Monday, 6:40 a.m., early shift. In the holding area, there are three mesh crates containing housings that were still considered acceptable on Friday. No one knows exactly which part marked the point where the hole fell outside the tolerance. So the parts are sorted, remeasured, and reworked while the line waits for approval. In the end, quality control knows the number of parts that were rejected. What they don’t know is the exact moment when the process began to go off track. It is precisely this gap that determines whether scrap remains a fixed cost or becomes an avoidable occurrence.
THE MOST IMPORTANT POINTS AT A GLANCE
|
IN SHORTSoftware for reducing scrap is not a single product category, but rather a combination of functions that address the root causes of defects: automatic recording of process values, linking to serial numbers, real-time statistical monitoring, and verification of manual work steps. This guide classifies six system categories by impact and timing of effect, presents the formulas for scrap rate, rework rate, and first-pass yield, identifies the process data that signal impending scrap, and calculates the return on investment using an example. |
Scrap reduction software refersto systems that detect or prevent defects in production as they occur, rather than simply documenting them. It accesses the real-time data stream from production: process parameters from machine controls, measurement values from inspection equipment, torque and rotation angle from the screwdriving system, and acknowledgments at manual workstations. This data forms the basis of an early-warning system that intervenes before a part becomes scrap.
Traditional quality management software operates at a different stage. It manages inspection plans, complaints, corrective actions, audit reports, and documents. While this is essential for effective quality management, it primarily operates downstream: the defect has already occurred by the time the system records it. Our comparison of quality management software provides an overview of the market for these systems.
Three features make the difference. First, automatic data collection without manual entry, because any manual transfer causes delays and errors. Second, linking each value to the serial number or batch so that deviations can be assigned to a specific component. Third, real-time evaluation with a clear response: a warning, blocking the part, or halting the work step. Our guide describes how this logic fits into the overall quality assurance process in production.
| Key Metric | Value | Source |
|---|---|---|
| Quality-related costs | 15 to 20 percent of revenue, up to 40 percent in some cases | American Society for Quality (ASQ) |
| Costs of poor quality in manufacturing companies | On average about 15 percent, ranging from 5 to 35 percent of revenue | Institute of Industrial and Systems Engineers (IISE) |
| Cost increase per defect per stage of the value chain | approximately a factor of 10 | Rule of Tens for defect costs (rule of thumb) |
| Typical minimum capability Cpk | 1.33; for special characteristics, 1.67 | IATF 16949 framework, AIAG, and VDA |
Software reduces scrap and rework when it addresses the primary cause of defects. In practice, six system categories are suitable for this purpose: Manufacturing Execution System (MES), Computer-Aided Quality (CAQ), Statistical Process Control (SPC) software, process data management, digital worker guidance, and AI-based anomaly detection. They differ in terms of the type of error they address and the point in the process at which they intervene.
| Category | Targets | Time of Intervention | Threshold |
|---|---|---|---|
| MES | Setup and order errors, incorrect program versions | during production | Quality characteristics are often recorded only roughly |
| CAQ | Recurring errors, lack of follow-up on corrective actions | after the error, in inspection planning | primarily has a downstream effect |
| SPC | Process drift, increasing variation | during production | requires automatically recorded measurement values |
| Process data management | undetected parameter deviations, gaps in traceability | during and after production | only becomes effective once all sources are connected |
| Digital worker guidance | Mix-ups, omissions, incorrect sequence | before and during the task | Applies only to manual workstations |
| AI anomaly detection | multidimensional patterns that do not capture threshold values | Proactive, during production | Requires a clean data history |
The table shows why the question of which software is “the right one” falls short. A plant with a high proportion of automated processing loses quality primarily due to creeping process changes. This is where SPC and process data management provide the greatest leverage. Our article on SPC in manufacturing explains how control charts and intervention limits work on the production line. In contrast, an assembly plant with many variants loses quality due to handling errors, and this is where digital worker guidance has an immediate impact.
AI-supported methods complement traditional limit monitoring. They identify combinations of parameters that are unremarkable individually but, when taken together, indicate a problem. This approach is known as predictive quality. To learn what it can achieve and what data foundation it requires, read our introductory article “What Is Predictive Quality?”
