How can AI learn business definitions from the dashboards and queries we already have?

Summary

• The most direct path is to let AI read the work you've already done. Genie Ontology automatically extracts business knowledge from the dashboards, SQL queries, notebooks, and pipelines already in your workspace, then organizes it into a shared context layer AI can use to answer questions.
• No upfront modeling is required. It produces answers from what it can infer on day one, so a data team doesn't have to hand-build a semantic model before the system is useful.
• Every learned definition carries an authority score. Databricks weighs where a definition came from, how often people rely on it, how closely it ties to certified assets, and how fresh it is, so conflicting definitions resolve toward the most trusted one.
• Answers stay permission-aware. Each learned snippet is gated by Unity Catalog permissions, so AI only uses definitions from assets a given user is already allowed to see.
• Genie Ontology is in Public Preview. Treat the behavior described here as an emerging capability that can change, and confirm current details in the documentation.

How can AI learn business definitions from the dashboards and queries we already have?

The most direct way is to let AI read the work your teams have already done. Genie Ontology, in Public Preview, automatically extracts business knowledge from the dashboards, SQL queries, notebooks, and pipelines already in your Databricks workspace and turns it into a shared context layer for answering questions. Instead of an analyst hand-writing every definition, the system infers how your business actually defines its terms from how people already query and report on data. You get useful answers early, and you raise accuracy over time by governing the few definitions that have to be exact.

What does it mean for AI to learn a business definition?

A business definition is the real meaning behind a term: what counts as an 'active user', how 'revenue' is calculated, when a 'lead' becomes 'qualified'. Those rules are usually implicit, buried in the SQL behind a dashboard or the logic of a saved query. To answer a question correctly, AI has to recover that meaning instead of guessing at it. Learning a definition means pulling it from the assets where it already lives, attaching it to the tables and metrics it describes, and knowing which version to trust when two teams disagree.

What Genie Ontology learns from your existing assets

Genie Ontology reads the assets you already maintain and extracts small, reusable pieces of context. The public documentation gives examples of the kinds of knowledge it infers.

Type of inferred knowledge	Example it can learn
Metric definition	An 'active user' is a distinct user, deduplicated across platforms.
Business rule	A 'lead' counts as qualified only once a demo is booked.
Authoritative source	Answer revenue questions from the certified Finance Genie Agent.
Relationships and trust signals	Which tables join on which keys, and which assets are certified or deprecated.

The extraction runs continuously, so as your dashboards and queries change, the learned context keeps pace with them.

How Databricks approaches this

Genie Ontology combines two kinds of context. The first is inferred context: snippets it extracts automatically from tables, queries, dashboards, notebooks, and Genie Agents. The second is modeled context you define and govern in Unity Catalog: metric views for measures, Pages for business terms, and domains to scope them. The published guidance is to model the 'head' and let Genie infer the 'tail': you deliberately curate the definitions that cannot be wrong, and rely on inference for the long tail of everything else.
Each learned snippet gets an authority score. Databricks describes ranking definitions by where they came from, how often people rely on them, how closely they tie to certified and widely used assets, and how fresh they are. When two definitions conflict, the higher-authority one wins, and a definition you assert by hand takes priority over an inferred one. Every snippet is gated by Unity Catalog permissions, so AI draws only on assets a given user can already see, and each answer cites the sources behind it. Because Genie Ontology is in Public Preview, treat these behaviors as an emerging capability and confirm the current details in the documentation.

How to get better answers over time

You don't have to model anything to start, but a few habits sharpen what AI learns:

• Certify trusted assets such as metric views, notebooks, and Genie Agents to raise their authority.
• Deprecate stale dashboards and saved queries so AI steers away from them.
• Capture the definitions that must be exact as governed metric views and Pages.
• Keep documentation on existing assets rich, since a well-described asset yields richer context.

For a staged approach, see Operationalizing Genie Ontology in your data stack.

FAQs

Do we have to build a semantic model before AI can answer?

No. Genie Ontology infers context from your existing assets on day one. Modeling metric views, Pages, and domains is a way to raise accuracy over time, not a prerequisite for getting answers.

Where does it get the definitions?

From assets you already have: dashboards, SQL queries, notebooks, pipelines, and Genie Agents. It extracts metric definitions, business rules, and authoritative sources, then organizes them into a context graph.

What happens when two dashboards define the same metric differently?

Genie Ontology ranks the competing definitions by authority and answers with the most trusted one. Definitions you curate by hand override inferred ones, which is why governing your critical metrics is worth the effort.

Does it respect data permissions?

Yes. Each snippet is gated by Unity Catalog permissions, so AI only uses knowledge from assets a user is allowed to see, and it cites its sources.

Is Genie Ontology generally available?

No. It is in Public Preview. Availability and behavior can change, so validate current capabilities in the Databricks documentation before relying on them.