Why Banks Struggle to Scale AI Beyond Pilot Projects

A new report highlights the challenges banks face in moving AI from pilot projects to full-scale deployment. While 88% of organisations use AI in at least one function, only 7% have rolled it out enterprise-wide. The gap stems from the complexity and scale of production systems in banking. Financial institutions have long used machine learning and advanced analytics, predating the rise of generative AI. However, the public release of large language models shifted customer expectations, demanding smarter interactions with banks.

Deploying AI into live banking processes has exposed issues like data inconsistencies and latency. Traditional analytics can tolerate these flaws, but AI cannot. The primary barrier to success is not weak models but fragmented or low-quality data at scale.

Boards have come under pressure to define an AI strategy quickly. Regulators have also tightened oversight, focusing on explainability and resilience. As AI grows, economic and governance challenges arise, including volatile costs from consumption-based infrastructure and the need for embedded oversight.

European banks may have an edge in AI deployment. Regulations such as the EU AI Act and DORA provide clear guidelines for design and implementation. Institutions that can operate AI reliably and at scale are likely to lead in the coming decade, not just those that experimented early. The banking sector continues to grapple with the shift from AI trials to widespread adoption. Success will depend on overcoming data fragmentation and ensuring consistent, high-quality data. Clear regulatory frameworks and scalable operations will shape the future of AI in banking.