Why it matters
Without AI, trend detection is manual and slow, near misses may be underreported, buried services stay undetected, and corrective actions are delayed, increasing risk on trench safety and UXO management.
From design to safety
Design
AI-supported data collection design
The system reads the incident database, RIDDOR entries, and near-miss forms to shape a risk taxonomy and a baseline reporting schema for written schemes of investigation; it defines data dictionaries, field validators, and form templates for consistent capture. It models historical event trees to test warning thresholds against future data.
- accident reports
- near misses
- RIDDOR
- investigations
Build
In-field risk logging and checks
During trench setup and excavation, it reads daily safety logs, service maps, contamination notes, gas readings, and weather impacts to trigger immediate investigations, adjust watchlists, and update trench design matrices; it supports automatic geolocation tagging and service-detection prompts.
- accident reports
- near misses
- RIDDOR
- investigations
Operation
Ongoing trend monitoring and reporting
The platform ingests ongoing incident logs, context sheets, client reports, and post-excavation notes to produce risk-adjusted trend dashboards, alerts for recurring patterns, and assignment of corrective actions with accountability trails.
- accident reports
- near misses
- investigations
- trends
Maintenance
Data quality and archive care
It validates records, reconciles post-excavation findings with HER entries, and flags missing reports for post-excavation assessment; it periodically re-indexes archive PETRA items and validates scans against original context sheets.
- investigations
- trends
- RIDDOR
- corrective actions
Safety
Safety compliance and sign-offs
It screens trench safety, buried services, contamination, and UXO considerations; qualified persons review actions before planning approvals are granted; it maintains an auditable trail for regulator reviews and client reporting.
- accident reports
- corrective actions
- investigations
- RIDDOR
Part by part
- 1
accident reports
- Watch for
- Inconsistent severity coding, missing date fields, or delayed reporting of a fault or injury.
- AI action
- Flag late or underreported incidents and auto-assign investigators.
- 2
near misses
- Watch for
- Patterns of near-miss events across trench faces, services, or spoil heaps.
- AI action
- Cluster near misses into risk categories and prompt preventive actions.
- 3
RIDDOR
- Watch for
- Non-compliance with statutory reporting timelines or incomplete cause analysis.
- AI action
- Validate completeness and trigger escalation to a qualified person.
- 4
investigations
- Watch for
- Investigation gaps, delayed root-cause analysis, or unresolved corrective actions.
- AI action
- Recommend targeted root-cause methodology and assign responsible party.
- 5
trends
- Watch for
- Emerging persistent risk patterns across sites or weeks.
- AI action
- Generate weekly trend reports and highlight high-risk corridors.
- 6
corrective actions
- Watch for
- Unimplemented actions or actions lacking due dates.
- AI action
- Track action owners, deadlines, and verification of completion.
Worked example
Catching an unmarked service before trenching
Illustrative scenario with sample data.
During a development-led trenching program, unexpected utilities and contaminated soils were found; early detection avoided strikes and contamination spread.
Signal in
incident_id=INV-1082; date=2026-08-14; type=near miss; location=Trench 3; risk=buried service; factors=poor service maps, dust exposure; fieldsCompleted=yes
System out
alert generated: risk category buried service; investigation assigned to safety officer; corrective action: verify utility maps and escalate to client HER for service verification.
Human sign-off: Project manager reviews the investigation plan and signs off on the corrective actions before field teams proceed.
How to roll it out
Phase 1
Weeks 1 to 4
Establish data schema, ingest legacy records, and calibrate risk scoring with safety culture leaders.
Phase 2
Weeks 5 to 12
Deploy in-field prompts, dashboards, and escalation workflows; train site supervisors and safety officers.
Phase 3
Weeks 13+
Mature monitoring, continuous improvement loops, and archive deposition aligned with HER procedures.
Where people stay in charge
- Decisions about clinical or legal reporting must be signed off by a qualified safety officer or the client’s heritage advisor, not delegated to automated systems.
- No determination of fault, liability, or closure of investigations may be made by AI; only the qualified person may decide.
- All data handling and retention decisions must comply with regulatory requirements and project-specific data management plans.
Questions owners and crews ask
How does AI help with accident reporting timing on site?
AI flags missing or late reports, prompts responsible staff to complete entries, and routes notifications to the safety officer for timely handling.
Can AI distinguish between minor near misses and serious incidents?
Yes. It analyzes severity fields, context, and recurrence patterns to categorize incidents and prioritize investigations appropriately.
Who approves corrective actions suggested by the system?
The site safety officer and project manager jointly approve corrective actions, with sign-off by the client heritage manager when required.
How does this integrate with RIDDOR reporting and HER records?
The system validates completeness, formats submissions, and triggers escalation to ensure timely RIDDOR reports and alignment with archive entries.