Skip to main content

AI readiness

When your data is self-describing and semantically rich, it becomes far easier for both humans and AI systems to interpret and apply it effectively. This page provides a comprehensive checklist for making your shared data AI-ready, with a focus on Databricks Genie and other AI-powered tools.

Why AI readiness matters​

Good metadata and semantic clarity enable:

  • Natural language querying: Users can ask questions in plain English and get accurate SQL
  • Reduced ambiguity: AI tools understand exactly what each column and table represents
  • Faster onboarding: New consumers can explore and understand your data without extensive documentation
  • Higher trust: Verified answers and clear definitions build confidence in query results

AI readiness checklist​

Core metadata​

  • Table descriptions — Every table has a description explaining its purpose and primary use case
  • Column descriptions — Every column has a description with meaning, units, and format where applicable
  • Clear naming — Column names are unambiguous (avoid generic names like value, data, type)
  • Proper data types — Dates and numbers use proper types (not strings)
  • Tags — Tags applied for discoverability (domain, sensitivity, refresh cadence)

Semantic clarity​

  • Primary keys — Primary keys explicitly defined for each table
  • Relationships — Foreign key relationships documented (join paths between tables)
  • Table grain — Grain explicitly stated (one row = one customer, one transaction, etc.)
  • Aggregation rules — Document how metrics should be aggregated (SUM, AVG, COUNT DISTINCT)
  • Time zones — Time zone handling documented for all timestamp columns

Unity Catalog metrics​

Unity Catalog metric views allow you to define reusable business calculations that can be referenced consistently across queries and tools.

  • Key metrics defined — Business metrics (revenue, churn, LTV) defined in Unity Catalog
  • Calculation logic — Formulas, filters, and exclusions documented for each metric
  • Consistent usage — Metrics used instead of ad-hoc calculations in downstream queries

UC metrics are not Delta-sharable—distribute definitions via Volumes or Asset Bundles.

Genie configuration​

Genie Spaces allow you to configure how Genie interacts with your data.

  • Genie Space created — Space configured with relevant tables for your data product
  • Sample questions — Common questions and verified SQL answers provided
  • General instructions — Domain-specific logic documented (e.g., "always filter by is_active = true")
  • Trusted assets — Priority tables and views marked as trusted
  • Terminology — Domain-specific terms mapped to table/column names

Genie configurations are not Delta-sharable—distribute configs and sample questions via Volumes or Asset Bundles.

Data quality signals​

  • Freshness indicator — last_updated or similar timestamp available
  • Completeness — Expected completeness documented (e.g., "99% of rows have email")
  • Known limitations — Data gaps or known issues documented
  • Expectations — Data quality expectations defined (DLT expectations, constraints)

Enablement​

  • Sample notebooks — Notebooks demonstrating common queries and use cases
  • Verified queries — Known-good SQL for common business questions
  • Join patterns — Recommended join paths documented
  • Edge cases — Tricky scenarios and their solutions documented

Business context​

Beyond technical metadata, AI tools benefit from business context:

  • Glossary alignment: Define what terms mean in your domain (e.g., "customer" = active account, not trial)
  • Calculation definitions: Document business logic (e.g., churn = no activity in 90 days)
  • Time period conventions: Clarify fiscal year vs calendar year, reporting periods
  • Regional considerations: Note when data is segmented by geography or business unit

Quick start​

If you're just getting started, prioritize these five items:

  1. Table descriptions - What is this table and what is it for?
  2. Column descriptions - What does each column mean?
  3. Primary keys - How do I uniquely identify a row?
  4. Table grain - What does one row represent?
  5. Sample queries - Show me how to use this data

These five items will dramatically improve AI tool performance and consumer onboarding.

What's next​