China Merchants Bank Chief Information Officer Zhou Tianhong: The iteration cycle for large model applications has been shortened to 8 days

(Source: Beijing Business Today)

Beijing Business Today reported (by Meng Fanxia, Zhou Yili) On March 30, China Merchants Bank held its 2025 annual performance briefing. The bank’s Chief Information Officer, Zhou Tianhong, provided a detailed introduction of the progress in its digital transformation and AI strategy implementation. He said that China Merchants Bank has adhered to its “technology to empower the bank” strategy for many years. The “15th Five-Year Plan” digital finance development plan clearly sets the goal of building an intelligent bank as the core task for the next five years, and it has also comprehensively implemented the AI first strategy.

Zhou Tianhong pointed out that since the Chairman, Qiu Jianmin, made the forward-looking proposal to build the industry’s first intelligent bank in 2023, CMB has actively laid out model-based application development. In 2025, it achieved notable progress based on the maturity of foundational large models. To implement the AI first strategy, China Merchants Bank made detailed arrangements for advancing large model applications. It comprehensively sorted out 1,588 independent work items across the bank, classified them into three categories—high value, medium value, and low value—according to quantifiable standards based on how effectively a large model can contribute, and prioritized the rollout of high-value scenarios. By the end of 2025, it had cumulatively deployed 856 large model application scenarios.

Zhou Tianhong said that China Merchants Bank plans to fully roll out high-value work items in 2026 and accelerate the deployment of low-value scenarios to drive end-to-end improvements in important business processes.

Regarding the characteristics of large model applications, Zhou Tianhong emphasized that they differ significantly from traditional software development. In essence, large model applications are probabilistic applications, so they inherently have a relatively high degree of uncertainty. They therefore require continuous debugging and ongoing iteration. Based on China Merchants Bank’s experience, a large model application typically needs to complete six iterations before it can be put into production. By optimizing its large model engineering system, since 2025, China Merchants Bank has shortened the application iteration cycle from the 2024 average of 32 days to 8 days in 2025. Efficiency has improved to one quarter of the previous level, significantly accelerating deployment.

Zhou Tianhong said that data show that in 2025, the bank’s average daily input/output Token volume increased 10.1 times compared with 2024. The latest total average daily input/output Token volume reached 26 billion, ranking among the leading level in the banking industry. “Overall, in 2025, large model applications at our bank have already fully begun to play their role, and we have made very good progress in both improving quality and efficiency,” Zhou Tianhong said. He noted that, as of cumulative totals, large model applications had replaced more than 15.56 million hours of human labor, equivalent to saving the workload of more than 8,000 full-time employees. He also cautioned that large models still have a “hallucination” risk. China Merchants Bank is taking multiple measures to build secure, reliable, and trustworthy large model applications. In 2026, it will continue to increase investment and fully implement the goal of building an intelligent bank.

A massive amount of information and precise analysis—available in the Sina Finance App