Custom Bag Quality Assurance: Building a QA Plan With Your Partner

Most creators who order custom bags believe they have quality control because they intend to inspect the shipment. Intention is not a system. When a lot arrives with crooked panels, the questions that decide the outcome are procedural: what counts as a defect, how many are tolerated, who has the authority to fail the lot, and what happens next. A QA plan answers those questions in advance, in writing, so that nobody is negotiating under time pressure while 500 or 5,000 units sit packed in cartons. This article is about building that document with your production partner, not about the inspection itself. The inspection technique side sits in our Bag QC and Compliance programme, and the specific arithmetic of AQL sampling is covered in Custom Bag QC: AQL for Kickstarter Campaigns. What follows is the layer above both: the agreement that makes an inspection result actionable.

Why a QA plan beats a checklist

A checklist is a list of things to look at. A QA plan is an agreement about what happens when the looking finds something. The distinction sounds academic until you watch a dispute unfold. An inspector reports 11 major defects in a 500-piece lot. The production side says 11 is acceptable for that lot size. You say it is not. Without a written plan both parties are arguing from memory and from what they believe is industry practice, and the argument is settled by whoever has more leverage at that moment, which is usually whoever is holding the goods. With a plan, the answer is a number written down three months earlier: at 501-1,200 pieces, code letter J, a sample of 80 units accepts on 5 and rejects on 6. Eleven fails. The conversation ends.

There are three failure modes a checklist cannot resolve and a plan can. The first is an ambiguous standard: the checklist says check print quality, but nobody defined how much registration offset is acceptable, so every unit is arguable. The second is absent decision rights: the inspector finds a problem but has no authority to stop a line or release a payment, so the finding is advisory and gets ignored when the calendar is tight. The third is no rework path: the lot fails and nobody agreed in advance who pays for sorting, who pays for re-inspection, or how many days that adds.

The cost of writing the plan is small relative to the cost of not having it. Sampling runs 6-10 working days for a standard construction and 12-15 working days for complex builds, and bulk production runs 35-50 days after deposit and sample approval. A single unresolved dispute can consume the slack in that window entirely, because re-inspection, sorting and rework are sequential, not parallel. A plan drafted during the sampling window costs you a few hours when you still have slack, and it removes days of ambiguity later. Our campaign bag specialist team treats the plan as a deliverable of the sampling stage, not an optional extra, precisely because that is the last point at which changes are cheap.

The header block: scope, references and revision control

Every plan should open with a header block that makes the document identifiable and binding. It needs a plan ID, a revision number, a date, the product name, the styles and colourways covered, the total quantity, the purchase order reference, and the names of the two parties. That sounds bureaucratic, and the reason to do it anyway is that bag programmes change. A campaign that starts with one colourway frequently adds a second after stretch goals unlock, and each colourway carries its own 500-piece minimum. If your plan says the Order without saying which revision and which colourways, then the stretch colourway is governed by a document that never mentioned it.

Revision control is the part most creators skip and later regret. The rule is simple: every change to the product, the material, the tolerance or the acceptance number creates a new revision, and the revision is dated and acknowledged in writing before it applies. Version 1.0 is the plan agreed at sample approval. Version 1.1 might record a lining colour change. Version 2.0 might record a change of zipper supplier. No verbal change is in force until it appears in a revision. If you take one habit from this article, take that one, because the majority of disputes we see trace back to a change that was agreed in a chat message and never entered the document.

The references line deserves particular care. A plan that does not name its reference artefacts is a plan that can be inspected against anything. Name the approved sample by its ID and the date it was sealed, name the spec sheet by version, and name the artwork file by version. If you have not written a spec sheet yet, do that first; the method is set out in How to Write a Bag Spec Sheet for a Fast, Accurate Quote, and the sampling workflow that produces the sealed reference is described in Bag Sampling and Prototyping for Crowdfunding. The plan then points at artefacts that already exist.

