Why Looking for Safety Incidents Is Too Late

A safety program is built around prevention. Its record is built around the incident.

Something happens — a near miss, a strike, an injury. It gets logged, investigated, root-caused, corrected. Over time you accumulate that record: incident reports, near-miss logs, corrective actions. This is competent work. It's what a serious safety function is supposed to do, and most of the discipline in the field is built around doing it well.

But there's a structural problem hiding inside it, and it has nothing to do with effort.

By the time something is an incident, the information you needed arrived too late to be cheap.

An incident is a condition that finally got unlucky

An incident is rarely the first time the condition occurred. A forklift and a pedestrian don't cross paths dangerously for the first time on the day someone gets hurt. That corner has been a conflict point for weeks. The restricted zone that finally caused an injury had been entered, briefly and without consequence, many times before. The PPE gap that turns up in an investigation was present on that line long before it mattered.

Operations are continuous. Incidents are instantaneous.

An incident is not the beginning of a problem. It's the moment a condition that was already present happened to line up badly in time. The condition was doing its work quietly the entire time. The incident is simply the one moment that announced itself.

Which means the thing most safety programs are organized to capture — the incident — is the least informative version of the problem. It's rare, it's noisy, and it's lagging. It tells you a condition existed only after that condition finally caused harm.

What this looks like on one aisle

Consider a simple example.

A camera that's already mounted over one aisle — installed years ago for security — watches people cut across a forklift lane to reach the break room. Any one crossing is nothing. A person walked through a lane; it happens constantly; no one would ever write it up.

But operations don't happen in isolated moments. They unfold over time.

Reviewed across two weeks, that same crossing shows up 34 times. Almost all of them fall in the ten minutes right after shift change. Almost all of them come through the same doorway, because that door is the shortest path to the break room and the lane is in the way.

No one was hurt. No incident was ever filed, because on the definition every safety program uses, nothing happened.

That pattern — 34 crossings, one doorway, one ten-minute window — is a recurring condition. It was invisible not because no one was watching, but because no one had time to watch that aisle for two weeks and understand what kept happening.

Once it's understood, the decision it points to is small and obvious: move the break-room door, restripe the walkway, or hold forklift moves for those ten minutes. Cheap changes, made before anyone gets hurt. A follow-up review a month later tells you whether the crossings actually stopped, or simply moved somewhere else.

That is the whole argument, on one aisle.

The real unit of safety is the recurring condition

Step back from the aisle and the shape holds everywhere.

Conditions that recur — a pedestrian entering a restricted zone at shift change day after day, a forklift and a person sharing the same corner every afternoon, PPE gaps clustering on one line, one crew, one time of day — are frequent where incidents are rare. They're patterned where incidents are random. And, critically, they're observable before anyone is harmed.

Time turns individual interactions into operational patterns.

A recurring condition is the signal. An incident is one outcome when that signal goes unrecognized or unaddressed in time.

If you could see the conditions, you'd rarely need to wait for the incident to know where your risk actually lives.

From observation to decision

The aisle example runs through a fixed sequence. Every recurring condition does:

One observation may be worth a look.

Time gives that observation meaning.

A recurring condition deserves attention.

A condition that persists after you've acted on it deserves a different decision.

This is the layer that sits between detection and prevention — the piece that's missing when a safety team has only individual alerts on one side and serious incidents on the other, with nothing structured connecting the two.

Why safety programs have always started with incidents

Not because anyone chose poorly. Because for most of the history of workplace safety, recurring conditions were hard to observe continuously, while incidents announced themselves.

You cannot manually watch every aisle, every hour, every shift. No safety lead has ever had the hours.

The limitation wasn't judgment. It was time.

So the profession optimized around the only signal that reliably surfaced on its own — the incident, which announces itself by definition. Everything downstream — the logs, the investigations, the metrics — is built on the one thing you could count on to show up.

The rear-view mirror wasn't a mistake. It was the only mirror available.

What's changed

The observation layer is already installed.

Many industrial, warehouse and construction facilities already run cameras across much of the floor — for security, operations and liability. The aisles, the zones and the dock doors are, in a literal sense, already being watched. Much of the observation infrastructure needed to surface recurring conditions is already in place.

What cameras capture naturally is not a single event. They capture time.

The value comes from understanding what keeps happening within it.

The objective is not to score individual workers or assign fault. It is to identify where visible operating conditions keep creating exposure so the site can improve the control.

What hasn't been solved is the step after sensing. Continuous footage is not the same as reviewed evidence. Hours of video that no one has time to review tell you little about which conditions recur, where they concentrate or how they evolve.

The gap was never the camera.

The gap is turning continuous observation over time into structured, reviewed evidence that reveals recurring operational conditions people can actually act on.

That is a different problem.

And it's the one worth solving.

The shift

The question a safety program asks defines what it can do.

"What incidents happened?" is a rear-view question. It's necessary — you still investigate, you still comply — but it can only ever describe a past you didn't get to change.

"What keeps happening?" is a different kind of question.

It recognizes that operations are continuous, that patterns emerge through time, and that the best opportunities for intervention usually exist before an incident has a chance to occur.

None of this makes incidents disappear, and it would be dishonest to claim it does. Conditions will still sometimes line up badly no matter how well you watch them.

But the earlier recurring conditions become visible, the more of your decisions get to be cheap ones — a changed route, a moved rack, a retimed task — instead of expensive ones made in the aftermath.

Seeing a recurring condition earlier also changes the economics of prevention. The business case does not depend on assuming that every observation would have become an incident. It depends on whether the condition is frequent, concentrated and consequential enough to justify action, considering potential incident consequence, operational disruption, investigation and corrective-action costs, and the cost of allowing the condition to continue. A follow-up review then shows whether the observed condition changed after the site acted.

That earlier moment, before the incident, is where the real work of prevention has always belonged.

Until recently, organizations simply couldn't observe operations continuously enough to stand there.

Today, in many facilities, they already can.

The recurring condition, not only the incident, is the unit of safety worth measuring. That is the premise Edgentik is built on: operations create interactions, time turns those interactions into patterns, and those patterns reveal where action may be justified long before an incident becomes the first thing anyone notices.