Inefficiencies in Manufacturing: Hidden Losses

What are inefficiencies in manufacturing?

Inefficiencies in manufacturing are the avoidable losses — in time, material, quality, and energy — that stand between the plant's theoretical capacity and its actual output. Every factory has them; no factory has ever quantified all of them accurately without measurement. Three decades of industrial-automation work has taught me one pattern that repeats in every single plant I have ever connected: the losses are always larger than the team believes, and the gap between perceived and real is what kills the business case for improvement before the improvement work even begins.

I have spent since 1991 connecting machines — first with Simatic S5 and COROS visualisations, then via OPC UA and IoT gateways into cloud MES. Hundreds of line commissionings, seven countries, every type of brownfield you can imagine. The moment I look forward to in every project is the same one: the day the first real data comes online and the customer sees the gap between what they thought was happening and what is actually happening. That moment is where improvement starts. Everything before it is guessing. This article is about how to think about inefficiencies honestly — not as a Lean taxonomy exercise, but as the measurement problem they really are.

The taxonomy that actually shows up on the shop floor

The classical Lean taxonomy — the seven wastes of Muda, the Six Big Losses of OEE — is a useful starting point and every engineer should know it. But in 25 years of walking into plants with fresh data, I have never once encountered a plant where the losses arranged themselves neatly into the academic categories. The losses that actually appear fall into a different split, driven by whether they are visible to the operator at the machine or invisible without measurement.

The pattern in the right-hand column is the lesson of this article. The losses that cost the most are usually the ones you cannot see from the shop floor without proper measurement — micro-stops, speed loss, ramp-up loss, idle energy. The ones that are obvious — breakdowns, visible scrap — are usually not the biggest single contributor. Plants that optimise only what they can see will spend their improvement budget on 20 % of the problem while the other 80 % keeps running untouched.

The measurement paradox — why honest OEE goes down in week one

Here is the pattern I warn every customer about on the day we commission the first machine. The reported OEE on Monday morning, after the first week of real measurement, will almost always be 15-20 percentage points below what the plant had been reporting before. This is not a problem with the system. It is the system working correctly for the first time. The old number was an estimate, usually an optimistic one, usually built from manually-classified downtime and a good-count that did not include micro-stops or speed loss.

I have seen this so many times that I now say it in the kick-off meeting: "Your OEE is going to drop in the first week. That is good news, not bad news. It means you are finally seeing reality. The real number was always this low — you just did not know it." The plants that understand this point improve. The plants that argue with the new number — that try to "fix the measurement" so it matches the old number — never improve, because they have chosen comfort over truth.

The field observation from roughly 200 plant connections: the perceived OEE is almost always 15-20 points higher than the measured OEE at go-live. The missing capacity — the "phantom capacity" — is real, it is recoverable, and it is the reason the business case for improvement was stronger than anyone realised. But you cannot recover capacity you do not see. Every single connectivity project I have led has been fundamentally about turning invisible losses into visible ones. The improvement work that follows is almost anticlimactic by comparison.

The invisible losses — where the big numbers hide

Three loss categories dominate the "invisible" side of the split, and they are the ones worth naming in detail because every plant has them and almost no plant sees them clearly.

Micro-stops. Stops under three minutes individually. An operator reaches into the machine to clear a jam, a sensor triggers and the line pauses for 40 seconds, a light curtain registers a reach-in. None of it feels like "downtime" to the operator — the machine was "basically running." In aggregate, across a shift, across dozens of machines, across a week, micro-stops typically account for 8-15 % of total loss time in discrete assembly plants. In packaging lines they can exceed 20 %. None of it appears in manual downtime logs because operators do not classify 40-second stops. Only event-level capture at the machine surfaces it.

Speed loss. The machine is running — green light, normal state — but not at nameplate rate. Maybe the operator reduced the feed because the material was off-spec last shift. Maybe a tool is dull and nobody noticed. Maybe the cycle time has drifted up over six months, 200 milliseconds per cycle, and it never crosses an alarm threshold. Speed loss typically accounts for 5-12 % of theoretical capacity in plants that have never measured it, and it is almost always discovered retroactively once cycle-level data is available. It does not feel like a problem in the moment. It is a problem.

