Scrap Rate vs. Rework Rate: Costs & OEE Impact

What scrap rate and rework rate actually measure

Scrap rate is the share of produced units that are defective beyond economical correction and have to be discarded — written off as a material, labour and energy loss. Rework rate is the share of units that failed an inspection at first pass but can be brought back to specification by additional work, additional cycle time, or additional material. Both are quality-loss metrics; they describe the same underlying problem (the process did not produce a conforming unit on the first try) but they hit the plant in fundamentally different ways. Scrap is loud — it shows up in the material write-off line, the COGS variance, the morning report. Rework is quiet — it consumes capacity that nobody booked against quality, lengthens lead times that nobody attributes to defects, and defends a Final Yield number that looks healthier than the plant actually runs.

The reason this distinction matters more in 2026 than it did twenty years ago: in a digitalised plant, both metrics are measurable in real time, separately, with their own counting rules. There is no longer a technical excuse to lump them together as "quality losses" the way paper-based quality reports did until the early 2000s. The plants that exploit this measurement separation get a sharper diagnostic instrument. The plants that don't keep losing margin to the part of the iceberg that stays under the waterline.

Scrap vs. rework vs. concession vs. yield loss — four terms, four management decisions

Before any rate calculation, the disposition vocabulary needs to be clear. A non-conforming unit can end in one of four states, each with different cost, schedule and customer implications.

The disposition decision is made by a Material Review Board (MRB) or equivalent approval workflow in regulated environments, and by an informal shift-level call in unregulated environments. Either way, the four outcomes carry very different cost and customer-risk profiles, and reporting them as a single "quality loss" number — as many ERP-driven scorecards still do — destroys the diagnostic value of every one of them. Concession in particular tends to disappear from internal reports while remaining the most expensive long-term outcome because field returns surface six to eighteen months later, far from the plant that made the call.

The cost asymmetry — why rework usually hurts margin more than scrap

The intuition most people start with is that scrap is more expensive than rework, because scrap is a total loss while rework recovers the unit. The arithmetic at the unit level supports this: a scrapped unit loses 100 % of the invested material and labour, a reworked unit loses only the additional rework cost. The arithmetic at the plant level often reverses the conclusion, for two reasons that the unit-level view ignores.

First, rework consumes capacity that the plant has already paid for. Every minute of rework cell time, every operator hour applied to fixing yesterday's mistakes, every machine slot occupied by a re-pressed part is a minute that cannot be sold to a paying customer. In a capacity-constrained plant — which is most plants most of the time — this opportunity cost dwarfs the unit-level material savings. A reworked unit that "saves" €40 in material may consume €120 in capacity that could have produced a fresh unit at €200 contribution margin. The accounting books a €40 saving; the P&L books a €160 missed contribution. The unit-level view celebrates the saving. The plant-level view does not.

Second, rework lengthens lead time. Every unit that loops through a rework cell adds variance to the throughput schedule. In a Just-In-Time or Just-In-Sequence environment — automotive Tier 1, food and beverage to retailer DCs, pharma packaging to wholesaler — late delivery costs are step-function: either the truck leaves on time with the unit on it, or it doesn't. Rework that pushes a single unit past the cut-off can trigger expedited freight, contractual penalty clauses, or a missed customer slot whose cost is orders of magnitude greater than the unit value. None of this appears in the rework rate calculation. All of it appears in the customer relationship.

The blunt heuristic that survives 30 years of this argument: scrap is a unit-cost problem; rework is a capacity-and-lead-time problem. They are not interchangeable, and treating them as a single bucket misallocates the improvement effort to whichever component the corporate template happens to highlight.

