One meaning for every record

Before an organisation can trust its AI, it has to agree on what its own data means. The data dictionary is the least glamorous and most important AI investment.

Drawers of a card-index file cabinet
Photo: Maksym Kaharlytskyi on Unsplash

Every large institution we work with has the same hidden problem. Ask three departments for the number of active assets, beds, vehicles or cases and you will get three answers. None of them is wrong. Each department simply defines the word differently.

People cope with this through experience and phone calls. AI cannot. A model trained on inconsistent definitions will produce confident, well-formatted and quietly wrong answers.

Five problems, one root cause

When we audit data estates, the same five issues appear again and again:

  • Data locked in separate systems and sites.
  • No shared names or definitions for data.
  • Little live data, and systems that do not talk to each other.
  • Gaps in security, classification and compliance.
  • No central rules for AI and data.

Most of these trace back to the second point. Without shared meaning, integration becomes a series of one-off translations, security rules cannot be applied consistently and governance has nothing to govern.

A common data dictionary is not documentation. It is the contract that lets every system, and every model, speak the same language.

Shared ontologyFinance systemField recordsAsset registryCitizen services
Many systems, one shared meaning for every record.

What good looks like

A useful data dictionary is alive. It is connected to the systems it describes, it records where each field came from (its lineage), and it carries quality checks that run every day, not once a year. Crucially, it is owned by the business, not by the IT team alone.

Start small, but start

You do not need to define every field in the organisation before you begin. Pick the decision that matters most, map the data it depends on and agree the meaning of those terms first. Each pilot adds to the dictionary, and the dictionary makes each next pilot faster.

Want to explore how this applies to your organisation? We share our approach with qualified organisations in a private briefing.