Where adoption actually sits
The ONS found in July 2026 that 13% of construction businesses use AI, against 58% in information and communication. RICS reported in 2026 that most construction organisations using AI are still at early pilot stage, and that embedded use is rare.
Those two findings together describe something more specific than slow adoption. A pilot is not embedded use. A firm that has tried a chatbot on a tender response and a firm that produces every site diary through a checked workflow both appear in adoption statistics, and they are not doing the same thing. When you read that a competitor is using AI, it is worth knowing which of those they mean.
Why the gap is commercially interesting
Low adoption is usually reported as a failing. It is also an opportunity, because it means the firms adopting properly now are not competing with a crowded field. It also means most of what is sold as construction AI has never run on a live site, and that is the part worth being careful about.
The problem the industry is actually trying to solve
The case for doing anything at all rests on how much time documentation takes. Re-flow and YouGov found that UK construction workers spend around 90 hours a year each on paperwork, at a cost of about £1,500 per employee per year. The ONS reports that UK construction productivity has grown less than 0.5% a year since 1997, against 2.8% for the wider economy. The Construction Leadership Council estimates that digital tools which reduce admin could improve SME construction workforce efficiency by 5%, worth £2.2 billion.
The labour market closes off the obvious alternative. CITB estimates the UK will need more than 251,500 additional construction workers by 2029, so hiring people to absorb documentation competes for the same scarce staff you need building. That is the real argument for automating documentation rather than staffing it.
What is genuinely working
The work that responds is repetitive, document-shaped and already being done by someone expensive. In practice that is a short list.
- Site diaries built from spoken notes and photographs before the manager leaves site, rather than written from memory the next morning.
- Subcontractor packs cross-checked against themselves, which finds the contradictions that reading one document at a time cannot.
- Onboarding checks run before a trade mobilises, where the failure is always the same: something expired, something missing, something contradicting another document.
- Drafting from a firm's own approved templates and previous submissions, so the starting point is site-specific rather than generic.
What these have in common is that a person still decides. The output is a draft or a set of findings, and a competent person reviews and signs it.
What is being oversold
Three claims deserve scepticism. The first is any suggestion that software can hold a duty. A competent person is a legal role held by a human being, and no software takes on a CDM 2015 duty. A supplier implying otherwise is describing something you should not buy.
The second is the platform that replaces everything. Most firms already own more software than they use, and the admin load persists anyway, because the problem is usually the workflow rather than the absence of a tool. Anything that produces documentation should feed the systems you already run, including your Common Data Environment, rather than asking you to move.
The third is the demonstration on a clean file. A pack assembled to show a product well tells you nothing about a hundred-page submission that arrived the day before a start on site. Ask to see it run on your own paperwork.
The safety case for better records
There is a reason this matters beyond hours saved. HSE reported in Construction statistics in Great Britain 2025 that 35 construction workers died in 2024/25, and that the construction fatal injury rate is around 4.8 times the all-industry rate. The HSE fee for intervention rate rose to £188 per hour on 1 April 2026.
When something goes wrong, the question is what you checked, when, and what you did about what you found. Records that are complete and consistent are not an administrative nicety in that situation. They are the evidence.
How to judge a supplier
A short set of questions separates most claims quickly. Does it run on your documents, or only on theirs? Does a named person approve every output? Does it feed your existing systems, or replace them? Is the module you are being shown live on a real site, or in development? Will they tell you when the answer is that you do not need them?
That last one matters more than it sounds. If your firm has one clearly defined problem and a tool that already solves it, buying the tool is the right answer, and a consultancy telling you so is being useful rather than unhelpful. Our comparison of consultancy against software sets out where each one genuinely fits, including the cases where we are the wrong choice.
The honest summary: adoption is low, the problem is real and measurable, a narrow set of document work responds well today, and a good deal of what is marketed has not been tested on a live site. None of that is a reason to rush, and all of it is a reason to start with something small and verifiable.
A sensible starting point
Pick one document type you produce every week, measure what it currently costs, and change only that. Our free tools exist for exactly that purpose: the Construction Admin Cost Calculator sizes the problem in about two minutes, and the free RAMS pack check runs a real cross-document review on one of your own packs. Neither commits you to anything, and if the numbers turn out to be small, that is a genuine result worth having.