Quality problems in cross-border sourcing rarely appear out of nowhere. In our experience, most of them can be traced back to a checkpoint that was skipped: a specification that was never written down, a sample that was approved by photo only, or an inspection booked after the goods were already sealed in cartons. Quality control is not a single event at the end of production. It is a chain of small, unglamorous checks, and the chain is only as strong as its earliest link. This checklist walks through the sequence we apply to orders placed with Chinese suppliers, from supplier verification to the moment the container doors close.

1. Before You Order: Verify the Supplier, Not Just the Price

A low quotation tells you almost nothing about whether the order will ship on spec. Before discussing price in detail, verify the basics:

• Business registration. Ask for the business license and check that the company name on it matches the name on the proforma invoice and the bank account you will pay. Mismatches are a red flag that deserves an explanation.
• Factory or trading company? Both can be legitimate partners, but you should know which one you are dealing with. A simple test: ask who will handle quality issues if the goods fail inspection, and how.
• Export experience. A supplier that regularly ships to your market will already understand its labeling, packaging, and documentation requirements. Ask which markets they export to and what documentation they usually provide.
• See the operation. A video call with a walkthrough of the workshop or warehouse reveals more than a polished company brochure. You are looking at scale, organisation, and whether the products on the floor resemble what is being quoted.
• Payment terms. Treat requests for full prepayment to an unverified supplier with caution. Staged payments — a deposit, with the balance tied to a passed inspection — are a common structure for good reason.

2. Samples: Lock the Standard Before Production Money Moves

Every dispute about finished goods eventually becomes a dispute about what was agreed. The sample stage is where you remove that ambiguity.

Approve a physical sample, not photographs. Then write down the specification the sample represents: materials and grades, dimensions and tolerances, colours referenced to a standard such as Pantone where colour matters, surface finish, function, and any parts or accessories included. Both sides keep a sealed reference sample — one at the factory, one on your side — so that "different from the sample" can be checked against an object rather than argued from memory.

Pay attention to what the sample does not show. A hand-finished sample proves the supplier can make the product; it does not prove the production line will make ten thousand of them the same way. That is what the later checkpoints are for.

3. During Production: The Mid-Production Check

Most buyers either inspect too early, when nothing is finished, or too late, when everything is packed. The useful moment is in between: once production is underway and the first finished units are coming off the line.

A mid-production check answers three questions. Are the actual raw materials and components the ones specified, or quiet substitutes? Do the first finished units match the sealed sample in dimensions, colour, and function? And is the production pace consistent with the promised delivery date? A systematic error — a wrong colour batch, a substituted component, a misprinted mould — caught at this stage costs a conversation. The same error caught after packing costs a shipment.

4. Before Shipment: Pre-Shipment Inspection and AQL

The pre-shipment inspection (PSI) is the checkpoint most buyers know by name, and it is usually performed using AQL sampling — the Acceptable Quality Limit method standardised in ISO 2859-1 (ANSI/ASQ Z1.4). Instead of opening every carton, the inspector draws a statistically defined sample based on lot size and inspection level, then classifies defects as critical, major, or minor.

Many importers of general consumer goods work with General Inspection Level II and acceptance limits of 0 for critical defects, 2.5 for major defects, and 4.0 for minor defects. Treat these as a widely used starting point, not a universal rule: the right limits depend on your product, your market, and how much risk a defective unit creates. Whatever limits you choose, put them in the purchase order before production starts, so the supplier produces against a known standard.

A thorough PSI covers more than defect counting: quantity verification, workmanship and function tests appropriate to the product, measurement checks against the spec sheet, barcode and label scanning, packaging integrity, and a carton drop test to simulate handling in transit.

5. Packaging and Labeling: Where Compliance Actually Fails

A surprising share of sourcing problems are not product problems at all — they are printing problems. Cartons held at destination because the country-of-origin marking was missing, retail packaging whose warning labels do not match the destination market's requirements, barcodes that do not scan.

