Quality · 2026

Warranty is the most expensive inspection station in the plant

Automotive quality is usually discussed as inspection: cameras, gauges, sampling plans. The number that decides whether a programme is profitable sits at the other end of the process — in what the customer finds after delivery.

The number that sets the ceiling

Warranty claims are the least ambiguous quality metric in the industry, because they are paid in cash. The industry-wide claims rate was around 2.0% of vehicle sales revenue in 2023, with a long-run global passenger-vehicle average nearer 2.5%. Ford, the highest of the major US automakers, runs near 3.5% — for every $100 of vehicle revenue, roughly $3.50 goes to fixing something that broke. Suppliers carry their share of that through recovery claims, which is why a tier-one’s quality performance is a commercial position, not a departmental KPI.

2.0%

industry-average warranty claims rate as a share of vehicle sales revenue (2023)

~2.5%

long-run global passenger-vehicle average

20–30%

typical error rate of manual visual inspection

Manual inspection has a known error rate

Manual visual inspection is generally reported with an error rate of 20–30%: inspectors differ from one another, and the same inspector differs from themselves across a shift. That is not a criticism of inspectors; it is a property of sustained human attention. Automated visual inspection reports detection accuracy above 90% across multiple defect types, and one comparative study recorded roughly 30% higher detection accuracy and 50% shorter inspection time, with inspector-to-inspector variability removed entirely. On a line running faster than a few parts per minute, this stops being a preference.

But sorting is not quality

A vision system at the end of the line sorts good parts from bad ones. It does not make fewer bad ones. Scrap still costs material, machine time and capacity; the process that produced the defect is unchanged. The familiar rule of thumb in quality engineering is that the cost of a defect grows by roughly an order of magnitude at every gate it passes — cheap at the station that made it, expensive at final assembly, and worst of all in the field. Inspection buys containment. Prevention is where the money is.

Moving the decision upstream

1. 01

   In-process, not end-of-line

   Measure the characteristic while it can still be corrected, on the station that produces it.

2. 02

   Parameters, not only parts

   Monitor the process variables that generate the defect — temperature, torque, tool wear — not just the defect itself.

3. 03

   One owner per failure mode

   The top failure modes each get a named engineer, a due date and a closed loop, reviewed weekly.

4. 04

   Field data flowing back

   Warranty findings and 8D outcomes update the control plan, not just the report that files them.

Digital transformation, honestly framed

Most quality digitalisation programmes fail in the same way: a platform is bought, dashboards are built, and the loop from anomaly to corrective action still has no owner. The alternative is unglamorous and works. Pick one characteristic on one critical process. Instrument it where the decision can still change the outcome. Agree the target — parts per million, scrap rate, warranty claims on that failure mode — before the work starts, and hold the deployment to it. Then do the second one.

If your quality data cannot answer “which process parameter moved before this defect appeared?”, you are measuring outcomes, not causes.