Human Error

"Human error" is the most common finding in incident investigation and the least useful. It appears as a conclusion — the cause was human error — when it is actually the point at which the investigation should have started. Something made a competent person do the wrong thing on that particular day, and that something is what could have been changed. This Human Error Toolbox Talk (Safety Talk / Tailgate Talk) is about what sits underneath the phrase.

Here is the distinction that carries this whole talk: you cannot train away a slip. A slip is what happens when someone knew exactly what to do, intended to do it, and their hands did something else. Reminders, instruction and discipline all address intention — and intention was never the problem. The only thing that reduces slips is changing the work so that the slip either cannot happen or does not matter.

Three different failures wearing one name#

Human factors research distinguishes error types that behave completely differently, and lumping them together is why so many corrective actions fail.

Slips — the action went wrong. The right plan, wrongly executed. Reaching for the correct valve and turning the one beside it. Picking up the wrong tool from the bag. Typing the right number in the wrong field. Slips happen most to experienced people doing familiar work, because familiar work runs automatically and automatic actions can be captured by habit.

Lapses — the action was missed. Memory failures. Forgetting a step, losing your place after an interruption, failing to replace something you removed. Lapses cluster around interruptions and multi-step sequences — which is why they defeat permits, isolations and reinstatement.

Mistakes — the plan itself was wrong. The action was carried out exactly as intended, and the intention was wrong. Misjudging a load, misreading a drawing, applying a method that suited the last job but not this one. Mistakes come from incomplete knowledge or a wrong picture of the situation, which is why they are the error type that training genuinely does address.

That gives three different fixes, and using the wrong one is the common failure:

Retraining someone after a slip is the safety equivalent of shouting at a spanner. Adding a checklist after a mistake is worse than useless, because it makes the wrong plan more reliably followed.

And then there are violations#

Errors are unintended. A violation is a deliberate departure from the method — and it is a different thing with a different cause, usually a real problem the worker was handed. That belongs to a separate conversation about behaviours and the conditions behind them; the point to hold here is simply that you cannot fix an error with a rule, and you cannot fix a violation with a checklist. Deciding which one you are looking at is the first move.

Why error is not a moral question#

The most important shift is from asking who did this to asking why did it make sense at the time. Not as generosity — as method. Human factors work on error consistently treats errors as consequences rather than causes: the visible end of a chain that includes the design of the task, the tools, the information available, the time allowed and the state of the person.

Two things follow that matter on site.

Competent people make errors, at a predictable rate, and always will. Error rates rise with fatigue, time pressure, interruption, noise, poor lighting and unfamiliarity — every one of which is a condition somebody chose. A system that only works when nobody errs is not a safe system; it is a lucky one.

The person who erred is the best source of information and the worst target of action. They are the only one who knows what the task actually looked like from inside it. Punish them and you lose the account — and every future account from everyone watching.

Designing so error does not matter#

The professional answer to human error is not fewer errors. It is work that tolerates them.

Make the wrong action impossible. Different connectors for different gases, keys that only fit one lock, guards that must be closed for the machine to run. This is the strongest control there is, because it does not depend on anyone.

Make the wrong action obvious. Colour, labelling, orientation, physical separation of similar-looking things. Most slips involve two things that were close together and looked alike.

Make recovery possible. A step that can be checked before it becomes consequential — test before touch, a second person on the isolation, a pause before energising.

Protect against interruption. Lapses are interruption-shaped. Finish the sequence, or restart it from a defined point, and never resume from memory.

Reduce reliance on memory. Written sequences for anything with steps that matter, and sign-off at the points where getting it wrong is expensive.

Manage the conditions that raise error rates — fatigue, time pressure, noise, lighting, unfamiliar work. This is where most of the available improvement actually lives.

Where the duty sits#

There is no OSHA standard on human error — it is a human-factors concept rather than a regulated one. What the standards require is the system in which errors are caught:

1926.20(b)(2) — frequent and regular inspections by competent persons designated by the employer. Inspection is an error-recovery mechanism: it finds what individuals missed.

1926.21(b)(2) — instruction in the recognition and avoidance of unsafe conditions, which is the control for mistakes specifically.

1926.32(f) — the competent person who can identify hazards and has authority to correct them, which is what allows an error to be caught and fixed rather than merely noticed.

Section 5(a)(1) — recognised hazards. Once a predictable error mode is known, the condition that produces it is recognised.

What can go wrong?#

"Human error" recorded as the cause, so the investigation stops and the condition survives.

Retraining after a slip, which addresses intention when intention was fine.

A checklist added after a mistake, making the wrong plan more reliable.

Punishing the person who erred, which removes the only good account of what happened.

Look-alike items stored together — the classic slip generator.

Multi-step work with no written sequence and no sign-off.

Interruptions treated as normal in the middle of isolations or permits.

A system that requires zero errors to stay safe, with nothing behind the person.

How do we manage this properly?#

Classify before you correct. Slip, lapse, mistake or violation — each has a different fix, and using the wrong one wastes the opportunity.

Ask what made it make sense at the time, from inside the task. That is where the fixable condition is.

Design out the slip. Separate, label, differentiate, and make the wrong action physically impossible where you can.

