From data to decisions: AI and supervision

Article by Elizabeth McCaul, Member of the Supervisory Board of the ECB, for Revue Banque

26 February 2024

European banking supervision is committed to exploring the potential of AI to make the work of supervisors more efficient

In today’s digital age, new data are being generated at an unprecedented and exponential pace. The question is no longer about whether or not to use artificial intelligence, but rather about how it can be used most effectively and responsibly. AI offers enormous opportunities, promising to drastically improve both the efficiency and quality of a wide variety of work-related processes. It can analyse vast amounts of data quickly and accurately, improve risk identification by detecting patterns in the data, support decision-making and automate repetitive tasks – all of which can enhance the work of banking supervisors. However, we also know that using AI entails risks which are not yet fully understood.

Banks are also facing the same dilemma. AI can improve the experience they offer their customers, increase their operational efficiency and strengthen their risk management processes. But challenges and risks abound as they seek to leverage AI’s capabilities – from data governance risks (concerning, for example, confidentiality and the reliability of training data) to emerging operational, model management and accountability risks. Banks are increasingly finding that, in order to remain competitive, they must embrace AI while delivering on their risk management responsibilities.

So what does all this mean for banking supervisors?

The short answer is that we must adopt a future-proof approach to understanding and using AI. We should use it to enhance our internal supervisory capabilities and gain greater insights into the risks facing supervised banks as they, in turn, also deploy AI. These risks are wide-ranging, affecting business models, governance frameworks, risk management processes and capital adequacy, as well as financial stability more broadly. Crucially, the role of ECB Banking Supervision is to ensure that banks remain safe and sound. It is not for us to dictate which business models banks adopt and which technologies they use. What we can do, however, is draw on the power of AI to decipher data, understand risks and speed up processes, freeing up more time for human analysis and judgement in an increasingly complex world.

At an early stage we recognised the need to embrace digital innovation and AI to make European banking supervision more efficient and effective. We introduced an ambitious digital agenda to enhance our analytical capabilities. We invested in a portfolio of supervisory technology (SupTech) applications to supervise a complex banking sector and manage an ever-expanding pool of data and tasks. And we focused on the people who would be using this technology – 14 applications and platforms have been developed in the past three years, serving more than 3,500 users across the ECB and the national supervisors.

Currently, our AI applications enable us to query supervisory data and employ chatbot functionalities for supervisory regulations and methodologies. In the field of textual analysis, for example, our Athena application translates and analyses the content of supervisory documents. It can combine this content with information from other sources, such as public media, allowing supervisors to augment their insights into banks and their risks.

In the field of big data analytics, GABI generates and optimises regression models on a large scale, enabling supervisors to base their analyses on a much larger set of models and draw more insightful comparisons than in the past. Of course, these models are still checked by human beings.

And in the field of network analysis, NAVI generates network diagrams to illustrate relationships in the data. This enables users to visualise the often complex ownership structures of supervised banks, and combines data from numerous sources to provide a comprehensive overview of bank owners and interdependencies.

Other more targeted SupTech tools include Heimdall, which supports experts in processing the vast amounts of information received to assess the fitness and property of members of the management body, and Medusa, which facilitates the drafting and consistency checks of reports following internal model investigations.

Our investment in technology has allowed us to build cutting-edge infrastructure to scale-up AI applications through the Virtual Lab, a cloud-based collaboration platform offering machine learning capabilities and a user-friendly environment for sharing and developing code. Not only has the Virtual Lab enhanced collaboration within European banking supervision, it also enables the deployment of technology such as generative AI (a type of AI made popular by applications such as ChatGPT).

Generative AI has the potential to support the work of supervisors by facilitating their day-to-day tasks. In 2023 we collected more than 40 potential use cases from supervisors and have developed several proofs of concept that demonstrate the potential of generative AI. The use cases we have sought to address include instantaneously retrieving answers to questions on supervisory methodology, with clear references to internal methodologies, and automatically translating queries written in plain language into code in order to find specific data points.

The second of those use cases is particularly interesting. Agora, our single data lake for European banking supervision, currently requires users to have some understanding of programming to access the database. But with the help of generative AI, which can automatically translate natural language queries into scripts, supervisors with no programming experience can ask Agora where to find very specific data points. This is just one example of how generative AI, and AI in general, can make traditional technologies easier to use and put SupTech at supervisors’ fingertips.

Of course, it goes without saying that these tools are not intended to replace supervisors. Human judgement and expertise are and always will be key to ensuring a reliable outcome.

Fostering a digital culture across our organisation is crucial to our success. We regularly organise dedicated training courses for both ECB and NCA staff with globally recognised providers such as Coursera and INSEAD Business School. The goal is to enhance our supervisors’ digital skills and raise awareness about recent technological developments and the latest relevant regulations (such as the EU’s new AI Act). We also organise an annual conference that brings together leading digital experts from the supervisory community, academia and industry to strengthen our partnerships and facilitate the development of best practices.

When developing and implementing AI we are obviously mindful of the associated risks. For instance, if we are to maintain trust in AI tools it is essential that they are transparent and that we can explain how they work, given the potential “black box” nature of this technology. For this reason, we are working hard to provide strong organisational support and clear guidelines for the use of AI in banking supervision. At the same time, we are continuously strengthening our IT security to ensure we can safely host AI tools.

Looking ahead, we will continue investigating the possibilities and challenges of using AI, in cooperation with supervisory authorities across Europe. We aim to harness the technology’s full potential to make our supervisory work as efficient and effective as possible.