Why banks struggle to scale AI beyond pilot projects

Banks have used machine learning and advanced analytics for decades, long before generative AI gained attention in 2022. The arrival of large language models raised customer expectations for more natural, conversational banking services. This shift has pushed boards to define AI strategies while regulators demand greater transparency and control. The release of public large language models forced banks to rethink how they interact with customers. Many institutions had already adopted AI in some areas, yet only 7% have managed to deploy it across their entire enterprise.

Scaling AI has revealed deep structural challenges. Successful pilot projects often collapsed when expanded, exposing issues like data fragmentation and inconsistent integration into live processes. The problem is rarely the models themselves but the ability to manage complexity securely and cost-effectively at scale.

European banks now hold a potential edge due to regulations like the EU AI Act and DORA. These rules provide clear parameters for design, helping institutions build reliable and repeatable AI systems. However, economic constraints and governance demands, such as cost volatility and oversight requirements, remain significant hurdles. Banks that establish solid foundations for scalable and secure AI operations will likely lead in the coming years. Success depends on absorbing complexity predictably and within strict economic and regulatory boundaries. The focus has shifted from building models to ensuring they function effectively in real-world conditions.