From a contract food and beverage operation in Baden-Württemberg, somewhere around 2003: the corporate KPI was scrap rate, single number, reported weekly to the divisional headquarters in another country. The plant had been running at 1.2–1.5 % scrap for three years, the headquarters was satisfied, the plant manager was satisfied, the divisional VP cited the plant in his quarterly review as a quality benchmark. Then a SYMESTIC rollout introduced separate counting for first-pass yield, rework events and final disposition — three numbers where there had been one. The picture inverted within two weeks of clean data. Scrap rate was indeed 1.4 %. Rework rate was 11.3 %. One in nine units was being looped back through a manual finishing station that had been informally added to the line two years earlier and never reported as a quality cost — operators called it "the rescue station" and the shift logs recorded its activity as "minor finishing work" rather than rework. The plant had not been a quality benchmark; it had been the most successful Hidden Factory in the division. Once the rework was visible, the cost calculation was easy: the rescue station consumed two FTEs across three shifts, occupied 14 m² of plant floor, and added an average of 9 minutes of lead time to one unit in nine, which translated into approximately €380,000 per year in fully loaded operating cost the corporate report had never seen. The plant manager's first reaction was the one I have heard at every comparable plant since: "we always knew about the rescue station, we just never thought of it as quality." That sentence is the reason scrap-rate-only KPIs fail. They reward exactly the behaviour that produces the rescue station — operators "saving" parts to keep the visible number low, while the invisible number quietly capitalises into permanent organisational structure. The fix in that plant was not to shut the rescue station — it was needed in the short term — but to put rework on the same dashboard as scrap, with the same visibility, escalated to the same management level. Within six months the rework rate was 4.1 %, the scrap rate had risen to 1.7 % (because the operators stopped rescuing borderline parts), the total quality loss in cost terms was down 60 %, and the rescue station was running one shift instead of three. The corporate report still showed a "worse" scrap number, which required a six-month conversation with the divisional VP to explain. That conversation is the part of these projects that nobody warns you about, and the part that determines whether the improvement sticks. Honest measurement is unpopular before it is rewarded, and the gap is usually about two reporting cycles long. The plants that survive the gap get the long-term benefit. The plants that bury the rework number to protect the scrap number get the rescue station back within a year.

The OEE attribution rule — where each loss type lands

The cleanest way to think about scrap and rework in an MES context is to follow them into the OEE calculation, because OEE is where most plants notice the impact first.

• Scrap → Quality factor. A scrapped unit is counted in production volume but not in good output, which directly reduces the Quality component of OEE. This is mechanical and well understood.
• Rework → Performance and Availability factors. The reworked unit usually does appear in good output (after correction), so Quality is unaffected. Performance is degraded because the rework cell consumes cycle time per unit that the standard cycle does not include; Availability is degraded if the rework cell is on the main line rather than a parallel cell, because the line stops or slows during the loop.

The numerical consequence: a plant that reduces its scrap by 1 % and increases its rework by 1 % may report a higher OEE — Quality went up, Performance and Availability changed less than the Quality gain — even though total quality loss in cost terms rose. This is the OEE blind spot that the metric inherits from any aggregate KPI: components can move in compensating directions and the headline number conceals the trade. Reporting scrap rate, rework rate and OEE side by side, with all three trends visible, is the structural defence. Reporting only OEE allows the rescue-station pattern to recur invisibly. See OEE: definition, calculation & practice for the broader framework this fits inside, and Rolled Throughput Yield (RTY) for the systemic metric that captures both losses in a single first-pass-or-not number.

The "saving parts" pathology — why a falling scrap rate can be a warning sign

The most common pattern in plants that report scrap-only quality KPIs: operators learn, within months, that the metric is scrap, and they optimise their behaviour for it. This is a perfectly rational response to the incentive structure. The result is that borderline units that ten years ago would have been scrapped are now "saved" through informal rework — a tap with a hammer, a quick re-grind, a re-test until the unit passes, a manual touch-up that was not in the standard work. Each individual save is small. The aggregate effect over a year is a permanent informal rework infrastructure that consumes capacity, lengthens lead times, depresses Performance OEE, and never appears in the quality report.

The diagnostic indicator that this is happening: a scrap rate that has been falling steadily for several quarters while throughput per labour-hour is flat or declining. If quality is genuinely improving, throughput should rise (less time spent on bad parts). If throughput is flat, the "improvement" is being absorbed by informal rework rather than reaching the customer. The harder confirming evidence requires direct shopfloor observation — walk the line during second shift, count how many parts get touched twice — but the throughput signal is usually sufficient to justify the investigation. Plants that monitor scrap and rework as separately reported, separately approved metrics catch this pattern in months. Plants that monitor only scrap discover it years later, usually during an external benchmark exercise that reveals the labour productivity gap.

Counting rules — when does rework become scrap?

