Every safety manager knows the phone call. A supervisor's voice, a little too flat, a little too careful: "We've had an incident." What follows is a familiar and unwelcome sequence — the walk to the scene, the cordon, the statements, the stopwatch on the ten-day RIDDOR clock, the form that turns a bad day into a permanent line on the company record.
By the time that call is made, prevention is over. The only thing left is paperwork and consequence.
RIDDOR — the Reporting of Injuries, Diseases and Dangerous Occurrences Regulations 1995 — is a good and necessary law. It gives the HSE the data it needs to see where British workers are being hurt. But it is, by design, a record of the past. It counts the fractures, the over-seven-day absences, the specified injuries, the dangerous occurrences that came within a breath of catastrophe. It tells you, with grim precision, where you have already failed.
The interesting question is not how to file the report faster. It is how to catch the incident in the minutes, hours and days before it ever becomes reportable.
The reportable incident has a long tail
No serious injury arrives out of nowhere. Behind the single amputation logged under RIDDOR sits a well-documented pyramid: a scattering of minor injuries beneath it, and beneath those, a broad base of near-misses and unsafe acts that harmed no one — this time.
That base is where the story is written. The forklift that passed a fraction too close to a pedestrian on Tuesday. The operator who climbed into the racking without a harness on Wednesday. The pallet stacked a course too high on Thursday. None of these are reportable. None generate a form. And precisely because they generate no paperwork, they are usually invisible to the very people whose job is to prevent the Friday when one of them finally connects.
The reportable incident, in other words, has a long tail of warnings. The problem has never been the absence of warnings. It has been that nobody was watching closely enough, for long enough, to see them all.
Human vigilance does not scale
A conscientious supervisor might witness a dozen unsafe acts on a walkround. The other several thousand that occur across a shift, on cameras nobody is watching, in aisles nobody happened to be standing in, pass unrecorded. This is not negligence. It is arithmetic. No human team can watch every camera, on every shift, without blinking, forever.
This is the gap that a computer-vision layer is built to close. SecureSafety adds an AI perception layer to the CCTV you already have — no new hardware, no rip-and-replace — and it watches every frame of every feed, continuously, on-premise, without fatigue. It does not get bored at 3am. It does not look away when the near-miss happens in the corner of a camera nobody was assigned to.
When it sees a vehicle and a pedestrian on a collision path, a worker entering a red zone, a missing helmet, a person on the ground, a plume of smoke, it flags it in real time — as an alert to intervene now, and as a data point in the pattern that predicts the RIDDOR event you have not yet had.
From reporting to pre-empting
Reframe the whole exercise and RIDDOR stops being the finish line and becomes the thing you are racing to prevent.
See the near-misses you are currently blind to. Every close call becomes a logged, time-stamped, reviewable event rather than a moment that evaporated the instant it passed. You cannot manage what you cannot count.
Find the hotspots before they find you. When the system shows that the same blind corner has produced forty vehicle–pedestrian conflicts this month, you have a location, a cause and a business case — long before that corner produces the collision that produces the report.
Intervene in the moment. A real-time alert to a supervisor's phone can stop the unsafe act while it is still merely unsafe, not yet an injury. The best RIDDOR report is the one you never had cause to write.
This detection was not built in a laboratory. It was forged offshore, on the drill floors of oil and gas — heavy moving steel, zero tolerance for error, lives in the balance on every tour. It has since run in the operations of a national oil major, a major international port and an international airport, at a measured error rate below 0.05%, and in the field it has been associated with reductions in unsafe behaviour of around 90%. When the environment punishes every mistake, you learn to catch the warning, not just the wound.
What your regulator, and your board, actually see
There is a quieter benefit worth naming. An organisation that can show a falling curve of near-misses — not just an absence of injuries, but active, dated evidence that hazards were spotted and closed out — is telling a very different story to an HSE inspector, an insurer or a board than one whose only safety data is its RIDDOR log.
A clean RIDDOR record can mean a safe site. It can also mean a lucky one. The difference is whether you were watching the long tail all along.
RIDDOR will always have its place: the law requires it, and the data serves the wider good. But the goal was never to file better reports. It was to have fewer reasons to file them.
RIDDOR and AI monitoring: closing the gap between reporting and prevention
What RIDDOR data tells you — and what it cannot
RIDDOR data is valuable for identifying which hazard categories are generating reportable incidents at your site, and for benchmarking your incident rate against industry norms. What it cannot tell you is where the near-misses are concentrating, because near-misses are structurally under-reported in every organisation. The ratio of near-misses to reportable incidents in most industrial environments is estimated at 300:1 — meaning for every RIDDOR-reportable injury, there are around 300 near-miss events that never appear in the data. AI monitoring makes those 300 visible, which is where the prevention opportunity actually lies.
Using AI data to meet RIDDOR's investigation obligation
When a RIDDOR-reportable incident does occur, the investigation obligation under the regulations requires the employer to understand the cause and prevent recurrence. An AI monitoring system provides the investigation with context that traditional CCTV does not: not just footage of the incident itself, but a near-miss history showing how frequently the same hazard had occurred before the injury, the specific times and locations where the risk was concentrated, and the response actions (or lack of them) that followed previous near-miss alerts. This context is both operationally useful for root cause analysis and legally relevant in the event of an HSE investigation.
The prevention dividend: fewer RIDDOR reports
The most direct financial benefit of reducing RIDDOR-reportable incidents is not the reporting compliance burden (which is modest) but the insurance, productivity and management costs that each reportable injury carries. A systematic reduction in near-miss frequency — the mechanism through which AI monitoring reduces incidents — reduces RIDDOR-reportable injuries proportionally, because incidents and near-misses are not independent: they are the same hazard, at different points on the severity distribution. Reducing the near-miss rate reduces the incident rate.
Implementation checklist for RIDDOR compliance and prevention
- Near-miss category mapping: define which AI detection categories correspond to the RIDDOR-reportable incident categories most common on your site — this is the mapping that lets you track whether near-miss reduction is translating to incident reduction
- Investigation support protocol: define how AI monitoring data is accessed and preserved when a RIDDOR-reportable incident occurs — the near-miss history for the location and the alert log for the period preceding the incident should be preserved as investigation evidence
- RIDDOR report cross-reference: when a RIDDOR report is filed, review the AI monitoring data for the location and time window to confirm whether a near-miss alert was generated before the incident — if it was, document the response action; if it was not, review whether camera coverage or zone configuration should be adjusted
See what your cameras have been missing — book a demo.
