• Milos Dunjic

Using AI and machine learning to monitor transactions

Since I was unfortunately unable to participate at Center for Financial Professionals' ( X-Tech 2019 Conference, in Las Vegas in early April, I was more than happy to provide some thoughts to Risk Insights on how AI and Machine Learning can improve monitoring transactions

NOTE: The original article was published on

Can you please tell the Risk Insights readers a little bit about yourself, your experiences and what your current professional focus is?

I have over 25 years of experience in senior technology roles in payments and capital markets space, where I have developed and launched innovative, award winning digital payment and stock trading solutions. I won TD Bank Group’s “Inventor of the year” 2017 award, and I lead TD’s Enterprise Payments Technology Innovation team. I have a master’s degree in Electrical Engineering with major in Computer Science from the University Of Belgrade.

What, for you, are the benefits of attending a conference like the ‘X-Tech 2019 Convention’? What can attendees expect to learn from your session?

The opportunity to speak at any conference, including X-Tech 2019 Convention, is always an opportunity for me to share my own industry experience but also the opportunity to validate my current views and potentially calibrate them through the connecting and discussion with the industry peers. I am always looking forward to connect and learn new stuff. There’s always a good thing, because our learning never stops.

In your opinion, how can AI and machine learning be used to monitor transactions?

The very definition of machine learning is that it is a software system which is able to adapt to new circumstances and to detect and extrapolate patterns, based on previously observed data (learning), without need to change the rules encoded inside the software. As such, it has significant potential in being applied to monitoring transactions – in payments, capital markets and insurance as well.

Traditionally most of these systems were developed and deployed by explicitly programming known rules into the software. As we are moving through the era of rapid digitization, the attack vectors on transaction space keep evolving and are becoming more sophisticated by the day. It is becoming very expensive to have software developers and analysts keeping up with those changes and keep manually updating the software. It is simply not scalable anymore.

First, the system designers cannot anticipate all possible situations that the system might find itself in. Second, the designers cannot anticipate all changes in the environment around the system over time; for example, a program for predicting future stock market prices must learn to adapt when conditions change from boom to bust, to recognize early signs, at the same time as trading patterns and velocity change rapidly.

Therefore we need new approach, and techniques of machine learning seems best suited for the job.

What are the implications of monitoring millions of transactions and flagging threats?

Historically banks have been using transaction monitoring systems for decades, which were based on pre-defined set of rules that require the output to be manually checked. The success rate was generally low. On average, only small percentage of the transactions flagged by the legacy fraud monitoring systems ultimately reflect a true crime or malicious intent. Most fell under category of false positives. Extensive manual processes were required, but as the transaction volumes are increasing at rapid pace, companies have struggled to keep up with the changes.

By contrast, today’s machine-learning solutions use predictive rules with the ability to automatically recognize anomalies in data sets, based on the learned patterns from the historically observed (learned) data. These advanced algorithms can significantly reduce the number of false positives and reduce need for manual Interventions

However, since machine learning and AI have the potential to eliminate manual processes significantly, and are bringing high levels of automation, the financial institutions must get ready in order to be able to explain to the customer why they were rejected for a loan, why their payment did not go through, etc. That requires increased level of sophistication of the customer support staff, new sophisticated tools providing insights to the front line employees into the outcomes of machine learning algorithms.

Could you provide insights on gaining a better perspective on markets?

Attending conferences is always an opportunity to get insights and learn from the experience of industry peers. Following posts of industry experts on LinkedIn, Internet portals is also a good strategy that I am myself utilizing on a daily basis.

How do you see the technology and innovation space evolving in the next 6-12 months?

Machine learning and Artificial Intelligence will definitely continue be a focus of many corporate innovators, from intelligent NLP based advisory chatbots, fraud detection and transaction monitoring solutions in payments, trading and insurance space, investment robo advisors systems, and also securities trading strategies. We shall also monitor whether machine learning can help Blockchain space, where it potentially may improve efficiency of consensus algorithms, and make Blockchain faster and more usable for everyday payments

The machine learning is the technology that can sit next to the legacy system, consume the transaction data and do its thing, without need to replace the legacy. As such, I see it as ‘legacy friendly’ augmentation, which makes it easier to come up with compelling business cases for using it.

21 views0 comments