- Separate three workflow states: candidate found, fields compared with official registration, and transaction identity reviewed with payment recipient check
- An English-name search returning no results does not prove a business doesn't exist; the absence is a task for the procurement team to request official Chinese registered name and Unified Social Credit Code
- Identity verification of a legal entity does not assess manufacturing capacity, product conformity, creditworthiness, or delivery ability, which require separate evidence
Free workspace
Turn your team’s expertise into content.
Record interviews, organize footage, and write with AI on a free trial of the MarketScale platform for qualifying companies. No demo required, no credit card.
An automated supplier-risk workflow can start with the wrong company. Before a procurement team asks what an AI system knows about a supplier, it needs to establish which legal entity the system is describing. A returned search result and a matching name are useful evidence, but they answer narrower questions than a completed identity check.
This distinction matters for buyers working with Chinese suppliers in particular. The English name on a quotation may be a brand, a translation, or a trading label rather than the identifier needed to locate an official registration record.
A recent test illustrates the gap. Researchers searched English names for 106 China-based trailer manufacturers drawn from a larger list of 2,589 China-based manufacturers sourced from NHTSA vPIC. This was not a representative survey of Chinese suppliers generally, but a narrow test of one category.
The searches returned at least one candidate for 61 of 106 inputs, or 57.5%. Among those 61 returns, the top result's English-name field matched the input for 47, or 77.0%. Across all 106 inputs, that same numerator produces 44.3%.
These percentages measure different things. The 57.5% is a search-return rate. The 77.0% is a name-field match rate among returned candidates. Neither is a verified-identity rate. The results do not establish a separate identity check for every candidate. Two same-platform runs produced identical counts, which supports repeatability within that test, not agreement across databases or over time.
A subset of 22 entries with both full names and brand or short names showed a similar pattern. Full-name searches returned candidates for 13; brand or short-name searches returned candidates for 20. More search results did not mean more correct identities. The data does not measure how often a brand search selected the wrong firm.
Keep three decisions separate
A practical supplier record can distinguish three states.
First, a candidate found: a search produced a possible company. Store the input, platform, query date, and candidate record. Do not silently turn a search hit into an approved supplier.
Second, fields compared: the buyer has compared the supplied Chinese registered name and 18-character Unified Social Credit Code with a registration source. Record the source and any unresolved differences. A matching English field alone is insufficient for this step.
Third, transaction identity reviewed: the buyer has checked how the registered entity relates to the seller named in the contract and the intended payment recipient. If another entity is involved, document the relationship and authority relevant to that transaction rather than guessing from similar names.
These are proposed workflow states, not outcomes measured by the study above. They help a reviewer see what has actually been checked and what is still missing. A database can support this distinction with separate fields instead of one ambiguous "verified" flag.
Treat an empty result as a task, not a verdict
If an English-name search returns nothing, ask the supplier for its Chinese registered name, business licence, and registration code. An empty result is not proof that a business does not exist. Conversely, a populated result is not proof that the person who contacted the buyer represents that company.
Consider a common scenario: a quotation uses a short English brand while the contract names a Chinese legal entity. The reviewer should preserve both names, record the source linking them, and check the transaction documents. Automatically replacing the contract name with the first search result would erase the very discrepancy that needs attention.
When information remains unresolved, that status should stay visible to the person approving the purchase. AI may help extract fields or surface candidates, but the approval record should still show the evidence used and the decision made.
Identity checks do not settle the whole purchase
Even a well-supported entity match does not establish manufacturing capacity, product conformity, creditworthiness, or an ability to deliver an order. Those require their own evidence. Separating search, comparison, and transaction review prevents confidence in one narrow step from spreading to questions it did not answer.
For procurement teams evaluating automation, a useful acceptance question is this: can the system show why this record belongs to the proposed counterparty, while leaving uncertain cases unresolved? A higher search-return count alone cannot answer it.
