Home › Field notes › Sourcing Bags from China: Twelve Signals That Should Stop a Deal

Modular Bag Pro · 2026-10-10 · 3,045 words · 14 min read

Most sourcing losses are visible on paper months before they appear in a shipment, which is why signal-reading beats price comparison as a qualification method. Twelve signals matter in practice, and none of them requires an investigator: a certificate whose scope excludes the product category, a unit price that cannot survive a material arithmetic check, a sample delivered in four days while technical questions go unanswered for two weeks, a schedule promise made without reference to material lead time, and banking details that do not match the contracting entity. Every one is neutral in isolation and diagnostic in combination. The gate that contains all of them is documentary, and it closes before a deposit moves on a 500-unit order. Each signal is diagnostic rather than accusatory, and a serious production partner answers all of them without strain. The exercise costs about a week. The positive pattern is stated too, because signal-reading works by contrast rather than by suspicion.

Why signal-reading beats price comparison

Price comparison is the default qualification method and the weakest one. Two quotations for a similar bag can differ by thirty per cent for entirely legitimate reasons: a different fabric class, a different hardware specification, a different inspection level, or a different payment structure. Comparing the bottom line without unpacking those inputs rewards whoever has been most optimistic about the specification, which is rarely the party a buyer wants to reward.

Signal-reading asks a different question. Instead of asking which number is lowest, it asks which number is explainable. A defensible quotation can be decomposed: material at a stated class, hardware at a stated brand and size, labour at a stated operation count, overhead, and margin. A quotation that resists decomposition is not necessarily wrong, but it is unverifiable, and unverifiable is the state in which surprises live.

The signals below are observable in any sourcing geography. This article frames them around China because that is where the overwhelming majority of bag production capacity sits, and because the volume of correspondence a buyer generates there makes the signals easier to read. Nothing in the list is a comment on any individual business; each entry is a check that a serious production partner in China will pass without difficulty and an unserious one will not.

The commercial frame is worth fixing early. Standard terms in this category are T/T 30/70 FOB Xiamen, with MOQ 500 per reference and a sampling stage of 6-10 working days extending to 12-15 for complex builds, a build window of 35-50 days, and release held at AQL 2.5. A counterparty whose terms deviate sharply from that frame is not automatically wrong, but the deviation is a signal worth an explanation.

Selection rule: Selection rule: Read signals before comparing prices, because a quotation that cannot be explained by material or process is describing a risk nobody has named yet, and no saving on 500 units justifies an unverified counterparty.

Documentation signals: certificates that do not survive a check

Certificates are the first thing offered and the last thing verified, which is exactly backwards. A certificate is a scoped statement: it names an issuing body, a certified entity, a scope of activity, and a validity period. Two of those four fields carry most of the information, and both are routinely skipped.

Scope is the first. An ISO 9001 certificate whose scope covers wholesale, consultancy, or a product category unrelated to bags does not describe the facility that will cut and sew the order. A BSCI audit report names a specific site and a specific date; if the named site is not where production is proposed, the report describes a different workplace. An OEKO-TEX Standard 100 certificate is issued against a defined material set; if the certificate names a generic fabric reference rather than the construction being quoted, it supports no claim about the unit that ships.

Validity is the second, and it is checked in a minute. Audit dates lapse, certificates are withdrawn, and an old PDF circulates long after it stops meaning anything. For chemical evidence under REACH (EC 1907/2006) and for market-specific regimes such as California Prop 65 and CPSIA, the report should also name the material batch tested, because a report covering a different batch supports nothing.

The response to a failed documentation check is not accusations. It is a request: ask for the certificate with scope visible, or for a report naming the material batch. A counterparty that produces it quickly has demonstrated the thing that matters. One that argues about why it is unnecessary has produced a stronger signal than the missing document did.

