The recall is the treatment. The launch is the disease.

Ford crossed eighty recalls by mid-October. One campaign every three and a half days, with the F-150 appearing in them again and again. Here is an uncomfortable observation from someone who has sat on the receiving end of that arithmetic: nobody gets to eighty with eighty unrelated problems. The number is a systems diagnosis wearing a news headline. Volkswagen was pulling thousands of cars for battery-fire risk the same month; Chery had a type approval suspended over a rear axle. Marques change. Mechanisms do not. Eighty campaigns are the visible output of a launch system whose gates stopped gating, billed to the regulator and the customer instead of to the programme that earned them.

Eighty recalls, a handful of root causes

Recalls cluster – around launches, around design changes, around new suppliers and new software. When I took over quality at SNOP I inherited critical customer escalations. The kind that arrive with red headers and a standing weekly call. Within a quarter we were at zero. Not because I closed every complaint, but because we stopped treating complaints as the unit of work: underneath the red headers sat three genuine failure classes, each masquerading as a dozen separate fires.

Every 8D on my desk closed an instance. The class stayed open until someone owned it as a class, killed the mechanism and verified the kill across every part number it could touch. Recurrence is the tell. A manufacturer that recalls the same subsystem twice, or recalls different subsystems sharing a supplier, a change point or a launch window, is not unlucky – it is telling you exactly where its launch discipline is thin.

Leading quality initiatives across a 2,000+ workforce, multi-site, taught me the same lesson at altitude: recall patterns are an organisational signature, not weather. Bad luck does not repeat quarterly along the same part families. Habits do.

Where recalls are born: the launch system

By the time a defect reaches a recall, every question being asked is downstream. The answers were decided upstream, in the rooms where the gate reviews lived.

Start with the gates. A phase exit that has never sent a programme back is a signature collector, not a gate, and a readiness review that cannot fail cannot protect anything.

The PFMEA tells the same story one document later: the failure mode appears in the file one revision after it appeared in the field, because nobody re-rated it at the last engineering change, the new tool, the second source. Suppliers come next. The copy tool is quoted, ordered and running at rate before anyone has demonstrated the process at rate, and a PPAP signed on a pre-series line at half cycle is paperwork, not proof. And when the calendar presses, the test plan gets thinned until it fits the date. The residual risk rides out in the customer's car.

I built a greenfield QA/QC department for a 900+ employee plant, and the first thing a greenfield teaches you is that readiness is a list of conditions, not a feeling. Either the process demonstrated capability at rate on the real line, or it did not. VDA 6.3 asks precisely this question, in less friendly language.

The invoice nobody sends

The rule of ten holds in euros. A defect caught at the wall costs single digits – scrap, sort, an hour of QRQC. The same defect caught by a dealer costs hundreds of euros per vehicle: parts at warranty rates, retail labour, logistics both ways, loaner cars, regulatory reporting, engineering hours diverted from the next launch. Multiply by six figures of vehicles. Then add the line that never reaches a ledger, the customer who quietly buys something else next time.

A recall is the launch system's own handwriting, returned to sender at dealer rates.

Containment feels decisive – airfreight, red bins, a war room, visible motion. Prevention feels slow: a PFMEA session, a supplier readiness audit, a gate that holds a programme two weeks. The accounting only catches up at warranty, which is why the fast option wins the meeting and the slow option wins the year.

At SNOP we took defect costs down 70% with QRQC and A3 – fast containment at the wall, structured problem-solving behind it, a board that made recurrence impossible to hide. That was not heroics. It was the arithmetic of catching classes at the wall instead of instances at the dealer.

Own the class, not the case

The public recall database is a PFMEA input this industry leaves on the table: every entry is a failure mode with a real occurrence rating and a detection ranking measured in field exposure. A quality director needs one Monday morning for it. List three years of your own campaigns, map each against launch events, engineering changes and new sources, and look for the classes. Then give each class an owner at director level and put recurrence on the wall where the plant can see it. That is the whole checklist.

The recall is the treatment. The launch is the disease. Eighty campaigns in a year are not eighty verdicts on eighty parts – they are one verdict on a system, delivered in regular instalments. Fix the launches: gates that gate, FMEAs that anticipate change, suppliers proven before volume. Then the recall count becomes what it should always have been, a number nobody outside your company has a reason to know. Until then, the public keeps auditing it for free.