J By Janson Wang, CEO of ASG Dropshipping · 26 min read
How to verify a china supplier for the first time and how to re-verify one you already use are two different jobs. This piece covers the second one only.
First-time verification asks whether the company is real. Re-verification before scaling asks a narrower question: on this SKU, at this site, on this route, what does the evidence say about a larger order?
Below is a Five-Signal Scale-Readiness Review. You define one test scope, name who owns each signal, then read five signals — capacity evidence, cycle time, batch quality, packaging revision conformity, and exception handling — as Green, Amber, Red or Unknown.
I’m Janson, CEO of ASG Dropshipping. We run warehouses in Shenzhen and Dongguan and work with Shopify sellers whose order volume is climbing past the point where their current supplier was configured. This review is the one I run before I let anyone move volume.
Quick Answers About Re-Verifying a China Supplier Before Scaling
Does a passed identity verification prove a supplier can handle more volume?
Is re-verification before a volume increase a recognized practice?
Can an approved sample predict bulk quality on a reorder?
What is the smallest unit this review works on?
Should a Red signal stop a volume increase by itself?
Key Takeaways
The review does not score a supplier. It produces one of three decisions: Hard Stop, Controlled Ramp, or Proceed.
Unknown is its own state. A supplier who cannot produce evidence has an evidence problem, which is not the same as a capacity problem.Every signal needs an owner. Factory, sourcing agent, fulfillment site, carrier and seller fail in different ways, and the fix is different for each.
This review works on physical, supplier-attributable performance. It does not cover returns caused by product design, carrier loss and damage after acceptance, listing errors on your side, or warehouse receiving compliance.
You already did some version of this. A business licence. Maybe a factory audit, maybe your agent’s. A sample that came back correct. A Trade Assurance line on Alibaba. Maybe less than that, if you found the supplier through a marketplace and the first orders simply arrived.
Either way, the same thing happened. You answered "is this supplier legit." You did not answer "does this supplier hold at a larger order."
The gap has a shape, and other people have described it. eCopy puts the timing plainly: "A bad supplier doesn’t usually announce itself early. The first few orders often go fine — it’s the fortieth order, or the one during a demand spike, that reveals a supplier who can’t actually sustain quality or shipping times at volume." The "fortieth order" there is eCopy’s figure of speech. Read it as timing, not as a trigger number.
Two published sources also name the moment to re-check. Maple Sourcing’s FAQ says re-verification makes sense "when your contact person or the ownership changes, or before you significantly increase order volume," and that "a verification is a snapshot, not a lifetime pass." Zignify Global Product Sourcing, writing on LinkedIn, lists "treating verification as a one-time event" as a failure mode, and suggests re-verifying at least once a year and before any major reorder or product change.
So the triggers are documented. What is usually missing is the instrument.
What the identity check confirmed What it does not tell you
The legal entity exists and is registered
Whether the person running production, or the ownership, has changed since the audit
A facility exists at the audited address
Which share of your SKU is made at that site and which share goes to subcontractors
They can produce a sample matching your specification
Whether a bulk lot, under schedule pressure, matches that sample
They fulfilled orders at your current volume
Whether cycle time, defect rate and packaging conformity hold at a larger volume band
A dispute window exists on the platform
Whether the supplier’s own exception path degrades at peak
Left column: what a verification report says. Right column: what the five signals below are for.
Most re-verification goes wrong before the first measurement. It goes wrong because "the supplier" is not one thing.
One supplier can hold one legal entity, two production sites, a subcontractor you have never been told about, a warehouse that is not the factory, and two shipping routes with different handover behaviour. Mix those together and you get an average that describes nothing.
So write the scope down first. One scope, one reading.
Scope field What you record Why it changes the reading
Legal entity The company on the invoice and the contract Ownership and contact changes are a documented re-verification trigger Production or fulfillment site The address where this SKU is actually made and packed A second site can behave nothing like the audited one SKU or product family The specific SKU, or a family that shares materials and process Defect and packaging behaviour do not transfer across unrelated SKUs Fulfilment route Carrier, service level, and the handover point Cycle time and exception types differ by route Baseline and test volume band Units or orders per week now, and the band you want to reach The whole review is one baseline-to-test volume comparison, not two moods Observation window Start and end dates, non-overlapping with the baseline window Overlapping windows double-count the same orders
Then name the owners. This is the part most re-verification skips, and skipping it sends every failure to the factory by default.
