Cannot reconcile to a reported figure
Resolve this first. Everything else is optimisation of something that is not yet trusted.
Power BI · Health check
Semantic model health check
Twenty checks on whether your Power BI model is trustworthy, performant and ready for Copilot to reason over.
Why run this
What this checklist is for
A semantic model is now an interface for people and increasingly for AI. Ambiguity that a human analyst worked around silently produces confidently wrong answers when an assistant reasons over it.
These checks cover correctness, performance and AI-readiness together, because they turn out to be the same discipline.
Run it with
Finance or operations director
And with
Power BI developers and analysts
And with
A practice leader who uses the reports
0 of 0 complete · 0%
The checklist
20 checks, in the order we would run them
Tick only what you can genuinely evidence today. An item you intend to do is not an item you have done, and scoring yourself generously here only produces a comfortable number and an uncomfortable project.
Section 1
Correctness
Whether the numbers are right.
Section 2
Performance
Whether it is usable.
Section 3
Governance
Whether it stays correct.
Section 4
AI readiness
Whether an assistant reasoning over this would be trustworthy.
What your score means
Read this against the number above
These bands are deliberately blunt. The middle band is where most organizations honestly sit, and it is a perfectly reasonable place to proceed from — provided the gaps are written down with owners rather than carried as optimism.
0–59%Significant gaps
Do not proceed yet. More than four in ten items are unaddressed, and the ones that fail here are usually the foundational ones that make everything after them harder.
60–84%Mostly ready, with known gaps
Proceed on a defined scope, with the outstanding items written into the plan as risks with owners and dates. This is the most common honest position.
85–100%Ready
The remaining gaps are small enough to handle during delivery rather than before it. Confirm the unticked items are genuinely minor rather than simply unexamined.
Your score highlights automatically as you tick items above. Nothing is saved, sent or tracked — refreshing the page clears it.
Closing the gaps
If you could not tick these, start here
The four items below are the ones whose absence causes the most trouble downstream. If your unticked items include any of these, they are worth addressing before the rest.
Bidirectional relationships
Review each one. They cause ambiguity, poor performance and — where row-level security is in play — security that does not apply as intended.
Cryptic naming
Rename in business terms. This was cosmetic when only analysts read the model; it is functional now that Copilot does.
Never tested with natural-language questions
Spend twenty minutes asking it partner-level questions. It is the most revealing check on this list and almost nobody runs it.
Want a second opinion on your score?
We will run the health check against your model and test it with the questions your partners actually ask, then give you a prioritised list of what to fix.
Talk through your result Read the related guides We reply to every message within one business day.
Keep going
Related checklists
CRM adoption health check
Twenty questions that establish whether your CRM is a working system or an expensive reporting obligation.
June 13, 2026 · 20 checksOpen
Cloud operating model readiness
Twenty checks on whether your team can actually run the cloud estate you are about to build.
December 6, 2025 · 20 checksOpen