AQL 2.5 explained, without the mystique

What does AQL 2.5 actually allow a supplier to ship?

AQL 2.5 is a sampling plan, not a promise of zero defects. According to the Wikipedia entry Acceptable quality limit, it defines how many defective units in a random sample trigger rejection of a lot — it does not inspect every unit and it does not guarantee that every piece conforms. Ask which characteristics are critical, who draws the sample, and what happens to the batch when the limit is exceeded; those answers matter more than the number. The four gates and their rejection rules are in the four-stage QC method.

What does AQL 2.5 actually measure, and what does it not?

AQL stands for acceptable quality limit — the worst tolerable process average, expressed as a percentage or a rate, at which a lot-by-lot sampling plan still accepts lots most of the time. According to the Wikipedia entry Acceptance sampling, this is the branch of statistical quality control that decides lot disposition from a sample rather than a full count. It is a property of the *sampling scheme*, not a promise about your shipment. An AQL of 2.5 does not mean a shipment contains 2.5% defects; it means the scheme is designed to accept most lots coming from a process running at that quality level.

That distinction is not pedantry. It is the difference between buying from a process and buying from a lottery. [Montgomery, 2020] frames acceptance sampling in exactly these terms: it manages the risk of accepting a bad lot and rejecting a good one, at a cost, and it is always inferior to controlling the process.

Which four numbers make an inspection plan, and which one gets forgotten?

A sampling plan has four moving parts, and only one is usually discussed. Ask about all four.

Inspection level

How large a sample is drawn relative to the lot. General level II is the common default; level III draws more, level I fewer. A supplier quoting AQL 2.5 without naming the level has not specified the plan.

Sampling scheme

Single, double or multiple. Single draws once and decides; double allows a second draw before rejecting, which is gentler on borderline lots.

Switching rules

Whether lots move to tightened inspection after failures and back to normal after a clean run. Without switching rules a plan is static, and static plans are easier to game.

Defect classification

Critical, major and minor defects carry different AQLs. A single number covering all three means a scratched box is being treated like a water-resistance failure.

Why can sampling not replace a 100% function test on this product?

Here is the practical limit. If you draw fifty units and find three defects, you know something about the process. You know nothing about unit fifty-one. For most product categories that residual risk is acceptable and priced in. For a body-contact device with a motor, a battery and a seal, it is not — which is why the inspection gates used on our range apply AQL sampling *on top of* a 100% final function inspection rather than instead of one.

The measurement we publish is the outgoing defect rate held below 0.1%, achieved because every finished unit is function tested — including charging, motor noise and the water-resistance rating applicable to that model. Details are on the what a batch is actually tested for, and the pre-shipment documents that evidence it are listed in the documents that ship with the order.

How do you write the inspection plan into your own purchase contract?

• Fix the plan in the contract: standard, level, scheme, and the classification with its own AQL per class.
• Classify defects before inspection begins, not after you disagree about one.
• Keep the failed units. Physical evidence settles a dispute faster than photographs.
• Ask whether critical defects — seal failure, battery safety — have an accept number of zero. They should.

What does first article inspection catch before the lot is run?

AQL sampling decides whether a finished lot ships. It says nothing about whether the process that made the lot is the one you approved. First article inspection (FAI) closes that gap: before a new tool, a changed tool, or a moved production line runs a full order, the supplier builds a small number of units and proves them against the drawing, the material specification and the control plan. PPAP — the Production Part Approval Process, an AIAG standard used across manufacturing — is the packaging of that evidence into one submission a buyer can sign off.

First article inspection

A one-off verification of the initial units against the design record: dimensions, material, performance and appearance, each checked rather than sampled.

PPAP submission

The dossier that accompanies the first article — design record, process FMEA, control plan, measurement studies and the inspection report — assembled so the buyer approves the process, not just the part.

Change-point control

The rule that a new tool, a new material batch or a moved line triggers a fresh first article rather than a waived one. Most quality escapes happen at a change nobody logged.

First article inspection and PPAP are not alternatives to AQL lot sampling; they are the gate in front of it. A lot that was never first-articled can pass AQL and still be the wrong product. For the documents that travel with an approved order, see the certification pack.

How do you classify critical, major and minor defects before you inspect?

The classification decides the plan, and the plan decides the outcome, so it has to be agreed while nobody has an interest in the answer. In practice three classes do the work, and for a powered body-contact product the boundaries are not a matter of taste.

The failure mode this prevents is the one where a supplier applies a single AQL to everything and a lot passes with a leaking unit in it, because a leaking unit was counted as one defect among several allowed. A plan with no critical class does not have a lenient standard; it has no standard for the only failures that matter. This is the same logic that makes a 100% function test necessary alongside sampling — see why sampling cannot replace a full test for how the two sit together on our own line.

When a lot is rejected, what happens next and who pays for it?

The standard answers the statistical question and stops there. It says a rejected lot is segregated and may be resubmitted after rectification, and that resubmission is inspected under tightened severity. What it does not say is who performs the screening, how long it takes, and what happens if the second inspection also fails — and those three questions are where the cost of a rejection actually sits.

1. Segregate and count. The lot is quarantined and the failing units are identified, not the whole lot. A factory that cannot say how many units failed has not inspected, it has rejected.
2. Screen or rework. The usual remedy is 100% screening of the lot against the failed characteristic, removing or repairing the defects found. This is the expensive part: 3,000 units screened by hand is days of labour, and it is charged to somebody.
3. Re-inspect under tightened severity. The resubmitted lot moves to tightened inspection, which draws a larger sample and accepts fewer defects. Taking a fresh sample at normal severity resets the risk you just paid to find.
4. Decide the commercial consequence. Who carries the screening cost, whether the delivery date moves, and whether a second rejection cancels the balance. These are contract terms, and their absence is why rejections escalate into disputes.

References

1. Montgomery, D. C. (2020). Introduction to Statistical Quality Control, 8th ed. Hoboken, NJ: Wiley.
2. International Organization for Standardization (1999). ISO 2859-1: Sampling procedures for inspection by attributes — Part 1. Geneva: ISO.
3. Shewhart, W. A. (1931). Economic Control of Quality of Manufactured Product. New York: D. Van Nostrand Company.
4. Juran, J. M. & Godfrey, A. B. (1999). Juran's Quality Handbook, 5th ed. New York: McGraw-Hill.