The newest plant wins on quality. Nobody's broken it yet.

Here is an uncomfortable observation. When Jim Farley said Mexican and Chinese auto plants are outpacing North American facilities on quality, half the industry reached for the labour-cost explanation and the other half reached for the tariffs. Both sides are arguing about geography while the actual variable is sitting in a binder labelled "Deviation Log 2008–2019" in a back office nobody has visited since the plant manager who opened it retired.

I have built a greenfield quality system for a 900-person plant. I have watched the same standard — AS9100 — produce radically different outcomes at sites that are, on paper, identical. The difference is never the postcode. It is how much institutional scar tissue a facility has accumulated and how much of that has hardened into permanent exception.

What a greenfield quality system gets right by default

When I stood up the QA/QC department at SNOP from scratch — no prior workforce, no inherited procedures, no dusty calibration cabinet with mystery gauges — the advantage was not enthusiasm or budget. It was the absence of compromise. Every control plan I wrote reflected the process as it was actually designed to run. Every escalation path terminated at a decision-maker who was alive and employed. PFMEA sessions were built by people in the room because they understood the failure modes, not because they had inherited a document from a supplier audit in 2014 and updated the date field.

A new plant's quality system works because it has not yet been negotiated down. Nobody has widened the tolerance on a torque spec because a maintenance manager argued the tooling was "basically fine." Nobody has replaced the containment procedure for a suspect batch with "call quality and see what they say." The 8D has not been quietly downgraded to an email chain.

This is why Farley's benchmark plants perform. Not because of wages. Because many of them are new. New facilities get discipline for free, and discipline is what produces the defect-cost numbers I have lived through — 70% reductions, zero critical escalations in a quarter, audit findings halved in a single cycle.

Your oldest plant runs on exceptions nobody remembers approving

Here is what happens to a quality system over fifteen or twenty years. Each year, a handful of deviations get approved. Some are legitimate — a tooling modification that actually improves capability, a supplier change that is properly validated. But a meaningful percentage are compromises made under delivery pressure, and the compromise is never reversed. It gets extended. Then extended again. Then someone updates the control plan to reflect the compromise, and the compromise becomes the standard. Nobody notices because the person who approved the original deviation left in 2019.

I call this exception creep, and it is the single most underestimated quality risk in legacy manufacturing. I have walked into plants running AS9100 where the documented process and the actual process barely resemble each other. The routing says one thing. The floor does another. The quality team has a procedure that references a gauge decommissioned three recalibration cycles ago. Compliant on paper, compromised in practice — and the next external audit is the first time anyone reads the procedure end to end.

A quality system does not collapse. It is negotiated away one signature at a time.

The mechanism is not malicious. Operators adapt to solve real problems. Engineering signs off because the alternative is stopping a line that is behind schedule. Quality concedes because fighting every battle costs political capital that is finite. The result is a plant that looks like it is running the standard but is actually running a patchwork of thousands of individual decisions, each one defensible in isolation, collectively corrosive.

How to audit for system decay

The standard surveillance audit will not catch this. AS9100, IATF 16949, ISO 9001 — they verify that a system exists and is followed. They are not designed to detect that the system itself has quietly degraded. You need a different approach, run by someone who has built systems from zero and therefore knows what intact looks like.

Start with routing verification. Pull the top five production routings in your highest-volume plant and walk every step on the floor. Compare what the routing says should happen with what actually happens. Document every discrepancy. If you find more than three per routing — and in a legacy plant, you will — you have system decay. At Airbus, routing verification KPIs alone collapsed internal lead time by 97% when applied with discipline. The tool is not exotic. The discipline is.

Pull the last twenty closed nonconformances and trace whether the corrective action actually addressed the root cause or simply contained the symptom. Look for 8D reports where D5 and D6 — permanent corrective action and prevention — are copy-pasted from D4. That pattern tells you the organisation is going through the motions on structured problem-solving. QRQC and A3 are only effective when the people running them have the authority to stop production. In a decayed system, they do not.

Audit the exception register. Count the open deviations, check their average age, find out how many have been extended more than twice. If the answer to any of these is "we are not sure," the system has already surrendered ground it does not know it lost.

Geography is not the answer

Farley is right that something is broken. He is wrong about what. You can move a plant to Sonora, to Shenzhen, to anywhere you like — within five years, if the quality architecture is allowed to erode the same way, you will have exactly the same problem in a different time zone. I have seen the pattern repeat across automotive and aerospace, across continents, across standards. The variable is not where you build. The variable is whether the system you designed on day one is still the system running on day four thousand. Go audit your oldest plant. Find out what it has quietly surrendered. That is where your next recall is hiding.