AI-Powered RAMS Software — How AI is Changing Health and Safety Documentation
Health and safety documentation has always been time-consuming. Writing a professional Risk Assessment and Method Statement from scratch — identifying every relevant hazard, researching the applicable UK regulations, describing step-by-step methodology and ensuring nothing critical has been missed — can take an experienced site manager several hours. For a small contractor with multiple jobs running simultaneously, that time pressure often leads to documents being rushed, copied from previous jobs or completed with gaps that could matter when things go wrong.
AI is changing this. Not by replacing the competent person — the law does not allow that and nor should it — but by doing the heavy lifting on the parts that are time-consuming and systematic, leaving the genuinely human judgement calls to the people who are best placed to make them.
This article explains exactly how the AI in RAMS-flow works, what it does well, what it deliberately leaves to you, and why that balance matters for legal compliance.
What the AI Actually Does
When you describe your scope of work in RAMS-flow — for example, "installation of a temporary bar structure at an outdoor event venue, including delivery of equipment by 18-tonne HGV, assembly by a team of four, connection to venue power supply and operation throughout a two-day event" — the AI does not simply search a database of generic risk assessment templates. It analyses what you have described and generates content specific to those activities.
Hazard Identification
The AI identifies the hazards that are genuinely relevant to the described scope. For the example above it would identify vehicle movements and pedestrian interaction during delivery, manual handling during assembly, working at height if any elevated structure is involved, electrical connection and temporary power, crowd management during operation, and emergency evacuation from a temporary structure. It does not pad the document with irrelevant hazards that could apply to any job — a common problem with template-based approaches.
Control Measures Referenced to UK Regulations
For each hazard the AI generates control measures that reflect current UK health and safety legislation. The system has been trained on and cross-references specific regulations including:
- LOLER 1998 — for any lifting equipment, vehicle tail lifts or MEWPs involved in the work
- PUWER 1998 — for work equipment, machinery and tools
- Work at Height Regulations 2005 — applying the correct hierarchy of avoid, prevent, mitigate
- COSHH 2002 — for any hazardous substances, cleaning products or fuels
- Manual Handling Operations Regulations 1992 — applying the TILE assessment framework
- CDM 2015 — for construction activities including temporary structures
- Gas Safety (Installation and Use) Regulations 1998 — for any gas work
- Electricity at Work Regulations 1989 and BS 7671 — for electrical installations
- Confined Spaces Regulations 1997 — for any confined space entry
- Regulatory Reform (Fire Safety) Order 2005 — for hot work and occupied premises
The AI identifies which of these regulations apply to the described scope and incorporates the relevant requirements into the control measures — without adding irrelevant regulatory references that do not apply to the work.
Methodology Generation
The AI generates structured methodology sections — numbered, sequential steps describing exactly how the work will be carried out. These are specific to the described scope, not generic procedure headings with placeholder text. Each step references the relevant safety measures, equipment requirements and regulatory considerations for that stage of the work.
PPE and Training Requirements
Based on the activities described, the AI suggests appropriate personal protective equipment referenced to EN standards, and the training qualifications required for the specific work — IPAF for MEWP operation, Gas Safe registration for gas work, IPAF or PASMA for scaffolding, and so on. It does not generate a generic PPE list — it identifies what is genuinely required for the described activities.
What We Have Deliberately Programmed the AI NOT to Do
This is where RAMS-flow's approach differs from simply using a general-purpose AI tool to generate health and safety documents. We have spent significant time identifying the mistakes that AI makes when generating H&S content and explicitly training the system to avoid them.
It Does Not Prescribe Mandatory Annual Retraining
A common error in AI-generated H&S documents is stating that annual refresher training is mandatory for all activities. HSE guidance under ACOP L117 (for forklift operation) and general training best practice is clear — refresher training should be competence-based and needs-led, not automatically annual. RAMS-flow's AI has been explicitly programmed to generate competence monitoring language rather than blanket annual retraining requirements.
