AI safety monitoring on site: PPE, height and forklift proximity
No supervisor can watch every camera on every shift. AI can, and on higher-risk sites (construction, warehousing, manufacturing) that difference is measured in injuries that never happen.
What is at stake, in the official numbers
The Health and Safety Executive's own statistics set the scale:
Behind the totals, around 623,000 workers are injured each year. Most of those incidents are preceded by a visible condition: the missing harness, the blocked walkway, the forklift and the pedestrian in the same aisle. Visible is the operative word, because on most sites a camera was already pointing at it.
What AI monitoring actually watches for
Connected to a site's existing cameras, the software watches continuously for the agreed set of safety events, including:
- PPE compliance: hard hats, hi-vis and other required equipment, detected in real time rather than sampled by spot checks.
- Working at height: people near unprotected edges or outside safe zones, the scenario behind the leading cause of worker deaths.
- Exclusion and no-go zones: anyone entering an area that should be empty, from crane radii to energised plant.
- Forklift and vehicle proximity: moving plant and pedestrians converging, flagged before contact, not after.
- Slips, trips and obstructions: blocked routes and exits, standing water, materials where they should not be.
Alerts route to the people who can act, on site, immediately. Over time the pattern of alerts becomes something else valuable: an objective picture of how the site actually runs, for toolbox talks, training and, where the operator chooses, for their insurer. The full detection list is on Use cases.
Does it work?
The engine behind our platform is built by a specialist video analytics company whose platform already runs on industrial sites internationally, across construction, logistics, marine and aviation operators. Their published figures include a 73% reduction in total recordable incident rate and more than 4.2 million hazards detected. Those are the developer's own published numbers from their deployments, not ours; we keep what is proven and what is new clearly separated on our Proof page.
The insurance angle
Fewer incidents mean fewer and smaller claims, which is why this technology increasingly arrives on site through insurance rather than as a standalone purchase: the insurer funds monitoring because prevention costs less than paying claims, the same logic behind insurer-funded leak sensors in home insurance. If you buy commercial cover, ask your broker about it; if you are the broker or insurer, start with the broker view or the insurer view.
Privacy, up front: in the deployments we work with, faces are blurred by default, footage is processed only for the agreed use cases, and retention and access are set per deployment. The system exists to catch hazards, not to watch people. Details on Security, privacy and data.
Common questions
Does AI safety monitoring identify individual workers?
Not in the deployments we work with. Faces are blurred by default and the system reads behaviour and hazards, not identities. The point is to catch the unsafe condition, not to name the person in it.
How is a PPE detection camera different from spot checks?
Spot checks sample moments; a camera watches every hour of every shift. Detection covers hard hats, hi-vis and other agreed PPE, with sensitivity tuned to the site, so patterns show up that no supervisor could see, and alerts arrive in time to act.
Can forklift proximity detection use our existing warehouse cameras?
Usually, yes. The software runs on most existing cameras over standard network protocols, watching for moving plant and pedestrians in the same zone and alerting when they get too close.
Does this replace our safety team?
No. It gives the safety team continuous eyes and an objective record. Decisions, interventions and culture remain theirs; the technology removes the physical impossibility of watching every camera at once.