Artificial intelligence is one of the technological paradigms that is set to transform financial services in the years to come. This includes every aspect of the bank, ranging from customer interaction through chatbots, process automation and risk and credit assessment just to name a few. However, the compliance function is one of the areas where AI will have a profound impact on banking. This is far from my area of experize, so bear with me as I try to look into the compliance department from my laymans perspective.
Authors note: When speaking about artificial intelligence in the compliance function, AI is used as a collective term including machine learning and natural language processing (NLP).
Compliance has always been an important function in the banks, but ever since the financial crisis of 2008, the cost of compliance has been accelerating. According to Federal Financial Analytics, compliance cost for the six largest US banks doubled from 2007 through 2013 from $35 billion to $70 billion, an increase of 102 percent. In 2015, the Financial Times estimated that some of the world’s largest banks each spent an additional $4 billion a year on compliance since the financial crisis.
The rules are often complex and difficult to understand and apply. And much of the process remains highly labor-intensive, when even the most automated solutions are often incompatible with other systems and, even today, most still depend heavily on manual inputs. At JP Morgan alone, the compliance headcount has nearly doubled the last five years from 23 000 in 2011 til 41 000 in 2016.
This will accelerate even more as technology drives changing customer behavior and increase complexity. Todays compliance functions are used to dealing with individuals and business as entities. With the promise of IoT and connected (every)thing this will change dramatically with machine to machine transactions. Gartner predicts that there will be 25 billion connected devices by 2020, and according to Mastercard, every connected device is a potential payment device.
The changing landscape of financial services will most definitely usher in a an array of updates to both existing and new compliance documents. For each new or updated regulatory requirement, someone has to read it, analyze it, understand the impact on the organisation, and then undertake and manage whatever actions are needed to ensure compliance. Needless to say, this approach does not scale and requires a vast amount of manual labor, often performed by skilled and experienced staff.
There are several ways AI could be applied in the implementation and revision of regulatory compliance. Even though large volumes of unstructured data is challenging for unsupervised machine learning, automated metadata extraction could help us understand and identify what products, topics and parts of the organization are likely to be affected by the given regulatory requirement. Furthermore, entity identification could classify who the actors in the regulatory documents applies to, this may be customers, employees, and also machines. Finally, natural language processing must be able to analyze legal documents and thus «understand» the context of the regulations to determine who they apply to as well as and what products, and processes they refer to.
The real promise of AI in the compliance function is in process automation, and one of the areas where AI shows promise is within Anti Money Laundering (AML). More specific in reducing the number of false positives, which has proven to be a growing headache as the number og obligations grow.
One of the characteristics of money laundering is that it is rarely limited to the activity of a single person, business, account, or a transaction. Therefore detection requires behavioral pattern analysis of transactions occurring over time and involving a set of (not obviously) related real-world entities. This has proven to be a challenge for unsupervised machine learning, as the process is only as good as the data, at the same time supervised machine learning within AML has proven to be ineffective as the nature of money laundering is constantly evolving. However there are a number of unsupervised machine learning techniques that could increase precision.
Bank of America has warned that even with state of the art AML monitoring systems the compliance people will be flooded with alerts without any context, and the number of false positives that need manual processing makes the AML function time-consuming and costly. In order to solve this, Bank of America has created an in-house system that compiles information from different sources such as its monitoring reports, news reports, alerts issued by regulators, and then turns that feedback into in order to increase the range of available and relevant datasets and is claiming to have increased conversion rates from 5-7% to 70% in number of alerts that require an action.
The complexity of money laundering makes it important to utilize a technology that has the ability to track the behavior of each individual involved and link them to one another. One such AI technology is Smart Agents. Smart Agents are virtual representations of each entity in the system and work collaboratively to gather collective intelligence.
Automating AML is also one of the issues IBM aims to solve with the acquisition of Promontory Financial Group. According to IBM, they plan to have the Promontory employees apply their expertise to train Watson in order to strengthen Watson Financial Services in regtech.
According to Booz Allen Hamilton, Machine learning will revolutionize AML. Rather than relying on a growing number of analysts manually investigating alerts from monitoring systems, the future of compliance should rather rely on automatic detection of suspicious activity, making it more difficult for criminal activity to go undetected.