Signal Primary owner Contributing owners
D1 · Capacity evidence Factory or production site Sourcing agent (what was actually committed), seller (forecast quality) D2 · Cycle time Fulfilment site Factory (material readiness), carrier (acceptance scan), seller (order release timing), integration app (status write-back) D3 · Batch quality Factory Material supplier, subcontractor, inspection party, seller (specification clarity) D4 · Packaging revision Seller or brand owner (approving the revision) Packaging printer, fulfillment site (which stock is picked), agent (handoff record) D5 · Exception handling Split by measure — see section 7 Seller, warehouse, carrier, customer, payment provider
A signal with no named owner is a complaint. A signal with a named owner is a task.
Here is the whole instrument on one page.
Signal Only measures Unit of measurement Evidence you need
D1 · Capacity evidence Whether committed capacity, stated with evidence, covers your test band
Site × SKU × week
Committed production, bottleneck rate, shift plan, material lead time, subcontracting share
D2 · Cycle time Elapsed time on one clock, on normal orders only
Order, grouped by SKU × site × route
Two timestamps per order, plus a reason code on every exclusion
D3 · Batch quality Nonconformance on a matured cohort of a single production lot
SKU × production lot × matured cohort
Lot identifier on shipped units, complaint log, confirmed cause per case
D4 · Packaging revision Whether shipped packaging matches an approved, traceable revision
Packaging component × approved revision × market or channel × effective date
Revision register, written supplier confirmation, physical or photographic samples
D5 · Exception handling Supplier response separately from end-to-end customer remedy
Exception case
Escalation thread timestamps, named responsible person, closure record
Two scope notes before the signals. These five signals are platform-agnostic. What is Shopify-specific here is only where D2’s timestamps live; on WooCommerce or a headless setup the same events sit in your fulfillment log. And the review tells you when to act, not what your contract lets you do about it — payment terms, quality clauses and rework responsibility live in the PO, and a Red reading is only as enforceable as that document allows.
Four things this review does not measure. Naming them keeps the model honest.
Returns caused by product design. If the product is hard to use or fits badly, the supplier can execute correctly and returns still climb.Carrier loss and damage after acceptance. D2 stops at verified carrier acceptance. What happens downstream is a separate contract.Listing errors on your side. A wrong variant or a bad size chart produces complaints that read like defects.Warehouse receiving compliance. D4 measures what your customer unboxes. Whether the carton passes a receiving specification at a third-party warehouse is a separate check upstream of the customer.
I say those out loud because the last time our team built a model like this, we stretched it. We pushed "wrong size chart on the product page" into a packaging dimension, and it distorted both. Admitting the boundary works better than widening the box.
You are about to ask a supplier whether they can take more. They will say yes. That answer costs them nothing, so it tells you nothing.
D1 fixes that by separating three things that usually arrive fused together.
Answer quality — is the reply specific, or is it reassurance?Evidence sufficiency — can they show the numbers behind the reply?Actual capacity — what the evidence, once produced, says about your test band.
Those three fail differently. A supplier who answers vaguely and then produces a clean production schedule had an answer problem. A supplier who answers confidently and cannot produce anything has an evidence problem. Only the third case is a capacity problem.
Ask for the evidence, not the opinion. For your SKU, at the named site, across the test window:
Committed production already promised to other customers in that window
The bottleneck process and its rate — the machine, line or station that sets the ceiling
The shift and labor plan, including when a claimed extra shift actually started
Material lead time for the larger order, which is often the real constraint
Which share of the expanded order goes to subcontractors, who they are, and whether they have made this SKU before
Rework capacity, because a lot that fails inspection has to go somewhere
How peak season scheduling changes the answers above
Get the arithmetic right on "twice as large." Fictiv frames supplier capacity evaluation with a version of this question — whether a supplier can fill your current order on time and process a subsequent order twice as large. Their published band is to avoid suppliers under 60% or over 80% utilization, stated in 2023 and stated for injection molding specifically. Treat that as the shape of the question. Your category, and your year, may sit somewhere else entirely.
The arithmetic still transfers. If your next order is twice your current one, and it overlaps a current order that is still in production, the site is carrying something closer to three times your normal concurrent load. Ask which of those two orders moves if material arrives late.