The operational definition that has to be decided before any rework rate calculation has meaning: at what point does a unit that has been reworked once, twice, three times, get reclassified as scrap rather than as cumulative rework? The defensible rule, used by most regulated and quality-mature non-regulated plants:

• First rework attempt: counted as one rework event, unit remains in flow.
• Second rework attempt on the same defect: automatic engineering review trigger; usually still rework but with documentation requirement.
• Third rework attempt: automatic scrap classification regardless of whether the unit could theoretically be saved further. The accumulated cost has exceeded the unit value, and continuing rework is destroying capacity.

The exact thresholds vary by industry — pharma typically caps rework at one attempt for batch-record reasons, automotive Tier 1 typically allows two for non-safety-critical features, foundry and forging operations sometimes allow three or more because the material value is high relative to labour. The discipline is less about the specific number and more about having any number at all, applied consistently, automatically logged in the MES rather than left to operator judgement at the moment of decision. Plants without a defined N-iterations rule produce rework rate numbers that are statistically meaningless, because the same physical unit can be counted as 0, 1, 2 or 5 rework events depending on which operator happened to handle it.

Cost of Poor Quality — the Juran framing that connects everything

Joseph Juran's Cost of Poor Quality (CoPQ) framework, in continuous use since the 1950s, is the management accounting structure that makes the scrap-rework distinction actionable at the financial level. CoPQ has four components: prevention, appraisal, internal failure, external failure. Scrap and rework both sit in internal failure — losses caught before the unit reaches the customer. The Juran insight that has not aged: internal failure costs are typically 5 % to 30 % of revenue in industries that don't actively measure them, and the visible component (scrap) is usually the smaller half. The invisible component (rework, expedited freight, plant overtime to compensate, capacity bookings against rework) is the larger half and the harder one to extract from standard accounting because it is distributed across multiple cost centres and rarely tagged as quality cost.

The practical use of the Juran framing in 2026: when proposing a quality investment to management, the business case has to include both halves of internal failure. A scrap-only business case usually under-estimates the recoverable savings by a factor of two to three, and proposals at that level are easy to defer. A complete CoPQ business case — scrap plus rework plus the lead-time and capacity consequences — is usually compelling enough to clear the approval threshold without further negotiation. The investment in measurement infrastructure that surfaces the rework component pays for itself in the first improvement project it enables, often in months rather than years.

How this fits into the SYMESTIC platform

SYMESTIC captures scrap, rework and concession as three separate event types in the production data stream, with configurable counting rules (including the N-iterations cap on rework attempts) and explicit linkage to the disposition workflow. Each event is attributed to a station, a defect code, an operator, a shift and an order — the attribution dimensions that make the data usable for both root-cause analysis and management reporting. The OEE calculation respects the attribution rule above (scrap to Quality, rework to Performance/Availability) and the dashboards display scrap rate, rework rate and OEE side by side rather than collapsing them into a single quality number, so that the rescue-station pattern described above cannot establish itself invisibly. For regulated-industry deployments the disposition workflow is integrated with the Material Review Board approval chain and recorded in a 21 CFR Part 11 / EU GMP Annex 11 compliant audit trail. Bidirectional ERP integration (SAP R/3 via ABAP IDoc, Microsoft Dynamics/Navision, Infor/InforCOM, proAlpha) ensures that scrap and concession events update inventory and order status in the ERP automatically, and the bidirectional connection to quality systems — the Böhme & Weihs CASQ-it integration at Meleghy is a concrete example — captures inspection-driven rework events that would otherwise remain invisible to the MES. The architectural principle is one I have argued for since the SYMESTIC platform's first redesign: any quality loss that is not measured separately gets optimised against indirectly, and the indirect optimisation is almost always worse than the direct one.

FAQ

Is rework usually more expensive than scrap?
At the unit cost level, scrap is more expensive — total loss vs. partial recovery. At the plant cost level, rework is usually more expensive — capacity consumption, lead-time variance, opportunity cost of the rework cell, and the administrative cost of the disposition workflow. The asymmetry holds in capacity-constrained plants, which is most manufacturing operations most of the time. The unit-level intuition is correct in isolation; the plant-level reality usually inverts it. The reliable rule of thumb: if the plant is selling everything it can produce, rework costs more than scrap. If the plant is sitting on excess capacity, scrap costs more.

How does scrap impact OEE differently from rework?
Scrap reduces the Quality component of OEE directly — the scrapped unit is counted in production but not in good output. Rework usually does not affect Quality (the reworked unit eventually appears in good output) but degrades Performance (extra cycle time consumed by the loop) and Availability (line stops or slows during the rework). A plant that converts scrap into rework can report a higher OEE while spending more on quality losses overall, which is the OEE blind spot the side-by-side reporting of scrap rate, rework rate and OEE is designed to eliminate.

