Central banks have long been early adopters of artificial intelligence (AI) and machine learning (ML), using these technologies to generate insights for statistics, research, and policy well before AI became a popular topic.

A new report by the Bank for International Settlements (BIS) examines how these authorities are leveraging AI, highlighting four main areas where the technology is being utilized: data and statistics collection, macroeconomic and financial analysis, payment system oversight, and supervision and financial stability analysis.

According to the report, over 85% of central banks in advanced economies (AEs) are using big data for economic research, a share that stands at about 70% for central banks in emerging market economies (EMEs). These figures make it the most common area of adoption.

Financial stability is the next most frequent area of use, with approximately 80% of AEs and 55% of EMEs applying big data techniques. This is followed by monetary policy, adopted by around 75% of AEs and 45% of EMEs, and statistical compilation, with usage rates of roughly 70% and 65%, respectively.

Data collection and statistical analysis

Around the world, central banks are collecting data from a large variety of sources both to use internally and make available as a public good. To ensure high-quality data for analysis and reporting, many authorities are increasingly using ML techniques.

One notable example is the use of isolation forests, a variation of random forests, on large and granular data sets. These models are highly scalable and can identify outliers regardless of the shape of the data’s distribution, making them valuable for detecting anomalies in derivatives data or benchmarking outliers identified through manual analysis.

Macroeconomic and financial analysis

Central banks are also using ML for macroeconomic and financial analysis, particularly in nowcasting where the goal is to estimate current economic conditions to support monetary policy. Because access to timely data remains a bottleneck, ML models are helping bridge this gap by seamlessly transforming unstructured data into structured and high-frequency indicators, allowing central banks to track inflation expectations through social media sentiment, construct economic sentiment indices from financial news, and decompose inflation drivers using neural networks, among other applications.

The use of granular sources also deepens insights into sectoral and regional trends. For example, ML algorithms can mine job posting databases or e-commerce portals to trace wages and hiring across occupations and industries, shedding light on technology-driven displacement, re-employment speeds and wage formation.

Supervision and financial stability

AI and ML are also transforming how central banks carry out supervision and financial stability functions. These tools are supporting tasks such as document processing, knowledge management, and document review, helping analysts process large volumes of information including news articles, internal bank documents, and supervisory reports. They also enhance market surveillance, allowing authorities to seamlessly detect patterns in trading data or market sentiment that may signal volatility or asset bubbles.

More than 20% of supervisory authorities are using AI to automate supervisory process, according to a recent survey. Around 15% apply AI in financial risk assessment, while approximatively 11% use the technology for environmental, social and governance (ESG) reporting, and risk horizon scanning.

AI-driven supervisory platforms

To harness the full potential of AI, the report notes that a growing number of central banks around the world are developing unified AI platforms to manage unstructured supervisory data.

The European Central Bank (ECB)’s Athena, for example, consolidates public and supervisory documents into a single system, facilitating supervisory analysis of unstructured data, and enabling sentiment analysis and topic identification.

In the US, the Federal Reserve’s Language EXtraction Engine (LEX) improves supervisors’ access to pertinent material scattered across millions of files and shortens review times.

Finally, in Brazil, the central bank has developed ADAM, a ML tool that’s capable of examining the entire credit portfolio of a supervised firm and identify credit exposures with inadequately recognized expected losses. The solution can review 3 million exposures to customers in just 24 hours, a task that would take a team of 10 experienced inspectors about 30 years.

BIS Innovation Hub’s AI projects

To support central banks and financial supervisors, the BIS Innovation Hub (BISIH) has explored various AI applications across diverse domains, including ESG reporting, regtech, and cybersecurity.

Project Neo, a collaboration between the BISIH Swiss Centre and the Swiss National Bank (SNB), aims to examine how AI and granular data from companies can support central banks in monitoring economic activity, generate economic insights and enhance short-term forecasts of key macroeconomic statistics, such as inflation, GDP, and consumption.

Project Gaia, developed by the BISIH in collaboration with the Deutsche Bundesbank and the ECB, explores how AI-driven text extraction can provide high-quality, accessible data at scale for a wide range of applications in money and finance. One focal point is climate-related risks, where a lack of global reporting standards currently makes comparison difficult.

Project Aurora, led by BISIH’s Nordic Centre, applies AI, ML privacy enhancing technologies and network analysis to develop new ways of combating money laundering. Phase one of the project, which concluded in 2023, demonstrated that these technologies can detect up to three times more complex money laundering schemes and reduce false positives by up to 80%.

Finally, Project Raven, another project led by BISIH’s Nordic Centre, aims to use AI to help authorities comprehensively assess the cyber security and resilience maturity readiness of their countries’ financial systems, identify areas of action and trends over time, and streamline regulatory reporting through intelligent automation.

Privacy, cybersecurity, talent shortage among top challenges

Despite these advances, central banks face a number of challenges in adopting AI and ML, especially in areas including privacy, cybersecurity, and algorithmic bias. They must also balance trade-offs between using external versus internal AI models, as well as in collecting and providing in-house data versus purchasing them from external providers.

Hiring talent is another critical challenge, with nearly nine in ten central banks reporting heavier recruitment headwinds, especially for cybersecurity, IT, fintech and AI roles.

Many are closing the gaps by blending permanent staff with consultants, contractors and remote specialists, and by emphasizing their public interest mission, unique data assets and training opportunities to candidates who might otherwise choose the private sector.

Central banks have steadily increased their budgets for AI and ML initiatives over the past years, and will continue to do. Three years ago, only about 2% of central banks allocated more than 5% of their total budget to AI and ML projects. Today, that proportion stands at about 10%. Over the next three years, this share is expected to exceed 40%, with more than 5% of institutions projected to dedicate over 10% of their budgets to AI and ML initiatives.

Featured image: Edited by Fintech News Switzerland, based on image by freepik via Freepik