Testing every unit sounds expensive until you price a single escape at a customer site. Then the arithmetic reverses, usually decisively.
Sampling is a rational response to cheap, high-volume, stable products. Industrial assemblies are usually none of those things: batches are small, mix is high, and the cost of a failure at the customer is measured in machine downtime rather than in the price of the part.
The cost of testing every unit is straightforward: fixture development, cycle time per unit, and the operator time to handle a failure. In the programs we run, that is typically a small share of the conversion cost, and it drops as the fixture matures.
The saving is harder to see because it is a cost that does not happen. One escape that reaches a machine builder's final assembly can consume a day of two teams. One that reaches an end customer consumes a service trip, a replacement unit and a portion of the trust that took years to build.
There is a second benefit that rarely gets counted: data. A test record per serial number turns quality from an opinion into a trend. Drifts appear before they become failures, and a field question resolves to a batch instead of a fleet. That data is the actual product of testing; the pass/fail is just the headline.
— GANI Manufacturing engineering team