How Handbag QC Changes Between Sampling and Bulk Production

QC does not drift gradually from the sample room into the bulk run; it crosses a hard line. In sampling, every unit is examined and every dimension either matches the drawing or does not, so the working standard is certainty. Once a line is producing 500 pieces per style / colourway, nobody can examine everything, so the standard becomes probability managed to ISO 2859-1. Nearly every question that later wrecks a bulk run was answerable during the 6-10 working days of sampling, before a single bulk hide was cut. What follows is the decision order, then each step of the handover.

Two stages, two sets of rules

The Decision Order, Stated Once Up Front

Two teams acting at two speeds are the reason handbags fail after approval rather than before it. Sampling rewards patience because one artisan can spend four hours resolving a folded corner. Bulk rewards repetition because the same operation is performed 500 to 20,000 times by several people with different habits. Nothing about the physics changes between the two; what changes is who has time to look. The seven steps below are ordered deliberately, and doing them out of order costs days rather than minutes.

• Decide what acceptable means in numbers before bulk/material is committed, not after the first complaint.
• Separate material approval from bag approval, because a beautiful unit made of the wrong lining still fails later.
• Seal and duplicate the reference unit so both sides can argue from one object rather than from photographs.
• Read the first thirty bulk pieces in detail, since that window contains the whole cost of the run.
• Switch deliberately to statistical counting once output stabilises, using one agreed standard.
• Instrument the three operations that drift, then give the team authority to stop.
• Hand the tolerance set to final sampling unchanged, so the last gate measures the same thing the line measured.

Those seven lines are the whole method. Everything below explains why each sits where it does and what it costs to skip it. Experienced readers will notice that only one of the seven concerns final sampling, which is deliberate: by the time goods are packed, almost every expensive decision has already been taken, and a rejection at that point is a verdict on work done forty days earlier.

Step One: Why Sampling Runs on Absolutes and Bulk Runs on Samples

A sample round produces one to three units. Examining them fully costs perhaps two hours and catches everything from a mis-set label position to an adhesive that has not cured. Bulk produces hundreds of units built from dozens of hides, several component lots and three or four operators per station, so the same standard of examination is impossible and also conceptually wrong. Bulk QC asks a narrower question: is this production process stable enough that the whole lot can be accepted at a defined risk level?

That is why the vocabulary shifts. In the sample room a fault is a fault. In bulk it becomes a grade, either critical, major or minor, and it is counted against a number derived from lot size. ISO 2859-1 fixes both figures so nobody negotiates them on the day: from a lot of 281 to 500 pieces the code letter is H, giving 50 units to examine, with three or fewer defects passing the lot at AQL 2.5 and four failing it. The mathematics replaces argument, which is its real value.

One consequence surprises founders: a passed lot may legally contain rejects. AQL 2.5 describes tolerance for a defect rate, not a promise of perfection, and the programme absorbs that residual through customer returns rather than through counting everything twice. Choosing to sample does not mean accepting bad work; it means pricing the remaining risk consciously and then driving that risk down through process control rather than through inspection volume.

Step Two: Turn the Approved Unit Into Written Numbers

Approval is only meaningful against a specification. A signed off sample without dimensions invites exactly the argument that statistical counting cannot resolve, because neither side can say what within tolerance meant. The conversion takes about ninety minutes and covers five areas: body dimensions with tolerances in millimetres, materials by reference rather than description, hardware finish against a plated swatch, construction details such as stitches per centimetre, and the marks that identify the unit as yours.

• Dimensions. Width, height, gusset, handle drop and strap length, each written with a plus or minus figure, since everything measurable needs one.
• Materials. Named by supplier reference and finish code, with a retained swatch signed at both ends rather than adjectives like soft.
• Hardware. Finish, plating thickness expectation and the swatch it was approved against, which settles most later disputes in one glance.
• Construction. Seam allowance, stitch density and thread ticket, because these are invisible in a photograph and decisive in wear.
• Marking. Position, size and method for every logo, care label and bar code, dimensioned from two fixed reference edges.

