Most plants track scrap rate, report it monthly, and are surprised when the number improves while the P&L does not. The reason is arithmetic: scrap is the cheapest form of poor quality to see and often the smallest share of what poor quality actually costs. The parts you throw away are visible. The hours spent reworking, re-inspecting, re-setting, chasing a supplier, and apologising to a customer are spread across a dozen cost centres where nobody adds them up.
The short version: stop optimising the scrap number and start costing the full failure. When you total the four buckets below, the ranking of your problems usually changes — and the fix that mattered most is rarely the one the scrap report pointed at.
The four buckets of the cost of poor quality
Quality cost splits into four categories, and the split is the whole point, because money moves between them.
Internal failure — everything that goes wrong before the part leaves: scrapped material, rework labour, re-inspection, sorting a suspect lot, the machine time lost re-running a job, the expedited replacement material. This is where most of the hidden money sits.
External failure — everything that goes wrong after the part leaves: returns, warranty, credits, containment at the customer's site, freight both ways, the engineering time spent on the 8D, and the commercial damage that never appears on any ledger. External failure is dramatically more expensive per part than internal failure, which is why detection at the last possible moment is a bad strategy even when it works.
Appraisal — the cost of finding out: inspection, gauging, test equipment, audits, the sort operation you added last quarter. Appraisal is not free and it is not prevention. It buys information, not conformance.
Prevention — the cost of making the failure unlikely: process capability work, fixture and tooling improvements, mistake-proofing, training, maintenance that keeps a process in control, better supplier qualification.
The pattern worth internalising is directional, not numerical: money spent in prevention reduces spend in the other three, and money spent in appraisal mostly moves failures from external to internal. Any specific ratio you have seen quoted is somebody else's plant — measure your own.
Why scrap rate is the wrong headline metric
Scrap rate answers "what did we throw away?" It does not answer "how much of what we made was right the first time?" Those are different questions, and the gap between them is where rework lives.
A line running at 2% scrap can be running at 80% first-pass yield if a fifth of the output goes through a touch-up station that nobody counts as a defect because the part eventually ships. That touch-up station has an operator, a bench, a queue, and a lead-time penalty, and it is invisible in the quality report because the metric was chosen badly.
First-pass yield — the proportion of units that pass through the process correctly with no rework, no touch-up, no re-test — is the honest number. It is uncomfortable when you first measure it, which is the sign that it is telling you something the old metric was not.
Two related habits make it more useful. Roll first-pass yield across the whole process rather than per station, because a chain of individually respectable stations can still deliver a poor overall yield. And connect it to your throughput picture: rework consumes capacity, so a quality problem often shows up first as an availability or performance problem in your OEE numbers rather than as a quality complaint.
Detection fixes vs prevention fixes
When a defect appears, the fast response is almost always detection: add an inspection, add a check sheet, add a second set of eyes. It works, it is cheap to implement, and it is why plants accumulate inspection steps like sediment.
The problem is that detection fixes have running costs forever and decay quietly. Inspectors get tired, the check sheet gets filled in retrospectively, the extra step gets dropped the first week the line is behind. A prevention fix — a fixture that only accepts the part one way, a connector that cannot mate incorrectly, a sensor that stops the cycle if a step was skipped, a process setting that no longer needs an operator judgement — costs more once and then costs nothing.
The practical discipline: for every corrective action, write down whether it prevents the defect or detects it. If your last ten corrective actions were all detection, your quality cost is going up even though your defect reports look busy. Mistake-proofing at the source is the single highest-leverage move available to most small and mid-sized plants, and it usually costs less than people expect because it is a bracket, a stop, a keyed feature — not a system.
Tracing a defect to the process step that made it
Reducing scrap requires knowing where it was created, not where it was found. Those differ more often than not, and the distance between them is why a defect survives several improvement attempts.
Work backwards from the defect's physical characteristics to the operations that could have produced them, then check whether the pattern fits a cause: does it follow a machine, a cavity, a fixture, an operator, a shift, a material lot, a changeover? A defect that appears on one cavity is a tooling problem; the same defect distributed evenly across cavities is a material or process-window problem. That is a different investigation and a different fix.
Where the pattern is not obvious, run a disciplined root cause analysis rather than a debate. And keep containment and correction separate: quarantine and sort to protect the customer today, then do the real work, and do not let the calm that follows containment cancel it.
A weekly scrap review that survives a busy plant
Most scrap-reduction programmes die because they are too heavy. What survives is short, regular, and owned.
- One hour, same time every week, same people. Production, quality, maintenance, and whoever can approve a small spend.
- Pareto first, in money, not in pieces. Rank by cost, including rework hours at a real labour rate, not by defect count. The ranking will surprise you at least once.
- Two items only. The top item and one recurring nuisance. A list of fifteen actions is a list of zero actions.
- Each action is named, dated, and classified as prevention or detection.
- Close the loop before adding new work. Re-measure the specific defect four weeks after the fix. If it did not move, the cause was wrong — reopen it rather than quietly starting something else.
- Log the near-misses. The sort that caught a lot before it shipped is a free lesson in a process that is not capable.
None of this requires software. A whiteboard, a rate for rework labour that finance agrees with, and a standing hour will outperform a dashboard nobody opens. If you already run a formal system, this fits inside it — the point is cadence and honesty about cost, which is the same discipline behind building quality into the process rather than inspecting it in at the end.
FAQ
Is some scrap unavoidable? Process-inherent losses like start-up parts, purge, or trim exist in many operations. Separate them from defect scrap in your reporting, because mixing the two makes both numbers meaningless and hides real movement.
Should rework be counted as scrap? Not as scrap, but it must be counted somewhere. Track it as its own line with hours attached. A plant that reworks heavily and scraps little often has a higher quality cost than one with the opposite profile.
How do I cost rework labour credibly? Agree one loaded hourly rate with finance and use it consistently. The precise figure matters far less than using the same figure every week, because the purpose is ranking problems, not accounting.
Where should a plant with no data start? One part number, one week, manual tally at each station: passed first time, reworked, scrapped. That single week of honest data usually reorders the improvement list on its own.
What about supplier-caused defects? Cost them the same way, then take the total to the supplier as a cost conversation rather than a defect count. Incoming issues are prevention opportunities at qualification, not just appraisal opportunities at receiving.
Pick your highest-volume part, measure its first-pass yield for one week, and cost the rework hours honestly. For more on building quality into production instead of inspecting it out, work through the operations guides on Manufax.
