What near misses can teach us, and why most organisations only learn half the lesson
This blog was co-authored by Dominic Furniss and David Embrey.
Near misses are often called “free lessons”. Something went wrong, the system got close to harm, but nobody was hurt and nothing was lost. The people involved are still around to talk to, they are more likely to speak openly because there is no serious fallout, and there is usually enough organisational interest to look into what happened.
That is the theory. In practice, the lesson is only free if you actually learn it. And in our experience, organisations vary a great deal in how well they do this. Some do not collect near misses at all. Some collect them far more effectively than they learn from them. Others have established processes for triaging near misses and analysing them well.
This is one of the themes of a new book, Learning from Near Misses: Cross-Sector Reflections to Support Safety Management, edited by Nick Woodier (CRC Press). It brings together perspectives from aviation, healthcare, rail, nuclear, mining, construction, agriculture and process safety. David Embrey and I wrote the process safety chapter, on using Safety Critical Task Analysis (SCTA)to learn from near misses. Below are some of the ideas from that chapter that we think travel well across sectors.
More reports do not mean more learning
Much of the early enthusiasm for near-miss reporting came from Heinrich’s accident triangle: the idea that for every serious accident there are many more minor incidents and near misses beneath it. Deal with the base of the triangle, the logic goes, and you reduce the chance of the event at the top.
This led some organisations to set reporting targets and incentives. Managers had monthly quotas; staff received bonuses or scratch cards for reports. Unsurprisingly, the volume of reports went up. The quality of learning did not necessarily follow, and incentive schemes are open to gaming.
The deeper problem is the assumption underneath. It is not clear how thousands of slips, trips and falls help prevent a major chemical release. Both Texas City and Deepwater Horizon occurred at organisations whose data showed good performance on occupational safety. Personal safety and major hazard safety are different things, which is why near misses need triage, not just counting. We explored this gap between gathering information and acting on it in Are we collecting data, or are we learning?
Did you fail safe, or fail lucky?
A useful question to ask of any near miss is whether the system failed safe or failed lucky.
Using Reason’s Swiss cheese model, a system fails safe when a barrier or a recovery stops the sequence. For example, a tank-filling calculation is wrong, there is no supervision, the operator misses the high-level alarm, but the high-level trip stops the fill. Or operators spot smouldering and act quickly enough to stop it escalating.
A system fails lucky when every barrier fails and harm is avoided only by chance. A heavy load falls from scaffolding, or an explosion occurs, but nobody happens to be there.
These are very different near misses. The first tells you something worked. The second tells you nothing did. Treating them the same, whether in a reporting system or a risk rating, loses a lot of information. Recovery in particular deserves more attention than it usually gets, as we discussed in The Railway Men of Bhopal: how recovery shaped the impact of disaster.
Looking beyond the one path that went wrong
Traditional investigation methods, such as root cause analysis and Five Whys, tend to trace the specific sequence of events and fix the holes it passed through. That matters, but it leaves questions unanswered (we looked at how deep an investigation needs to go in The second story has layers). What about the other holes in the system? Was the event an unusual combination of conditions, or an accident waiting to happen in normal work?
This is where a task-based approach helps. SCTA is usually used proactively to assess human failure in critical tasks, but it can be adapted for investigation. At HRA we use a version called TABIE (Task Analysis Based Incident Evaluation), which works in three broad stages:
- Map the task and the events using hierarchical task analysis and swim lanes, built with the people who actually do the work.
- Identify the failure modes involved, using a failure mode library as a prompt (for example, “action too little/too much” when a valve was not opened far enough).
- Evaluate the Performance Influencing Factors (PIFs) such as distraction, procedures, training, labelling and fatigue, that made those failures more likely.
The third stage is where the extra value comes from. If distraction is high at one step, it is probably high at others. The analysis reveals vulnerabilities across the whole task, and across similar tasks elsewhere, not just the step where things nearly went wrong. For a worked example, see our SCTA in reverse series on the Herald of Free Enterprise disaster.
Three stories from practice
The operator’s little black book. After a smouldering near miss in a chemical column, the site’s own investigation found that a valve had been left open, and that the procedure did not match how the task was really done. We built a task analysis with the engineers and quickly found the plant had changed since the procedure was written, and none of the valves were labelled. But when we walked the task through with operators, they did not recognise our analysis at all. One took a small black book from his pocket with his own notes. The engineers’ view of the task was functional and technical; the operators’ was situational: do these steps on floor one, then these on floor two. The lesson: build the task analysis with the people who do the task, or you analyse the wrong task.
The step that was not analysed. A client had carried out their own SCTA after an explosion during a heater startup. Nobody was present, so it failed lucky. The cause was the heater being left in automatic mode, so it started up too quickly. Yet the client’s analysis had not examined that step at all; it had been merged with others. The lesson: screen for critical steps and give them the attention they deserve.
Beyond the engineering fix. After an electric heater was left in a superheated state, a client sensibly added automatic protection to limit energy supply. What they had not addressed was that operators got poor feedback on which mode the heater was in, and the HMI was problematic. The lesson: strong engineering controls are welcome, but the human factors issues still need fixing.
Five quality markers
From this work, we have found five markers of a good near-miss analysis using SCTA. None is novel, but they are easy to lose sight of:
- Involve frontline staff, to bridge the gap between work-as-imagined and work-as-done (sharing stories helps here, see The power of story in SCTA).
- Do a proper task analysis that goes beyond the current procedure.
- Screen for risk, to find the critical steps and focus effort there.
- Consider multiple failure modes for higher-consequence steps.
- Look beyond training and procedures, applying the hierarchy of control and improving the conditions people work in.
Why this matters across sectors
The specifics here come from process safety, but the questions apply anywhere. Are you counting near misses or learning from them? Do you know which ones failed safe and which failed lucky? Are your investigations fixing one path, or revealing how the system really works?
The book is well worth reading for the contrasts between sectors: aviation’s long-established reporting culture, healthcare’s shift towards learning from what goes right, rail’s case for why reporting matters, and more.
Learning from Near Misses is available from Routledge.
Take it further
Go beyond “root cause: human error”. Near misses are only as useful as the investigations behind them. Our blog Beyond “root cause: human error”: five principles for investigators, written with Julie Avery, sets out the principles we use to take investigations past the human error label. It is also the basis of our free 60-minute webinar, which we are running twice to cover different time zones. Join the waiting list to hear about dates first. Build the skills in your team. Our SCTA training covers task analysis, failure modes and PIFs in depth, and our SHERPA software