Material approval deserves its own line even though it appears inside this list. A tote may pass every dimension while using a lining whose chemistry is not what was declared, and the failure then arrives as a compliance problem rather than a defect. Where EU sales are intended, the restricted substances named in REACH Annex XVII should be written into the same specification, with a schedule for laboratory confirmation. Approving materials separately from the finished unit is the habit that prevents this class of surprise.

Step Three: Read the First Thirty Units Before Packing Starts

The pilot run is the last cheap moment in the whole programme. At QUANZHOU JUNYUAN BAGS this handover is scheduled deliberately: our validated partner factory, Quanzhou Yingming Bags, builds the first thirty pieces on production tooling with production components, and every one of them is examined against the tolerance set before the balance is cut. Thirty is enough to expose operator-to-operator variation and too small to waste much material if a tooling change is needed.

What the pilot reveals is rarely dramatic and almost always cheap to fix at this point. A strap that has been set 3 mm short across all thirty units is a pattern or jig correction, not a defect list. A lining panel that pulls at one corner on four of thirty indicates operator method, which a two-minute demonstration resolves. Left until final sampling, the same two faults become either a reworked lot or a negotiation nobody planned for, with a container already reserved against it.

Give the pilot a hard gate in the schedule. Where a line is expected to produce 500 pieces in eleven days, holding four hours at the front for this reading costs less than half a day and typically removes between sixty and eighty percent of what would later be counted as defects. Programmes that skip it do not save time; they convert a four hour check into a forty hour sorting exercise performed under pressure.

Step Four: Hand Over From 100% Checking to Statistical Counting

Once the pilot is released, checking cannot continue at 100% without throttling output, so coverage declines in a planned way. A practical ramp used on structured handbags looks like this: one unit in five examined through the first 200 pieces, one in twenty through the next 800, and thereafter a defined pull every two hours plus the standard final sample. The ramp is not laziness. It concentrates attention where the risk actually sits, which is early output before habits settle.

Notice that the final sample is random. Systematic selection from the top of each carton reaches units packed first and misses late-packing drift, so the drawer is chosen from the whole lot and not from what is closest to the tape machine. Teams working under ISO 9001 will recognise this as standard process discipline; for everyone else it is simply the difference between a count that means something and a count that flatters everybody.

Step Five: Three Measurements That Catch Drift Early

Every two hours, five units come off the line and three things are measured. Pick observables that move before defects appear, rather than counting finished faults after they happen. The useful three for most leather and PU constructions are folded edge thickness, handle symmetry and thread tension, and none of them requires expensive equipment: a caliper, a steel rule and your hands will do.

1. Folded edge thickness. A rise of about 0.3 mm against the reference unit signals adhesive build-up or a double fold, both of which later crack in cold transit.
2. Handle symmetry. Measured drop from the top edge on both sides within 2 mm; asymmetry indicates a jig drifting rather than a cutting fault.
3. Thread tension. Pressed flat along a seam, the stitch line should sit without puckering; visible ridges mean the machine has gone off during the shift.

Record the three figures on one sheet per shift rather than in memory. Five readings a day over eleven production days produces 165 observations, which is enough to see that faults cluster after a material change or after a break rather than evenly through the day. That pattern is actionable in a way that a single final report never is, and it costs one line per hour of somebody's time.

Step Six: Stop Triggers Writing Themselves Into the Contract

A stop trigger says: when this happens, work pauses and nobody needs permission. Without one, a line operator who spots escalating faults has to find a supervisor, a supervisor has to find a coordinator, and the coordinator is asleep in another time zone. With one, the same fault costs twenty paused minutes instead of two hundred defective units. Three triggers cover almost everything on a handbag line.

• Critical defects. Anything sharp, anything that fails a specified pull test, anything legally non-compliant stops the line immediately and gets photographed before the unit is touched.
• Same major fault three times in one hour. Pattern beats coincidence; stop, diagnose and restart with a corrected method rather than sorting later.
• A measurement outside tolerance on two consecutive reads. Confirm with a third unit, then hold the batch made since the last good reading for separate assessment.

Write these three lines into the purchase order, along with who pays for the paused time. That second half is what makes them real, because nobody stops a line when doing so is unbudgeted. Embedding two working gates into one agreement is easier than it sounds and is covered in more detail in the piece on running inline and final inspection together.

