New trappings to an old mule: A growing large bank curtails current account mule fraud, reduces losses, and manages reputation risk using machine learning models.
Context
“Mules” are individuals, recruited, often unknowingly, to launder illegal money. In this case, individuals or small business owners were often deceived into opening a current account and then using it to funnel illicit funds. While the mules believed they were carrying out legitimate transactions, they were in fact part of a money laundering scheme. Despite stringent KYC processes and using third party bureau data our clients were seeing the occasional mule show up in their systems.
Challenge:
Firstly, manual process of account screening and on boarding left a process gap. The mule is unaware that they are being used and lack any unusual behaviour. They are often carefully chosen and have clean or limited financial histories. In addition, the fraudsters use synthetic or stolen identities, further complicating the detection process at account opening. All this makes it a challenging problem for banking operations as well as risk and fraud units.
Solution
We started with the mule cases identified by the banks risk containment unit as well as the FINRA and MHA databases. Attributes gathered both from the customer journey and credit bureau, rolled up at varying grain fed into one model. Extensive data from the phone network fed into another model. A priority ‘reject’ list was drawn from a combination of both these models resulting into a 12% reject rate.
Benefit
Financial crimes impact a bank across the board. Direct losses, recovery costs, regulatory fines and of course a damage to the reputation. A 12% reject covering about 52% of potential mules was certainly an encouraging start. The models also made it easy on the risk containment unit by auto-processing about 50% prospects. As with the all fraud models, success is a moving target, so the good work continues.
