Santander cuts AI response costs to three cents

Santander’s AI Lab has reduced the reported cost of generating an artificial intelligence response from €5 to three cents through work with Nvidia. The Spanish bank is testing smaller, specialised language models for specific tasks instead of relying on large, general-purpose models.

In one customer service experiment, Santander reported a 50% reduction in costs and improved conversation quality. The AI Lab tests emerging technologies before the bank considers wider deployment.

Santander and Nvidia have agreed to a three-year framework covering foundation models, AI agents, fraud detection, anti-money-laundering systems and real-time decision-making. The companies are also developing methods to assess which models and AI agents suit particular tasks.

Santander has built a system called Model Arena to compare open-weight AI models on banking tasks. It is using Nvidia’s NeMo Customizer to fine-tune models with the bank’s own data for specific types of work.

“Cost per intelligence output is a key metric,” Aser Blanco, Nvidia’s global banking lead, said. He described Santander’s approach as using smaller models trained on the bank’s data, tested against banking tasks and operated at a scale suitable for a global bank.

Iñaki Bernal, Santander’s chief AI officer, said the companies had the infrastructure, software, research and operational scale to develop AI applications for banking services. Santander has not disclosed how many customer service interactions were included in the cost comparison.

Santander announced a target in February to generate €1 billion in business value from AI between 2026 and 2028 through lower costs and higher revenue. Ricardo Manjón, the bank’s chief data and AI officer, said Santander generated €35 million of that value in the first three months of 2026 and expects to reach €200 million by the end of its financial year.

The bank plans to focus on a limited number of AI applications, measure their financial effect and expand those that produce results across the group, Manjón said.

In October, NatWest Group said it was using Nvidia’s AI technology in a voice-based customer service tool. The Spending Insights Tool is being tested with 75% of customers of NatWest’s Royal Bank of Scotland brand. It is designed to recognise UK accents, pronunciations, everyday language about money and financial terms. The system combines Nvidia’s technology with NatWest’s research and customer data.