Tech for RetailMeet us there
Opinion · AI on data

You don’t need a semantic layer

In the data community, the consensus is that, to use AI on your data, you need a semantic layer.

For most companies, it’s probably something they will have to do down the line. But the truth is, it’s not necessary to start, and for many of them, it’s useless.

If you are a small company, you will probably be fine without one.

If you just want to use AI with your IT / data teams, you can start without one.

The truth is, before the AI revolution, most companies never had a semantic layer.

Humans relied on shared, messy context for thousands of years. They analyzed some data, gained some insights, sometimes made a decision… and then moved on.

No formal “active customer” definition. No “owner”. Nothing to maintain.

It seems bad, but creating and maintaining a semantic layer is not doing the work.

At the end of the day, data is only useful if it helps make a better decision and improve the business. Maintaining a semantic layer is like having clean code: maybe it helps, but it’s rarely the turning point.

Formalization is not useless. Contracts bind two parties. Accounting standards exist for investors and tax authorities. Civil law exists for strangers who share neither context nor trust.

Data works the same way. We should formalize to build trust between different parties:

  • financial and regulatory reporting
  • KPIs tied to bonuses and targets
  • group-wide indicators in large organizations

Leave the rest to context: a catalog, a glossary, the company’s vocabulary. An agent with that context can rebuild the calculation.

Full guideAI on enterprise data: do you need a semantic layer?

Why AI fails on data, the six levels of governance, and which level each use requires.