← Inside PlanOps

AI governance

AI in construction: competence, evidence and the steps between

Ian Yeo2 min read
AI in construction: competence, evidence and the steps between - PlanOps notes from Digital Construction North, shown beside a site professional marking up printed documents with a pen next to a laptop showing the same document highlighted

Yesterday I spoke at Digital Construction North in Manchester. My talk was called AI without a record is a liability, and its central question was simple: when AI helps someone make a project decision, can we show what informed it?

A specification summary in a closed chat, quantities pasted into Teams, or a method statement saved without its history can all influence work. The useful record connects the request and source revisions, what the tool actually received, its original output, the human review, and approval for a defined purpose.

I shared an example from our own testing. A processing fault meant no document content reached the AI, yet it produced a convincing RAMS review. Checking that a file was uploaded would not have exposed the problem. We needed to know what reached the review step. I have written about that fault, and the controls we built after it, in How we stop AI making things up.

Competence and the gap between the document and the data

James Wright-Zhang's session, Back to the source: AI competency and the professional brain, brought a closely related challenge. A professional checking the answer needs to recognise omissions, question assumptions and understand the limitations of the tool. His session also explored the gap between the document a person sees and the information AI reads.

It left me thinking about the steps between the source and the answer. In PlanOps, our AI tasks already involve distinct stages. The next question is which intermediate results should be visible, checked or challenged before the work moves on.

Where to start

There is a practical foundation for this. ISO 19650 information management guidance covers revisions, status and controlled exchanges. The RICS professional standard on the responsible use of AI in surveying practice has been in effect since March 2026, and the new CIBSE AI position statement reinforces professional judgement and verification.

My takeaway is to start with the AI uses affecting safety, design, compliance or significant commercial decisions. Check the evidence, the review and the route into the project record.

A good record cannot make a bad decision safe. It gives us something we can examine, challenge and improve.

AI GovernanceConstruction AIEvents

Tell your project what you need.

Start free with 50 IU. No credit card, no training, no waiting.

Start free