Reducing defects in China fulfillment requires identifying where the failure originated and where it could have been detected. A wrong component from the factory, a damaged carton during handling and a wrong variant picked for an order need different corrective actions.

Begin with a defect record that preserves those distinctions. A single “quality problem” code cannot show whether the next investment should be in supplier controls, packaging, product mapping or warehouse training.

Agree what the rate counts

Define the population, inspection method and unit of measurement. A defective-unit rate counts units with at least one relevant defect. A defects-per-unit measure counts individual findings, so one unit can contribute several defects.

In an illustrative inspection, 1,000 units contain 35 nonconforming units and 45 recorded defects. The defective-unit rate is 3.5%, while defects per inspected unit are 0.045. Neither figure should be described as the operation’s customer return rate.

Sampling results and complete inspection results also have different limitations. NIST’s acceptance-sampling overview describes sampling as a lot-decision process. It does not turn a sample finding into a guaranteed shipment-wide percentage.

Classify cause and severity separately

Use a short, stable failure taxonomy tied to evidence. Possible categories include product manufacturing, inbound identity, storage damage, pick error, pack failure and transport damage. Include an unresolved category while the cause is genuinely unknown.

Severity must remain separate from frequency. A rare safety issue can require immediate containment even if a frequent cosmetic defect dominates the count. A ranked frequency chart should never hide a critical finding.

For each record, retain SKU and variant, lot, supplier, location, process step, discovery point and photographs or measurements. Avoid assigning blame based solely on where the defect was found. A cracked item noticed at packing may have been damaged before receiving.

Identify the lot or other defensible population that could share the problem. Hold the affected inventory and prevent its release through the selling or warehouse systems.

Document who may narrow the hold and which evidence supports that decision. Releasing unaffected stock can be reasonable, but the distinction should follow traceable production or handling information rather than commercial pressure.

If a defect concerns safety, regulatory compliance or a possible recall, involve the responsible specialists. An ordinary warehouse rework instruction is not an adequate response to an unresolved safety question.

Test a cause with process evidence

Suppose an illustrative review finds wrong-size garments concentrated in one pair of adjacent bins. Check receiving records, barcode mappings, label similarity and observed pick instructions before concluding that operator attention is the sole problem.

A useful corrective action might change the location arrangement or require a variant-specific scan. Training alone may leave the underlying ambiguity unchanged. Conversely, a scanner cannot correct a barcode that was assigned to the wrong SKU.

For packaging damage, compare the failure location with the approved pack design and handling sequence. The corrective action should address the observed mechanism, not simply add more cushioning to every product.

Verify the action on comparable exposure

Record the baseline, change date, eligible product population and verification method. Compare similar order mixes and inspection coverage. A falling count after a quieter sales week is not evidence that the process improved.

For an illustrative follow-up, inspect comparable quantities of the affected variant before and after the mapping correction. Report both the finding count and inspected denominator, including any changes in selection method. Do not claim the difference proves causation if other processes changed at the same time.

NIST’s OC-curve guidance also explains why sampling decisions retain statistical uncertainty. Keep that uncertainty visible when interpreting a small follow-up sample.

Close the corrective-action record only when the change is implemented, its effect has been checked and its control owner accepts responsibility for maintaining it. A lower dashboard number is useful; an explained and sustained process change is what prevents recurrence.

Sources and verification links

Operational rules can change. Check the linked primary sources and confirm shipment-specific requirements before booking.