If you buy CNC machined parts in batches, the inspection question that decides your risk is simple: how many pieces should be measured, and what result lets you accept the lot? A CNC inspection sampling plan answers that. It is not a promise that every part is good; it is a statistically defined rule that trades inspection cost against the risk of accepting a bad lot. This guide explains lot definition, critical versus noncritical features, acceptance rule, measurement capability, and change control — the five decisions that turn “we inspect parts” into a plan you can actually audit.
Why sampling exists at all, and where it breaks down
Acceptance sampling uses statistical sampling to decide whether to accept or reject a production lot of material. It is normally applied when testing is destructive, when the cost of inspecting every unit is very high, or when inspecting every unit takes too long. Critically, checking every unit does not guarantee full compliance, and it is time-consuming and costly. That is the honest boundary of any inspection claim: even full inspection passing does not prove every feature on every part is within specification.
A sampling plan buys you a defined, repeatable decision — not certainty. Buyers who understand that distinction write better purchase orders and get fewer surprises at incoming inspection.
Step 1 — Lot definition
The lot definition is the foundation. A sampling plan has no meaning unless “the lot” is a meaningful, homogeneous population. For CNC parts, a defensible lot is usually one production run of one part number, made on one machine, from one material heat/batch, with no unrecorded process change in between. If the same part number was run across two machines, two shifts with different tooling, or interrupted by a setup change, that is normally two lots — not one.
Why this matters: sampling statistics assume the units in a lot share a common process. Mixing sub-populations can hide a defect cluster in one sub-group and make a passing sample meaningless. When you specify sampling on a PO, state the lot basis explicitly: per part number per shipment, per manufacturing run, or per date code. Ambiguity here is the most common source of incoming-inspection disputes.
Step 2 — Critical versus noncritical features
Not every dimension deserves the same inspection intensity. Split the drawing into two groups:
- Critical and key characteristics — features whose failure affects safety, function, fit to a mating part, or a regulated requirement. Typical examples: a sealing diameter, a thread class on a load-bearing joint, a bore that sets a bearing fit, a datum that drives assembly. These usually justify tighter sampling, dedicated gauging, or in some cases verification of every unit.
- Noncritical features — cosmetic surfaces, non-functional radii, marking, or dimensions with generous tolerance that cannot cause a functional failure. These normally use looser sampling, often the general sampling plan applied to the lot.
First article inspection is the mechanism that confirms the critical-feature list is actually producible before a lot is released for ongoing sampling. First article inspection is a design verification process that verifies the first production parts meet specification; in aerospace it is formalized as AS9102. Failing to lock critical characteristics first is how a sampling plan gets applied to the wrong features.
Step 3 — The acceptance rule
The acceptance rule is the arithmetic that converts sample results into a decision: sample size n, acceptance number Ac, and rejection number Re. The lot is accepted if the number of defects falls below the acceptance number; otherwise the lot is rejected.
The best-known attribute plans are ISO 2859-1 (the international AQL standard) and its US counterpart ANSI/ASQ Z1.4 (successor to the historical MIL-STD-105). These plans are indexed by lot size and by an AQL (acceptance quality limit) selected by the buyer. Two things are often misread:
- AQL is not “the maximum defect rate we will accept.” It is the quality level that, under the standard’s operating rules, corresponds to a high probability of acceptance. A plan can still accept a lot that is worse than the AQL some of the time.
- Switching rules matter. The standards describe normal, tightened, and reduced inspection and rules for switching between them based on recent lot history. A plan without switching rules is only half of the standard.
You also choose between attribute plans (count defective units, pass/fail) and variables plans (measure the continuous value and compare to specification). Variables plans can reach a decision with smaller samples when the characteristic is measurable and roughly normally distributed, but they assume a distribution shape and are more sensitive to that assumption. State which type you require, because “sample 32 pieces” means something different in each.
Step 4 — Measurement capability
A sampling decision is only as good as the measurement behind it. If the gauge cannot reliably distinguish a good part from a bad one, sampling — or even checking every unit — produces noise. Measurement system analysis (MSA) is a thorough assessment of a measurement process, typically a designed experiment that identifies the components of variation in that process. The standard reference for automotive-grade MSA is the AIAG MSA manual; ASTM and ASME also publish related procedures.
