A Taxonomy Isn't Paperwork

“Taxonomy” sounds like the least exciting word in this whole conversation.

It's also one of the things that determines whether your AI agents can be trusted.

It's easy to treat taxonomy definition as a documentation task. Something to get through quickly before moving on to the “real” engineering work.

That's a mistake.

A taxonomy gives data elements across your organisation clear, consistent names and meanings.

Without it, an AI model can end up working with a vocabulary that means different things in different systems.

And when the same information is described inconsistently, the consequences go beyond confusion.

Unreliable outputs. Agent misfires. Behaviour that's difficult to trace and debug.

The good news? You rarely need to start from scratch.

Established industry standards already provide useful foundations:

- UK Open Banking: OBIE

- US Open Banking: FDX

- European Open Banking: Berlin Group NextGenPSD2

- Global financial messaging: ISO 20022

- Telecommunications: Open Digital Architecture

Where an existing standard doesn't fit, a custom taxonomy built around your organisation's own business language can provide the consistency you need.

The important part isn't whether the taxonomy is standard or custom.

It's whether every system is speaking the same language.

For financial institutions trying to move AI from experimentation into production, that consistency is part of the infrastructure needed for reliable, governed AI access.

Not an administrative exercise to complete along the way.

So, has your organisation adopted an industry standard, developed a custom taxonomy, or combined elements of both?

What has worked best when connecting AI to your existing systems?