Anchoring the plan to one approved sample

A QA plan without a physical reference is unfalsifiable, because every judgement becomes a matter of taste. The fix is a sealed pre-production sample that both parties treat as the definition of correct. In practice this means three things. First, the sample is produced from the actual bulk materials and the actual bulk tooling, not from a close substitute. Second, it is signed, dated and photographed. Third, there are at least two sealed copies, one retained by our production team on the floor and one held off the floor, so that a damaged or altered reference can be detected.

The sample stage is also the moment to fix tolerances in writing, because it is the moment when you can still see the thing being tolerated. Fabric bag dimensions move with cutting and sewing tension, and a plan that says 45 cm tall without saying plus or minus anything will produce an argument. Agree a tolerance band, typically in the range of a few millimetres to a centimetre depending on construction, and write the measurement method as well: where the tape is placed, whether the bag is empty or stuffed, whether the handle is included. The mechanics of choosing those bands are covered in How to Handle Size Tolerances in Fabric Bag Manufacturing, and the approval workflow itself in How to Request and Approve a Pre-Production Sample for Bags.

Colour is the other reference that must be sealed rather than described. Written colour names and even Pantone references do not fully determine appearance, because the same reference reads differently on cotton canvas, on coated polyester and on recycled rPET. The plan should therefore name a sealed fabric swatch for each colourway in addition to any numeric reference, and should state the viewing condition for comparison, since colour judgement under a warm warehouse lamp and under daylight can differ enough to change a decision. Where a colourway is added later, it gets its own sealed swatch and its own plan revision, and it carries the same 500-piece minimum as the original.

Finally, the plan should say what the sample is not. A pre-production sample confirms construction, materials and proportions for that style. It does not certify the whole lot, it does not certify carton packing, and it does not certify that the bulk will match on a day when the line is running at full speed with a different operator on the zipper station. Those are separate gates, and conflating them is how creators end up approving a beautiful sample and receiving an inconsistent shipment. The distinction between approval events and inspection events is explored further in Custom Bag Pre-Production Samples: What They Prove.

Defect classes and the acceptance numbers

The middle of the plan does the real work: it defines what a defect is and how many are allowed. The standard approach divides findings into three classes. Critical defects are unsafe, illegal or render the item unusable, and the tolerance is zero, meaning a single occurrence fails the lot. Major defects are ones a backer would notice and complain about, and these are typically held to AQL 2.5. Minor defects are visible close up but acceptable at arm's length, and these are typically held to AQL 4.0. Writing these definitions into the plan, with bag-specific examples under each, removes most of the judgement calls before they happen.

The examples matter more than the definitions. Two people can agree that a major defect is one a backer would notice and still disagree about whether a 2 mm print offset on a side gusset qualifies. If the plan lists slight print offset as minor and states the offset tolerance in millimetres, the inspector classifies it in seconds and the classification is not negotiable at the end of the run. Build the example list during sampling, when you can look at real units and write down what you actually see, rather than inventing hypotheticals in the abstract. A useful starting list is discussed in Custom Bag Defects: What Tolerance Is Acceptable Under AQL 2.5.

One further clause belongs here: the mixed-defect rule. A lot can pass on majors and pass on minors and still be unacceptable in combination, because a unit carrying two minor defects often reads as one major defect to the person opening the box. State how combined findings on a single unit are treated. A common formulation is that two minor defects on the same unit count as one major. Whatever you choose, write it down, because it is exactly the kind of rule that gets improvised at the inspection table if it is not already fixed.

Sampling arithmetic under ISO 2859-1

Once defect classes exist, the plan needs sampling arithmetic: how many units get examined and what count passes. The widely used basis for bag orders is ISO 2859-1 single sampling at general inspection level II with an AQL of 2.5 for major defects. The lot size determines a code letter, the code letter determines the sample size, and the sample size determines the accept and reject numbers. The important structural point for a plan is that the numbers are derived, not chosen. You do not get to decide that 20 units feels like enough. You look up the lot size and you get the sample.

