Yes, robots are on their way to take our jobs.That’s a good thing, and we should be grateful because the jobs they’re taking are pretty bad.Do you really want to go back to manually monitoring, flagging, and investigating the world’s daily bank transfers for financial fraud and money laundering schemes? DBS Bank, Singapore’s largest bank, most emphatically does not.The company has spent years developing a cutting-edge machine learning system that heavily automates the tedious process of “transaction surveillance,” freeing up human analysts to perform higher-level work while operating in delicate balance with the industry’s antiquated financial regulations.It’s fascinating material.Working with Artificial Intelligence by Thomas H.Steven M. Davenport and Steven M.Miller is chock-full of similar case studies from a variety of tech industries, examining commonplace human-AI collaboration and providing insight into the potential consequences of these interactions.
Working with AI: Real Stories of Human-Machine Collaboration by Thomas H. MIT PressSteven M. Davenport and Steven M.Miller.The MIT Press has granted permission for this reprint.Copyright until 2022.
DBS Bank: AI-Powered Transaction Monitoring
Since the Bank Secrecy Act, also known as the Currency and Foreign Transactions Reporting Act, was passed in the United States in1970, governments around the world have held banks accountable for preventing money laundering, suspicious cross-border flows of large sums of money, and other types of financial crime.DBS Bank, Singapore’s and Southeast Asia’s largest bank, has long prioritised anti-money laundering (AML) and financial crime detection and prevention.”We want to make sure that we have tight internal controls within the bank so that perpetrators, money launderers, and sanctions evaders do not penetrate the financial system, either through our bank, through our national system, or internationally,” said a DBS compliance executive.
“The Limitations of Rule-Based Surveillance Monitoring Systems”
For many years, the area of DBS that focuses on these issues, known as “transaction surveillance,” has used AI to do this type of work, as has the case at other large banks.This function evaluates alerts generated by a rule-based system.The rules evaluate transaction data from a variety of bank systems, including those for consumers, wealth management, institutional banking, and payments.The rules flag transactions that match conditions associated with an individual or entity conducting suspicious transactions with the bank—those involving a potential money laundering event or another type of financial fraud.Rule-based systems, also known as “expert systems” in the past, are one of the oldest types of AI, but they are still widely used in banking and insurance, among other industries.
Every day, rule-based financial transaction surveillance systems of this type generate a large number of alerts at DBS and most other banks around the world.The primary flaw of rule-based surveillance systems is that the vast majority— up to 98 percent — of alerts generated are false positives.Some aspect of the transaction sets off a rule, resulting in the transaction being flagged on the alert list.However, further investigation by a human analyst reveals that the alerted transaction is not suspicious.
The transaction surveillance analysts must investigate each alert and review all relevant transaction data.They must also take into account the profiles of the individuals involved in the transaction, their previous financial behaviours, anything they have declared in “know your customer” and customer due diligence documents, and anything else the bank may know about them.Following up on alerts takes a lot of time.
If the analyst determines that a transaction is legitimately suspicious or fraudulent, the bank is required by law to file a Suspicious Activity Report (SAR) with the appropriate authorities.This is a high-stakes decision, so the analyst must get it right: if wrong, law-abiding bank customers may be incorrectly notified that they are being investigated for financial crimes.On the other hand, if a “bad actor” is not identified and reported, it may result in problems with money laundering and other financial crimes.
For the time being, rule-based systems cannot be eliminated because most countries’ national regulatory authorities still require them.However, DBS executives realised that there are numerous additional sources of internal and external information available to them that, if used correctly, could be used to automatically evaluate each rule-based system alert.This could be accomplished with machine learning, which can handle more complex patterns and make more accurate predictions than rule-based systems.
Using AI Capabilities from the Next Generation to Improve Surveillance
DBS began a project a few years ago to combine the new generation of AI/ML capabilities with the existing rule-based screening system.The combination would allow the bank to prioritise all rule-based system alerts based on a numerically calculated probability score indicating the level of suspicion.The machine learning system was trained to identify suspicious and fraudulent situations based on recent and historical data and outcomes.The new ML-based filtering system had been in use for just over a year at the time of our interviews.The system evaluates all rule-based system alerts, assigns a risk score to each alert, and categorises each alert into higher-, medium-, and lower-risk categories.This type of “post-processing” of rule-based alerts allows the analyst to determine which ones should be prioritised right away (those in the higher-and medium-risk categories) and which can wait (those in the lowest-risk category).This ML system has an important capability in that it has an explainer that shows the analyst the evidence used in making the automated assessment of the likelihood that the transaction is suspicious.The AI/ML model’s explanation and guided navigation assist the analyst in making the best risk decision.