Dragonway Global, a sourcing agency writing for Shopify sellers, phrases the SLA half of this well: "What’s your processing time SLA, and what happens when you miss it? — You want a number (24–72h) and a concrete remedy, not ‘we always try our best’." That 24–72h window is Dragonway’s own framing for the sellers they work with, not an industry norm. The portable part is the pairing. A number without a remedy is a wish.
The same guide notes when this question is premature: "When you don’t need one yet: under ~5–10 orders/day, or still testing many products." Their read, their thresholds. Below that kind of volume, D1 is mostly about establishing a baseline.
Reading D1. Green means the evidence exists and covers the test band. Amber means the evidence exists and is tight, or the subcontracting share moves under load.
Red means the evidence contradicts the commitment — a lead time quietly extended last month, a shift that started the day you asked. Unknown means they cannot produce the evidence.
Unknown is not Red. It is a different problem with a different fix, and the fix is usually a document request, not a supplier change.
Capacity headroom checks are not new. They are ordinary procurement practice, borrowed here for ecommerce reordering. What this piece adds is the place to run them: on a supplier you already use, before you raise the PO.
D2 is the cheapest signal to collect and the easiest one to corrupt. It corrupts when you mix clocks.
Look at the events available to you: paid, ready to ship, marked fulfilled in Shopify, tracking number created, carrier acceptance scanned. Five events, and sellers routinely average across three of them. The resulting number moves when nothing in the factory has changed.
Pick one start event and one end event, then keep them.
Clock element Use Do not use
Start Order released to the supplier and accepted by them Customer payment time, which includes your own release delay End Verified carrier acceptance — a scan, not a label Tracking number created, or “fulfilled” set in your admin Grouping SKU × site × route One blended store-wide average Comparison Two non-overlapping windows of similar length Trailing windows that share orders Statistic Median, plus P90 and P95 The mean alone, which one stalled order can move
A label is not a handover. Dragonway’s protocol says it directly: route live orders through the supplier and measure "processing hours, delivery days, tracking quality (google the tracking number yourself), defect/complaint rate." Many trackers show movement while the parcel sits on a dock. The scan is the evidence.
Separate write-back failure from physical delay. This is the trap that makes D2 look like a supplier problem when it is a systems problem.
In a public Shopify Community discussion, a participant wrote on 2026-05-30 that "the reason someone held it, split the shipment, changed the address, or overrode shipping is sitting in a staff note or nowhere at all." This is public forum context, not an ASG customer testimonial.
If your fulfillment app fails to write a status back, your dashboard shows a delay that never happened in the factory. Check a sample of slow orders against the supplier’s own record before you attribute anything.
Exclude with a reason code, never silently. Customs hold, address correction, stockout, recall, remake, seller-side hold. Each excluded order needs a code, and each physical exception moves to D5 rather than disappearing. Silent exclusion is how a signal gets tuned to the answer you wanted.
A prerequisite, stated plainly. If your supplier confirms shipment by chat message and you mark orders fulfilled by hand, you do not have a start timestamp. Build that first. Until then, D2 has no clock, and a review that reports Unknown is more useful than one that reports a number you made up.
Reading D2. Compare the current window against the baseline window, and both against what your product page promises. A widening median with a widening P95 is a capacity story. A stable median with a fat P95 is an exception-handling story, which belongs to D5. The size of gap you tolerate depends on your category, your price point and your delivery promise — set it yourself, and write it down before you look at the data.
Shopify’s own supplier management documentation, as of 2026-09-02, tells sellers to "monitor supplier reliability, quality, and delivery times" and to "provide suppliers with advance notice for large orders or seasonal requirements." That is the platform saying monitoring is your job. Shopify names no threshold, and neither will I on their behalf.
D3 is where a supplier drifts quietly. Your first-order sample stays perfect on the shelf while the third reorder degrades.
Aqualora names the mechanism: "Your approved production sample was correct. That does not mean the bulk production batch will match it. Factories sometimes substitute materials, change suppliers, or adjust production parameters between sample approval and bulk production…" The ellipsis is theirs. Zignify puts the same idea in one line: "Samples are produced under attention. Mass production is produced under pressure."
First, fix the vocabulary. Three words get used interchangeably here and they are not the same object.
Production lot — one manufacturing run of one SKU, under one set of conditions. This is the thing that varies.PO — one order you placed. A single PO can span several lots if the supplier splits production.Shipment — one consignment leaving the site. A shipment can carry units from more than one lot.
