Small Batch Bag QC: The Same AQL 2.5 Standard at 500 and 5,000 Pcs

What AQL 2.5 Actually Means

The most common misunderstanding we hear from creators is that AQL 2.5 means someone checks 2.5% of the order, or that it means 2.5% of the bags are allowed to be defective. Neither is right. AQL stands for acceptance quality limit, and it is a threshold expressed as a defect rate that the sampling plan is built to accept most of the time. The number 2.5 refers to a rate, not to a share of cartons opened.

The sampling plan is what translates that rate into an instruction someone can follow. Under ISO 2859-1, you first find the sample size code letter for your lot at general inspection level II, then read across the single sampling table for your AQL to get a sample quantity and a pair of numbers: the accept count and the reject count. If the inspector finds defects at or below the accept number, the lot passes. At or above the reject number, it fails. There is no negotiation in the middle, which is exactly the point of using a published standard.

For a creator, the practical consequence is that the sample size is decided by the lot size, and the tolerance is decided by the sample size. Nobody gets to choose to look at twelve bags and call it done. On a 500-piece lot, the plan says 50 units, and 50 units is what goes on the bench. That is 10% of the order being physically handled, measured and flexed, which surprises people who assumed small orders get small checks.

The other half of the definition is what counts as a defect, and this is where most disputes start. AQL 2.5 is normally applied to major defects, which are the flaws a backer would notice and complain about: a skipped stitch, a crooked panel, a zipper that catches, a strap that is 8 mm short. A separate and far stricter limit applies to critical defects, and minor cosmetic issues sit under a looser one. Read as one number, AQL 2.5 is meaningless. Read as three classes with three limits, it is a workable specification.

This matters to a crowdfunding campaign in a specific way. A backer who receives one bad bag out of a thousand tells 200 people about it on the campaign comments. A backer who receives a good bag tells nobody. The asymmetry is why a published AQL is worth insisting on even at 500 pcs, and why Low MOQ Bag Production for Creators treats inspection as part of the standard programme rather than an optional extra.

The ISO 2859-1 Sample Size Table, Read for Small Lots

Here is the part of the standard that creators can use directly. The table below shows the sample size code letters and single sampling numbers for AQL 2.5 at general inspection level II, across the lot sizes a small brand actually orders. Keep it open when you review a QC report.

Two things jump out of that table. The first is that the absolute sample size grows far more slowly than the lot. Going from 500 pcs to 5,000 pcs multiplies your order tenfold and multiplies the inspection work fourfold, from 50 units to 200 units. The standard is deliberately built that way, because the statistical information in 200 units about a 5,000-unit lot is comparable to the information in 50 units about a 500-unit lot.

The second is the share of the lot inspected. At 500 pcs you physically examine one bag in ten. At 5,000 pcs you examine one in twenty-five. That sounds like the small order gets better treatment, and in one narrow sense it does, but it is the wrong way to read the table. What the standard holds constant is not the effort but the acceptable quality level. Both plans are tuned so that a lot running at roughly 2.5% major defects will usually pass, and a lot running materially worse will usually fail.

There is a third detail that catches small brands out, and it is the arrow system. For very small samples the ISO tables contain arrows that send you to a different sample size or a different accept and reject pair. A 120-piece lot at level II falls into a range where the table points you somewhere other than a straight read across. If your QC report shows an odd sample size for a small lot, that is the arrows at work, not an inspector improvising.

Level II is the default and the one you should assume unless you have agreed otherwise. There are three general levels, and level I halves the sample while level III roughly doubles it. Some creators ask for level III on a first run, which is a reasonable thing to want, though it does add inspection hours and therefore cost. The honest framing is that AQL 2.5 to ISO 2859-1 at level II is the standard our production programme runs, and it is a defensible specification to put in a purchase agreement.

At 500 Pcs You Inspect 50: What 3 Accept and 4 Reject Says

Take the case that matters most to a first campaign. A 500-piece lot sits in the 281-500 band, which gives code letter H, a sample size of 50 units, an accept number of 3 and a reject number of 4. The inspector pulls 50 bags at random from across the cartons, not from the top of the first one, and counts major defects across them.

