Everybody saw the near-miss. Nobody wrote it down.
A 350-person shop floor had near-miss reporting fall to almost nothing, so I redesigned the reporting route and the culture around it, instead of scheduling another safety course.
The problem
Formal near-miss reporting on a 350-person shop floor had fallen to almost nothing. Walk the same floor between shifts, though, and you would hear near-misses discussed all the time: what nearly caught someone's hand on the cutting line, what slipped last Tuesday and just missed the guide rail. People knew exactly what had nearly gone wrong. They just would not write it down.
The organisation's answer was to schedule more safety training. It made sense on paper. It was also wrong. Every operative on that site had already sat through statutory induction and refresher training, and the reporting form had been covered in both. So I stopped treating this as a knowledge gap, because the evidence did not support one, and went looking for what actually stops a person picking up a pen.
I found two things, reinforcing each other. Fear: a report was widely believed to end up as a black mark in a supervisor's file, not a safety input. And friction: the process was paper-based, so a five-second observation meant leaving your station, finding a form, usually in a supervisor's office, and handing it to a named team leader in person. That turned noticing something into a five-minute act of exposure. Nobody trains that fear away. You have to redesign it out.
The learner and the constraints
The person this had to work for was mid-shift, not mid-course. Three shifts, varied tenure, no assumption that everyone reads fluently or comfortably under pressure. Whatever I built had to survive being glanced at for two seconds next to a machine, not read carefully in a break room. I was not designing for the person who already reports confidently. I was designing for the person who has something to say and currently has every reason not to say it.
The question I set myself: how do I make it safe and fast to report something, when the system currently asks people to be brave every single time they do?
The approach, and why
I did not add training. Everyone on that floor had already had it, twice.
- Diagnose before designing. I mapped the evidence against Gilbert's Behaviour Engineering Model, separating what sits with the person from what sits with the environment, before proposing anything. Every symptom landed on the environment side: process friction and blame fear. Nothing pointed to a knowledge or skills gap. That is what let me rule training out, not a preference for a different answer.
- Name the culture problem properly. "Fix the culture" is not a plan. Just Culture is: it draws a clear line between honest reporting or an honest mistake and genuinely reckless behaviour. I folded a short briefing into the root-cause summary, addressed directly to team leaders, on what to do in the sixty seconds after a report lands on their desk, because that response is what decides whether the next report gets submitted.
- Redesign the reporting route, not just the training around it. I mapped the paper process step by step and replaced the two costliest steps, finding a form and handing it to a named supervisor, with a shared, reduced-attribution reporting route (no named-supervisor routing). I want to be straight about what that promises and what it does not. On a single site of around 350 people, a report that names a line and a rough time can still be worked out by anyone who knows the shift pattern. So I never call this anonymous. It reduces attribution. The rest of the protection comes from what a team leader does with it, which is exactly why the Just Culture briefing sits next to the process redesign and not on its own.
- Match the format to the moment. A near-miss is a "something has just gone wrong" moment, not a "learn something new" moment, so I built for that: a single-page job aid at the point of work, not a course leave-behind, and a short scenario rehearsal rather than a knowledge module.
- Design for the whole shift, not the confident reader. Icon-led, low-text, built to be understood at a glance across three shifts of mixed tenure. If the least confident reader on nights cannot use it standing at the machine, it has not worked.
- Measure the behaviour that leads, not the one that lags. The evaluation is built around near-miss report volume, a Kirkpatrick Level 3 behaviour measure, because it responds quickly to whether people feel safe reporting. The recordable incident rate sits in the dashboard as context only, never as a plotted result, because it moves slowly and for reasons that have nothing to do with reporting behaviour.
Artefacts
A root-cause analysis one-pager with the team-leader briefing folded in, a before/after process-flow diagram of the reporting route, a laminated job aid, a scenario-based microlearning storyboard, and an evaluation dashboard mock-up that keeps the leading and lagging indicators honestly separate. All available to walk through on request.
The outcome
The honest number is the one I started with: near enough zero. Everything the redesign is meant to move sits in front of a real cohort that has not run yet, so it stays marked in amber until it has.
What I took from it
Awareness was never the gap here either. People could describe a near-miss accurately, hours later, over tea. What stopped them was fear and friction, not ignorance, and you do not fix either with a slide deck. You fix fear by telling a team leader exactly what to do in the first sixty seconds after a report lands. You fix friction by making reporting take five seconds instead of five. And you tell the truth about what you can promise: on a site this size, reduced attribution is the honest offer. Anonymous is not, and I would rather say the smaller true thing than the bigger comfortable one.
Further reading: microlearning in the flow of work, measuring behaviour, not completion.