D3 measures the lot . That requires a lot identifier travelling with shipped units, on the carton or in the packing list. Without it, you can measure that complaints rose. You cannot measure which run caused it.
Second, separate two rates that look alike.
Observed complaint or return rate — what customers reported, on a defined cohort.Confirmed supplier-attributable nonconformance — cases where a cause was established and it sits with the supplier.
The first is easy to collect and includes buyer’s remorse, damage in transit and listing mismatch. The second is the one D3 acts on. Reporting the first as if it were the second is the most common way these reviews mislead.
Third, wait for the cohort to mature. A lot shipped this week has barely entered its return window. Comparing it against a lot from three months ago compares an infant to an adult. Define a maturity date — a fixed number of days after delivery that fits your category’s return behaviour — and only compare cohorts past it. Set a minimum sample too. Below it, the honest reading is Unknown.
Fourth, grade severity. A cosmetic blemish and a failed safety-relevant component are not the same event at different frequencies. Separate them, because some of them have their own rule: a single confirmed safety, regulatory, unauthorized material or label event — or traceability broken badly enough that you cannot say what shipped — is a Hard Stop on its own , regardless of rate.
On inspection cadence. TradeAider, a company that sells inspection services, publishes a tiered cadence tied to defect rate.
In their words, you can skip inspections "only after a supplier has completed six orders at under 2% defect rate, and even then only on every second order — never two orders in a row," with every-third-order cadence requiring "twenty completed orders at under 1% defect rate plus stable supplier management," and "a single failed inspection resets you to every-order cadence for the next three shipments."
Three boundaries on that. The numbers arrive as a set — six with 2%, twenty with 1%, plus the reset — so lifting the 2% out on its own breaks it.
The vendor sells more inspection, which is a commercial interest sitting on the same side as the recommendation.
TradeAider also describes this cadence as mapping onto ISO 2859-1:2026 switching rules for reduced inspection, which they characterize as five consecutive lots passing at normal inspection. I have not opened the standard — it sits behind a paywall — so that reaches you as TradeAider’s description of their own commercial cadence. This article has not verified that the six-order and three-shipment counts are equivalent to anything in the standard.
Aqualora adds a discipline for suppliers you reorder from more than twice a year: "maintain a running quality scorecard. Track inspection pass rates, defect categories, and corrective action response times. Share this data with the factory." Sharing it is the operative part. It tells the supplier you are watching a trend rather than a single lot.
For the sampling arithmetic underneath all of this — how many units to inspect at which AQL level for a given lot size — we covered that separately in how to cut returns without slowing fulfillment . This section is about which lots to look at. That one is about how deep to look.
Reading D3. Green means matured cohorts are stable and confirmed supplier-attributable cases are flat. Amber means a movement you can trace to a known variable — a new material lot, a seasonal change, a subcontractor switch you were told about. Red means confirmed supplier-attributable nonconformance has moved on a matured cohort and nobody can explain it. Unknown means the lot identifier is missing, or the cohort is too young or too small.
D4 gets skipped until it bites, usually as two customers holding visibly different boxes for the same SKU.
Reframe what it measures. D4 is not a vote on whether packaging looks consistent. It asks one question: did the packaging that shipped match an approved revision, and can you prove which one? An unapproved or untraceable version is the failure. Two versions in circulation during a planned, documented transition is not.
The unit is packaging component × approved revision × market or channel × effective date. A carton, an insert, a label and a polybag can each sit on different revisions at the same time. A compliance label can differ by market by design.
Two kinds of revision, two levels of tolerance.
Regulatory and identity revisions — ingredient panels, safety marks, country of origin, barcodes, importer details. Getting these wrong is a compliance event, and it can block a market.Aesthetic and commercial revisions — a new logo lockup, a seasonal colour, a revised insert. These matter to brand experience and rarely to legality.
Mixing those two into one alarm produces noise. Separate them and D4 becomes usable.
Manage the transition rather than banning overlap. RuntoDropship, in Private label dropshipping , describes the discipline: "When you update your packaging design — new logo, revised color, seasonal variant — the version transition is managed deliberately. Old stock is depleted or segregated before the new version enters active use. No mixed-version orders reach customers during the transition window."
The control point is the artwork handoff.
SKUWorks documents a procedure used by teams who manage this well:
Create the new revision under a version identifier such as REV05.
