A plant manager points to his OEE figure: 78%. “Solid,” he says. Three days later, the project team stands with him on the production line and counts along. The machine is running, but for 40 minutes per shift, it produces parts that later end up in rework. The OEE system counts these parts as good. So the efficiency isn’t 78%, but significantly lower—it’s just not being measured honestly anywhere.
The market promises a quick fix for such gaps: a dashboard, AI, a platform—and productivity will rise on its own. That’s rarely wrong, but almost never the whole story. Increased efficiency in production doesn’t come from a single tool, but from a combination of honest measurement, clean data, and processes that relieve the burden on the people on the production line rather than monitoring them.
Anyone who has seen enough production facilities from the inside knows the patterns. The biggest efficiency losses are almost never where reporting suspects them to be. They’re hidden in setup procedures, in data discontinuities between systems, in errors that go unnoticed until it’s too late, and in knowledge that exists only in individual minds.
THE MOST IMPORTANT POINTS IN A NUTSHELL
|
IN A NUTSHELL
|
|
CONTENTS OF THIS ARTICLE |
Before you adjust a single setting, you need to know whether your measurement is even accurate. In practice, it often isn’t.
OEE, the classic efficiency metric, multiplies availability, performance, and quality. The problem lies in the details: If rework is treated as an internal process rather than being recorded as scrap, it isn’t factored into the quality component. A line with a reported OEE of 78% often actually sits between 60% and 68% once you honestly factor in pseudo-output and unreported rework.
A second blind spot is the time resolution. Those who view efficiency solely as a shift or daily average fail to see short, recurring downtimes, which collectively represent the largest source of loss. In automotive projects, experience shows that these micro-downtimes of less than two minutes account for 15 to 25% of total availability losses, yet they do not appear in any daily report.
|
COMMONLY UNDERESTIMATED COST FACTORS
|
As soon as the assessment is honest, a recurring pattern emerges. Three levers appear in nearly every plant visited.
Setup time is dead capital—the equipment is idle and producing nothing. Structured setup optimization based on the SMED principle—that is, the separation of internal and external setup—reduces setup times by 30 to 50% in practice. This is not primarily a software issue in the initial phase, but rather one of methodology and discipline. Software then helps ensure that the optimized process remains reproducible.
In many plants, production data is handled three times: once at the machine, once in an Excel spreadsheet, and once in the ERP system. Every transition costs time and leads to errors. Closing these gaps not only saves time but also improves data quality. This is exactly where process data management comes into play.
An error that is only detected during the final inspection has already passed through all preceding stages of the value chain. The rule of ten for error costs states: With each process step, the cost of correcting errors increases tenfold. Early detection at the source is therefore one of the most cost-effective levers for improving efficiency.
|
Efficiency Leverage |
Typical Effect |
Prerequisite |
|
Setup time optimization (SMED) |
30 to 50% shorter setup time |
Methodological discipline, documented processes |
|
Eliminate data gaps |
Less data entry effort, higher data quality |
Integrated process data collection |
|
Early error detection |
Error costs reduced by a factor of 10 |
Inline measurement, anomaly detection |
|
PRACTICAL TIP: Manufacturing OS The Manufacturing OS’s IPM process data management system captures production data directly at the source, eliminating the need for data transfers. This means the data foundation for setup analysis and early defect detection is consolidated into a single system rather than three. Click here for more information. |
||
Every strategy mentioned in the previous section will fail at the same point if the data is inaccurate. An efficiency analysis is only as good as the numbers on which it is based.
Consistent production data means: the same variable is measured identically everywhere, unambiguously assigned to an order, a component, and a specific point in time, and stored without manual entry. Only then can sources of loss be accurately pinpointed rather than merely guessed at.
This is not just a theoretical claim. At the BMW Group, database archiving with CHRONOS measurably reduces storage costs because quality-relevant production data is stored in a structured and audit-proof manner rather than being scattered and redundant. Clean data is therefore not only a prerequisite for efficiency—it actually reduces costs.
|
MASTER DATA CHECKLIST
|
Standard Reference: IATF 16949 requires documented information in Section 7.5 and complete traceability in Section 8.5.2. ISO 9001:2015 requires data-driven decisions in Section 9.1. Those seeking to increase efficiency can meet these requirements simultaneously by ensuring clean data.
