Author: Richard Harmon, vice president and global head of financial services at Red Hat
The financial services industry has been an extensive consumer of advanced analytics, or predictive AI, for decades. We are now seeing the first inroads in generative AI-powered solutions that will lead to significant improvements in customer service, operational efficiency and resilience. Consistency and standardization in frameworks and platforms is key.
Driving resilient modernization with AI
The global challenge that central banks and regulators face is to ensure the stability of the global financial system. The forthcoming regulatory changes being brought globally, at a staggered pace, will address new types of potential systemic risks, such as cloud concentration risks, AI usage, climate change and future hidden correlated risks emerging from the ever-changing global financial system.
Applying AI to the automation of processes will be key for enabling efficiency and resilience. AI-enabled event-driven automation will help institutions address forthcoming application portability requirements for critical third party ICT. Demonstrating this on a regular basis is a key regulatory requirement of the EU’s Digital Operational Resilience Act (DORA).
On the infrastructure side, the focus continues to be on modernizing legacy applications to become more agile. Replatforming or rewriting software for cloud is an area where generative AI can support, by translating software into different programming languages, or helping to retrain IT staff with coding assistants that can take on basic work. AI can also optimize where applications are deployed according to key business challenges such as sustainability metrics, operational efficiency and cost concerns.
The use of AI can also bring significant benefits to operations and service teams, for example with AI-powered assistants handling more complex customer inquiries or AI that helps automate data aggregation and analysis for more accurate and timely financial reporting.
AI will also play an important role in a bank’s ability to manage risks from short term external macroeconomic shocks to longer term transformational changes such as climate change. AI in these situations requires access to large diverse data sets that can yield new insights for improved predictive capabilities.
The potential of synthetic data
As AI continues to develop it will have the capability to dramatically expand and improve what applications can do. However, there are challenges where data is siloed within institutions due to privacy requirements, intellectual ownership requirements and growing sovereignty requirements.
Fortunately, AI itself will be able to help overcome the lack of high quality, fit-for-purpose data sources: through the creation of “synthetic data”. This is when existing data sources are replicated to anonymize for privacy and make data shareable. This will allow certain applications to break beyond the limitations of purely historical data to develop richer insights into possible future emerging situations.
An area of application for synthetic data is financial crime. If a machine learning or AI model is built from a data set that contains a limited number of cases of a certain type of financial crime, it can predict the recurrence of these specific cases with high accuracy. Adding AI-based simulations generating hundreds of thousands or millions of synthetic scenarios can provide potentially more robust detection systems.
The AI explainability challenge
Beyond data, significant challenges exist in order to make AI models trustworthy and resilient. One key area is a global research effort toward explainability in AI (XAI). The goal of XAI is to develop frameworks to make AI decision-making transparent and accountable to improve trustworthiness, decision-making and regulatory compliance.
On the regulatory side, the recently-passed EU AI Act is the first major effort to build out a comprehensive regulatory framework for AI systems wherein the higher the level of risk, the stricter the rules governing AI compliance. High-risk AI systems have to address a wide range of requirements that include explainability as well as documentation, data governance, human oversight, risk management and auditability. As similar AI-focused regulatory frameworks get created in other regions, there will be the need to ensure some degree of global consistency to encourage continued innovation in AI capabilities and solutions.
The possibilities of AI are almost unlimited, as long as the right teams, processes and solutions are in place. Just as many leaders are considering a hybrid cloud strategy supported by one common platform and automation layer to give them freedom and flexibility, the same applies to AI: giving teams across the business a consistent, centralized experience when training, maintaining, fine-tuning and deploying AI models in production. A holistic environment protects against having too many failure points, and helps standardize testing and validation to meet the control and transparency requirements that are so crucial to secure regulatory approvals as well as get AI to scale.
