Industry Observation | Over 500 application scenarios implemented, ICBC explores a new AI paradigm

The financial innovation in the era of artificial intelligence goes beyond the support and application of technology in finance; more importantly, it is about redefining the financial paradigm through technological innovation. ICBC’s technology investment in 2025 reached 28.588 billion yuan, exceeding 20 billion yuan for the fifth consecutive year, continuing to lead the industry in investment intensity.

Text by | Deng Jia

Edited by | Li Chenxi

Over 500 AI (artificial intelligence) application scenarios have been implemented, with trillions of Token (word element) financial datasets and nearly 300 intelligent risk control decision-making scenarios…

On March 27, ICBC (hereinafter referred to as “ICBC”) released its 2025 annual report, presenting impressive results in AI application.

Since the beginning of the year, the “lobster craze” triggered by OpenClaw has pushed the wave of artificial intelligence to new heights. People’s expectations for AI have shifted from “answering questions” to “completing tasks,” and the demand for technological implementation has never been more urgent.

This raises a question: how can various industries achieve digital transformation in the era of artificial intelligence?

ICBC President Liu Jun pointed out in his signed article that true digital transformation requires a complete reconstruction of underlying logic and methodology, achieving a fundamental innovation from “improvement” to “reconstruction.”

In this context, Liu Jun believes that financial innovation in the era of artificial intelligence is not just about finance’s support and application of technology; more importantly, it is about redefining the financial paradigm through technological innovation. Therefore, the future direction of finance is not just “financial innovation,” but rather “innovative finance,” along with more disruptive new methodologies and new system integrations.

In practice, as the leading “head goose” of the national financial team, ICBC is deeply advancing the “Leading AI+ Action Plan,” exploring a new paradigm of AI empowering finance.

In the past year, ICBC seized the significant breakthrough opportunity of generative AI, building a set of AI-native financial service ecosystems centered around intelligent agents, promoting the large-scale application of AI.

The annual report shows that ICBC has successfully implemented over 500 AI application scenarios in more than 30 business areas.

An industry insider commented that ICBC not only broke through the technical bottleneck of AI transitioning from “concept” to “actual productivity,” but also explored a standardized AI-native transformation path for the banking industry, becoming a leader in the digital transformation of the industry.

Behind the forefront technological exploration is robust performance support.

By the end of 2025, ICBC’s total assets reached 53.48 trillion yuan, becoming the first bank in the world to surpass 50 trillion yuan. In terms of performance, ICBC achieved an operating income of 801.395 billion yuan in 2025, an increase of 1.9% over the previous year; net profit was 370.766 billion yuan, an increase of 1.0% over the previous year.

Thus, ICBC’s technology investment in 2025 reached 28.588 billion yuan, exceeding 20 billion yuan for the fifth consecutive year, continuing to lead the industry in investment intensity.

At the 2025 annual performance release conference, ICBC’s management introduced that the bank upgraded its four-year “Digital ICBC (D-ICBC)” strategy to “Intelligent ICBC (AI-ICBC),” and will promote it as a key focus in 2026.

“This upgrade signifies that ICBC is keeping pace with the times and advancing with the country, fully embracing artificial intelligence and striving to enhance the level of intelligent development,” said Zhao Guide, Vice President of ICBC. “For ICBC, intelligent development is not an option; it is a compulsory question, a strategic choice for us to seize opportunities and take the initiative.”

Intelligent Upgrade: From D-ICBC to AI-ICBC

In January 2026, ICBC launched the mobile banking “Smart Enjoy +1.0” version, in which the upgraded “Gong Xiaozhi” has the capability for intelligent dialogue interaction with customers, becoming one of the few AI financial applications that interface with customers in the industry.

According to ICBC, the “Smart Enjoy +1.0” version of “Gong Xiaozhi” can understand customer needs in natural language, providing intelligent services covering scenarios such as financial consulting, business handling, and problem-solving.

Since the advent of ChatGPT, generative AI applications capable of conversing with users have emerged one after another, but due to the inherent “hallucination” flaws of large models, their application in finance has been strictly limited.

To address this, ICBC has independently built a full-stack AI security detection system that is “quantifiable in risk, simulative in attacks, and verifiable in defenses,” supporting intelligent customer service for mobile banking and client assistants at branches, becoming the first in the industry to pass the备案 of the Cyberspace Administration.

“Gong Xiaozhi” interfacing with customers is a breakthrough for industry-level financial large models and a milestone in ICBC’s exploration of intelligent upgrades in financial services.

Going back to early 2025, ICBC initiated the “Leading AI+ Action Plan,” announcing the construction of a new generation of enterprise-level business empowerment model centered around the “Gongyin Zhiyong” large model, comprehensively promoting the intelligent upgrade of financial services and exploring the Gongyin paradigm for large model applications in the financial industry.

The person in charge of ICBC’s “Gongyin Zhiyong” disclosed in a signed article that the bank has explored and formed a “1+X” large model application paradigm and supporting engineering solution.

“1” refers to the super-intelligent agent, which serves as the intelligent hub, adopting a modular expandable multi-agent collaborative framework (MoA intelligent agent framework) to decompose, plan, and execute complex financial tasks, achieving a success rate of over 90%. “X” refers to specialized intelligent agents, assembled for use by the intelligent hub, including knowledge retrieval, data analysis, document writing, intelligent search, system API calls, etc., with knowledge question answering satisfaction exceeding 90% and conversational query accuracy over 95%.

Industry insiders believe that large models are the “brains” of AI, while intelligent agents are the “hands and feet” of AI, executing the commands of large models as “digital employees.”

