Construction’s interest in artificial intelligence has centered on estimating – the step that turns a set of drawings into a price. But an estimate is only as sound as the quantities underneath it. The step that produces those quantities, known as the takeoff, is where much of the industry’s pricing still goes wrong. Measuring walls, openings, fixtures, and finishes off a drawing is slow, often rushed, and rarely checked. Any tool that promises a faster estimate inherits whatever errors that process contains.
Adoption itself remains shallow. In a 2025 survey of more than 2,200 construction professionals by the Royal Institution of Chartered Surveyors, 45 percent said their organizations use no AI at all, about a third were still in early pilots, and just 1 percent had scaled AI across projects. Respondents named poor data quality as a barrier alongside skills shortages and integration problems. That suggests the inputs, not only the software, are holding adoption back.
In the UK, the cost of pricing errors falls on a sector already under strain. Construction recorded the highest or second-highest number of insolvencies of any sector in England and Wales throughout 2025. According to the Insolvency Service, roughly 3,950 construction companies became insolvent in the 12 months to November. Specialist trades such as electrical, finishing, and demolition subcontractors accounted for the largest share. Insolvencies have many causes, but for businesses operating on thin margins, a mismeasured job can decide whether the work makes money at all.
Armin Aghili works on AI-based takeoffs at Nestimate, a London construction technology company and one of a number of businesses building software for this step. He argues that part of the problem lies in how many estimation tools are built. In his view, companies have built estimation products on top of large language models designed to process text, when reading and measuring a drawing is closer to a computer vision problem. He says the results include tools that misidentify building elements, such as reading a toilet symbol as a door. He also says they lose accuracy on longer document sets as the models run up against their memory limits. Whatever the underlying architecture, buyers across the industry now have to ask whether a given tool can reliably read a drawing set.
When Speed Beats Accuracy
For smaller residential builders, the problem begins with how work is won. Speed tends to matter more than precision in deciding who gets the job, and the tolerance for error that creates carries through every later stage. Aghili says some contractors price jobs from a handful of photos, without formal measurement. He estimates that more than 60 percent of the time, the work goes not to the most accurate quote but to the first reasonable one.
The same imprecision shows up next in material orders. Aghili says builders he works with treat ordering 20 to 25 percent more material than needed as normal, and much of that surplus is thrown away. “I recommend going near a building site, go near their skips where they keep the rubbish,” he says. “You will see brand new things in that skip.” The pattern is long documented in the UK. Government and WRAP estimates put the share of materials delivered to construction sites that go unused at around 13 percent, a figure still cited in industry research.
In some markets, measured quantities barely drive price at all. In London, Aghili says, quotes for the same work can vary several-fold between streets a short walk apart, depending on what nearby homes sell for. Builders price to the margin a location will bear. When a quote is anchored to the neighborhood rather than to the quantities, there is little pressure to measure accurately in the first place. The cost of that inaccuracy surfaces later, in wasted materials and lost margin.
Where Document Volume Becomes the Risk
On larger commercial projects, the failure takes a different form. The problem is less guesswork than volume. These projects generate documentation on a scale that makes it hard to keep one reliable record of what has been scoped, changed, and paid for. Aghili describes cases in which a subcontractor was paid twice for the same work because no one could tell from the paperwork what had already been settled. The difficulty is familiar well beyond any one contractor. Arcadis’s annual construction disputes reports have repeatedly identified errors and omissions in contract documents as a leading cause of disputes.
Aghili says variation management is where larger contractors lose margin most visibly. Variation management means tracking the difference between what was originally scoped and what is actually built, and he ties it to the same root as the smaller builders’ problem. Duplicate payments are a record-keeping failure, but variation losses are a baseline failure. Every change on a project is priced against the original quantities. If those quantities were wrong at the start, each variation carries the error forward instead of correcting it.
Why the Tools Lack Specialists
The tools aimed at these problems face a constraint of their own: the people who build them. Lack of skilled personnel was the most commonly cited barrier to AI adoption in the RICS survey, named by 46 percent of respondents. Aghili sees the gap on both sides. He says technology companies entering construction often lack working knowledge of how buildings are scoped and priced. Meanwhile, construction companies building in-house AI capability tend to hire generalist developers rather than computer vision specialists, the discipline most relevant to reading and measuring drawings.
Recruiting those specialists is its own challenge. Aghili says top computer vision researchers have better-paid and more prestigious options elsewhere. “You’re not going to get an A-player who could have worked for Google for an insane salary to now come to construction and do building measurement,” he says. As a result, domain knowledge and technical specialization rarely sit on the same team. That gap may help explain why tools that perform well in a demonstration can struggle with a real drawing set.
The Unverified Baseline
Across all of these problems, the industry is layering new tools onto inputs it has not verified. Aghili says he has spoken with quantity surveying companies that spent 20 months evaluating AI products without finding one reliable enough to build into their workflow. “The industry needs to really go backward a little bit before they can start moving forward,” he says.
The check he describes applies regardless of which software a company uses, or whether it uses any. Before adding estimation tools, a company should confirm that its own team’s measurements and quantities are accurate. When companies do check, Aghili says, they almost always find that they are not.
That creates a cycle that is easy to misread. A company adopts a tool on unverified inputs, gets unreliable estimates, and concludes that AI does not work. The more basic failure was assuming the quantities were sound. With only a sliver of the industry having moved past pilots, the real test may have less to do with which software a builder chooses than with whether the measurements going into it can be trusted. For a sector with one of the highest insolvency counts in England and Wales, getting that foundation right is a margin question before it is a technology question.
About the Expert: Armin Aghili works on AI-based takeoffs at Nestimate, a London construction technology company.
This article is based on information provided by the expert source cited above. It is intended for general informational purposes only and does not constitute legal, financial, or real estate advice. Readers should conduct their own research and consult qualified professionals before making any real estate or financial decisions.