Quotation signals: what an implausible number looks like

An implausible quotation is usually not a lie; it is an unstated assumption. The unit price assumes a lighter fabric, a generic hardware set, a lower inspection level, or a larger order than the buyer discussed. The price is real for the product assumed and wrong for the product specified, and the gap surfaces as a change request once the deposit has been paid.

The arithmetic that exposes this is simple and takes fifteen minutes. Take the fabric area from the pattern, multiply by a realistic fabric price for the class specified, add hardware at the stated brand and size, add a labour estimate from the operation count, and compare the result with the quotation. A quotation materially below that floor is telling the buyer something about its assumptions, and the useful move is to ask which assumption differs rather than to accept the number.

Structure carries signal too. A quotation that itemises setup separately from unit price, states the sampling fee of USD 50-150 and confirms it is refundable against the order, and names tooling and screens within the USD 300-2,500 band is describing real costs. A quotation that folds everything into one unit figure is harder to compare and impossible to renegotiate fairly when the scope changes, because nobody can say what the original scope was.

Bottom line: Bottom line: Decompose any quotation sitting well below the others into fabric, hardware, labour and packing, and ask which line was reduced, because an unexplained gap in an FOB Xiamen price is always answered somewhere in the bill of materials.

Sampling signals: fast samples, slow answers

Sampling is the most informative stage of qualification, because it is the only stage where the counterparty has to do engineering rather than sales. Two signals matter, and they pull in opposite directions.

The first is speed without substance. A sample that arrives in three days is often a stock unit relabelled, not a build from the buyer's specification. Legitimate sampling for a conventional reference takes 6-10 working days, and 12-15 working days where a moulded part, a custom pull or a welded module enters the build. A sample delivered faster than that window is worth checking against the tech pack dimension by dimension, because the fastest way to win an order is to send something that looks right.

The second is the inverse pattern: a sample that arrives on schedule while technical questions go unanswered. Ask three specific questions at sample stage, about the reinforcement behind an anchor, the tolerance on a spacing repeat, and the coating system on the shell. A counterparty that answers with numbers, drawings or test references is describing an engineering function. One that answers with reassurance is describing a sales function, and the difference will surface again the first time something goes wrong.

Change behaviour during sampling is the third signal, and it is the one that predicts the rest of the relationship. Every first sample needs corrections; that is normal and not a signal. What matters is whether corrections are tracked, whether the revised sample carries a revision note, and whether the specification document is updated to match the physical change. A sample revised verbally, three times, with no document updated, is a preview of how the production run will be managed.

Our production team issues a revision note with every sample iteration for that reason, and buyers should expect one as a matter of course. It is not bureaucracy. It is the only artefact that lets a reorder eighteen months later match the unit that was approved.

Verdict: Verdict: Judge a counterparty on the second question rather than the first, and treat a fast sample followed by a slow technical answer as the clearest available signal about how a problem in the 35-50 day production window will be handled.

Capacity and schedule signals

Schedule claims are easy to make and hard to verify, and the gap between the two is where launch dates are lost. The standard build window for a bag programme is 35-50 days from deposit, and the number is dominated by material lead time, not by sewing. A counterparty quoting the bottom of that band without naming the fabric it has already booked is quoting an aspiration.

The diagnostic question is about dependencies rather than dates. Which materials are in stock, which are on order, and which have not been booked? A production partner in China with a booked fabric position can name the mill and the delivery week. One without it will describe the total build time as though cutting could begin tomorrow, because in principle it could, if the material were there.

Capacity claims deserve the same treatment. A plant running 7 production lines with 149 machines and 137 people across 4,950 m² of SGS-verified floor can describe its monthly output, and a serious counterparty will state where an order sits in that queue rather than promising it is never a problem. The question that produces a real answer is about the queue: what else is committed in this window, and where does this order sit?

Banking, entity and change-order signals

Money movement produces some of the cleanest signals available, because banking details are factual rather than persuasive. The check is simple: the account name should match the legal entity on the contract and on the business licence. A request to pay a personal account, an account in a third country, or an entity with a different name from the contracting party is a discrepancy that deserves a written explanation before any transfer.

