The document work AI actually handles well

Forty-four per cent of contractors say they don't know how to use AI in their business. Another 43% say there are no clear use cases. Those numbers come from a recent trades contractor survey, and they're worth sitting with, because the use cases exist, they're just not being talked about plainly.

What happened

A few guides aimed at UK contractors have been circulating that do something useful: they name the actual tasks general AI tools handle well, without dressing it up. The short list is tender document summarising, RAMS drafting, daily report write-ups, meeting minutes, RFI drafts, variation assessments, and cost plan narratives.

Nothing exotic. All of it is document-heavy, repetitive, and low-stakes enough that a human reviewing the output catches any errors before they matter. The guides recommend starting with Claude or ChatGPT, not specialist construction software, for most of this work, and only moving to a purpose-built tool if the volume justifies it.

The one caveat worth noting: the AI needs your actual project information fed into the prompt. Generic questions get generic answers. Paste in the spec section, the sub's scope letter, or the meeting notes, and the output is usable. Ask a vague question and you get a vague answer you have to rewrite anyway.

Why it matters for your shop

For a company doing $5M to $90M in work, the estimating department is usually one or two people carrying a heavy document load. Tender packages run long. RFI logs pile up. Someone is writing the same boilerplate language in every sub request for information.

None of that requires judgment, it requires time. That's where this fits.

What it doesn't do: quantity takeoffs, pricing, or anything that needs a number you can stand behind. The tools that do takeoffs are separate, and they're still earning trust. For now, the honest value is in the writing and reviewing layer that wraps around the numbers, the stuff that takes an estimator two hours after the real work is done.

If your estimator is staying late to write up a tender submission narrative or pull together a variation log, that's the place to start.

One thing to try this week

Take the last RFI your team wrote and the spec section it referenced. Open Claude or ChatGPT, paste both in, and ask it to draft a response. Don't use the output as-is, read it, correct it, and send your version. The point is to see how much of the first draft it can handle before your estimator touches it. Most people who try this once keep doing it. Austin

Source: https://buildcopilot.ai/guides/how-to-use-ai-in-construction