Get Microsoft Purview set up

Microsoft Purview is Microsoft’s suite of solutions for data security, data governance and compliance. It has two components for finding and managing data: Data Map, which uses scans and classification to retrieve metadata from your sources, and Unified Catalog, where people can search for data and request access.

Sensitivity labels, DLP and auditing are also part of Purview. But it is a tool: agreements on ownership and definitions belong in data governance, while the underlying platform is part of our data foundation.

Microsoft lists Azure SQL, SQL Server, Oracle, Snowflake, SAP ECC, SAP S/4HANA, Salesforce, Power BI and Fabric, among others. For systems on your own network, you install a self-hosted integration runtime, sometimes with additional drivers.

For Azure Databricks, Purview reads from Unity Catalog: catalogues, schemas, tables, views and columns, plus lineage from notebook runs. Note that if you delete a table from the source, it will not automatically disappear from Purview during the next scan.

A classification identifies what a column contains, such as a name or a credit card number. Purview has more than 200 built-in classifications, and you can create your own using a pattern or a list of terms. A sensitivity label indicates how sensitive something is, such as Public or Confidential, and can enforce encryption or a watermark.

Labels appear in the metadata and stay with the content. In Fabric and Power BI, you can apply them to reports, semantic models and lakehouses; anyone applying labels in Power BI needs a Pro or PPU licence.

Yes. In Word, Excel, PowerPoint and Outlook, a label can be suggested or applied while someone is working; for SharePoint, OneDrive and Exchange, a policy runs in the background. A sensitive information type can identify a Dutch citizen service number, for example, from its pattern, check digit and nearby keywords.

Run such a policy in simulation mode first. That lets you see which files would receive a label before it is applied, so you can adjust rules that are too broad or too narrow.

Fabric has its own governance: the OneLake catalogue, with domains, endorsement and lineage. Microsoft describes it as the first place to manage Fabric data. Purview adds sensitivity labels for all Fabric items, DLP policies for lakehouses, warehouses and semantic models, and an audit log of user activity.

To read how we set up a Fabric environment with workspaces, permissions and labels, see Microsoft Fabric implementation.

Purview shows where personal data is stored, such as Dutch citizen service numbers in tables or exports. This helps you determine what data you process and where. Your data protection officer or legal adviser should assess what the GDPR requires of you.

Microsoft 365 Copilot and agents recognise sensitivity labels and take labelled data into account during use. It is therefore best to set up labels before rolling Copilot out widely. For agreements on AI use, see AI governance.

We are a Microsoft Partner for Fabric and a Databricks Partner, and we build data platforms on Azure, including for Vivara and the Koninklijk Concertgebouworkest. Purview fits with those platforms because it can scan Unity Catalog and Fabric. We do not yet have a separate Purview case study among our case studies.

If we set up the platform, we can also manage it afterwards through management and continuous improvement.

If your data sits in a single Fabric environment or a single Databricks workspace, the OneLake catalogue or Unity Catalog will often already cover discoverability and permissions. And if no one is willing to take ownership of a domain, a scan will only populate a list that nobody maintains.

Start with the agreements instead: who manages which data, and what does each term mean? If the data itself is unreliable, first read about data quality.