Notes from the floor

When behavior goes wrong, the reflex is to write a procedure. Another policy, another required sign-off, another rule. But a procedure cannot fix a problem that is really about what people value and how they behave when no one is checking.

In a lot of operations everything is a fire. Every issue is escalated, every problem is urgent, every day is reactive. But when everything is urgent, nothing is, and the genuinely important work of prevention never gets done because the loud keeps crowding out the vital.

It is comforting to believe that a 100 percent inspection catches 100 percent of defects. It does not. Human visual inspection typically catches somewhere around 80 percent, which means relying on it as your control quietly guarantees escapes.

Traceability systems get sold as quality. But being able to trace a defect after the fact is not the same as preventing it. You can have perfect traceability and no control at all, and confusing the two leaves you with an excellent record of problems you never stopped.

It is one of the most repeated slogans in the field: quality is everyone's responsibility. It sounds right. In practice it is often how quality becomes no one's responsibility, because a job that belongs to everyone belongs to no one in particular.

Experience is a genuine asset, right up until it becomes the reason nobody questions an assumption. The person who has seen it a hundred times is also the person most likely to pattern-match a new problem to an old answer and stop looking.

Everyone treats quality and speed as opposites, a dial you turn toward one at the expense of the other. In most operations that is simply false. The fast way and the right way are usually the same way, and the belief that they conflict is what creates the conflict.

The control plan was carefully built at launch, approved, and then quietly frozen. Meanwhile production drifted, tooling wore, methods evolved, and new failure modes appeared. The document on file describes a process that stopped existing years ago.

Operations track defects, scrap, and returns, the things that went wrong and got counted. Meanwhile the richest signal, the near misses, the deviations, the quiet adjustments operators make to keep things running, goes unrecorded. The most useful data is the data nobody keeps.

When a defect shows up in production, the reflex is to fix it on the floor. But a large share of what gets called a manufacturing quality problem was actually decided in design, long before anyone made a part. You cannot manufacture your way out of a design that was never producible.

The standard response to a defect is to add a control. Another check, another step, another sign-off. Each one feels like progress. Stack up enough of them and you have built a process so complex that it becomes its own source of new defects.

Every time you accept a deviation to keep the line running, waive a requirement under schedule pressure, or let a late corrective action slide, you teach your supplier what you actually reward. Suppliers behave the way you train them to, and most of that training is accidental.

Most quality effort is aimed inside processes, at the machine, the operator, the inspection. But a large share of real problems do not happen inside any one step. They happen in the gaps between them, in the handoffs nobody owns.

The customer is always worth listening to. But listening is not the same as agreeing. Sometimes a complaint is about a part that fully meets the requirement, and treating every customer claim as a real defect quietly wrecks your process and your data.

Lean gets deployed as a polite word for cutting. Reduce headcount, strip the buffers, call it efficiency. Done that way it delivers a short-term number and a long-term mess, because it removes the slack that was quietly protecting quality and delivery.

We train people to fill out an FMEA, run an 8D, build a control chart. Then we are surprised when the FMEAs are shallow and the 8Ds fix nothing. We taught the form and skipped the thinking the form was supposed to capture.

Scrap gets counted because you throw it away. Rework gets absorbed, because the part still ships. So it hides in labor hours and normal routine, uncounted and unquestioned, often costing more than the scrap everyone worries about.

There is the quality system on paper, and the quality system that actually runs the floor. In a lot of organizations they are two different things, and the gap between them is where problems live and audits get survived.

A Cpk of 1.67 on the report looks like control. But capability is only meaningful if the process was stable, the measurement was trustworthy, and the sample actually represents production. Break any of those and the number is decoration.

The skilled operator who quietly makes bad parts good is a gift and a liability. Every save they make is a problem the system never has to fix, right up until they retire and the defects arrive all at once.

Trace enough defects back and a pattern appears. The check that got skipped, the step that got rushed, the marginal part that got passed. Behind a surprising number of quality problems is not a quality failure at all. It is the schedule.

Every experienced team has a private list of the tolerances nobody really holds, the callouts that do not match function, and the prints everyone quietly ignores. The specification says one thing. The floor has decided what actually matters. That gap is a real risk.

Automation is sold as a quality solution. It is not one. A robot running an unstable process does not fix the instability. It repeats it, precisely and quickly, and hands you the same defects at a higher rate and a higher cost.

The instinct is to measure a quality department by how much it catches. The better measure is how little it needs to. A quality function that is indispensable as a last line of defense has usually failed at its real job, which is to make itself unnecessary.

When something goes wrong, the fastest explanation is that a person made a mistake. Write down operator error, add retraining and a reminder to be careful, close the action. It feels like an answer. It is one of the most expensive habits in manufacturing.

Six Sigma has a real toolkit and a real track record. The problem is that a large share of the projects run under its banner wrapped a simple problem in a heavy process and reached an answer a competent person could have found in an afternoon.

Boards on the wall, a suggestion system, kaizen events on the calendar, a slide counting projects completed. Far fewer programs actually move the numbers the business cares about year over year. Most of it is improvement theater, and everyone can feel the difference.