DBS also created new tools to aid in the investigation of alerted transactions, such as a Network Link Analytics system for detecting suspicious relationships and transactions involving multiple parties.Financial transactions can be visualised as a network graph, with the people or accounts involved as nodes and any interactions as links between them.This relationship network graph can be used to identify and further assess suspicious patterns of financial inflows and outflows.
Simultaneously, DBS has replaced a labor-intensive approach to investigation workflow with a new platform that automates much of the support for surveillance-related investigation and case management for the analyst.CRUISE is a system that combines the results of the rule-based engine, the ML filter model, and the Network Link Analytics system.
Furthermore, the CRUISE system provides the analyst with simple and integrated access to the relevant data from across the bank that is required to follow up on the transactions under investigation.The bank also captures all feedback related to the analyst’s work on the case within this CRUISE environment, and this feedback helps to further improve DBS’s systems and processes.
The Analyst’s Impact
Of course, these advancements make analysts much more efficient in their alert review.It was not uncommon for a DBS transaction surveillance analyst to spend two or more hours investigating an alert a few years ago.This time included the time spent gathering data from multiple systems and manually compiling relevant past transactions, as well as the time spent evaluating the evidence, looking for patterns, and making the final decision on whether or not the alert appeared to be a genuine suspicious transaction.
Analysts can resolve about one-third more cases in the same amount of time after implementing multiple tools, including CRUISE, Network Link Analytics, and the ML-based filter model.Furthermore, for high-risk cases identified with these tools, DBS is able to catch the “bad actors” faster than before.
The DBS head of transaction surveillance shared the following insight into how this differs from traditional surveillance approaches:
At DBS, our machines can now gather support data from various sources across the bank and display it on the screen of our analyst.The analyst can now easily see the relevant supporting information for each alert and make the correct decision without having to search through sixty different systems for the data.Machines can now do this for analysts much faster than humans.It simplifies the analysts’ lives and sharpens their decisions.
Due to practical constraints, transaction surveillance analysts could only collect and use a small portion of the data within the bank that was relevant to reviewing the alert in the past.Today, with our new tools and processes, the analyst at DBS can make decisions based on instant, automatic access to nearly all relevant data about the transaction within the bank.They see this data on their screen, nicely organised and condensed, with a risk score and the assistance of an explainer that guides them through the evidence that led to the model’s output.
DBS invested in a skill set “uplift” for staff involved in the development and implementation of these new surveillance systems.The transaction surveillance analysts, who had expertise in detecting financial crimes and were trained in using the new technology platform and relevant data analytics skills, were among those who benefited from the upskilling.The teams contributed to the design of the new systems, beginning with front-end work to identify risk typologies.They also provided input to identify the data that made the most sense to use, as well as where automated data analytics and machine learning capabilities could be most useful.
When asked how the systems would affect human transaction analysts in the future, the DBS compliance executive stated, “Efficiency is always important, and we must constantly strive for higher levels of it.”We want to handle the transactional aspects of our current and future surveillance workload with fewer people, then reinvest the extra capacity in new areas of surveillance and fraud prevention.There will always be unknown and new dimensions to bad financial behaviour and bad actors, and we must devote more time and resources to these areas.To the greatest extent possible, we will accomplish this by reinvesting the efficiency gains we achieve through our more traditional transaction surveillance efforts.
The Next Stage of Transaction Monitoring
The bank’s overall goal is for transaction monitoring to become more integrated and proactive.Rather than relying solely on rule-based engine alerts, executives want to use multiple levels of integrated risk surveillance to monitor holistically from “transaction to account to customer to network to macro.”This combination would enable the bank to find more bad actors in a more effective and efficient manner.
The compliance executive went on to say that money launderers and sanctions evaders are constantly coming up with new ways to get around the law.To stay ahead of these emerging threats, our people must collaborate with our technology and data analytics capabilities.We want to use the time our people have been spending on the time-consuming, manual aspects of reviewing alerts to keep up with emerging threats.
Human analysts will continue to play an important role in AML transaction surveillance, though their time management and human expertise will evolve.
“It’s really augmented intelligence, rather than automated AI in risk surveillance,” the compliance executive said.We do not believe we can eliminate human judgement from final decisions because assessments of what is and is not suspicious in the context of money laundering and other financial crimes will always be subjective.We can’t get rid of the subjective element, but we can reduce the manual work that the human analyst has to do when reviewing and evaluating alerts.
“What We Learned From This Case”
A system that generates a large number of alerts, the majority of which are false positives, does not save human labour.
Multiple types of AI technology (in this case, rules, ML, and Network Link Analytics) can be combined to enhance the system’s capabilities.
Companies may not reduce the number of people doing a job even if the AI system significantly improves its efficiency.Employees can instead use the extra time to work on new and more valuable tasks in their jobs.
Human judgement may not be eliminated from the evaluation process because there will always be subjective elements in the evaluation of complex business transactions.