Eight Hours a Week. Every Week. And Most Firms Still Haven't Automated It.

New research from AI consultancy HeyBRB, published in the UK Admin Drain Report 2026, finds that UK tradespeople lose up to 8 hours per week to manual admin tasks. That's roughly 10 working weeks a year.

The estimated cost: GBP 17,000-25,000 in lost billable hours per business, per year. The biggest culprits are invoicing and chasing payments - precisely the tasks that are most amenable to automation. The finding that stands out most, though, is the spending data. Despite this level of loss, most small building firms spend less than GBP 40 per month on automation software.

HeyBRB has launched a free calculator to help tradespeople quantify their own admin backlog. The point isn't the tool - it's the framing. When admin loss is expressed in weeks and pounds rather than minutes, the conversation about automation changes.

Skanska UK president and CEO Katy Dowding has made a clear-eyed argument this week: a culture of inclusion makes construction sites safer, because people who feel belonging look out for one another. Her team is putting both principles into practice.

At 105 Victoria Street in London, Skanska has deployed the Schindler Robotic Installation System for elevator work - removing workers from one of the higher-risk tasks on any commercial build. Dowding is also explicit about AI's longer-term role in the skills crisis: using AI and robotics to take humans out of the most dangerous positions, including electrical work, heavy lifting, working at height, and people-plant interface situations.

The context matters. The UK construction industry remains its deadliest sector, with 35-50 fatalities per year over the past decade. Electrical accidents have risen 75% in recent years. These aren't abstract statistics for a firm the size of Skanska.

Chris Whyborn, Head of Cybersecurity Services (UK and Europe) at TUV SUD Business Assurance, has set out the regulatory and ethical landscape for AI deployment in facilities management - and it's more complicated than most FM teams currently appreciate.

The EU AI Act doesn't directly bind UK-only organisations, but equivalent UK frameworks are incoming. AI in FM can touch predictive maintenance, energy optimisation, health and safety compliance, and space management - all valuable applications. But the risks include data protection violations, bias in automated decision-making, and erosion of public trust if deployments aren't transparent.

Whyborn's core argument is that technical safeguards alone aren't enough. Organisations need to integrate ethical and legal compliance across the full AI lifecycle - from procurement and deployment through to ongoing monitoring and audit.

Siemens has built its industrial AI strategy around a problem the wider industry consistently avoids discussing: data quality. Gartner projects that 60% of AI projects unsupported by AI-ready data will be abandoned by end of 2026. In manufacturing and construction, the AI project failure rate already stands at 76.4%.

The culprits aren't the AI tools themselves. They're OT/IT integration gaps and IoT data quality. In construction terms, that translates directly: fragmented project data, inconsistent naming conventions, information locked in siloed systems, and sensor data that was never properly structured for analysis.

This is the uncomfortable backstory to every ambitious AI deployment story. The tools are real. The use cases are proven. But the data infrastructure that makes them reliable is absent in most construction firms. Siemens has been explicit about this in a way that most vendors are not, because acknowledging it is central to their strategy of building AI on top of properly structured industrial data.

Johnson Controls has published its 2026 AI and Digitalization in Facilities Management Report, based on a survey of 1,020 US business leaders and FM professionals. The headline numbers are useful context for UK FM teams thinking about where the sector is heading.

42% of business leaders and 47% of FMs currently use AI for predictive maintenance. 75% use workplace management technology for space management and planning - up from 70% last year. The conversation has shifted from return-to-office debates to building performance and employee productivity as primary outcomes.

But, the most telling finding is about friction: 33% of business leaders cite ease of integration as the top frustration with current AI systems. The tools work. Getting them to talk to the systems that already exist in buildings is the problem.

This maps directly to the data readiness issue above. AI tools for FM - predictive maintenance, energy optimisation, space planning - are proven. The barrier isn't capability. It's integration with legacy building management systems, inconsistent data formats, and the absence of structured data from older equipment.

Deloitte's latest analysis notes that AI intelligence is becoming "embodied" - moving from software into physical systems in factories, warehouses, and supply chains. By 2035, autonomous robots could cost as little as $13,000 (roughly GBP 10,000), making widespread deployment in construction economically viable for the first time.

BMW is already testing humanoid robots on assembly tasks. The construction robotics landscape is also maturing: this week's Construction Digital round-up of leading robotics companies includes Boston Dynamics Spot for site inspection, Dusty Robotics' FieldPrinter for AI-powered layout printing, and Hadrian X for autonomous bricklaying - all live, commercial products.

The shift from "$13,000 robot" as a headline to "what does this mean for site productivity" is probably five to eight years away for the UK construction mainstream. But the companies doing their first robotics pilots now will have a significant head start when the cost curve arrives.

A new benchmark from the ARC Prize team puts a useful frame on the current state of AI. ARC-AGI-3 tests whether AI systems can approach entirely new, interactive tasks without preparation. Humans solve 100% of them. Current frontier models score under 1%.

This isn't a flaw in the benchmark - it's the point. The test measures zero-preparation generalisation: the ability to encounter something genuinely new and work it out from first principles. Current AI models excel at tasks they've seen variations of before. They struggle when the environment is interactive, feedback is sparse, and there's no prior training exposure to draw on.

For construction professionals thinking about where to trust AI tools: this matters practically. Construction projects are full of novel situations - site conditions that weren't anticipated, RFIs that don't fit a standard template, programme impacts that cascade unexpectedly. AI tools are powerful for well-defined, repeatable tasks. They're not yet capable of the adaptive reasoning that experienced site managers apply to genuinely unfamiliar problems.