Near Miss: Meaning, Examples and Reporting

Ask ten people on a large site what a near miss is and you will get ten answers, several of which describe something else. Some will describe an unsafe condition they walked past. Some will describe an event where somebody was hurt, but only slightly. Some will say their organisation does not use the phrase at all and calls it a close call, a good catch or a potential incident. None of them are being difficult. The vocabulary in this corner of health and safety is genuinely muddled, used inconsistently between organisations and often inside one organisation, and that is not a cosmetic problem. It is why so much near-miss data turns out to be unusable when somebody finally tries to analyse it.

The message up front: a near miss is an event that could have caused harm or loss but did not. Its value is that it carries the information of an accident without the cost. Most of what determines whether an organisation captures that value sits in two places: whether reporting is easy and genuinely blame-free, and whether people can see that their reports change anything. The technology is the least interesting part.

This article is a general explanation written from implementation experience, not legal or safety-engineering advice. Reporting obligations, and the design of a reporting system that satisfies them, must be established from your own jurisdiction and your own management system with competent advice.

1. Near miss meaning: what a near miss actually is

A near miss is an unplanned event that had the potential to cause injury, ill health, damage or loss, and did not. Nothing was broken, nobody was hurt, no product was lost, and in most cases no record would exist at all unless somebody chose to create one. That last clause is the whole reason near misses are difficult to manage: an accident announces itself, and a near miss only exists if a human being decides to tell you about it.

Three elements are doing the work in that definition, and it is worth separating them because arguments about classification usually come down to one of the three.

• It is an event. Something happened at a point in time. A spanner fell. A reversing truck stopped short. A valve was opened. This distinguishes a near miss from a hazard or an unsafe condition, which is a state that persists rather than an event that occurred.
• It had realistic potential to cause harm or loss. Not theoretical potential, because on that test everything qualifies and the data becomes noise. The question is whether, with a small and plausible change of circumstance, this event would have hurt somebody or destroyed something.
• No harm or loss actually occurred. As soon as there is an injury, ill health, damage or loss, however minor, you are dealing with an accident rather than a near miss, and it belongs in a different bucket.

There is also a linguistic oddity worth naming, because it affects behaviour. The phrase "near miss" is slightly misleading: it was a near hit. The event nearly caused harm; it did not nearly avoid it. The wording quietly invites people to dismiss the event as a non-event, which is the opposite of the intended message, and that is precisely why a number of organisations rename the category. "Close call" and "good catch" are both attempts to keep the seriousness while removing the invitation to shrug. I have no strong view on which label to adopt, and a very strong view that an organisation should pick one and define it.

2. The family of related terms, and why the muddle matters

Most near-miss data problems are classification problems, not collection problems. An organisation collects a large volume of reports, finds that a substantial share are unsafe conditions filed as near misses and a handful are minor injuries filed as near misses to protect the injury statistics, and concludes the data cannot support any conclusion. It cannot, and the fault is upstream in the definitions.

Two of these deserve a pointer to a fuller treatment, because each is a subject in its own right. The behaviour-versus-state distinction, and the reason "unsafe act" is a weak stopping point for any analysis, is worked through in the unsafe act versus unsafe condition guide. The hazard-versus-risk confusion, which is probably the most common terminological error in the whole field, is handled in hazard versus risk.

3. Why near misses are worth this much attention

The business case for taking a no-consequence event seriously is not sentimental, and it does not depend on any statistical claim. It rests on four fairly plain observations.

They carry much of the information of a serious incident, with none of the cost. The failed control, the confusing procedure, the missing isolation, the blind corner, the worn sling: these are usually identical in an event that hurt somebody and in one that did not. What differs is often nothing more than where a person happened to be standing. An organisation that only learns from events with consequences is buying its learning at the highest available price.

They are far more numerous. Whatever the ratio is in a given workplace, and it is not knowable in advance, events without consequence plainly outnumber events with consequence. That gives a larger sample to reason from and a better chance of seeing a pattern before it produces an injury.

They surface conditions and behaviours before harm occurs. This is the only category of safety information that is genuinely leading rather than lagging. Injury statistics tell you what has already happened; a near-miss report can tell you about a failing control while there is still time to act.

They are frequently the only visible evidence that a control is being worked around. This one is underrated and, in practice, the most valuable of the four. When a procedure is impractical, people quietly invent an alternative that gets the job done. That workaround is usually invisible to management, because it works. It becomes visible on the day it nearly does not. A near miss is often the first honest indication that the written method and the actual method have diverged, which is information no audit against the written method will ever produce.

