Injuries are lagging indicators. Near-misses are the same events without the harm, they are far more numerous, and they are the only advance warning available.
Why reporting is the foundation
Near-misses are numerous relative to injuries, so they give a usable signal in a period where injuries would give none.
They identify conditions before someone is hurt.
They are free to investigate, because nobody is off work.
They only exist if people report them, which is entirely a matter of what happens next.
What destroys reporting
Attribution and consequence. A report that leads to a conversation about the reporter's conduct is the last report that person makes.
No visible response. Reports that disappear stop being made within a quarter.
A long form. Reporting must take under a minute.
Requiring a supervisor's involvement, which filters out anything involving the supervisor.
Statistics used as targets. A facility with a near-miss reduction target has created an incentive to under-report, and it will succeed.
What sustains it
A short, easy route — a card, a code, a two-field form.
Anonymous option available, even if most people do not use it.
Visible response. Publish what was reported and what was done, including where nothing was.
No consequence for the reporter, stated and demonstrated in the first hard case.
Thank people, specifically.
Treat a rising reporting rate as success, which it is, and say so before the rate rises so that the interpretation is established in advance.
What automated detection adds
Proximity events from vehicle systems, which capture near-misses nobody would report.
Impact events, including minor ones.
Speed exceedances.
Pedestrian incursion detection.
These fill the gap for events people do not notice or do not consider worth reporting, and they are not a substitute for reporting, because the human report includes the reason.
Combining the two
Automated detection tells you where and when.
Human reports tell you why.
Use detection to find the locations, then ask the people who work there what is happening. That combination produces findings neither source gives alone.
Where detection shows a cluster and nobody has reported it, that is a reporting culture finding as much as a safety one.
Investigating properly
Conditions first, person last, every time.
Ask what made the unsafe action reasonable, because it usually was, given the pressures.
Look for the pattern, not the instance. A single near-miss is noise; five at the same corner is a finding.
Involve the people who work there in deciding the remedy, which produces remedies that get used.
Close the loop publicly, which is what sustains reporting.
The measures
Reporting rate, trended, with a rise interpreted as good.
Detected events, by location and hour.
Time from report to response.
Proportion of reports resulting in a change.
Ratio of detected to reported events, which indicates how much is going unreported and is the most diagnostic number in the set.
The detected-to-reported ratio
The most diagnostic number available on reporting culture.
Count events detected automatically — proximity, impacts, incursions.
Count events reported by people.
The ratio indicates how much is going unreported.
A widening gap means reporting is degrading, whatever the absolute numbers do.
Track it monthly, and treat a widening gap as a culture finding rather than a technology one.
Report it, because it is the measure that makes an invisible problem visible.
Closing the loop publicly
What sustains reporting, and the step most often skipped.
Publish what was reported, anonymised, and what was done about it.
Including where nothing was done, and why, which is the part that buys credibility.
Post it where the reports came from, not in a management report.
Name the change, not the reporter.
Do it on a schedule, so the absence of a response is visible rather than assumed.
Reports stop within a quarter where nothing visibly follows them, and restarting takes years.