Free self-check
The Power BI Governance Check
Twenty questions about your Power BI and Fabric estate, answerable in an afternoon. Built on Microsoft’s Fabric adoption roadmap and the points we audit on a platform review. Answer for the workspaces people actually use, not the ones on the governance diagram; a report nobody can find is not governed, whatever the policy says.
- The estate: workspace inventory, lineage you can trace, semantic models reused rather than copied. One certified dataset that three reports point to behaves very differently from three near-duplicates drifting apart after the next refresh.
- Controls: tenant settings, sensitivity labels, endorsement, DLP, and row-level security someone else has tested, not just switched on and left.
- People and process: who owns standards, and the path a new report builder takes from a blank workspace to a certified one.
- Platform economics: capacity consumption, refresh failures, and the licensing mix, the three places a Fabric estate quietly overspends.
Common questions
We use Power BI but not Fabric yet. Does the check still apply?
Yes. The tenant settings, endorsement, DLP, and row-level security questions apply to a Power BI Premium or Pro estate on their own; the Fabric-specific items just don't count against you if there's no Fabric capacity yet.
Who should answer it, the BI team or the business owners?
Whoever owns the workspace people actually open, not the one on the governance diagram. A workspace that looks compliant on paper but that three finance analysts have stopped trusting behaves like an ungoverned one.
What if we score badly on platform economics?
Capacity consumption and refresh failures are usually the fastest wins, cheaper to fix than a full model rebuild. The Power BI and Fabric Platform Review prices out exactly which of the three areas, estate, controls, or economics, is costing the most.
Does a low endorsement score mean our reports are wrong?
Endorsement measures trust and discoverability. A correct report with no certification still reads as unreliable next to three uncertified copies nobody can tell apart.