The game of cat and mouse between the regulators and banks against money launderers has now moved to a new level – all thanks to the emergence of AI and machine learning technologies.
AI and machine learning technologies have been around for a long time now, but have lately started coming into prominence in the world of financial services. Banks and financial services companies are under constant regulatory pressure to implement ever more stringent regulations to curb the flow of illegal money through their counters.
Currently, the processes for anti-money laundering (AML) and know your customer (KYC) are often both tedious and time consuming. Many financial institutions still rely on a combination of manual and non-intelligent, non-adaptive automation – this trawls through heaps of data to monitor for suspect transactions and ensure compliance to regulations. These emergent AI and ML technologies offer a more intelligent approach to automating banks’ monitoring and compliance capabilities.
Current Challenges
The financial services industry plays an important role in governments’ efforts worldwide in controlling and eliminating the infusion and circulation of illegal money into formal financial systems. Governments are constantly evolving regulatory restrictions and monitoring requirements, for example for the EU’s Fifth Money Laundering Directive (5MLD) and regular updates to the US Patriot Act and Sanctions regulations.
Thus, banks and financial services companies find themselves constantly on the treadmill of upgrading their systems and processes to monitor and comply. Against this backdrop, those looking to avoid detection are trying more and more innovative ways to slip through the monitoring net.
The penalties to banks that let such a transaction slip through are very severe – and given such a high price for failure, banks have taken a very conservative approach to dealing with suspect and potentially suspect transactions. This has led to large volumes of false positives in addition to the genuine ones, and unravelling these has become one of the largest concentrations of manual effort for banks. In an increasingly fast-paced world, where customers expect services in record time, this has the disadvantage of reduced processing speeds, missed SLAs and poor customer experience.
Banks employ significant numbers of operations personnel trained in monitoring transactions, picking out potentially suspicious ones and working through each to decide if they are false positives or transactions needing to be stopped. This is often based on a set of well-defined rules – and also on the expertise of the operations personnel trained to pick-out the suspicious ones from the rest. The operators use a combination of a deep knowledge of the client, their business and associated transaction flow patterns to spot those that don’t conform to the normal pattern.
Banks have also leveraged automation to augment and amplify human efforts in sifting, sorting and using deterministic approaches to this monitoring effort – and such automation have largely been rule-based and non-intelligent (i.e. no ability to learn) and non-adaptive (using that learning to drive better conclusions). Coupled with this is also the risk of the ‘human-fatigue factor’ inherent in largely manual operations, that may cause a few suspicious transactions to slip through the net. This is precisely where AI and machine learning can help the banks.
Streamlining AML with AI and ML technology
AI and ML technologies enable banks to implement ‘intelligent automation’ that can learn – either through self-learning or by being taught to determine if a suspected transaction is a false positive or not by a human supervisor. There is also ‘adaptive automation’ that can apply such learning, adapt its rules and then improve its classifications for the future.
Most banks are conducting proofs-of-concept and pilots to test the efficacy of using these technologies. These experiments involve using these approaches to develop algorithms that are run on large quantities of past real-world data and trained using supervised learning techniques, letting an experienced human operator to teach them the right from the wrong conclusions. Training using large quantities of real-world data enables these algorithms to narrow the deviation from the correct outcomes of such transactions, processed earlier by human operators.
In some scenarios unsupervised learning approaches can also be used to learn from past transactional data and the associated outcomes. It is important therefore that the quality of transactional data used in the learning process is good enough, and it is important to use datasets that offer a variety of patterns, to improve the quality of the learning.
These algorithms will have to be put through rigorous testing to determine the ‘dependability factor’ before they can be used to replace human operators. Until this happens, these algorithms can be used to assist human operators in pre-classifying potentially suspect transactions into low, medium and high risk categories, helping improve the efficiency of human operators.
When such technologies are employed at scale, they can offer enormous benefits. Firstly, they improve the overall quality of transaction monitoring and compliance, as they can read and make sense of large quantities of structured and unstructured data, and conduct real-time analysis of transactions to classify potentially suspicious ones and grade them as low, medium and high risk categories. This enables prioritized processing by human operators.
Technologies can also significantly reduce the risk of the human-fatigue factor, as AI and ML solutions will have much higher threshold for fatigue – if any at all. They can also learn to spot newer patterns of potentially suspicious transactions through continuous learning, both supervised and unsupervised.
Ultimately, the major impact on banks will be to reduce the overall number of people deployed in AML and KYC operations in banks – this not only saves costs, but enables banks to redeploy those staff who previously worked on manual processing into higher-level, creative, problem solving roles. With customers wanting more instant, seamless experiences than ever before, banks should be using their best staff to find new ways to innovate and meet customer demand – not to carry out manual processing tasks that machines can do faster.
It’s not news that investing in AI and machine learning technologies will have a positive impact on financial organizations – and in a landscape of ever-expanding regulation and growing customer demands, this is more necessary than ever. However, if banks and financial services firms don’t jump on the technologies now, organizations won’t be able to harness the scale of that impact.
Jayakumar Venkataraman
Managing Partner, Financial Services & Insurance Practice, IC Europe
Jayakumar has more than 23 years of experience in the banking and financial services industry. He leads our Financial Services & Insurance practice in Europe. He has been with Infosys for over 15 years, all of it in consulting under various guises – DCG, IC, MCS and Lodestone. Prior to joining Infosys, Jayakumar worked as a corporate banker at Citibank, American Express and BNP Paribas in the areas of transaction banking operations, credit risk management and corporate relationship management.
