Reading the Lot, Not the Part: Acceptance Sampling for Chinese CNC Deliveries

A UK buyer reviewing a Chinese CNC delivery faces a familiar choice: inspect everything and absorb the cost, or accept a supplier's assurance that the batch is good. Neither position is a sampling plan. Acceptance sampling is a deliberate, documented way of taking evidence from a lot and deciding whether to accept or reject it. ISO 2859-1 gives the mechanics, but the commercial judgement belongs to the buyer. The sample is not the batch; it is an inference about the batch, and the purchase order should say so plainly.

Why a Good Sample Proves Very Little on Its Own

A supplier can produce a perfect first article and still ship a batch with variation in a critical bore, an inconsistent thread class, or a cosmetic finish that drifts across the run. CNC machining is a process, not a single event. Tool wear, fixture wear, material batch differences, operator changeovers and deburring practice all affect the lot after the sample has been approved.

Acceptance sampling therefore belongs at the point of lot acceptance, not at the point of quotation. It should be agreed before the purchase order is placed and written into the order as a measurable requirement. A phrase such as all parts must be free from defects is not a sampling plan. It is a wish, and it usually produces either a dispute or a supplier who learns to present inspection records rather than improve the process.

Classify Defects Before You Count Them

ISO 2859-1 works with attributes: a part is conforming or nonconforming against a defined characteristic. That means the buyer must define defect classes before the sample size has any meaning. For machined parts, a practical three-class structure is usually enough.

• Critical defects affect safety, regulatory compliance or irreversible function. Examples include wrong material grade, missing heat treatment, cracks, or a feature that could cause assembly failure under load. Many UK buyers set critical AQL at zero and require either 100 per cent verification or process evidence for the characteristic.
• Major defects affect fit, form or function but do not create an immediate safety risk. A bore outside tolerance, an incorrect thread class, a missing operation, a wrong chamfer that prevents assembly, or a surface finish outside the drawing are typical major defects.
• Minor defects affect appearance or housekeeping without compromising function. Light tool marks in a non-cosmetic area, a small burr that can be removed during assembly, or a handling mark on a hidden face may be minor, provided the drawing and finish schedule support that judgement.

The classification must be written down. If the supplier and the buyer disagree about whether a mark is major or minor, the AQL table cannot resolve the argument. It will simply formalise it.

Choosing Inspection Levels and AQLs Without Defaulting

ISO 2859-1 provides general inspection levels and special levels. General level II is a common starting point for routine machined lots. Special levels S-1 to S-4 suit expensive or destructive tests, or characteristics where a smaller sample is commercially necessary. Reduced inspection can be earned through a stable history, and tightened inspection should follow a run of rejections, but both need to be stated in the contract rather than left to the supplier's discretion.

AQL values are also commonly misread. An AQL of 1.0 does not mean the buyer will accept one per cent defective parts. It is a reference point used with the sample size and acceptance number in the standard. The buyer should not choose an AQL from habit. Critical characteristics often justify zero acceptance or 100 per cent inspection. Major characteristics might sit at 0.65 or 1.0 for functional parts. Minor characteristics might sit at 1.5 or 2.5. These are examples, not a universal schedule. The correct values depend on the part, the consequence of failure, and the process capability the supplier has demonstrated.

For CNC parts, variable data such as CMM reports and calliper readings are useful, but they do not replace an attribute sampling plan unless the contract says how that data will be used. A first article report is not a lot acceptance record. It is evidence about one part at one moment.

Writing the Plan into the Purchase Order

The sampling plan should appear as a short annex to the purchase order. It should name ISO 2859-1, the inspection level, the AQL for each defect class, the lot size and how the lot is formed, the sampling point, and the records the supplier must retain. It should also state who inspects, what equipment is acceptable, and what happens if the lot is rejected.

A workable annex will usually cover:

• the definition of a lot, so that the supplier cannot combine several production runs into one presentation;
• the inspection stage, for example before packing and before shipment;
• the requirement for critical characteristics to be verified by 100 per cent inspection or by agreed process controls;
• the acceptance and rejection numbers, or a clear reference to the ISO 2859-1 tables;
• the prohibition on resampling a failed lot without written agreement;
• the disposition of nonconforming parts, including sorting, rework, scrap and cost responsibility;
• the right to audit inspection records and to witness inspection on the first batch or on tightened inspection.

This turns AQL from a quality department slogan into a commercial term. It also gives the UK buyer a defensible position if a delivery arrives with a defect rate that the supplier later describes as normal.

Where AQL Gives False Confidence

AQL is weakest where the batch is small or the part is expensive. For low-volume lots, the sample size required by the standard may be close to the whole batch, or the statistical confidence may be so limited that acceptance means very little. For high-value batches, the cost of a single failure can dwarf the saving from a small sample. In those cases, the buyer should not hide behind an AQL. A better approach is 100 per cent inspection of critical features, a documented process audit, and a first article inspection that actually challenges the process.

AQL also says little about latent defects. It will not confirm material certification, heat treatment integrity, plating adhesion, or the absence of subsurface cracks unless those characteristics are separately controlled. Nor does it resolve subjective cosmetic disputes. A finish schedule and boundary samples are needed for that.

Acceptance sampling is a decision rule, not a quality system. For UK buyers sourcing CNC parts from China in 2026, the sensible position is to use ISO 2859-1 where it fits, classify defects honestly, write the plan into the order, and reserve 100 per cent inspection or process validation for the characteristics where a sample cannot carry the risk. The aim is not to inspect more. It is to know exactly what the sample is allowed to prove.