adjust thresholds to Continually challenge the state and effectiveness of AML programs,
of the art contains costs and ensures superior
reduce false positives. Models must be tuned and tested as things compliance with regulations.”
Lessons learned change. Maybe the original model worked
should be frequently well for a while, but now it has become less Put the governance in place
When delegating many operational
fed back to the system effective. Or there have been changes in
the customer base you don’t know about. processes to the machines, you must
for continuous Maybe feedback from investigators shows monitor how the machines are
improvement. that some scenarios need to be fine-tuned. making decisions.
“So, I have this model. Great. It’s in place, “Machine learning? Firms can do that
it’s performing. It’s helping me focus on today,” said Suplee. “But it’s also applying
quality investigations,” said Hale. “But what all the rigor that goes around it, making sure
am I doing to validate and challenge it, to you have controls in place. That’s the part
ensure that it is still the best model that fits that I often see missing in the industry. For
my business process? Am I looking at this kind of a program to work, you have
“With AI, we definitely see different techniques? Am I tuning the to have good controls on the back end,
alert-to-SAR and case-to- existing model? And then finally, am I making sure that you’re testing and
documenting all of this?” validating the process.”
SAR ratios improve, with
a clear reduction in false Through multiple iterations using different
positives. In many cases, ML techniques, you test and determine AI techniques need to be
you may actually be
the best-performing model (known as evaluated, monitored and
the champion), which may unseat the
reviewing more alerts, existing model.
governed with appropriate
but you’re looking at the oversight and controls
Feed results back into the system in place.
right activity that really for continuous improvement
poses the risk to your Was a particular rule or risk factor helpful
institution, and you’re in the decision to either investigate or not Be prepared to demonstrate
finding the things that investigate? Did the SAR end up being the controls
based on the system-identified trigger, or “You need to be able to show regulators,
add the most value.” was it something else discovered during examiners and your internal audit teams
Carl Suplee, Director of Product the investigation – something the moni- evidence of how you arrived at a decision
Management for Fraud and toring program missed? What attributes and that controls are in place to ensure that
FinancialCrimes Solutions, SAS were associated with false positives and decision is a good one,” said Suplee. “You
false negatives? What new intelligence have to demonstrate that you’re following
items should get more or less attention? a standard model validation process and
documenting model testing and changes
“Learning from resolved case decisions
along the entire life cycle of the model.
is crucial to harnessing AI to continuously
You have to be able to show how a result
improve onboarding, detection and investi-
was arrived at, and be able to replicate it.
gations,” said Ray of Celent. “Combining the
“These aren’t new concepts, but they take “I don’t necessarily see AI replacing transaction
on new emphasis given the relative lack of monitoring systems, but rather as a supplement,
transparency in autonomous, self-learning
approaches.”
being integrated to work together in an AML/CFT
program. The best result for the institution is in
Don’t let AI feel intimidating bringing those two processes and capabilities
“AI is nothing to be scared about,” said together.”
Suplee. “It’s technology you should be
embracing. Your organization might already Carl Suplee, Director of Product Management,
be using it in some other capacity. And you Fraud and Financial Crimes Solutions, SAS
can leverage it throughout the entire life
cycle of your monitoring program, all the
way from the data to the process to the back
end and then back around. There are a lot of
places where AI can come into play. It isn’t
about replacing human beings. It’s about
replicating some of those manual processes
for more efficiency and consistency and
delivering the insights that make your work “False positives are demonstrably costly.
more productive.” However, false negatives – where AML
systems fail to spot actual suspicious
Closing thoughts
cases – may carry higher risk and be hard
Leading-edge analytics and hybrid
modeling techniques detect AML and to quantify. As criminals become smarter,
CFT threats faster and more accurately. the ability to quickly identify new and
Embedded ML enhances every step of the
previously unknown patterns will be para-
process – continuously fine-tuning detection
algorithms, streamlining and automating mount in AML. Advanced analytics such
investigative processes, and boosting as AI, ML and RPA offer opportunities for
overall performance while reducing cost.
real improvements and efficiency gains in
By applying advanced analytics and AML operations.”
powerful machine learning on a unified
Arin Ray, Senior Analyst, Celent
platform, your organization can uncover
more financial offenses, reduce false posi-
tives and run more efficient investigations.
SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA and other countries.
® indicates USA registration. Other brand and product names are trademarks of their respective companies. Copyright © 2019, SAS Institute Inc.
All rights reserved. 110495_G96437.0419