PRACTICAL MISTAKESMany companies first implement a CAQ system and are surprised that the scrap rate barely decreases. The reason: A CAQ system organizes the response to errors; it does not prevent them. To reduce scrap, you first need data from the process itself. Only then does the tracking of corrective actions have a solid foundation. |
FREE WHITEPAPERManagement of Quality-Related Production Data This white paper shows you how to integrate process and inspection data along the production line, detect deviations before scrap occurs, and permanently reduce production errors. |
Without clear metrics, it’s impossible to describe the initial state or measure the success of software. Three metrics form the foundation, while a fourth evaluates multi-stage processes. It’s important that all metrics are calculated based on the same reference quantity and the same time period.
| Metric | Formula | Meaning |
|---|---|---|
| Scrap Rate | Reject quantity divided by total quantity, multiplied by 100 | Percentage of parts that are permanently lost |
| Rework rate | Number of units requiring rework divided by total quantity, multiplied by 100 | Percentage of parts that require additional work |
| First Pass Yield (FPY) | Good parts in the first pass divided by the input quantity, multiplied by 100 | Percentage of parts that are good without any intervention |
| Rolled Throughput Yield (RTY) | Product of the FPY for all process stages | Throughput quality across the entire process chain |
An example using hypothetical values: A production line manufactures 10,000 units per month. 180 parts are scrapped, and 420 parts are reworked and subsequently approved as good. The scrap rate is thus 1.8 percent, and the rework rate is 4.2 percent. In the first run, 9,400 parts are good without any intervention; the first-pass yield is therefore 94.0 percent. Although 98.2 percent of the parts are ultimately shipped, every seventeenth part required additional effort.
This is precisely where the most common distortion lies. Rework performed directly at the workstation does not appear in any report. Quality pioneer Armand V. Feigenbaum coined the term “hidden factory” to describe this phenomenon: capacity that is exclusively devoted to correcting defects. Anyone who calculates the rework rate based solely on booked rework orders therefore systematically underestimates it.
For automotive suppliers, recording this data is mandatory anyway. IATF 16949 requires, in Section 8.7.1.4, a risk analysis prior to deciding on rework, a documented process for approving rework in accordance with the production control plan, and records of the use of reworked products. Software that automatically records rework along with the serial number, cause, and result fulfills this requirement as a matter of course and simultaneously provides the accurate rework rate.
Failure analysis in production begins with an uncomfortable realization: The point where a defect is discovered is almost never the point where it originates. A housing fails the final inspection; the cause lies three stations earlier in a worn-out tool. That is why a reliable analysis is organized by root causes, not by where defects are found. Only in this way does every corrective measure improve production quality where the defect actually originates.
| Cause Category | Typical Defect Pattern | Data source for the analysis | Appropriate software category |
|---|---|---|---|
| Process and Machine | Dimensional deviation due to temperature drift or wear | Process parameters, measured values over time | SPC, process data management |
| Human Factors and Handling | Mixed-up variants, missed work steps | Confirmations, scanning operations, work sequence | Digital worker guidance |
| Tools and Equipment | Incorrect torque, tool past its inspection interval | Tightening curves, tool inspection data | Process data management, tool inspection |
| Material and batch | Variation in raw parts, defective supplier batch | Batch assignment, incoming goods inspection | Process Data Management, CAQ |
The Pareto analysis has proven effective for prioritization: In most companies, a small number of defect types account for the vast majority of defects. To identify the root cause, the Ishikawa diagram is used, along with the question “Why?” until the technical root cause becomes apparent. IATF 16949 requires a documented problem-solving process in Section 10.2.3; in the automotive industry, the 8D method is typically used for this purpose.
All of these methods stand or fall on access to the history of the affected part. If process values, inspection results, and operator acknowledgments are linked via the serial number, narrowing down the cause takes minutes instead of days. Our article on traceability in production shows how this chain of evidence is established. The article on process data analysis in quality assurance describes the patterns that can be identified in the data.
Scrap rarely occurs suddenly. It usually announces itself in the process data over the course of hours or days, long before a characteristic exceeds the tolerance limit. Software for reducing scrap is designed precisely for this purpose: to automatically detect and report these early warning signals while good parts are still being produced.
| Signal | What it means | Recommended response |
|---|---|---|
| Trend on the control chart | Several consecutive measurements are rising or are on one side of the mean | Investigate the cause before the action limit is reached |
| Increasing variation | The process is becoming unstable, even though the mean is correct | Check machine condition, clamping devices, and material |
| Decreasing Cpk | Process capability is decreasing; the tolerance is being utilized more tightly | Re-evaluate capability; shorten the inspection interval |
| Changes in the screw-in curve | The rotation angle or curve shape deviates, even though the final torque is correct | Inspect the thread, component, and tool |
| Tool nearing end of service life or inspection interval | Wear or lack of calibration is imminent | Schedule replacement or inspection |
| Batch or shift changes | New operating conditions in the process | Targeted inspection and comparison of the first parts |
The rules that apply to trends depend on the set of rules being used. Common statistical process control rules interpret a sequence of seven to nine data points on one side of the centerline as an indication of a systematic shift. It is crucial that these rules are applied automatically and for every critical characteristic, not just during the weekly review.