Protect sequences from interruption, and never resume a permit or isolation from memory.

Put a check before the consequence — test before touch, second-person verification, a deliberate pause.

Fix conditions, not people. Fatigue, pressure, lighting, noise and unfamiliarity are the levers that move error rates.

Never let "be more careful" be the corrective action. It is the same failure as "be more aware": it names the outcome and changes nothing.

Keep the account safe. People report their own errors only where doing so is free, and those reports are the most valuable safety data a site will ever get.

Before you start#

• Confirm whether anything you will use today looks like something else nearby.
• Confirm whether this task has steps that must be done in order, and whether they are written.
• Confirm what happens if you are interrupted halfway through.
• Confirm whether there is a check between a wrong action and its consequence.
• Confirm you understand the plan, not just the steps — mistakes come from the plan.
• Confirm how tired or rushed you are, because both raise the error rate.
• Confirm you could report your own error here without it costing you.
• Confirm what would catch it if you got this wrong.

Talk it over#

• What is the easiest thing on this site to pick up or operate by mistake?
• When were you last interrupted in the middle of something that mattered?
• What would happen to you if you reported your own mistake?
• If you got this task wrong today, what would catch it?

The bottom line#

Human error is the start of an investigation, not its conclusion. Three different failures wear the name: a slip (right plan, wrong action — most common in experienced people doing familiar work), a lapse (a step missed or forgotten, clustering around interruptions and multi-step sequences), and a mistake (the plan itself was wrong, executed exactly as intended). They need different fixes — design for slips, checklists and interruption protection for lapses, training and information for mistakes — and using the wrong one is why corrective actions fail. You cannot train away a slip, because instruction addresses intention and intention was never the problem; and adding a checklist after a mistake is worse than useless, because it makes a wrong plan more reliably followed. Errors are consequences rather than causes, arriving at a predictable rate that rises with fatigue, time pressure, interruption, noise, poor lighting and unfamiliarity — every one a condition somebody chose. So design work that tolerates error: make the wrong action impossible, then obvious, then recoverable. And remember that the person who erred is the best source of information and the worst target of action.

Frequently asked questions about human error#

What is the difference between a slip and a mistake?

A slip is the right plan wrongly executed — you knew what to do, intended to do it, and the action came out wrong. A mistake is the wrong plan correctly executed — the action was exactly what you intended, and the intention was wrong. They need opposite fixes: design for slips, better information and training for mistakes.

Why can't training prevent slips?

Because training changes what someone knows and intends, and a slip is a failure of execution by someone whose knowledge and intention were both correct. Slips actually happen most to experienced people doing familiar work, because familiar actions run automatically. The controls that work are physical: separation, distinct labelling, and making the wrong action impossible.

What causes lapses?

Interruptions and memory load. A lapse is a step missed or forgotten, so it clusters around multi-step sequences and around anything that breaks concentration part-way through. That is why permits, isolations and reinstatement are the classic lapse territory, and why sequences should be finished, or restarted from a defined point, rather than resumed from memory.

Is a violation the same as an error?

No. Errors are unintended; a violation is a deliberate departure from the method, usually because the worker was handed a real problem and the deviation was the cheapest solution available. The practical rule is that you cannot fix an error with a rule, and you cannot fix a violation with a checklist — so working out which one you are looking at comes first.

Should someone be disciplined for an honest error?

If you do, you lose the account. The person who erred is the only one who knows what the task looked like from inside it, and that information is the most valuable output of the whole event. Punishing them removes it — and removes every future account from everyone who was watching.

What does error-tolerant design mean?

Work arranged so that errors either cannot happen or do not matter. Three levels, strongest first: make the wrong action impossible (different connectors, interlocks, keys that fit only one lock); make it obvious (colour, labelling, physical separation of look-alike items); and make it recoverable (test before touch, second-person verification, a deliberate pause before the consequential step).

Which conditions raise error rates?

Fatigue, time pressure, interruption, noise, poor lighting and unfamiliarity — and every one of them is a condition somebody decided on. That is where most of the available improvement lives, because a system that stays safe only while nobody errs is not safe, it is lucky.

Download the human error toolbox talk PDF#

Get this human error toolbox talk as a print-ready PDF — available in English, Spanish, Portuguese, and Turkish. Print it, hand it to the crew, and collect signatures on the included attendance sheet.

Download the PDF — free account required. New members get 5 free downloads.

Related toolbox talks#

• Common Unsafe Behaviors
• Common Unsafe Conditions
• Situational Awareness
• Fatigue Management

Sources#

• OSHA, 29 CFR 1926.20 — General safety and health provisions: https://www.osha.gov/laws-regs/regulations/standardnumber/1926/1926.20
• OSHA, 29 CFR 1926.21 — Safety training and education: https://www.osha.gov/laws-regs/regulations/standardnumber/1926/1926.21
• NIOSH, Construction Safety and Health: https://www.cdc.gov/niosh/construction/default.html

Written by FieldSafetyTalk's safety professional — a CSP, ASP, CHST and OSHA Authorized Outreach Trainer with 14+ years of international construction safety experience across federal, heavy civil, and industrial projects.

Hazards covered