From "Digital" to "Intelligent Data" AI+ Strategy Reshapes the Financial Paradigm

Ask AI · How digital-intelligence strategy is reshaping the competitive landscape of the banking industry?

Economic and Industrial Daily reporter Zhang Manyou, Beijing

Industrial and Commercial Bank of China (601398.SH) will continue to upgrade its four-year “Digital ICBC” (D-ICBC) strategy into “Digital-Intelligence ICBC” (AI-ICBC). China Construction Bank (601939.SH) has already built nearly 400 AI scenario applications. Agricultural Bank of China (601288.SH) has built an enterprise-level AI digital staff member “Yiming”… 2025 is a pivotal year for the banking industry to fully embrace generative artificial intelligence (AIGC), represented by large language models (LLMs). A new picture of the banking industry—remade by AI—is gradually taking shape.

In this technology-driven transformation, the integration of finance and technology no longer remains at the surface-level digitization overhaul, but begins to reach the core of business logic and operating models.

Industry insiders believe this is not only a technological iteration, but also a strategic choice for banks to seize the initiative amid the tide of the times. Whoever can better balance innovation and risk, and can more efficiently turn cold algorithms into service with warmth, will gain the upper hand in future financial competition.

Strategic Up-leveling: From Efficiency Tools to the “Must-Answer Questions” for Banks

With the rapid development of large models and intelligent agents, in 2025 the banking industry is undergoing a profound transformation—from “efficiency tools” to “underlying logic.”

ICBC vice president Zhao Guide said, “For ICBC, digital-intelligence is not a multiple-choice question—it’s a must-answer question. It is our strategic choice to seize the initiative and take control.” He further explained that this upgrade is mainly based on three considerations: first, keeping pace with the trends of the times and actively seizing the overall direction of digitalization, networking, and intelligence; second, implementing national strategies and advancing the “Artificial Intelligence+” initiative in a tailored, localized manner; third, deepening reform and transformation, injecting strong impetus into building a world-class modern financial institution with Chinese characteristics across the whole bank.

Zhao Guide said that in 2025, ICBC launched and implemented the “Voyager AI+” initiative. It insists on combining top-level design with grassroots innovation, promoting the mutual reinforcement of technological innovation and application enablement, and actively building new quality productive forces in finance. On the technology side, ICBC insists on breaking new ground upward, building an industry-leading “ICBC Zhiyong” technology system that is full-stack, independently developed, and fully controllable. In terms of computing power, this system mainly builds an elastic computing pool for large models based on domestic computing power, achieving minute-level switching between training and inference modes. In terms of models, the bank has integrated more than a dozen industry mainstream models, carried out extensive second-stage training, and built an enterprise-level foundational model matrix that is more knowledgeable about finance and more knowledgeable about ICBC. It also constructed an agile and easy-to-use intelligent agent creation platform. On the data front, the bank built an enterprise-level artificial intelligence knowledge system, creating a trillion-level Token financial dataset with high quality, large scale, and broad coverage. On the security front, it improved governance effectiveness and built a full-chain security and risk-control system for AI applications, effectively covering areas such as security of technology infrastructure, data security, model security, and application security.

ICBC’s strategic upgrade is not an isolated case—an AI racing contest is underway among major banks.

Agricultural Bank of China president Wang Zhiheng said the bank is accelerating the construction of intelligent survey and review report templates, with an automated generation proportion of report data exceeding 70%, covering ten major business types such as inclusive small and micro credit underwriting and group lending. This greatly reduces the workload of manual drafting by grassroots credit staff. At the same time, the bank is building an enterprise-level AI digital staff member “Yiming” to empower relationship managers to serve customers even better.

Lei Ming, vice president of China Construction Bank, introduced that, facing major opportunities from the development of artificial intelligence, China Construction Bank is deeply advancing the “Artificial Intelligence+” initiative, approaching it from three aspects and promoting deep application of AI technologies across various fields in a systematic, and scalable way. It is strengthening basic capabilities: in terms of computing power, it insists on being moderately ahead of schedule and is推进ing the construction of a high-availability, high-elasticity intelligent computing cluster with “four locations and five centers.” In terms of algorithms, it introduced generative large models such as DeepSeek, Qianwen, and Zhipu, forming a model system that coordinates large models and small models, and blends generative AI with decision-oriented AI. In terms of data, it has built an enterprise-level knowledge base with intelligent search capabilities. By the end of 2025, China Construction Bank had built nearly 400 scenario applications.

In practice, AI empowerment provides strong support for frontline business development. At the application level, ICBC insists on rooting itself downward, following value orientation, and promoting successful implementation of artificial intelligence in more than 500 scenarios. For example, in investment and trading, it strongly promoted a financial markets intelligent quotation assistant, with the degree of trading intelligence reaching 96%, and the number of trades increasing 50% year-on-year. In customer acquisition and marketing, it built an intelligent marketing assistant for personal customer managers, forming a new service mode of human-machine collaboration that drives an increase of over one trillion yuan in transaction value for key products. In risk prevention and control, the credit intelligent assistant provides intelligent support for business-element information throughout the entire process for more than 20k credit staff across the bank. In operational efficiency improvement, it upgraded intelligent customer service and operational assistants; the share of intelligentization in key operational business types under centralized operations exceeds 60%, improving service efficiency and also reducing operational risks.

Lei Ming pointed out that in the realm of operational management, the coverage rate of AI assistants in response to branch outlets’ issues has reached 99.42%, with daily average visits exceeding 100k. In risk management, it has built an end-to-end “AI+ risk control” management model. Relying on generative large models, the bank achieved double-digit growth in the volume of approved business acceptances in 2025, and average processing time decreased by more than 30%. In addition, it continuously strengthened security and compliance protection, building a multi-dimensional control system covering business, data, models, and network security.