If three or fewer major defects turn up, the lot passes and ships. If four or more turn up, it fails, and the whole 500-piece lot goes to rework or to a second inspection. There is no partial credit for a lot that produced three and a half. The reason the two numbers are adjacent rather than identical is that the standard needs a decision rule with no gap, so reject is always one more than accept.

What does four defects in 50 tell you? On the face of it, an 8% observed rate, which is far above the 2.5% the plan is built around. In sampling theory that jump is expected and is not an overreaction, because a plan with a 3-accept threshold is designed to pass lots that are genuinely at or under 2.5% most of the time, and to fail them some of the time as well. The accept number is not a promise that your lot contains fewer than 3 defects in 50. It is the level of evidence at which the standard says stop.

This is where creators get nervous, and understandably so. A 500-piece lot with a real defect rate of 3% has a meaningful chance of throwing four defects into a 50-unit sample purely by chance, and failing a lot that was mostly fine. That is not a flaw in the standard, it is the known behaviour of acceptance sampling, and the industry answer to it is not to relax the numbers but to use the switching rules: tighten after a failure, return to normal after a run of passes, and reduce once a supplier has a long clean history.

There is also a practical answer, which is better than the theoretical one. A first lot at 500 pcs should be preceded by a pre-production sample and a first-piece check on the line, because those two steps remove the defects that would otherwise arrive at the inspection bench in the first place. Sampling at the end is a safety net, not the mechanism that makes the bags good.

At 5,000 Pcs You Inspect 200: Same Standard, Different Shield

Now take the other end of the comparison. A 5,000-piece lot sits in the 3,201-10,000 band, which gives code letter L, a sample size of 200 units, an accept number of 10 and a reject number of 11. Ten defects out of 200 is an observed rate of 5%, and that is the fail point. Compare it with the small lot, where four defects out of 50, an observed rate of 8%, is the fail point.

The difference is the whole lesson of this article. The two plans look inconsistent if you read them as percentages, and they are perfectly consistent if you read them as sampling plans. A 50-unit sample is a noisy instrument. It needs to allow a relatively high observed rate before it will declare a lot bad, because random variation alone can produce three defects in 50 from a good lot. A 200-unit sample is a sharper instrument, so it can set the fail point lower and still treat good lots fairly.

Both instruments are aimed at the same target quality of roughly 2.5% majors. The large lot's plan discriminates more sharply between a good lot and a bad one; the small lot's plan discriminates less sharply and therefore lets through more marginal lots, and fails more decent ones. Understanding that asymmetry stops you from concluding that a 5,000-piece order is inspected less carefully. It is inspected differently, with a tool suited to its size.

For a growing brand the implication is encouraging. As you move from 500 to 1,000 to 5,000 pcs across re-orders, the inspection you receive becomes statistically more powerful without you asking for anything. The same AQL 2.5 to ISO 2859-1 line in your purchase agreement keeps working, and the sample size under it rises from 50 to 80 to 125 to 200. You do not need to renegotiate quality terms as you scale; you need to keep the ones you have.

It also means the reverse. If you are comparing two programmes and one offers only AQL 4.0 while the other offers AQL 2.5, that is a real difference and not a rounding one, because at the same lot size the 4.0 column has a higher accept number. The standard is worth naming precisely in your agreement, which is why we name ISO 2859-1 alongside the AQL level rather than quoting a number alone.

Why a Small Lot Is the Statistically Harder Place to Hit AQL

There is an uncomfortable truth about small-batch inspection that nobody puts in the sales material: a 500-piece lot is the hardest place to produce a clean AQL result, for reasons that have nothing to do with the skill of the sewing line.

The first reason is that small lots have no averaging. A 5,000-piece run settles into a rhythm by the second day and produces consistent work for weeks. A 500-piece run can be over in two days, and the early units come off the line while stations are still being set up. Setup units are the units most likely to carry errors, and in a small lot they are a larger share of the total.

The second reason is the sampling noise already described. With 50 units on the bench, a single bad hour can put four majors into the sample and fail the lot, where the same hour spread across 5,000 units would have been absorbed. The standard compensates by setting the accept number generously relative to the sample, but it cannot eliminate the effect.

The third reason is specific to bags. Every new material lot behaves slightly differently. A fabric roll with a different stiffness, a zipper from a new batch, a webbing dye lot that runs a shade darker: each of these changes how the operators work, and in a short run there is no time to adapt. This is one more argument for the two practical controls that do most of the work, which are a pre-production sample approved before bulk starts and a first-piece check in the first hour of the run.

