A founder once sent me a PDF at eleven at night. A scanned business licence, red chop in the corner, registered capital in the millions, eighteen-character Unified Social Credit Code sitting at the top like a serial number on a safe. The message said: “Checked out. Wiring the deposit Monday. Anything I’m missing?”

Everything, basically.

The licence told him a legal entity existed in a specific district, that someone had filed paperwork, and that a named individual was the legal representative. It told him nothing about whether that entity owned a production line, whether that line had capacity in Q2, whether the person on WeChat had any authority to commit it, or whether the goods would clear customs in Rotterdam with the right certification. He had verified existence and called it diligence. Those are different activities, and the gap between them is where Western companies lose money in Asia.

I have been building and buying in China and Southeast Asia for twenty years. Consumer hardware with the Jean-Michel Jarre venture, where one product became a line of eight and every one of them had to come off somebody else’s factory floor. A location-based platform in mainland China that grew past 300,000 users on infrastructure and partnerships I did not own. Digital transformation for a private healthcare group opening sites across the mainland. Now a climate-tech company with a hardware supply chain. In all of it, I have never once been burned by a company that did not exist. I have been burned, repeatedly, by companies that existed perfectly and performed badly.

What the registry actually tells you, and what it does not

Start with the obvious, because skipping it is negligence. If you want to know how to check if a company is registered in mainland China, the primary source is the National Enterprise Credit Information Publicity System at gsxt.gov.cn, run by the State Administration for Market Regulation. Commercial aggregators like Qichacha and Tianyancha sit on top of the same filings and are easier to search, especially if your Chinese is weak. You are looking for the Unified Social Credit Code (eighteen characters, it replaced the old separate business, tax and organisation codes), the registered legal representative, the registered address, the business scope, the registration status, and any litigation or abnormal operations flags.

Three things in there matter more than most people realise. The first is business scope. A Chinese company may only legally operate within its filed scope. If the scope says trading and consulting and you are buying injection-moulded parts, you are not buying from a factory. You are buying from an intermediary, which may be fine, but you should know it before you negotiate price. The second is registered capital, which until recently was subscribed rather than paid in, meaning a company could file an impressive number without a cent in the bank. The 2024 Company Law amendment tightened this with a five-year payment requirement, but you should still read the figure as an intention, not a balance. The third is the legal representative’s name, because that name needs to match the signature and the chop on anything you sign, and it needs to match the beneficiary name on the bank account you pay.

That last point alone would have saved several companies I have advised. Payment instructions that route to an account in a different name, often an offshore Hong Kong trading entity, while your contract is with a mainland manufacturer, are not an administrative convenience. They are a structural problem. If the money does not go to the counterparty on the contract, you have no counterparty.

So do the registry check. It takes an afternoon and costs almost nothing. Then accept what you have: proof of existence, not proof of performance.

Reference triangulation, which means three calls and not three logos

Every factory deck I have ever received had a customer logo wall. Some of those relationships were real, some were a single sample order from 2017, and at least one I know of was a logo lifted from a trade show banner.

What I ask for instead is this: introduce me to two of your current customers in a market similar to mine, and one you lost. The third request is the one that matters. A supplier who can talk about a customer they lost, and describe why without blaming the customer, is a supplier with self-awareness and probably a quality system. A supplier who claims they have never lost a customer is either lying or has very few customers.

When I get the calls, I do not ask whether they are happy. Everyone says yes. I ask specific operational questions. How many times did your delivery date move in the last twelve months, and by how many days. What happened the last time you rejected a batch. Who picked up the phone when it went wrong, the sales contact or someone with authority. How long did it take to resolve, in days. If the reference cannot answer those in concrete terms, either the relationship is thin or the reference is staged.

And watch what the supplier does with the request. A serious partner will hesitate, then come back in a few days with one or two names and a reason why others declined. A supplier who produces three enthusiastic references within an hour has either coached them or invented them. I have seen both.

The trial order designed to fail safely

Here is the part most Western founders skip because it feels slow. It is the single highest-return thing you can do.

Do not ask for a free sample. Samples come out of the engineering room, built by the best technician in the building with unlimited time. They tell you the factory’s ceiling, not its floor. What you want to know is what the production line does on an average Tuesday.

