AI in construction: what the software actually does

AI has arrived on UK construction sites, and the most established use of it is not a robot. It is software that checks what has actually been built against what the drawings and the programme say should be there.

Here is what that looks like. A technician from the supplier walks the site with a 360-degree camera, filming as they go, and uploads the footage when they are done. Software compares it against the build programme and the 3D model, and returns a list of differences: what is in place, what is not, and what is behind.

That is Buildots. Its published UK work includes the BEACH building at Royal Bournemouth Hospital, delivered by IHP, the joint venture between VINCI Building and Sir Robert McAlpine, and a Wates Residential scheme at Shuttleworth Road in Wandsworth, which Wates describes on its own site as 71 homes for the council.

What that changes is not whether a job gets checked independently. On a hospital scheme there is already a client project manager, a cost consultant, building control and warranty inspection. What changes is that an independent check becomes continuous and cheap, instead of periodic and expensive.

We went through 22 companies working on AI in construction. What follows is what each of them says on its own website, which is marketing, so it is reported here as claims rather than as results. Where a UK project is named, we checked it against the client's own site as well.

Progress tracking: knowing what is actually built

Progress is already recorded on a job in a dozen ways: percentage complete against activity, short-term look-aheads, subcontractor returns, site diaries, and monthly valuations for payment. The difficulty was never an absence of numbers. It is that the earliest and most immediate account of progress comes from the people who will be judged on it.

Software has not removed the need for someone to decide what counts as complete. What it has done is change where the first version of that answer comes from.

Buildots takes that furthest. Founded in 2018, it has a London office and states on its own about page that it has raised $166 million. It describes the capture as follows: "Certified Buildots technicians scan the full site with 360 cameras and upload them immediately after completion." The software then compares the scan against the programme and the model, and reports the gap.

Andy Towers, Chief Engineer at Sir Robert McAlpine, is quoted on its site: "The real power of Buildots is that it provides an objective view of what is actually happening on the project." Objective is the word doing the work. What is being sold is not the camera, it is an account of progress that did not come from the contractor.

OpenSpace works from a wider range of inputs, listing "Photos, videos, voice notes, 360° walks, drone flights" as the raw material, and pins them automatically to the point on the drawing where they were taken. A given spot can then be pulled up as it looked on a given date. That matters for disputes, and for anything that gets covered up: pipework behind a wall, fixings under a floor.

Worth noting how OpenSpace describes what Track actually does, because it is more honest than most of this category: it "combines jobsite imagery with computer vision and expert human verification". There are still people in the loop. They are simply not the people being measured.

Both companies have been buying. OpenSpace announced its acquisition of Disperse, the London progress-tracking company, on 28 October 2025, and now sells that technology as OpenSpace Track; disperse.io redirects to openspace.ai. Buildots announced its purchase of the workforce platform Genda on 16 October 2025, and sells it as Buildots Field. Two acquisitions in one month is not a trend on its own, but the direction is towards fewer, broader platforms rather than more point tools.

Predicting the programme instead of reporting it

nPlan is London-based, founded in 2017 by two people who met through Entrepreneur First.

It forecasts how a construction programme is likely to run, using a statistical model its own site says is trained on 750,000 historical schedules representing more than $2 trillion of construction spend. The output is blunt: your programme says the job finishes in March, and the model, having seen how comparable projects behaved, gives you a different date.

It names Laing O'Rourke, Galliford Try, BAM, Skanska, Kier, Network Rail, HS2 and Anglian Water on its own site, and raised a Series A of $18.5 million led by GV, Alphabet's venture arm. This part of the market is not about seeing the site at all. It is about using past programmes as evidence for the next one, rather than filing them and starting fresh.

Instrumenting the material itself

Two companies work on the materials rather than the management: what the material is doing once it is in the building, and what came through the door in the first place.

Converge describes itself as "a UK construction technology company based in central London", founded in 2014. It casts wireless sensors into concrete as it is poured, and they report how fast that concrete is gaining strength.

That sounds like a small thing. It is not. Concrete strength governs when the temporary moulds can come off and when the next lift can go up, and the traditional way of judging it is a lab test on a sample cube plus a margin of caution. A Laing O'Rourke engineer is quoted on Converge's own site saying the data means "we're striking formwork as soon as possible, rather than waiting for lab results which can come in days after strength has been reached". Its sensors have gone into central London schemes including 105 Victoria Street and the 40 Leadenhall tower.

Qflow, also London, founded in May 2018, works on what comes through the door instead. Someone photographs a delivery ticket or a waste transfer note on a phone as it arrives, and paper that used to end up in a folder becomes live data on materials, waste and carbon. Its own description of the problem it replaces is recognisable to anyone who has done it: "the current reporting process requires someone on site to physically gather Waste Transfer Notes and material delivery receipts and take them to the project office for the document controller to input into a spreadsheet". Its published case studies name Berkeley Group at Kidbrooke Village, Canary Wharf Contractors at Wood Wharf, and Skanska.

