In contrast, while machine learning and cognitive automation are more
complex and take longer to achieve, greater benefits are possible. And since
the various intelligent automation classifications can be implemented concurrently, institutions that take the full journey will see immediate benefits that
will help offset the costs of implementing these more complex components.
Ways to Integrate Intelligent Automation into
Financial Crime Compliance
A transparent and easily explainable intelligent automation framework, regardless of the level of complexity, is imperative when planning a strategy to
incorporate within a financial crime compliance program. This is essential
given the need to inform and educate senior management and regulatory
agencies regarding how it will be integrated within the program, and to be
sure it is aligned with the institution’s risks. This transparency should extend
not only to the models and algorithms that will predict outcomes, but to the
risk factors that provide the data to make those predictions, as well. But be
wary of “black box” solutions that may be highly complex and risky.
Fundamentally, every decision and action that is executed must be completely
auditable and rationalized in “human-readable” language so that all outcomes
are fully understood and can be justified against any scrutiny. Using the example
of transaction monitoring, every time an alert is flagged as likely to be a false
positive, a financial crime officer must be able to easily uncover the reason
why the model made that determination. The inability to do so will not only
potentially expose the institution to additional risk, but it will make it harder
to support the conclusions made by the models and subsequent actions taken.
With this in mind, there are at least three areas within a financial crime
program where financial crime officers may find that the various types of
intelligent automation can help reduce costs and increase efficiencies and
Transaction Monitoring (TM)
Transaction monitoring is a prime example of where financial crime compliance can benefit from enhanced technology. Typical AML transaction
monitoring platforms are designed to consider rules-based typologies and
scenarios, which not only require constant tuning and updates, but since
they are typically more simplistic and rule-based, they can also fail to take
into account a multitude of risk factors. This often results in a large number
of false positives for humans to resolve.
■ ■ ■ RPA—Bots can scan the internet and public due diligence sites to collect relevant data. They can also compile due diligence results into an
electronic case file for an analyst’s review. Deploying the bot saves the
analyst valuable time.
■ ■ Automate highly repetitive manual
alert resolution tasks.
■ ■ Complete tasks autonomously using
■ ■ Interface directly with existing
■ ■ Design, test, implement quickly
with relatively low investment or
■ ■ Reduce human factor significantly.
■ ■ Use machine learning models to
enhance current TM rules post
processing with the most predictive
■ ■ Use models to provide the likelihood
of whether the alert is a false or true
positive, speeding up human analysis,
allowing for more efficient alert
review and escalation.
■ ■ Streamline model risk management
and simplify regulatory requirements
with the use of accepted, proven
■ ■ Incorporate more advanced models
to enable the use of structured
and unstructured data to support
elements of self-learning.
■ ■ Automate transaction monitoring
through decision support and
advanced algorithms that incorporate
advanced self-learning capabilities
and Natural Language Processing to
interpret unstructured content.
■ ■ Ingest, consider and interpret
massive amounts of data on which to
formulate hypotheses, well beyond
the capabilities of human review.
■ ■ Increase coverage and uncover
emerging risks by considering
patterns, events and factors; reduce
■ ■ Establish base domain knowledge
prior to solution deployment, establish
feedback mechanism to train machine