Models are inherently complex tools that underpin board reporting, regulatory compliance and financial decision making. In the UK, models must adhere to the Prudential Regulation Authority's (PRA) requirements outlined in SS1/23, effective since May 2024.
Many international firms need a consistent framework to enhance comparability and support informed financial decision-making at a group level. This encompasses international approaches such as the US’s longstanding SR11-7, as well as evolving principles and standards from the UAE, Canada, India, and Japan. Regardless of the region, model risk rules broadly align to promote sound governance, effective risk management, and strong internal controls.
Achieving widespread compliance and ensuring effective model risk management is a substantial challenge. Forward-looking firms can leverage emerging technology to monitor these models throughout their life cycle across key areas, such as governance and oversight; model risk identification and classification; model development and implementation; validation and independent review; performance monitoring, change management and documentation. This approach ensures integrity, reliability and responsible use across these areas.
Successful model risk management hinges on a comprehensive understanding of the model universe, including the identification of all models, their respective purposes, and their criticality to the firm. A model inventory provides that crucial central authority, outlining model methodology, assumptions, limitations, documentation and validation reports.
However, many model inventories still rely on manual processes, which heightens the risk of incomplete data, poor traceability and weak controls. Embracing emerging technology, such as generative AI, is essential to enhance the model inventory and boost model risk controls. Firms that don’t keep pace with these advancements may encounter challenges with regulatory compliance and could find themselves ill-equipped to make tough business decisions.
Generative AI can use digital fingerprinting to identify models that may be missing from the existing inventory. This includes automated validation checks to ensure that metadata – such as core repository references, or model names in documentation – is complete and accurate. Tasks like this have historically been carried out manually, increasing the risk of potential errors or omissions.
It’s also essential to maintain other organisational databases, project management systems and financial records to track where and how teams are applying these models. AI can cross-reference a wide range of sources against the central model inventory for consistency across all systems.
First and foremost, firms need a transparent record detailing how a team intends to use a model. This is crucial, as many models are repurposed for uses they weren't originally designed for, thereby introducing significant risks. Machine learning, which is more quantitative in nature, is being used to look at trends in usage patterns to identify disparities between predicted and actual use, promptly flagging potential misuse. It can also track underlying data sources and generate automated alerts when these sources change. This gives senior stakeholders the information they need to intervene and evaluate whether a model should be retired or replaced.
Generative AI, being language based, can create code and produce adaptive models that automatically adjust based on evolving data flows and market conditions. This is useful for scenarios where inputs regularly need updating, allowing firms to allocate resources to more complex tasks.
Visualisation tools, such as PowerBI or Tableau, use heatmaps to pinpoint redundant models or make better use of existing ones. They can also identify any hotspots, where a model may be experiencing unusually high rates of usage. This functionality can help firms with their risk assessment and make it easier to identity critical models for greater oversight. Creative use of emerging technology and AI in financial services can create opportunities to mitigate model risk and ensure compliance.
Many firms use version control systems to monitor technical changes and support model risk management, but there's room for further enhancements. By leveraging generative AI and machine learning, firms can automate change logs and create new versions in response to significant changes in the model or its parameters. The automatic tracking of version management will provide deeper insight into model adaptation, improving traceability and reproducibility to support model risk management processes.
Machine learning serves as a valuable tool to detect unusual patterns or discrepancies within the model inventory. Visual dashboards can identify any exceptions, changes or new models and send real-time alerts to all relevant stakeholders. Generative AI can build on this, with dynamic alert thresholds based on trigger criteria that evolves in line with changing model use. This approach minimises the likelihood of false alarms and allows model owners to address potential issues more promptly.
Effective model risk management is crucial for informed financial decision making, and is a regulatory imperative. Therefore, leveraging emerging technology to demonstrate robust oversight and governance is essential. Machine learning can simplify these processes by automating inventory audits to identify any documentation gaps. For example, it can assess a model’s digital footprint and automatically prompt model owners for necessary supporting information. Additionally, teams have the flexibility to run these audits on demand, tailored to firm’s size, business activities and risk profile.
To successfully integrate emerging technology and generative AI into model risk management, firms must ensure widespread buy-in. This may require a culture shift to demonstrate the value of these tools and their capacity to address specific pain points. Targeted training through workshops, online learning and interactive sessions will help individuals’ understanding of their role and key use cases.
When assessing the use cases for emerging technology and generative AI, it’s essential to recognise their dual applications. On the one hand, these tools can enhance the accuracy and completeness of the model inventory, thereby supporting broader model risk management through greater oversight. On the other, they can demonstrate regulatory alignment to SS1/23’s model identification principle, demonstrate the governance framework and facilitate validation processes.
With emerging applications of AI in financial services, it’s crucial to strike a balance in the use of these capabilities and mitigating the associated risks. Firms that can achieve this equilibrium effectively will gain greater confidence in their financial decision making processes and be better equipped to make bolder choices for long-term growth.
For insight and guidance on the use of generative AI in financial services and to support model risk management, contact Vivian Lagan and Miles Davis.
