Rob Norton, UK director, PlanRadar asks what happens when the dust settles on a contract.
Great news: the meeting went well. The vendor walked the room through a polished interface, the use cases mapped neatly to the challenges described in the brief, and the senior team left the boardroom genuinely excited. Sign-off came quickly and the rollout email delivered, but six months later the tool lies dormant on site servers, while teams carry on doing what they've always done. So where did it go wrong?
The untold story
AI procurement in construction has a structural problem. The people making buying decisions are rarely the people who will use the tool at 7am on a muddy jobsite with a subcontractor querying a drawing and three inspection tasks already overdue. Demonstrations are typically optimised for presentation before procurement committees, but not for operating in live project conditions. They are smooth, controlled and compelling in ways that real construction work is not. Frankly, a tool that performs brilliantly under demo conditions may be incompatible with the pace and pressure site teams experience.
Simply, when tools don't fit, teams don't use them and (frustratingly) usually there's no resistance or complaint. Teams just take the line of least resistance and revert to what they know: spreadsheets, WhatsApp threads, even 'good old' pen and paper. It all represents a passive implementation failure, currently the most common outcome of AI adoption in construction.
The scale of what's happening really matters. According to 2026 RICS data, around two-thirds of construction organisations now use AI, though 39% are stuck in pilots that never reach full deployment. Concerningly, fewer than 1% have AI embedded organisation-wide! Meanwhile, the AI market in construction is projected to grow from $1.8 billion to $12.1 billion by 2030, it all shows there’s is a serious gap between investment ambition and operational reality.
Counting the cost
AI tools rely on consistent, structured data flowing through them regularly to generate insight, so sporadic use, or partial use by some teams on some sites, produces little that's useful. Our own research found 77% of construction companies already struggle with inconsistent quality processes, and a tool that gets installed but is not habitually used makes that problem worse, adding another data gap on top of the ones already there.
Poor quality data cost contractors an eye-watering $1.6 trillion globally in 2020; it’s a huge figure and can be seen in the amount of rework, delays, disputes and project overruns. The industry’s unnecessarily shooting itself in the foot; failed AI adoption is a glaring missed chance to start addressing a cost construction has been carrying for years.
To be fair, most firms are not entirely disengaged with AI and the CIOB has observed that construction leaders are using tools like ChatGPT for drafting correspondence and basic research. However, the transformational applications, risk monitoring, document analysis, quality assurance at scale, these opportunities remain largely untouched.
What good actually looks like
Effective adoption does not look like a successful pilot with hand-picked users and a dedicated implementation team. What you actually need is a site manager picking up their phone to log an issue without being reminded to do it. It has to become habitual, unsupervised and used day-in-day-out. If a tool requires a prompt, a chase email or a refresher session to stay in rotation, it has already lost, so the test is whether it earns its place in an existing workflow, rather than impress a procurement panel.
As Brabners' Building Links event put it: "You won't lose your job to AI. You'll lose it to someone who uses AI efficiently and effectively." It’s this framing that applies to firms and the individuals too. The divide between companies using AI well and companies accumulating a graveyard of underused tools will define competitive positioning in this sector over the next decade.
The question procurement teams need to start asking is "what does this look like six months after go-live, with no implementation support, on a site that's behind programme?" If the vendor can't provide a convincing answer, then the demo was the best it was ever going to get.