Notes on AI in construction

AI learns from the information you give it, so the state of your data shapes what is possible.

Questions worth asking

• Is the data stored in one place, or scattered across inboxes and folders?
• Is it labelled consistently, such as photo locations, drawing revisions and cost codes?
• Do you have enough history from past projects to learn patterns?
• Are you allowed to use it, including client and subcontractor information?

Gaps do not rule out AI, but they do change the plan. Often the first step is improving how data is captured.

Computer vision can help spot patterns that are easy to miss: workers without protective equipment, people entering restricted areas, or the same hazard appearing again and again.

What it does not replace

It does not replace supervision, training or a good safety culture. Camera angles, lighting and weather affect results, and any system will sometimes be wrong. The goal is to give safety teams better information, not to take people out of the loop.

Involving site teams early, and being open about what is monitored and why, matters as much as the technology.

Construction projects generate huge volumes of text. A document assistant lets people ask a question in plain language and see the relevant passage.

What to insist on

• Answers that cite the exact document and section.
• Access rules that match who may see what.
• Clear handling of superseded revisions.
• A habit of checking the source before acting on an answer.

Ready-made tools are a sensible choice when your process is standard and the tool fits it. Custom AI makes sense when your workflow, data or standards are distinctive, or when the value comes from connecting several systems.

A simple test

If you would have to change how your team works to suit the tool, look closer at a custom option. If the tool slots in with little change, you may not need to build anything.