Check artwork proofs character by character before printing, including the origin marking (goods entering the United States, for example, must be marked with their country of origin), any mandatory warnings for your product category, and net quantity and manufacturer details where required. If the goods are headed to Amazon FBA or a retailer, confirm their current packaging and prep requirements at the time of the order — these change, and last year's spec is not evidence. Finally, verify that what was approved on the proof is what appears on the physical packaging during the PSI, not just on the product.

6. The Pitfalls We See Most Often

• Specification drift. The sample is approved, then materials are quietly adjusted to protect the supplier's margin. Written specs and a mid-production check are the countermeasure.
• Subcontracting. Your order is passed to a smaller workshop you never vetted. Ask directly whether production stays in-house.
• The holiday rush. Orders squeezed in before the Chinese New Year shutdown are produced under time pressure and shipped with skeleton staffing on the other side of the holiday. Build the calendar into your lead time.
• Inspection booked too late. A PSI arranged when goods are already packed can only report; it cannot fix. Book inspections against the production schedule, not the shipping deadline.
• No loading check. Goods that passed inspection can still be damaged by poor container loading. For full-container orders, loading supervision is cheap insurance.

None of these steps is complicated. The discipline is in running all of them, every time, including on the repeat order with the supplier who has never let you down — because that is precisely the order nobody checks.

Planning a first order from China, or trying to fix a recurring quality problem with an existing supplier? This checklist is the framework we start from. If you would like a second pair of eyes on a specification, a supplier shortlist, or an inspection plan, you are welcome to reach out through our contact page — no obligation, and we will tell you plainly if we are not the right fit for the order.

← Back to Insights

For a consumer brand outside the United States, Amazon US looks deceptively simple: open a seller account, ship inventory, publish a listing. Brands that treat it that way usually discover the hidden sequence the expensive way — inventory produced before the trademark was filed, goods held at customs over a missing test report, a listing launched with no review strategy and no advertising structure behind it. The steps below are not difficult, but their order matters, and several of them have lead times you cannot compress once you discover you needed them. This is the sequence we walk brands through before their first shipment leaves.

Step 1: Validate the Market Before You Validate the Product

Start with demand, not with your catalogue. Look at what already sells in your category on Amazon US: the price bands where volume concentrates, the review counts and ratings of the top listings, and — most usefully — the complaints inside three- and four-star reviews. Recurring complaints about size, durability, instructions, or packaging are product briefs written by the market itself.

Then run the numbers on a spreadsheet, not on optimism. Product cost, freight to a US warehouse, import duties, Amazon's referral fee for your category, FBA fulfilment fees by size tier, storage, and a realistic advertising cost. Amazon publishes its fee schedules in Seller Central, and they are revised periodically, so pull the current figures rather than relying on anyone's blog post — including the margins you see quoted in courses. If the product only works at a price above the band where competitors sell, the market has answered your question. Believe it.

Step 2: File the Trademark Early, Then Enrol in Brand Registry

This is the step with the longest lead time, so it goes near the front. Amazon Brand Registry — which unlocks A+ Content, Brand Analytics, Sponsored Brands advertising, and the platform's brand-protection tools — requires a trademark that is registered, or at minimum pending, with the relevant trademark office; for the US store, that means the USPTO. Trademark examination takes months, not weeks, and it cannot be rushed after the fact.

Practical implications: file before you print packaging, choose a name you can actually clear (run a proper search first — a name that collides with an existing mark is a rebrand waiting to happen), and note that brand owners based outside the US are required to work through a US-licensed attorney for USPTO filings. While the application is pending, you can generally proceed with account setup and product preparation in parallel; what you cannot do is recover the months if you start the filing after the goods arrive.

Step 3: Compliance and Certifications: Category by Category

There is no single "Amazon certificate." Requirements stack up from three directions — US federal and state law, your product category, and Amazon's own policies — and they differ sharply by product:

• Electronics may require FCC authorisation for radio-frequency devices; anything with a battery adds transport and safety documentation.
• Children's products fall under CPSIA requirements, including third-party testing by a CPSC-accepted laboratory and a Children's Product Certificate.
• Cosmetics, supplements, and food-contact products intersect with FDA rules on registration, labeling, and permitted claims.
• Certain products sold into California need Proposition 65 warnings if they contain listed chemicals above threshold levels.

