Perspectives
An ontology before a model.
21 November 2026 · Neurai
Choosing a model is the last decision and the easiest to change. Understanding the organisation is the first, and it is the one that makes AI accountable.
The first question in most AI projects is which model to use. It is the least important one. Models improve every few months and can be swapped in a day. What cannot be swapped is an understanding of the organisation the model is meant to serve.
What a model does not know
A language model knows language. It does not know that your second warehouse counts stock on Thursdays, that a promised date in the sales system is a hope until the purchase order is confirmed, or that orders above a limit need a second signature.
Those facts live across systems and in people’s heads. Until they are written down in a form a machine can follow, every answer is an estimate.
What the ontology holds
The ontology is a structured, living model of the organisation: its data, entities, processes and decisions, mapped as they actually operate. Each fact carries its source.
A question is answered by following the relationships that matter to it, not by finding passages that resemble it. The answer names the records it rests on.
Why this makes AI accountable
An answer that cites its records can be checked. An action proposed against a stated rule can be approved or refused by the person the rule names. Without the ontology there is nothing to check an output against, and trust becomes a matter of tone.
Build the understanding first. Then choose the model, and expect to choose again.
More perspectives