So place a paid order, small enough that losing it entirely would annoy you rather than hurt you, but large enough to require the actual line rather than the sample bench. For most hardware that means a few hundred units, not five. Pay the standard deposit. Set a real delivery date. And then, roughly a third of the way through the run, introduce one deliberate, modest change. A packaging revision. A minor tolerance tightening. A label language addition.

What you are measuring is not whether they accommodate it. You are measuring the shape of the response. The good outcomes look like this: they come back within a day with a cost and a schedule impact, quantified, or they tell you clearly that the change is not possible mid-run and offer the next window. Both of those are a functioning organisation. The bad outcomes are silence, instant agreement with no cost and no timeline change, or agreement followed by delivery of the original spec and an explanation that the change was communicated too late.

Instant agreement is the one that fools people. It reads as service. It usually means nobody has read the spec carefully enough to know what the change costs.

While that order is running, test the other thing: call at nine in the evening local time. Not to be difficult, but because that is when problems actually surface in a factory running two shifts. If the boss or the production manager picks up, you have a direct line into the plant. If your contact is a salesperson who goes dark after six and reappears at nine the next morning with a polished answer, you are talking to a layer, and there is somebody behind that layer you have never met. In my hardware years, the single best predictor of a supplier relationship that would survive a crisis was whether I had the mobile number of someone who could walk to the line while we were talking. Ask them to video call from the machine. A partner who is actually producing your goods will do it. A partner who has quietly subcontracted the run to a cousin’s workshop in the next province will find a reason not to.

Payment terms are the real contract

I have signed memoranda of understanding in three languages. I have never once enforced one. In practice, cross-border litigation against a mid-sized Chinese or Vietnamese supplier is slow, expensive, and rarely worth the management time for anything under a very large number. Your contract is not the document. Your contract is the schedule of payments and what remains unpaid when something goes wrong.

The industry default of 30 percent deposit and 70 percent against the bill of lading copy is written for the supplier. It means your money is gone before anyone independent has looked at the goods. I push for a three-part structure: deposit on order, a milestone payment on passing a third-party inspection at the factory before shipment, and a final balance after delivery or after a short acceptance window. Even when I cannot get the last tranche, inserting a pre-shipment inspection gate between the deposit and the bulk of the money changes behaviour immediately. A third-party inspection costs a few hundred dollars. It is the cheapest insurance in the entire supply chain, and the way a supplier reacts to the proposal tells you a great deal before you have spent anything.

Pay in the counterparty’s name, to a bank in the jurisdiction where the company is registered, with the invoice matching the contract entity. If anyone asks you to split the payment across entities or route it through a personal account, stop and reopen the whole question.

The three red flags that have cost me money

The first is the factory visit that keeps moving. Dates shift, a trade fair comes up, the line is being retooled, meet us at the office in the city instead. I have accepted that excuse before and paid for it. The office is not the business. If you cannot stand on the floor and see your product being made, assume it is being made somewhere else by someone you have not vetted.

The second is the counterparty who agrees to every specification without a single objection. Real manufacturers push back. They tell you your tolerance is unachievable at your price, that your colour match will drift, that your lead time assumes a material that is on allocation. Frictionless agreement is not enthusiasm, it is an unread document, and the disagreement arrives later, at your cost.

The third is the mismatch between entities: contract with one company, invoice from another, bank account in a third name, export declaration in a fourth. Sometimes this is legitimate structuring through a trading arm. Often it is the sign that your actual manufacturer is a layer or two removed from anyone you have spoken to, which means that when you need to escalate, there is nobody to escalate to.

None of this is exotic. It is a registry check that takes an afternoon, two or three reference calls that take a week, a paid trial order that takes a month and costs a few thousand, and payment terms that you negotiate once and reuse forever. Set against a first production run, a certification failure at the border, or a season missed, that is nothing.

The licence proves the company exists. Everything that determines whether you get paid back is behavioural, and behaviour only shows up when you give someone a small, real, slightly inconvenient opportunity to disappoint you. Do that on purpose, early, while the cheque is still small.

I write from twenty years of building businesses between Europe and Asia. If your company is facing this, start a conversation.