What is concession and why does it matter?
Concession (also called "use as is") is the disposition where a non-conforming unit is approved to ship without correction, usually after engineering and customer review. It is the third disposition outcome after rework and scrap, and the most dangerous in cost terms because it has no visible production cost — the unit ships, the order closes, the disposition is buried in the Material Review Board record. The hidden cost surfaces months later as field returns, warranty claims and customer complaints, far from the plant that made the call. Plants that don't track concession volume separately are systematically blind to a quality cost that may exceed scrap and rework combined.

How is rolled throughput yield related to scrap and rework rates?
Rolled Throughput Yield (RTY) is the systemic metric that captures both losses in a single number — the probability that a unit passes every step of the process right the first time. Both scrap and first-pass rework events count as RTY failures, regardless of whether the unit eventually ships. Scrap and rework rates are the component diagnostics; RTY is the chain-level consequence. A complete quality scorecard reports all three side by side because they answer different questions: "what was lost?" (scrap), "what was rebuilt?" (rework), "what worked first time?" (RTY).

What is the most common mistake in measuring scrap and rework?
Reporting scrap as the sole quality KPI. Operators rationally optimise for the metric that gets reported, which means borderline units get "saved" through informal rework rather than scrapped. The scrap rate falls, the rework rate rises invisibly, and the plant accumulates a permanent informal rework infrastructure — the "rescue station" pattern. The fix is to report scrap rate and rework rate as two separately escalated metrics, with rework treated as the more strategically important of the two because it is the one that consumes capacity. The second most common mistake is treating yield loss in process industries as scrap, which over-states scrap by an order of magnitude in industries with significant trim, purge or transition waste.

How many rework attempts should be allowed before a unit is scrapped?
The defensible rule is two attempts maximum for non-safety-critical features, with automatic engineering review on the second attempt and automatic scrap reclassification on the third. Pharma typically caps at one attempt for batch-record reasons; foundry and high-material-value operations sometimes allow three or more. The exact number matters less than having any rule applied consistently and logged automatically — without an N-iterations cap, rework rate numbers are statistically meaningless because the same physical unit can be counted as 0, 1, 2 or 5 events depending on the operator handling it.

Should rework cells be on the main line or off-line?
The architectural rule that minimises the OEE distortion: rework off-line, in a parallel cell, with its own labour and its own capacity reporting. On-line rework consumes Availability and Performance directly, conflates production and rework cycle times, and makes the Hidden Factory invisible because it shares the same machine and operator records as primary production. Off-line rework is more administratively complex (a separate cost centre, a separate routing, a separate operator group) and produces dramatically cleaner cost data. The correct decision is the one that produces actionable cost visibility, which is almost always off-line.

Related: OEE: definition, calculation & practice · MES: definition, functions & benefits · OEE software · MES software compared · Rolled Throughput Yield (RTY) · Recipe management · Work plan · Change control · Role-based access control · Production metrics module · Process data module · Production control module · Automotive · Food & beverage · Metal processing · Plastics processing · For operational excellence · For production managers · For COOs & plant managers. External reference: ASQ – Cost of Quality (the American Society for Quality's public reference on Juran's CoPQ framework).

About the author

Uwe Kobbert

Founder and CEO of SYMESTIC GmbH (Dossenheim). Over 35 years in the manufacturing industry. Dipl.-Ing. Nachrichtentechnik/Elektronik. Career: Consultant at SAS (1989–1992, Heidelberg), Department Lead Industry at STERIA Software Partner (1992–1995) with direct responsibility for process control and MES in food and beverage manufacturing, founder and CEO of SYMESTIC since 1995. Led the platform's rebuild as cloud-native on Microsoft Azure in the mid-2010s; today the platform runs in 18 countries on four continents with 15,000+ connected machines. Nominated for the Großer Preis des Mittelstandes. Self-financed, no external investors. Expertise: Manufacturing Execution Systems, OEE, shopfloor management, cloud-native manufacturing software, Industry 4.0, Lean production, process control systems, ERP-MES integration, JIT/JIS processes, batch production, automotive, food and beverage. · LinkedIn · Großer Preis des Mittelstandes profile