While much focus has been dedicated
to improving the identification and
thwarting of external bad actors, banks
must better monitor for and prevent unethical and illegal activities from close-in business partners and even its own
employee base. Consider the recent case
of a large bank where employees fraudulently opened millions of bank and
credit card accounts unbeknownst to
clients. The bank employees have been
accused of creating accounts customers
did not request, funding those accounts
by transferring money from other accounts without notifying customers,
creating PIN numbers for unknown and
unwanted debit cards, and setting up fictitious email addresses to obfuscate their
fraud. The firing of those employees involved should suggest that management
would not tolerate this unethical activity.
However, many are speculating that with
such activities happening over a period
of several years, means that some managers turned a blind eye to a situation
of ongoing fraudulent activity or were
potentially a part of the problem.
Data can and should be harnessed
by banks to produce valuable insights
regarding the behaviors and actions of
people and entities—both legitimate
and illegitimate. If the right questions
are being asked, there is no excuse for
this type of fraud or unethical behavior
to exist on a systemic basis. Data sci-
ence, including big data, artificial intelli-
gence and machine learning, can enable
institutions of all industries and sizes to
better manage risks, to find fraud and
criminality, and to therefore support the
needs of compliance, management and
governing boards.
Applying Data Science to
the Banking Industry
Data science methodologies can alleviate
the costly dependency on human bank
investigators and their manual, limited
ability to more efficiently and directly
discover internal and external bad actors
and suspicious activities. The insights
revealed by a data science-powered
solution can then be conveyed through
APIs, reports, dashboards and rich visualizations, enabling banks to take the
required actions to reduce their risk and
associated costs. In today’s environment
with the proper data science solution in
place, investigators could report findings
to senior management in a timely manner after a data science resolved alert.
Identifying Employee
Fraud with Data Science
Techniques
The identification of unauthorized
client accounts is a relatively simple
problem to address. From a data sci-
ence perspective, it is not a complex
issue, and can be quickly identified with
access to a relatively small number of
data sets, all of which are in any bank’s
possession. Here are four key potential
data sets and data science techniques
that can be utilized to ensure unethical
internal behavior is identified:
■ ■ ■ Email account names. Fuzzy logic
and anomaly detection can highlight
an inordinate amount of accounts with
fraudulent or invalid email addresses.
For example, if a checking account application, a mortgage application, and a
credit card application have very different data or email addresses, this could
raise an alert. Banks can set alerts that
are triggered with, for instance, three or
more data disparities, and software can
also alert if an email address is changed
just before a new account is created. Of
course, many people maintain different
email addresses, but analysis in aggregate may lead to a problem.
■ ■ ■ Account dormancy. Banks can utilize time series analysis, a model used to
predict future results based on previously observed values, to identify an abnormal amount of unused accounts. This
tool can be adapted to flag accounts that
have had no activity for long periods of
time, after being open, say in the previous 12 months, as well as a heavy pattern
of account escheatment. This could be
a potential sign that the customer did
not ask for a product and does not know
what he has. Statistically significant
amounts of both would indicate that
customers had accounts of which they
were not aware. Of course, you might
not want to wait 12 months–you can
also review for inactivity for accounts
opened for 30 or 60 days in order to take
immediate action.
■ ■ ■ Customer complaints. Natural
language processing can pull out phrases
Using Data Science to Mitigate
Bank Employee Fraud
BANKS TODAY hold vast and ever-multiplying stores of data, including emails, text messages, Customer Relationship Management (CRM) records, client activity information, and much, much more. This data, if analyzed correctly, could reveal both enormous value and
uncover unidentified risk. In efforts to meet the requirements and expectations
of regulators, as well as legislators, shareholders and customers, banks have
spent billions on scores of data management, data mining, data analytics, and
reporting solutions and tools. Despite these investments, banks continue to face
information challenges and significant operational, legal, and reputational risk.