For practical buying, two numbers anchor the discussion:
- Gauge repeatability and reproducibility (GR&R) — how much variation comes from the gauge and the operators rather than the parts. A common rule of thumb is that GR&R should be a small fraction of the tolerance; where it is not, the inspection result is partly measuring the measurement system, not the part.
- Process capability (Cp/Cpk) — a statistical measure of whether a process can produce output within specification limits. Capability has meaning only when the process is in statistical control. A sampling plan applied to a process that is drifting is a description of history, not a control.
Ask what the supplier’s measurement capability is for the critical characteristics, on which instrument, and whether it was verified. Do not accept a capability figure without the instrument, the tolerance, and the study basis attached — a Cpk number with no context is not evidence.
Step 5 — Change control
Sampling is calibrated to a specific process and its documented capability. When the process changes, the plan’s assumptions — and sometimes the correct sample size — change with it. Change control is the discipline that decides when a lot is still “the same lot” statistically and when it must be re-qualified.
Trigger a re-qualification (typically a fresh first article plus a revised or reset sampling baseline) when any of the following occur:
- A new or changed machine, fixture, or work offset;
- A new material heat, supplier, or material condition;
- A tooling change that can move a critical dimension;
- A drawing or specification revision, or a tolerance change;
- A detected nonconformance, corrective action, or a shift to tightened inspection under the AQL standard;
- A long production gap that invalidates the previous control assumptions.
Good change control records what changed, when, who approved it, and what evidence re-established control. That record is what lets you trust the next shipment’s sampling result.
When full inspection may be justified
Sampling is a cost-control tool, but it is not always the right one. Full (every-unit) inspection is more defensible when: the failure mode is safety-critical and the consequence of a single escape is severe; the process capability is unknown or poor; the lot is a first article or pilot run; a customer or regulator requires verification of every unit; or the feature is cheap to gauge automatically. Even then, remember the standard limitation — checking every unit does not guarantee full compliance, so pair it with process control (statistical process control) rather than treating it as a wall against defects.
A practical rule: use sampling where the process is stable and the failure cost is bounded; use every-unit inspection plus process control where an escape is intolerable; and use neither as a substitute for qualifying the process in the first place.
Putting the plan into a purchase order
To make a sampling plan enforceable, specify: the lot basis; the list of critical characteristics and the sampling method for each; the standard and AQL (for example ISO 2859-1 or ANSI/ASQ Z1.4) with inspection level and switching rules; the measurement instrument and required capability for critical features; the acceptance rule (n, Ac, Re); the inspection-report content; and the change-control triggers that require re-qualification.
Pair it with the documentation you already expect against good practice — a CMM inspection report for geometry, a first article report for new parts, and a measurement-system study for critical features. If you are still defining which dimensions belong in the critical list, the linked guides below cover tolerance stack-up, CMM report reading, first article checklists, and surface-roughness callouts, which are the natural companions to any sampling plan.
Key takeaways
- A CNC inspection sampling plan defines sample size and an acceptance rule; it does not guarantee zero defects.
- Define the lot as one homogeneous run; mixed lots invalidate the statistics.
- Split critical versus noncritical features; severity drives inspection intensity.
- Use ISO 2859-1 / ANSI ASQ Z1.4 with stated AQL, inspection level, and switching rules; AQL is not a guaranteed defect ceiling.
- Verify measurement capability (MSA, GR&R) and process capability before trusting any accept decision.
- Re-qualify on machine, material, tooling, or specification change.
Related reading: First Article Inspection for CNC Parts: A Buyer Checklist, How to Read a CMM Inspection Report for CNC Machined Parts, CNC Machining Quality Inspection Report: Buyer Guide, What Is Your Machining Accuracy and Tolerance, CNC Tolerance Stack-Up: How Buyers Can Prevent Assembly Fit Problems, and Surface Roughness Callouts for CNC Parts (Ra, Rz, Direction and Measurement).
