■■ ■ Machine Learning—Machine learning augments human
decision making, building upon RPA. At this stage, machine
learning has bearings on the investigative process through the
review of triggered activity. And, it can be used to operationally
automate aspects of the review process. In reviewing and assessing historical outcomes of investigations, the machine can be
deployed to build statistical models that incorporate gathered
data and calculate a likelihood for disposition, either closure
or escalation. Those transactions that have a high likelihood
for escalation would be subject to further review by humans
who apply judgment to the resolution. That judgment is then
assessed in terms of how the models should be updated. Since
false positives tend to be pervasive in transaction monitoring
systems, machine learning models can provide institutions with
significant gains. They do this by quickly identifying alerts for
closure along with the rationale for that conclusion.
■ ■ ■ Cognitive Automation—Institutions need to build upon the
foundation of alerts, previously dispositioned cases, and any of
the machine learning models, to the extent they are already in
place. Cognitive automation relies on an institution’s knowledge
base from which complex reasoning algorithms can be developed. A domain knowledge base is generally specialized to the
institution. It consists of all of the underlying structured and
unstructured models which have been learning and adapting
to the institution’s risks, outcomes, processes and procedures.
Because of this foundation, the machine does not need to
limit its monitoring to the risks the institution already knows,
has identified and captured in a rule or set of rules. Rather,
with cognitive automation, the machine leverages the domain
knowledge to assess the transactions, applying reasoning to
establish a confidence level against the relevance of its findings.
If the pattern is new, the machine would flag the transactions
for human review.
As a result, cognitive automation is a key to finding new and
emerging financial crime risks. It is through cognitive automation
that one truly addresses risk.
Know Your Customer (KYC)
Financial institutions devote substantial time and resources to
performing KYC during onboarding and periodic review intervals. Depending upon the size of an institution and its volume
of new customers, institutions may dedicate hundreds of hours
per month to K YC tasks, often supplemented by contractors and
consultants. Time investments for each customer typically range
from a few hours for a low-risk customer to upward of 24 hours
for high-risk customers.
KYC typically includes:
1. Process steps for conducting external due diligence;
2. Screening customers and often related parties (such as controllers and eventual beneficial owners;
3. Clearing of identified negative news (which can be hundreds
of pages of documentation requiring review); and
4. Reaching out to the front office or intermediary team to obtain
requisite information that adheres to internal protocols to
meet regulatory requirements.
The same is true for each periodic review to be performed.
■ ■ ■ RPA—KYC processes tend to be comprised of highly repetitive
tasks, which are ripe for intelligent automation to augment
and help expedite. As a result, many institutions have already
identified elements of their KYC process where RPA can assist,
such as importing documents from public news sources and
negative news screening.
■ ■ ■ If properly setup, RPA can enable KYC analysts to devote more time to areas of onboarding that require
deeper analysis, such as clearing negative news results or assessing gaps in documentation needed.
RPA could also reduce or eliminate the need to
contact customers repeatedly, resulting in a better
■ ■ ■ When implementing intelligent automation to
supplement KYC work, financial crime compliance
personnel would continue to perform targeted testing of
results achieved using automation to understand the precision
with which the machines perform, or to refine the parameters
that are being used in order to achieve greater accuracy. Over
time, it may be possible to scale back the testing or quality assurance (QA) reviews as the accuracy and consistency increase.
■ ■ ■ Machine Learning—Machine learning can automate extraction
of data from unstructured documents. This, coupled with RPA,
can result in a more reliable and more efficient customer risk
rating process. If the KYC customer file and risk rating can be
updated in a more rapid manner, institutions can then migrate
to more of a real-time risk assessment. This will enable a more
accurate analysis of the customer’s actual risk at a point in time.
■ ■ ■ Cognitive Automation—With the RPA and machine learning
solutions in place and functioning well, the machine can apply
judgment based on the domain knowledge base. For example,
using semantic language processing to evaluate negative news
articles allows for a very diverse set of sources to be used to
gather articles. And at the same time, cognitive automation
can help identify the most relevant articles. Over time, with
feedback from financial crime QA staff, the technology can
be refined to further improve accuracy and reliability. Cognitive automation technology can also be used to identify KYC
outliers that could be risk indicators (e.g., a customer that is
stated to be regulated in a jurisdiction like Cayman Islands
without a record available via the Cayman Monetary Authority).
Through greater automation and technology’s ability to learn
and re-calibrate, financial crime officers can better prioritize
their KYC efforts and the information they obtain. As a result,
they can be more reflective of actual risks, with a robust audit
trail of analysis to justify any changes.
Innovation today means considering new
approaches supported by technology
to help alleviate compliance problems,
improve accuracy and efficiency, and help