Sectors: tunnelling
AI for tunnelling contractors
Tunnelling carries the heaviest documentation load in UK construction, because almost everything underground is governed, monitored and recorded. Aldbry produces and checks those records and keeps every engineering and safety judgement with the people qualified to hold it.
The documentation load
The documentation load in tunnelling
A regime where the records are part of the control measure, not an administrative afterthought.
01
BS 6164 documentation
Health and safety in tunnelling brings its own expectations for method, ventilation, fire precautions and emergency arrangements, each with records that have to agree with one another.
02
Shaft sinking and segment RAMS
Separate method statements for sinking, for ring build and for the interfaces between them, usually from more than one subcontractor.
03
Ground movement and monitoring
Instrumentation readings, trigger levels and the response actions attached to them. The value of the record is in it being current.
04
Emergency and rescue arrangements
Escape, rescue and firefighting arrangements referenced across several documents, which is exactly where a gap goes unnoticed.
Where AI helps
What comes off your team
Repetitive, document-shaped work, applied inside the way your firm already works.
Produced and checked automatically
- Cross-check that trigger levels and response actions state the same figures in every document that mentions them
- Flag an activity with no matching rescue or emergency arrangement in the pack
- Check site-specific method statements against your own templates
- Build the daily record from what the shift captured underground
Which modules are live
Site Diaries, RAMS Assurance and Subcontractor Onboarding are live on UK construction sites with a small number of contractors.
Inspections and Quality, Progress Reporting and Handover Packs are in development. We do not describe a module as live until it is running on a real site.
Where the line sits
What stays with competent people
AI drafts, checks and flags. People decide, and the accountability does not move.
- Tunnel and shaft designEngineering design and its checking belong to qualified engineers.
- Trigger level responseDeciding what happens when an instrument reads a trigger is a judgement, not a lookup.
- Rescue planningSpecialist work, made by people who will be accountable for it.
- Every CDM 2015 dutyUnchanged.
Common questions
Questions we hear in tunnelling
How can construction firms use AI to cut admin?
Construction firms cut admin with AI by applying it to the documents they already produce every week, not by replacing the people who produce them. The work that responds best is repetitive and document-shaped: site diaries written from spoken notes, subcontractor packs checked for missing insurances and expired cards, RAMS cross-checked for conflicts and gaps before a competent person signs them off, and progress records assembled from what was already captured on site. Aldbry applies AI inside your existing workflows rather than asking you to move to a new platform, and a competent person reviews and signs off every output. If you want to size the problem before spending anything, the Construction Admin Cost Calculator gives you a headline figure in a couple of minutes.
How do you check a subcontractor RAMS pack?
Read the pack as one submission and compare the documents against each other, because that is where the real problems sit. The checks that matter are the figures that must agree across documents, plant and substances that need a matching assessment somewhere in the pack, controls that are missing from the document where they count, references and revisions that do not line up, and sign-offs that are absent or out of date. Aldbry does this as a managed service and inside Corvleo, and the output is an evidenced review sheet rather than a verbal opinion, so you can show what was checked. The decision to accept the pack stays with your competent person.
Is it safe to use AI on RAMS and safety documents?
Yes, provided AI checks and flags while competent people write, review and sign off. AI is good at finding missing sections, inconsistent figures, expired certificates and requirements that have not been addressed, and it does not get tired at the end of a long pack. It is not a competent person and should never be treated as one. Aldbry designs every workflow so that a named person approves the output, and Corvleo presents findings for review rather than issuing decisions. Accountability under CDM 2015, LOLER 1998, COSHH 2002 and your other duties stays exactly where it sits today.
Can it check monitoring and trigger level records?
It checks that the documents agree with each other, which is where trigger levels most often go wrong. If a trigger value or a response action is stated one way in the monitoring plan and differently in a method statement, that conflict is flagged with both documents quoted. What the reading means on the day, and what to do about it, is an engineering judgement that stays with your people.
Our packs run to hundreds of pages. Is that a problem?
Length is the argument for cross-checking rather than against it. A pack that takes a day to read properly rarely gets a day, and the errors that matter are contradictions between documents that each look correct alone. The review reads every document against the others and returns findings grouped by severity, each naming the documents involved and quoting what they say, so a person can go straight to the two pages that disagree.
Next step
Send one tunnelling pack and see what a cross-check finds.
The free RAMS pack check reviews one subcontractor pack and sends the findings back. Or book a short call about the records your work generates each week.