Your AEC data is already ready enough to start

This is Part 2 of the three-part series 'AEC AI IS NOT A MODEL PROBLEM', a critical review of the habit, common in AEC organizations, of postponing AI adoption on the grounds that the data is not ready. Unifying naming conventions, tidying classification systems and building ontologies may all be necessary work, but none of it has to come first. On that premise, and drawing on official sources such as Speckle's object-based data model, buildingSMART IDS and ISO 19650, the article sets out an alternative approach: the question comes first.

It treats data readiness not as a single organization-wide state but as a ladder of semantic certainty — exploratory reasoning, verification of repetitive work, and high-stakes deterministic meaning — and argues that a knowledge graph should be introduced only at the moment its cost is justified, such as multi-hop relationship traversal or cross-domain identification. Formalization does not remove ambiguity; it relocates it. The piece closes with a practical principle: semantic debt is not repaid in one go, but paid down in the order set by the frequency and cost of the errors it causes.