Scrap Reduction: The Hidden Margin Lever

What is scrap reduction?

Scrap reduction — sometimes called material waste reduction, scrap and rework reduction, material yield optimisation, or in the OEE vocabulary "quality loss reduction" — is the structured discipline of eliminating material that is produced but cannot be sold. It covers the obvious categories (defective parts rejected at end-of-line inspection, setup scrap from the first pieces of a changeover, cutting waste and offcuts in material-removal processes) and the less-obvious categories that are typically the biggest hidden losers (startup yield loss, in-process degradation, material shrinkage from handling and storage, over-consumption that never even registers as "scrap" because it was never counted as a part). Every gramme of material purchased that does not leave the plant as a sellable product is scrap — regardless of whether the system classified it as such.

I have spent 25 years chasing scrap across four continents. Six Sigma Black Belt at Johnson Controls starting in 2001, where DMAIC on scrap problems was literally my weekly work. Global MES and traceability lead across China, Mexico, Tunisia, Macedonia, France and Russia at Johnson Controls and Visteon, where every plant had its own scrap-reporting habits and none of them matched each other. The finding is consistent enough that I now treat it as a law: the scrap number a plant reports is almost always wrong, and it is almost always understated by 30 to 50 percent. The biggest lever in scrap reduction is almost never better process control — it is better measurement of the scrap that is already happening and already invisible.

The three categories of material loss — and why they don't map to a single workflow

The serious mistake in scrap-reduction programmes is to treat all material loss as one number — "we produced 10,000 parts, we scrapped 320, scrap rate is 3.2 %." That number is not wrong; it is just not actionable. The three categories of material loss have structurally different causes, different owners, different countermeasures, and different capture mechanisms. A programme that doesn't separate them optimises against the aggregate and usually picks the least-important lever.

The financial reality underneath this table: in most discrete-manufacturing plants I have audited, defect scrap represents 20–35 % of total material loss, and process yield loss represents 50–70 %. Yet plants typically spend 80 %+ of their scrap-reduction effort on defect scrap, because it is the only category that shows up in the standard report. The yield loss — often the largest category — is hidden in the monthly material reconciliation where nobody with improvement authority is looking. The first lesson of scrap reduction is therefore: separate the categories before improving any of them, because the aggregate number tells you the wrong story.

Why most scrap data understates reality by 30–50%

This is the finding I now lead every scrap-focused engagement with, and it is almost always met with disbelief for the first three months of the engagement. The reported scrap rate in most plants is systematically too low — not occasionally, not regionally, systematically. The distortion is not theft or fraud; it is a product of how manual scrap reporting interacts with the realities of shop-floor work. Four specific failure modes produce the gap, and they compound.

A hard-earned lesson from Tunisia, 2008. I was running a DMAIC project on a headliner line at Johnson Controls. Reported scrap rate was 2.1 %. The plant had been working on it for two years, had an active kaizen team, had posted bar charts everywhere, and the line held steady at 2.1 % plus or minus a quarter point. I spent one full shift doing what Six Sigma manuals tell you to do and nobody actually does: I weighed every scrap bin myself at the end of the shift, weighed the raw material in, and reconciled against good parts out. The real scrap rate was 3.6 %. The 1.5-percentage-point gap — a relative underreporting of 42 % — came from three places: parts that broke during the pull from the press to the trim station and were put back in the raw-material bin rather than the scrap bin; setup scrap from the first 40 parts of every changeover that was never booked because "it's always like that;" and a small daily loss from off-spec material that was reworked rather than rejected. Every one of those sources was rational at the operator level. Cumulatively, they hid 1.5 percentage points of scrap from every improvement programme the plant had ever run. The breakthrough in that project was not a better process — it was counting the scrap that was already there.

The cumulative effect is why a plant reporting 2 % scrap is usually really running at 3–4 % scrap, and why improvement programmes built on the reported number rarely deliver their projected savings — they are targeting one-half to two-thirds of the actual loss. The correct first step in any scrap-reduction programme is a one-shift reconciliation audit: weigh in, weigh out, count good parts, and calculate the real number. Every plant I have ever audited this way has produced a different number than the one on the dashboard. The question is not whether your scrap data is wrong, it is by how much.

The DMAIC framework applied to scrap reduction

Scrap reduction is the single best-fit problem for Six Sigma's DMAIC methodology — Define, Measure, Analyse, Improve, Control. Not because the methodology is fashionable (it isn't anymore), but because the structure maps exactly onto the four failure modes of scrap programmes: unclear problem, unreliable data, undisciplined root-cause analysis, and no sustainment mechanism. I have run DMAIC on scrap problems in foundry, injection moulding, stamping, assembly, SMT, and blister packaging. The pattern of where the work actually sits is consistent enough to be worth stating explicitly.