Ramp-up loss. After every changeover, every break, every weekend shutdown, the line takes some time to reach full rate. The "some time" is rarely measured. In plants with frequent changeovers, ramp-up loss can consume 3-8 % of available time — and the fix is usually procedural, not technical, once the loss is visible. But again, without cycle-by-cycle capture immediately after restart, the loss blends into "normal operation" and nobody works on it.

These three categories together typically account for 15-25 % of total equipment loss in plants that have never measured them. That is not a rounding error. That is the difference between a plant making its margin targets and one that is not.

The diagnostic question — do you know, or do you estimate?

Every improvement conversation I have ever had on a shop floor runs into the same fork. The plant team has a view about where their biggest losses are. The data has a different view. Either the two align — in which case the improvement work is obvious and proceeds — or they do not align, and the first task is to figure out which view is wrong. Almost always, the estimate is wrong and the data is right. Not because the operators are careless, but because the human eye systematically underweights short, frequent events and overweights long, dramatic ones. A six-hour breakdown feels like "the problem." Two hundred 45-second micro-stops over the same week feel like "just normal running." In time-equivalent, they are the same number. In visibility, they are not.

The diagnostic question I apply on every site visit is this: can the plant team tell me, without guessing, what their five largest loss categories were this week, in minutes, ranked? If yes — the measurement infrastructure is there and the improvement work is the next conversation. If no — there is no point having the improvement conversation yet. Losses that are not measured are not addressable. They are just folklore.

A real case: Carcoustics International

Carcoustics International is an international Tier-1 supplier of acoustic and thermal solutions for the automotive industry, with production sites in Germany, Poland, Slovakia, the Czech Republic, Mexico, the USA and China. The process mix is exactly the kind of heterogeneous brownfield that makes inefficiency capture hard: injection moulding, cold foaming, stamping — completely different process dynamics, completely different control systems, some lines modern, some decades old. The engagement started with a proof-of-concept at the Haldensleben plant and scaled within six months to more than 500 connected machines across every site.

The technical approach is worth describing in detail because it addresses exactly the "invisible losses" problem. The OT connectivity was built on IXON IoT gateways using the MQTT protocol into Microsoft Azure. This meant no SPS intervention on the hundreds of existing machines — a hard requirement in automotive brownfield where touching the control system is disallowed during production. Every machine state, every cycle, every stop was captured at event level and streamed to the cloud MES in real time. Digital setup-process support was added at the injection-moulding cells to make changeover loss visible for the first time. Bidirectional integration into SAP R/3 via ABAP IDoc mapped every machine cycle to the originating production order, so loss analysis could be done at the order level — not just at the machine level.

The results after six months are a textbook example of what happens when the invisible losses become visible:

The 8 % availability improvement is the number that tells the real story. It was not achieved by buying new machines, adding shifts or hiring more maintenance staff. It was achieved by making the losses visible on 500+ existing machines, then letting the plant teams act on what they could finally see. The modular SYMESTIC catalogue meant the Carcoustics team could extend the connectivity to additional sites themselves, without returning to the vendor for every new rollout — which is the sustainable pattern for inefficiency elimination across a multi-plant enterprise.

FAQ

What are the main types of inefficiency in manufacturing?
The losses that actually show up in plants fall into eight categories: unplanned downtime, planned downtime, micro-stops, speed loss, startup/ramp-up loss, quality loss (scrap and rework), material loss, and energy loss. The Lean "Seven Wastes" (Muda) and the OEE "Six Big Losses" are useful taxonomies, but the practical split that matters is not academic — it is visible versus invisible, because the invisible losses (micro-stops, speed loss, ramp-up) are usually the biggest and the ones plants have never measured.

Why does reported OEE drop when we start measuring it properly?
Because the old number was an estimate, usually an optimistic one. Manual downtime classification misses micro-stops, speed loss is invisible to the eye, and rework time is rarely included. Event-level capture at the machine surfaces all of it. The drop of 15-20 percentage points in week one of real measurement is not a problem with the system — it is the system telling the truth for the first time. The plants that improve accept the new number and work on it. The plants that argue with the new number do not improve.

