model path) Build the model. Create a Direct Lake semantic model called RetailSalesModel over dimproducts, dimstore, and factsales. Add the factsales → dimstore and factsales → dimproducts many-to-one relationships.

iv. Land the streaming data. Create an Eventhouse called TelemetryDataEH, then a KQL table called FreezerTelemetry from FreezerTelemetry.csv.

v. Create the ontology. Move on to tutorial part 1, create an ontology and generate entity types like Store, Product, and SaleEvent, then bind the live freezer telemetry from Eventhouse.

vi. Ask in natural language. Hook the ontology up to a Fabric data agent and ask things like ”What is the top product by revenue across all stores?”

Where to go next ?

• Fabric IQ ontology overview

• Required tenant settings

• Fabric data agents

The tutorial takes a couple of focused hours and is the fastest way to feel the difference between querying a lake and asking a librarian.