The single most-often skipped phase is Measure, and it is also the phase where the project's success is determined. A DMAIC project built on the reported scrap number reaches Improve, implements something, sees the reported number move, declares victory, and the reported number then drifts back up within a year because the underlying loss was never really quantified and the countermeasures were aimed at the wrong root causes. A DMAIC project that starts with a genuine reconciliation audit — three days of work at most — produces countermeasures that actually hit the loss and stay hit.

Quality at source vs end-of-line detection

There are two fundamentally different philosophies for catching scrap, and the choice between them determines 80 % of the economics of a scrap-reduction programme. End-of-line detection finds the defective parts after they are complete — through inspection, testing, or customer returns. Quality at source detects the problem at the process step that caused it, before the downstream value is added. Both have their place; the ratio between them in your plant largely determines how much scrap you can economically eliminate.

The fourth approach is the one most plants haven't yet exploited and is usually the fastest-payback intervention. A plant producing 3 % defect scrap on an automated line almost always has the signal in the PLC data to predict or explain 60–80 % of those defects — a specific alarm, a parameter excursion, a cycle-time variance — but the signal is sitting in the PLC log and the defect is sitting in the quality system, and nobody has joined them. Joining them, which is technically a straightforward MES task, typically surfaces the root cause of the majority of defects in the first two weeks of operation. This is not theoretical; it is the Neoperl pattern: PLC-alarm-to-defect correlation yielded 15 % scrap reduction directly from making previously-invisible causality visible.

What this looks like across the SYMESTIC installed base

Across the 15,000+ machines connected to the SYMESTIC platform, the scrap-reduction pattern is consistent. Automated scrap capture at the process step where it is produced — not at the end-of-shift paperwork — is the baseline; defects are either detected by in-line sensors and booked automatically, or classified by the operator at a shop-floor terminal with one-tap context-aware reason selection. PLC alarms are automatically correlated with downstream quality events. Yield loss is tracked at the BOM level in near-real-time rather than reconciled monthly, so over-consumption surfaces at the shift where it happens rather than in the following month's finance report.

The outcomes from the named references are consistent enough that I now quote them as expected ranges, not exceptional results. Neoperl (assembly automation, Müllheim) landed 15 % less scrap and 15 % higher productivity through PLC-alarm correlation with defects. Klocke (pharma packaging, Weingarten) saw 12 % output improvement and 8 % availability improvement within three weeks, with scrap reduction as a meaningful contributor to both. Meleghy (automotive forming and joining, six plants across four countries) reduced stops by 10 % and improved output by 7 %, with scrap tracking at the cycle level as the measurement foundation. The capital expenditure across these interventions was essentially zero; the material that had always been scrapped simply became visible at the point of production, which made it addressable for the first time.

FAQ

What is scrap reduction?
Scrap reduction is the discipline of eliminating material that enters production but cannot be sold — defective parts, setup losses, process yield loss, handling damage. It is sometimes called material waste reduction, scrap and rework reduction, or material yield optimisation. In OEE vocabulary, scrap is the primary component of the quality loss bucket. Serious scrap reduction is not about eliminating the final-inspection reject rate; it is about eliminating the material losses across all four categories, of which end-of-line defects are usually only the smallest.

What are the main categories of material loss in manufacturing?
Four categories with structurally different causes and countermeasures. Defect scrap — parts that fail inspection; typically 20–35 % of total material loss. Setup and startup scrap — first pieces of a changeover before the process stabilises; Category 5 of the Six Big Losses. Process yield loss — material consumed above the theoretical BOM through sprue, flash, offcuts, over-dosing, evaporation; typically 50–70 % of total material loss and usually the invisible category. Handling and storage loss — material damaged in transport or degraded in storage. Separating these categories is the first step of any serious programme, because the aggregate scrap rate tells you the wrong story about where the loss is.

Why is reported scrap usually lower than real scrap?
Four systematic distortions, all rational at the operator level, all producing the same underreporting effect. Rework masquerading as good (10–20 % gap), setup scrap silently absorbed into the outgoing order (5–15 %), handling breakage returned to the raw-material bin rather than the scrap bin (5–10 %), and yield loss below the reporting threshold reconciled only in monthly finance closes (10–30 %). Cumulatively, a plant reporting 2 % scrap is usually really running at 3–4 %. The first step of any scrap programme should be a one-shift reconciliation audit — weigh material in, weigh material out, count good parts — to produce a verified baseline before any improvement is attempted.

