The Transformative Role of Generative AI in Banking, with Potential Risks for Customer Support
In the ever-evolving world of banking, the use of artificial intelligence (AI) has become increasingly prevalent. One area where AI is making a significant impact is in the realm of generative AI, which is transforming the way banks like Citi make data-driven decisions. However, while generative AI has proven to be a powerful tool for internal processes, its potential risks have led Citi to exercise caution when it comes to customer support.
Promiti Dutta, head of analytics technology and innovation at Citi, spoke about the transformative power of generative AI during VB’s AI Impact Tour in New York. She highlighted the paradigm shift that occurred when data science and analytics were put at the forefront of decision-making, thanks to generative AI. This shift created a culture of curiosity around data, with employees across the organization proposing AI projects.
Citi has identified three main areas where generative AI can have a meaningful impact. The first is “agent assist,” where large language models (LLMs) provide call center agents with summarized notes about customers or help them find information quickly. While not customer-facing, this application still enhances the customer experience by providing agents with valuable insights.
The second area is automating manual tasks through LLMs, such as summarizing compliance documents and helping employees locate specific information. This automation streamlines processes and increases efficiency within the bank.
Lastly, Citi has created an internal search engine that centralizes data and allows analysts and other employees to derive data-driven insights more easily. The bank is now integrating generative AI into this tool, enabling employees to use natural language to create real-time analysis.
Despite these successful applications of generative AI internally, Citi remains cautious when it comes to external use, particularly in customer support. The inherent quality of generative AI to “hallucinate” poses risks when precision is crucial. Citi recognizes that a wrong suggestion or recommendation could erode customer trust and interest in the bank.
While some industries may find it acceptable to use LLMs for external communication, Citi believes that maintaining a human in the loop is essential. The bank aims to learn the limitations of the technology and gradually expand its use as it matures. Citi acknowledges the need for thorough testing and proofing due to the highly regulated nature of the banking industry.
In contrast to Citi’s cautious approach, Wells Fargo has embraced generative AI for its Fargo virtual assistant, which provides answers to customers’ everyday banking questions. Wells Fargo’s use of LLMs in its virtual assistant and other products demonstrates a different perspective on the risks and benefits of generative AI.
Generative AI has also prompted Citi to reevaluate its use of cloud resources. While the bank has traditionally preferred on-premise infrastructure, it is exploring the use of OpenAI’s GPT models through Azure’s cloud services. Additionally, Citi is exploring open-source models that allow the bank to bring models in-house and utilize its on-premise GPUs.
To ensure responsible use of generative AI, Citi has established an internal task force that reviews all projects related to the technology. This rigorous process involves the bank’s chief executive, Jane Fraser, and other executives to ensure financial investments are made responsibly and that customer safety is prioritized.
Microsoft, a sponsor of VB’s AI Impact Tour, commends Citi’s approach to generative AI. Sarah Bird, Microsoft’s global head of responsible AI engineering, acknowledges the ongoing work to fix instances where LLMs make mistakes. Microsoft is investing time and effort into improving the effectiveness of grounding errors, which occur when LLMs provide incorrect information.
As generative AI continues to transform the banking industry, it is clear that both its benefits and risks must be carefully considered. Citi’s cautious approach to external use reflects the need to prioritize customer trust and the importance of responsible AI deployment. The ongoing advancements in the field, along with efforts to address grounding errors, promise a future where generative AI can be more widely utilized.
VentureBeat, the organizer of the AI Impact Tour, aims to serve as a platform for technical decision-makers to gain knowledge and transact in the transformative enterprise technology space.