LinkAI - The Network Miner
Help your community!
We are a team of developers and data scientists from TransferWise:
Ester Eggert (AML/CTF product manager), Krister Jaanhold (AML/CTF data scientist), Jonas Tuhte (AML/CTF data scientist), Vadym Vasyliev (Asia currencies lead engineer), Erki Hindo (AML/CTF engineer), Ivan Lissitsnoi (AML/CTF engineer).
Have you ever wondered how your customers are connected to each other? How transactions are linked? If yes, here is a tool to help you in these efforts.
What is LinkAI?
LinkAI is a tool for identifying suspicious customers and transactions to detect money laundering and terrorist financing. LinkAI provides two main functionalities:
1) Network visualisation as an investigation tool for understanding the links of an individual customer to other customers
2) Network mining tool powered by advanced machine learning algorithms outputting all potentially suspicious customers
Why should financial institutions use LinkAI?
Money laundering and terrorist financing is rarely conducted using one single account. To avoid detection, criminals distribute the funds between several accounts. Some links between the criminal accounts are straightforward (shared recipients etc.). However, others are more subtle and difficult to detect. This is where LinkAI comes into play.
It is of course important to emphasise that not all links are suspicious. It is perfectly normal for connecting links to exist between different customers. By learning from historical data, LinkAI is able to clearly distinguish between normal networks and those indicative of money laundering or terrorist financing.
How does LinkAI distinguish between good and bad networks?
On the background LinkAI derives a number of features for each customer based on the specific network metrics. These include basic features like the size and diameter of the network as well as more complex features like network secrecy and link importance.
LinkAI then looks at historical labelled data (i.e. cases where it is known if a customer is suspicious or good), and detects similar patterns for the customers/transactions currently loaded into the system.
How do you know this works?
Similar approaches have been applied in terrorism investigations (e.g. 9/11 attacks). In those cases the network is formed of real world events and movement of suspects. However, these investigations are retrospective. The attacks have already happened. With LinkAI you can track the networks of events consisting of financial transactions while the networks are still in their early phases. Thus it enables proactive prevention of money laundering and terrorist financing.
What about auditability? I have heard that machine learning is a black box.
Traditionally yes, but LinkAI includes functionality, which makes complex machine learning models easily interpretable. Therefore you are always aware of new typologies and risks detected by LinkAI.
Can I start using it right now?
Yes. The code is open source. You can download it now, run it, upload a CSV file, and instantly get back results.
LinkAI has been successfully tested on anonymised transaction data from real financial institutions.