This week's report on heat stress in Cambodian garment factories describes plants that clear every compliance audit on the books while the heat quietly rewrites the process underneath them: hands slower by mid-shift, error rates climbing, workers ending the shift in a condition no checklist row captures. Here is the uncomfortable part. An automotive or aerospace audit would pass those same plants. ISO 9001 parks ambient conditions in clause 7.1.4 – the dustbin clause on the environment for the operation of processes – and IATF 16949 adds barely a sentence. Temperature is filed under welfare. Fatigue is filed under welfare. Neither ever reaches the control plan, which is exactly where a 34 °C afternoon does its work.

I know this because I ran quality and health and safety as a single department at SNOP – a 900-person greenfield plant, one leader owning the defect pareto and the incident log at the same time. They told one story. Two chapters, one plot.

Heat is a process input, not scenery

Start with the physics, because physics does not care how we file paperwork. Steel grows about 11.5 microns per metre per degree Celsius; aluminium roughly twice that. A 300 mm aluminium part straight off a 34 °C line, measured immediately in a 20 °C room, carries about a tenth of a millimetre of pure temperature in the reading – a third of a ±0.15 mm tolerance band spent before the process itself has contributed a single micron. This is why ISO 1 fixes 20 °C as the reference temperature for dimensional measurement, and why, when I built the QA and QC laboratories from scratch at that plant, the first room I specified held 20 °C around the clock.

One summer the gauge study on our bore gauges drifted from 19% to 29% study variation in ten weeks. Same gauge. Same appraisers, same parts. We spent a fortnight accusing the gauge of instability, then the operators of inconsistency. Then a €40 data logger clipped beside the bench read 20.1 °C at 07:30 and 27.6 °C at 15:00 – the air handling could not hold setpoint against a south-facing wall. The gauge was fine. The room was a process variable wearing a lab coat.

The operator is an input too

The gauge story has a human twin. Plot the defect pareto by hour of day instead of by defect code and watch what shape appears. Ours climbed from early afternoon: escapes, missed torque verifications, mislabelled containers – the error families that live on sustained attention rather than machine capability.

The incident log had the same silhouette. Near-misses clustered late shift; dehydration headaches stacked up in the first-aid book.

Two charts, one department – the only reason anyone ever laid them on top of each other. In most plants the overlay is structurally impossible: quality reports to one director, safety to another, and the shared cause dies in the white space between two organisational boxes. I did not need Phnom Penh to learn that. A Polish July in a plant full of presses and welding cells supplies all the evidence a pareto requires.

The fix was not a leaflet. We wrote the heat protocols as reaction plans, in the same grammar a control plan uses for any other input: a measured trigger rather than a feeling, actions with a name against them – rotation, scheduled hydration, first-off and final dimensional checks moved into the cooler hours – and a record somebody audits. The afternoon bump in the hourly pareto flattened without a single speech about quality culture. We changed an input; the output followed, as outputs do.

Why an audit cannot see a season

An audit samples a moment. A window of days, booked months ahead, walked mostly at mid-morning, ambient conditions reduced to a checkbox – lighting adequate, temperature adequate – initialled on professional opinion. Q4 surveillance visits are being booked now, which means a fair share of the world's certified plants will be examined at the most forgiving point of their thermal year. Nothing sinister in that. An audit is an instrument with a sampling plan, and the sampling plan does not span August.

So treat the environmental envelope the way you already treat calibration. Log temperature at the critical stations, with sensors that sit on the MSA calendar themselves. Repeat the GR&R in the coldest month and again in the hottest, and treat the drift between the two studies as a finding. Put ambient conditions into the PFMEA as causes, with occurrence rated against the worst month rather than the audit week. Write the heat protocol as a reaction plan and audit its records of use like any other control.

An audit is a photograph of your factory at ten in the morning. Your defects develop at three in the afternoon.

Key takeaways

  • Log ambient temperature beside every critical gauge and station, with calibrated sensors that sit on the MSA calendar – data, not the "adequate" checkbox.
  • Repeat the GR&R in the hottest and the coldest month of the year; drift between the two studies is a finding about the facility, not the appraiser.
  • Write heat protocols as control-plan reaction plans: measured trigger, defined action, named owner, auditable record. A leaflet has none of the four.
  • Add environmental conditions to the PFMEA and rate occurrence against the worst month of the year – a rating based on the audit week is fiction.

Conditions are a question of quality ownership before they are a question of welfare. The moment temperature moves a dimension or moves a defect rate, it is a process input, and process inputs belong to the people who own the process – in the PFMEA, the control plan, the MSA calendar. The Cambodian plants in this week's report are not outliers; they are the control group. After the record northern summer that just ended, every plant on the planet ran some version of the same experiment, certified or not. Until heat has a line in the PFMEA, the audit keeps passing at 10 a.m. and the process keeps failing at 3 p.m. in August.