None of this makes a small lot a bad idea. It makes it a place where the process controls matter more, because the end-of-line inspection is a weaker safety net than in a long run. The programme we run for creators puts its effort into those controls, which is also how we keep sampling within 6 to 10 working days and hold the bulk window at 35 to 50 days without surprises at the end.

Defect Classes: Critical, Major, Minor and Where Bags Fall

A defect count is meaningless until you have agreed what to count. The convention in soft goods is three classes, and bags distribute across them in a fairly predictable pattern.

The critical row is the one to read twice. A single critical defect fails a lot regardless of how good the other 499 bags are, because a critical defect is not a cosmetic issue, it is a safety or compliance one. For children's backpacks that can include small parts or drawstrings; for tech bags it can include a battery pouch that does not meet the relevant transport rule. Our Custom Kids Backpacks for Kickstarter: Safety and Sizing page covers how compliance reshapes the inspection plan in that category.

The major row is where the AQL 2.5 counting happens, and it is where your agreement should be most specific. Vague major definitions produce disputes, because an inspector and a brand owner will draw the line between major and minor differently. A written defect list with photographs, agreed before bulk, removes most of that disagreement. It is a half-day of work that prevents a two-week argument.

The minor row is where experience shows. Loose threads and small marks are inevitable in sewn goods at any volume, and a plan with no minor allowance produces failed lots for cosmetic reasons that no backer would ever have noticed. A separate AQL 4.0 for minors is the conventional balance, and it is what we use unless a creator asks for something stricter for a premium reward tier.

One practical note about measurement. Size and weight tolerances belong in the major class and should be written as numbers, not as adjectives. A bag specified at 45 cm tall with a tolerance of plus or minus 1 cm is inspectable. A bag specified as true to size is not. If you want a clean inspection on a first run, write the tolerances down.

What an AQL 2.5 Pass Does Not Cover

This is the section creators most often need and least often get. An AQL pass is a statement about the sample, not a warranty about every unit, and there are four things it plainly does not do.

• It does not guarantee zero defects in the uninspected units. A lot passing at 3 majors in 50 is a lot that very probably contains some majors in the 450 that were not opened. That is the nature of acceptance sampling, not a defect in the service.
• It does not cover defects that appear later. Colourfastness after sun exposure, coating delamination after six months, a zipper that fails in month three: none of these are visible on the inspection bench and none are caught by counting.
• It does not verify compliance. Regulatory testing for the market you are shipping into is a separate activity with its own laboratory reports.
• It does not check that the bag does its job. AQL counting is largely visual and dimensional. Whether a strap survives a loaded pull is a functional test, not a count.

The fourth point deserves its own emphasis because it is where bag inspection differs most from inspection of, say, a moulded plastic part. A bag is an assembly of fabric, thread, hardware and closures, and most of what can go wrong with it is discovered by using it rather than by looking at it. That is why a serious inspection programme layers functional checks on top of the AQL count: zipper cycle tests, strap and handle pull tests, load tests with the bag filled to its stated capacity, and rub tests on printed or coated surfaces.

Those functional checks usually run on a small number of units, commonly three to five from a lot rather than the full AQL sample, because they are slow and sometimes destructive. They are not a substitute for the sample and the sample is not a substitute for them. A programme that only counts is cheap and misses things; a programme that only tests pulls is thorough and has no idea about the other 495 bags. You need both, and the cost of both is why a 137-staff operation across 7 lines can run them without adding weeks to the schedule.

Our Kickstarter Backpack QC: What AQL 2.5 Inspection Checks page lists the checks in the order an inspector performs them, which is useful if you plan to attend an inspection yourself.

The 500-Piece Inspection Day: What Actually Happens

It helps to know the shape of the day, because most of the anxiety around QC comes from not knowing what is happening while it happens. On a 500-piece lot, a full AQL inspection with functional checks is typically a single working day on the bench, sometimes two if the bag has many components.

