A semantic layer audit finds where meaning is owned, reused, trusted, and broken.
Start with the decisions, not the catalog. Pick the operating loops that matter: forecast, retention, margin, implementation, support severity, product usage, renewal risk. Then list the terms and metrics those loops depend on.
For each one, ask a few plain questions:
- What does this term mean in business language?
- Who owns the definition?
- Which source or calculation is authoritative?
- Where is it reused: dashboards, reviews, workflows, exports, agents?
- What are the known exceptions?
- When did it last change, and who was told?
- What breaks if two teams define it differently?
The audit should also look for duplicate definitions hiding in code. Search dashboard filters, dbt models, spreadsheet formulas, CRM fields, workflow rules, and recurring review decks. If the same business term is rebuilt in five places, the company does not have a semantic layer. It has scattered habits.
Do not turn the audit into a platform selection exercise. Tools can help later. First find the operating damage: review time wasted on definitions, metrics that cannot be reconciled, automation that uses the wrong object, teams that keep private versions of shared concepts.
The output should be small and actionable: the top contested terms, the owner gaps, the highest-risk duplicates, and the next definitions to put under custody.
A good audit makes the company slightly less mysterious to itself. That is the whole point.
This is part 10 of 10 in The Semantic Layer.
