Santander has partnered with Nvidia to develop new AI capabilities for banking, focusing on evaluating and deploying specialized AI models. The collaboration, announced on Monday, aims to determine the most efficient AI models for specific banking tasks, reducing costs while maintaining performance. Santander has shifted from large general-purpose models to smaller, task-specific models, cutting serving costs from €5 per interaction to approximately €0.03 in selected experiments.
The bank's AI lab has created a 'Model Arena' to assess open-weight models against banking tasks and identify the best model for each workload. The lab is also specializing in models using Nvidia's NeMo Customizer. The partnership has already shown success in a customer-service experiment, improving conversation quality and reducing costs by over 50% through a more efficient AI model.
Looking ahead, Santander and Nvidia plan to continue collaborating on foundation models, AI agents, fraud prevention, real-time decisioning, and advanced simulation technologies, including banking digital twins. Aser Blanco, Nvidia’s global banking lead, emphasized the importance of cost per intelligence output, highlighting Santander's achievement in deploying smaller, specialized models trained on the bank's own data.
Recent research by PYMNTS Intelligence indicates that 95% of large financial services firms have widely deployed or embedded newer AI tools. However, challenges remain, with 30% of companies citing data quality and fragmentation as obstacles to further AI deployment.