Read that table against your real order. A first campaign order frequently lands at or near the 500-piece minimum per style and colourway, which puts it in code letter H: 50 units examined, accept on 3, reject on 4. A 3,000-unit order moves to code letter K: 125 units, accept on 7, reject on 8. Notice what does not scale: the sample size grows far more slowly than the lot. At 500 pieces you inspect roughly a tenth of the lot; at 20,000 pieces you inspect well under two per cent. That is the nature of acceptance sampling, and the plan should say so plainly, because a creator who assumes inspection means every unit checked will draw the wrong conclusion from a passing report.

Two clauses complete this section. The first is switching rules: tightened, normal and reduced inspection, and the conditions that move you between them. A supplier with a history of passing lots can move to reduced inspection; a failed lot typically triggers tightened inspection on the next one. Writing the trigger into the plan prevents the argument about whether this run gets the same treatment as the last. The second is the zero-tolerance carve-out: whatever the AQL for majors, critical defects fail the lot on a count of one, and compliance-related findings where a documented certificate is missing also fail regardless of the sample count. The reasoning behind these numbers in a campaign context is set out in Custom Bag QC: AQL for Kickstarter Campaigns.

The four gates inside the production calendar

A plan is only useful if its checks land at moments when the result can still change something. In a bag programme there are four such moments, and the plan should name each one with a trigger, an output and a decision. Gate one is material and component approval, before bulk cutting, where fabric, lining, webbing, zippers and hardware are confirmed against the sealed references. Gate two is the first-piece check, in the first hours of bulk sewing, where a completed unit off the real line is compared against the approved sample. Gate three is the in-line check, typically at 10-30 per cent finished. Gate four is the final random inspection, at 80-100 per cent packed.

The gates answer different questions and the plan should say which, because that determines what happens to the result. Gate one asks whether to start. Gate two asks whether the line setup is right. Gate three asks whether to continue. Gate four asks whether to release the balance payment and ship. Only gates one, two and three can prevent defects; gate four can only detect them, by which point the remedy is sorting, rework or remake. A plan that funds gate four and skips the earlier gates is a plan that has chosen detection over prevention, which is a legitimate choice on a tight budget but should be a conscious one. The trade-off is examined in detail in In-Line vs Final Inspection for Bag Orders.

Mapping gates onto the calendar is where plans usually go wrong, because the arithmetic is unforgiving. Sampling takes 6-10 working days for standard constructions and 12-15 working days for complex ones. Bulk production takes 35-50 days after deposit and approved sample. Within that 35-50 day window, gate two must happen in the first one to two days, gate three somewhere around the middle, and gate four in the final days before packing completes. A failed gate three with rework can consume a week. Build the plan with the gate dates as actual dates, not as percentages, and check them against the shipping date you promised backers. Practical scheduling guidance sits in Fast Turnaround Custom Bags: What Timelines Are Realistic.

Roles, authority and who signs what

The most common structural weakness in a bag QA plan is not technical. It is that nobody wrote down who decides. An inspector issues a report; a report is a recommendation. Somebody has to hold the authority to stop the line, to accept a corrective action, to release the balance payment, and to reject the lot. Name each of those authorities in the plan beside a named role, not a job title that three people share. On your side that is usually one person, and on ours it is the programme coordinator, with our production team responsible for executing the corrective action.

1. Material or component deviation is found at gate one: our production team proposes a substitute, the creator approves or refuses in writing, and a refusal creates a plan revision.
2. First-piece mismatch at gate two: the line stops, the setup is corrected, and a new first piece is checked before sewing resumes.
3. In-line major rate above the accept number at gate three: production pauses, the defect cause is identified on the floor, and the affected units already made are segregated for sorting.
4. Final inspection fail at gate four: the lot is held, a corrective plan is written with dates, and the balance payment is not released until re-inspection passes.
5. Any critical defect at any gate: the lot is rejected outright and the remedy is remake or full sort, not partial rework.

The escalation ladder belongs in the same section. Write three rungs: the finding is raised with the floor supervisor on the day; if unresolved within a stated number of working days it moves to the programme coordinator; if still unresolved it moves to a named person on each side with authority to commit cost. Give each rung a time limit in working days, because an escalation path without a clock is just a longer version of the original argument. Two working days per rung is a reasonable default for a 35-50 day production window, and it is the kind of number that should be in the plan rather than in somebody's head.

