Near-miss reporting: why your current system probably underreports by 90%

Here's something that should bother every safety manager: you almost certainly know about the incidents that went wrong. The ones that resulted in an injury, a lost time event, a workers' compensation claim. Those get recorded because they have to be.

But what about the hundreds of events that nearly went wrong?

The forklift that turned a blind corner at speed, missing a pedestrian by a metre. The pallet that slipped off a rack but landed in an empty aisle. The worker who stepped into an exclusion zone while a vehicle was reversing, but happened to step back in time.

Those events contain more useful safety information than every injury report in your filing cabinet combined. And the research is clear: your current reporting system is almost certainly missing the vast majority of them.

The safety triangle tells us what we should be seeing

The relationship between near misses and serious injuries is one of the most studied concepts in workplace safety. Frank E. Bird's landmark 1969 study analysed 1.75 million accident reports from 297 companies and established a ratio of 1 serious injury to 10 minor injuries to 30 property damage events to 600 near misses.

Heinrich's earlier work in 1931 proposed a similar relationship: 1 major injury for every 29 minor injuries and 300 no-injury incidents.

A 2003 ConocoPhillips Marine study pushed the concept further, finding that for every single fatality, there were approximately 300,000 at-risk behaviours.

The exact ratios are debated (and vary by industry), but the underlying principle is consistent: for every serious injury, there are hundreds or thousands of precursor events that, if captured and addressed, could prevent the serious injury from ever occurring.

Now here's the problem. If Bird's ratio holds roughly true, a site that records 2 serious injuries and 10 minor injuries per year should also be seeing at least 120 near misses. Most sites report far fewer. Some report fewer than 10.

That's not because near misses aren't happening. It's because the reporting system isn't capturing them.

What the research says about underreporting

The underreporting problem isn't limited to near misses. Even recordable injuries are significantly underreported in most workplaces.

Leigh, Marcin, and Miller (2004) estimated that the U.S. Bureau of Labor Statistics system misses between 33% and 69% of all workplace injuries. A Washington State study found that 90% of employers failed to correctly record work-related injuries, and research across multiple industries has found that workers report only about 63% of serious occupational injuries.

The AFL-CIO's 2025 Death on the Job report states it plainly: underreporting of workplace injuries and illnesses is widespread, with the true toll estimated at 5.2 million to 7.8 million cases per year in U.S. private industry alone.

BLS's own research acknowledges the problem: a follow-up survey of employers found considerable misunderstanding of recordkeeping requirements, inconsistent log-keeping, and that small establishments are least likely to maintain records at all.

If recordable injuries are underreported by 33% to 69%, imagine the scale of underreporting for near misses, which carry no legal reporting obligation and rely entirely on voluntary worker participation.

That's where the 90% figure comes from. It's not a stretch. It's a conservative estimate based on comparing the near-miss volume Bird's ratio would predict against what most organisations actually capture through manual systems.

Why manual near-miss reporting doesn't work

The reasons are practical, cultural, and structural. And once you understand them, the scale of the gap makes complete sense.

People are busy

A warehouse operator running a forklift at pace doesn't stop to fill out a form every time something happens that could have been worse. They're focused on throughput, on their next pick, on the clock. Unless someone was injured, the moment passes and the near miss goes unrecorded.

There's no clear incentive (and sometimes a disincentive)

Research on underreporting identifies several reasons workers don't report: time pressure, doubts about eligibility or relevance, concerns about reputation, fear of income loss, and worry about career prospects. In workplaces with safety incentive programmes tied to low incident rates, there's an active disincentive to report anything at all. The BLS Hidden Tragedy report documents cases of intimidation and harassment of workers who report injuries.

Definitions are fuzzy

What counts as a near miss? Different workers have different thresholds. An experienced forklift driver might consider a close pass "normal." A new starter might not recognise a near miss when it happens. Without a clear, consistent definition, reporting depends on individual judgement, and individual judgement varies wildly.

Paper and portal systems add friction

If reporting a near miss means walking to an office, finding the right form, filling in 15 fields, and hoping someone reads it, most people won't bother. Even digital systems (apps, portals) require the worker to stop what they're doing, open the system, describe the event, and submit. The friction alone kills most reports before they start.

Nobody does anything with the data

This might be the biggest one. When workers report near misses and nothing visibly changes, they stop reporting. A 2024 study published in the Journal of Occupational Health found that companies that responded adequately to reported near misses had significantly fewer subsequent accidents. The flipside: when companies failed to respond, workers perceived reporting as pointless, and accident rates increased.

Why near misses matter more than incident data

If you're running your safety programme primarily on injury data, you're driving by looking in the rear-view mirror. Injuries tell you what already went wrong. Near misses tell you what's about to go wrong.

The National Safety Council makes the case simply: near misses reveal hazards that might go unnoticed until it's too late, and correcting unsafe conditions before an injury occurs is always preferable to investigating after one.

There's also a statistical argument. Injuries are (thankfully) rare events. For most sites, the sample size is too small to identify meaningful trends. You might have 5 recordable injuries across a year, spread across different types and locations, and there's not enough data to see a pattern.

