Page 3 - Konect Science & Technology Magazine Cover

Unsupervised ML for detecting Teller

Fraud in Banking 21

Written by

Frauds carried out by bank employees are a huge global problem.
Association of certified fraud examiners report puts the cost of
banking fraud at around $70bn a year – and cases involving bank
insiders account for about 70 percent of that total. To cover all the
internal fraud scenarios will need an entire course in itself. But I am
going to cover a very specific case of internal fraud which is not
known to most people in the fraud space. And how advanced
machine learning algorithms in unsupervised learning techniques
(yes, you heard it right !.. It is not the boring supervised learning
algorithms). All right, now
if that excites you then what I am going to say next in few
paragraphs is going to sound no less than a movie. Consider the 48
below scenarios:

1.Teller is servicing withdrawal requests of $1000. The customer is
an old lady of 75 years who just signed the withdrawal slip and
handed it to the teller. While processing the slip teller informs the
customer that there is some issue with the system and it is taking
some time. At that moment, the teller turns aside, fills out a
different slip of $1200, and forges the signature of the customer
from the original slip. Hands over $1000 to customer and pockets
the additional $200.
If you are thinking that banking – being an old
2. John Doe is a regular customer of a bank and has multiple mammoth industry – would already be taking care of
accounts with the bank. His relationship manager (RM) has good such frauds then you are not wrong. Banks have lots
repo with him and he tries to give the best service. One day his RM of checks and balances to track this kind of fraudulent
offers him an upgrade on his debit card and starts processing his behavior. After all, one of the primary duties of the
request by asking for IDs and authenticating him in the system. But bank is to keep the money of customers safe.
then he issues an instant debit card without his knowledge (there
are machines that print the debit card instantly … I know !). RM But then again, the challenge is to catch this behavior
sweet talks to John and managed to reset the PIN number of this in the early stages and nip them in the bud.
instant card issued. RM empties John’s bank balance by using Debit Conventional methods like video surveillance, daily
Card at POS, online, and withdrawal. Oh, by the way, RM was smart audits of the cash boxes, etc. will put a lot of burden
to change the phone number associated with that account number. on operational costs.

Also, the cost would outweigh the benefits if the
3. A customer just left the teller counter when the teller saw that average amount of such fraud is low, which is usually
there is a huge balance in another account associated with the the case. And in the age of digital banking, this can be
customer. It is the closing hour of the bank and he starts snooping a matter of survival for traditional banks. And to top it
the account details and also calls his colleague to see about other all, some fraud like in scenario 1 is regulated. For e.g.,
details and products owned by the customer. He starts keeping a Financial Exploitation of Elderly and Vulnerable Adults
list of these kinds of customers and siphon off small bits of $200 regulations in US mandates to take extra efforts
from multiple accounts during the closing hours or early joining against fraud to elderly (age >60) and vulnerable
hours of the bank branch. After 3-4 months, he resigns and joins adults (like a minor or mentally/ physically challenged
another bank. person).

DARVIX CONNECT | 4 www.konectmag.com