Decide also whether you are using your own inspector, ours, or a third party, and write that into the plan with the reporting line. Each has a place. An independent third party has no interest in the outcome and is the right choice when the relationship is new or the order is large. Our own inspection is faster and cheaper and is appropriate when the plan is mature and the construction is proven. Using both is not wasteful on a first run: an in-line visit from our side to catch setup problems and a final visit from an independent inspector to govern payment release. The decision framework is laid out in Should You Hire a Third-Party Bag QC Inspector for Kickstarter.

The test plan: what gets tested and how often

Visual inspection against an AQL sample catches appearance and construction faults. It does not catch a zipper that fails after three weeks of use, a strap that pulls out under load, or a print that rubs off. Those need tests, and the plan should list them by name with a frequency and an acceptance criterion. A workable bag test plan has four tiers: per new material, before bulk cutting; per lot, during inspection; per shipment, before it leaves; and per compliance claim, whenever you make a claim on the campaign page.

• Zipper cycle test on 3-5 units per style per lot, opening and closing a stated number of times without binding or splitting.
• Strap and handle pull test on 3-5 units, loaded to a stated weight and held for a stated duration without attachment failure or visible deformation.
• Colourfastness to rubbing on printed, dyed and coated surfaces, submitted before bulk for any new material and repeated if the material changes.
• Seam strength check on the load-bearing seams, particularly where webbing meets the body panel.
• Carton drop and packing check before shipment, testing the master carton rather than the bag.
• Compliance documentation review for any claim you make publicly, matching certificates to the actual production lot.

Frequency is where plans get vague, so be specific: a test that says as needed will not happen. Tie each test to one of the four tiers and to a gate. Per-new-material tests happen at gate one and gate only; if the material changes at any point, the test repeats. Per-lot tests happen during gate four on the stated number of units. Per-shipment tests happen after packing and before the goods leave. Per-claim tests happen once per claim, before the campaign page goes live, not after. Detailed methods and realistic pass criteria are described in Bag Material Testing: Fabric, Zipper and Print.

One clause protects you more than any individual test: the material-change trigger. If any input material changes after the approved sample, for any reason including a supplier shortage, the material tests repeat and the plan gets a revision. Most quality failures that reach backers after a passing inspection trace back to a substitution that happened quietly between sample and bulk. Making the trigger explicit converts a silent substitution into a documented event with a test attached, which is a very cheap form of insurance on a lot that will be in backers' hands for years.

Non-conformance, rework and the reject decision

The reject decision is the clause that gives the rest of the plan force, and it needs to be written before anyone is emotionally invested in a shipment. Define three outcomes. Accept: findings are within the AQL numbers and the lot ships. Accept with correction: findings exceed the numbers but the defect is sortable or reworkable, so the lot is sorted or reworked and re-inspected. Reject: the defect cannot be corrected by sorting or rework, or a critical defect is present, and the remedy is remake or a commercial settlement.

Then define the mechanics of the middle path, because it is the one that actually happens. Who sorts: our production team, on the floor, against a defect reference provided by the inspector. How the sort is verified: a 100 per cent sort followed by a fresh AQL sample drawn from the sorted lot, not a re-check of the original sample. How long it takes: sorting a 500-piece lot is a matter of a day or two; sorting 5,000 pieces is a week or more. Who pays: sorting and rework cost sits with whoever caused the defect; re-inspection cost typically sits with the party requesting it unless the plan says otherwise. Write all four, because each one is a negotiation if it is not a clause.

Be honest in the plan about what rejection costs you, not only what it costs the other side. Rejecting a lot inside a 35-50 day window usually means the remake cannot be completed inside the original window, which means the shipping date you published moves. A plan that makes rejection easy but says nothing about the calendar consequence can push a creator into a worse outcome than the defect. So write the decision tree with dates attached: if the lot fails at gate four on day 42 and the ship date is day 48, then remake is not available and the realistic options are partial shipment, sort-and-ship with a disclosed remedy, or a moved date with a backer update. The communication side of that scenario is covered in Kickstarter Shipping Advice: Updates Backers Trust.

