Chris M. / 13 August 2026

What if buildings were designed like products?

A building has a specification long before anyone gets on site. Drawings and a specification on a traditional job, Employer's Requirements and Contractor's Proposals on design and build, a schedule of works, and a contract binding the lot together. Anyone arguing that construction suffers from a shortage of documentation has never been handed a tender pack.

So the useful question is not what a product has that a building lacks. It is what a product gets to do that a building never does, which is get made a second time.

A product is perfected by repetition. The first run is wrong, the hundredth is better, and by the ten thousandth every defect has been designed out by people who watched the earlier ones fail. Tooling gets paid off across the run. A bespoke building gets built once, on a plot that will never be built on again, by a coalition of firms assembled for that job and disbanded at the end of it. There is no second run to learn from, and no amount of specification writing creates one.

Has anyone actually tried building houses like products?

Yes, with a great deal of money, and the results are a matter of public record.

The dominant reason is the same in three of the five, and it is not about design. Legal and General said it could not sustain the significant running costs of the large factory in Selby given insufficient orders, and that without the necessary scale of pipeline it was not sustainable to keep producing modules. TopHat cited a continued reduction in future pipeline. Ilke's administrators pointed at inflation and land supply while going out of their way to say the company had a strong product.

The other two are different, and the argument is weaker if you pretend otherwise. Urban Splash's administrators blamed design issues causing production defects whose rectification costs could not be passed on. Caledonian's collapse was attributed to cost inflation and a cyber attack during funding negotiations.

Nor was pipeline the only headwind. Volumetric unit costs never fell below traditional construction at the volumes actually achieved, so the learning curve that makes products cheap never arrived. Lenders, valuers and warranty providers stayed cautious about modular homes, which is a demand problem rather than a planning one. Then the 2022 rate shock removed the affordable housing and build to rent buyers who had been absorbing much of the output.

What survives all that is still substantial. Every one of these firms built a factory, and a factory has to be fed. The idea was not tested to destruction. The capital structure around it was.

Can you recall a building?

No, and this is the part of the product analogy that deserves more attention than it gets.

A product company that ships a systemic defect issues a recall. It knows which units are affected, because the production record says so, and it fixes them. Construction has no equivalent mechanism, and standardisation replicates a defect exactly as faithfully as it replicates a good detail.

Caledonian Modular went into administration in March 2022, and its business was bought by JRL Group the following month. In August 2023, seventeen months later, the Department for Education closed two schools the firm had built. One of them, Sir Frederick Gibberd College, said the DfE was no longer able to verify the structural safety of the building, hence its advice to close.

That is the strongest argument against the product framing, and it is not an argument for doing nothing. It is an argument for traceability: a record of what was built, to which revision, on whose instruction, that outlives the firm that built it. The idea belongs to Dame Judith Hackitt, whose 2018 review gave it a chapter and a name, the golden thread of building information. CIOB's Code of Quality Management endorses it for the delivery team and backs RIBA's proposed chain of custody for passing responsibility for quality through the design and construction team. The statutory duty followed in 2022.

Is construction's productivity problem real, or a measurement artefact?

Mostly real, and the honest answer matters because the convenient one is available.

ONS analysis of construction productivity puts the scale of it plainly. Measured against 1970, output per hour in the market sector as a whole reached an index of 250 by 2020. Construction reached 114.6. Construction's multi-factor productivity finished at 69, roughly a third below where it started fifty years earlier. On a separate measure, output per hour indexed to 1997, the construction of buildings sub-industry fell furthest of all.

The convenient defence is that this is a classification problem. There is something in it: ONS confirms that offsite manufacturing sits outside the construction industry as defined, and that firms specialising in modern methods may be allocated elsewhere. CIOB makes the same point in its productivity position, noting the gains from prefabrication will not show in construction's figures, and that delivering buildings more productively might even reduce measured construction productivity.

ONS's own view is that widening the scope would likely have little impact, though the aggregate it actually computes adds only architectural and engineering services, not offsite manufacturing. Take that for what it is worth. It is not a demonstration that the whole flat line is real, and it is not nothing either.

ONS does point at a candidate. Over 50% of construction's intermediate consumption is subcontracting expense, by far the highest of any UK industry, and ONS says this could cut either way: more specialisation and competition, but also more fragmentation and higher communication costs.

ICE arrived at the same territory from the infrastructure side. Its systems approach to infrastructure delivery proposes an eight component delivery model, and the eighth is that data oils your project: owners should define the vital data, agree how it is shared through the supply chain, and expect that collaborating around shared data increases productivity and enables integration.

Where does the repetition actually sit on a bespoke job?

Not in the building. In the process that delivers it.

A contractor may never build the same building twice, and does not need to. They run the same delivery process on every job. The same subcontractor RAMS arrive late and get reviewed against the same checklist, the same ITPs get signed off, the same RFIs and technical queries get chased, the same site diaries get written up at seven in the evening, the same subcontractor quotes get levelled.

The clearest case is the payment cycle, because it is mechanical and the consequence of getting it wrong is immediate. Applications arrive, the due date runs, the payment notice falls due, and the pay less notice has to be served before the final date for payment. Miss it and the notified sum becomes payable in full, which is how smash and grab adjudications happen. That is a six figure exposure created by a calendar and an inbox, and there is no judgement in the calendar part at all.