Log the reason for the change, such as an ingredient panel updated per compliance review.
Obtain compliance and ops approval again.
Send a fresh handoff email stating that REV05 supersedes REV04.
Ask the supplier to confirm in writing that only REV05 will be used.
Archive the exact pack that was sent under a "Supplied to Vendor" folder.
SKUWorks estimates the added step costs roughly fifteen minutes and can save thousands in reprints — their estimate, offered without a method, so treat it as illustrative.
In my experience the archive step is where drift starts. A corrected file goes to a supplier contact over chat. That contact leaves. The next person reaches into a shared folder and pulls whatever is there. The correction existed. It was never stored where the next person would look.
Zhypacking ties D4 back to D3: "Inspection paperwork becomes the reorder baseline. If results live only in chat screenshots, the next PO will re-argue the same defects. … Buyers who keep this file cut reorder drift more than buyers who only keep unit price history. … Link to next PO — Attach the form when reordering so the factory sees the frozen standard."
If the only record of last quarter’s registration issue is a screenshot in a group chat, the next lot repeats it.
Reading D4. Green means shipped packaging matches an approved revision and the register documents that match. Amber means an overlap inside a documented transition window, with dates. Red means an unapproved revision shipped, or a regulatory or identity element is wrong. Unknown means you cannot say which revision shipped last week — which, for a brand, is close enough to Red to act on.
D5 gets confused with D2 constantly, so start with the split. D2 measures normal orders. D5 measures the orders where something went wrong. Different pile, different clock.
Then split D5 itself, because it contains two measures with different owners.
Supplier response — the supplier owns this.
Time to acknowledgement by a named responsible person, not an autoresponder
Containment: what stopped the same failure reaching the next shipment
A corrective plan with a stated cause
A committed completion date, and whether it was met
Customer remedy — end-to-end, shared ownership.
Time until the customer has the correct item, or the refund has cleared
Touched by you, your warehouse, the carrier, the customer’s own responsiveness and the payment provider
Averaging those two together produces a number that blames the factory for a payment provider’s settlement window. Measure them apart. Only the first one is a supplier signal.
What counts as an exception. Carrier issue, wrong item shipped, missing item, damage in transit, a customer requesting an exchange, or any order excluded from D2 with a physical reason code. What does not count: a customer changing their mind, a listing error on your side, or a payment dispute.
The best diagnostic question I have seen for the supplier half comes from Dragonway: "What happens in Q4? — You want to hear a straight answer about peak capacity, staffing, and which of their SLAs loosen. Anyone claiming Q4 changes nothing is lying."
That last sentence is Dragonway’s, quoted verbatim, and the absolutism is theirs rather than mine. The useful part is the test underneath it. A supplier who has thought about peak can tell you which promises stretch and by how much.
China Sourcing Pro adds a related point for fragile categories: "If you sell fragile products, ask your fulfillment partner for their damage rate data and packaging protocol before committing. … Surprise traffic spikes without preparation always end badly." Again, "always" is theirs. What I will say is narrower: the surprise is the failure. Spikes are forecastable if someone is looking.
Reading D5. Look at the distribution, not the average. One long resolution buried among fast ones usually marks a category of problem with no owner.
Green means named acknowledgement, containment and a met completion date across your exception set. Amber means acknowledgement is fast but closure drifts, which is an authority problem rather than an attention problem.
Red means a category of exception with no documented path at all. Unknown means too few exceptions in the window to read — which is good news about the period, not evidence about the supplier.
You will not see five Greens. You will see a mixed picture, and the question is what to do on Monday.
There is no score here, and that is deliberate. Multiplying frequency by loss by risk by severity produces the shape of a decision without its substance. The inputs are not calibrated to a common unit, so the product is judgment wearing a decimal point.
There is also no vote count. "One Red is a signal, two Reds are a decision" is arithmetic pretending to be reasoning. Two low-confidence Ambers on small samples are not equal to one confirmed safety failure. Signals are read on their merits, not tallied.
Each signal gets one of four states.
State What it means What it is not
Green Evidence exists, is in scope, and supports the test band A promise that nothing will go wrong Amber Movement you can trace to a named, bounded cause A softer Red Red Evidence contradicts the commitment, and nobody can explain it A verdict on the company Unknown The evidence needed does not exist or is not mature A pass, and not a fail either
Then the scope produces one of three decisions.
