How can AI answer questions using live company data?

Summary

• AI answers from live data when it queries the governed source directly instead of a stale export. On Databricks, AI/BI Genie generates read-only SQL against the current tables in your lakehouse.
• Unity Catalog governs every read, so each person's answer reflects only the live data they are permitted to see.
• Genie Ontology (Public Preview) is the context layer that helps Genie interpret each question using your metrics, glossary, and definitions, so terms resolve the way your business means them.
• Context is inferred automatically from existing metric views, dashboards, SQL queries, and Genie Agents, so heavy upfront modeling isn't required to start.
• Answers include citations, so users can see which sources produced them and confirm the result.

How can AI answer questions using live company data?

AI answers questions from live company data when it connects a natural-language agent directly to your governed data and queries it in place, instead of working from a stale export or a hard-coded snapshot. On Databricks, AI/BI Genie does this by generating SQL that runs read-only against the current tables in your lakehouse, with Unity Catalog governing every read. Genie Ontology, currently in Public Preview, is the context layer that helps Genie interpret each question the way your business actually means it. The payoff is an answer grounded in the data as it exists right now, with citations you can check.

What answering from live data actually requires

"Live" means two things: the data is current, and the meaning is correct. Reading fresh rows is not enough if the agent misreads what "active customer" or "revenue" is supposed to mean. A system that answers from live company data needs a direct connection to the governed source of truth, a shared layer of business context so terms resolve consistently, permission checks so each person sees only what they should, and transparent sources so an answer can be verified.

Key capabilities to evaluate

Capability	Why it matters
Queries the live source, not a copy	Answers reflect current data, so decisions don't rest on a stale snapshot.
Shared business-context layer	Metrics and terms resolve the same way for everyone, so "active user" means one thing.
Permission-aware answers	Access follows each user's identity, so live data is never over-shared.
Works across structured data and documents	Real questions mix tables with policies, notes, and reports.
Cited, checkable responses	Users can see which sources produced an answer and confirm it.
Little upfront modeling to start	Context can be inferred from assets you already have, so value doesn't wait on a long project.

How Databricks approaches this with Genie and Genie Ontology

Genie connects to data registered in Unity Catalog, including managed and external tables, views, materialized views, and metric views. When someone asks a question, Genie translates it to SQL and runs it read-only against those current tables, so the answer reflects live data rather than a cached result. Because the query is governed SQL, Unity Catalog enforces each user's permissions on every read.
Interpreting the question correctly is the harder half, and that is the job of Genie Ontology (Public Preview), described in the docs as "the unified context layer that gives the Genie family of AI experiences a business-aware map of your organization." It combines two kinds of context:

• Unity Catalog semantics you define and certify: metric views for governed metrics, a business glossary and Pages for shared definitions, and domains that group related assets.
• Inferred context that Genie extracts automatically from "metric views, dashboards, SQL queries, and Genie Agents," organized into a living map of how your company uses its data.

Genie ranks these context snippets by authority, weighing where a snippet came from, how often it is used, how closely it ties to certified assets, and how fresh it is, then feeds the most relevant ones into the question. Snippets are "gated by Unity Catalog permissions, so Genie only uses snippets extracted from assets you have permission to see," and you can click citation icons in a response to see which sources answered. See Introducing Genie One, Genie Ontology, and Genie Agents.

Use cases

• Everyday operating questions. Sales, finance, and operations teams can ask about current pipeline, spend, or inventory and get an answer without filing a request for a new report.
• Executive self-service. Leaders can ask follow-up questions against live numbers and open the cited sources, rather than waiting for an analyst to refresh a dashboard.
• Consistent metrics across teams. Because definitions live in metric views and the glossary, the same question returns the same defined metric for everyone entitled to it.

FAQs

Is Genie Ontology generally available?

Genie Ontology is in Public Preview. Its behavior and availability can change, so validate it against your own governed data before relying on it for critical reporting.

Does answering from live data mean Genie changes my data?

No. Genie generates read-only SQL to answer questions. It reads current data through Unity Catalog and does not modify your tables.

How does Genie keep answers current?

Genie queries the live tables in Unity Catalog at question time rather than a stored snapshot, so answers reflect the data as it stands when you ask.

How does Genie know what a business term means?

Genie Ontology supplies the meaning. It draws on governed definitions you certify (metric views, glossary, domains) and context it infers from your existing dashboards, queries, and agents, then answers using the sources you are permitted to see.