What Happens When a Bag Order Fails AQL: Reject, Rework or Return?

Every buyer handles a rejection eventually, and almost nobody has agreed in advance what the word means. The inspector writes "lot rejected" against the plan; the factory writes "minor issues, can be fixed"; and the two sentences describe the same set of bags. What follows is three emails, a delay nobody budgeted for, and a relationship dented by an argument that was entirely predictable.

The way out is to treat a failure as a routing problem. The plan divides faults into classes; the purchase order should divide outcomes along the same lines, with a named owner and a number of days attached to each. When that exists, a failed report triggers a decision instead of a negotiation.

What Does "Failed AQL" Actually Mean?

It means this and only this: against a stated plan, for the sample size drawn from that lot, the count of findings in at least one class reached the reject number, or a finding in the zero-tolerance class appeared. Nothing more. It says nothing about the other units in the lot, nothing about whether the goods are usable, and nothing about blame.

Because that reading is narrow, a failure has to be translated into a commercial sentence by somebody with authority. A count of majors above the reject number usually means the defect rate across the whole lot is probably too high to ship and the lot must be touched again. A single critical finding means the lot cannot ship regardless of how good the rest of it is, until the specific hazard is dealt with. Those are different problems with different remedies, which is why one word "reject" is not enough.

Who Decides What Happens Next?

The party that owns the goods at the moment of inspection, which in practice is set by the shipping term and by who booked the visit. Under a FOB Xiamen arrangement with a buyer-booked pre-shipment inspection, the buyer decides, because risk has not yet transferred and no payment balance is due until the goods are released. Where the mill booked the visit as part of its own release process, the decision is still the buyer's to confirm before the balance is paid.

The practical test is simple: whoever is asked to release the payment decides the outcome, and should say so in writing inside twenty-four hours of the report. Silence is itself a decision, and it is the expensive kind, because production capacity carries on being allocated elsewhere while everybody waits.

What Can Be Reworked on a Bag Order?

Rework means the unit is made correct without remaking it. On sewn goods that covers a surprising amount: a strap end can be reopened and re-stitched with a bartack where a bartack was missed; a lining seam opened at the turning hole can be closed properly; a care label in the wrong position can be moved if the panel allows it; a logo that has lifted at one edge can be re-pressed if the adhesive system tolerates a second cycle.

What decides whether rework is available is access and evidence. Access: can the operator reach the faulty element without opening a seam that then has to be re-closed to standard? Evidence: does the reworked area pass the same check again, and does it leave a mark the customer will notice? A re-wheeled zipper slider leaves nothing behind. A re-pressed heat-transfer print can leave a visible halo around the edge, which turns a rework into a new fault.

What Cannot Be Reworked?

Three categories. Anything that changes the material itself: a shell colour outside the shade band, a coating that has blocked or transferred, a fabric weight below specification, an insulation laminate that has delaminated. Anything where opening the unit destroys more than fixing it gains: an internal frame, a welded liner joint inside a cooler construction, a bonded seam that cannot be re-welded once cut. And anything in the zero-tolerance class, where the unit must be accounted for rather than corrected — needle fragments, sharp burrs on metal, a restricted substance above the limit.

Those last items deserve a sentence more. A metal-detection pass does not rescue a lot with a needle fragment in one unit; it tells you one unit was found. What matters next is the needle-control system: how broken needles are accounted for, whether every fragment is taped to a card, and whether the missing fragment has been reconciled. That is a question about records, not about the bag.

What Are the Five Outcomes, Plainly?

Strip away the wording that vendors use and there are five things that can happen to a lot with findings. Which one applies is mostly determined by the class of the finding and by whether the fault is reachable. Writing the five outcomes into the order, each with its own condition, removes most of the argument that follows a rejection.

When Is Sorting Enough?

Sorting — sometimes called 100% inspection of the lot for one specific fault — is the right answer when the defect is visible, confined to a known population, and not a systemic property of the run. Examples: a print that only failed on one machine head, so only units from that hour are suspect; a hardware batch with a plating fault that can be identified by eye; a colour lot that was cut on one specific roll.

Sorting works because it converts a statistical problem into a counting one. It fails when the fault cannot be reliably identified, when it only shows under load or after use, or when sorting requires judgement that two inspectors will make differently. If the sort instruction cannot be written in one sentence a temporary worker would understand, it is not a sort plan, it is a wish.

Two conditions make sorting honest. First, sort against the same written definition and the same reference samples used in the original visit. Second, re-inspect the sorted lot under the original plan before release — a sort plus no reinspection is just an opinion about your own work.

When Should You Remake Instead?