Near misses, when captured at sufficient volume, give you a large enough dataset to identify patterns, pinpoint high-risk zones, compare shifts, and track whether interventions are working. They're the leading indicators that every safety framework talks about but most organisations struggle to generate.

The Campbell Institute has pushed this thinking further with the concept of SIF potential (serious injury and fatality potential). Not all near misses are equal. A dropped pen is not the same as a forklift-pedestrian near miss. The near misses that matter most are those where, if circumstances were slightly different, the outcome could have been a serious injury or fatality. Isolating those events from the noise is where the real prevention value lies.

What changes when detection is automatic

This is the fundamental shift that computer vision AI introduces to near-miss reporting: it removes the reliance on manual observation and voluntary reporting entirely.

When cameras are monitoring your facility continuously, a near miss between a forklift and a pedestrian is detected and recorded whether anyone reports it or not. It doesn't depend on the worker's willingness to fill out a form, their judgement about what counts, or whether their supervisor follows up. It happens automatically, consistently, 24 hours a day.

The result is a dramatically larger and more accurate dataset. Instead of 5 or 10 manually reported near misses per month, organisations using inviol typically see hundreds of safety events captured across their sites, giving them the volume of data they need to spot trends, compare shifts, and identify high-risk zones through heatmaps.

Crucially, this data feeds directly into coaching workflows. At inviol, the goal isn't just detection; it's turning each event into a coaching moment. When a supervisor can pull up blurred, privacy-protected footage of a real near miss and discuss it with their team during a shift briefing, the conversation is grounded in evidence, not assumptions.

This is the missing link in most near-miss programmes: the feedback loop. Workers see that events are being captured, that they're being discussed, and that changes are being made as a result. The 2024 Journal of Occupational Health study showed exactly this: when companies respond visibly to near misses, safety outcomes improve.

From compliance to insight

In New Zealand, the Health and Safety at Work Act 2015 defines notifiable incidents as unplanned or uncontrolled events that expose a worker to serious risk. WorkSafe NZ requires PCBUs to notify these events as soon as possible. In Australia, similar obligations exist under WHS legislation. In the U.S., OSHA's recordkeeping requirements cover injuries and illnesses but rely heavily on employer self-reporting.

But compliance is the floor, not the ceiling. The organisations that are genuinely reducing risk aren't just meeting their notification obligations. They're building systems that capture the events below the regulatory threshold: the near misses, the close calls, the at-risk behaviours that never trigger a report but collectively signal where the next serious injury is coming from.

That's the shift inviol enables across warehousing, logistics, manufacturing, and retail operations. Not replacing your reporting system, but filling the gap that your reporting system was never designed to cover.

What would your safety programme look like with 10 times the data?

Think about that for a moment. If you're currently capturing 10 near misses a month through your reporting system, what decisions would you make differently if you had visibility of 100? Or 500?

You'd know which intersections are the most dangerous, and at what times. You'd know which shifts have the highest near-miss rates. You'd know whether the new traffic management plan is actually working or whether people are reverting to old habits. You'd have the data to justify the layout change you've been requesting for months. And you'd have evidence that your organisation is actively identifying and managing risk, which is exactly what HSWA and WHS legislation require.

If you want to see what that looks like in practice, book a demo with inviol and we'll show you how safety teams are closing the near-miss visibility gap across their sites.

Frequently Asked Questions

What percentage of near misses go unreported?

Research suggests the vast majority of near misses are never captured by manual reporting systems. Frank E. Bird's 1969 study established a ratio of 600 near misses for every serious injury, yet most organisations capture only a small fraction of that expected volume. Combined with BLS research showing that even recordable injuries are underreported by 33% to 69%, conservative estimates suggest that 90% or more of near misses go unrecorded.

Why don't workers report near misses?

Research identifies several barriers: time pressure during shifts, unclear definitions of what counts as a near miss, fear of blame or reputational damage, safety incentive programmes that discourage reporting, friction in reporting systems (paper forms, multi-field portals), and a lack of visible follow-up action from management. When workers see that reports lead to no change, they stop reporting.

What is Bird's safety triangle?

Frank E. Bird's 1969 study of 1.75 million accidents from 297 companies established a ratio of 1 serious injury to 10 minor injuries to 30 property damage events to 600 near misses. The principle suggests that reducing the frequency of events at the base of the triangle (near misses and at-risk behaviours) reduces the likelihood of serious injuries at the top.

How does AI improve near-miss reporting?

Computer vision AI uses existing CCTV cameras to automatically detect and record safety events like forklift-pedestrian near misses, exclusion zone breaches, and speeding without relying on manual observation or voluntary reports. This removes the reporting barriers (time, friction, judgement, fear) and produces a much larger, more consistent dataset that can be used for coaching, trend analysis, and risk reduction.

Are near misses legally reportable in New Zealand?

Under the Health and Safety at Work Act 2015, PCBUs must notify WorkSafe NZ of "notifiable incidents," which are defined as unplanned or uncontrolled events that expose a worker to serious risk. This covers serious near misses (such as structural collapses, uncontrolled substance releases, or plant failures) but does not cover the majority of everyday near misses like close-proximity pedestrian-vehicle events. Those events, while not legally notifiable, are exactly the ones that predict future serious injuries.