AI-Powered COSHH Assessments for Construction Sites

COSHH assessments are one of the most time-consuming compliance tasks on a construction site. For every hazardous substance and dust-generating process, you need a documented assessment covering identification, risk evaluation, control measures, health surveillance requirements, and emergency procedures. On a typical project, that can mean dozens of individual assessments. AI-powered COSHH assessment tools are changing this by generating comprehensive, regulation-compliant assessments in minutes rather than hours. This guide explains how they work and how to use them effectively.

The COSHH Challenge in Construction

The Control of Substances Hazardous to Health Regulations 2002 require employers to assess and control exposure to hazardous substances. In construction, the range of substances is vast:

• Silica dust from cutting concrete, brick, stone, and morite
• Wood dust from cutting, sanding, and routing timber
• Cement (alkaline burns, chromium VI sensitisation)
• Welding fumes (manganese, chromium, nickel)
• Solvents in paints, adhesives, and sealants
• Diesel exhaust emissions from plant and vehicles
• Asbestos in refurbishment and demolition (separate regulations but similar assessment process)
• Biological agents (leptospirosis from rat urine in excavations, pigeon droppings in refurbishment)

Each of these requires a specific assessment that considers the work activity, exposure route, exposure duration, existing controls, and the adequacy of those controls. Creating these assessments manually -- reading safety data sheets, cross-referencing workplace exposure limits, identifying appropriate controls from the hierarchy -- takes significant time and expertise.

The result? Many sites either use generic assessments that do not reflect their actual working conditions, or they simply do not produce assessments at all. Both approaches leave workers at risk and the organisation exposed to enforcement action.

How AI COSHH Assessment Tools Work

AI COSHH assessment tools like Site Manager AI use trained language models that understand UK COSHH regulations, HSE guidance documents (EH40, COSHH Essentials, INDG136), workplace exposure limits, and construction-specific hazard profiles. When you input a substance or activity, the AI generates a complete assessment that includes:

1. Substance identification -- the hazardous substance(s) involved, their classification under CLP Regulation, and their hazard statements
2. Exposure assessment -- how workers are exposed (inhalation, skin contact, ingestion), the likely exposure level relative to the WEL, and the exposure duration
3. Health effects -- what harm the substance can cause, both acute (immediate) and chronic (long-term)
4. Control measures -- recommended controls following the hierarchy: elimination, substitution, engineering controls, administrative controls, and PPE
5. Health surveillance requirements -- whether health surveillance is needed and what form it should take
6. Emergency procedures -- first aid measures, spill response, and fire safety considerations
7. Review schedule -- when the assessment should be reviewed

Practical Example: Silica Dust Assessment

To illustrate how an AI COSHH tool works in practice, consider a silica dust assessment for concrete cutting:

Input: "Cutting concrete blocks with a table saw, external location, 4 hours per day"

The AI would generate an assessment that identifies respirable crystalline silica (RCS) as the hazardous substance with a WEL of 0.1 mg/m3, notes that dry cutting concrete blocks can generate airborne RCS concentrations 10-50 times the WEL, and recommends specific controls: water suppression on the table saw (reduces dust by 85-95%), positioning the saw so that prevailing wind carries dust away from the operator and other workers, RPE (minimum FFP3 disposable respirator) as a secondary control, and health surveillance in the form of annual spirometry testing.

The assessment would reference HSE guidance CIS36 (Silica) and the Construction Dust Partnership guidelines. It would note that dry cutting of concrete is subject to HSE enforcement action and is never acceptable regardless of RPE provision.

Generating this assessment manually would require reading the safety data sheet for the specific concrete product, consulting EH40 for the WEL, reviewing HSE guidance CIS36 and INDG463, and structuring the findings into a compliant assessment format. An AI tool produces the same output in under a minute.

Quality Assurance: Reviewing AI-Generated Assessments

AI-generated COSHH assessments are a powerful starting point, but they require human review before implementation. Key checks include:

• Site specificity -- the AI generates assessments based on the information you provide. If your site has unusual conditions (confined spaces, proximity to occupied buildings, poor ventilation), add these to the input or edit the output accordingly.
• Product accuracy -- if you are using a specific commercial product (rather than a generic substance), check the assessment against the product's safety data sheet. The AI works from general data; the SDS has product-specific information.
• Control feasibility -- the AI may recommend controls that are not practical on your site. For example, local exhaust ventilation might be recommended for an activity that takes place in an open area where it is not feasible. Adjust controls to reflect what is actually achievable while maintaining adequate protection.
• Regulatory updates -- check that the WELs and regulatory references are current. WELs are periodically revised (the wood dust WEL was reduced from 5 to 3 mg/m3 in 2020), and the AI's training data may not reflect the very latest changes.

Integrating COSHH with Other Safety Documentation

COSHH assessments do not exist in isolation. They should be linked to related safety documents:

• Risk assessments -- the COSHH assessment should be cross-referenced in the task-specific risk assessment
• Method statements -- the method statement should specify the COSHH controls as mandatory steps in the work process
• Toolbox talks -- use AI-generated toolbox talks to communicate COSHH findings to the workforce
• Training records -- document that workers have been trained on the hazards and controls identified in the COSHH assessment
• Health surveillance records -- link surveillance requirements to the specific substances identified in the assessment

Site Manager AI maintains all of these documents in a single integrated platform, making cross-referencing automatic and ensuring nothing falls through the gaps. This is significantly more efficient than managing separate paper or spreadsheet-based systems for health and safety documentation.

Common COSHH Gaps AI Helps Close

• Process-generated hazards -- many sites assess chemical products in containers but miss hazards generated by work processes (cutting dust, welding fumes, diesel exhaust). AI tools prompt for these if you input the work activity rather than just the substance name.
• Biological hazards -- leptospirosis risk from excavation near watercourses, histoplasmosis from pigeon droppings in refurbishment, and tetanus from soil exposure are frequently missed. AI assessments include these when the work context is specified.
• Combined exposures -- workers are often exposed to multiple hazardous substances simultaneously (e.g., silica dust + diesel exhaust + cement). AI tools can assess combined exposure profiles, which is difficult to do manually.
• Review triggers -- AI-generated assessments include specific review triggers (change of product, change of work method, incident or near miss, new scientific evidence) that prompt timely reassessment.

COSHH Legal Compliance with AI

An AI-generated COSHH assessment is legally compliant provided it is reviewed and approved by a competent person, reflects the actual conditions on your site, is communicated to the workers affected, and is reviewed at appropriate intervals.

The HSE does not prescribe a specific format for COSHH assessments. What matters is that the assessment is suitable and sufficient -- that it identifies the hazards, evaluates the risks, and specifies adequate controls. An AI-generated assessment that meets these criteria is no less valid than a manually written one. Indeed, AI-generated assessments are often more thorough because the AI systematically covers all required elements, whereas human assessors may overlook sections under time pressure.