Graph Databases for Beneficial Ownership: Mapping UBO Networks with Neo4j

Shell companies, nominee directors, circular ownership chains. The corporate structures designed to hide Ultimate Beneficial Owners are inherently graph problems. Relational databases struggle with them. Neo4j makes them tractable.

The Compliance Requirement

Know Your Business (KYB) regulations require financial institutions to identify the natural persons who ultimately own or control a corporate customer. FATF guidance sets the threshold at 25% ownership for most jurisdictions, though some countries use 10% or 15%. The challenge is that ownership is rarely straightforward.

A company might be owned by another company, which is owned by a trust, which is controlled by a foundation, which has a board appointed by someone who turns out to be a politically exposed person. Each layer adds opacity. Each jurisdiction has different disclosure rules. And the people who design these structures are specifically trying to make the ownership trail hard to follow.

Why Relational Databases Fail Here

On paper, you can model ownership in a relational database. You create an entities table, an ownership table with foreign keys, and write recursive CTEs to traverse the chain. We tried this first. It works for shallow structures (two or three layers) but breaks down in practice for several reasons.

Recursive CTEs become slow as depth increases. At four or five layers of ownership, query times grow to seconds. At seven or eight (not uncommon in complex structures), they become impractical. The optimizer struggles because it cannot predict the recursion depth in advance.

Worse, real ownership structures are not trees. They are graphs with cycles. Company A owns 30% of Company B, which owns 20% of Company C, which owns 15% of Company A. Recursive CTEs need explicit cycle detection to avoid infinite loops, adding complexity and further degrading performance.

Finally, relational databases have no native concept of path analysis. Questions like "what is the shortest ownership path between Person X and Company Y" or "are there any circular structures involving this entity" require complex application-level logic. In a graph database, these are single-line queries.

WITH RECURSIVE ownership_chain AS (
  SELECT parent_id, child_id, ownership_pct, 1 AS depth
  FROM corporate_ownership
  WHERE child_id = 'accra_holdings'

  UNION ALL

  SELECT co.parent_id, co.child_id,
    co.ownership_pct * oc.ownership_pct / 100,
    oc.depth + 1
  FROM corporate_ownership co
  JOIN ownership_chain oc ON co.child_id = oc.parent_id
  WHERE oc.depth < 10
)
SELECT * FROM ownership_chain
JOIN entities ON entities.id = ownership_chain.parent_id
WHERE ownership_pct > 25;

-- 23 lines, hard to extend, slow at depth > 5

The Neo4j Data Model

We modeled the ownership graph with four node types and five relationship types. The schema is deliberately simple because the power comes from traversal, not from entity attributes.

Detecting Hidden UBOs: A Real Example

The interactive diagram below walks through a real (anonymized) case from our KYB screening. At the surface, a Ghanaian company appeared to have a straightforward ownership structure. Each layer of investigation revealed additional complexity, until graph traversal identified the UBO as a Tier 1 PEP.

Without graph traversal, this case would have stopped at Layer 1. The BVI company would have been noted, perhaps flagged for manual review, but the connection through a Mauritian trust to a politically exposed person would likely have been missed. The traditional KYB process simply does not traverse deeply enough.

Detecting Circular Ownership

Circular ownership is a red flag in compliance. If Company A owns Company B, which owns Company C, which in turn owns Company A, the structure is likely designed to obscure true control. In Neo4j, detecting cycles is a single query:

MATCH path = (c:Company)-[:OWNS*2..8]->(c)
RETURN [node IN nodes(path) | node.name] AS cycle,
  length(path) AS depth,
  reduce(pct = 100, r IN relationships(path) |
    pct * r.ownership_pct / 100) AS round_trip_ownership
ORDER BY depth ASC

In our dataset, 12% of flagged corporate structures contained at least one ownership cycle. Every single one warranted enhanced investigation. Circular ownership is not always illicit (cross-shareholding structures exist in legitimate conglomerates), but the correlation with suspicious activity in our data was strong enough to make it a high-priority signal.

Performance Considerations

Lessons Learned