Two cautions. First, a supplier's existing test reports may be useful evidence, but reports must correspond to your exact product, model, and materials — a certificate for a similar item from the same factory is not transferable. Second, Amazon increasingly requests compliance documents during listing creation or after a complaint; finding out at that point that you need a test you never commissioned costs weeks. Budget testing into the product plan from the start.

Step 4: Freight, Customs, and Getting into FBA

Decide your shipping structure deliberately. Under DDP terms your supplier arranges everything to destination, which is convenient but leaves you dependent on their broker and their declarations — know whose name is on the customs entry, because compliance responsibility follows it. Many brands prefer to control the chain through their own freight forwarder and customs broker once volumes justify it.

Ship the first order smaller than your ambition. A modest first FBA shipment validates the whole pipeline — production, freight, customs clearance, FBA check-in, listing accuracy — at survivable cost if something is off. Confirm Amazon's current inbound requirements (box labels, carton weight and dimension limits, prep rules for your category) at ship time; they are updated periodically and non-compliant shipments get delayed or refused. Keep buffer stock logic simple at this stage: your real reorder point only becomes clear after the first sales cycle.

Step 5: The Listing and the Cold Start

A new listing has no reviews, no sales history, and no algorithmic trust. Plan the launch as a system, not a hope:

• Assets first. Main image that reads clearly at thumbnail size, supporting images or video that answer the questions reviewers of competitors keep asking, and A+ Content once Brand Registry is active.
• Copy that matches how buyers search. Title and bullets built around the terms customers actually use, in plain language, without keyword stuffing that makes the listing read badly.
• Reviews, compliantly. Amazon's Vine programme is the legitimate route to early reviews for enrolled brands. Incentivised reviews, review manipulation, and friends-and-family purchases violate policy and risk the account — there is no safe version of this shortcut.
• Advertising as data collection. Sponsored Products campaigns at a controlled test budget exist to learn which search terms convert. Expect the first weeks to buy information as much as sales, and price that into the launch plan.

Step 6: Cash Flow and Risk Control

The financial rhythm of Amazon surprises many first-time brands. Payouts arrive on Amazon's settlement cycle, not when the customer pays; referral and FBA fees are deducted along the way; and successful products create their own cash crunch, because the reorder must be financed while the previous cycle's revenue is still settling. Model the cash conversion cycle before you scale spend.

On risk: keep account health as a standing agenda item (respond to performance notifications promptly, keep compliance documents filed and current), avoid building the entire business on a single ASIN, and treat inventory as the risk it is — unsold stock in FBA accrues storage fees and is far cheaper to prevent than to liquidate. A brand that survives Amazon is usually not the one with the best launch week, but the one whose second and third orders were sized by data rather than by the launch week's excitement.

If you are weighing up a US Amazon launch — or your first attempt stalled at the trademark, the compliance documents, or the cold start — we are happy to look at where your product stands against this sequence and tell you honestly what we think the gaps are. You can reach us through our contact page.

← Back to Insights

Most writing about AI in customer service is either a vendor's promise or a sceptic's warning. Neither is much use to a trading company handling real inquiries across time zones, in three languages, about products with forty-page specification sheets. This article describes how AI support is actually deployed in day-to-day foreign trade work — the four jobs it does well, the boundaries we draw around it, and the data hygiene that has to come first. Nothing here requires exotic technology. The gains come from putting the AI in the right places and keeping it out of the wrong ones.

1. First Response: Speed, With a Human-Ready Summary

In export sales, the first response time shapes whether an inquiry becomes a conversation. An inquiry that arrives at 2 a.m. your time and waits nine hours for a reply has often been answered by two competitors in the meantime.

This is the safest and most valuable place to use AI: an immediate, honest first response that acknowledges the inquiry, confirms the key details back to the buyer, and asks for whatever is missing — product or link, quantity, destination port or country, target specification, and any deadline. While the buyer reads that, the inquiry is summarised and routed internally with the buyer's details, the product interest, and the unanswered questions flagged. The salesperson who picks it up in the morning starts from a briefing instead of a blank page. The rule we hold to: the automated first response never quotes prices or lead times. It buys speed without creating commitments a human has not reviewed.

