★ AQL sampling · factory audit · scorecard

OEM quality control

What a sample can, and cannot, prove.

The quality-control detail behind the OEM overview: what a sample proves, AQL sampling under ISO 2859-1, telling a plant from a trading company, the supplier scorecard, and the audit badges that decide who you will talk to.

We don't sell the products we review. Ever. · Are you a brand? Put your device on our bench →

This page is the quality-control detail behind the OEM and private-label overview: what a sample can and cannot prove, AQL sampling under ISO 2859-1, telling a plant from a trading company, the supplier scorecard, and the audit badges and approved-supplier list that sit underneath them.

Sampling: what a sample proves, and what it cannot

A stock sample ships in three to five business days and tells you three useful things: whether the format suits the use it is for, how the motor feels at the setting a customer will actually use, and whether the noise level is tolerable in a room with thin walls. It cannot tell you whether the batch you receive will match it, because a stock sample comes off a line that has been running.

A custom sample is a different instrument and it is worth being precise about what to demand. The sample must be made on production tooling, not hand-finished from a prototype mould, and it must be accompanied by the first-article measurements for the dimensions that matter to your packaging.

A hand-finished sample is the most common way a first order disappoints: the sample was beautiful because a technician spent two days on it, and the production unit is what a 40-second cycle produces. Ask for a production sample, approve the first article in writing, and keep the approved unit — it is the reference for any dispute later. Two documents make that reference enforceable, and both have to exist before the first production run.

The first is the tech pack: the drawing, the material call-outs, the tolerances, the surface finish and the packing configuration the plant is actually building to, in the plant’s own language rather than yours — and these are the same numbers the retailer draws the planogram from, months later, whether or not you sent them. The second is a change control rule — any deviation from the tech pack, however small and however obviously an improvement, is quoted and approved in writing before it is built. A tech pack nobody signed is a memory, and every “small improvement” that skips the rule is a specification change you will first notice in the third container.

Sample charges are stated separately on the quotation, and a stock sample is credited against the bulk invoice for the same SKU. An open-ended sample charge with no credit is not a red flag by itself, but a supplier who will not separate it from the unit price usually will not separate anything else either.

Quality control: AQL, inspection, and telling a plant from a trading company

The line on every catalogue SKU is the same: incoming inspection on components, in-process checks during assembly, final inspection on every unit, and outgoing inspection before packing, with a stated outgoing defect rate under 0.1% and a twelve-month warranty on business orders. On top of the per-unit line, sampling inspection follows ISO 2859-1, normally at critical 0, major 2.5 and minor 4.0 for consumer goods.

AQL zero on critical means a single safety defect fails the batch, which is the only sensible reading of the standard for a device that goes near the body. The sampling level that sets the sample size is a second decision, separate from the AQL number, and it is the one buyers forget to put in writing: general inspection level II is the working default in ISO 2859-1, level III widens the sample for a first run, and the special levels S-1 to S-4 exist for tests that destroy the unit.

Agree the defect classification sheet before production starts — whether a mark is a critical defect, a major or a minor is your call to make in writing, not the inspector's to make at the carton — because the identical sample passes or fails depending on the class the defect is put in.

An AQL is not a ceiling on the defect rate: it is the quality level a sampling plan is designed to accept most of the time.

A 10,000-unit lot: 200 units sampled, and the acceptance numbers

Critical (AQL 0) 0 defects
Major (AQL 2.5) 10 defects
Minor (AQL 4.0) 14 defects

Ten major defects in a 200-unit sample is an accept: 5% observed passes an AQL of 2.5.

A 10,000-unit lot: 200 units sampled, and the acceptance numbers. The full chain, for one lot size: 10,000 units sits in the 3,201–10,000 band, which is code letter L at general inspection level II, which is 200 units under ISO 2859-1 single sampling on normal inspection. Read the AQL columns at that sample size and the acceptance numbers are 0 for critical, 10 for major and 14 for minor. Ten defective units in a 200-unit sample is 5% — and it is still an accept at an AQL of 2.5, which is the mistake almost everyone makes with these numbers. Critical is run at zero acceptance, and it is the only one of the three lines that works the way people assume all three do.

Book a third-party pre-shipment inspection, at your cost, before the goods leave. It is the cheapest insurance in the order and the clearest single signal of what kind of counterparty you have: a plant welcomes it because the report protects them too, and a trading company resists it because the report describes a factory they do not control. What happens once the container is on the water — the MABD date, the OTIF score and the deductions that follow a documentation slip — is set out on retail readiness, and what a repeat order costs in cover and cash is on the second order, and what a delisting costs once the stock is no longer on a shelf is on the delisting arithmetic. Six signals are worth gathering before a first deposit, and they are set out as a figure on our for-brands page — a business licence whose scope matches the product, a live unscripted video walk of the floor, a sample made on production tooling, a batch-level test report, an invoice beneficiary that matches the licence, and an audit above roughly $10,000.