That is also why near-miss reporting belongs inside a management system rather than beside it. ISO 45001:2018 as amended by Amd 1:2024 (an international voluntary standard, certifiable, and law in no jurisdiction by itself) requires hazards to be eliminated and risks reduced through the hierarchy of controls at clause 8.1.2, with the same principle set out in the US voluntary consensus standard ANSI/ASSP Z10.0-2019 at section 8.4. Near-miss reports are one of the better mechanisms for discovering that a control somewhere in that hierarchy is not holding in practice. The tiers themselves are covered in the hierarchy of controls guide.

4. The accident triangle: the useful idea and the disputed numbers

Anyone who has sat through a safety induction has seen the triangle: a single fatality or serious injury at the apex, a larger band of minor injuries beneath it, a larger band of near misses beneath that, and a broad base of unsafe acts and conditions. It is drawn with specific numbers attached, and those numbers are repeated everywhere, in inductions, toolbox talks, vendor slides and a great deal of published safety material.

Here is the honest position, worth stating plainly because almost nobody does. I am not going to publish any of those ratios, and you should be careful about doing so either. The specific published figures originate in early twentieth-century and mid-century industrial studies. They are very widely repeated, they are also genuinely disputed, and their data, methods and applicability to any particular modern workplace have all been questioned. They are not measurements of your organisation. Repeating a ratio as though it were one is the most common accuracy failure in writing on this topic.

What survives the criticism is the shape of the idea, and the shape is useful. Events without consequence are far more numerous than events with consequence, and because they cost nothing when they occur they are a better source of learning per unit of pain than the rare serious event. That is the argument for a reporting system, and it does not need a number to work. There is also a second, more substantive criticism that matters more than the arithmetic, and it is the part of this article I would most want a safety manager to take away.

The practical consequence is a two-track approach. Run the reporting system to learn broadly, and separately and deliberately verify the handful of controls standing between the operation and a catastrophic outcome. Those verification activities are not driven by report volume, they are driven by the consequence of failure, and they belong in the risk assessment process rather than in the incident process.

5. Near miss examples

The examples below are illustrative, written to show the shape of the category rather than to describe any real event. They are weighted towards maintenance and facilities work, because that is where most readers of this site spend their time, and because maintenance activity is unusually rich in near misses: it involves working on equipment that is normally guarded, energised or in service, often under time pressure, often by people who do not work in that area every day. No values, no injuries and no real cases appear here.

Notice a common thread. In most of these, the reason nobody was hurt has nothing to do with any control working. It has to do with where a person happened to be, or when they happened to look. That is the test I would apply to any candidate report: if the only thing standing between this event and an injury was position or timing, it is a near miss and it is worth capturing.

6. Near miss reporting: what a system needs to actually work

This is the heart of the subject. Most organisations have a near-miss reporting mechanism on paper. A minority have one that produces useful data. The gap between the two is rarely technical.

It must be easy and fast. Friction is the primary suppressor of reporting, and it operates quietly, because nobody refuses to report. They simply run out of the will to do it. A system requiring a login, a desktop machine, mandatory fields the reporter cannot answer and fifteen minutes of their time will collect almost nothing, and what it collects will come from the few people whose job includes reporting. The realistic target is a report submitted from where the person is standing, in the time it takes to describe the event in two sentences, with classification and analysis done later by somebody whose job that is. Every mandatory field is a tax on reporting, paid in missing reports.

It must be genuinely blame-free in practice, not in policy. Every organisation has a no-blame statement. Very few have a no-blame reality, and the workforce knows which kind it works for with considerable accuracy. There is one test, and it is not the wording of the policy: what happened the last time somebody reported something embarrassing, something that revealed their own mistake, or something that implicated their supervisor. Whatever happened on that occasion is the operative policy, and it has already been communicated to everybody who needed to know. It is worth noting that this principle is not a soft one. IEC 62740:2015, the international standard on root cause analysis (a voluntary standard, adopted in Europe as EN 62740:2015, and not law anywhere by itself), explicitly excludes the assignment of responsibility or liability from the scope of the analysis. The separation of learning from blame is built into the recognised method, not bolted on as a cultural aspiration.

Feedback must close the loop. The fastest reliable way to kill a reporting system is for somebody to report something and hear nothing. They have taken a small risk and spent their time, and the silence answers them: this was pointless. Every report needs an acknowledgement, and every report needs an outcome communicated back to the person who raised it, even when the outcome is that no action will be taken and here is why. A reasoned no is survivable. Silence is not.

