Why Near-Misses Are the Most Valuable Safety Data You're Not Using
Nothing happened…
-The forklift passed close to a pedestrian, but there was no collision.
-An operator entered a restricted area for a few seconds, but no one was injured.
-A pallet temporarily obstructed a safety route, but it was moved before becoming a problem.
-A worker approached a machine without the expected protective equipment, but the situation was corrected immediately.
From a traditional safety reporting perspective, these events may disappear almost as quickly as they occurred:
-No injury.
-No accident.
-No downtime.
-No formal incident.
But from a preventive safety perspective, they may contain some of the most valuable information available inside the factory.
They are near-misses.
And when near-misses are detected, contextualized and analyzed as patterns rather than isolated events, they can help HSE teams understand where risk is developing before an accident occurs.
Accidents Tell Us What Already Happened
Traditional safety metrics are essential.
Number of injuries.
Lost-time incidents.
Accident frequency.
Severity rates.
Days without accidents.
These indicators help organizations understand historical safety performance.
But they share an important characteristic:
they usually measure events after something has already gone wrong.
An accident is therefore a lagging indicator.
It tells us that a risk became a real event.
Near-misses provide a different perspective.
They can reveal the conditions that repeatedly appear before an accident ever happens.
A forklift crossing too close to pedestrians.
Operators repeatedly entering a hazardous zone.
Emergency exits that are frequently obstructed.
PPE requirements that are inconsistently followed.
Unsafe interactions around moving machinery.
None of these situations guarantees that an accident will occur.
But repeated exposure can reveal where the safety system deserves attention.
One Near-Miss May Be Random. Fifty Are a Pattern.
Imagine a warehouse intersection.
A forklift passes close to a pedestrian once.
It may be an isolated event.
Now imagine that similar interactions happen 38 times over two weeks.
Mostly between 14:00 and 16:00.
Mostly at the same intersection.
Mostly when outbound logistics activity increases.
The safety question changes.
It is no longer simply:
“Did an accident happen?”
It becomes:
“Why are risky interactions repeatedly occurring in this location?”
Perhaps the pedestrian route is poorly positioned.
Perhaps visibility is limited.
Perhaps material staging temporarily blocks the normal path.
Perhaps traffic increases during a particular production window.
Perhaps floor markings are no longer adequate for the actual logistics flow.
The individual event provides limited information.
The pattern provides context.
The Problem: Most Near-Misses Are Never Recorded
Traditional near-miss reporting often depends on people noticing an event and deciding to report it.
That creates an unavoidable limitation.
Many events are too brief.
Some appear insignificant.
Some occur in areas without supervisors.
Others become part of everyday routine.
A worker may think:
“Nothing happened, so there is nothing to report.”
Over time, potentially useful safety signals remain invisible.
This does not mean employees are behaving irresponsibly.
It means manual reporting cannot realistically capture every interaction occurring across a complex manufacturing environment.
Modern Smart Safety technologies can help close part of this visibility gap.
Computer Vision Can Turn Events into Safety Data
Computer vision can continuously analyze specific operational areas and identify predefined safety events.
For example:
pedestrian-forklift proximity;
entry into restricted zones;
obstructed emergency routes;
missing PPE in designated areas;
unsafe presence near machinery;
traffic moving through predefined hazardous intersections.
The objective should not simply be to generate more alerts.
If every event becomes an alarm, the result can quickly become alert fatigue.
The more valuable opportunity is to transform individual observations into structured safety data.
Where did the event occur?
When?
How frequently?
Under which operating conditions?
Is the same pattern increasing?
Does it occur during particular shifts?
Is it concentrated around specific production or logistics activities?
This turns computer vision from an event detector into a source of preventive safety intelligence.
From Individual Events to Risk Patterns
Suppose a Smart Safety system detects repeated forklift-pedestrian proximity events.
Looking at one event provides limited information.
Looking at hundreds of observations over time can reveal patterns.
Perhaps 70% of the interactions occur at two intersections.
Perhaps one location is particularly problematic during shift changes.
Perhaps another becomes critical when temporary material is staged nearby.
Perhaps the number of events increases significantly during periods of high logistics activity.
Now the HSE team has something actionable.
Not a prediction that an accident will definitely happen.
But evidence that exposure to a particular risk is recurring.
That distinction is important.
Predictive safety should not pretend to predict individual accidents with certainty.
Its value lies in identifying conditions and patterns associated with elevated risk.
Frequency Alone Is Not Enough
Not every near-miss deserves the same priority.
A frequently occurring low-severity event may require attention.
A rare interaction with potentially severe consequences may also require immediate investigation.
This means HSE teams need more than event counts.
Risk prioritization can consider several dimensions:
Frequency — how often does the situation occur?
Exposure — how many people or operations are exposed?
Potential severity — what could happen if the event developed into an accident?
Recurrence — is the pattern stable, increasing or decreasing?
