Failure cost is designed in. The line just delivers it.

Walk past any scrap bin on any shop floor and you are not looking at parts. You are looking at decisions – a tolerance, a material call, a fastener count, a surface finish nobody could hold at rate – signed off months earlier by people who never saw the bin and never will. Every plant I have worked in, automotive or aerospace, the bin told the same story. The line does not create failure cost. It delivers it, on schedule, with full traceability if you ask nicely.

This month the trade press is running its usual circuit – AI vision inspection headlining the semiconductor shows, factory automation forecast to swell by another third of a trillion dollars within a decade – and, more interesting than any of it, a design-review software vendor touring the message that review discipline is a manufacturing cost lever. That last item is the tell. When companies that profit from inspection start selling upstream discipline, they are conceding what the scrap bin already knew: no amount of downstream intelligence recovers cost committed upstream.

Where failure cost is actually born

The cost-commitment curve is old engineering economics, and it has not aged a day. Roughly 70% of a product's failure cost is locked before the first part exists – in the tolerances, the joint design, the material stack, the fastener count settled in a 40-minute meeting that started late because the projector would not connect. Everything after that is bookkeeping. End-of-line metrics cannot see any of it. First-pass yield, PPM, scrap rate: they measure the residue of upstream choices, not the choices. A plant can hit every monthly quality target while running a design that taxes it €40,000 a month in sorting, containment and premium freight. Warranty then pays retail for decisions nobody in the plant remembers making.

At WITTE Automotive I inherited a warranty failure that had been fixed three times. Containment held, the A3s closed, the Q-Wall flagged every escape – and within months the failure came back. It died for good in a 45-minute PFMEA session, killed by one question from a process engineer: why does this tolerance exist? Nobody in the room could defend it. It had been inherited from a drawing two generations old; the function did not need it, the supplier could not hold it economically, and our own gauge could barely measure it. We changed the tolerance and the failure mode ceased to exist – no inspection, no sorting, no line-side quarantine. The QRQC, A3 and Q-Wall discipline around that work cut failure cost substantially. But the biggest single kill came from one question, asked upstream, by someone finally permitted to ask it.

Design review as theatre, not a gate

Most plants run design review as a signature carousel: a calendar invite, twelve attendees, forty minutes, a PDF on a screen, initials collected in sequence. The gate is recorded as passed. Nothing was gated. A real gate has four components, and I have audited enough IATF 16949 and AS9100 programmes to know their absence by smell. PFMEA linkage: every open action traceable to a named failure mode, not to somebody's comfort. Open-issue closure with evidence, not with the next meeting's agenda. A decision log – who decided, on what data, accepting which trade-off – so that when the escape surfaces two years later, the fingerprints exist. And the only component that matters under pressure: someone in the room with the authority to say not yet and the organisational backing to survive having said it.

A scrap bin is an archive of meetings that ran out of time.

Aerospace leans harder on this because the gate carries liability. Automotive keeps it theatrical because programme timing is treated as untouchable, and the review is the cheapest thing on the altar.

Moving upstream without starting a war

Design engineering will defend its meetings the way production defends cycle time, so do not arrive with a flip chart and a sermon. Three moves actually work. Seat a quality engineer at release with a genuine veto – advisory attendance is how the carousel got its signatures. Run A3 on design changes the way you run 8D on escapes: containment, root cause aimed at the decision process itself, corrective action against the review that let the change through, verification that the gate now behaves differently. And measure prevention with the same rigour you measure scrap – failure modes killed at PFMEA, tolerance challenges sustained, open-issue age at release, cost avoidance posted in euros beside scrap in the same monthly pack.

At SNOP we cut defect cost by 70%, and the part of that number I am proudest of came less from inspection – though we did build the QA/QC organisation from bare concrete for 900 people – than from QRQC findings that traced escapes back to design-release decisions and then changed the release discipline itself. The QRQC room found the symptom within 48 hours. The money was in the meeting upstream. Across two decades of APQP and gate audits, one constant holds: most plants account for prevention as an expense and scrap as a surprise.

None of this argues against the vision systems. They will keep improving, and this month's metrology press reads like an arms race of catch rates. But a 99.9% catch rate is a recurring cost on every part, forever, while a tolerance decision is a one-time act. Inspection is rent. Design discipline is purchase.

The inspection AI will keep getting better, and it should. It will still never be cheaper than the meeting you keep cancelling. When the next warranty campaign lands or the next bin fills, walk it backwards – past the line, past the supplier, past the containment – until you reach a room, a projector and forty minutes. Audit the room where the cost was committed, not just the line that absorbed it.