How does DMAIC apply to scrap reduction?
DMAIC is the single best-fit methodology for scrap problems because its five phases map exactly onto the four failure modes of scrap programmes. Define produces a scope narrow enough to be solvable — one product family, one process, one failure mode, not "reduce plant scrap." Measure produces a verified baseline, not the reported number. Analyse produces the 2–3 root causes that explain 70 %+ of the loss. Improve implements targeted countermeasures on those causes. Control sustains the gain through SPC, automated capture, and a control plan. The most-skipped phase is Measure, and it is also the phase where the project's success is determined.

What is the difference between end-of-line detection and quality at source?
End-of-line detection finds defects after the part is complete — through final-stage testing, inspection, or customer returns. Quality at source detects the problem at the process step that caused it, before downstream value is added. End-of-line is cheap to set up but expensive per defect, because the full value of the part is already sunk by the time the defect is caught. Quality at source — via in-process SPC, poka-yoke error-proofing, or PLC-alarm correlation with defects — is moderate to set up but near-zero per-defect cost once running. The economic argument almost always favours shifting detection upstream; the execution question is whether the process has measurable signals upstream to detect against.

What is PLC-alarm correlation with defects?
PLC-alarm correlation is the technique of cross-referencing alarms raised by the machine controller with downstream quality results, so a defect can be attributed to the specific PLC event that preceded it. In highly automated processes, 60–80 % of defects are caused by events already visible in PLC telemetry — a parameter excursion, a sensor signal, a cycle-time variance — but the PLC log sits in the OT layer and the defect sits in the quality system, and the two are never joined. Joining them is a straightforward MES task that typically surfaces the root cause of the majority of defects in the first two weeks of operation. The Neoperl case study at SYMESTIC produced 15 % scrap reduction from exactly this intervention, with effectively zero capital expenditure.

What is the role of SPC in scrap reduction?
Statistical Process Control is the in-process detection mechanism that catches process drift before it produces defects. SPC charts track measurable process parameters — dimension, temperature, pressure, cycle time — against statistical control limits, and signal when the process is drifting toward the specification boundary before any part is actually out of specification. A well-configured SPC system converts reactive defect detection (find scrap after it is made) into proactive process control (prevent scrap from being made). SPC's effectiveness depends entirely on selecting the right parameters to chart, which is an engineering-design question, not a statistical question — the most common failure mode is charting many parameters that have no correlation with the defects that actually occur.

Is yield loss the same as scrap?
Technically no, functionally yes. Yield loss is the difference between the material consumed (per BOM-plus-actual reconciliation) and the material embedded in sellable product — sprue, flash, offcuts, evaporation, over-dosing, shrinkage. Scrap in the narrow sense is defective completed parts that fail inspection. Yield loss is usually not called scrap in accounting, but it is material that entered the plant and did not leave as product, which is the economically meaningful definition of scrap. In most plants, yield loss is 2–5× larger than defect scrap and almost entirely invisible to the improvement programme. A scrap-reduction programme that does not address yield loss is addressing the smaller half of the problem.

How does SYMESTIC support scrap reduction?
Automated scrap capture at the cycle where it is produced — in-line sensor detection for automated rejection, one-tap context-aware reason selection at the shop-floor terminal for manual classification. PLC-alarm correlation with downstream quality events as the native data model, not a reporting overlay. Yield loss tracked at the BOM level in near-real-time, so over-consumption surfaces at the shift where it happens rather than in the following month's finance close. SPC chart generation from the captured process parameters. Typical outcome across industries is a 5–15 % scrap reduction in the first six months, stacked across making invisible losses visible, surfacing root causes through alarm correlation, and shifting detection upstream from end-of-line to in-process. See SYMESTIC Production Metrics.

Related: OEE · Six Big Losses · Six Sigma · SPC · Lean Production · Value Stream Mapping · Downtime Analysis · MES · SYMESTIC Production Metrics

About the author

Christian Fieg

Head of Sales at SYMESTIC. 25+ years in manufacturing — maintenance engineer and Six Sigma Black Belt at Johnson Controls, global MES and traceability lead for 900+ machines and 750+ users across China, Mexico, Tunisia, Macedonia, France and Russia, Manager Center of Excellence for the global MES programme at Visteon, Sales Manager MES DACH at iTAC, Senior Sales Manager at Dürr. At SYMESTIC since 2021. Author of "OEE: One Number, Many Lies" (2025). · LinkedIn