Business Graphs for KYC Checks

Network Risk > Watch Lists

Network Risk > Watch Lists

How clear a picture of financial risk do sanction lists provide? Not very clear, it turns out. In this post we’re going to use data visualization to discuss the potential of business graphs and risk scoring for KYC, showing why we think business networks are important for every bank in terms of the visibility they give into high risk entities.

Doing business with sanctioned entities is a serious matter targeted by existing KYC products, but sanctioned entities do not act alone. Financial crime is conducted by criminal networks spanning the globe. Other than asking sanctioned entities who they do business with or collaborating with law enforcement… how do we map their networks of associates?

Sanctioned entities are illicit participants in the global economic network formed by relationships between businesses. Criminology has shown that legitimate business networks are just as important in mapping criminal networks as intelligence from law enforcement agencies. Fortunately, data describing legitimate business is much easier to collect. When we map sanctioned entities onto a network of legitimate business, we reach a clearer understanding of the risk potential of multiple orders of magnitude more entities than appear on sanction lists.

The business network surrounding entities in red who are sanctioned by the US Treasury Office of Foreign Asset Controls (OFAC).

In the image above, red nodes are entities (companies and individuals) on the US/OFAC sanction list. The large dark gray nodes are their first degree neighbors and the smaller light grey nodes are their second degree neighbors.

While we know the red nodes are high risk, the grey nodes are more likely than other nodes to be high risk customers as well. Are all of these nodes automatically high risk? Compute the 2-hop neighborhood of all OFAC nodes, categorize everyone on it as high risk, call it a day? No, that score would be awful. Risk scoring is not that simple. To be a high risk entity a confluence of factors must align such as a pattern of association with known bad actors. What the visualization shows is the known unknown of risky entities. We know these nodes are higher risk than others – but which ones are high risk? This is the problem Deep Discovery was founded to crack! Deep Discovery is building a risk scoring system that takes this and many more powerful patterns into account.

We’re building a multi-billion edge business graph of identities describing the global economy. We’re pioneering new methods in identity resolution to combine corporate registry filings, trade manifests, employment data and other datasets from across the world into an identity graph. Deep Discovery entities aren’t just names, they are clusters of data points all representing a single identity that corresponds to a real corporation or individual.

This network is the raw data for our risk scoring system. We employ Graph Neural Networks (GNNs) to calculate a risk score based on both the properties and relationships of this business graph. Our automated risk scoring system can accelerate the KYC process while providing increased coverage of sanctioned entities via the network analysis described above and another approach we’ll talk about in our next post – adverse facts versus adverse media.

A risk score is not enough to automate KYC. We believe people don’t act on predictions they can’t understand. We A risk score is not enough to automate KYC because people do not act on predictions they don’t understand. They don’t trust them. We provide an explanation of every risk score using network analysis demonstrating what sources of risk contributed to a risk score.