How to scope quality control for mixed SKUs: part 6

Key takeaways

1. Treat each SKU family as its own sampling lot, because acceptance sampling mathematics assumes units drawn from one homogeneous process.^[1]
2. Weight sample sizes by order value, product complexity, and defect history instead of splitting inspector time evenly across lines.
3. Anchor the scoping method in a documented quality process so mixed-SKU decisions are repeatable, reviewable, and auditable.^[2]
4. Set per-stratum accept and reject numbers before the visit so borderline findings do not trigger renegotiation on the factory floor.
5. Record results per SKU even when the shipment gets a single overall verdict, preserving the detail that drives corrective action.

A mixed-SKU shipment looks like one inspection job on the booking form, but statistically it is several jobs sharing a container. Each SKU family usually comes off its own line, its own material batch, and sometimes its own subcontractor, which means defects in one family tell you little about defects in another. Treating the whole container as a single homogeneous lot breaks the basic assumption that sampling math relies on.

The practical consequence is misallocated effort. Junior QC plans often divide inspector time evenly across SKUs, or pull one flat sample from the shipment total, which over-checks simple, mature lines and under-checks complex or troubled ones. The risky SKU is exactly where a systematic problem hides, and a thin sample there is how escapes happen.

Risk-based scoping fixes this without adding inspection days in most cases. You group SKUs into strata, assign each stratum a sample size driven by its quantity and its risk weight, and write per-stratum accept and reject decisions before anyone travels. What follows is a repeatable method for doing that, plus the documentation habits that make results comparable from one shipment to the next.

Why even splits fail in mixed cartons

An even split feels fair, but fairness to SKUs is not the objective; finding defects is. When five SKUs share one container, their risk profiles are rarely equal: one may be a mature repetitive item, another a first-time new development sample scaled into production, a third a decoration-heavy variant with a history of finish complaints. Giving each the same twenty samples spends as much effort confirming the safe bet as probing the doubtful one.

The statistics agree with common sense here. A sample drawn from a mixed pile cannot represent any single SKU family, so its accept or reject verdict answers a question nobody asked: whether the average of unlike things is acceptable. Worse, a defect cluster in one small SKU can be diluted below the acceptance threshold by clean units from the rest, producing a pass that certifies nothing about the weak line.

The fix is stratification: divide before you sample. Each stratum gets its own sample size, its own accept number, and its own verdict, and the shipment decision becomes a reasoned combination of stratum results plus business judgment about value and urgency. This is more planning work up front, but it is the only approach whose numbers mean what they claim to mean.

Build risk strata before you count samples

Stratification is a planning exercise done at a desk, ideally days before the inspection, because it needs the packing list, the production layout, and the complaint history in one view. Resist the temptation to let the inspector improvise strata on arrival: decisions made under time pressure on the factory floor consistently overweight whatever is staged nearest the door. A short written scoping sheet, agreed with the supplier, protects both sides.

Group by process, not by packing list order

Start from how the goods were made, not how they are packed. SKUs that share a production line, material batch, and operator team belong in one stratum even if they ship in different cartons, because they share failure modes. SKUs from different workshops or subcontractors belong apart even when they look similar on paper, because their defect causes are independent. Ask the supplier for the production layout per SKU family; the answer usually redraws your strata.

Score each stratum on three axes

Rate every stratum on commercial value at risk, technical complexity, and quality history, using a simple high-medium-low scale rather than a fragile numeric model. Value covers unit price times quantity plus the cost of a stockout at destination. Complexity covers part count, tight tolerances, surface finishing, and any regulated claims such as food contact or child safety. History covers prior rejections, complaint rates, and whether the factory is new to you. Two highs out of three earn the stratum an upgraded inspection level.

A worked allocation for a six-SKU shipment

When pooling samples is defensible

Pooling means drawing one sample across several SKUs and judging them together, and it is legitimate in exactly one situation: when the pooled SKUs are interchangeable for quality purposes. Same line, same materials, same checklist, same failure modes. Spare parts from the same mould run and colour variants of an identical body are classic candidates, because a defect mechanism in one predicts the same defect in the others.

Where SKUs differ in process or risk, pool the paperwork but not the sampling. You can still report them compactly and ship them together while keeping each stratum's sample and verdict independent. Many buyers ask a provider of third-party product inspection in China to review the pooling logic at booking stage, so the factory hears one consistent plan instead of negotiating methodology with the inspector on arrival.

Document the pooling rationale in one paragraph on the scoping sheet: which SKUs were pooled, why they qualify as homogeneous, and what trigger expands the sample. That paragraph is what lets you defend the decision months later if a customer complaint questions whether the goods were properly checked.^[1]

From packing list to inspection plan in five moves

1. Demand the full breakdownList every SKU with quantity, carton count, production workshop, and whether it is finished and packed. Refuse to scope from a one-line total; mixed-SKU math needs the breakdown.
2. Form risk strataMerge SKUs that share process and failure modes, split those that do not, and score each stratum high, medium, or low on value, complexity, and history.
3. Size samples per stratumRead the sample size and acceptance number from the sampling table for each stratum quantity and level, then upgrade one level for any stratum scoring two highs.
4. Fix decisions in writingWrite stratum verdicts, the shipment rule, rework permissions, and the pooling rationale on one page, and get written supplier agreement before staging.
5. Execute stratum by stratumStage goods by stratum, draw cartons at random within each, and log per-SKU findings so the next order inherits real defect data.

Frequently asked questions