Using Model Health
This guide walks you through analyzing your model with the Model Health dashboard, understanding the scores, and working through findings to improve your model.
If you're preparing your models for AI consumption - whether through Copilot, MCP servers, or natural-language query tools - Model Health is the place to start. The AI Readiness score tells you how well-structured your model is for AI tools, and the findings show you exactly what to fix. For more on why this matters, see the Model Health feature overview.
Step 1: Open Model Health
Click the Model Health tab in the main workspace area.
Step 2: Choose a Source
Use the Source dropdown to select what you want to analyze:
- Working Model: Your local, in-progress model.
- Environment servers (e.g., Develop, UAT, Production): A model published to one of your configured environments. Useful for comparing your working model against what's currently deployed.
Step 3: Run an Analysis
Click one of the two analysis buttons:
- Run Analysis: A rules-based scan that checks for common structural and formatting issues. Runs locally with no AI key required.
- Run AI Analysis: Everything in Run Analysis, plus an AI-powered review that catches issues rules can't - like suspicious descriptions, inconsistent data types across related tables, or commented-out DAX. Requires an AI key configured in Settings > AI or through BYOK.
Step 4: Read the Score Cards
After the analysis completes, six score cards appear across the top of the dashboard:
AI Readiness
A weighted score (0–100) measuring how prepared your model is for AI-powered tools like Copilot. Click the (i) icon to see the exact breakdown:
| Factor | What It Measures |
|---|---|
| Description coverage | Percentage of fields with descriptions |
| No many-to-many relationships | Avoiding relationship types that confuse AI |
| No bidirectional relationships | Simpler filter direction for AI to reason about |
| No bridge tables | Cleaner star schema structure |
| Synonym coverage | Alternative names for columns and measures |
| Date table present | Whether a proper date table is defined |
Each factor is scored 0–100% and multiplied by its weight. Weights are configurable in Settings > Modeling.
Model Quality
A weighted score (0–100) measuring overall modeling best practices. Click the (i) icon for the breakdown:
| Factor | What It Measures |
|---|---|
| Measure descriptions | Percentage of measures with descriptions |
| Naming clarity | Consistent, clear naming conventions |
| Structure | Star schema adherence, table organization |
| Relationships | Proper cardinality and filter direction |
| Date table | Proper date table configuration |
| Format strings | Measures and columns with explicit formatting |
| Display folders | Measures organized into logical folders |
Weights are configurable in Settings > Modeling.
Documentation
Shows how many fields in your model have descriptions (e.g., "105 / 105 fields documented (100%)"). Click Generate Descriptions to use AI to fill in missing descriptions - see Model Documentation (AI) for details.
Relationships
A summary of your model's relationships: how many are active, how many are inactive, and whether any are bidirectional.
Structure
Flags structural issues at a glance: orphaned tables, missing date tables, bridge tables, and measures placed on dimension tables.
Ontologies
Shows whether your model is part of an ontology. Ontologies improve AI discoverability across models - see Ontologies.
Step 5: Review the Tables Tab
The Tables tab gives you a table-by-table breakdown:
| Column | What It Shows |
|---|---|
| Mode | Storage mode (Import, DirectQuery, etc.) |
| Rows | Row count (when available) |
| Columns | Number of columns |
| Calc Cols | Number of calculated columns |
| Measures | Number of measures |
| Hierarchies | Number of hierarchies |
| Missing Desc | Count of fields without descriptions |
| Table AI Score | How AI-ready this table is |
| Table Quality | Overall quality score for this table |
Use the Show hidden tables toggle to include tables that are hidden from report authors.
Step 6: Work Through Findings
The Findings tab is where you'll spend most of your time. Findings are grouped by severity:
- Error: Issues that will cause problems (e.g., no date table defined, orphaned tables)
- Warning: Issues that should be addressed (e.g., measures without format strings, bidirectional relationships, default summarization on numeric columns)
- Suggestion: Improvements to consider (e.g., unnecessary columns imported, missing synonyms, unreferenced hidden measures)
Each finding has a code (e.g., ERR001, FMT002, PERF003) and a count showing how many objects are affected. AI findings are prefixed with AI- and come from the Run AI Analysis.
Filtering findings
Use the Severity and Category dropdowns to narrow the list, or type in the search box to find specific findings.
Acting on a finding
- Click the chevron (>) on a finding to expand it and see which specific objects are affected and a detailed explanation.
- Click Go to to navigate directly to the affected object in your model.
- Click Resolve to mark an AI finding as resolved and remove it from the list.
Step 7: Check Relationships
The Relationships tab shows all relationships in your model, making it easy to audit cardinality, filter direction, and whether each relationship is active or inactive.
Exporting Results
Click the Export button in the top-right to export the dashboard results.
Tips
- Start with Errors first - these are the most impactful issues.
- Run the standard analysis first to fix structural issues, then run the AI analysis for deeper insights.
- Re-run the analysis after making changes to see your scores improve.
- Customize the score weights in Settings > Modeling to match your team's priorities.