Engineering & Technology · Reliability, Safety, Failure & Maintenance

Human factors, organisational conditions and learning from incidents

An investigation records that a technician missed a step. That statement describes part of an event sequence, but leaves the useful questions open: what information was visible, which tasks competed for attention, what was practical at the time, and how could the system make a similar omission less likely or less consequential?

Use Quality Engineering, Experiments & Process Evidence; Engineering Design Methods & Trade-offs; and Material Selection, Degradation & Failure. Read probability as a fraction of comparable cases, distinguish a rate from a count and retain units in energy and inventory balances. Models use invented teaching records and supplied thresholds; they do not approve a real safety or maintenance decision.

Back to the chapter

After reading, you can

Examine how task design and organisational conditions shape performance, and turn incident findings into verified improvements.

Describe the conditions of the task

Human factors connects people’s capabilities and limitations with the equipment, task and organisation. Interface ambiguity, distraction, workload, fatigue, competence and communication can influence performance. Calling someone careless skips those conditions and provides little direction for improving a system used by other people tomorrow.

A slip or lapse is an unintended action or omission during execution; a mistake involves an unsuitable plan or decision. A deliberate departure from a rule raises another set of questions, including whether the rule was workable and understood. These categories can overlap in an event. Training can help with knowledge, but cannot by itself eliminate a confusing control layout or make an impossible workload feasible.

Make a resource constraint visible

The figure treats a workload as a simple inventory of comparable tasks. It does not model human error. If work arrives faster than it can be completed with the required checks, unfinished work accumulates somewhere: a queue, deferred inspection, hidden rework or unmet service. Counting the queue helps expose the mismatch without attributing it to an individual.

An apparent increase in throughput obtained by omitting quality checks changes the definition of completed work. Compare qualified completion, available tools, information and interruptions. Ask workers about adaptations that allow the service to continue, and whether those adaptations introduce exposure or conceal a deteriorating margin. A numerical staffing ratio alone cannot answer those questions.

Starting backlog 8 tasks, constant incoming work 6 tasks/h. Solid case has qualified completion capacity 4 tasks/h, so B=8+2 t over 0–4 h. Dashed case has supplied qualified capacity 8 tasks/h and backlog B=max(8−2 t, 0), reaching zero at 4 h. The task-inventory model assumes comparable tasks, complete quality checks and available resources; it predicts neither human-error probability nor a safe staffing level. Capacity increases are supplied conditions, not a direction to hurry work or remove checks.

Make learning change the future system

After an incident or near miss, reconstruct the available information and conditions before judging decisions with knowledge acquired later. Protect the distinction between the event record, a causal hypothesis and an action proposal. Include organisational decisions such as competing targets, maintenance resources and change control when the evidence supports their involvement.

Learning requires an action owner, a defined change and evidence of effectiveness. A revised instruction that nobody can use is not an implemented improvement. Verify the changed interface or process under representative conditions and examine later results. An increased number of reports can reflect better reporting rather than more incidents; retain reporting definitions, exposure and the willingness to report when interpreting a trend.

Follow a benign incident from observation to a usable change

Use a separate invented training desk that hands out reusable art-supply boxes. Its record contains full box identifiers BX-170 and BX-179, but display A truncates both to BX-17…. The supplied event log shows selection of BX-179 for a request for BX-170 after an interruption. A colleague compares the full paper identifier and catches the mismatch before handover. Staff accounts say the visible prefixes looked identical. These observations establish a near miss and an ambiguous information display; they do not establish intent, an error probability or every contributor to the selection.

The supplied release record adds an organisational condition: approval checked that a result arrived, but did not include a handover task using similar identifiers or interrupted work. Staff who perform the task were not included in that check. This explains a gap in the acceptance process, rather than assuming that either the software team or the clerk knew the later outcome in advance. The table links those supported conditions to a change and an owner.

The interface owner implements display B with the full selected identifier available visibly and through the agreed accessible text, plus an explicit comparison with the requested identifier before handover. The service owner involves desk staff in defining representative interrupted and similar-identifier tasks and in assessing whether the added comparison is workable. The release owner makes these task records part of the next acceptance decision and allocates time to perform them. A reminder to be careful alone would leave the original ambiguity and missing acceptance case intact.

