The 2026 Inclusion·Outer Temple Conference highlighted a pivotal theme on September 9th: the practical application of AI in the financial sector. During the "AI-Native Era: Exploration of Credit Economy Applications" forum, Zhang Jianhua, director of the Financial Development and Regulatory Technology Research Center at Tsinghua University's PBC School of Finance, asserted that financial institutions failing to adopt AI will inevitably be left behind.
The forum, co-hosted by Qiantang Credit and Ant Consumer Finance, centered on real-world uses such as personal credit management, consumer scenario risk control, and intelligent customer service. A clear consensus emerged among attendees: while AI has entered financial production workflows, it has not yet reached the point of replacing human decision-making. The collective view is that advancing AI from basic task execution to sound judgment and, eventually, to constrained autonomy requires establishing trustworthy checkpoints at every stage to prevent system failures. Zhang also noted that technological progress and business innovation often outpace regulatory frameworks, meaning issues like model hallucinations, algorithmic black boxes, and the potential for AI to amplify existing risks must all be incorporated into governance strategies.
Where to start
In practice, AI is already making inroads into finance through credit management and scenario-based risk control. Li Zhen, general manager of Qiantang Credit, shared that the company is transitioning from a financial data service provider into an operator of social credit infrastructure, focusing on authorized public data operations and the development of trusted data spaces.
In the consumer finance space, Lin Jianan, vice president and chief risk officer at Ant Consumer Finance, detailed how AI is being used to identify genuine customer needs, optimize differentiated credit offerings, and improve risk assessment. The company's scenario-based risk controls now cover over 230 key areas, including daily spending, major purchases, travel, hobbies, and further education. In 2026, scenario-based targeted credit limit increases grew 30% year-on-year, with default rates for these users running 20% lower than the overall average. The Huabei "Little Red Flower" feature now supports limit increases through the submission of over 1,000 document types, with future plans to introduce conversational interfaces where AI automatically extracts, verifies, and makes real-time decisions.
Beyond incremental efficiency gains, a more structured approach is needed. Ning Jiangbin, deputy general manager of Alibaba Cloud Intelligence's New Financial Industry Solutions division, proposed a phased roadmap for credit intelligence as AI agents become more involved in lending. This includes L1 for perception and execution, L2 for verification and reasoning, and L3 for autonomous intelligence within defined constraints, all aimed at driving AI toward decision-making that is verifiable, auditable, and reversible.
Determining the true production value of AI
Zheng Bo, chief AI scientist at SPD Bank's headquarters, offered a framework for assessing whether AI applications have genuinely entered production. He suggests the key indicators are: whether AI is embedded in core business processes, whether it performs critical tasks previously handled by humans, and whether it improves customer experience, operational efficiency, and risk management. Looking ahead, Zheng predicts that the next three years will be a watershed for AI in banking, defined by production capacity, value operations, and organizational adaptability. The leading banks, he argues, may not be those with the most AI agents, but those with faster organizational learning, where the expertise of a top relationship manager can be rapidly replicated across the entire workforce.
Setting boundaries is as important as driving efficiency. Zhu Le, CTO of Qiantang Credit, emphasized that as AI becomes more sophisticated, it will gain growing "exploration rights," "advisory rights," and bounded "execution authority." However, he stressed that final decision-making power cannot currently be delegated to AI. Zheng Bo echoed this sentiment, stating that in high-risk banking operations like credit approval and capital trading, AI can provide analysis and recommendations, but critical decisions must remain with human oversight.
Comments