How many checkpoints you buy matters more than which firm you buy them from, because three of them answer three different questions and buyers usually purchase only one. A factory audit before the deposit answers who you are paying and whether this site can make the thing: the entity matching the licence, the equipment and the capacity, whether work is subcontracted, and the last three defect records rather than a quality manual.

A pre-shipment inspection on finished, packed goods answers whether the batch matches the signed golden sample, and it is scheduled at 80–100% completion so rework is still possible. Loading supervision answers what actually crossed the container door, and under which seal. Skipping the first and buying the other two is the standard way to inspect goods made by the wrong supplier. A during-production check at 20% to 50% completion — DUPRO — is the cheap addition that catches a material substitution before it is repeated across the whole run.

Scoring a supplier: seven dimensions, and the weight each one earns

The audit above is a decision; a scorecard is the record you keep. Give every dimension a weight that adds to 100, score each from 0 to 5, and the composite is the sum of score divided by five, times the weight. None of that arithmetic is proprietary. What makes it useful is that the weights are fixed before the quotes arrive, so a persuasive salesperson cannot move them afterwards.

Supplier scorecard: seven weighted dimensions, and what each score means
#DimensionWhat earns a 5WeightWhat a 1 is telling you
1Certificate pack, per destination marketThe complete set is in hand and named on the documentation, including the test report for the exact variant20The pack depends on the order being placed first — which means it will never be shown to you
2Final-inspection defect rateA published rate under 0.1% that a third-party report confirms18No number exists, so the first batch is the measurement
3Delivered against quoted lead timeShips inside the quoted window on the second and third order, not only the first15The quote was a sales figure; the real lead time appears after the deposit
4Independent pre-shipment inspectionWelcomes it, books the slot, and has done it before12Resists it, or offers to handle the inspection internally
5Tooling ownership and transferOwnership, amortisation and transfer price are all stated in the contract12The mould is the factory’s, and leaving means starting again
6Commercial termsInvoice beneficiary matches the licence, deposit at or below 30%, Net-30 offered from the second order13Personal account, 100% up front, or a beneficiary in a third name
7Response and document turnaroundAnswers a technical question in writing within two working days, with the document attached10Fast on price, silent on specifications

Two quotes from the same round, scored the same way, usually separate further than the price difference suggests. Supplier A answers within the hour and quotes 8% less. Supplier B takes two days and sends a certificate pack nobody asked for yet. On the seven dimensions A lands at 60.6 and B at 84.2 — and the price advantage A is selling disappears the first time a container has to be reworked.

The same two quotes scored: the cheaper one loses on the dimensions that cost money later
DimensionWeightA: cheaper, fasterB: slower, documentedA × weightB × weight
Certificate pack per market203512.020.0
Final-inspection defect rate18247.214.4
Delivered against quoted lead time15349.012.0
Pre-shipment inspection accepted12254.812.0
Tooling ownership and transfer12357.212.0
Commercial terms134310.47.8
Response and document turnaround105310.06.0
Composite, of 10010060.684.2

Three rules keep the scorecard alive past the first order. Score per purchase order rather than once a year, because the second order is where a first order’s promises are actually tested. Hold the weights fixed for at least a year, or the ranking turns into a way of justifying a decision already made. And record the score you actually gave, not the one you would defend in the meeting — a scorecard that flatters every supplier measures nothing. The RFQ checklist that starts the process is on this page: target market, volume band across the first three orders, what you want changed, target landed cost, packaging status, and the documents you already hold.

Two documents decide whether a shortlist survives contact, and neither is the one suppliers lead with. The first is the ISO 9001 scope statement — not the certificate badge, but the line naming which processes at which site were assessed, because a scope that reads “trading of consumer goods” describes a trading company holding a quality certificate.

The second is the social-compliance audit: BSCI (amfori’s Business Social Compliance Initiative) and SEDEX, reported as SMETA, cover labour standards, health and safety, environment and business ethics at one named address, and they are re-audited rather than granted once. Ask for both as PDFs carrying the site address, and read the date. Neither substitutes for a batch-level product test report, which is the document a border or a marketplace actually asks for.

The result belongs in an approved supplier list — the register of vendors audited, qualified and approved for a specific product, with the approval conditional on audit findings and corrective-action deadlines. Without one, approval is a conversation; with one, it is a record a procurement note can cite.

Re-score it per purchase order rather than once a year, because the second order is where a first order’s promises are tested, and a list that is never re-scored is a decoration. Where the product is a first run, book the inspection before shipment and nominate the AQL inspection level in the purchase order rather than leaving it to the factory’s standard — the sample size, and therefore what the inspection can actually detect, is set by that level.

Quality questions before a first deposit

What a sample proves, AQL sampling under ISO 2859-1, telling a plant from a trading company, the seven-dimension supplier scorecard, and the audit badges that decide who you will talk to.

How do I tell a real factory from a trading company before the first deposit?