Original learning record: observation, proposed response and evidence have different roles.
Supported condition	Change and accountable role	Evidence needed
Both identifiers displayed as the same prefix	Interface owner preserves full identity and makes the requested/selected comparison available	Known A/B configuration; full visible/accessible identifier and selected-record agreement
Acceptance omitted similar-ID and interrupted handover tasks	Service/release owners agree task cases with staff before acceptance	Retained cases, participant/task conditions, anomalies and release decision
A near miss was caught before handover	Service owner retains discrepancy and escalation paths	Task outcome records distinguish correct completion, stopped mismatch and unresolved difficulty

Calculate the backlog without predicting errors

Start with eight unfinished tasks. Supply incoming work at six tasks/h and qualified completion capacity at four tasks/h, with comparable tasks and constant rates. Net accumulation is two tasks/h. After four hours, backlog is 8+(6−4)×4=16 tasks.

For a separate configuration with qualified capacity eight tasks/h, net removal is two tasks/h and the same initial backlog reaches zero after four hours. The model assumes the additional capacity has suitable resources and retains the checks. It supplies no human-error probability, safe staffing level or instruction to hurry. Its purpose is to reveal a persistent demand-capacity mismatch.

Now supply a completed B evaluation record: six distinct task conditions, each represented by four scripted handovers, give 24 tasks. Conditions include similar identifiers, interrupted work, a changed selection, keyboard-only navigation, an unavailable visual display with alternative text access and restart. The stated rule is that an item must not be handed over with an unresolved identifier mismatch; a separate service goal seeks independent correct completion. The record includes the exact B version, requested and selected identifiers, available information and outcome for each task. No wrong-identifier handover occurs. Twenty-two tasks finish correctly without assistance; in two interrupted tasks the user cannot confidently resume and stops for help.

The recorded interface and mismatch-response criterion is met in these 24 supplied tasks, but the independent-completion goal is unresolved for those two cases. The owner keeps them open, examines what state is missing after interruption and revises the recovery design before repeating the affected cases. It would be misleading to close every action because the display text was changed. These constructed, selected tasks are not independent random samples of all work: neither zero wrong handovers nor 22/24 completion estimates a population error rate or proves the redesign alone caused an improvement.

Try the reasoning

Why is “operator error” an incomplete stopping point?
Show the answer and reasoning
It names a contribution but does not explain task, equipment and organisational conditions that made the event possible or how to improve them.
Can retraining alone eliminate a slip caused by two indistinguishable controls?
Show the answer and reasoning
It may help awareness, but the confusing design remains. The task and interface should be examined directly.
Does a rise in reported near misses necessarily mean performance worsened?
Show the answer and reasoning
No. Reporting practice or willingness to report may have changed. Comparable definitions, exposure and context are needed.
In a later case the full requested and selected identifiers are correctly displayed, but a wrong physical box is chosen because the two shelf labels are indistinguishable. Does that finding justify only repeating the display-text change?
Show the answer and reasoning
No. The new evidence locates an ambiguity in the physical task/interface that the earlier display change does not remove. Retain what still works, examine label-to-record mapping with the staff who retrieve the boxes, assign the relevant physical/service owner and define a representative retrieval check. Do not infer a new error probability or blame a person from this one event.
A report says “B passed because no wrong handover occurred in 24 tasks.” What should the action record also retain?
Show the answer and reasoning
It should retain the exact criterion and configuration, the six selected conditions, the two tasks that stopped for help, and the separate independent-completion goal. The mismatch criterion has bounded supporting evidence; the interruption-recovery gap still needs an owner and follow-up. Counts of selected tasks are not a reliability probability, and an implemented change is not the same as complete effectiveness.

What to carry forward

Explain performance in its working conditions, then verify that the resulting change is usable and effective.

References

Teaching resources and research supporting this lesson’s explanations and examples.

Health and Safety Executive3 sources