Home > Blogs > Eliminating false positives for banks in their AML drives

One of the biggest regulatory and compliance challenges financial institutions face today is the high rates of false positives generated from their AML (anti-money laundering) monitoring systems. Despite designed to identify suspicious transactions involving illicit proceeds for illegal purposes, traditional TMS (transactions monitoring systems) are fundamentally antiquated.

They lack the ability to assess the context of a transaction by a customer, the agility to react to rapidly evolving digital patterns used by money launderers and the ability to understand clearly, why a transaction can be possibly suspicious. Over the last few years only, global cases of money laundering have amplified manifold, leading enforcement agencies to update the regulations in respective countries. The EU’s latest “blacklist” due to the $221 billion money-laundering scandal involving Denmark’s Danske Bank is one such example . (Quartz)

Closer home, banks have spent ~$321 billion in fines since 2008 in lieu of regulatory failings, terrorist financing, money laundering, and market manipulation. In fact, as per UNODC, ~2 to 5% of the GDP worldwide is laundered globally each year. (Business Standard)

In an environment of rising regulatory demands and spiking screening volumes, artificial intelligence (AI) and machine learning (ML) can be the only viable option to accurately detect suspicious transactions. There are a few challenges, however.

False positives: The double-edged sword for financial institutions

Almost all banks and financial institutions are implementing advanced verification systems, adding stricter criterions to accept new customers and increasing PEP (Politically Exposed People) screenings. The scrutiny of customer public records has gained so much momentum that banks include negative publicity/news as an assessment factor. This becomes a double-edged sword for banks and financial institutions. Because as banks inspect more deeply, they also need to tread the thin line of customer privacy violations. The more questions banks ask, the more uncomfortable the customers become.

Of course, AI has its benefits in battling financial crimes in banks, i.e. it can improve the effectiveness & efficiency of investigations and improve their risk management practices. There is however the common element of “false positives”, where banking systems end up flagging a legal transaction as discredited. Inaccurate data in an environment of data overload, therefore, is a growing concern. Consistently ineffective legacy systems have resulted in astronomical budgets, causing dropped stock values, leading to fading consumer trust and long drawn out resolutions.

However, as PWC pointed out recently in a research document, AI maturity is a hindrance for banks and financial institutions. It points out despite being aware of AI as a cheaper, faster and smarter option to tackle financial crime, there is a lot of confusion around how to harness it. The olden ways of rule-based filtering technologies are now inadequate, and inflexible to support real-time interventions, as they depend on the expert judgement from human beings. Until the recent past, banks have reasonably not taken complete advantage of AI solutions due to the concerns of transparency with “black box” models. Of course, poor data leads to poor outcome, but to avoid adopting AI unless the data is less than perfect, also removes your competitive edge.

In the age of implementation however, with Machine Learning, banks can now create a holistic viewing panel of their customers’ form static KYC documents and transactional dynamic data in a completely compliant way. Even more, an ML engine can be used on top of an existing infrastructure to run independently without troubling current operations.

Getting realistic about ML deployment in AML

Today banks do not have to shred their existing system to replace new advanced technologies to enhance legacy ones. Many banks and financial institutions have made strategic investment in data science companies that specialize in real-time fraud detection. There is also a general acceptance of machine learning and artificial intelligence in AML regulations and fraud detection.

With money laundering, cyber-attacks and breaches becoming a global menace, banks need to be sure of their AML and compliance budgets before jumping in. While ML and AI are known for being great at fully automated systems, banks need rather to start looking at them as a human augmentation tool. Soon AI and ML will possibly become critical and a core component for compliance-conscious financial institutions worldwide.

This article has been published in DATAQUEST.


Article by


Pankaj Upadhyay

Vice President - Data Science, BI & Analytics