Definition
A semantic layer encodes entities, relationships, metrics, permissions and provenance in a form every system above it can reason from. It is the operating system the enterprise has been missing.
Why it matters
Without it, every report, model and agent rebuilds the world from scratch — and disagrees. With it, meaning becomes a shared substrate. Every tool above becomes interchangeable; every answer becomes defensible.
Symptoms
- Every dashboard needs a footnote explaining its own definitions.
- New models require weeks of "data understanding" before any modelling.
- Agentic tools contradict each other because they read different versions of the same term.
Typical mistakes
- Treating the semantic layer as a BI feature rather than an enterprise substrate.
- Letting each domain own definitions without a single arbitration authority.
- Building the layer without provenance — meaning without traceability erodes trust.
Examples
An insurer defines "premium" seven times across policy, billing, actuarial, tax, reinsurance, reporting and CRM. The semantic layer defines it once, notes the seven contextual variants, and every downstream system references the same source.
How Zero-Gravity approaches it
We install a semantic layer with a single accountable owner, start from the decisions that most need it, and expose the definitions to humans, applications and agents through the same interface.