Finally, record how disputes are settled when the two sides read the same report differently. The cleanest mechanism is a defined re-inspection by a named independent party, with the cost following the outcome: if the re-inspection confirms the failure the cost sits on the production side, and if it overturns the failure it sits with the party that raised it. That single sentence removes the incentive to argue, because arguing becomes expensive for whoever is wrong. Put it in the plan at the start, when both parties are optimistic, rather than at the moment when one of them is holding a failed lot.

The documentation pack you should receive

A QA plan is incomplete without a defined output. State exactly what documents arrive, in what form and by when, and treat a missing document as a non-conformance in its own right. The minimum pack for a bag order is an inspection report with the lot size, code letter, sample size, accept and reject numbers, and a classified defect count; photographs of every defect found with a scale reference; a measurement table for the sampled units against the stated tolerance; a packing and carton count; and any test or compliance certificates referenced in the test plan.

Timing matters as much as content. The report should arrive before the balance payment is released, not after, because the report is the instrument that governs that release. On a T/T 30/70 structure, where 30 per cent is paid at order and 70 per cent against shipment, the natural control point is: report received, report reviewed, report passed, then the 70 per cent is released. If your plan allows payment first and report second, you have removed the only leverage the plan creates. Write the payment term and the report condition in the same clause so that they cannot be separated operationally.

Specify the format too, because a report you cannot read quickly is a report you will not read. Ask for defect counts broken out by class rather than a single total, photographs that show the defect and the unit in the same frame, and a one-line conclusion stating pass, conditional pass or fail. Then give yourself a review procedure: a fixed number of working days to accept or contest the report, after which silence counts as acceptance. Two working days is a workable default. Learning to read these documents fast is a skill in itself; the method is set out in Reading a Bag QC Report in Ten Minutes, and the packing and carton side is covered in Carton Marking and Master Carton Specs.

Keeping the plan alive through re-orders and revisions

The plan is a living document and the most common failure is that it is written carefully for the first order and then ignored. Two events should trigger a mandatory review. The first is any product change: a new material, a new colourway, a new hardware supplier, a revised print file. The second is any inspection outcome, pass or fail, because both carry information. A pass with zero majors suggests the plan may be inspectable at reduced level next time. A fail suggests the defect class list was incomplete and needs the new example added.

Re-orders are the case where a stale plan does real damage. A campaign that succeeds usually re-orders, often in a new colourway unlocked by a stretch goal, and each colourway carries the 500-piece minimum per style and colourway. The temptation is to re-use the existing plan unchanged. Resist it for one specific reason: the second run is frequently made by a different line with different operators, after materials have been re-sourced, and the defect profile will not be identical. Open the plan, increment the revision, confirm the sealed reference is still the current one, and re-issue it. The review takes an hour; the alternative is discovering at gate four that the plan described a product you are no longer making.

Set a review cadence as well, even when nothing changes. An annual read-through, or a review at every fifth lot, is enough to catch the slow drift where tolerances that were once argued over become assumed and then forgotten. Record who attended, what changed and what did not. Over a programme of several campaigns this file becomes the single most valuable artefact you own: it is the difference between starting every order from scratch and starting from a documented history of what has actually gone wrong and what fixed it. Our programme coordinator keeps that history across campaigns so that a returning creator does not have to rebuild the wheel.

If you are at the stage of choosing a production partner rather than writing a plan with one, the plan structure in this article doubles as a question list. Ask a candidate partner which of these sections they already work to, whether they can produce a sealed pre-production sample within 6-10 working days, what AQL they default to, who signs off a gate three failure, and what the report pack contains. A partner who answers those five questions concretely will be straightforward to work with. A fuller version of that due-diligence list sits in Sourcing Partner Questions to Ask Before Ordering Bags.

Frequently Asked Questions

People Also Ask