Microsoft Purview
Microsoft Purview combines data governance — a Data Map that scans and classifies assets and a Unified Catalog above it — with data security and compliance solutions, billed through Azure meters for capacity, scanning, and governed assets.
LicenseCommercial
DeploymentManagedServerless
PricingSubscription
Operational complexityLow
WorkloadInteractiveBatch
Use it when
- The estate is Microsoft-shaped: Fabric, Azure SQL, and Power BI share classifiers, sensitivity labels, and connectors with the compliance tooling you already run.
- Governance must span security and compliance as well as cataloguing: DLP, eDiscovery, and the catalog live on one platform.
- Glossary terms should carry policy: a term applied to a data product pushes its attached policies onto the assets underneath.
- Costs should scale with what is governed rather than with who looks: billing is per governed asset per day plus processing units.
Think twice when
- Data lives mostly outside Azure: around forty-five connectors, each rated per capability, and few support classification, lineage, policies, and live view together.
- You want a lightweight catalog; this is three product families sharing a portal.
- Uneven lineage coverage would undermine trust; the connector matrix is where the gaps are admitted.
How it runs
Managed and serverless in Azure. The Data Map scans and classifies sources; the Unified Catalog organises governance domains, data products, and glossary terms above it. Quality scans run on managed Spark with AI-assisted profiling and rule generation; self-service access requests cover a whole data product in one approval.
Details
CompareHow Microsoft Purview answers the questions Data Catalogs turns on.
Data Catalogs
| How it works | |
| Metadata model | Two layers with different jobs, the Data Map is the scanned technical inventory, and the Unified Catalog is the governance layer above it, organised into governance domains that hold data products, glossary terms, critical data elements and OKRs. Some Azure and Fabric sources appear through live view with no scan at all |
| Lineage | Collected rather than inferred: from scans where the source supports it, and from the pipelines that moved the data: Data Factory, Synapse pipelines, Fabric, Airflow, Power BI and Azure Machine Learning. Coverage is uneven by source and the connector matrix is where that is admitted |
| Search and discovery | Search and browse the Unified Catalog by governance domain, by data product or by keyword, with an AI copilot taking the question in plain language; the free version is browse-and-search only, with advanced search reserved for enterprise |
| Business glossary | Terms that carry policy rather than only meaning, Microsoft's framing is that they moved from static to active objects, so a term applied to a data product pushes its attached policies down onto the assets underneath |
| Quality checks | A full module: out-of-the-box rules across completeness, consistency, conformity, accuracy, freshness and uniqueness, AI-generated rules and AI-assisted profiling, scans scheduled hourly to monthly, and scores rolled up from column to asset to data product to governance domain, with alerts and an actions centre. It runs on Apache Spark 3.5 and Delta Lake 3.2.1, and caps at 200 rules per asset |
| Custom metadata | Custom classifications and custom scan rule sets including custom file types and parsers, managed attributes on assets, and REST APIs over the Data Map, though the free version restricts API asset creation to data curators and reserves the full API surface for enterprise |
| Connections | |
| Connectors | Around forty-five sources across Azure, databases, files, and services and apps, and the per-source matrix matters more than the count, since each is rated separately for classification, sensitivity labelling, policies, lineage and live view, and very few support all five |
| Access | |
| Policy and compliance | Access policies attach to data products, critical data elements and glossary terms, so one rule covers a business concept rather than a table; underneath, the Data Map applies classifications and sensitivity labels per source, and data owner and protection policies where the source supports them. The security and compliance halves of Purview share those classifiers |
| Access requests | Self-service access requests from inside the Unified Catalog, one request covers a whole data product rather than each table in it, and the access policies attached to the product, its critical data elements and its glossary terms decide the answer |
| Cost | |
| Billing unit | The clearest meters in this capability, unique governed assets per day, where a governed asset is a table or view linked to a governance concept and counts once however many concepts attach to it, plus Data Governance Processing Units for data quality and health runs, each 60 minutes of compute in Basic, Standard or Advanced. Data Map capacity units and vCore-hour scanning are the classic model and stop applying once you move to pay-as-you-go |
Alternatives to Microsoft Purview