A firm running thirty jobs a year runs that process thirty times a year. That is a production run. It is the one genuinely repeating thing in a bespoke construction business, and almost nobody treats it as repeatable.

Does this hold for a volume housebuilder?

It holds differently, and the difference proves the point.

A volume housebuilder does repeat. The same house type goes up hundreds of times across dozens of sites, which is closer to a product than anything drawn for a single client. The obvious objection to this argument is that Barratt Redrow and Bellway solved it years ago.

They solved it for the building. They did not solve it for anything the building sits on.

A factory controls its inputs. A site controls almost none of them. Nobody standardises an archaeology find, a knotweed survey, a ward councillor, or a technical approval sitting with a highways authority. The factory analogy stops at the site boundary, and the administrative load on the far side of that boundary is where the hours go.

The bespoke element never leaves UK construction. It changes address. On a one-off building it sits in the design. On a housing site it sits in the land, the planning and the discharge of conditions.

Why do generic tools keep failing on this work?

Because a product built for someone else's process asks the contractor to adopt that process.

That is a fair trade when the work is standard and a bad one when the entire premise of the job is that it is not. The tool assumes the variability away, the team works around it, and within a quarter the real process has moved back into email and spreadsheets while the system holds a tidy fiction. That is why generic CRMs fail builders.

The research points the same way. The Enterprise Research Centre's study of technology adoption and productivity in UK SMEs found impact varies sharply by technology, that some combinations result in lower productivity with pairings such as VR and AR alongside automation showing diminishing returns, and that benefits are not automatic. Its conclusion is that careful selection and implementation of the right tools matter more than adopting more of them.

Construction's own numbers say the fit is not there yet. RICS research on AI in construction, covering more than 2,200 professionals globally in 2025, found 45% of organisations reporting no AI use at all and just 1% having scaled it across projects. The barriers reported were cost and effort, a lack of skilled people, and inconsistent approaches across supply chain partners. None of those are model problems.

What is the honest case against automating any of this?

It is stronger than most vendors admit, and it comes in three parts.

Adoption is wide but shallow. ONS figures on AI in UK businesses show self-reported use among businesses with ten or more employees rising from around 12% to around 35% since late 2023, but only 10% of those using it report using it extensively, and the average adopter uses barely more tools than when the series began.

Reported benefit and measured money are different things. DSIT's AI adoption research surveyed 3,500 UK businesses, around one in six of which use AI. Among the 700 that do, 75% reported improved workforce productivity while 77% reported no change in revenue yet. Both numbers are self-reported perceptions rather than audited outcomes.

Pilots fail often. MIT's NANDA initiative reported in 2025 that 95% of organisations were getting zero return from generative AI pilots, though that rests on 153 survey responses, 52 interviews and a review of 300 disclosed initiatives, which is a small and self-selected base for an organisation-level rate, and the paper is preliminary and not peer reviewed. Gartner forecast in July 2024 that at least 30% of such projects would be abandoned after proof of concept by the end of 2025, citing poor data quality, inadequate risk controls, escalating costs and unclear business value. That was a forecast rather than a measurement, and its deadline has now passed without a published retrospective.

What both describe is an integration failure rather than a technology failure, which is where RICS arrived from the opposite direction. The response is not to skip the work but to run it properly, which means a pilot with real scope and numbers agreed before it starts.

Does the arithmetic work on a bespoke job?

The prize is error, and error in UK construction is expensive.

The Get It Right Initiative puts the total cost of avoidable error at between 10% and 25% of project cost, somewhere between £10bn and £25bn a year, with its own research pointing to around 21%. That research dates from 2015 and 2016, and published estimates of rework cost vary widely, so treat the precision carefully. Even the low end is a large number.

GIRI's own ranking of root causes puts inadequate planning first, then late design changes, then poorly communicated and poorly coordinated design information. That is worth sitting with, because it cuts against the comfortable reading. Much of it is not design quality so much as design information failing to move cleanly: a wrong revision on site, an instruction nobody recorded, a variation under JCT or a compensation event under NEC that became a dispute because nothing contemporaneous was written down.

Mark Farmer wrote the 2016 review that told the industry to modernise or die, and served as the government's MMC Champion for Homebuilding from 2019 to 2023. Even that 2016 report, remembered for its ultimatum, argued that the culture of data silos within the industry needed breaking.

Asked about the sector in September 2024, after the factories had closed, he pointed to hybrid pre-manufacturing such as panellised and podded solutions as the direction of travel. He also pointed to the latent opportunity of building traditionally but more productively, without necessarily changing current techniques, with standardised designs and pattern books as the easiest route to it.

Standardised information, not industrial plant. Keep the architecture bespoke, keep the plot and the planning and the client's ambition bespoke, because that is the value being sold. Then stop rebuilding the machinery around it from a blank page thirty times a year, because that part was never bespoke to begin with.

That is the work AI Metric does: drafting from a contractor's own document library rather than from nothing, holding the sequence when the programme tightens, and putting the records somewhere the business can actually reach. Construction Metric is the product version of the same idea, turning what a site team already sends during the day into a structured daily record. One caveat worth stating plainly, because it decides whether any of it is appropriate: a WhatsApp capture route suits firms whose site information genuinely lives there, and a Tier 1 with a client-mandated common data environment and its own disclosure rules is a different conversation. Which one you are is the first question, not the last.