Location — is the event concentrated in a specific area?
Time — does it occur during particular shifts or activities?
The objective is not to create an artificial safety score that replaces professional judgment.
It is to help HSE teams identify where investigation may have the greatest preventive value.
A Heatmap Can Reveal What Incident Reports Cannot
Imagine reviewing a map of the production and logistics areas.
No accidents were recorded during the last month.
Traditional incident statistics therefore appear positive.
But the near-miss data shows something else.
One warehouse intersection has an unusually high concentration of pedestrian-forklift interactions.
A second area shows repeated access to a restricted machine zone.
Another frequently experiences temporary obstruction of a designated safety route.
No one has been injured.
That is precisely why the information is valuable.
The organization has an opportunity to investigate before the statistics change for the worse.
Near-Miss Data Can Improve the Workplace, Not Just Monitor It
The purpose of Smart Safety should not be to watch employees more closely.
It should be to understand how the working environment creates or amplifies risk.
If repeated near-misses occur at one intersection, the solution may not be telling workers to “be more careful.”
The better intervention could be redesigning the traffic flow.
Adding physical separation.
Moving material staging.
Improving visibility.
Changing pedestrian routes.
Adjusting production or logistics timing.
Adding warning systems.
Revising a procedure.
The most useful safety data often points toward system improvements, not individual blame.
This distinction is essential for building trust around safety technology.
Privacy and Governance Matter
Computer vision in the workplace requires careful governance.
The objective should be clearly defined.
Which safety events are being detected?
Why is the information collected?
How long is it retained?
Who can access it?
Is individual identification actually necessary?
In many safety analytics scenarios, the operational value comes from understanding aggregated patterns rather than identifying specific individuals.
For example:
27 risky interactions occurred at Intersection B.
Most happened during the afternoon shift.
Frequency increased after the logistics layout changed.
That information can support preventive action without turning safety analytics into employee surveillance.
Technology should strengthen the safety system while respecting the people working inside it.
Human HSE Expertise Remains Central
AI can identify patterns that would be difficult to observe manually.
But a pattern does not explain itself.
If near-misses increase near a production area, the HSE team still needs to understand why.
Is the layout responsible?
Has production volume increased?
Did the material flow change?
Are operators using an unofficial shortcut?
Is visibility poor?
Is a procedure unrealistic under actual operating conditions?
AI can help answer:
Where should we look?
Human expertise determines:
What should we do about it?
This is the same principle that should guide industrial AI more broadly.
Technology expands visibility.
People provide operational judgment.
Measuring Whether Safety Actions Actually Work
Near-miss data also creates another opportunity.
It can help measure the effect of preventive interventions.
Suppose an HSE team identifies repeated forklift-pedestrian interactions at one intersection.
The company changes the layout and installs a physical pedestrian barrier.
What happens next?
If the number of risky interactions falls significantly over the following weeks, the team gains evidence that the intervention improved the situation.
If the pattern remains unchanged, further investigation may be necessary.
This creates a continuous improvement loop:
Detect → Analyze → Intervene → Measure → Improve
Safety becomes more than reporting.
It becomes an operational improvement process.
From Lagging Indicators to Leading Signals
A mature safety strategy needs both perspectives.
Lagging indicators remain important.
Organizations must understand accidents, injuries and their consequences.
But leading indicators provide an opportunity to act earlier.
Near-misses.
Repeated unsafe interactions.
Exposure patterns.
Restricted-zone entries.
Recurring PPE gaps.
Obstructed safety areas.
These signals do not predict the future with certainty.
They provide evidence about where risk is repeatedly emerging.
That evidence can help HSE teams decide where to investigate before an incident forces the issue.
The Safest Event Is the One That Never Becomes an Incident
The traditional safety process often begins with an event.
An incident occurs.
It is reported.
The causes are investigated.
Corrective actions are introduced.
The organization learns.
Near-miss intelligence creates the possibility of moving that learning process earlier.
The event is detected.
The pattern is identified.
The risk is investigated.
Preventive action is taken.
And ideally, the accident never occurs.
That is the real promise of predictive safety.
Not predicting exactly when someone will be injured.
But giving safety teams better signals about where intervention may be needed before someone is.
Conclusion
Factories already generate far more safety information than appears in traditional incident reports.
Much of it exists in the small events that disappear because nothing happened.
A forklift passed close to a pedestrian.
A restricted zone was entered.
A safety route was temporarily blocked.
An unsafe interaction occurred around a machine.
Individually, these events may appear insignificant.
Collectively, they can reveal how risk develops across the factory.
Smart Safety capabilities integrated into solutions such as SkyMes can help transform these observations into structured information, allowing HSE teams to identify recurring patterns, investigate high-exposure areas and measure whether preventive actions are actually reducing risk.
Because waiting for an accident gives you certainty.
Understanding near-misses gives you something more valuable:
the opportunity to act before certainty arrives.