The launch of DeepSeek’s R1 open-source model has prompted fresh discussions about the future of AI in financial services, particularly around the potential for smaller language models to reshape how firms approach AI adoption.

Chris Probert, partner and global head of data and generative AI at Capco, suggests that the release of R1 highlights both the rapid pace of AI innovation and the increasing viability of smaller models as an alternative to large language models (LLMs).

SHIFT TOWARDS SMALLER LANGUAGE MODELS

Probert notes that firms previously saw LLMs as the dominant AI solution, but R1 now provides an alternative that could better align with financial services use cases. Smaller models offer firms the ability to leverage their own training datasets, avoiding reliance on externally controlled datasets from large AI providers.

The ability to use proprietary training data could have several advantages, including:

• Developing more tailored models to accommodate customer and compliance requirements across different markets
• Enhancing data security and privacy by maintaining greater control over data storage and processing
• Gaining a competitive edge by building AI models that utilise unique internal datasets
• Improving transparency and explainability in AI decision-making, which is increasingly important in a regulated industry

AVOIDING VENDOR LOCK-IN AND FUTURE-PROOFING AI STRATEGIES

The emergence of R1 also reinforces the argument for firms to adopt an agnostic approach to AI development. Probert highlights the risks of vendor lock-in, advocating for a strategy that allows firms to swap models easily, ensuring adaptability as AI technology evolves.

“Firms should focus on scalable enterprise solutions that allow easy model swaps, providing flexibility while also minimising transition costs. By embedding modularity and interoperability into the solution architecture early on, organisations can future-proof their AI investments against rapid advancements in the field,” he said.

Probert also suggests that financial services firms will increasingly adopt a “model of models” approach, selecting AI systems based on benchmarking against specific business needs. This method enables firms to optimise AI performance while ensuring compliance and maximising return on investment.

KEEPING PACE WITH AI ADVANCEMENTS

The release of R1 has been described as AI’s “Sputnik moment,” signalling a period of accelerated change. Probert believes financial services firms will face the challenge of keeping pace with AI developments while maintaining flexibility in their adoption strategies.

“To build for adaptability and high pace of change you can’t be locked into a single vendor, and a ‘model of models’ approach will be the optimal path forward. Robust model benchmarking will be crucial, allowing financial services organisations to evaluate which AI models best align with their specific use cases, maximise performance, and deliver the highest return on investment.”