AML and graphs - meant to be together
One of the main information layers in the AML process is clients' transactions analysis. Old school approach suggests digging in huge tables and transactions' logs trying to find some suspicious deals or proofs of the client's profile. More advanced way is to add some aggregation results as top partners of the client by number of payments or by volume, or compare top countries for incoming or outgoing payments. Working with the tables is hard and time consuming. But the main drawback of the "transactions table" approach is that you can see only direct payments so an analyst has limited field of view and ignores very useful information about his partners' activity.
The graph is the mathematical structure of the interconnected objects. Graphs are all around us, your Facebook friends - a graph (people as objects and friendship as connections), your and your friends' phone calls - a graph (phone numbers as objects and phone calls as connections), football game - also could be represented as graph (players as objects and passes as connections). Now think about clients and partners as objects and payments between them as connections. All payments activity of all bank's clients is a one huge graph. Selection of the one client and those clients who made payments to him or received payments from him, as also made some payments in between is a subgraph showing all connections of the first client and his "neighbors".
Graph has many advantages over a table in analysis process:
it is visual, so it is easier for human to analyze it;
it is faster to process comparing to tables;
some informative characteristics could be calculated for the graph;
graphs also provide topological information about interconnections.
Summing up, graph is a great way to present transactions data to the AML analyst for the further inspection. And graphs provide some otherwise unobtainable risk ratios - such as the longest transactions chain or the presence of the circle pattern in payments.
In the next posts I will show some real life examples how graphs can help find suspicious transactions and clients.