Du Xiaoman CEO Zhu Guang: Applying reasoning large models in the financial field requires delving into core businesses | Frontline

Author｜Huang Nan

Editor｜Yuan Silai

The financial industry highly relies on technology. In this field, AI + tool applications have been used for a long time, providing more efficient and intelligent service methods for financial institutions. For example, in the past year or more, large models have been used in the ways of information processing and content production to improve the customer service experience, making robot conversations more smooth; and helping financial planners and staff of financial institutions to organize documents, meeting minutes, and provide financial assistant services.

However, although these services play a certain role in reducing costs and improving efficiency, they are still peripheral applications of financial business, not part of the core business, and have not been able to deeply change the core areas of the financial industry.

On October 28, the 2024 Hong Kong FinTech Week opened at the Hong Kong AsiaWorld-Expo. Zhu Guang, the CEO of Du Xiaoman, pointed out at the main forum of the Tech Week that "The application of the new wave of generative AI technology represented by the o1 reasoning large model in the financial field will go from peripheral scenarios to core businesses, directly affecting the quality of core decision-making in the financial industry, and bringing a huge breakthrough in the product and service experience for customers, while reshaping the fintech industry."

Zhu Guang, the CEO of Du Xiaoman, speaking at the main forum of the FinTech Week

For an industry, the huge changes brought about by generative AI technology cannot be separated from two major premises. Zhu Guang analyzed that one is that the core customer experience must undergo a huge change; the second is that it must have an impact on the core decision-making of the business.

For example, in the credit business, for employees, potential customers are not scarce because people with capital needs will actively seek services; but to find high-quality customers that are compatible with the services, it is necessary to conduct accurate customer profiling and use large models to identify high-quality customers on the network. From the user's perspective, when the large model can have an impact on their financial service experience and have a significant impact on core business decisions such as risk decisions and business decisions, the potential of the large model can be truly released.

Among them, since OpenAI released the GPT-o1 large model, the "thinking ability" of the model has been significantly improved, and more complex logical reasoning and problem decomposition can be realized in the financial scenario. Taking the risk control scenario as an example, through the reasoning large model, Du Xiaoman can analyze the customer's credit report and bank statement, infer the customer's repayment ability, and finally give a risk control decision suggestion on whether to approve the application, and the decision time can be shortened to be realized in the fastest 1 minute.

Zhu Guang said that the current large model has mastered risk control knowledge and has reasoning ability, "It can read credit reports and check bill flows like a professional auditor, and even interpret big data on the Internet, think and capture the correlations between data, and generate the basis and conclusion of risk judgment,"

In addition, in the field of quantitative investment, high-value factors can be mined and investment algorithms can be optimized through large models. In the insurance field, personalized product design and underwriting decisions can be made according to the customer's situation. It can be seen that the application and implementation of large models in the financial field is still continuously expanding. Embrace AI and large models first, reduce the threshold for using AI, in order to solve more complex problems in actual scenarios.

In terms of improving the customer experience, Du Xiaoman has developed a small intelligent cartoon robot that will always be displayed on the desktop of the customer service staff. When the customer service staff has emotional fluctuations such as excitement or anger during the service process, the robot can immediately detect these emotional changes through voice recognition technology. At this time, the expression of the cartoon robot will turn red and prompt the customer service staff to slow down the speaking speed and control the emotions. According to the actual measurement, this measure can significantly reduce the customer complaint rate and ensure the service quality of each call.