Putting Both Stages on One Calendar

Sampling consumes 6-10 working days, extending to 12-15 for framed, woven or heavily hardware-led constructions. Bulk consumes 35-50 days. Those two numbers are the only ones a planner needs for eighty percent of programmes, with the remainder driven by when materials land. Materials are the quiet variable: a lining arriving eight days late does not shorten the bulk window, it removes choice from it, which is why approval order matters more than approval speed.

Build one calendar with three fixed dates rather than two windows: the date the tolerance set is signed, the date the pilot runs, and the date final sampling may begin. Everything else flexes around those three. Programmes built this way rarely argue about expediting, because the delay is visible two weeks before it happens instead of two days after it was unavoidable.

What Survives the Handover and What Must Be Rewritten

The tolerance set survives intact from sampling to bulk to final inspection. Anything else duplicating it creates two standards, and two standards means every disagreement resolves in favour of whichever one is more convenient at the moment. Rewriting for bulk is limited to coverage percentages, stop triggers and reporting rhythm; the numbers themselves never move without a signed revision, and every revision is dated so that units made before it can still be assessed fairly.

What genuinely needs rewriting each season is the indication that something new might go wrong: a new hardware finish, a new tannery, a first order above 5,000 pieces, or a change in the operator mix during peak months. Each of those deserves a short added note in the tolerance set and an extra half day of pilot inspection. Stable programmes do exactly that, which is why their bulk defect counts drift downward season after season rather than holding steady.

Conditioning: The Twenty-Four Hours Before Anyone Measures

Every dimension on a tolerance sheet is a statement about a material at a particular moisture content, and most arguments about measurement are really arguments about when the unit was measured. Leather absorbs and releases moisture with the room, and a panel that left the line at 30 per cent relative humidity will read differently from one rested overnight at 50. The convention below costs nothing but twenty-four hours and removes a whole family of false failures.

• Condition flat for 24 hours at 23 plus or minus 2 degrees Celsius and 50 plus or minus 5 per cent relative humidity before any dimension is recorded.
• Condition the reference unit with the lot, so one object is not being compared against another in a different state.
• Never measure cold. Units taken from an unheated enclosure read up to 2-3 mm different over a 400 mm panel until they come to room temperature.
• Write the conditions into the tolerance set, so an inspector working three months later is replicating the situation rather than inventing one.

Reading Operator Variation Without Blaming Anyone

The first thirty pilot units are read unit by unit, but their most useful output appears when those same reads are attributed to the station that made them. Three reads per station across thirty units is enough to separate a method problem from a material problem, because a method problem concentrates and a material problem spreads. The table is the shape this takes on a five-station line.

Three faults concentrated in one station is not three complaints; it is one demonstration waiting to happen. The standard response is a two-minute method correction followed by an extra read at that station for the next 200 units, which costs about a quarter of an hour of attention. Skipping the attribution turns the same signal into a general workmanship complaint, which can only be answered with a full sort at USD 0.35-0.60 per piece and two wasted days.

Keeping the Measuring Tools Honest

The three cheap instruments named in the earlier steps are only useful while they remain instruments. A schedule is all it takes, and it fits on the same sheet as the readings: verify, date, and note anything found out. The expectations below are the ones worth writing into the tolerance set so a later disagreement has something to be checked against.

1. Calipers to gauge blocks weekly, accurate to 0.02 mm, with the check written on the shift sheet rather than remembered.
2. Steel rules only for body dimensions. A cloth tape stretches 1-2 mm per metre, which is enough to fail a good unit against a 3 mm tolerance.
3. Cutting dies resharpened after 3,000-5,000 strokes, and patterns re-cut after 800-1,200 pulls, before edge quality drifts into the panel dimension.
4. Zero jigs against the retained unit at the start of every shift, since a loosening jig is the origin of most 3-5 mm setting errors.
5. Colour under one source. Keep a D65 light box in the inspection area, so colour is not judged under whichever lamp happens to be lit.

Frequently Asked Questions

People Also Ask