Remake when rework would leave visible evidence, when the underlying material is wrong, or when the numbers say so. That last one is arithmetic: if eighteen per cent of the units in a lot must be opened and re-closed, the rework touches more labour than cutting and sewing new units, takes longer, and lands with a higher risk of introducing secondary faults. At some proportion the answer stops being repair.

There is also a lead-time argument that points the same way. Bulk production runs 35-50 days from written sample approval, and a partial remake inside an existing slot is often faster than a full one — but only if materials are on hand. Fabric, lining, webbing and hardware are ordered against the original quantity with a cutting allowance, typically a few per cent. If the remake exceeds that allowance, new material has to be ordered and dyed, and the calendar restarts. Ask for the allowance figure before choosing this path.

What Happens to the Shipping Date?

Every path costs calendar days, and the decision that spends least money is rarely the one that protects the date. Sorting costs one to three days for a small lot and up to a week for a large one. Rework costs whatever it takes to get the line back plus re-operating time — commonly three to ten working days. A partial remake costs close to the original production window for the affected quantity. A return with full remake costs the whole window again.

Two things decide how much of that delay is real. The first is booking position: whether the original production slot can be held while the question is resolved, which is a commercial decision the supplier makes and worth securing early. The second is the booking window itself — the container or vessel slot. A four-day rework that misses the weekly cut-off costs the same as a nine-day one, because both land on the same vessel.

For that reason, the sensible order of operations is: establish defect counts per class, establish how many units each path touches, work out whether the recovery finishes before or after the next booking cut-off, and only then choose. Where a recovery lands after a fixed sailing slot, paying for air freight on a partial quantity to protect a launch can be the rational answer, provided the cost bearer is named in advance.

Who Pays for Rework and Reinspection?

Default position under a T/T 30/70 arrangement: the factory bears the cost of correcting its own work, and the buyer bears the cost of reinspection, because reinspection exists for the buyer's confidence. That split is reasonable and worth writing down, because it is predictable and neither side has to argue about it later.

Where the fault came from the buyer's own specification or a late change, the split reverses. A colour chosen against an unreachable reference, a logo position moved after approval, or a tolerance tightened after the first article was signed will produce findings that are genuinely not the supplier's doing. This is why the approved sample, countersigned and dated, matters more than any clause: it fixes what the work was supposed to be.

Two costs are routinely forgotten. Freight for a returned or partially returned shipment, which nobody budgeted, and storage for goods waiting on a decision, which accrues daily and nobody budgeted either. Name both in the clause even if the answer is "as incurred by the responsible party", so the principle is at least agreed.

What Should the Failure Rule Say in the Purchase Order?

Short, specific, and written before anything has gone wrong. It should state the standard and plan; define a critical finding with a zero acceptance number; set the accept and reject pair for majors and minors; and then name the outcome per class. Something along the following lines carries most programmes: on a rejection, the supplier shall within two working days submit a written cause statement and a corrective proposal; the goods remain unshipped until the proposal is agreed and carried out; the lot is then re-presented under the original plan; the cost follows responsibility.

Add the two mechanical details that make it work. Timing — how many hours the supplier has to respond, and who has authority to accept the proposal. And identification — every finding referenced to a numbered photograph and a location on the unit, so that later conversations are about photographs rather than adjectives.

What Records Should a Failed Visit Leave Behind?

More than the verdict. The file should hold the report with sample size and code letter, the full defect list with class and count per category, photographs referenced to each finding, the random-number or location method used to draw cartons, the measured values rather than pass marks alone, the cause statement, the agreed corrective proposal and its date, and the reinspection report with its own number. Six months later that file is what settles a recurrence.

It also earns its keep as a trend record. Count majors per hundred units across lots and the direction becomes visible: whether faults concentrate in one operation, one colour, one material batch, or one week of the schedule. A programme with four lots of history can predict its own failures; one that files reports and nothing else cannot.

How Does One Failure Change Later Lots?

Through the switching rules, if the plan says they apply. ISO 2859-1 is built as a continuing series: two rejections among five or fewer consecutive lots move the programme to tightened inspection, where the same sample size faces a stricter accept pair, and five consecutive acceptations return it to normal. Buyers often write AQL 2.5 into an order and never reference the switching rules, which leaves a useful mechanism unused.

The second effect is on the checklist. Whatever the fault was, it now has a line, a method and a limit, and it sits at the top of the list for the following three lots. That is how checklists get better — each failure buys a permanent improvement in detection, provided somebody updates the document the same week rather than promising to.

And the third effect is on the plan for the next order. If a construction failed twice at the same seam, the answer is not a tighter level; it is a reinforcement in the tech pack. Sampling detects. Only specification prevents.

Frequently Asked Questions

People Also Ask