Human error in root cause analysis: why it is rarely the real root cause

Open any deviation log and count how often the root cause field says 'human error', 'operator error' or 'procedure not followed'. At many sites it is a third or more of all closed deviations, and the CAPA attached to almost every one of them is 'retrain the operator'. Inspectors know this pattern well. It is one of the most reliable ways to turn a set of individually closed deviations into a single finding against the quality system.

This guide explains why human error is a description rather than a cause, how to classify what actually happened, the questions that take an investigation past the person to the conditions around them, and what a defensible conclusion looks like when a person really did make a mistake.

Why inspectors push back on 'human error'

EU GMP Chapter 1 (1.4(xiv)) expects an appropriate level of root cause analysis during the investigation of deviations, and says that where human error is suspected or identified as the cause, this should be justified, having taken care to ensure that process, procedural or system based errors or problems have not been overlooked. The burden of proof sits with the investigator. 'Human error' is not the default answer when nothing else is found; it is a conclusion you have to earn by ruling the other causes out.

ICH Q10 makes the same point from the other direction: CAPA should come from investigation of the cause, and its effectiveness should be evaluated. If the cause is 'a person got it wrong', the only available action is to change the person, which in practice means retraining. When the same error then recurs with a different, freshly trained person, the record shows that the cause was never found.

Human error is a symptom, not a cause

A person entering the wrong weight, skipping a step or signing for a check they did not do is the point where a problem became visible. It is the start of the investigation, not the end of it. Competent, trained people make predictable errors under predictable conditions: unclear instructions, poor equipment design, time pressure, interruptions, fatigue, look-alike materials, or a process that relies on memory instead of a check.

The useful question is not 'who made the error?' but 'why did a trained person, acting in good faith, think this was the right thing to do at that moment, and why did nothing catch it?' Both halves matter. Most serious deviations involve an error and a missed detection: a second-person check that was signed but not performed, an in-process control that could not detect the problem, or a review that looked at the record instead of the data.

The four types of human error

Classifying the error is the step most investigations skip, and it decides which corrective action can work. The widely used model from human factors research separates unintended actions from intended ones:

• Slip: the person intended the right action but executed it wrongly. Example: selecting the adjacent line on a touchscreen, or transposing two digits when typing a batch number. Caused by attention and design, not knowledge.
• Lapse: the person forgot a step or lost their place. Example: missing a line clearance step after an interruption. Caused by memory load, interruptions and long, unstructured procedures.
• Mistake: the person did what they intended, but the intention was wrong because their knowledge or judgement was wrong. Example: applying an old dilution after a method change. Caused by training, procedure clarity or the way change was communicated.
• Violation: the person knowingly deviated from the procedure. Usually a 'routine violation' where the procedure is impractical and everyone works around it, sometimes a 'situational violation' under pressure. Rarely malicious, and almost always a sign that the procedure or the workload is the problem.

Retraining only addresses mistakes, and only some of them. It does nothing for slips and lapses, because the person already knew what to do. It does little for routine violations, because the person also knew the rule. That is why sites that answer every human error with retraining see the same deviations come back.

Questions that find the real cause

A structured set of questions makes the investigation repeatable and gives the reviewer something to check. Ask them of the event, not of the person, and record the answer to each one, including the ones that come back 'no'.

1. Procedure: was the instruction available at the point of use, current, unambiguous and in the order the work is actually done? Could two trained people read it differently?
2. Training and qualification: was the person trained on this version, and did the training include doing the task, not just reading the SOP? When did they last perform it?
3. Equipment and interface: do labels, screens or controls invite confusion? Are look-alike materials, containers or settings stored or displayed next to each other?
4. Environment: noise, lighting, temperature, PPE limiting dexterity or vision, cramped layout?
5. Workload and timing: was the shift short-staffed, was there time pressure, overtime, a night shift, a changeover being rushed?
6. Interruptions: was the person interrupted or asked to switch tasks mid-step?
7. Communication and handover: did a shift handover, a recent change or a verbal instruction feed into the event?
8. Detection: what check should have caught this, why did it not, and was the check designed to detect this failure at all?
9. History: has the same or a similar error happened before, in this area or elsewhere on site, with a different person?

If any answer points to a condition that would make the same error likely for another trained person, that condition is a contributing or root cause and needs its own action. If the history question returns earlier events with different people, human error is almost certainly the wrong conclusion.

Choosing CAPA that works: the hierarchy of controls

Once the error type and the conditions are known, choose the strongest action that is practical. Actions higher in this list remove the opportunity for error; actions lower down rely on people remembering to do better.

1. Eliminate the step: remove the manual transcription, the calculation or the handover entirely.
2. Engineer it out: interlocks, barcode verification, system range checks, physically different connectors or containers.
3. Simplify and make errors visible: redesign the form or screen, reorder the procedure to match the work, add a checklist at the point of use, separate look-alike items.
4. Detect it: a second-person check that is designed to catch this failure, an in-process control, a review of the data rather than the printout.
5. Train: targeted and practical, when the investigation showed a real knowledge gap, ideally combined with one of the actions above.

When human error is the honest conclusion

Sometimes a well-designed process, a clear procedure and a trained, rested person still produce a one-off slip. That conclusion is acceptable when the record shows the work: the questions above were asked, each contributing factor was considered and ruled out with evidence, the history check found no similar events, and the detection controls worked as designed. In that case, record it as an isolated human error, state why no systemic cause was found, and trend it.

Trending is what protects you. Human error deviations should be trended by area, task, shift, error type and time of day, not just counted. A cluster of lapses on night shift or slips on one piece of equipment is a systemic cause that no single investigation would have shown.

A short checklist for the deviation reviewer

• The error is classified as a slip, lapse, mistake or violation, with the reasoning.
• Each contributing factor question has a recorded answer, including the negatives.
• The history of similar events across the site was checked and is referenced.
• The failed or missing detection control is identified, not just the error.
• The CAPA sits as high in the hierarchy of controls as is practical, and retraining alone is justified if used.
• An effectiveness check with a measurable criterion and a time frame is defined.
• Human error as a final conclusion is explicitly justified, as EU GMP 1.4(xiv) requires.

Human error investigations are where a quality system shows whether it learns. Sites that ask why the error made sense at the time, and fix the conditions, see their repeat deviations fall. Sites that retrain and close see the same events, with different names on them, at the next inspection.

Frequently asked questions