Share this post:
The vast potential to redesign workflows in the banking industry through innovative technologies such as Artificial Intelligence (AI), Big Data Analytics and Robotics Process Automation is well known. The critical question is, how this potential can be realized within the own organization. Here, we see two decisive approaches:
- The use of innovative technologies should always be done in close interaction with sufficient internal and external data as well as human process expertise
- The goal should not be to optimize individual process steps in isolation, but to consider holistic workflows (end-to-end) and, based on this holistic view, to embed innovative technologies in the individual steps of the workflow.
If these approaches are successfully implemented, the entire organization will benefit. We at IBM summarize this concept under the term Intelligent Workflows. The added value generated by Intelligent Workflows ranges from an improved customer experience and leaner operating processes to optimized process quality and cost structure. The goal of our new blog series is to illustrate the general idea of Intelligent Workflows using concrete use cases from the financial services sector. In Part I, we look at existing use cases for the intelligent redesign of anti money laundering and fraud prevention processes:
Intelligent Automation of Anti Money Laundering & Fraud Prevention for Banks
The fight against money laundering and fraud is a crucial part of the daily operations of financial institutions. On the one hand, it is a matter of meeting the continuously increasing requirements of the regulators in order to avoid delicate fines. On the other hand, banks have a strong self-interest in detecting cases of fraud at an early stage, in order to avoid the resulting financial and reputational damages. With the advancing digitization of business processes and the increasing volume of transactions, the detection of fraud cases is becoming more and more complex. Experience has shown that banks face the following challenges when screening transactions and business partners:
- High manual effort with monotonous research and input processes
- High manual effort in the review of transactions falsely identified as fraud („False Positives“)
- Lack of transparency and traceability in the internal processing of potential fraud cases
One approach to address these challenges is the redesign of due diligence with exponential technologies such as AI and big data analytics. The following paragraphs outline three concrete use cases already being implemented by banks today:
1. Use Case: Collection of Relevant Data
In addition to the data already available within internal databases, external data, e.g. from sanctions lists, company directories and search engine results, are also relevant for the comprehensive screening of a counterparty in financial market transactions. Our observations show that the collection and structuring of this data is often still performed manually by analysts. Artificial Intelligence today is not only capable of independently aggregating data from various sources and structuring it based on its relevance, but is also able to perform a first deeper analysis of the content based on Natural Language Processing (NLP). In addition to receiving an AI-generated list of relevant research results, analyst can now also be presented with a first assessment of the general sentiment of the content found. Through the targeted integration of AI, the process for analysts thus no longer begins with the time-consuming (and thus costly) research, but with value-adding analysis and evaluation. The example of a leading US bank confirms the enormous savings potential of automated research in practice. By intelligently automating a large part of the manual research and data entry process, the bank managed to reduce the time required for due dilligence by 60%.
2. Use Case: Improved Risk Evaluation
Beyond accelerating data structuring, the use of AI can also help reduce the high rate of transactions that are incorrectly marked as fraud. These false positives currently represent a major problem for many banks. Experience shows that the false positive rate of many institutions is well over 90%. Separating these from actual fraud cases, currently still requires a lot of manual effort and is therefore extremely costly. A concrete starting point to solve this problem is the use of AI-based unsupervised learning. Here, the AI is shown past cases of suspicious transactions – both false positives and actual fraud cases – without any further specifications. The system then independently searches for features that are particularly common either in false positives or actual fraud cases. If the AI identifies a group of transactions that have similar characteristics but are highly unlikely to be fraudulent, appropriate adjustments can be made in the control systems to prevent these transactions from being reported in the future. This process results in a gradual reduction of the high rate of false positives. Moreover, transactions that are still flagged as suspicious can be better prioritized for human analysts. The higher the AI-generated probability that the case is an actual fraud case, the higher it is prioritized for manual processing.
While the approach described in the previous paragraph focuses on improving the screening of individual transactions, in the long run financial institutions should focus their fraud prevention efforts less on the analysis of individual transactions and more on gaining holistic knowledge of their counterparties. One way to do so is to consolidate and structure counterparty data from various sources, such as databases from other business units. Banks that follow this approach with available Big Data & Analytics tools have quickly gained new insights into the structures and behavior of individual customers or customer groups. It will also be important to move away from a static evaluation at a specific point in time to a continuous and automated evaluation of counterparties. This way, conspicuous behavior changes that may indicate criminal activities can be detected faster. With the targeted application of AI to reevaluate and prioritize previously flagged transactions and a stronger focus on the structures and continuous behavior of its counterparties, a bank from the UK not only succeeded in reducing its internal false positive rate by 70%, but also in detecting 50% more actual cases of fraud.
3. Use Case: Increased Traceability and Transparency
Since internal anti money laundering and fraud prevention processes are subject to consistent regulatory audits, it is essential for financial institutions to be able to present past screenings transparently and comprehensibly to external auditors. In the past, we have observed two concrete ways in which financial institutions can add value through intelligent automation at this step of the AML process: On the one hand, it is easier to ensure that all documents that contributed to a specific decision are documented and archived, when they are processed in a standardized and automated manner. This process redesign ensures relevant documentation to be available for audits, that usually take place weeks, months or even years later. In addition to automated archiving, banks have succeeded in using AI components and especially the Natural Language Processing approach to receive first fully AI-generated dossier from the large number of archived documents, providing a structured overview of audit-relevant data.
With regard to the increased transparency and traceability of money laundering and financial crime audits, banks have achieved measurable successes through the targeted use of AI and Robotics in the past. For example, one of the top 20 US banks states that the need to retrace investigation steps has been reduced by 50% since comprehensive audit information is documented in an automatically generated investigation file.
Intelligent workflows ensure that employees have to spend less time on monotonous research and documentation tasks and can thus devote more time to value-adding activities such as analysis and strategic planning. The ultimate goal, as shown with the presented use cases, is to reduce operating costs and processing times. This is especially true for the areas of compliance and risk management, as these represent large cost blocks without direct revenue potential.
The basis for the successful application of exponential technologies such as AI and Big Data Analytics is the efficient use of internal and external data while leveraging existing process expertise. With helpful insights from pioneering industries, banks are now increasingly recognizing the potential of their available data and are beginning to transform their legacy infrastructure and data governance concepts. The more successful these efforts are, the more successfully existing workflows can be automated and optimized.