Every construction project solves problems the next one will face. Yet much of that knowledge remains trapped in records, software and individual experience, then disperses with the project team.
Buildots, a construction technology company that uses computer vision and AI to track progress, is trying to change that through its Intelligence Lab. The research hub analyses aggregated and anonymised project data to produce benchmarks and insights that contractors can apply elsewhere.
Its first findings show the potential. In data-centre construction, Buildots reports a 20% to 50% gap between planned and actual weekly output for mechanical, electrical and plumbing (MEP) work. The final 20% of an activity takes 27% of its duration, while the best-performing MEP teams work up to three times faster than average. Because the figures come from Buildots’ own project data, their value will depend on how representative and comparable the underlying work is.
Collecting information is only the first step. The harder task is turning it into knowledge the organisation can reuse.
Why project knowledge disappears
Construction companies have spent years digitising projects. But temporary teams, changing supply chains and fragmented systems make it difficult to retain what each project teaches.
Mark Coates, vice-president of infrastructure policy advancement at Bentley Systems, says the industry is “very good at creating data”. It is less effective at treating that data as “a long-term asset”, he adds.
Coates says a typical major project can generate “enormous quantities” of design, commercial, programme, asset and operational information. Much of it remains locked within applications, project teams, contracts or supply-chain organisations. When a project finishes and the team disperses, he says, “a significant proportion of that accumulated knowledge effectively disappears with it”.
Contractors are beginning to build the systems and skills needed to prevent that loss.
AI does not correct poor information management
Balfour Beatty is developing AI-powered ‘smart agents’ that integrate data to support on-site decision-making. It also committed £7.2m to Microsoft 365 Copilot in 2025 as part of its wider digital strategy. The first smart-agent trial is reviewing inspection and test plans on the A9 Dualling project in Scotland.
Skanska UK has expanded its Data Academy and introduced an AI apprenticeship to help employees make faster, data-informed decisions. Such programmes matter because tools cannot create organisational memory unless staff know how to question, interpret and reuse their output.
From data to decisions
The Construction Leadership Council is pushing the sector in the same direction. Its Information Management Initiative aims to improve data practices across the built environment, while Construct AI is developing a non-commercial lab in which organisations can explore AI use cases together.
Buildots’ platform connects building information modelling, schedules and site data to give project teams a common view of progress. The wider challenge is making that information consistent enough to compare across projects.
That information must be reliable.
“AI does not correct poor information management,” says Coates. “In many circumstances it simply exposes it more quickly.”
Useful AI depends on common standards, reliable data, clear provenance and systems that allow information to move between projects and organisations.
Technology is only part of the work. Contractors must also change how people use information and make decisions. AI can make patterns easier to identify. It cannot decide which ones matter.
The real prize is using information that already exists to make better decisions earlier
“Construction productivity will not fundamentally change because we automate the writing of another report or save a few minutes producing meeting notes,” he says. “The real prize is using information that already exists to make better decisions earlier.”
In practice, that means identifying programme risk before it becomes a delay, detecting cost pressure or understanding why an intervention succeeded or failed elsewhere.
The Balfour Beatty and Skanska initiatives do not yet amount to a shared organisational memory. But they show some of the capabilities it requires: integrated data, repeatable checks and employees able to use AI critically. Without those changes, digital tools can add another layer of administration rather than remove one.
Learning across projects
The larger opportunity comes when contractors can compare performance across many projects rather than review each one in isolation.
That is the proposition behind the Buildots Intelligence Lab. By aggregating anonymised data, it aims to identify recurring problems, reliable interventions and the points at which performance varies most.
The question then changes from whether a project is on schedule to why comparable projects perform differently – and what the better-performing teams do differently.
The test is whether this knowledge improves the economics of construction. Coates says organisations should judge AI-driven productivity by outcomes. These include fewer design iterations, earlier decisions, less rework, more reliable programmes, greater cost certainty and better use of scarce engineering and delivery expertise.
Construction will always involve uncertainty, judgement and people. But each project should leave behind reliable evidence about what worked, what failed and why, so the next one starts with more knowledge than the last.
The advantage is making sure that when one project ends, the lessons it leaves behind do not.
Every construction project solves problems the next one will face. Yet much of that knowledge remains trapped in records, software and individual experience, then disperses with the project team.
Buildots, a construction technology company that uses computer vision and AI to track progress, is trying to change that through its Intelligence Lab. The research hub analyses aggregated and anonymised project data to produce benchmarks and insights that contractors can apply elsewhere.
Its first findings show the potential. In data-centre construction, Buildots reports a 20% to 50% gap between planned and actual weekly output for mechanical, electrical and plumbing (MEP) work. The final 20% of an activity takes 27% of its duration, while the best-performing MEP teams work up to three times faster than average. Because the figures come from Buildots’ own project data, their value will depend on how representative and comparable the underlying work is.
