Machine vision / Cost of quality
The Real Cost of a Defect Escape: Warranty, Recall and Customer PPM
A defect costs roughly ten times more to fix at every stage it is allowed to advance. Prevent it at the machine and it costs one unit. Catch it inside your own four walls and it costs ten. Let it reach the customer and it costs a hundred, and that hundred is not a repair bill, it is return freight plus replacement plus an RMA investigation plus a containment sort at your cost plus a PPM entry that follows you into the next quotation. This article puts ringgit figures on each stage of that stack, shows what your customer's PPM scorecard is really measuring, and reframes inspection as the cheapest insurance premium on the line.
The numbers at a glance
01 / The rule of ten
A defect never gets cheaper to fix
Quality engineering has one rule of thumb that survives every industry it is tested in. The money, time and effort needed to fix a defect multiply by roughly ten at each stage it is allowed to advance: about one unit of cost to prevent it at the source, ten to catch it later inside your own process, a hundred once it escapes to the customer. The dollar signs stand for orders of magnitude rather than literal invoices, and the ratio is directionally consistent across industries: prevention at the machine is an order of magnitude cheaper than internal rework and two orders cheaper than an escape.
Read that carefully, because the implication is not the obvious one. It is not that quality is expensive. It is that where you catch a defect matters far more than whether you catch it. Two plants with identical defect rates and identical inspection budgets can differ by an order of magnitude in cost of quality purely because one detects at the station and the other detects at the customer.
On an electronics line the ladder is easy to see. A missing 0402 caught before the oven costs an operator ten seconds. The same part caught after reflow costs a rework station, a soldering iron, an unplanned thermal cycle and a rework count against a customer specification that typically caps cycles per location at two or three before pad lift becomes the argument. The same part caught by your customer costs everything in the next section.
02 / The cost stack
What one escape actually triggers
This is the table to put in front of anyone who thinks inspection is an overhead. Each row is the same single defective unit, caught at a different point.
| Caught at | What it costs | Who absorbs it | Order of magnitude |
|---|---|---|---|
| Prevented at the process | A parameter adjustment. No unit is consumed and no labour is spent recovering value | Process engineering | 1× |
| In-process, pre-reflow | Seconds of operator time. The part is still cheap to move and the assembly has little value added | The line | ~1-2× |
| Post-reflow, in-plant | Rework station time, thermal cycle, a rework count against IPC-7711/7721 limits, or scrap of a unit carrying full processing cost | Manufacturing | ~10× |
| Final test / audit | Disassembly, retest, re-labelling, schedule disruption, and the value of everything added after the defect occurred | Quality and planning | 10-30× |
| Customer incoming inspection | Return freight, replacement unit, RMA administration, an 8D investigation, and a PPM entry on the supplier scorecard | You, visibly | ~100× |
| Customer's production line | Line-stop charges, sorting of finished goods at your cost, premium freight for replacement stock, containment reporting, and an on-site audit | You, expensively | 100-1,000× |
| The field | Warranty claim, service labour, recall exposure, supplier recovery claim, and the commercial cost of a scorecard downgrade at re-quotation | You and your insurer | Uncapped |
Magnitudes follow the 1-10-100 convention, extended for the containment and field stages that dominate automotive and semiconductor supply. Substitute your own recovered costs from the last twelve months of RMA records; most plants find the real multiplier higher than they assumed.
The row that changes budgets is the sixth one. A defect that stops your customer's line is not priced as a defect, it is priced as a disruption, and the invoice includes people you never employed. That is the row that turns a RM 4 component into a five-figure event, and it is the row that never appears in a cost-per-inspection calculation.
03 / The hidden total
Cost of poor quality is 15 to 20 percent of revenue
Individually, escapes look like exceptions. Aggregated, they are one of the largest line items in the business. According to the American Society for Quality, most companies carry quality-related costs equal to 15 to 20 percent of total sales revenue, with high performers holding below 10 percent and troubled plants exceeding 30 percent. The widely cited Juran figure for cost of poor quality lands at 10 to 30 percent of revenue in mature manufacturers.
The reason nobody argues about this number is that nobody sees it. It does not appear as a line called quality. It hides in scrap accounts, rework labour, warranty reserves, containment activity and the soft cost of firefighting, spread across ten different cost centres. Armand Feigenbaum's term for it was the hidden factory: the 15 to 20 percent of production capacity consumed by rework and correction that never appears in standard accounting. You are already running that factory. It has no revenue.
There is a second insight buried in the framework, and it is uncomfortable. Most struggling plants spend the large majority of their quality budget on failure and appraisal, which means paying to make bad parts and then paying to find them. Detection alone plateaus, because detection does not reduce the rate at which defects are produced, it only increases the rate at which they are found before shipment. Inspection is the gate that stops the escape. Process control is what stops the defect. You need both, in that order of urgency and the reverse order of investment.
04 / Warranty and recall
The bill your customer sends back to you
Warranty is where escapes become a published number. Warranty costs of large manufacturers typically average about 2 percent of revenue, and the automotive industry-wide claims rate sat at 2.0 percent of vehicle sales revenue in 2023, with a five-year average claims rate of 2.23 percent. Two percent of revenue is not a rounding error. For most E&E suppliers it is comparable to the entire net margin.