It Does Not Specify Fixed Exclusion Zone Distances
Stating that a "5-metre exclusion zone" is universally safe for forklift operations, or a "2-metre exclusion zone" for any particular activity, presents a fixed distance as a universal standard when it is not. The required exclusion zone depends on the specific equipment, load dimensions, turning radius, site layout and operating conditions. RAMS-flow's AI generates site-specific exclusion zone language that requires the user to determine the appropriate distance for their specific situation.
It Does Not Average Risk Scores
Averaging hazard scores to produce an overall risk rating is misleading and dangerous. A document with one Red-rated hazard and five Green-rated hazards does not have an overall Amber rating — it has an overall Red rating because the highest individual hazard governs. RAMS-flow calculates and displays the highest individual hazard score as the overall rating, not an average.
It Does Not Confuse Pre-Use Checks with Statutory Examinations
For lifting equipment, a common AI error is generating "annual inspection" as a catch-all control measure without distinguishing between the operator's daily pre-use checks (which are not inspections in the legal sense), planned maintenance (which is schedule-based) and the statutory thorough examination required under LOLER 1998 (which must be carried out by a competent person and produce a written report). RAMS-flow's AI has been trained to make this distinction explicitly.
It Does Not Reference IIRSM as a Forklift Competence Standard
IIRSM (the International Institute of Risk and Safety Management) is a professional membership body, not an equipment operator certification scheme. Referencing IIRSM certification as evidence of forklift competence is incorrect. RAMS-flow's AI references ACOP L117 and the three-stage training requirement (basic, specific-job and familiarisation) as the correct standard for forklift operator competence.
It Does Not Present PPE as a Primary Control
Personal protective equipment is always the last resort in the hierarchy of control — after elimination, substitution, engineering controls and administrative controls have been considered. RAMS-flow's AI generates control measures in the correct hierarchy order and does not lead with PPE as the primary or only control for a hazard.
What the AI Leaves to You — and Why
The AI handles the systematic and time-consuming parts of RAMS document creation. The parts that require genuine human judgement are deliberately left to the competent person reviewing the document.
Site-Specific Conditions
The AI does not know the specific layout of the site, the condition of the access route, the proximity of other trades, the history of incidents at that location or the specific capabilities of the individuals who will carry out the work. These site-specific factors must be assessed and incorporated by the person who has actually visited the site or has direct knowledge of the working environment.
Equipment-Specific Information
The AI can identify that a LOLER thorough examination is required for lifting equipment and generate the appropriate control measure. It cannot confirm the date of the last thorough examination for your specific machine, the SWL marked on your specific equipment or whether your specific equipment has any current defects or limitations. These details must be added by the person with knowledge of the actual equipment being used.
Personnel Competence Confirmation
The AI can identify that Gas Safe registration is required for gas work and generate the appropriate control measure. It cannot confirm that the specific individuals named in the document hold the correct category of Gas Safe registration for the specific work type, or that their IPAF card covers the specific MEWP category to be used. Competence confirmation must be completed by the responsible person.
Risk Score Judgement
The AI generates suggested risk scores based on the described activities and stated control measures. These are starting points, not final values. The frequency, probability and severity scores must be reviewed by the competent person in the context of their specific site, equipment, personnel and working conditions. A score that is appropriate for one site may not be appropriate for another carrying out nominally the same task.
Final Approval
Every AI-generated document in RAMS-flow is placed in an Awaiting Review status before it can be used. It cannot be linked to a RAMS document, printed or submitted to a client until a competent person has reviewed and approved it. This is not a feature that can be bypassed — it is built into the workflow because the legal duty to produce a suitable and sufficient risk assessment sits with the employer, not with the software.
The AI RAMS Review
Beyond generation, RAMS-flow includes an AI-powered document review that analyses completed RAMS documents and risk assessments for quality and compliance issues. This review checks for the common errors described above — risk averaging, mandatory annual retraining language, fixed exclusion zone distances, missing emergency arrangements, logical contradictions in control measures and hazards that should have been included given the scope of work.
