Connecting AI to your data · Layer 2
Semantic layer
A shared reference that defines your business terms, such as “active customer” or “invoice”, with an owner, so people and AI count the same thing.
A way to picture itIt is the organisation’s official dictionary: without it, finance, sales and the AI all talk about “revenue”, but each one counts something different.
An AI agent has to report the number of invoices unpaid for more than 60 days. Accounting counts invoices issued, the collections tool adds credit notes, the sales system leaves out subsidiaries: three figures for a single question. The finance department sets one definition, owned by the head of accounts receivable and tested on a month already closed. The agent, the dashboard and internal audit now get the same result.
The accuracy principle assumes you know what each data item means: a vague definition produces wrong figures, including about people (EU · GDPR, art. 5). The record of processing activities lists categories of data; it stays theoretical if those categories have no shared definition (EU · GDPR, art. 30). For high-risk systems, the AI Act requires governance of training, validation and testing data, including the assumptions about what the data are supposed to measure and represent (EU · AI Act, art. 10). A versioned definition with an owner is stronger evidence of control than an explanation given after the event.
“When your agent calculates a metric, where does the definition it applies come from, who on our side approved it, and how are we told if it changes?”
A data dictionary describes technical columns. A semantic layer describes business concepts, their calculation rules and their owner, and it is used directly by tools and agents.
It is a richer form of semantic layer: it describes business objects, their relationships, the rules that link them and the actions possible on each. It is particularly useful for agents that must act, not just count.
With the dozen or so terms that keep coming up in your committees and dashboards, and for which two departments report different figures. Every gap you close is an immediate gain, with or without AI.
Educational explainer, not legal advice. Up to date as of 29 October 2026.