Microsoft Purview Explained: Data Governance, Security and Compliance Architecture
Author : Vaden Consultancy | Published On : 19 Nov 2026
Data rarely stays in one place. A sales dataset might start in a lakehouse, feed a semantic model, appear in Power BI reports and end up as an Excel export on a laptop. Governing that journey takes more than good intentions. Microsoft Purview is Microsoft's family of tools for governing, protecting and managing data. Knowing how its parts fit together helps teams avoid costly misconfiguration. This guide explains how Purview works, how it relates to Microsoft Fabric, Power BI and the wider Power Platform, and what to plan before implementation.
What Microsoft Purview Covers
Purview groups its capabilities into three areas:
- Data governance: Data Map and Unified Catalog help people find and trust data.
- Data security: Information Protection, Data Loss Prevention (DLP) and Insider Risk Management protect sensitive data.
- Risk and compliance: Audit, eDiscovery and Data Lifecycle Management support legal and regulatory duties.
All of these are managed in the Microsoft Purview portal, which replaced the older compliance and classic governance portals. Menus have moved, so older tutorials can mislead. Data Estate Insights, for example, is now Reports under Unified Catalog. If a solution card is missing, suspect a role or licence gap before assuming the feature does not exist.
How Data Governance Works
Data Map scans data sources and records their metadata: names, schemas, classifications and lineage. Sources include Azure SQL Database, Azure Data Lake Storage, many non-Microsoft systems and Fabric itself. Scanning a Fabric tenant surfaces workspaces, lake-houses, warehouses, semantic models and reports.
A key design point: Data Map and Unified Catalog hold metadata only. Their roles never grant access to the underlying tables.
Unified Catalog is the business-facing layer on top. Stewards organise scanned assets into governance domains (such as Sales or Finance), bundle related assets into data products, and attach glossary terms and critical data elements so that differently named columns share one business meaning. It also provides data quality rules and scores, health management reports, and self-service access requests. A free version exists for evaluation and testing, but scanning, collections and automatic classification require the enterprise version.
Data Security: Labels, Protection Policies, DLP and Insider Risk
Sensitivity labels (General, Confidential, Highly Confidential) are created once in Purview and reused across services. In Fabric and Power BI, manual labelling works on all items. Default labelling and downstream inheritance cover all Fabric items with some limitations, so a label on a lakehouse can flow to dependent semantic models and reports. Mandatory labelling is fully supported for Power BI items only.
A label alone does not block anyone inside Fabric. Access control comes from a protection policy tied to a single label: named users and groups keep access to labelled items, and everyone else is blocked. Today these policies cover native Fabric items and Power BI semantic models, but not reports or dashboards.
DLP scans Fabric data for sensitive information and can show a policy tip, raise an alert, or restrict access (the restriction action was still in preview at the time of the source guide). Technical limits matter here:
- The workspace must be on a Fabric (F) or Premium (P) capacity.
- Only Delta tables in Fabric items are evaluated.
- DirectQuery and live connection semantic models are not evaluated.
Insider Risk Management looks for behaviour patterns rather than sensitive content, such as bulk report downloads followed by label removal, or lakehouse data shared outside the organisation. Its Fabric indicators require pay-as-you-go billing.
Risk and Compliance
All Fabric user activity, including viewing, exporting, sharing and changing settings, is recorded in the Purview audit log, which is where you answer "who exported that report?" For eDiscovery and Data Lifecycle Management, Fabric coverage centres on Copilot and AI interactions (prompts and responses). It is not a mechanism for purging old warehouse rows, which remains a data engineering task.
Purview Alongside Fabric, Power BI and the Power Platform
Microsoft's current position is that the OneLake catalog is the primary place to discover and govern Fabric data, using domains, endorsement, lineage and governance insights in its Govern section. Purview complements it with labels, protection policies, DLP, Insider Risk, audit, and cataloguing across Fabric and non-Fabric systems. The old Purview hub inside Fabric has been retired.
For Power BI development, labelling strategy is therefore a design decision, not an afterthought. Exports to Excel, PDF, PowerPoint, Analyze in Excel and .pbix keep the label and any encryption, whereas .csv and .txt exports do not.
The wider Power Platform needs the same layered thinking. Activity from Power Automate flows can be investigated through Purview's audit capabilities (confirm current coverage in Microsoft's documentation). Connector control, however, sits in the Power Platform admin center's data policies, which are separate from Purview DLP.
A Practical Implementation Approach
- Inventory and scope. Identify workspaces, capacities and sources. Confirm that Fabric workloads sit on F or P capacity.
- Define the label taxonomy. Keep it small and meaningful. Decide where inheritance and default labels apply.
- Protect. Link protection policies to labels, and remember that reports are not yet covered, so workspace and app permissions still apply.
- Detect. Scope DLP policies to specific workspaces and sensitive data types, such as email addresses in a customers table.
- Catalogue. Scan the tenant with Data Map, then curate domains, data products and glossary terms in Unified Catalog.
- Operate. Review audit logs, alerts and health reports on a schedule, and train owners and stewards.
Licensing Considerations
Purview uses two billing models, and most organisations use both. Per-user licences (Microsoft 365 E3 or E5) cover Microsoft 365 content, and sensitivity labels need an Information Protection licence. Pay-as-you-go billing through a linked Azure subscription covers data outside Microsoft 365, including Fabric. DLP and protection policies are charged per asset in scope per day, and Unified Catalog is charged per governed asset per day plus processing units for quality and health management.
Asset counting surprises teams. With 10 workspaces of 50 assets each and a DLP policy scoped to 5 of them, 250 assets are counted, whether or not they match the policy. Check Microsoft's current pricing page before budgeting.
Common Pitfalls
- Assuming a label blocks access without a protection policy.
- Expecting DLP to cover every semantic model.
- Scoping DLP to the whole tenant "to be safe" and paying for every asset.
- Treating Unified Catalog roles as data access.
- Following guidance written for retired portals or the old Fabric hub.
Conclusion
Purview is best understood as layers: metadata governance, classification and protection, detection, and audit. Each has its own technical limits and billing behaviour. Start with a clear label model, scope policies deliberately, and align Fabric, Power BI and Power Platform controls so that governance holds from source data to final report or app.
Frequently Asked Questions
What is Microsoft Purview used for?
It helps organisations govern, protect and manage their data. Data teams use it to catalogue and curate data, security teams use it to label and protect sensitive information, and compliance teams use it for auditing, legal searches and retention.
Do I need Purview if I already use Microsoft Fabric?
Not for day-to-day governance of Fabric data, because the OneLake catalog handles discovery, domains, endorsement and lineage. You need Purview for sensitivity labels, DLP, Insider Risk Management, the audit log, and cataloguing data across Fabric and other systems.
Does Purview work with Power BI?
Yes. Power BI items support sensitivity labels, semantic models are covered by DLP and protection policies, and Power BI activity is recorded in the audit log. Data Map scans also pick up semantic models and reports.
Is a sensitivity label enough to protect my data?
No. Inside Fabric, a label alone does not block access. You also need a protection policy linked to that label, or encryption that travels with exported files.
Is there a free version of Microsoft Purview?
For data governance, yes. The free version offers a basic catalog and is meant for evaluation and testing. Scanning, collections, automatic classification and the full Unified Catalog require the enterprise version.