The dashboard is open, the metrics are reviewed, the actions are assigned, and the people in the room may go weeks without setting foot on the floor. It feels efficient. It is one of the quieter ways a quality function loses touch with reality.

A quiet quarter feels like a good one. No angry calls, no returns, no escalations. The absence of complaints gets read as evidence that quality is fine. It is one of the most dangerous inferences in the business.

Adding inspectors feels like doing something about quality. It rarely is. Inspection sorts good from bad after the fact. It does nothing to the process that made the bad parts in the first place.

Zero defects makes a great banner and a poor strategy. Chasing the slogan drives the wrong behavior. What you actually want is a capable process, and that is a different target entirely.

A clean audit and a healthy quality system are not the same thing. One measures whether you follow your procedures. The other is whether your procedures actually prevent problems. Plenty of certified plants have a spotless certificate and recurring escapes.

PPM is the headline quality metric in most plants. It is also one of the easiest to game and the easiest to misread. A falling PPM can hide a rising problem, and a single number was never going to tell you where your quality actually stands.

When prevention works, nothing happens. No defect, no complaint, no fire to put out. That invisibility is exactly why prevention is the first thing questioned in a budget review, and why so many plants stay stuck firefighting.

When a process misbehaves, the instinct is to change one setting at a time. It feels scientific. It is slow, and it misses the answer, because the real cause usually hides in the interactions that one-at-a-time testing cannot see.

Every dimension measures in tolerance, the report is all green, and the parts still will not assemble. The culprit is usually not the gauge. It is a drawing that measured the wrong thing.

Most quality problems are born months before production, in the gap between designing a product and figuring out how to make it. APQP is the discipline built to close that gap on purpose, before a customer sees a defect.

Most manufacturers run a supplier scorecard. Far fewer have one that changes how suppliers behave, which is the only reason to have one. The difference is what you measure, how you share it, and what happens when the number is bad.

Calibration is the least glamorous system in quality and one of the most important. Every measurement is a comparison against a standard, and calibration is what keeps that standard honest.

Today the United States turns two hundred and fifty. I came here at sixteen with no English. A career in quality, two degrees earned later in life, and a firm I founded all trace back to a bet this country makes on people it has never met.

Sensors on every machine and AI on every image. Some of it is real and useful. Some of it is a faster way to make the same defects. The difference is whether the process underneath is capable.

Most manufacturers already pay for Microsoft 365. Far fewer use it to run quality. Here is how Planner, Power Automate, Power Apps, and Power BI fit, and where the new AI features need a firm hand.

Every capability study, control chart, and pass-or-fail decision rests on one assumption: that the measurement is real. A measurement system analysis checks whether it is.

Inspection asks people to catch every error. Mistake-proofing changes the process so the error cannot happen, or cannot pass unnoticed. Only one of those actually holds.

I spent time on the floor at Automate 2026 in Chicago as a quality engineer, not a robotics buyer. That lens changes what stands out, and what to be careful about.

The control plan is the one document that says, for every characteristic that matters, how you keep the process good and what you do the moment it goes wrong. Most are written to satisfy an auditor. That is a waste.

A layered process audit checks the actual process, often, by many people, at the point where the work is done. It catches drift before it becomes scrap, and it is underused outside automotive.

A process FMEA is supposed to stop failures before they reach a customer. Too often it becomes a form filled in after the line is already running. Here is how to keep it a real prevention tool.

The two indices look almost identical and get confused constantly. The gap between them is trying to tell you something about your process.

OEE turns three kinds of loss into one number. The number is useful. The benchmark behind it gets quoted with far too much confidence.

Most plants track the scrap they can see. The larger cost hides in rework, sorting, and the failures that reach the customer.

A problem gets closed, then comes back two months later. Usually the team fixed the symptom and missed one of the two root causes.

Lean Six Sigma works when it solves a real problem on your floor, not when it becomes a belt-collecting exercise. Here is how to tell the difference.

IATF 16949 is ISO 9001 plus a demanding layer of automotive rules. Here is the realistic path, and the parts that trip suppliers up.

A quality management system fails when it becomes a binder no one opens. Here is how to build one that runs the way your plant actually works.

Most supplier defects are planned in, not made in. PPAP and APQP exist to catch them before the first production run, if you use them honestly.

A quality dashboard is not about prettier charts. It is about seeing a problem early enough to act. Here is what actually belongs on one.

SPC is not more inspection. It is how you separate normal variation from a real signal, so you act when it matters and leave a stable process alone.

The honest ranges for a small-to-mid manufacturer, what really drives the number, and why the certificate fee is the smallest part of the bill.

Certification day gets the attention. The audit a year later is the one that catches teams off guard. Here is why, from a lead auditor's chair.

Two industries, two languages for proving a part is right the first time. The plain-English difference, and when each one applies.

"Gap assessment" sounds vague. Here is exactly what you get, why it matters before an audit, and how to tell a useful one from a box-checking exercise.

A recurring defect usually does not mean the corrective action failed. It means it solved the wrong problem. Here is how to tell the difference.