If you supply components rather than finished goods, the temptation is to read those figures as somebody else's problem. That reading is out of date. OEMs have become markedly more diligent about supplier recovery, and the scale is instructive: in the Chevrolet Bolt battery recall, General Motors recovered 2.7 billion US dollars in claims costs from its battery supplier, about 90 percent of the total cost of the recall. A recall is a chain event, and the chain is now routinely walked back to the component that failed.
The arithmetic of a recall is also unforgiving in a way that surprises people. A trivial part on a large population is a large number: a recall requiring a RM 45 part and eighteen minutes of dealer labour, applied across a single model year, comfortably clears eight figures once notification, logistics, legal exposure and buy-backs are counted. Nothing about that cost is proportional to the part. It is proportional to how many units shipped before somebody noticed.
Which is the real argument for inspecting at line rate rather than sampling. Sampling limits your knowledge to the units you looked at. Escape populations are defined by the units you did not.
05 / Customer PPM
The scorecard is the real specification
Every discussion about acceptable quality ends at the same place: the number on your customer's supplier scorecard. PPM, defective parts per million, is the standard language of that scorecard, and every automotive OEM communicates quality targets in PPM. IATF 16949 does not mandate a specific figure, but it requires that you monitor performance, define targets and demonstrate improvement, which in practice means the customer sets the number and you meet it.
| Sector or tier | Typical target | What it means in units |
|---|---|---|
| Medical devices | <10 DPPM | Fewer than 10 defective parts in every million shipped |
| World-class supplier | <25 PPM | On 10 million parts a year, roughly one defective part per production day |
| Automotive Tier 1 | <50 DPPM | 0.005 percent. A 500,000-part shipment may contain 25 defects |
| Electronics assembly | <100 DPPM | 0.01 percent, which most scorecards still describe as room for improvement |
| Six Sigma reference | 3.4 DPPM | The theoretical world-class figure, genuinely reached only by the most disciplined processes |
| Automotive semiconductor | 10 DPPB | Parts per billion. The bar moved from 10 DPPM because a vehicle carries close to 1,000 ICs that must last fifteen years |
| Corrective action trigger | >1,000 PPM | 0.1 percent. Above this, most customers open a formal corrective action |
Sources: published sector benchmarks for DPPM targets and the automotive semiconductor move toward parts-per-billion screening. Your contractual target is whatever your customer's scorecard says, which is frequently tighter than the sector benchmark.
Two of those rows deserve attention from anyone shipping out of Selangor or Penang. The automotive semiconductor row is the direction of travel for the whole sector: where 10 DPPM used to be a solid benchmark, the bar is now 10 DPPB, three orders of magnitude tighter, driven by the roughly 1,000 integrated circuits in a modern vehicle that must survive fifteen years of service. Sampling cannot measure a target like that. Only 100 percent inspection with a per-unit record can.
The corrective action row is the commercial one. Crossing it does not simply cost you the parts. It costs an 8D, a containment plan, a customer audit, controlled shipping status, and in many cases a hold on new business awards until the trend closes. That is the moment a quality problem becomes a sales problem.
06 / The escape maths
Your detection rate is your PPM
Here is the calculation that connects inspection method to scorecard. Escapes are not accidents, they are arithmetic: volume, times defect rate, times the share of defects your gate fails to catch.
Take a line shipping 200,000 units a month at a 0.1 percent defect rate. That is 200 defective units produced monthly, which is entirely normal and says nothing bad about the process. What decides your scorecard is the gate. Manual visual inspection detects 70 to 80 percent of defects under production conditions, so at the optimistic end 40 units ship defective: an escape rate of 200 PPM. Automated optical inspection above 95 percent detection lets roughly 6 to 10 ship: 30 to 50 PPM.
| Inspection gate | Detection | Units escaping / month | Escape PPM | Scorecard position |
|---|---|---|---|---|
| Sampling audit only | Partial by design | Unknown, and unbounded between samples | Unmeasurable | Cannot demonstrate control |
| Single manual inspector | 70-80% | 40-60 | 200-300 | Above the corrective-action line for most automotive customers |
| Two inspectors in tandem | ~96% | 8 | ~40 | Meets Tier 1, at double the labour cost |
| AOI, programmed classes | 95-99%+ | 2-10 | 10-50 | Meets Tier 1 and produces the per-unit evidence to prove it |
Now price the difference. Between a single manual inspector and a tuned AOI gate, this line avoids roughly 34 escaped units a month. At a conservative RM 250 per escape, covering freight, replacement, RMA administration and sorting, that is RM 8,500 a month, or RM 102,000 a year, before a single containment event. One customer line-stop or one sort of finished goods pushes it several times higher in a single month. The full break-even arithmetic, including machine and verification costs, is set out in what an AOI machine costs in Malaysia.
The escape you should price is not the average one. It is the worst one your product can cause, multiplied by the probability that your current gate misses it. That is how insurance is priced, and inspection is insurance.
