Question one
YesCan AI actually do this work?
Eight selected public cases sit behind this workflow, across contractors, developers and industrial clients in the UK, Europe, the United States and Japan.
38thMove · Workflow
Construction progress & reality capture.
Using cameras, drones and AI to record what has actually been built, compare it against the plan, and catch problems while they are still cheap to fix. This is one of the better-evidenced workflows in the library: named firms have reported operational use, and the gains they report are large enough to be worth testing.
Start here
The three questions worth asking.
Question two
5 of 8Has anyone really done it?
Five of the eight accounts report operational or scaled use rather than trials, including NCC, Mace, Vinci, Hensel Phelps and Intel.
Question three
Six figuresWhat benefit do they report?
The largest single claim is $342,000 of labour on one airport project. Others report saved hours or reduced rework cost. They measure different things, so each is named and attributed below.
The evidence
What firms actually report.
This guide examines eight selected accounts, rather than every construction-related entry in the database. Read the overall evidence summary.
These are the headline figures from the public sources, shown with who said them. They are not audited results, and they are not a forecast for your projects. They report operational use across several named firms and outcomes worth testing; timing and financial claims retain their individual limitations.
Hensel Phelps · Track3D
$342kLabour saved on one airport project.
Reported alongside 2,964 hours and three major reworks avoided at San Francisco International. The figures come from the technology supplier, not an audit.
See card EC-092 →NCC · Buildots
70%Reduction in manual progress reporting.
The source does not establish that 70% is a reduction in working time. It also claims 2.3 times more tasks completed on time. Supplier-published.
See card EC-029 →Mace · Buildots
4,200Hours described inconsistently.
The supplier calls 4,200 hours both saved work and a gap in task completion. The meaning is unresolved; this is not treated as a measured time saving. The project is unnamed.
See card EC-030 →Vinci Construction UK · OpenSpace
6,000Documentation hours saved a year.
A supplier estimate, across 45 projects in the supplier’s account. The breadth of use is the more reliable signal here.
See card EC-042 →Kajima · drone and AI
75%Cut from one materials-tracking cycle.
One task on one named project, reduced from about two hours to 30 minutes. Confirmed by the contractor rather than a supplier.
See card EC-082 →Intel · Buildots
4.3%Reduction in rework cost per plant.
A supplier-reported rework-cost percentage, not a reduction in total construction cost. No cash amount or independent audit.
See card EC-010 →Seven of the eight cases rely on figures published by the technology supplier. That does not make them wrong, but the public sources do not independently check the calculation, the starting point or the cost of putting it in. Repeated supplier examples are not independent replications of the reported benefit.
What to do about it
How to find your own number.
The evidence supports a small, well-defined test where progress information is a real problem. Agree the result your board would need before starting.
- Pick one project and one thing to improve, such as hours spent producing progress reports, and measure it today.
- Run a limited comparison with a clear owner, an agreed definition of acceptable output, a pass mark and an end date. Include capture, checking and correction time.
- Scale only if the result clears the threshold after tool, integration, training and support costs. Agree who keeps the benefit.