Entity resolution: telling same-name companies apart

Why names are not identifiers

Company names are not unique across jurisdictions, and in several they are not unique within a jurisdiction either. Japan removed its similar-trade-name (類似商号) restriction in 2006, so two unrelated 株式会社 can share a name at different addresses. US companies are registered state by state, so an Acme Technologies Inc. in Delaware and another in Texas are different legal persons. Trading names, brand names, former names and group names add further noise: a brand on a website may belong to a parent, a subsidiary or a licensee.

The practical rule is that a name is a search key, never a match. A record is attached to a company only when a registration number, or a combination of address, officer names and dates, confirms it. OpenDossier records the basis of every match in the entity match field of the evidence chain so that a reader can see whether the link rests on a number or on circumstantial agreement.

When a match cannot be confirmed, the record is held as a candidate and listed under open questions rather than silently included or dropped.

Registration numbers by jurisdiction

These identifiers are the primary keys for a dossier. Check formats with the issuing authority, since they change.

Hong Kong adopted a Unique Business Identifier in December 2023: the Business Registration Number now identifies companies on the Companies Registry portal, alongside the older company number.

CJK names and romanisation

Chinese, Japanese and Korean company names create three distinct problems.

Script variants. Traditional and simplified characters (臺 / 台, 國 / 国) and Japanese shinjitai forms mean the same name can appear in several spellings across registries, courts and the press. Search both forms and record which form the authority uses.

Romanisation systems. A Taiwanese company may be romanised in Wade-Giles, Hanyu Pinyin or Tongyong Pinyin depending on the document and its date; a Japanese name may follow Hepburn or Kunrei-shiki. Hsieh, Xie and Shay can be the same surname. Treat romanised forms as aliases and resolve through the original script where possible.

Unregistered English names. In Taiwan the legal name on the MOEA register is the Chinese name; an English name is a trading name, often registered separately for import and export purposes, and may be shared or changed freely. In Japan the registered 商号 may be in kanji, kana or Latin letters, and the English name a company uses abroad is frequently not the registered one. A dossier should list the registered-script name as primary and every romanised or English form as an alias, each with the source where it was seen.

Common mis-matches and how they happen

• Parent and subsidiary. A SOC 2 report issued to Example Holdings Pte. Ltd. does not cover Example Technologies KK unless the report scope says so. Read the scope section, not the cover.
• Former names. A 2021 judgment against a company's former name will not surface under the current name unless the search includes name history. Companies House, ACRA and ASIC expose previous names; use them.
• Same name, different state. US litigation searches by name alone pull in unrelated companies in other states. Confirm with the state of incorporation and officer names from the complaint.
• Brand-level press. News often names a product or group brand. Map the brand to a legal entity before attaching the story.
• Dormant shells with the same name. A struck-off company with the same name in the same jurisdiction is a classic false positive in insolvency searches.
• Transliterated sanctions hits. Sanctions lists carry many alias spellings; a fuzzy hit on a common Chinese or Arabic name needs date of birth, nationality or an identifier before it is reported as a match.

Each of these is avoidable when the record is tied to a number or an address plus officer, and each is reported in the dossier with the basis of the match or the reason it was rejected.

A resolution procedure that holds up

OpenDossier applies the following order of evidence for each candidate record, and the chain states which step produced the match.

1. Exact identifier match (UEN, company number, 法人番号, CIK, ACN, NZBN, LEI, BRN). This is conclusive.
2. Cross-identifier match through a document that carries two identifiers, for example a Form D that names a Delaware corporation, or a Japanese registry extract that shows the 法人番号 alongside the 商号.
3. Registered address plus incorporation date plus at least one officer name agreeing across the two sources.
4. Name plus jurisdiction plus a distinctive business description, held as probable and listed as an open question until a stronger link is found.

Anything weaker than step 4 is excluded and noted. The procedure is also run in reverse when a company claims its profile: the claimant must prove control of a domain or an officer email that the registry already associates with the entity, so that a claimed profile cannot be attached to a company by a stranger.

What to ask the company

Where public identifiers do not settle a match, the fastest route is a short list of questions to the counterparty, logged in the Trust Workspace so that answers are dated and attributable.

• The full registered name in the registered script and every English or romanised form in use, with the documents where each appears.
• All registration numbers held, including tax numbers used on invoices and any LEI.
• A group chart naming each legal entity, its jurisdiction and number, and which entity will sign the contract.
• Former names and dates of change.
• Which entity owns the website domain and the trademarks used in marketing.

These answers are recorded as company-stated claims. Where they agree with the registry they lift the match to verified; where they disagree, the contradiction is shown and the company is asked to explain. A counterparty that cannot say which of its entities will contract with you has told you something useful before diligence has begun.