2. Product Q&A: Answers Grounded in Your Own Documents

Buyers ask the same product questions hundreds of times: dimensions, materials, packaging options, minimum order quantities, certifications held, care instructions. AI handles this well if it is grounded in your own approved material — specification sheets, product manuals, and a maintained FAQ knowledge base — and instructed to answer only from those sources.

The grounding matters more than the model. An AI that improvises will eventually invent a dimension, a certification, or a compatibility claim, and in trade an invented specification becomes a claim dispute after shipment. Set the system up so that when the answer is not in the approved documents, it says so and escalates to a human rather than filling the gap. Keep the knowledge base versioned like any other controlled document: when a spec changes, the old answer has to change everywhere at once.

3. Multilingual Communication: Draft, Don't Send Blind

Trading companies routinely serve buyers in languages nobody on the team speaks natively. AI drafting has largely solved the first-draft problem: a reply drafted in the buyer's language, in a professional register, in seconds. It has not solved the verification problem.

Our working pattern is draft-and-review. The AI produces the reply in the target language; a human checks the parts that carry consequences — numbers, dates, specifications, payment and delivery terms, and any sentence that commits the company. Maintain a small glossary of your fixed terminology (product names, material terms, trade terms) so translations stay consistent across conversations and colleagues. Tone is the other thing to watch: a literally correct translation can still read as abrupt or evasive in the buyer's business culture, and the human reviewer is the one who carries that judgement.

4. Order Status and After-Sales Triage

A large share of routine messages are status questions: Has production finished? Has the vessel sailed? Where is the tracking number? These are lookups, not conversations. If the AI is connected to your order records, it can answer them accurately and instantly at any hour, attaching the relevant document — invoice, packing list, bill of lading copy — where appropriate.

The value is in the triage as much as the answer. Messages that signal a problem — a delay that affects the buyer's own commitments, damaged goods, a documentation error, dissatisfaction with quality — should not be handled at machine speed with machine wording. Route anything containing a complaint, a claim, or a change to an agreed order straight to a human, with the AI's role limited to summarising the thread so far. Fast, accurate answers on the routine 80 percent buy your team the time to be careful with the 20 percent that actually decides whether the customer reorders.

5. Where We Draw the Human Line

The dividing line we use is simple: the AI may inform, but it may not commit. Negotiation, price quotations, delivery promises, compensation, and anything involving fault or money belong to a person — not because the AI cannot produce plausible sentences, but because these are the moments where judgement, context, and accountability live. A buyer who discovers that a "promise" in your chat was generated by a system that had no authority to make it has learned something about your company you did not intend to teach.

Practically, this means configuring escalation triggers (price requests beyond a reference range, complaint language, legal or compliance questions), making sure the handover includes the full context so the buyer never has to repeat themselves, and telling buyers plainly when they are chatting with an automated assistant. Honesty about the automation is not a weakness; buyers object to being misled, not to being answered quickly.

6. Data and Privacy: The Unglamorous Prerequisite

AI support touches your most sensitive commercial material: buyer identities, pricing, order histories, product documents. Treat the plumbing with the same seriousness as the conversation quality.

Use business-grade accounts whose terms address how your data is handled and whether it is used for training; do not paste customer personal data, pricing tables, or unreleased product files into consumer tools with unclear retention. Limit which staff can connect which data sources, keep access logs where the platform offers them, and decide in advance how long conversation records are kept and where. Buyers in some markets — the EU most visibly — have their own data-protection expectations and regulations; being able to explain, in one paragraph, what happens to a customer's data in your support system is becoming part of looking like a professional supplier.

The trading companies getting real value from AI support are not the ones with the most automation. They are the ones that automated the repetitive layer carefully, kept humans visibly in charge of commitments, and can explain exactly how the system is set up. That is a much shorter distance to travel than the hype suggests — and it starts with one inquiry queue, not a transformation programme.

If you are considering AI support for your own trade business — or you have a tool in place that your team does not quite trust — we are glad to compare notes on what has worked in practice. You can reach us through our contact page.

← Back to Insights