Putting the model on the site

XYZ Reality, founded in 2017 and headquartered in London, built a headset called the Atom that projects the 3D model onto the site in front of you. Its own description is "a construction safety headset, augmented reality displays and in-built computing power", letting teams "view and position holograms of 3D design models to millimeter accuracy onsite". You can see where a duct is meant to run, in position, at full size, and check what has been installed without measuring back from drawings.

It is the only hardware product described here. The company has since narrowed its focus towards data centres and similar work, where being a few millimetres out is expensive. It names Mace and Cundall among its partners.

The estimator's side of the business

The last of these is the part of construction that happens before anyone reaches site.

Togal.AI does takeoff automatically, meaning it measures the quantities of material off a set of drawings. Its own headline is "Takeoff in Minutes. Not Days." Done by hand, that job can eat an estimator's week. Beam, built by Attentive.ai, does the same for trades including HVAC, plumbing and steel, and does it partly as a service rather than purely as software: "99%+ accurate, QA-checked takeoffs and estimates in 24-72 hours". A human estimator reviews the output before it comes back to you. (Note the web address, ibeam.ai. The similarly named Beam Global, an unrelated electric vehicle charging company, is at beamforall.com.) Handoff aims at residential contractors and remodellers, bundling estimating together with the job paperwork that surrounds it.

Of everything described here, estimating is the easiest case to make. A takeoff is a repetitive, rules-based measuring job done by an expensive person under time pressure, which is close to the ideal shape for automation.

The gap between what exists and what gets used

All of this is running on real projects. And construction is still among the lowest adopters of AI in the country.

The Office for National Statistics, reporting on 20 July 2026 from June fieldwork, found "over half of businesses in information and communication (58%) reporting using AI, compared with much lower levels in construction (13%)". Among UK businesses with ten or more employees across all sectors, the same release puts AI use at around 35%.

The main reason is who these products are sold to. Almost everything above is bought by main contractors, priced against project or portfolio size, and reached through a sales conversation rather than a sign-up page. The technology is genuinely working, and most of the industry will never see it.

One caveat about all of it. Our method was reading each company on its own website, which establishes what these products are and who is publicly using them. It does not establish what they save. The case studies are the vendors' own, and we found no independent measurement of the return on any of them.

The institutions are catching up

Two of construction's professional bodies have now taken a position, and they have taken different ones.

The Royal Institution of Chartered Surveyors published its first global standard on responsible AI use, "in effect for all members and regulated firms from 9 March 2026". Its reach is narrower than most of the coverage implied. Section 1.2 states that it "sets requirements for members and RICS-regulated firms", so it binds surveyors and RICS-regulated firms rather than the construction industry at large. Where it does apply, it requires competence in the systems being used, a risk register reviewed at least quarterly, and a written record of the risks that remain when a supplier will not answer questions about its own system. That last requirement is the interesting one, given how little most of the companies above disclose.

The Chartered Institute of Building went the other way, publishing a voluntary playbook written "for built environment organisations of all sizes".

What it actually changes

Take the categories together. Buildots and OpenSpace produce a record of what has been built, captured continuously rather than reported weekly. nPlan gives a view of how the programme is likely to run that does not come from the people running it. Converge puts sensors in the concrete so it reports its own strength, and Qflow turns delivery tickets into data as they arrive. XYZ Reality puts the design in front of you at full size, so what was drawn and what was built can be compared on the spot. Automated takeoff removes a week of measuring from the front of a job.

The estimating tools are a straightforward time saving. The rest share something less obvious. Each takes an assessment that used to be produced by the party whose performance it describes, and moves it. Not away from people, as OpenSpace's reference to human verification makes clear. Away from the people with something riding on the answer.

That is the part worth watching, and it is a commercial change more than a technical one. If the first account of progress no longer comes from the contractor, the weekly progress meeting starts from a different place.

FAQ

Which of these are British? Four of the companies still trading independently. nPlan, Qflow and Converge are all London-founded, and XYZ Reality is headquartered in London. Disperse was British too, and has been acquired by OpenSpace. Buildots was founded in Israel and has a London office, and its strongest published case studies are UK projects. Togal.AI is Miami-based, Beam is built by Attentive.ai in the United States, and Handoff publishes US prices and US customers only. None of the three names a UK office, customer or price.

Is any of this proven, or is it still pilots? Further along than most sector AI. The progress-tracking and materials companies publish case studies on named UK projects, and at least one carries a named engineer quoted by name. But every one of those case studies is published by the vendor, and we found no independent measurement of what any of it saves. Treat them as evidence the products work, not as evidence of the return.

Why is construction so low in the adoption tables if all this exists? Because of who it is sold to. Pricing is negotiated against project or portfolio volume, most of these products publish no price and offer no self-serve route, and the sales model assumes a buyer with a procurement function. That leaves most of the industry outside the market.

What happened to Disperse? Acquired by OpenSpace, announced in October 2025. Its progress-analysis technology is now sold as OpenSpace Track, and its old domain redirects to the acquirer.

---

If you want to talk through what any of this means for your firm, book a conversation.