The review produces a score out of 100 and identifies specific improvements with their severity — critical issues that must be addressed before the document is suitable for use, warnings that should be reviewed, and suggestions that would improve the document's quality. Each improvement is linked to the specific section of the document where the issue was found.
This means that even where the initial generation misses something or a user has produced a document manually, the AI reviewer acts as a second line of quality assurance before the document is used on site.
Why the Human Always Decides
The legal framework in the UK is clear. The Management of Health and Safety at Work Regulations 1999 require the employer to carry out a suitable and sufficient risk assessment. The duty sits with the employer — a natural or legal person — not with software. No AI system can carry legal responsibility for the adequacy of a risk assessment.
This is not a limitation to work around — it is the correct approach. AI is exceptionally good at systematic tasks: identifying hazards from a large knowledge base, cross-referencing regulations, generating structured methodology, spotting inconsistencies. Humans are exceptionally good at contextual judgement: assessing site-specific conditions, evaluating individual capability, making proportionate decisions about risk in real-world situations where the context matters.
RAMS-flow is designed to give each the tasks it does best. The AI generates a comprehensive, regulation-aware starting point in minutes. The competent person applies their knowledge of the specific site, specific equipment and specific people to produce a document that is genuinely suitable and sufficient for the actual work. The result is better documentation, produced faster, with less risk of missing something important.
- RAMS-flow's AI generates hazard identification, control measures, methodology and PPE specific to your described scope of work — not generic templates
- The AI cross-references UK regulations including LOLER, PUWER, WAH 2005, CDM 2015, COSHH and others relevant to the described activities
- Common AI errors have been explicitly programmed out — no risk averaging, no fixed exclusion zones, no mandatory annual retraining, no PPE as primary control
- Site-specific conditions, equipment details, personnel competence and final approval always remain with the competent person
- Every AI-generated document requires competent person review and approval before it can be used
- The AI reviewer checks completed documents for quality and compliance issues as a second line of assurance
Can I submit an AI-generated RAMS document to a principal contractor?
Yes — provided it has been reviewed and approved by a competent person. Principal contractors are interested in the quality and completeness of the document, not how it was produced. An AI-generated document that has been properly reviewed, adapted to the specific site and approved by a competent person meets the same standard as a manually produced one — and often exceeds it because the AI is less likely to miss hazards or produce generic control measures.
What happens if the AI gets something wrong?
This is exactly why every AI-generated document in RAMS-flow requires competent person review before it can be used. The AI generates a comprehensive starting point — the competent person catches anything that does not reflect the specific site conditions, corrects any scores that do not reflect the actual residual risk, and adds information the AI could not know. The review process is the safety net. The AI generates; the competent person decides.
Is AI-generated content legally compliant?
AI-generated content that has been reviewed and approved by a competent person is legally compliant with the requirement to carry out a suitable and sufficient risk assessment under the Management of Health and Safety at Work Regulations 1999. The legal duty sits with the employer — the AI assists in fulfilling that duty but does not replace the competent person's responsibility to ensure the document accurately reflects the actual risks of the actual work.
How is RAMS-flow's AI different from using ChatGPT?
General-purpose AI tools like ChatGPT are trained on broad datasets and will generate plausible-sounding health and safety content, but they do not have the specific prohibitions and corrections that RAMS-flow's AI has. They may prescribe annual retraining as mandatory, specify fixed exclusion zone distances as universally safe, average risk scores, or reference incorrect competence standards. RAMS-flow's AI has been specifically configured for UK health and safety documentation with explicit rules that override these common errors.
Which industries does AI RAMS generation work for?
RAMS-flow's AI generation works for any UK industry — construction, events and entertainment, facilities management, hospitality, food production, logistics and warehousing, aviation ground handling, utilities and engineering. The AI identifies which regulations apply based on what you describe in the scope of work, so a roofing contractor gets WAH 2005 hierarchy applied automatically, a gas engineer gets Gas Safety Regulations requirements, and an events company gets temporary structure and crowd management considerations. The generation is scope-driven, not industry-limited.