Your Supplier's QC Report Says AQL 2.5: 6 Numbers to Check Before You Trust It

Every factory QC record says roughly the same thing: inspected, passed, ready to ship. It arrives as a tidy PDF with a table, a stamp and a signature — and most importers file it and release the balance. The problem is that a record written by the factory is a self-report, not evidence. Six numbers on that document tell you whether a real sampling plan produced it, or whether it was typed up the night before your container arrived. 🔍

A factory QC record is a claim, not evidence

In-house QC is a good sign. It means the factory is checking its own output and stopping some defects at the line. But the person writing that report is paid by the factory, sits next to the production manager, and has no reason to find a problem the day before a shipment is booked. Nobody audits their own homework, and that is a structural issue, not an accusation.

A supplier record tells you what the factory believes about its own work. A third-party inspection tells you what was measured, on which units, at what time, with photos you can look at — and only the second gives you leverage over the balance. 🧾

The 6 numbers to check on a supplier QC record

Numbers 1 and 2: sample size and the Ac/Re pair

The sampling plan is the backbone. ISO 2859-1 general inspection level II gives code letter L — a sample of 200 units — for any lot between 3,201 and 10,000 pieces. A 500-unit lot takes 50; a 20,000-unit lot takes 315. When the report shows the same sample number across wildly different lot sizes, that number was copied from a template, and the "inspection" behind it is unverifiable.

At AQL 2.5 with 200 units the acceptance pair is Ac 10 / Re 11: the batch passes with up to 10 defectives found, and fails from 11. Now the part that surprises importers — that is an acceptance threshold, not a quality guarantee. A lot truly carrying 4% defective units passes an AQL 2.5 check on 200 units roughly 8 times out of 10. At an 8% hidden defect level it still slips through about once in twelve. ⚠️

Numbers 3 and 4: defect classes and the escape math

A single blended defect rate hides the difference between a loose thread and an electrical fault. Real records split critical, major and minor, and critical defects are usually handled with a zero-tolerance rule no matter what the AQL says. If the report has one number, you cannot tell what was found, only what the writer wanted you to conclude.

Then do the escape math on whatever the record admits. Ten majors found in a 200-unit sample is a 5% point estimate. On a 5,400-unit order that is around 270 defective units, and once those are inside FBA the cost moves well beyond the unit price — refunds, return handling, removal orders, a dented star rating. Multiply 270 by a realistic landed unit cost of $9 and you are looking at roughly $2,400 of direct exposure before any review damage. That is the number to weigh against inspection time at $169 per man-day.

Numbers 5 and 6: photo timestamps and carton reconciliation

Where the samples came from matters more than how many. Units pulled from a tray beside the line are the best the factory has at that moment; units pulled from sealed finished cartons spread across pallets are what your customer will receive. A dated photo set with carton labels legible in frame lets you verify both the spread and the batch codes.

Reconciliation is pure arithmetic: shipped cartons multiplied by units per carton should equal the PO quantity, and any shortfall explained. A record that never does this arithmetic never verified a count either. 📦

What a "passed" record looked like when the container was already booked

A Guangdong factory shipped 5,400 decorative ceramic vases worth $38,300 with an internal record reading "AQL 2.5, passed, 3 minor defects". The inspector we sent pulled 200 units from sealed cartons across four pallet positions. Result: 17 majors — glaze pinholes on the visible face, painted band off by ΔE 2.4 against the approved sample, five units with a rock in the base — plus 5 criticals where wobble exceeded 2 mm on a flat surface.

Ac 10 / Re 11 exceeded, batch rejected. The factory's own sample had come from the first two pallets, at eye level, from the newest cartons. Rework of 480 units cost the factory about $200 in labour and five days; the re-inspection came back at 1.4% majors. Had the batch sailed at a conservative 5% defect level, roughly 270 units would have reached customers as refunds, returns and removal orders — call it $9,600 and a run of one-star reviews, against an inspection that costs a fraction of one man-day.

How to use the record instead of arguing about it

Three habits close most of the gap. Put the raw numbers — lot size, sample size, Ac/Re, class counts — into the purchase order so the standard is agreed before production. Tie final payment to a third-party report rather than the factory's own document, with rework responsibility named in writing. And insist that sampling happens on finished sealed cartons, not on a tray beside the line.

CloudSpects has completed 2,000+ inspections across consumer goods, electronics, furniture and building materials. Reports are typically delivered within about 24 hours of the visit, with dated photos and defect close-ups, from $169 per man-day. Run your sample size first with our AQL calculator, or send us the lot size and we will quote the visit. ✅

Frequently asked questions

Is a factory's own QC report worthless?

No. It is useful as a production signal and it usually shows the factory is at least looking at its own output. It is worthless as payment evidence, because the same company that writes it also gets paid when the order ships. Treat an internal record as a claim to be reconciled, and a third-party report as evidence you can act on.

What sample size should a 5,000 unit lot show on a QC record?

Under ISO 2859-1 general inspection level II, a lot between 3,201 and 10,000 units takes code letter L, which means a sample of 200 units. At AQL 2.5 that sample is accepted with up to 10 defectives found (Ac 10) and rejected from 11 (Re 11). A record that shows 200 samples for every lot size, regardless of quantity, was not calculated from a sampling plan.

Can a bad batch still pass AQL 2.5?

Yes, and this is the part most importers miss. Acceptance means the batch is no worse than the tolerance of the plan, not that it is good. A lot genuinely carrying 4% defective units still passes an AQL 2.5 check on 200 units roughly eight times out of ten. Tighten the AQL, increase the sample, or move the defect class if that risk is not acceptable for your product.

What should be in the photo set that comes with a QC record?

Dated photos that show the units inside sealed finished cartons, carton and shipping marks legible in frame, defects photographed individually with a scale reference, and the sampling spread across pallets or batch codes. A handful of glossy photos with no timestamps, no carton labels and no defect close-ups tells you nothing about the batch you are paying for.

How do I tie the balance payment to a real inspection?

Write it into the purchase order: final payment released against a third-party inspection report, not against the factory's own record. Name the sampling standard and level, the defect classes, and who pays for rework and re-inspection. CloudSpects reports are typically delivered within about 24 hours of the factory visit, so the payment clock is not usually the bottleneck.