Sampling answers a different question than 100% testing, and the wrong choice either wastes money or ships defects with statistical blessing. The decision follows from failure cost, not habit.
Sampling and 100% testing are often discussed as cheap versus thorough. That framing hides the real distinction: they answer different questions. A sampling plan tells you, with stated confidence, that a lot’s defect rate is probably below a threshold. 100% testing tells you that this specific unit, serial number visible on the label, passed. Which question does your customer need answered?
The decision follows from failure cost. If a defective unit means a warranty return and an apology, statistical lot control may be entirely adequate. If it means a field failure in a machine, a safety function that does not trip, or a recall measured in truck rolls, then the confidence interval is not the point — the individual unit is. That product gets tested, every unit, every time.
What makes 100% testing affordable is designing the test into the line instead of appending an inspection department to it. A test rig at the end of the cell, cycle-timed to the takt, executing a defined program — continuity, function, parameters against limits — costs seconds per unit. The same coverage performed manually offline costs minutes and drifts with attention.
The test record is worth as much as the test. Our rigs log every unit’s results against its serial number: measured values, limits, timestamp, program revision. When a question arrives from the field two years later, the answer is a database query, not an archaeology project. This is what serial traceability means in practice — not a label, but the data behind it.
Be suspicious of 100% visual inspection presented as 100% testing. Human visual inspection catches a fraction of defects, and the fraction falls with every hour of shift. It has a place as a final gate for cosmetic criteria. As the sole defence for a functional requirement, it is sampling with extra steps and worse statistics.
And measure whichever regime you run. A sampling plan nobody recalculates after a process change, or a test program whose limits were never revisited after the first design revision, both provide confidence they no longer earn. The plan is a living document; the day it stops being maintained, you are shipping on reputation.
— GANI Manufacturing engineering team