Three checks, and all of them are cheap. Ask for a live video walkthrough of the line that will actually run your order rather than a showroom tour; ask for the business licence and compare the registered address with the one printed on the quotation; and ask for a third-party factory audit report, or commission one if the order justifies the fee.

A pilot batch is the cheapest way to test the answer before you commit to the run. Factory acceptance of that request without scheduling already tells you what you need to know: a trading company cannot show you a line it does not own, and its customs data will show the same supplier buying in what it claims to make.

What AQL do you inspect to, and can I book a third-party inspection?

The line runs IQC, in-process IPQC, final FQC and outgoing OQC, with 100% final unit inspection and a stated outgoing defect rate under 0.1%. On top of that, sampling inspection follows ISO 2859-1, normally at critical 0, major 2.5 and minor 4.0 for consumer goods. Book a third-party inspection at your own cost before the goods leave, and tell us the date — a factory that resists an inspection is telling you which of the two you are dealing with. It is the cheapest insurance in the whole order and the single clearest signal that distinguishes a plant from a trading company.

How do I score and compare two factories that both look acceptable?

Use a scorecard with fixed weights, set before the quotes arrive. The seven dimensions that decide a first order are the certificate pack per destination market (20), the final-inspection defect rate (18), delivered against quoted lead time on the second and third order (15), whether independent pre-shipment inspection is welcomed (12), tooling ownership and transfer terms (12), commercial terms including the invoice beneficiary and Net-30 eligibility (13), and document turnaround (10).

Score each from 0 to 5 and multiply score divided by five by the weight. Two suppliers that look equal on price routinely separate by twenty points here, and the gap is in the columns that cost money after the deposit. The weighted table is set out in full under scoring a supplier on the quality page.

What do SEDEX, BSCI and ISO 9001 audit badges actually tell a buyer?

They answer three different risks, so read them as a set and not as one gold star. ISO 9001 is a quality-management system standard — described in the Wikipedia entry of that name — and tells you the factory runs documented, repeatable process control rather than heroics; it is about consistency, not about the product being good. BSCI and SEDEX (the latter built around the SMETA audit methodology) are social-compliance audits: they look at labour, working conditions and subcontracting, so a clean report is a signal that your brand will not be the one photographed in a bad supplier story.

None of the three is a product-safety certificate, and a factory can hold all three and still ship you a unit that fails your own spec — so treat them as the floor for who you will talk to, and let a third-party factory audit and a pre-shipment inspection carry the rest. A buyer who asks for the audit report reference number, not just the badge, is the buyer suppliers take seriously.

Does the sample fee come off the bulk order?

A stock sample is charged at the sample price and credited against the bulk invoice when the bulk order is placed against that SKU; a custom sample is charged at cost, because the tooling and the hand work exist only for that sample. Both are stated on the quotation as a separate line. The thing to refuse is a sample that is hand-finished and presented as representative of production — a sample made off production tooling is the only one that tells you anything about the units you will receive.

What is a should-cost model, and why build one before comparing quotes?

A should-cost model is your own arithmetic for what the unit ought to cost — material mass, motor grade, battery cells, assembly minutes, packaging, plus the tooling amortisation you are being asked to pay — built before you open the quotes. It turns a price comparison into a cost comparison: with a should-cost in hand you can see which quote is cheap because the material is thinner, and you can argue about the total cost of ownership rather than the unit price. Add the tariff code and the landed duty at your border, and two quotes that looked eight per cent apart are often the same number.

Which parts of the device should the factory make itself?

Ask, and expect a specific answer: an intimate wellness device is an assembly — motor, battery, control board, silicone body and housing — and a plant that moulds and assembles the silicone but buys the electronics in is telling you exactly where your quality risk sits. The concept is vertical integration, described in the Wikipedia entry of that name: the more of the critical path a supplier owns, the fewer parties can change your product without telling you.

It is not a reason to reject a factory that subcontracts the board — most do — but it is the reason to ask who signs off on a substitute component when the specified one is out of stock, and to get that answer in writing before the deposit rather than after the first shortage.

Why a sample is not a batch

Back to the OEM overview →

A sample proves a process can make the part; it does not prove the batch will match it, which is why statistical quality control exists as a discipline rather than a hope [Montgomery, 2020]. The habit of presenting any result as more certain than it is the failure mode the quality literature names and guards against [Heizer, 2023], and Juran's framing — quality as fitness for use, planned rather than inspected in — is the reason the defect classification sheet is agreed before production rather than after [Juran, 2010].

Sources cited

  • 《Acceptance sampling》 — Wikipedia. https://en.wikipedia.org/wiki/Acceptance_sampling
  • Montgomery, Douglas C. (2020). Introduction to Statistical Quality Control (8th ed.). Wiley, Hoboken, NJ.
  • Heizer, Jay; Render, Barry; Munson, Chuck (2023). Operations Management: Sustainability and Supply Chain Management (13th ed.). Pearson, Harlow.
  • Juran, Joseph M.; De Feo, Joseph A. (2010). Juran's Quality Essentials for Leaders. McGraw-Hill, New York.