The production manager’s dashboard says “OEE 78%”, the quality team’s report says “71%” and the finance presentation says “82%”. All three describe the same line in the same week. This is one of the quickest ways analytics projects lose trust, and it is usually caused not by the data itself but by scattered metric definitions.
What is a semantic layer?
A semantic layer sits between raw data and analytics tools and defines business concepts once, centrally. Metrics (OEE, unplanned downtime, unit cost), dimensions (line, shift, product, site) and the relationships between them are defined here; dashboards, reports and applications use those definitions.
The difference: instead of a formula buried inside every dashboard, the metric logic becomes a reusable asset managed in one place.
Why it matters especially in industry
- Metrics are complex: A metric such as OEE involves decisions on how planned/unplanned stops are classified, what the ideal cycle time is based on and which counter is used. (How is OEE calculated?)
- There are many sources: The same metric combines MES, SCADA, historian and ERP data.
- Comparison across sites is wanted: If each site calculates the metric differently, comparison is meaningless.
- AI relies on definitions: An AI assistant answering natural-language questions has to know what “OEE” means; without a definition it guesses.
A layered data architecture
| Layer | Role | Example |
|---|---|---|
| Source | PLC, SCADA, MES, ERP | Raw tags, work orders |
| Modeled data | Asset models, context, quality | pump-01.temperature (°C, good quality) |
| Semantic layer | Metric and dimension definitions | OEE = Availability × Performance × Quality |
| Consumption | Dashboards, reports, embedded analytics, AI | Line-level OEE dashboard |
A semantic layer is not effective without data modeling and contextualization: definitions are built on reliable, consistently named data.
How to build one
- Start with the 5–10 most contested metrics. Which metric shows you different numbers?
- Write the definition and name an owner. The calculation rule, scope (which stops count?), data source and responsible person.
- Standardize dimensions. Dimensions such as line, shift, product and site must match with the same keys across all sources.
- Define centrally. Let formulas live in the semantic layer, not in dashboards.
- Version and monitor. Keep definition changes on record so that “why is OEE different from last month?” can be answered.
Semantic layer with GoodData
GoodData Analytics places the semantic layer and metric definition at the core of the platform: metrics defined in one place carry through identically to interactive dashboards, embedded analytics and multi-tenant applications. Find the product features on our GoodData page and our industrial analytics and AI approach on the solution page.
Common mistakes
- Writing the definition only in a document and not reflecting it in the application.
- Each dashboard team keeping its own “temporary” formula that becomes permanent.
- Trying to fix data quality problems with a semantic layer. However good the definition, bad data gives bad results; data quality is a separate job.
- Trying to define too many metrics at once and inflating the effort.