Reports must visibly change things. Feedback tells one person their report was read. Visible action tells everybody else it was worth reporting. This is the only durable incentive in the whole system, and it does not require large changes. A guard refitted, a walkway cleared, a procedure amended, a barrier moved, with the connection to the report made explicit on a board or in a briefing, does more for reporting rates than any campaign. Where a report leads to a substantive change, the toolbox talk is a natural vehicle for closing that loop with the wider crew, and the toolbox talk guide covers how to run one that is not simply read aloud.

Of those nine, two decide the outcome. Feedback and visible action are what make a reporting system durable, and they are also the two most often treated as administrative afterthoughts. An organisation with a clunky paper form, a genuine no-blame record, fast feedback and a visible record of changes made will out-report one with a polished mobile application and none of those things, by a wide margin and every time.

7. What to do with the reports once they arrive

A near-miss system that collects well and processes badly is a slightly more expensive version of one that collects nothing, and it burns credibility on the way. Four things need to happen to the incoming flow.

Triage by potential consequence, not by what actually happened. This is the most important handling rule. Every near miss has, by definition, an actual consequence of nothing, so sorting by actual consequence sorts by nothing and a report about an undocumented safety-system inhibit lands in the same tray as a coffee spill. Sort instead by the worst credible outcome had circumstances differed slightly. A near miss with catastrophic potential deserves the response a serious incident would get, at the same seniority and with the same rigour. A process that cannot escalate a zero-consequence event to the level of a serious one will eventually miss the report that mattered.

Look for patterns as well as individual events. A single near miss is one data point and may have an obvious local fix. The same near miss recurring is a different and considerably more serious signal than three unrelated ones, because repetition means the local fix has not held, was never made, or was made against a symptom. It also tends to indicate a systemic cause: a procedure that cannot be followed as written, a layout that forces a workaround, a workload that makes the safe method slower than the available time. Pattern analysis is where a reporting system stops being administrative and starts being useful, and it depends mostly on consistent coding and consistent location and asset references.

Investigate proportionately. Not every report needs a formal investigation, and treating every report as though it does will collapse the process under its own weight within a quarter. Match the depth of enquiry to the potential consequence and to the presence of a pattern. Where a full investigation is warranted, the method, the evidence handling and the common failure modes of the process are covered in the incident investigation guide, and the causal analysis techniques that sit inside it, including how the recognised methods relate to one another, in the root cause analysis guide. Both of those are separate subjects and I am deliberately not reproducing them here.

Act, and manage the action to closure. A cause identified and no action taken is worse than no analysis, because the organisation now has a documented awareness of a risk it did nothing about. Corrective actions fix the instance; preventive actions address the conditions that allowed it; and both need owners, dates and verification that the change actually worked. The lifecycle, and the reason so many action registers fill with items closed without evidence, is covered in the CAPA guide. Where reports keep revealing hazards nobody had identified, that is a signal about the identification process itself rather than about the reporting system, and the hazard identification guide is the right place to look.

8. The measurement trap

Near-miss reporting is measured badly more often than it is measured well, and the bad measurement does active harm. The root of the problem is a category error about what the number means.

A near-miss report count is a leading indicator of reporting culture. It is not an indicator of safety performance. It measures how willing people are to tell you things. It does not measure how safe the operation is. Once that is understood, two conclusions follow immediately, and both of them are the opposite of the instinctive reading.

The first is that a rising count is usually good news. More reports normally means more trust, better awareness of what qualifies, and lower friction. Presenting a rising near-miss count to a board as deteriorating safety performance is a mistake that will, entirely predictably, result in pressure to reduce it, which will succeed, and which will reduce reporting rather than risk.

The second is that treating a low count as good news is the more dangerous error of the two. A low count occasionally means a small, simple and well-controlled operation. Far more often it means people have stopped telling you, and the organisation is now flying with the instrument disconnected. A site with very few reports and a busy, hazardous operation should be treated as a site with a reporting problem until demonstrated otherwise.

From those two follows the reason not to set a numeric target. Requiring a number of reports per person per period, or rewarding volume, reliably produces exactly what it asks for: a large number of trivial and sometimes invented reports, submitted near the end of the period, destroying the signal in the data and costing real effort to process. Once the workforce understands that the count is what matters, the count is what they produce, and the dataset does not recover quickly. For the same reason I will not publish a benchmark figure here. There is no credible universal number, rates vary enormously with industry, activity mix, definition and reporting maturity, and any figure quoted would become a target the moment it was read.