Decision When it applies What you do next
Hard Stop Any confirmed safety, regulatory, unauthorized material or label event; or traceability broken badly enough that you cannot say what shipped
Hold the volume increase. Contain the affected lots. These stop a ramp alone, without a second Red
Controlled Ramp Reds or Ambers that are explained, bounded and owned; or `Unknown` on a signal you can close with evidence
Raise volume in one bounded step, not to target. Hold the rest of the increase, split a bounded next PO with a verified backup where you have one, and raise inspection cadence for the affected SKU
Proceed Signals in scope are Green, and the `Unknown` count is zero on anything load-bearing
Move to the test band and re-read at the end of the next observation window
Controlled Ramp is the normal outcome. Most suppliers are neither disqualified nor ready.
Two cautions about correlation. When D1 and D3 move together, that is worth investigating — not concluding. First, movement together is not cause. The candidate explanations include labor changes, overtime, a shift in subcontracting share, a new material lot, equipment condition and a change in inspection practice. Correlated movement tells you where to look; it does not establish that one caused the other. Second, the relationship does not run backwards on command. Reducing load does not automatically restore quality.
The cadence rule inside D3 is not the same as this section. TradeAider’s reset is a within-signal action: a failed inspection tightens inspection. This section handles the across-signal question — what to do when D1 and D5 both move at once.
Shopify’s guidance closes a gap that sits outside every signal above: "Don’t rely on a single supplier for critical products" (per Shopify Help Center, as of 2026-09-02). If your review comes back Red and you have no alternative, the constraint you are facing was created earlier.
Repair or replace is a different decision, and it needs a cost you have not priced yet. A switch resets the instruments above to their pre-verification state: a new sample approval cycle, a new packaging revision issue, a new baseline window before D2 or D3 mean anything. That is weeks of blindness, and it belongs in the comparison.
If you want that comparison run against your own records rather than in the abstract, that is the conversation we have with sellers most weeks. Talk to the ASG team about your SKUs and sites .
The skeptic’s argument deserves its strongest form.
"The sample-versus-bulk observation is not new, and you did not find it." Correct. Zignify and Aqualora are two independent companies on two different sites saying versions of it in public. The observation is not the contribution. The synthesis is: the scope card, the owner map, the clock discipline, the four-state reading and the decision frame, assembled into one review.
"Capacity headroom checks are standard procurement. You did not invent them." Also correct. Fictiv writes about capacity as an ordinary input to supplier evaluation, and industrial buyers have run versions of this for a long time. What is different here is the object: a supplier you already use, re-read before a specific volume step.
"Most of your measurement methods come from a handful of vendors." Fair, and worth naming.
Dragonway’s guide supplies the live-order routing method and the SLA question. TradeAider supplies the inspection cadence. Aqualora supplies the scorecard. RuntoDropship, SKUWorks and Zhypacking supply the packaging revision discipline.
The component methods are theirs, cited at the point where each is used. What this piece contributes is the Five-Signal synthesis — the scope card, the owner map, the clock discipline, the four-state reading and the decision frame — which decides which of those methods gets run, on which scope, and what the reading means afterwards.
"A professional inspection service does this better." For batch quality on complex or high-risk SKUs, often yes. If you have the budget and the SKU justifies it, hire one.
The argument here is narrower: D1, D2 and D3 run on your scoped order records together with capacity and revision evidence you request from the supplier and the site, and where that instrumentation does not exist yet the signal is recorded as Unknown rather than skipped. Starting is what usually does not happen.
"Two suppliers running in parallel is simpler than one review." This is the strongest version, and it is a real alternative. Splitting volume across two suppliers tests them with live business rather than with a study. It costs more per unit, fragments your packaging revision control, and doubles the number of scopes you have to maintain. For sellers who can absorb that, it is a reasonable substitute for parts of this review.
Where "hire a service" stops working is D5. An inspection service measures goods, not an organization’s response behavior under pressure. That signal only exists in your own escalation records.
One last thing, since this piece has been asking you to check other people’s numbers. The 60%/80% band is from 2023 and stated for injection molding. The 24–72h SLA is one vendor’s own framing. The 2% and 1% inspection thresholds come from a company that sells inspection. If I ask you to hold a supplier to a standard of evidence, the same standard applies here.
What should I collect before placing a materially larger PO?
