Du Xiaoman CEO Zhu Guang: AI Moves Financial Services from "Tailored to Thousands" to "Tailored to Each Individual," Jointly Releases XuanYuan 4.5 Financial-Enhanced Large Model

On November 2, the 2026 Hong Kong Fintech Week opened at the Hong Kong Convention and Exhibition Centre. Key officials of the Hong Kong SAR government and dozens of entrepreneurs from the AI and fintech fields at home and abroad attended the event. "AI-native finance and agents" became a major topic of this year's conference. On the opening day, Du Xiaoman CEO Zhu Guang and Lin Chen, Vice President of the University of Hong Kong and Director of the Center for Financial Innovation and Development, held a dialogue on how AI can drive financial service innovation.

Zhu Guang believes that the important value of AI in finance is to move financial services from "one face for a thousand people" to "a thousand faces for a thousand people," and further toward "a thousand faces for one person": not only understanding the differences among different customers, but also continuously understanding the real needs of the same customer at different times and in different scenarios, so that personalized services benefit a broader population.

From "one face for a thousand people" to "a thousand faces for one person," what changes are taking place in financial services? In Zhu Guang's view, traditional services meet common needs with standardized products; in the internet era, customer segmentation provides similar products and services to similar people, namely "a thousand faces for a thousand people"; while AI digital humans can continuously accompany customers, confirm their real intentions through interaction, understand how the same person's needs change at different times and in different circumstances, and then adjust prices, limits and terms in combination with risk control, enabling financial services to move from "a thousand faces for a thousand people" to "a thousand faces for one person." At the same time, as service coverage expands, continuous interaction also brings more customer insight data, forming a flywheel that drives continuous improvement in demand understanding and service capabilities.

"This year, the biggest change in large models for us is that they have gone from 'answering questions' to 'delivering results,'" Zhu Guang said. What supports this change is distilling expert methods into reusable skills, agents autonomously invoking tools and verifying results, and the ability to continuously run long-duration tasks.

In risk control, Du Xiaoman has built a self-iterating and self-evolving system: from locating and attributing risk anomalies, to variable mining, customer segmentation, model and strategy development, and then to online monitoring; after business results flow back, improvements continue, with humans mainly making judgments and confirmations at key nodes. The cycle from risk anomaly attribution plus strategy development has been shortened from about two weeks to three to four days.

Zhu Guang believes that a deeper change is expanding the exploratory capability of risk control. In the past, experts, limited by experience and energy, could only verify some hypotheses; today, AI can continuously test hundreds or thousands of variable combinations and segmentation schemes, seeking better solutions that are difficult for humans to exhaust.

Supporting this exploration are underlying models with stronger financial expertise. On the opening day, Du Xiaoman also jointly released the XuanYuan 4.5 financial-enhanced large model with the Hong Kong University of Science and Technology and the Hong Kong Generative AI Research and Development Center (HKGAI). This is an industry-leading financial-enhanced large model based on RSI (self-evolution) technology. In core financial scenarios such as customer acquisition, risk control and customer management, XuanYuan 4.5 improves results by 30% compared with general models, at only 10% of the latter's cost. Du Xiaoman has also launched AI-native credit services based on XuanYuan 4.5 in the Mexican market.

Looking to the future, Zhu Guang believes that personal assistants may become a new entry point for financial services. Customers hand their needs to an assistant, which coordinates goods, services and financial capabilities across different platforms, while people make judgments and grant authorization. Financial services are expected to move from "people looking for products" to "components that satisfy people's intentions," invoked within ongoing life goals. For personal assistants to truly act on people's behalf, payments need to become an execution capability they can invoke: from customers "confirming each payment" one by one, to giving the assistant a goal and budget and "authorizing a task." The assistant executes within the authorized scope, and payment networks and financial institutions provide verification, settlement and follow-up services.

Lin Chen believes that as collaboration and resource invocation among agents continue to increase, payment networks for agents will gradually develop. Such a network needs to establish three layers of foundation: identity, authorization and settlement—clarifying the identity, capabilities and supported payment methods of agents, defining the permissions and execution boundaries granted by humans, and enabling payment settlement among agents. At present, relevant solutions are still in their infancy, with considerable room for innovation and entrepreneurship in technical improvement and scenario implementation; companies such as Du Xiaoman can participate by drawing on their accumulated financial scenario experience.

He also emphasized that security is an important direction for the integration of AI and Web3. As agent applications expand, code vulnerabilities and infrastructure risks need more attention. Smart contract auditing, infrastructure development and AI security are not only necessary conditions for supporting application implementation and safeguarding customer asset security, but will also bring new development opportunities.