AI for Regulatory Compliance at Banks: 4 Assessments Before Say OK

3 mins

Banking is one of the industries where artificial intelligence and machine learning find their applications at a rapid pace. Regulatory compliance within banking is an area which has become a costly and inefficient affair due to complex and never-static regulations, and this area is set to be disrupted with the emergence of RegTechs (regulatory technologies), which are touted as the next big thing in the financial services industry. Regulatory compliance solutions leveraging emerging technologies such as artificial intelligence and machine learning are reportedly a more than USD 100-billion opportunity in the coming years as global banks are seemingly spending huge on next-gen solutions in fear of heightened regulatory scrutiny. The key areas of focus for RegTechs are AML/CFT, fraud prevention, regulatory compliance automation, case management, employee surveillance and management information and reporting.

Banks are willing to give artificial intelligence-powered regulatory compliance solutions a try, but the banking leaders entrusted with key software procurements may find it difficult to choose the right one for their banks as they generally lack in formal artificial intelligence/machine learning background and training in data science. Here, we are trying to give some guidelines to business people from a non-technical background while assessing an AI-based solution. These will help decide whether an AI solution can be relevant to their business. However, if banks have their own in-house tech/AI experts, the business people should consult with them before procuring these solutions.

Match your problems with their expertise

It is important to list out the problems that a bank wants to resolve with the application of AI/machine learning and match them with the features that the solution vendor provides. A legitimate, focused vendor can explain the business problem it solves and how it uses AI/machine learning to solve that problem in a comprehensive manner to business people. On the other hand, those AI vendors with no clear road ahead often find it difficult to articulate the business problem they solve. Therefore, a bank should go through the value propositions of a vendor and understand if they align with its actual requirements.

Understand the troubles related to data

Without the proper kind of data and the proper amount of data, no machine learning solutions can work. Machine learning models should be trained with data in order to identify patterns in the data and predict outcomes or find solutions to a problem. Banks will have to provide a significant amount of their data to the AI solution vendor to train their machine learning models. Here, they will have to find out what kind of data will be used by the vendor and how it will be used. Most of the times, the data from banking systems cannot be readily deployed into a machine learning model because of format or structure issues. There might be a need to clean and prepare the data for deployment and it may take many months to complete. So, a bank should consult its IT/AI experts with regard to the availability, accessibility, AI preparedness, and data cleaning time and costs before taking a procurement decision.

Look for a reliable history

It is very important to assess the track record of the AI/machine learning product as these technologies are relatively new and many of their applications are yet to be reviewed to be successful. A vendor’s engagement with past and current clients can be found in the forms of press releases, case studies, client list on website, testimonials, etc. Case studies provide a better view of the solution and its capabilities, as they clearly mention the addressed problems and issues and quantifiable results. In the case of regulatory compliance, there will be clear metrics on improvement in efficiency and effectiveness. While the metrics may change from one case to another, they can give an idea about the efficacy of the solution. However, the lack of clients does not always mean that a vendor’s AI product is not worth testing. If the vendor has a referenceable contact and some evidence of positive results, it is worthier to be considered.

Know the difference between ‘black box AI’ and ‘transparent AI’

‘Black box AI’ is a concept in machine learning where even the designers of a machine learning algorithm cannot explain why the machine arrived at a particular decision. This inability to explain the machine learning model would create issues for banks, especially during regulatory scrutiny. Regulators often reject models that are not auditable. In contrast, there is ‘transparent AI’ or ‘explainable AI’ which makes it easy for auditors to understand the inner works of a machine learning model. Transparent AI uses techniques that make the model trustworthy and easily understandable by humans. While choosing an AI solution, it is imperative from the part of a bank to know if the solution has the required transparency to be audit-friendly. Tookitaki’s Anti-Money Laundering Suite and Reconciliation Suite are good examples of transparent solutions that demystify modern machine learning and bring algorithmic transparency by providing thorough explanations for predictions.

As AI and machine learning are technologies which are still in their early stages, banks should be diligent while purchasing and deploying solutions based on these. Artificial intelligence is a complex technology, and its full-scale assessment requires additional skills and technical knowledge which are not required for other technology procurement decisions. Proper audit on the solution’s compatibility with existing systems at the bank and the solution’s established capacity to solve business problems is crucial before enrolling for the service. The guidelines above are meant to help business people with no/limited AI knowledge as they get a proposal from an AI solution vendor, and we hope these tips will help them make a proper decision.