Databricks lays out seven criteria for choosing governance tools

The old problem of scattered data

Most enterprise data lives spread across data warehouses, data lakes, SaaS applications, and departmental spreadsheets, with no single system tracking what exists where or who owns it. The guide notes that data teams lose hours every week just hunting for assets, and sensitive data often goes unprotected because there's no consistent access control layer. Governance tools sit on top of all that, adding a searchable, shared layer so owners and stewards can see, classify, and protect assets no matter where the underlying data lives.

Six core capabilities

The guide argues a mature governance platform needs to cover six bases—and that skipping any one of them creates real operational gaps.

Capability	What it does
Cataloging & discovery	Organizes tables, files, dashboards, and models into a searchable index
Data lineage	Tracks how data moves from source through to reports and models
Access control & policy enforcement	Applies permissions centrally and consistently at the row, column, and attribute level
Quality monitoring	Continuously checks completeness, accuracy, and freshness, and flags anomalies
Compliance & auditing	Produces audit trails for regulations like GDPR and HIPAA
AI & agent governance	Extends the same access control and lineage tracking to models, prompts, and autonomous agents

Five categories of tools

Rather than sorting tools by brand, the guide groups the market into five categories based on scope and architecture. Standalone data catalogs connect broadly across many data sources but often rely on separate tools for access control or quality monitoring. Point solutions go deep on a single function—quality, lineage, or classification—but stitching several together tends to reintroduce the very fragmentation governance tools were supposed to fix, the guide warns. Enterprise suites bundle cataloging, quality management, master data management, and policy management into one product; modules built into ERP systems, like SAP Master Data Governance, typically manage a specific system rather than the full data estate. Platform-native governance builds cataloging, lineage, access control, and quality monitoring directly into the data platform itself, so it works on the same tables, files, and AI assets without needing to sync a separate system. The guide points to Databricks' own Unity Catalog as an example of this category within a lakehouse. Open-source governance tools let you inspect the internals and extend them yourself, trading lower licensing costs for heavier implementation and operational overhead.

Frameworks and tools aren't the same thing

The guide draws a clear line between a governance framework and a governance tool. A framework is the set of policies, roles, and standards that define how data should be classified, owned, accessed, and used. A tool is the software that actually enforces those rules across systems. Policy without tooling means someone has to manually check every table every day—a process that collapses once you're past a few dozen data assets. Tooling without policy just becomes an expensive catalog nobody follows.

Seven criteria for evaluation

Whatever tool you're weighing, the guide recommends scoring it first against seven criteria: scalability, integration, usability, granularity of policy enforcement, AI governance readiness, total cost of ownership, and vendor support. Scalability in particular means catalog freshness and query performance shouldn't degrade as data grows from gigabytes to petabytes—and that includes handling open table formats like Delta Lake and Apache Iceberg.

AI agents have entered the governance picture

It's no accident that the guide lists AI and agent governance as its sixth capability. As more AI agents read and act on enterprise data, the same access controls and lineage tracking long applied to tables now need to extend to model inputs and outputs, the guide explains. This capability is still maturing, and it's becoming the dividing line between tools that stop at structured data and tools built for the AI era.

Editor's view

What makes this guide interesting is that Databricks led with evaluation criteria rather than its own product name. It's not exactly hiding its marketing intent—Unity Catalog shows up as the sole example of the "platform-native" category—but the advice to build a feature-by-feature comparison table is genuinely useful regardless of who wrote it. A few years ago, a governance tool just needed to handle cataloging and access control. Now there's a new requirement to fold prompts and agent outputs into that same policy layer. Anthropic adding enterprise-managed authentication to Claude's MCP connectors, and GS E&C consolidating field data on Snowflake, both sit on the same trend line: as AI starts touching enterprise data directly, governance has to manage not just human access rights but model access rights too. For companies here, the practical move isn't rushing to adopt a whole new enterprise governance suite—it's checking whether the data platform you already use has access control and lineage tracking built for AI agents. The guide's warning about point solutions reviving fragmentation rings true. In the coming months, expect competing platforms like Snowflake and Microsoft to roll out similar AI and agent governance features on top of their own catalogs, following the same logic—because this capability has quietly become table stakes for governance tools.