07 / On your parts
Inspection as the cheapest premium
Framed as a cost centre, inspection loses every budget argument. Framed as a premium against an uncapped liability, it wins most of them. The premium is the machine, the verification labour and the programming time. The claim it protects against is a containment event, a recall recovery, or a scorecard downgrade at re-quotation. No plant buys fire insurance by comparing it to the sprinkler bill.
What decides whether the premium is worth paying is whether the gate is tuned to your parts. This is how CODETRACE machine vision and AOI is built: models trained on the customer's own good and defective parts, thresholds set against your acceptance criteria and your customer's standard, and JOVIS, our robotic vision inspection platform, providing micron-level geometry analysis where the defect is defined by height rather than appearance. Every unit carries an image and a measurement, which is also the containment evidence you will need the day a customer asks what else shipped.
Systems are deployed and tuned on site across Selangor, the Klang Valley and Batu Kawan, for semiconductor, electronics and automotive producers working to export customer PPM targets. CODETRACE is a member of the NVIDIA Inception programme. Bring the defect that reached your customer, and the honest answer about which gate stops it comes before any quotation.
08 / Where to start
Price your last escape, properly
Do not start with a machine. Start with your most recent escape and cost it honestly, including the hours nobody logged. Return freight and replacement. The 8D and the meetings. The sort, and who paid for it. The premium freight to recover the schedule. The engineer who spent two weeks on containment instead of the next product. Most plants find the total is three to five times the figure that ever reached a spreadsheet.
Then divide it by the number of units that gate inspects in a month. That is your true cost per unit of not inspecting, and it is the only figure that makes an inspection business case comparable to the alternative. Once you have it, the equipment question becomes a straightforward comparison: which gate can see this defect, at line rate, for less than that.
For the method comparison see AOI vs AVI vs manual QC, for the detection ceiling behind the escape arithmetic see your inspectors catch four in five defects, and for the operating cost that decides whether the gate stays switched on, the AOI false call rate model. All of it sits inside a wider factory automation plan, and quality is usually the station that justifies the rest.
You are not paying for the defects you catch. You are paying for the ones you ship.
FAQ / Cost of a defect escape
Questions, answered.
01What does one defect escape actually cost?
Cost multiplies by roughly ten at every stage a defect is allowed to advance. The 1-10-100 rule holds up well in practice: about one unit of cost to prevent a defect at the machine, ten to correct it inside your own four walls, and a hundred once it reaches the customer. On an E&E line that means a missing component caught before reflow costs an operator ten seconds, the same defect caught after reflow costs a rework station and a thermal cycle, and the same defect caught by your customer costs return freight, a replacement, an RMA investigation, a containment sort and a PPM entry on your scorecard. The precise ratios are illustrative, but the order of magnitude is consistent across industries.
02How much does poor quality cost a manufacturer?
The American Society for Quality puts total quality-related costs at 15 to 20 percent of sales revenue for most manufacturers, and the widely cited Juran figure for cost of poor quality runs 10 to 30 percent of revenue in mature manufacturers. High-performing plants with mature quality systems hold below 10 percent, while plants with serious quality problems exceed 30 percent. Most of it is invisible on the profit and loss statement because it sits in scrap accounts, rework labour, warranty reserves, containment activity and firefighting time that nobody books against quality.
03What PPM do automotive and semiconductor customers expect?
Automotive Tier 1 suppliers typically target under 50 DPPM, electronics assembly under 100, and medical devices under 10, with world-class suppliers holding below 25 PPM and anything above 1,000 PPM triggering immediate corrective action. Automotive semiconductor is the extreme case: where 10 DPPM was once a solid benchmark, the bar has moved toward 10 defective parts per billion, because a modern vehicle contains close to 1,000 integrated circuits that must survive fifteen years of service. The practical point for a Malaysian supplier is that the target is set by the customer's scorecard, not by your own process capability.
04Does a component supplier get charged for an OEM recall?
Increasingly, yes. OEMs have become far more diligent about supplier recovery, and the scale can be severe: in the Chevrolet Bolt battery recall, General Motors recovered 2.7 billion US dollars in claims costs from its battery supplier, about 90 percent of the total cost of the recall. Warranty costs of large manufacturers average around 2 percent of revenue, and the automotive industry-wide claims rate sat at about 2 percent of vehicle sales revenue in 2023. For a component supplier, that means the recall exposure is not somebody else's balance sheet problem.
05How does inspection reduce the cost of escapes?
By moving the point of detection upstream, which is the only variable in the 1-10-100 rule you control cheaply. Human visual inspection detects 70 to 80 percent of defects under production conditions, so roughly one defect in five structurally escapes. Automated optical inspection detects above 95 percent on programmed defect classes and produces an image and a measurement per unit, which also gives you the traceability to prove containment when a customer asks. CODETRACE builds machine vision and AOI inspection for semiconductor, electronics and automotive manufacturers across Selangor and the Klang Valley, trained on the customer's own parts, and is a member of the NVIDIA Inception programme.