How AI Foundation Models Are Changing the Financial Industry

The emergence of foundation models has expanded the possibilities for the integration of AI and finance. By reviewing current applications, discussing future trends, and analyzing potential impact, this chapter aims to address three key issues. First, the current application of foundation models in the financial industry. At present, the application of foundation models mainly focuses on the non-decision-making processes of simple business scenarios, such as intelligent customer service (AI agents) and operations assistant (copilots), etc. Deployment remains difficult in business scenarios that require deeper financial expertise, involve substantive financial advice, or entail core decision-making tasks. From the perspective of business processes, various financial institutions have been experimenting with applying foundation models to enhance front-office marketing and operations as well as middle and back-office operational support. In core analysis and decision-making processes, foundation models are more often used to assist front-end information collection and back-end content generation and output, thereby enhancing efficiency. From the perspective of business scenarios, foundation models have been deployed in payment, credit, investment research, investment advisory, insurance, and other areas. Second, the potential impact of foundation models on the future development of the financial industry. We believe foundation models will affect the financial industry in three aspects: (1) In terms of application trends, foundation models are likely to work collaboratively and complimentarily with smaller, specialized models and be deployed across more specific scenarios. At the same time, foundation models’ capabilities as copilots are expected to enhance and enable more business functions. In addition, AI agents are expected to upgrade user interaction experiences and reshape the business development models of financial institutions. (2) In terms of empowerment potential, we believe that foundation models may be relatively larger for wealth management and asset management, in the insurance and credit areas. Using a “demand-intermediary-supply” framework, we believe that foundation models may have greater potential to empower wealth management and asset management, and they also hold promise for generating value in insurance and credit. (3) In terms of industry landscape, foundation models are expected to improve the interaction experience for long-tail customers and create new service touchpoints and super applications. Meanwhile, we believe both a “Matthew effect” and a “multiplier effect” will emerge: Large institutions may be re-ranked, while some small and medium-sized institutions may have the opportunity to leapfrog. Third, the potential impacts of foundation models on the financial system in the future. In terms of efficiency, inclusion, and safety, foundation models are expected to reduce the information asymmetry within the financial system and improve overall efficiency. They are also expected to reduce user interaction barriers and enhance customer reach and pricing efficiency, thereby facilitating inclusive finance. Regarding safety, we believe the application of foundation models in finance is associated with multiple risks, including foundation models’ own technical risks, the potential for foundation models to exacerbate existing financial risks, and the emergence of new financial risks. Appropriate regulation and effective response will therefore be a top priority. We observe that regulatory authorities worldwide have already begun exploring and experimenting with adequate measures in certain aspects. Looking ahead, we recommend mitigating risks through measures such as strengthening multi-stakeholder collaboration to build shared infrastructure, establishing a tiered and categorized licensing regulation system, encouraging financial institutions to enhance internal risk management systems, and developing risk-control technologies.