Six things, and they take a week rather than a quarter. Your scope card. Two non-overlapping windows of order timestamps on one clock. Lot identifiers on recent shipments. A packaging revision register with dates. Your exception cases with all the timestamps both D5 clocks need — supplier response, and customer remedy end to end — rather than a fixed pair per case. And the supplier’s own capacity evidence for the test window. Missing items become Unknown, not assumptions.
What can a factory audit not prove?
An audit establishes that a site, a system and a set of documents exist on the day of the visit. It does not establish which share of your SKU is made there, how the site behaves at a larger order, whether packaging revisions are controlled between reorders, or how the organization responds when a shipment goes wrong. Those are behaviors over time, and they need records rather than a visit.
My supplier does not put batch or lot IDs on shipments. What now?
D3 reads Unknown until that changes, and no amount of complaint data fixes it. Ask for a lot identifier on the carton and in the packing list, applied from the next production run. RuntoDropship’s pre-volume checklist in Dropshipping challenges covers the same category of agreement — SKU and variant identity, specifications, packaging requirements, and who owns defective units — and settling those before a volume step is cheaper than arguing them after one.
How do I tell a factory delay from an integration write-back delay?
Take a sample of the slowest orders and compare three records: your admin timestamps, your fulfillment app’s log, and the supplier’s own dispatch record. If the supplier’s record shows dispatch on time and your admin shows a gap, the delay is in the write-back and belongs to your systems. If both agree, the delay is physical. Doing this on a sample of ten orders usually settles the argument.
When should I pause a planned volume increase?
Pause on a confirmed safety, regulatory or unauthorized material or label event, or when traceability is broken badly enough that you cannot say what shipped. Those justify a Hard Stop on their own. For everything else, a Controlled Ramp — one bounded step, raised inspection cadence, and a re-read at the end of the next window — usually gets you better information than either freezing or committing.
How to verify a china supplier you have never used and how to re-verify one you already rely on are different jobs with different instruments. This piece was about the second.
The measurements are ordinary. Two timestamps on one clock. Defects grouped by production lot on a matured cohort. A revision register with dates. Two clocks on your exception cases. What makes them work is the framework around them: the scope card, the owner map, the clock discipline, the four-state reading and the decision frame.
Run it before you raise the PO. A Red found in advance costs a delayed ramp. The same Red found afterwards costs a quarter.
If you want a second set of eyes on the scope card before you run it, start a conversation with our team .
Maple Sourcing — Supplier Verification in China (accessed 2026-09-02)
Zignify Global Product Sourcing — How to verify a Chinese manufacturer before you place an order (published 2026-05-14, accessed 2026-09-02)
eCopy — How to find suppliers for Shopify (published 2026-04-02, accessed 2026-09-02)
RuntoDropship — Dropshipping challenges (published 2026-08-21, accessed 2026-09-02)
RuntoDropship — Private label dropshipping (published 2026-03-13, accessed 2026-09-02; same publisher as the entry above, so the two are one source)
TradeAider — How Often to Inspect China Suppliers: A Risk-Based Frequency Framework (published 2026-04-24, accessed 2026-09-02; vendor sells inspection services)
Aqualora — Quality control for Chinese manufacturing inspections (published 2026-04-01, accessed 2026-09-02)
Dragonway Global — China sourcing agent for Shopify (published 2026-07-03, accessed 2026-09-02)
China Sourcing Pro — China-based fulfillment for Shopify dropshipping (published 2026-02-09, accessed 2026-09-02)
SKUWorks — How to Avoid Sending the Wrong Artwork Version to a Supplier (published 2026-04-10, accessed 2026-09-02)
Zhypacking — Pre-shipment inspection for paper packaging (published 2026-07-04, accessed 2026-09-02)
Shopify Help Center — Managing suppliers (as of 2026-09-02)
Fictiv — Five key factors to consider when performing a supplier evaluation (published 2023-06-28, accessed 2026-09-02; the 60%/80% band is stated for injection molding)
Shopify Community — Your Shopify store is growing, why does everything feel harder (thread started 2026-05-28, accessed 2026-09-02; quoted line is post 5, 2026-05-30)
The ASG context in this article — Janson’s operator perspective and the company’s warehouse work across Shenzhen and Dongguan — comes from internal company records. No ASG performance statistic, client outcome or before-and-after result is claimed anywhere in this piece. This article is not tax, legal or compliance advice.
