A semantic layer is the context layer between raw data and the people or AI models that use it. It keeps definitions such as revenue and active customer identical everywhere. This article explains why a data dictionary on its own goes stale, how the gold layer and a definition in code prevent that, and what that looked like at CentreBlock and Runderkamp.
Kees Pronk
CEO
A semantic layer is the context layer between raw data and the people (or AI models) who need to do something with it. It ties technical fields to business definitions, so that "revenue" or "active customer" means the same thing everywhere. At our clients we see that this is exactly the part that makes an AI project succeed or stall on conflicting numbers.
A semantic layer translates technical column names into understandable business terms, with unambiguous rules for calculation and meaning. Where a table shows "cust_stat_02", the semantic layer shows: "customer status, active within 90 days".
This layer typically sits above the cleaned data and below the reports, chatbots, or agents that work with it. Without that layer, every system or employee interprets a concept in their own way, with different outcomes as a result.
AI models (systems that recognise patterns or carry out tasks based on data) have no sense of business logic unless you record it explicitly. A language model does not know on its own that "revenue" at your company means excluding discounts.
In practice we see that a dashboard makes an error in a definition visible as a kink in a chart. An AI agent quietly works that same error into advice or an automatic action, without a warning. That is why a semantic layer weighs more heavily once you move from reporting to autonomous execution.
The biggest benefit is data consistency: everyone, from controller to AI agent, calculates with the same definition of the same concept. That prevents debates about whose number is correct.
A semantic layer also strengthens data governance (agreements about ownership and management of data). Record, per concept, who the owner is and who may make changes. Tie that to access rules such as row-level security, so sensitive definitions are only visible to the people who need them. Arrange this from the first implementation, not afterwards as a repair.
Start with the concepts that most often lead to debate, such as revenue, margin, or active customer. Record the calculation, the source, and the owner for each concept before you build a dashboard or an agent.
Work from a clean data foundation. Without tidied source data you only build consistent definitions on a shaky foundation.
The medallion architecture (a layered structure of data in Bronze, Silver, and Gold) gives the semantic layer a fixed place. The Bronze layer holds raw data, the Silver layer cleaned data, and the Gold layer the business-ready definitions the semantic layer rests on. Build the semantic layer on the Gold layer, not apart from it, so reports and AI agents consult the same definitions.
A data dictionary is the static list of definitions: a document that records what a concept means. A semantic layer is the active, technical translation of those same definitions into code, directly usable by systems.
The problem with a data dictionary on its own: it gathers dust. Someone changes a definition in an Excel file, but the dashboard or model behind it keeps using the old rule. A semantic layer ties the definition directly to the metrics and KPIs calculated with it, so a change takes effect everywhere.
At a marketing intelligence client such as CentreBlock (a marketing data platform), 11 KPIs were spread across separate reports, each with slightly different calculation rules. By recording those KPIs in one semantic layer, the debate about which number was correct disappeared and reporting time dropped from days to minutes.
At a food retail client such as Runderkamp, a similar pattern played out with about 1,400 products and 170 suppliers. Unambiguous definitions per product and supplier prevented each system from applying its own version of the truth.
A semantic layer is not a standalone technical project but the context layer that connects business definitions, the gold layer, and a data dictionary. Without that connection, a data dictionary remains a document that gathers dust, while an AI agent keeps calculating on outdated rules. Want to know how this works for your data foundation? Book a Discovery Call and talk it through in concrete terms.
Author
CEO
Boards often ask for a better dashboard, while the real split is signalling versus reporting. One shows what happened; the other acts when it counts.
Kees Pronk
CEO
A portal that works at three clients often stalls at thirty. How row-level security and a multi-tenant setup replace manual rights work.
Kees Pronk
CEO
One shared report, each user only sees the rows they are entitled to. How boards share data with staff, partners and customers without losing control.
Kees Pronk
CEO