Process capability serves as the unifying benchmark here. Our article on “Calculating and Interpreting Cpk and Ppk” explains how to correctly determine and interpret Cp, Cpk, and Ppk. If you want to quickly evaluate your own measurement series, you can use the free Cpk calculator. The step-by-step process capability study shows how a complete capability analysis is conducted.
There is a special consideration for screw connections. The final torque alone says little about the quality of the connection. Only the torque-angle curve reveals whether a thread is damaged or a component is missing. Our article on screw data management describes how this data is recorded and verified.
At manual workstations, rework arises primarily from four types of errors: mixing up components or variants, omitted work steps, incorrect sequence, and incorrectly configured tools. None of these can be detected using control charts because there is no continuous process value that drifts. This is where reducing rework through digital worker guidance comes into play.
The principle is error prevention rather than error detection. Section 10.2.4 of IATF 16949 explicitly requires a documented process for applying error-prevention methods, known in Japanese as Poka-Yoke. Digital worker guidance implements this requirement in software.
An error that cannot be made does not need to be detected, sorted out, or reworked.
This approach is particularly effective when dealing with a high number of variants, because the risk of mix-ups increases with each new version. Our article on variant diversity in assembly shows exactly how worker guidance prevents mix-ups. The article on digital worker guidance and faster onboarding describes a second benefit: New employees work according to standard procedures from day one, which significantly stabilizes the error rate during the ramp-up phase.
Cost-effectiveness can be clearly calculated using the key metrics from the previous section. The basic formula is: Annual savings equal the reduction in the scrap rate multiplied by the annual volume multiplied by the cost per scrap part, plus the reduction in the rework rate multiplied by the annual volume multiplied by the cost per rework. In addition, there are items such as savings from eliminated sorting operations, reduced inspection costs, and avoided customer complaints, which you should evaluate separately.
| Item | Initial Situation | After implementation | Annual Savings |
|---|---|---|---|
| Annual volume | 400,000 units | 400,000 units | |
| Scrap rate | 1.8 percent | 1.2 percent | 2,400 parts at 38 euros each: 91,200 euros |
| Rework rate | 4.2 percent | 2.8 percent | 5,600 interventions at 11 euros each: 61,600 euros |
| Total internal error costs | 152,800 euros |
These figures are intentionally conservative and are provided for illustrative purposes only. For your calculation, use the actual costs per rejected part—including material, processing already performed, and disposal—as well as the costs per rework item, such as labor, re-inspection, and logistics. On the cost side, in addition to licenses, there are the costs of connecting data sources, setting up inspection rules, and training employees.
The timing of detection often has the greatest impact. According to the “rule of ten” for error costs, the cost of an error increases by a factor of ten at each stage of the value chain. The Fraunhofer IPA reports that, based on the application of its error-process matrix at BMW AG, projected warranty costs were reduced by approximately 75 percent—achieved through an inspection chain that detects errors earlier in the process. Our article on the benefits and costs of real-time monitoring shows how real-time monitoring on the production line can be evaluated economically.
PRACTICAL TIPFirst, calculate only the internal failure costs for a single pilot line. This figure can be substantiated using existing data and is more readily accepted by management than an extrapolation across all plants that relies on estimated follow-up costs. |
FREE WHITE PAPERManagement of Quality-Related Production Data This white paper shows how to integrate process and inspection data along the production line, detect deviations before scrap is generated, and permanently reduce production errors. |
A comprehensive rollout across all lines simultaneously often fails due to complexity and a lack of a basis for comparison. A pilot project lasting about 90 days on a single line with high, easily measurable scrap has proven effective. This turns process optimization into a measurable project rather than an ongoing task with no results. The following roadmap serves as a guide; the duration of each phase depends on the existing equipment landscape.