Where the explanation is that a related entity handles collections, the resolution is equally simple: name the related entity in the contract, or obtain a written confirmation on letterhead identifying the relationship and the account. Legitimate arrangements survive this request easily. What it does is remove the ambiguity that makes a later dispute unresolvable, because a payment to an unnamed party is very hard to recover.

Change-order behaviour is the second signal in this group and it surfaces later, usually after the deposit. A change requested at that point is normal; the response to it is the signal. A counterparty that quotes the change in writing, with the cost and the schedule consequence, before performing it, is describing a controlled process. One that performs the change and presents the cost afterwards has taken the decision away from the buyer, and the same pattern will repeat at every subsequent decision.

Scope discipline closes this section. Change requests should be numbered, quoted, approved in writing and folded into the specification document. That habit costs almost nothing and prevents the most common late-stage dispute in the category, which is a disagreement about whether a change was requested, approved, or simply assumed.

Takeaway: Takeaway: Pay only to a bank account held in the name of the contracting entity, and treat any request to redirect a deposit or to split one transfer across accounts as a signal that ends the conversation before T/T 30/70 money leaves the buyer's bank.

Building a verification gate before the deposit

A verification gate is a short, fixed checklist that must close before money moves. Its value is not that it catches bad actors, which is rare. Its value is that it converts a relationship from trust-based to evidence-based at the one moment when the buyer still holds all the leverage.

The gate has five items and can be completed in a week. First, the entity check: contract name, business licence name and bank account name agree, or a written explanation exists. Second, the documentation check: the certificates offered carry a scope that covers the product and a validity date that has not passed. Third, the specification check: the tech pack is complete enough to inspect, with dimension callouts, material callouts and a named test method per claim. Fourth, the sample check: the approved unit is physically retained, signed, and paired with a revision note. Fifth, the commercial check: terms, incoterm, inspection level and payment structure are written down.

Sequencing matters. Entities and documents first, specification and sample second, commercial terms last. A buyer who negotiates price before the specification is closed is negotiating against an undefined product, and both parties will discover the definition later at greater cost.

Once the gate closes, the arrangement is ordinary. Thirty per cent moves on order to release material and book capacity, sampling runs 6-10 working days or 12-15 for complex builds, the build occupies 35-50 days, release waits on AQL 2.5 by ISO 2859-1, and the seventy per cent balance is paid against release after inspection passes. That structure is unremarkable, which is the point: unremarkable structure with documented evidence is what makes a first order boring, and boring is the outcome worth paying for.

What a good counterparty looks like from the buyer's side

It is worth closing on the positive pattern, because signal-reading works by contrast. A serious production partner is detectable early and the markers are behavioural rather than promotional.

First, it asks questions before quoting. A counterparty that requests the interface drawing, the dimension callouts and the intended duty cycle before producing a number is doing engineering. One that quotes from a photograph is guessing, and the guess will be corrected later at the buyer's expense. Second, it names its materials: mill, fabric class, hardware brand and size, and coating system, rather than describing them generically.

Third, it states what it cannot do. A counterparty that declines a tolerance it cannot hold, or a finish it cannot qualify, is providing information that is worth more than an optimistic acceptance. Fourth, it documents change: revision notes, numbered change requests, and an updated specification after every iteration. Fifth, it separates inspection from production, so that release depends on a measurement rather than on an assertion.

None of those markers is exotic, and none requires a lengthy audit to observe. They are visible in the first three weeks of correspondence, which is the cheapest possible place to see them. Buyers who read for them early spend the rest of the programme managing a schedule rather than managing a dispute.

Before the qualification email goes out, keep two documents open: the interface drawing standard the range must be built to, and the current modular range for reference constructions.

Spec rule: Spec rule: Build the verification gate before the deposit rather than after the deposit, and accept a counterparty only when entity records, certificate numbers, a measured sample at AQL 2.5 and two references all agree.