When discussing推进ing product and model innovation in “agriculture, rural areas, and farmers” finance, Wang Zhiheng said the bank is promoting the application of smart banking tools in the “agriculture, rural areas, and farmers” service domain, rolling out a “on-site + remote” survey model. It is constructing an agricultural-related data system covering satellite data, drones, and ground IoT, enhancing technology support for agricultural-related businesses and data supply capabilities.

Si Moxun, an IDC China Financial Industry Research Manager, summarized for reporters of The China Business Journal: in 2025, AI applications are no longer limited to the “efficiency tool” category. Instead, by using “intelligent agents” with autonomous decision-making and execution capabilities, they are推动ing credit, risk control, marketing, and operations to shift toward an RaaS business model. The core of this model is to directly link the income of the service provider to the actual business outcomes created for customers, rather than relying on the traditional software licensing or fixed service fee model. Meanwhile, with deeper deployment of large models and intelligent agents, the industry is shifting from “process-driven” to “data- and intelligence-driven.” In the risk control domain, it is changing from “rule-based defense” to “intelligent anticipation.” In mobile banking, it is shifting from “people looking for services” to “services finding people,” reshaping interaction logic.

Anchoring the “15th Five-Year and 6th Five-Year Plan” period: Seeking the best solution between security and innovation

In 2026, the banking industry’s deployment of artificial intelligence has already risen from tactical exploration to strategic planning.

Zhao Guide said that ICBC benchmarks against China’s “15th Five-Year Plan and beyond” (the “15th Five-Year and 6th Five-Year Plan” outline) and has preliminarily formulated the Group’s “15th Five-Year and 6th Five-Year Plan” plan, clarifying the main approach for building “Digital-Intelligence ICBC.” Put simply, it is “one new and three highs.” “One new” refers to “new quality productive forces driven by digital intelligence,” which is the driving force for building “Digital-Intelligence ICBC.” The “three highs” refer to the bank’s “high-quality development” as the goal, “high-level integrated safety across the Group” as the bottom line, and “high-efficiency governance through fusion of industry, technology, and digital intelligence” as the guarantee.

It is understood that to ensure the plan is implemented solidly, ICBC has clarified the bank-wide target tasks for the full year of 2026, focusing on four areas: “intelligence,” “smartness,” “smart computing,” and “smart experience.” It will continue implementing the “Voyager AI+” initiative. First, it will accelerate intelligent transformation, continuously optimize “ICBC Zhiyong,” build an enterprise-level data space, accelerate the improvement of the knowledge base, innovate in the creation of financial intelligent agents, and explore a new paradigm for AI enabling finance. Second, it will consolidate the smart foundation: iterate and upgrade the ECOS 2.0 smart banking ecosystem, accelerate evolution toward an intelligent native architecture system, and shift technology positioning from “backstage support” to “frontstage driving.” Third, it will expand the scale of smart computing: moderately and proactively do optimization for the computing power system, providing advanced, efficient, and secure computing power support for the development of digital intelligence. Fourth, it will build a smart experience platform: strengthen construction of key platforms such as mobile banking, accelerate the formation of a “one customer, one advisor” service model, and provide customers with high-quality services with one-point access and bank-wide response.

Bank of China (601978.SH) vice president Cai Zhao also expressed a strong determination to actively embrace the digital-intelligence wave. He said Bank of China will comprehensively implement the “Artificial Intelligence+” initiative, driving the bank-wide digital-intelligence transformation.

It can be expected that in the opening year of the “15th Five-Year and 6th Five-Year Plan” period, major banks will engage in even more intense competition around the deep integration of AI technology and business scenarios.

However, while the banking industry actively embraces AI, how to balance innovation and risk in a strong regulatory environment for finance has become an issue that cannot be avoided.

Chen Maochuan, an expert from China Insights Qianfan (financial industry consulting), believes that the core of balancing innovation and risk is to推进 AI applications in stages, with boundaries, under the premise of safety and controllability. He said first, it is necessary to view AI rationally, recognizing that finance is the essence of service. The core value of AI lies in improving the efficiency, precision, and coverage of financial services; it is reshaping service models, not the essence of finance. Second, regarding strategies for balancing innovation and risk, it is possible to build a scientifically systematic system across AI applications, governance, and defense. This includes establishing a management system covering the full lifecycle of AI, clarifying the boundaries of human-machine collaboration—where humans are responsible for final decision-making and for exception intervention. It also requires strict implementation of data classification and permission isolation, prioritizing an industry large-model plus local deployment approach. AI applications should be classified and tiered by risk level and value contribution, and innovation should be advanced in a differentiated manner by scenario. Increase investment in AI security and establish a risk emergency response and handling mechanism.

Si Moxun further emphasized the core principles of governance first and technological control. He believes that commercial banks, as financial institutions that manage operating risks, when推进ing the application of large models and intelligent agents, should follow the core principles of governance first and technological controllability. In governance first, it is necessary to establish an AI governance architecture, formulate AI full-lifecycle management regulations covering initiation, research and development, testing, launch, operation and maintenance, and withdrawal, and strictly follow the central bank’s “active and prudent, safe and orderly” eight-character guideline. It is also essential to落实 the three bottom lines: model interpretability, data not leaving the domain, and responsibility being accountable. In technological control, through technical means, AI risks should be controlled at the source. A real-time monitoring system should be established to watch for risks such as model accuracy degradation, concept drift, data privacy leakage, and adversarial attacks. At the same time, “red teams” should be regularly introduced to simulate malicious attacks or extreme market environments, testing the robustness of AI systems under stress scenarios.