Building a near-miss reporting culture

A near miss is an incident that gave you the information for free. Same conditions, same sequence, same lesson available — and no injury, no vehicle damage, no claim, no patient harm. It is the cheapest evidence a safety program will ever be offered.

It is also the most reliably missing. Nothing forces a near miss into the record. An injury creates paperwork whether anybody feels like filing it or not; a strap that slipped and was caught creates nothing at all unless somebody chooses to tell you. So a near-miss count is not a measurement of near misses. It is a measurement of willingness to report them.

That distinction matters because the two look identical on a dashboard. A service reporting four near misses a year is not safer than one reporting two hundred. It is certainly less informed, and the number on the chart reads as though the opposite were true.

Somebody has just had a close call. Nobody was hurt. They are back in the vehicle and there is a decision to make, and it takes about four seconds.

What they are weighing is not complicated. Does this cost me anything? Will it come back at me, at my partner, or at my crew? How long will it take? Does anything happen as a result, or does it go into a system nobody reads? And underneath all of it: what happened to the last person who did this?

Some of those are outside your control. Most are not. The cost question is set by what people have watched happen to colleagues. The time question is set by the form. The does-anything-happen question is set by whether the last report produced a visible change or silence.

And the ambiguous case resolves as a no. If somebody is unsure whether reporting is safe, they do not gamble on it for an event where nobody got hurt. This is why a policy that is technically fair but has never been demonstrated produces almost no reports: absence of evidence reads as risk, and the four seconds run out.

Sidney Dekker’s work on just culture makes a distinction that is easy to state and hard to operate: accountability and blame are not the same thing, and an organization can insist on the first without reaching for the second. Accountability is being able to give an account — to tell the story of what happened and why it made sense at the time. Blame is what happens when the account is used against the person giving it.

The common failure is announcing the first while keeping a process shaped for the second. The policy says no-blame. The form asks who was at fault. The investigation is conducted by the person who writes performance reviews. The outcome field offers counselling and retraining as its main options. Everyone reads the process rather than the policy, and they are right to.

The other failure is the opposite one, and it is worth naming because fear of it is why organizations hesitate. Just culture is not the absence of consequence. Dekker’s framing has room for genuinely reckless conduct being treated as such. What it removes is consequence for honest accounts of ordinary work going wrong — which is the overwhelming majority of what you will ever receive.

The test is not what your policy says. It is whether somebody in your organization can name a time when a person reported something that reflected badly on them and nothing bad happened. If nobody can name that, you do not have a just culture yet. You have an intention.

The strongest single predictor of whether somebody files a second near-miss report is what they observed after the first one.

Not whether it was acknowledged, though that helps. Whether anything visibly changed, or — when nothing changed — whether they were told why. A report that disappears teaches a precise lesson: this goes into a system, the system does nothing, and the four seconds are better spent elsewhere. That lesson spreads faster than any policy communication, because it is delivered by a colleague rather than a poster.

The closing of the loop does not have to be a fix. We looked at this, it is the third time it has come up, here is what we are testing is a closed loop. So is we looked at this and decided not to act, and here is the reason. Both are answers. Silence teaches people not to bother.

It is also cheap, and easy to skip, because closing the loop is rarely anybody’s explicit job and never appears on a dashboard.

Friction and time both matter, and they are separate problems.

Friction is the form. A near-miss report that asks the same twenty-two fields as an injury report is a form that says we do not really want these. If the event produced no injury, most of the injury paperwork is inapplicable, and requiring it anyway converts a two-minute act of goodwill into a twenty-minute chore at the end of a shift.

Timing is the other half, and it is underrated. A report filed six hours later is a materially different report — not because anybody is being dishonest, but because memory consolidates into narrative. The specifics that make a report analytically useful, the sequence and the incidental details, are the first things to go. The account you get the next morning is a summary of a story, and the story has already been simplified.

Which means the design goal is not a better form. It is the shortest possible path from the moment somebody thinks that was close to the moment the thing is recorded — and the fewer decisions you ask them to make on that path, especially decisions about how to categorize what happened, the more of them arrive.

Near-miss reporting going up is usually good news, and it will not look like good news on a chart.

The number will rise steeply when the program starts working, because you are not measuring events — you are measuring reporting of events, and the events were always happening. Somebody above you will see a line going up and read it as a safety decline, and if that reading lands before you have framed it, the program will be shut down for succeeding.

So say it in advance, in writing, before the first number moves. State plainly that the count is expected to rise, that the rise reflects reporting rather than incidence, and that a flat near-miss count alongside a steady injury rate is the genuinely concerning pattern. Agreeing what the numbers will mean before you have them is the same discipline that makes any measured change readable afterwards.

Better indicators of the culture itself: how quickly reports arrive after the event, what proportion come from people describing their own close calls rather than somebody else’s, and how many reports received a visible response. Those measure the thing you actually care about. The raw count mostly measures how many people currently trust you.