Artificial Intelligence has been increasingly playing a crucial role in predictive and behavioural analytics.
In a market context where the domain of available technologies expands along with available data, the increased accuracy of algorithms has become more impactful than ever in driving business decisions. Indeed, risk management and compliance are key areas where AI can make a difference in the way Financial Institutions identify, measure, and mitigate the risks they are exposed to. Financial and non-financial data can now be analysed with enhanced granularity and deeper analysis, helping organizations to make more informed decisions for their business.
AI and Intelligence automation predict issues before they occur. Their adoption has been observed to different extents in several areas, particularly in lending, regulatory compliance, and cybersecurity among others. In the credit context, lending is a big data problem at its core, hence naturally suited for machine learning. The more data is available about a borrower and similar customers, the better financial institutions can assess their creditworthiness. AI technologies can come into aid for the organization’s monitoring capabilities in corporate governance and regulatory compliance through Early Warning Systems that evolve into Early Learning Systems, which ultimately aim at preventing threats from materializing long before their predecessors could do. AI is extensively used also in cybersecurity to identify threats by detecting software with specific features and repel the attack accordingly (for example, threats are identified whenever large amounts of data and processing power are consumed).
When defining more technically what Artificial Intelligence is about its application to risk management, we can refer to a suite of statistical techniques that combine non-traditional data, large data sets, black-box models (with complex relationships among variables), and models with timeline structures that vary rapidly.
The domain of technologies available is wide, but it can be summarized into six clusters, namely.
- evolutionary programming (i.e. genetic algorithms, ant colony optimization)
- network segmentation “NS” (i.e. neural clustering)
- machine learning “ML” (i.e. deep learning and gradient boosting)
- natural language processing “NLP” (i.e. text analysis and screen scraping)
- cognitive RPA
- Visual analytics
Evidence shows that the broadest application of AI technologies in risk management and compliance stays with segmentation analytics, natural language processing, and machine learning.
To give a quick overview of the three most used technologies mentioned above, machine learning identifies the subset of capabilities that enable machines to learn by experience and acquire skills without human involvement. Deep learning and gradient boosting are broadly used techniques in this field.
As the name suggests, natural language processing is the capability of analysing and deciphering human language from written documents and verbal communication, which is recorded, extracted, categorized, and organized.
Segmentation analytics encompass the capability to predict behavioural patterns from customers with similar attributes. This is particularly useful in the lending space, as sound customer segmentation can lead to minimum exposure to credit risk. AI advanced analytics tools come into aid in the complex customer segmentation process, by automating the division of customers into smaller segments elaborating demographics (i.e. gender, race, age, occupation, income, etc.), psychographics (i.e. lifestyle, values, attitudes, etc.), geographic (country, region, state, zip code, etc.), behavioural (i.e. product usage rate, brand loyalty, etc.) and past business history variables (i.e. credit default records, purchasing history, etc.).
Looking at the adoption of risk & compliance-focused technologies in banking, we can look at wealth management, retail banking, ERM, and financial crime. Wealth management encompasses a multitude of business contexts which range from mass affluent services to fund management of ultra-high net worth individuals.
AI-based portfolio optimizations have become extremely popular as AI risk management capabilities in this business line. As an example, Aladdin by BlackRock can use machine learning to enable individual investors and asset managers to assess the levels of risk or returns in a particular portfolio of investments. Aladdin can automatically monitor over 2,000 risk-related factors per day (like interest rates or currency rates) and test portfolio performance under different economic conditions. Banks can use Aladdin to augment the capabilities of their human investment managers by giving them the capability to predict the performance of portfolios in real time.
Looking at AI solutions for retail banking, banks are getting benefits in behavioural modelling and credit risk calculation. Behavioural modelling includes checking for financial crime, managing credit, fraud, and addressing ALM and treasury needs for assets retail book. Machine Learning and Natural Language processing drive insights and minimize negative customer interactions across the retail sales lifecycle. AI is also offering interesting enhancement in early warning and risk reviews through augmented external risk factors.
Companies like GiniMachine and Affirm offer solutions in credit scoring. GiniMachine executed pilot projects to build accurate scoring models with minimal data points and without access to an applicant’s credit history. Some of the most promising and predictive parameters included the size of their company, the applicant’s industry and occupation, the total years they had been in business, the size of their family, and data from social networks ranging from their overall activity to the quantity and quality of their connections. GiniMachine has proven that it is possible to capitalize on information about borrowers that is collected from alternative sources to assess borrowers’ credibility and make effective lending decisions accurately and efficiently.
Moving to Enterprise Risk Management, AI can make a difference in non-regulated use cases like early warning signals generation and “what if” analyses. AI can construct a multitude of stress and scenarios library to scan market data points, scan benchmark results, and list areas of concern.
Kuliza’s Operational Intelligence suite is an example worth mentioning. It includes EWS that allows lenders to predict whether a default is likely to occur a specified number of days in advance, with a reasonable amount of confidence. The system is powered by a set of trained models that are available for lenders to apply to their portfolios to get accurate results for each account and supports multiple structured datasets, allowing data scientists to select the appropriate parameters for developing the model. The system runs the parameters through all selected models and provides the accuracy, precision, and other scores generated for each, allowing users to select the most suitable model based on their requirements. The outputs of the system can be integrated into existing loan management and/or risk management systems.
With respect to financial crime and compliance, AI has brought benefits to anti-fraud, AML and cybersecurity applications and increasingly to KYC.
Besides risk profiling, ID management is a capability within AML and KYC, which is central to financial crime. Customer profiles are enriched, and their information is used to tackle payment frauds, for example, by linking customers to their devices and identifying anomalies in performed transactions. Graph analytics can provide valuable views on the interrelationships between a variety of corporate entities, devices, individuals, and activities that wouldn’t be visible with standard statistical methods. Many types of financial crime analysis leverage large-scale automated graph management, also to address payment frauds.
Alert management is another capability that comes into aid in financial crime. ML allows prompt interventions for organizations to report and control suspicious activities, as it offers agility to respond to transaction monitoring alerts on a risk-based prioritization.
Eventually, a broad range of AI tools like NLP, ML, and graph analytics are useful for case investigation and suspicious activity report management (SAR), which is labour-intensive and can require a broad range of data sets - hence AI tools accelerate the investigation process.
Within financial crime and compliance, AI comes into aid also to fraud analytics. Besides generating risk profiles and appetites for all clients, pattern recognition and predictive modelling are other two valuable capabilities in the fraud analytics field.
They link behavioural patterns to specific parameters (variables estimated from a dataset, which define a model and its conditions) and properties. Notable companies include Trifacta and Digital Reasoning Systems. Trifacta’s data-wrangling solutions help data analysts accelerate the detection of suspect behaviour patterns by providing a superior approach to directly discover and transform complex, noisy data sets for analysis. Leveraging the full potential of Hadoop to cost-effectively store large, complex transaction data sets, Trifacta removes legacy inefficiencies and coding requirements enabling analyst teams to directly discover and transform data themselves. Banks can gain a holistic view of their customers’ relationships, accounts, transactions, and channels from various sources to differentiate fraudulent activity from normal behaviour.
A Hadoop “data lake” allows analysts to combine historical data with new data types to investigate patterns of abuse over time. When fraud occurs, the data lake provides an opportunity for investigators to work with a larger pool of diverse internal and external data sets to explore anomalies and find correlations and patterns that can assist in preventing future breaches.
To read more such insights from our leaders, subscribe to Cedar FinTech Monthly View