The setter on days could bring that line into spec by the third piece. Everybody else on that job needed twelve. If you are working out how to reduce scrap in manufacturing, that gap is where I would start.
Nobody at the plant filed this under quality. They filed it under “good setter,” the way plants do, and they were right that he was good. What they missed is that on a two-hundred-part run, the difference between his setup and everybody else’s was nine parts of scrap every single time the job changed over. Multiply that by the changeovers in a month and you have the real answer to how to reduce scrap in manufacturing sitting in one man’s hands, undocumented.
Why is your scrap report sorted the wrong way?
Every scrap report I have been handed sorts by defect code and part number. Sometimes by machine. That sorting answers “what went wrong,” which is the question quality asks.
It does not answer “when in the run did it go wrong,” which is the question that finds the money.
Sort the same data by position in the run and a flat picture becomes a slope. In most discrete plants, scrap is not evenly spread across a run. It piles up in the transition, while somebody brings the process in from memory, and then goes quiet until the next changeover.
How do you reduce scrap in manufacturing?
How to reduce scrap in manufacturing starts with finding out where in the production run your scrap is actually happening, before you try to reduce it. If it concentrates around the changeover, the cause is setup knowledge that was never written down rather than process capability, and buying inspection technology will detect the scrap faster without producing less of it. Capture how your best setter brings the job in, reconcile it against how everybody else does it, and make that the documented setup.
What does the research say about changeover transitions?
Enough to take the transition seriously as its own regime, and less than the headline suggests. A 2025 study in Applied Sciences by Andersson, Hvam, Forza and Mortensen applied logistic regression to production data from a large chemical manufacturer and found that product made during the ramp-up before a major changeover was associated with a 4.1-fold increase in the odds of quality failure. Before a minor changeover, product was nearly three times more likely to show quality issues.
Three things about that finding have to be said plainly, because the honest version is more useful than the loud one.
It is continuous process manufacturing, not discrete assembly, measured at the packed big-bag level in one company. It is an odds ratio, an association rather than a measured scrap percentage. And the window it measured is the ramp-up leading into a changeover, not the parts that come off the machine immediately after one.
So the study does not prove the claim a machine shop cares about. What it establishes is that production around a changeover behaves as a distinct and riskier regime, which is what every setter will tell you for free. The discrete version of that window sits just after the setup. Treating the two as the same phenomenon is a reasonable inference, not a measurement. That is what the sort below is for: it measures your plant instead of borrowing someone else’s.
What does scrap actually cost?
APQC’s open-standards benchmark for scrap and rework costs as a percentage of sales puts the cross-industry median at 1.0%, across a sample of 1,008 companies. That is a cost ratio rather than a scrap rate, and it is cross-industry rather than manufacturing-specific, so treat it as context.
One percent of sales sounds survivable until you set it beside your operating margin instead of your revenue. Run that division on your own numbers before deciding this is a small problem.
Why did Lean point everyone at the clock?
Because the clock was measurable and the method was not. Decades of quick-changeover work, SMED, internal versus external setup, single-digit minutes. Search for changeover improvement today and essentially every result is about duration.
That work is valuable and I have run it myself. But shortening the window does not change what happens inside it. A plant can cut changeover from ninety minutes to thirty and produce exactly the same nine scrap parts, faster.
There is also a trade nobody prices: pushing setup time down without capturing the setup method first pushes setters toward speed over verification. The clock improves and the front of the run gets worse. If how to reduce scrap in manufacturing is the actual goal, duration is the wrong target.
What is the market selling when you ask how to reduce scrap in manufacturing?
Search the phrase and the results converge on four answers: better inspection, real-time monitoring, predictive analytics, and operator training.
Three of the four sell you a faster alarm. Knowing sooner that you made a bad part is worth something, but it is not the same as making fewer.
The fourth is the right instinct aimed at the wrong target. You cannot train a setup method that has never been written down, so everybody ends up trained on whoever was free that week, which is the same inconsistency that produces the defects in the first place.
The position-in-run scrap sort
One afternoon, one spreadsheet, no budget.
- Pull the last 90 days of scrap transactions for two or three high-changeover jobs.
- Add a column: was this piece inside the first 25 pieces after a changeover, yes or no. Use your own number if 25 is wrong for your process.
- Add a second column: who performed the setup.
- Total the scrap in each bucket and express the front-of-run bucket as a share of all scrap on those jobs.
- Compare that share against the share of total production those first pieces represent. On a 200-part run, 25 pieces is about 12% of the output.
If the front of the run produces 12% of your parts and 40% of your scrap, you are looking at a setup-documentation problem wearing a process-capability costume. In one aerospace machine shop where I ran this sort, the front of the run carried 62% of the scrap on the jobs we looked at. That is one plant’s result from one afternoon, not a benchmark, and your number will be your own. It is also the cheapest honest answer to how to reduce scrap in manufacturing that I know of.
The second column matters as much as the first. If one name is consistently attached to the clean runs, you have located the knowledge that needs capturing, and you located it without a single meeting. That is the same move behind separating human error from root cause, applied to the setup rather than the investigation.
What do you do with the name in column two?
Do not promote him and do not make him the trainer. Both of those move the bottleneck without removing it.
Go and capture what he does. Not the steps, which are probably in the setup sheet already. The judgment: what he looks at first, what he listens for, which dimension he checks before the sheet tells him to, and the three adjustments he makes that nobody documented because they were obvious to him.
At SenseiLab this is the whole job. The setters and leads talk through the setup directly with an AI agent while they are performing it, in their own words, so tribal knowledge comes out attached to the steps instead of staying in one man’s hands. It is the same capture logic behind measuring rework properly, aimed at the most expensive four minutes of the run.
You can keep buying faster ways to find bad parts. Or you can spend one afternoon finding out when they are being made, and one week writing down how the person who avoids them does it.
The second one is cheaper, it is how to reduce scrap in manufacturing without buying anything, and it is the only one that survives him leaving.
FAQ
How do you reduce scrap in manufacturing? How to reduce scrap in manufacturing starts with finding where in the production run the scrap occurs. Sort 90 days of scrap by whether each piece was made in the first parts after a changeover and by who performed the setup. If scrap concentrates around the changeover, the root cause is undocumented setup knowledge rather than process capability.
Why is scrap highest around a changeover? Because the process has not stabilised and the person bringing it in is working from experience rather than a document. A 2025 study in Applied Sciences found product made during the ramp-up before a major changeover carried 4.1 times the odds of falling outside specification, in a continuous-process chemical plant. The setup method that closes that window quickly is usually held by one or two people and written down nowhere.
How do we reduce rework or defects caused by inconsistent processes? Reconcile the versions before you standardise anything. Inconsistent processes are usually several working methods held by different people, not an absence of method. Capture each version, decide step by step which becomes the standard, then document that.
What do scrap and rework typically cost a manufacturer? APQC’s open-standards benchmark puts the cross-industry median for scrap and rework costs at 1.0% of sales across 1,008 companies. That is a cost ratio, not a scrap rate, and it is cross-industry rather than manufacturing-specific. Treat it as context and compare your own figure against operating margin rather than revenue.
Does SMED reduce scrap? Not directly. SMED shortens the duration of the changeover window without changing what happens inside it. A plant can cut changeover time by two thirds and still produce the same number of bad parts at the start of every run.
Author bio
Diego Echenique is CEO and co-founder of SenseiLab, a knowledge capture and operational excellence firm in Aventura, Florida. He has spent more than 20 years in manufacturing operations across automotive, mining and heavy industry, launching plants and leading Lean and Six Sigma transformations in Argentina, Chile, Europe, the Middle East and Asia.
