November 29, 2024
We were delighted to attend the 8th SEE NPL + Finance Forum in Belgrade for the first time last week. Invited by the World Bank and the event organizers, we had the opportunity to discuss the transformative potential of AI in credit management and non-performing loan (NPL) sales. Below, we summarize the key insights from the forum on NPL-related discussions.
Unintended Consequences of New Regulation
In recent years, countries in Southeastern Europe and the Western Balkans have seen a sharp decline in NPL ratios, reaching historic lows. For example, Serbia’s NPL ratio, which stood at a staggering 23.0% in 2014, dropped to just 2.9% by mid-2024. Similar trends were observed across Southern Europe. While portfolio sales are rare, we heard of specific interest in selling single positions.
A key focus of the forum’s NPL panel was the impact of the new EU Directive on Credit Servicers and Credit Purchasers. Major debt collectors, such as EOS Matrix, highlighted the temporary stagnation in NPL activity caused by the directive in local markets like Croatia and Serbia. According to data from the Croatian National Bank, average annual NPL sales between 2020 and 2022 were approximately €200 million. However, in 2023, sales plummeted to €42 million, with only €10 million transacted in the first nine months of 2024. Encouragingly, Croatian banks have recently resumed sales, with expectations of increased activity in the fourth quarter.
Although Serbia is not directly bound by the EU directive, it tends to align its regulatory framework with EU developments. That said, regulatory heterogeneity remains high in the region, and NPL markets are typically localized, with limited cross-border activity. For example, Serbia prohibits the sale of loans to private individuals to non-bank entities, significantly constraining its NPL market. Conversely, such sales are allowed in neighbouring EU countries like Slovenia and Croatia.
The implementation of the EU directive in Croatia was described by panelists as a missed opportunity to stimulate NPL market growth. Croatia has introduced particularly stringent consumer protection measures, requiring credit servicers to provide detailed, transparent information to borrowers regarding loan transfers and servicing, while also ensuring strict compliance with data privacy laws. These measures temporarily discouraged banks from participating in the Croatian NPL market.
Nonetheless, there was broad consensus among panelists that the regulation of credit servicers ultimately benefits the public. Over time, it is expected to bolster banks’ confidence in selling NPLs to regulated entities, thereby mitigating reputational risks. Transparency measures introduced by the EU directive—such as the EBA NPL transaction templates—were also welcomed, with expectations of increased investor interest and improved NPL pricing.
AI Use Cases in Credit Management and NPL Sales
We presented several case studies showcasing the use of AI in NPL sales and credit management. While machine learning has long been used in the financial sector to analyze structured credit data and predict credit events, the adoption of Generative AI for managing unstructured data and documents remains limited.
We demonstrated how AI can deliver measurable benefits in automating key processes related to NPL sales and credit management. Here are some notable use cases:
1. Data Preparation, Validation, and Migration
AI supports the seamless transfer of credit data into internal data warehouses or from sellers to buyers as part of NPL portfolio transactions.
2. Smart Virtual Data Rooms for NPL Sales
Advanced AI tools streamline the organization and validation of complex loan portfolio sales involving millions of unstructured documents. Capabilities include automated index generation, file type validation, borrower and loan identification, and key value extraction to enrich loan data tapes.
3. Automated Financial Covenant Monitoring
AI enables automated workflows for monitoring financial covenants. This includes borrower portals for collecting updated financial statements, extracting legal clauses to identify covenant definitions, analyzing financial statements, calculating covenant KPIs, and integrating all data into early warning systems.
4. Synthetic Data Creation for Secure Sharing
Machine learning and language models can generate synthetic data to facilitate secure internal and external data sharing, while ensuring compliance with confidentiality, data protection, and bank secrecy regulations. The European Commission (https://digital-finance-platform.ec.europa.eu/data-hub) is working on a central data hub based on synthetic data that may benefit NPL markets in the medium to long term. We encouraged the participating central banks in the forum to engage with the Commission on this important topic.
Looking Ahead
Across Europe, NPL sales are at a cyclical low, primarily due to historically low NPL stock levels. While the introduction of the EU Credit Servicer Directive initially created uncertainty, market participants at the SEE NPL + Finance Forum expressed optimism about its long-term impact.
The directive is expected to increase transparency, foster investor confidence, and stimulate NPL markets over time. Furthermore, AI has the potential to significantly reduce transaction costs and accelerate processes, making it a critical enabler in the evolving landscape of credit management and NPL sales.