1. Carton selection. Cartons are chosen at random across the lot, not sequentially. The inspector records which cartons were opened so the draw can be audited.
2. Unit selection and external check. 50 units are pulled, and each is checked for overall appearance, shape, symmetry and obvious soiling before anything is measured.
3. Dimensional check. Key measurements are taken against the approved specification, usually height, width, depth, strap length and handle drop, at the stated tolerance.
4. Construction check. Seams, bar-tacks, binding, rivets and attachments are examined, with the stress points given the closest attention.
5. Functional check. Zippers are cycled, buckles are fastened and released, and straps are pulled on a sample of three to five units.
6. Defect logging and classification. Each finding is recorded by class and position, and tallied against the accept and reject numbers.
7. Report and decision. The lot passes, fails, or passes with a rework instruction for specific identified issues.

The classification step is where the written defect list earns its keep. An inspector who finds a 3 mm print offset needs to know whether that is a minor or a major in your specification before tallying it. If the list was agreed in advance, the tally is arithmetic. If it was not, it becomes a phone call, and the truck is waiting.

Where a creator can add real value is by attending once, for the first run, in person or over video. You will learn more about your product in one inspection day than in a month of email, because you will see the five things that go wrong most often and you will know what to specify more tightly next time. The second run is always smoother than the first, and this is most of the reason why.

A final practical point about timing. Inspection happens after packing is complete enough to draw cartons from, which means it sits near the end of the 35 to 50 days bulk window, not in the middle of it. If a fail triggers rework, the recovery time comes out of your delivery promise, which is why building a buffer into the campaign delivery date matters more than squeezing the production window.

Sampling Risk and the Cost of a Rejected Lot

What happens when the reject number is hit? The standard is clear and the consequences are worth pricing before you start, because a failed lot on a small order is a more serious event than the same failure on a large one.

There are three outcomes. The first is rework, where specific defects are corrected and the lot is re-presented for inspection. Rework is the common outcome for construction issues such as skipped stitches, poor binding ends or loose threads. It costs labour and time, typically several days, but it saves the inventory.

The second is replacement, where affected components are remade. A zipper batch that fails cycling cannot be repaired, so the units are re-zipped or remade. This is more expensive and slower, and it is the reason to approve component samples before bulk rather than after. The third is rejection and remake, which is rare, expensive, and usually the consequence of a specification that was wrong from the start, such as a size that does not fit the product it was designed for.

On a 500-piece lot, a rework event has a particular sting: it is a large share of your total order and there is no other shipment to hide it behind. A brand with a 5,000-piece order spread across three production rounds can absorb a problem in one round. A creator with one 500-piece run for a campaign with a fixed delivery promise cannot. The mitigation is not to demand a stricter AQL, it is to build the right buffer into the timeline and to insist on the upstream controls.

It is also worth understanding that the standard itself gets stricter after a failure. The switching rules move you to tightened inspection, which means a larger sample or a lower accept number on the next lot, and it takes a run of five consecutive passing lots under normal inspection before you can switch back. That is deliberate design: a supplier who has just failed gets scrutinised harder. It is a good sign if your production team volunteers this rather than waiting to be asked.

Our Kickstarter Bag Campaign Risks: Delays and Quality page treats the schedule side of the same problem, including how much buffer to build around a first production run.

The Checks That Sit Outside AQL 2.5

A first campaign benefits from a small set of checks that sit alongside AQL counting rather than inside it. None of them are exotic, and together they cover most of what a counting plan misses.

Notice the timing column. The cheapest checks are the earliest ones, and the expensive ones are the ones you discover late. A pre-production sample that reveals the bag is 2 cm too narrow costs a sample revision inside the 6 to 10 working days sampling window. The same discovery at final inspection costs a rework or a remake inside the 35 to 50 days bulk window, and the difference between those two is the difference between a campaign that delivers on time and one that posts an apology.

For creators planning a first order, the sensible sequence is straightforward. Approve a physical pre-production sample. Hold a first-piece check. Let the standard AQL 2.5 sample run at level II. Add three to five units of functional testing. Ask for the report with photographs, not just a pass or fail. That is a complete and defensible quality programme for a 500-piece run, and it is the one we apply whether the lot is 500 pcs or 5,000.

If you want the category-specific version of this, our Kickstarter Travel Bag QC: AQL 2.5 Checks for Luggage and Tote Bag QC: Strap Attachment and Handle Tear Tests pages set out what changes by product type.

Frequently Asked Questions

People Also Ask