9. Why people do not report

Every organisation with a reporting system also has a large volume of unreported near misses, and the reasons are consistent enough to list. This section is the honest part of the article, because the reasons are mostly rational responses to how the organisation actually behaves.

• Fear of blame for themselves. The most valuable reports are self-implicating and the least likely to be filed. So the blame-free record has to be demonstrated rather than asserted, and the first self-implicating report a manager receives is a test they are being set.
• Fear of getting a colleague into trouble. Loyalty within a crew usually outweighs loyalty to a process. So reports about other people's actions need a route that does not read as an accusation, and the response has to be visibly about the condition rather than the person.
• Not wanting to create work. Reporters know a report generates an investigation, questions and possibly a delay to the job. So proportionate handling is not merely efficient, it is a precondition of reporting at all.
• Not recognising the event as reportable. Many near misses are not suppressed, they are simply not seen as events. Somebody catches a falling object and carries on. So the definition has to be taught with concrete local examples rather than a policy statement.
• Previous reports going nowhere. The most common reason in mature organisations, and the most self-inflicted. So the feedback loop is not a nicety, it is what keeps the system alive.
• Time pressure. When the schedule is tight, reporting is the first discretionary activity dropped. So it must be short enough to survive a bad day, and supervisors must be seen to accept a small delay without visible irritation.
• Contractor status and concern about the contract. A reporter may worry they are creating a record that affects a scorecard or a renewal. That is a commercial disincentive on top of the ordinary ones, discussed further below.
• Language barriers. A process operating in one language collects from the people who speak it. So reports must be accepted in the languages the workforce actually uses, including verbally to a supervisor who files them, because a free-text form in a second language defeats even a willing reporter.

The common structure across all eight is that non-reporting is usually a rational response to an experienced cost. Nobody needs persuading that safety matters. They need the cost of telling you to be lower than the benefit they can see coming back.

10. Contractors and multi-employer sites

On most facilities and maintenance sites a substantial share of the higher-risk work is done by people who are not employed by the organisation that owns the site. That arrangement creates a specific and well-known reporting gap, and it has an uncomfortable shape: the people most exposed to the hazards are frequently the least likely to report them.

Several things combine. The route is often genuinely unclear, because there are two systems and no stated rule about which receives what. Access is often the practical blocker, since the client's tool sits behind a corporate login the contractor's workforce does not have. A report may be read as a performance issue against the contract. The relationship is shorter, so the blame-free record the client built with its own staff has not been demonstrated to these workers at all. And the contractor workforce is frequently the part most affected by language barriers.

What I would advise, without pretending it is easy to arrange: state one reporting route for the whole site, in writing, at induction, and make it work without a client login. Accept reports from anybody on site regardless of employer. Say explicitly that reports do not count against the contract, and then behave that way, because the first time a report surfaces in a commercial conversation the route closes permanently. Feed outcomes back to the contractor's workforce and not only to their management. And where reporting from contractor crews is conspicuously lower than from direct employees doing comparable work, treat that as a finding about the route rather than about the crews.

11. Where this meets maintenance management

There is an angle here that general safety material tends to miss, and it is the one most relevant to a maintenance or facilities readership: near misses and maintenance condition are closely coupled, in both directions.

Maintenance work generates near misses. It is intrinsically hazardous work: equipment that is normally guarded is opened, equipment that is normally energised is isolated, stored energy is released, access is temporary, and the person doing the work is often not the person who operates the plant every day. Isolation failures, permit scope mismatches, wrong-asset events and access-equipment problems are all recognisably maintenance near misses, and they cluster around the same few process steps. Where controlled work runs on permits, the integrity of that process is where a large share of the high-potential reports will originate, and the mechanics of tying permits to the work management system are covered in permit to work integration.

Deferred maintenance generates near misses. This is the connection worth making to anyone who has to argue for maintenance resource. A defect backlog is not a neutral administrative queue. It is a list of known degraded conditions that have been accepted for now: the handrail not yet repaired, the light not yet replaced, the leaking valve on the list, the guard held on with a temporary fixing. As that backlog grows and ages, the population of degraded conditions in the workplace grows with it, and near misses arising from those conditions can reasonably be expected to rise. I would not claim a measurable coefficient between backlog and near-miss rate, because I have not seen one credibly established and I am not going to invent one. I would claim that a maintenance manager who can show a rising high-potential near-miss trend alongside an ageing backlog has a considerably stronger resourcing argument than one showing the backlog alone. The measurement side of that, and the reason backlog age matters more than backlog volume, is in maintenance backlog and downtime tracking.