What are micro-stops and why do they matter?
Micro-stops are machine stops under three minutes — jams, sensor triggers, reach-ins, brief material interruptions. Individually they feel inconsequential to operators. In aggregate across a week, across dozens of machines, they typically account for 8-15 % of total loss time in discrete assembly plants and over 20 % in high-speed packaging. They are effectively invisible without event-level machine data capture, because no operator classifies 40-second stops in a manual log. The first time a plant sees its micro-stop Pareto is usually a shock — and usually the biggest single improvement lever available.

How do we identify inefficiencies we cannot see?
The honest answer is that you do not identify them by walking the floor or by reviewing production reports — you identify them by capturing machine events at cycle level and letting the data surface the patterns. Speed loss, micro-stops, ramp-up loss and idle energy consumption all require automated capture at the machine to become visible. Manual observation systematically underweights short, frequent events and overweights long, dramatic ones. The human eye is not the right instrument for this measurement.

How much inefficiency is "normal" in a well-run plant?
World-class OEE benchmarks are around 85 % for discrete manufacturing, meaning roughly 15 % of theoretical capacity is lost even in excellent plants. Typical mid-market plants run between 55 % and 75 % OEE when measured honestly — so 25-45 % of theoretical capacity is lost. The ranges differ by process type (batch, flow, job-shop, continuous) and industry, but the gap between perceived and measured is nearly universal. The question is never "do we have inefficiency" — it is always "how large is ours and where is it hiding."

Is the classical Lean taxonomy (Seven Wastes) still useful?
Yes as a thinking framework; no as a measurement framework. The Seven Wastes of Muda (overproduction, waiting, transport, over-processing, inventory, motion, defects) are excellent for training and for structuring improvement workshops. They are not directly mappable to machine-level data capture, which is where the quantification has to happen. In practice, the OEE-based split (availability, performance, quality) is the one that matches automated data capture, and the Lean categories layer on top of it for root-cause analysis once the quantitative baseline exists.

Can old machines even be measured for inefficiency?
Almost always, yes. This is the single most persistent misconception I encounter. Machines from the 1990s without any digital interface can be connected via digital I/O gateways that read existing signals (cycle-count sensors, alarm lines, safety-circuit states) without any PLC modification and without any production interruption. Typical installation time is 2-4 hours per machine. The idea that "our old machines can't deliver data" is factually wrong about 95 % of the time — it is a story the plant has told itself because nobody specialised in brownfield connectivity has looked at the problem yet.

How does SYMESTIC help identify and reduce inefficiencies?
Event-level capture at the machine — cycles, states, alarms, micro-stops — via OPC UA for modern controls, MQTT for IoT-gateway-connected assets, and digital I/O gateways for brownfield machines without any digital interface. No PLC intervention, no production downtime, 2-4 hour installation per machine. Automatic OEE, speed-loss, micro-stop and ramp-up-loss calculation across all connected assets. Pareto and waterfall visualisation as first-class report primitives so the biggest losses are always visible. Bidirectional ERP integration (SAP, Infor, proAlpha, Navision, Dynamics) so losses can be attributed to orders, products and customers. 15,000+ machines connected across 18 countries using exactly this pattern. See SYMESTIC Production Metrics.

Related: OEE · Six Big Losses · Seven Wastes (Muda) · Machine Data Capture (MDE) · Production Data Capture (BDE) · Micro-Stops · MES · SYMESTIC Production Metrics

About the author

Martin Brandel

MES Consultant at SYMESTIC. 30+ years in industrial automation — started as automation engineer in 1991 with Simatic S5 and COROS visualisation, automation projects at Hermos AG (paint lines and process plants across Eastern Europe and China), built and led the Automation Technology division at SYMESTIC for 11 years. Now a hands-on MES project lead from first enquiry to go-live, specialising in brownfield machine connectivity without PLC intervention. Dipl.-Ing. Communications Engineering. · LinkedIn