AI That Fits the Way Construction Already Works

Most construction teams don’t need a new way of working. They need technology that fits into the way they already work.

The gap isn’t about appetite

Two thirds of construction professionals now use AI in some part of their work, up from just over half the year before, according to RICS’s 2026 report. In the UK, the fall in organisations using no AI at all was the largest of any region. But only 19% use it regularly, and most organisations are still at the pilot stage. That isn’t because the industry is resistant to new technology. RICS puts it plainly: a year ago the gap was between planning to invest and starting. Now it’s between running a pilot and embedding it.

RICS also explains why that step is harder in construction than in commercial property, where regular use is 29%. Construction data resets at every project boundary, and income comes project by project, so each tool has to prove itself again on the next job.

The barriers haven’t changed either. Skills, integration with existing systems and data quality are still the top three. Privacy and security concerns have grown, from 22% to 30% in a year. People are waiting for tools that don’t ask them to relearn their jobs, don’t need the data cleaned up first, and don’t move their records somewhere they can’t account for.

It’s a theme we’re seeing more and more across the construction press. PBC Today and ENR have both made the same point recently: AI earns its place on a project when it fits into the work people already do, not when it sits on top of it. Bluebeam’s chief executive, writing in PBC Today, said the firms not yet using AI keep raising the same three concerns: data ownership, compliance and responsible use.

Why it matters now

The cost of getting this wrong is already known. The Get It Right Initiative puts avoidable error at around 21% of project costs, up to £25bn a year across UK construction. Its recent pilot with Kier, BAM, VolkerStevin and Taylor Woodrow avoided £92.6m of it. A missing sign-off caught before a pack goes in costs a fraction of one caught at handover.

The paperwork is growing too. RICS points to the Building Safety Act gateway regime as a clear early use case for AI, because of the extra documentation it demands. The golden thread is a legal duty, the Building Safety Regulator has been a standalone body since January, and the Building Safety Levy came in on 1 October. The evidence has to be findable, current and attributable, and the people producing it don’t have more hours in the day.

And safety hasn’t gone away. Construction still records more worker deaths than any other industry, 25 in 2025/26 according to HSE. RICS found it’s the area where professionals expect AI to matter least, at the same time as the obligations around it are rising fastest.

Build with the process, not over it

That’s how we’ve built BuildPrompt. We don’t replace your process or ask people to learn a new way of working. We build on what’s already there.

PromptStack, our workflow engine, is designed around this. It reads the systems and file types teams already use, structured or not, and builds hierarchy from the data as it stands. It maps how that information connects across sources, so teams can ask it directly and automate the processes they already run. Nothing needs to be migrated, restructured or relearned first.

Think of it as a grid that fills itself in. Each row is a record. Each column is a prompt, written the way the professional would have asked it. Each cell is a response, cited to the page, paragraph or cell it came from. Once a check is set up it keeps running, every night, as new records come in.

What that looks like in practice

Contractor assurance: contractor submissions are checked against requirements as they come in, inside the assurance process teams already run. On HS2, that took it from two weeks to 35 seconds per asset, an externally estimated £5m annual saving.
Quality: NCRs and ITPs are linked, so teams can see which inspections are actually catching defects, using the records they already keep. A pack that’s missing a sign-off is flagged before it’s submitted, with a line such as “three of the four required checks are recorded; the concrete pour sign-off is missing”, and the gap cited to the record it should have been in. On a construction joint venture, that took quality intelligence from three days to under two minutes.
Health and safety: reports from different systems are pulled together, so emerging hazards show up earlier without anyone changing how they report. The same approach at a leading UK airport took safety intelligence from five hours to 20 seconds.

People stay in charge

The engineer still makes the call. The assurance lead still owns the sign-off. Every response comes back with the evidence behind it, so decisions stay traceable and accountable. We take away the admin, not the judgement.

That’s what the profession is being asked for now. RICS’s new standard on responsible AI use puts responsibility with a named surveyor, not the system, and asks firms to record what data the AI ran on and where a person reviewed it. A citation on every response does that by default.

Your data stays where it is. BuildPrompt runs inside your boundary, and the model it builds of your organisation stays yours. We never pool it, publish it or use it to improve anyone else’s results.

Where to start

A good test is to look for where errors keep recurring and which decisions eat up the most time. In practice, these are the signals we look for with every client:

Unstructured data: key information is locked in PDFs, drawings, emails and spreadsheets, so it’s hard to find, compare or trust.
How data is being used: teams collect plenty of data but only use a fraction of it, or rely on a few people who know where everything is.
Slow manual processes: reviews, checks and cross-referencing take days or weeks when the result is needed now.
Corners being cut: there’s too much data or too little time to review it properly, so checks get sampled, stripped back or skipped.

Then pick one check that would be audited if it were wrong, and time how long it takes today. That’s the baseline for the first phase, and it’s the difference between a pilot that impresses and one that sticks.

Wherever these show up, there’s usually a quick win, and it can be delivered without changing how the team works.

Good AI shouldn’t change how people work. It should make the day better.

About BuildPrompt

BuildPrompt is the application layer for regulated industries. It reads the documents, spreadsheets, images, systems and live web pages an organisation already holds and returns cited, auditable intelligence, with every response traceable to the page, paragraph or cell it came from. Powered by PromptStack. ISO 27001, Cyber Essentials Plus and Crown Commercial Service Supplier.

AI that fits the way construction already works