Capture reports against the asset and the location. The practical step that makes the first two visible. Most systems record a description, a date and a rough location, which supports very little analysis. If a report can be attached to a specific asset or functional location and coded consistently, patterns tied to particular plant become visible: a panel that keeps producing electrical near misses, a pump set where isolation keeps going wrong, a plant room where access events concentrate. That turns reports into engineering decisions rather than reminders about care, and it puts the reports alongside the failure history for the same asset, which is where the interesting correlation lives. The coding discipline involved is the one described in failure codes: problem, cause, action, and if you are assessing what a system needs to do in the first place, this introduction to CMMS covers the ground. The capability matters far less than whether anybody codes the location consistently.

12. The statutory boundary: when a near miss is legally notifiable

This needs care, and it is where a purely internal view of near misses can go wrong. In many jurisdictions there are categories of event that cause no injury at all and are nonetheless formally reportable to a regulator. Certain failures of lifting equipment, certain releases of hazardous substances, certain collapses, certain electrical and explosion events and certain plant failures can fall into that class. The category exists precisely because these events carry high potential consequence regardless of their actual outcome.

The practical implication is that an event which is a near miss in your own terminology may simultaneously be a legally notifiable event, and the two classifications are decided by different authorities. Yours is decided by your management system. The other is decided by law.

The categories, the criteria and the notification arrangements are jurisdiction-specific, and I am not going to publish a threshold, a deadline or a form name here, because they differ by jurisdiction, they are revised, and getting one wrong in a published article is worse than not stating it. In Great Britain the duty sits with the Health and Safety Executive under the relevant reporting regulations, and Northern Ireland operates under separate instruments. In the United States, reporting duties to the Occupational Safety and Health Administration apply federally, with State Plan states operating their own arrangements that may differ. In the United Arab Emirates, occupational safety and health duties in the private sector sit under Federal Decree-Law No. 33 of 2021 on the regulation of employment relationships, administered by the Ministry of Human Resources and Emiratisation, with Article 13 carrying the occupational safety duties, alongside emirate-level frameworks such as the Abu Dhabi Occupational Safety and Health System Framework administered by the Abu Dhabi Public Health Centre. US and UK requirements have no legal force in the UAE; they are voluntary benchmarks only.

What I would do internally is straightforward. Establish which categories are notifiable in your jurisdiction, from the regulator's own current published material. Build that determination into triage, so the question "is this notifiable" is asked of every high-potential report at the point of receipt rather than discovered weeks later. Give one named role responsibility for making the call. And keep the internal and external classifications separate in the record, because an event can easily be both, and collapsing them tends to produce either over-reporting or, more commonly, silence where a duty existed.

The wider risk-management frame these judgements sit in is described in ISO 31000:2018, which is guidance and is not certifiable, so there is no accredited organisational certification against it, with assessment techniques catalogued in IEC 31010:2019. Both are voluntary international standards and neither creates any legal duty.

The idea to walk away with

A near miss is an event that could have caused harm and did not, and its entire value lies in the fact that it hands you the learning of an accident with none of the cost. Capturing that value depends on very little technology and a great deal of trust. One definition, applied consistently. A reporting route short enough to survive a bad day and open to everybody on site regardless of employer. A blame-free record that has been demonstrated rather than asserted. Feedback to every reporter, and visible change often enough that reporting looks worthwhile to the people watching. Triage by what could have happened rather than by what did.

And two things to hold firmly. The report count measures your reporting culture, not your safety performance, so a rising count is usually good news and a low count is usually a warning. And the triangle, whatever numbers are drawn on it, does not license the assumption that managing the numerous small events controls the rare catastrophic one. Serious events often travel by different routes, and those routes need controlling on their own terms.

Final thoughts

The organisations that do this well are not the ones with the best software. They are the ones where a technician can tell a supervisor something awkward in thirty seconds, and expects to hear back, and has seen things change before. That is a cultural asset that takes years to build and a single badly handled report to damage, which is why I would spend far more attention on how the first difficult report is received than on the design of the form.

If you want one practical starting point: go and find the last twenty near-miss reports your organisation received, and check what happened to each one. Not what was recorded as happening, what actually happened, and whether the person who reported it was told. That exercise usually answers the question of why the reporting volume is what it is, and it costs an afternoon.

Primary sources worth going to directly rather than through secondary summaries: UK Health and Safety Executive , US OSHA , ISO , IEC .

Related reading: Incident investigation: a complete guide, Unsafe act vs unsafe condition, Hazard identification methods and process, Hierarchy of controls, CAPA explained, Root cause analysis methods, Maintenance backlog and downtime tracking, Permit to work integration.