Adopting large language model-based AI systems (LLMs) presents significant opportunities for NatWest Group (NWG) to improve customer service, marketing personalisation, and operational efficiency through applications like chatbots and content generation.
Specific use cases where NatWest is integrating LLM-based AI systems include; Cora, NatWest’s customer chatbot that supports millions of customers every year; Ask Archie, NatWest’s internal colleague chatbot that supports our 60,000 employees; and our personalised messaging to customers and communication testing projects.
However, as a large financial institution, NWG faces challenges in developing robust, trustworthy AI systems that comply with regulations. These systems must be aligned to consistently adhere to the bank's style guidelines, tone of voice, and business rules without generating biased or illegal content. Additionally, large models pose technical challenges around transparency, explainability, and bias mitigation to ensure fair recommendations and communications that adhere to regulations like the FCA’s Consumer Duty rules.
A key challenge is developing and integrating effective information retrieval capabilities that are crucial for providing AI systems with relevant, accurate data to generate high-quality outputs. Additionally, training LLMs to consistently adhere to business rules and tone of voice requires deep technical expertise and extensive, high-quality training data, which can be difficult to source.