| Phase | Timeframe | Content | Result |
|---|---|---|---|
| Starting Point | Days 1 through 15 | Select a pilot line, create a Pareto chart of scrap causes, collect key metrics | Establish a reliable baseline for scrap rate, rework rate, and FPY |
| Integration | Days 16–35 | Connect machines via OPC UA, screwdriving controllers, and measuring equipment; define serial number as key | Automatic data stream without manual entry |
| Rules | Days 36 through 60 | Limit values and control charts for critical characteristics; operator guidance at the workstation most prone to errors | Early warning and error prevention during ongoing operations |
| Evaluation | Days 61 through 90 | Comparison with the baseline, deriving corrective actions, deciding on rollout | Decision template with measured effect |
Five key questions can help in selecting the system. Can the software read data directly from machines, tools, and test equipment? Does it link each value to the serial number? Does it respond in real time with warnings or locks? Does it cover both automated and manual workstations equally? And can data be exchanged with MES and ERP systems without creating new silos? Our article on MES and ERP integration in quality management describes how to achieve this integration.
The organizational aspect is just as important. The software must be embedded in the existing quality management system, with clear responsibilities for threshold values, approvals, and corrective actions. Our article on setting up a quality management system in manufacturing outlines the requirements involved.
The Manufacturing OS unifies the functions described in the previous sections as separate system categories into a single, shared database. The serial number serves as the consistent identifier: Every process value, every tightening curve, every inspection result, and every operator acknowledgment is linked to the component’s identity.
Through process data management in the CSP Manufacturing OS, machines and testing equipment are connected via OPC UA and REST, limit values are monitored in real time, and process metrics such as Cpk are continuously calculated from the production data. Digital operator guidance, as part of the platform, ensures that manual work steps are performed according to the specific variant and locks parts in case of deviations. Tool inspection in the quality assurance area ensures that only inspected tools are used, and AI anomaly detection flags patterns that traditional limit values fail to detect.
Because these functions are not four separate systems but rather components of a single platform, the effort required to integrate them is eliminated. Fault analysis draws on a complete history, and documentation for IATF 16949 is generated as a byproduct of production. Thus, the question of why parts were scrapped becomes the question of how to prevent the next batch of scrap.
SUCCESS STORYSchaltbau GmbH: Troubleshooting Through Process Optimization How a machine manufacturer uses end-to-end process data to more quickly pinpoint and eliminate the causes of defects. |
Systems that address the root cause of errors are particularly effective at reducing scrap and rework: software for statistical process control and process data management to combat process drift, digital operator guidance to prevent handling errors, and AI anomaly detection to identify complex patterns. MES and CAQ complement these functions by providing order context and tracking corrective actions. The greatest impact is achieved by a platform that links all data via the serial number.
Software for reducing scrap automatically collects process data from machines, tools, and workstations, links it to the serial number, and evaluates it in real time. It detects deviations before a part becomes scrap and prevents errors at manual workstations through guided work steps.
The rework rate is the number of reworked units divided by the total quantity produced, multiplied by 100. It is important to also record rework performed directly at the workstation; otherwise, the rate will be systematically underestimated.
The scrap rate indicates the percentage of parts that are permanently lost. The first-pass yield indicates the percentage of parts that are acceptable on the first run without rework. The first-pass yield is therefore the stricter metric because it accounts for both scrap and rework.
Useful data includes process parameters and measured values over time, screwdriving curves with torque and angle of rotation, tool inspection data, batch assignments, and acknowledgments at manual workstations. They only realize their full value when they are assigned to a specific component via its serial number.
The American Society for Quality estimates that quality-related costs account for 15 to 20 percent of revenue at many companies. Scrap and rework are included in these internal defect costs. The exact amount is determined by the quantity, the cost per scrap part, and the cost per rework intervention on your production line.
AI-powered anomaly detection can identify combinations of process parameters that are unremarkable on their own but, when combined, indicate a quality deviation. This requires a clean, complete data history. AI does not replace traditional limit monitoring; rather, it complements it.
A reliable effect can be measured in a pilot project lasting about 90 days on a single production line: establish a baseline, connect data sources, activate inspection rules, and compare the results to the baseline. The extent to which the scrap rate decreases depends on the dominant causes of defects on the line.
Section 8.7.1.4 of IATF 16949 requires a risk analysis prior to deciding on rework, a documented process for confirming rework in accordance with the production control plan, and records of the use of reworked products. Section 10.2.4 additionally requires the use of defect prevention methods.
NEXT STEPReducing scrap starts with the data In the white paper “Management of Quality-Related Production Data,” you’ll find the approach for comprehensively capturing process data, defining early warning signals, and addressing the root causes of rework. |