The second group of levers concerns the interaction between people, tools, and systems. This is where it is determined whether efficiency gains are reproducible or depend on individual shifts.
If every employee performs a task slightly differently, quality fluctuates—and with it, efficiency. Digital worker guidance specifies the process step by step and verifies critical steps. At Knorr-Bremse, worker guidance ensures quality in assembly; at Stadler Rail, it ensures the quality of the assembly processes as a whole. The result: shorter training periods for new employees and less error-related rework.
A worn-out tool produces defective parts, often over several cycles, before anyone notices. Systematic tool inspection shifts the focus from reacting to preventing and avoids entire batches of scrap.
AI-supported curve analysis detects deviations in process signals that a human cannot see during the current cycle. It flags the suspect component before it continues through the process. Important and required by regulations: AI provides decision support here, not autonomous approval.
|
WHEN AI-BASED ANOMALY DETECTION WORKS
|
Standard Reference: The EU AI Act classifies AI used in safety-critical production decisions as high-risk and requires transparency and human oversight. IATF 16949 Section 8.6 explicitly assigns the approval decision to a responsible person. Fully autonomous approvals by AI are not permitted in the automotive, medical technology, and aviation industries.
The first six levers only take effect when the underlying systems work together. As long as process data, operator guidance, tool inspection, and archiving run in separate silos, part of every efficiency gain is lost at the interfaces.
|
PRACTICAL TIP: Manufacturing OS CSP’s Manufacturing OS brings the four levers together on a single platform: process data management (IPM), operator guidance (PG), tool inspection (QST), and data archiving (CHRONOS). Instead of four isolated solutions, this creates a unified database that enables reproducible efficiency gains. Clients such as BMW, Mercedes-Benz, and Knorr-Bremse are already using components from this system in their production environments. |
The greatest efficiency gains never come from a faster machine, but from the fact that data finally converges in one place and everyone sees the same truth.
— Amadeus, Chief Technology Evangelist, CSP
Honesty is part of any consulting engagement. Software doesn’t automatically make production more efficient, and anyone who promises that has rarely stood on the production line.
No platform can compensate for a poorly balanced production line, unclear order control, or a lack of methodological discipline during setup. If the process isn’t right, a system merely digitizes the chaos more quickly. Data makes losses visible, but people are the ones who have to eliminate them. And while AI recognizes patterns, it bears no responsibility—that remains with the manufacturer under the EU AI Act and the 2024 Product Liability Directive.
The realistic approach: first stabilize the process, then measure it honestly, then create the data foundation, and only then automate and supplement with AI. In this order, the seven levers work together. In any other order, part of the investment goes to waste.
The most accurate way to measure efficiency gains in production is through OEE, which consistently records rework as a quality loss and captures micro-downtime lasting less than two minutes. Fine-grained time resolution is crucial, as shift or daily averages mask the most common sources of loss. A reliable measurement requires that all key performance indicators come from a single, unambiguous data source.
The fastest results are usually achieved through setup time optimization based on the SMED principle, as setup times can be reduced by 30 to 50% in practice without the need for initial investment in technology. This requires methodological discipline and documented processes. In contrast, eliminating data gaps yields the most sustainable results, as it simultaneously saves time and improves data quality for all subsequent measures.
Data quality is a prerequisite for any reliable increase in efficiency—not an optional add-on. Only when production data is consistent, unambiguously assigned, and captured without manual entry can sources of loss be accurately pinpointed rather than merely suspected. Poor data leads to optimizations in the wrong areas and thus results in double the cost.
AI boosts efficiency by detecting anomalies in process signals earlier than a human can during the production cycle; however, in safety-critical industries, it does not make autonomous approval decisions. The EU AI Act classifies such systems as high-risk and requires transparency and human oversight. AI thus serves as a decision-support tool; final responsibility remains with a qualified person.
Efficiency and quality are not opposites; rather, they are directly linked because every error destroys upstream value and triggers rework. According to the “rule of ten” for error costs, the cost of correction increases tenfold with each process step in which an error remains undetected. Early error detection therefore increases both efficiency and quality at the same time.
Depending on the starting point, double-digit percentage points across multiple levers are realistic—for example, 30 to 50% shorter setup times or error costs at the source reduced by a factor of ten. General promises regarding overall efficiency are unreliable because the effect depends heavily on the maturity of the processes and data. Reliable conclusions can only be drawn after an honest assessment of the current state.