Engineering blog · 2026-05-22

Cross-jurisdiction officer match — the KYB signal nobody surfaces

A single director sitting on 248 companies across Norway, the UK, and a tax haven is a shell-game pattern any KYB analyst would flag — if they could see it. We built the data layer that lets them see it.

What it does

Given a person's name, return every company across every jurisdiction where they hold a director / officer role. Joining the Norwegian Brønnøysund officer register, the UK Companies House officer index, and the OpenSanctions PEP list gives a cross-country graph that legacy single-country providers can't produce.

Why it matters

Real-world shell-game patterns we've found in our own data:

• One Norwegian individual appears on 248 active company boards, mostly registered at the same shared office address
• A UK officer of a freight-forwarding LLC is the same person as a Norwegian director of three holding entities — the connection is invisible unless you join officers across registries
• Sanctioned individuals listed in OFAC/EU lists show up under slight name variations as directors of nominally unrelated EU companies

How the match is built

Fuzzy by design. A person named "John A. Smith" can appear as:

• John Smith in NO brreg
• SMITH, John Andrew in UK Companies House (uppercase, comma-separated)
• J. A. Smith in a sanctions PEP listing

We use multi-method fuzzy matching: exact-full, first-initial-plus-last, last-only-with-disambiguation-by-birth-year, hyphenated variants, and SequenceMatcher ratio as a final fallback. Every match has a confidence score. We never reject — we return the candidate list and let the customer set their threshold.

The endpoint

GET /v2/companies/{canonical_id}/global-officers

{
  "canonical_id": "...",
  "officers_with_overlap": [
    {
      "officer": "JANE SMITH",
      "also_directs": [
        { "orgnr": "923609016", "name": "EQUINOR ASA" },
        { "orgnr": "913913771", "name": "DNB BANK ASA" }
      ],
      "count": 12
    }
  ]
}

Or, MCP-side, the global_officer_match tool returns the same shape to Claude / Cursor / any agent.

What's still hard

Person dedup across jurisdictions is genuinely difficult. The same person might have:

• Different name transliterations (Russian Cyrillic → English ASCII variants)
• Different birth years recorded in different registries (typos)
• Listed as a representative entity in one registry and a natural person in another

Our current approach is conservative — we surface all candidates with confidence scores, not a single "this is definitely the same person" answer. Anything above 0.80 is high-confidence; 0.50-0.80 is "candidate, verify"; below 0.50 we still return but flag as low-trust.

Try it

curl -H "X-API-Key: sandbox_try_2026" \
  https://api.nordicdata.cloud/v2/companies/8b0a130f-e6e9-4ed6-97f3-8ebbe06dfac3/global-officers

Full docs at /docs. Real KYB customers — let's talk: [email protected].