RICS Sounds the Alarm: Construction Is Drowning in AI Pilots It Can't Graduate

The numbers look good until you read them properly. The RICS AI in Commercial Property and Construction Report 2026, published on 18 August and drawing on more than 3,100 responses from industry professionals worldwide, confirms that AI use across construction has accelerated sharply. About two-thirds of construction professionals are now using AI in some form, up from just over half in 2025. In commercial property, the figure is higher still, with more than three-quarters reporting some level of use.

But then you get to the number that matters on site: regular, embedded deployment. In construction, that stands at just 19%. Commercial property manages 29%. Both sectors are running the same proportion of early-stage pilots, 39% each, but property is converting them into live workflows at a rate construction simply isn't matching.

Pilots Are Not Programmes

This is not a new observation for anyone who has spent time on a live project. Construction has been running AI trials for the better part of three years. What the RICS data makes starkly clear is that the gap between "we're exploring it" and "we use it every day" has barely narrowed. The industry's challenge, as RICS itself puts it, has shifted from persuading organisations to experiment to actually embedding AI into professional practice.

The barriers are not mysterious. Legacy technology is the first culprit. Research from London Build puts 42% of construction and engineering respondents on the record saying outdated systems are slowing their progress. You can't bolt a machine-learning layer onto a project extranet from 2014 and call it a digital transformation. The second barrier is governance. As firms move from no engagement to pilots, and from pilots to regular use, they run into data handling questions that earlier-stage organisations haven't faced yet. The RICS report flags that privacy and security concerns in construction rose from 22% to 30% between 2025 and 2026, a jump the report links directly to increasing use, not increasing anxiety.

So the sector is in an awkward position. The more firms actually try to use AI at scale, the more they discover why scaling is hard.

The Regulatory Tailwind - and Headwind

Building safety regulation is doing some of the heavy lifting that commercial pressure hasn't yet. The Building Safety Act 2022 and its Gateway requirements are generating documentation volumes that lean site teams can no longer manage manually. AI is being pulled in, not because firms are innovation-hungry, but because the compliance load is simply too high. That's a real adoption driver, and it's UK-specific.

At the same time, the EU AI Act became generally applicable on 2 August 2026. Construction firms operating in or supplying the EU market now face an obligation to assess whether their AI tools fall within the high-risk category, particularly tools used for safety-critical applications, worker monitoring, or automated decision-making. That's not a theoretical concern for a sector where the Health and Safety Executive attributes 39 of 144 UK workplace fatalities annually to construction, despite the industry representing just 7.3% of the workforce.

The RICS professional standard on the responsible use of AI in surveying practice came into force in March 2026, and the report highlights its growing importance. Compliance-led adoption is real adoption, but it isn't the same as strategic adoption, and the difference will show up in margins.

Vendors Are Betting on the Graduation Problem

The software market has read the same data. Procore, picked up by Construction UK Magazine on 11 August, announced three tiered Digital Coworker packages built explicitly to "meet customers at every stage of AI adoption." The entry-level Starter Pack bundles five ready-to-use agents covering submittal review, RFIs, daily logs, contract review, and deep search. The Pro tier expands to 20 agents across bidding, safety, quality, and risk. Enterprise adds Agent Studio for building custom workflows.

The more interesting piece is a capability called Procore Skills, which lets organisations train AI agents on their own standard operating procedures, using plain-language prompts or existing documents, so the system applies company-specific ways of working rather than generic defaults. That addresses one of the core reasons pilots stall: an AI that doesn't know your subcontractor approval process or your quality hold-point schedule is not going to replace the project manager's institutional knowledge.

The Trademark Signal

The intellectual property market is reading the sector's uncertainty, too. AIBD analysis of IPO trademark data shows UK Class 19 filings, covering non-metallic building materials and serving as a proxy for product innovation activity across the built environment, came in at just 262 in Q3 2026, a fall of 44.3% against the prior period (AIBD analysis of IPO (TMD) data, as of August 2026). That kind of drop in materials IP activity, set against surging interest in construction software and AI tooling, suggests the sector's innovation energy has migrated from physical products toward digital process. Which is either a sign of maturity or a warning that firms aren't investing in the physical fabric of buildings while they're distracted by the dashboard.

What Actually Has to Change

The 37% of construction respondents planning to increase AI investment over the next 12 months is an encouraging figure. But investment intentions have outrun implementation before in this sector. The RICS report is clear: the question now is governance, data infrastructure, and professional supervision, not whether AI works.

For mid-size contractors in particular, the path is narrowing. Firms without AI-enabled workflows are already finding it harder to win higher-value work that demands auditable delivery records and Building Safety Act compliance. The pilot window is closing. What the RICS data confirms is that most of the industry is still standing at the threshold.