At the 8th China Fintech Forum, held in Beijing on September 9 as part of the China International Fair for Trade in Services, Xiamen International Bank's Chief Information Officer, Wang FengHui, delivered a keynote address. He shared insights into the bank's distinct identity as a mid-sized institution and explored the technological foundation, particularly its AI-driven initiatives, that underpin its operations.

Wang began by highlighting the bank's unique background, explaining that although it is a mid-sized bank with a group asset scale of under 1.2 trillion yuan, the institution has deep ancestral ties to overseas Chinese communities. With its headquarters in Xiamen, the group also includes Macau International Bank and Chiyu Bank in Hong Kong, collectively operating over 150 branches and employing nearly 7,500 staff. Founded in 1985 as China's first Sino-foreign joint venture bank, the institution was largely established with capital from a prominent Fujian-based overseas Chinese philanthropist, Li Wenguang. The bank later transitioned to a domestic city commercial bank in 2013, yet it retains its strong cross-border focus, connecting mainland China with Hong Kong and Macau.

The bank's early advantage in cross-border services was formalized in 2001 when it became the first joint-venture bank approved to conduct comprehensive foreign exchange business. In 2017, it acquired Chiyu Bank, and in 2022, it established a dedicated overseas Chinese finance department, publishing the sector's first standard for this niche. This year, it became the first local corporate bank in Fujian to directly join the CIPS system, with its subsidiaries also becoming direct participants. The bank's strategic positioning relies on a robust set of licenses, including FT accounts in Shanghai FTZ, EF accounts in Hengqin, and securities licenses in Hong Kong. Wang noted that this unique licensing structure gives it a natural advantage in the cross-border financial sector. This is highlighted by the bank's "Cross-Border e-Station" platform, which has reached a cumulative transaction volume exceeding 500 billion yuan, and an instant remittance service to Macau that bypasses SWIFT.

Moving to the technical backbone, Wang emphasized that the bank began building its cloud computing foundation back in 2016. The cross-border platform runs on a cloud-native architecture using domestic technology. He also detailed the bank's AI-driven risk control system, known as "Sky Eye," which integrates real-time intelligent decision-making for credit approval and fraud prevention. While the bank has applied machine learning to process trade documents, Wang acknowledged that adding large language models has only solved about 95% of the challenges; complex elements like handwritten notes, blurred images, and nested tables within documents remain persistent pain points. He also reflected that the bank has established a first-level department dedicated to AI, integrating it with the existing data management center, a move he believes is crucial for institutional transformation.

A central theme of Wang's presentation was the bank's philosophy of "people first, value first," with the guiding principle that ultimate purpose is human potential, while AI serves as a supportive tool. He addressed the anxiety employees have about being replaced by AI, arguing that the "distillation" of tacit knowledge from employees into explicit, accessible forms is essential. To support this, the bank has structured its strategy around a "5+4+3" framework, which includes five key projects and a focus on knowledge engineering, alongside organizational safeguards. Xiamen International Bank believes it may be among the first in the industry to establish a dedicated AI department, focusing on converting existing systems and staff expertise into structured knowledge.

The bank is actively deploying a wide range of AI applications. Among these, AI coding has shown the most immediate returns, generating over 5 million lines of code and accounting for approximately 70% of new code. Other practical applications include using AI for credit report automation, which now achieves a rate of over 90%, and AI-assisted requirements writing, which performs best in specific scenarios. The bank currently has over 220 AI production scenarios in use, handling an average of 4.25 million agent requests annually. To maintain focus, they track two key "North Star" metrics: daily token consumption, which stands at about 1.6 billion, and the digital employee substitution rate.

Wang highlighted several developmental stages in the bank's AI journey, with the release of DeepSeek's R1 model acting as a turning point that confirmed the arrival of the AI era. The bank is actively exploring innovative solutions across multiple fronts. First, it uses a workflow-based intelligent body platform, which Wang argued remains essential for efficiency and accuracy, particularly for applications like AI-powered mobile banking. Second, they are developing a general agent tailored for vertical domains, which functions as a digital employee for risk control, offering data analysis and strategy generation. Third, they are building an internal enterprise assistant, akin to a private version of a public chatbot, but trained on their proprietary knowledge. Fourth, they have developed a desktop office agent for employees on secure terminals that cannot access the internet. This local agent, built to prevent hallucinations and ensure behavior safety, is already being rolled out and is expected to significantly boost productivity with document processing. Finally, the bank is also piloting commercial agents from major tech firms for less sensitive tasks, such as sentiment analysis.

To propagate its AI culture, the bank has selected 90 enthusiastic employees from all branches to act as change agents, sparking wider adoption. Wang shared his strategic foresight, predicting that AI's prevalence is inevitable and that the cost of computing and tokens will eventually decrease. He noted that while the bank must carefully consider how deeply it can blend AI with its banking processes to achieve a seamless experience, the real challenge lies in system architecture. He argued that systems must be designed for agents, not just programmers, which is why the "distillation" of human knowledge is a non-negotiable step. He also raised ethical considerations regarding accountability in human-machine collaboration as agents play a larger role.

In his concluding remarks, Wang presented his final insights. He underscored the bank's decision to merge the AI and data management teams, emphasizing that data and knowledge are two sides of the same coin and are both essential for creating a powerful "flywheel effect" for the organization. He also delineated two separate safety concepts: content security and behavioral security, particularly for general agents operating in a sandbox. Finally, he stressed the significant need to bridge the gap between individual and organizational efficiency improvements. He reiterated the bank's primary principle that AI is never the goal; it is purely a means to serve and uplift people, a mindset essential for mitigating employee anxiety and ensuring a healthy, ethical transition into the future.