According to the aforementioned responsible person, under the “1+X” framework, ICBC can flexibly assemble functions like building blocks based on scenario needs, rapidly constructing AI applications. As a result, the bank is able to scale the construction of intelligent agents, achieving a transformation from single-scenario empowerment to comprehensive business reshaping.

ICBC’s management disclosed that the bank has already implemented over 500 AI application scenarios in more than 30 business areas. For instance, in investment trading, it has vigorously promoted the intelligent inquiry assistant for financial markets, achieving a trading intelligence ratio of 96% and a 50% year-on-year increase in the number of transactions.

The large-scale application of intelligent agents relies on a powerful technological foundation. Since 2023, ICBC has successfully constructed a fully stack self-controllable “Gongyin Zhiyong” large model technical system, characterized by efficient computing power, model adaptability, rich data, and secure reliability.

Among them, the elastic computing power pool of large models, primarily based on domestic computing power, has achieved minute-level switching between training mode and inference mode; the trillion-level Token financial dataset provides ample “nutrients” for large model training, making ICBC the only bank selected for the National Data Bureau’s 2025 pilot list for innovative development of trusted data spaces.

Looking ahead, the 2026 Party Building and Management Work Conference of ICBC proposed, “Upgrade to create Intelligent ICBC, enhancing intelligent development momentum,” thus upgrading the four-year “Digital ICBC” strategy to “Intelligent ICBC.”

Zhao Guide stated at the performance release conference that “Intelligent ICBC” is an important component of ICBC’s 14th Five-Year Plan. Therefore, ICBC has defined the main construction ideas of “one new and three high,” namely: new quality productivity driven by intelligence, high-quality development of ICBC, high-level safety of the group as a whole, and high-efficiency governance integrating industry, technology, and data.

Risk Control Leap: From Risk Prevention to Efficiency Enhancement

In the construction of “Intelligent ICBC,” risk control is undoubtedly one of the business scenarios that has benefited the most.

In recent years, the pressure of credit risk control in the two major sectors of inclusive banking and personal loans has generally increased, but ICBC has achieved five consecutive decreases in its non-performing loan ratio.

By the end of 2025, ICBC’s non-performing loan ratio was only 1.31%, recording the lowest value since 2015.

ICBC Vice President Wang Jingwu stated that in recent years, ICBC has solidified the foundation for risk control in inclusive loans through various means, including continuously enhancing the foresight of risk control and the precision of risk monitoring and early warning through intelligent and centralized methods.

In 2025, ICBC accelerated the construction and promotion of an enterprise-level intelligent risk control platform, which currently covers nearly 300 risk control decision-making scenarios, transitioning risk control work from human prevention and control to technical prevention and intelligent control.

According to the aforementioned responsible person’s signed article, in the field of risk management, the bank has extracted the credit risk evaluation thought chain data aimed at customer managers, review and approval personnel, risk officers, and risk managers. By strengthening the training of the risk control capabilities of the underlying large model, it has constructed a dedicated large model for enterprise credit risk intelligent assessment. This model achieves comprehensive analysis of risk assessment across multiple professional dimensions, significantly improving the accuracy and recall rate of risk customer identification, while automatically generating readable, logically complete risk judgment descriptions to assist in risk management decisions.

In the field of credit management, ICBC has constructed a multi-agent collaborative matrix covering the entire credit process, providing over 20 intelligent agent services for four types of business scenarios: institutional answers, business marketing, customer analysis, and risk investigation monitoring, comprehensively enhancing the intelligent service level of the credit system. Among them, the intelligent loan review agent can intelligently analyze customer financial data and identify financial embellishments, improving financial risk analysis efficiency by 20%.

At the 2025 annual performance release conference, ICBC’s management disclosed that the credit intelligent assistant developed by the bank has provided intelligent support for the entire business element information for over 20,000 credit personnel across the bank.

The annual report indicates that the current “4E Center” (the risk view, measurement, early warning, and decision-making of the enterprise-level intelligent risk control platform) comprehensively covers the head office and branches, driving the overall non-performing loan ratio of the bank down by 3 basis points compared to the end of the previous year.

It is worth noting that the application of new technologies such as AI not only enhances the bank’s risk control capabilities and efficiency but also provides possibilities for breaking through the bottleneck of inclusive finance and lowering financing thresholds.

Taking the poultry farming industry as an example, under the traditional credit model relying on manual assessments, the poultry farming sector has long faced difficulties in assessment and financing. In 2025, a local branch of ICBC introduced AI image recognition technology, launching “White Duck e-loan 2.0,” which automatically identifies the number and health status of stock using an AI system, and dynamically adjusts the credit limit based on real-time monitoring data, enabling local white duck farmers to complete the entire loan process online.

The annual report shows that in 2025, ICBC explored the application of satellite remote sensing in agriculture, integrating technologies such as the Internet of Things and biometric recognition to monitor collateral and manage post-loan, further enhancing digital credit capabilities. By the end of 2025, ICBC’s agricultural loan balance reached 50.722 trillion yuan, an increase of 676.1 billion yuan over the previous year.

In addition, in the field of financial technology, ICBC also provides credit support for technology enterprises lacking collateral through intelligent financial products such as “Innovation Points Loan” and “Tech Innovation Cloud Intelligent Loan.” By the end of 2025, the balance of ICBC’s technology loans reached 6 trillion yuan, with a growth rate of 19.9%.

“We will certainly comprehensively and effectively validate the series of new technologies currently in existence on ICBC’s broad platform. Then, we will focus on applications, on the needs of everyone, and on the voices of investors, diligently refining the technology, ultimately serving the real economy and the people’s aspirations for a better life,” Liu Jun stated at the performance release conference. (Data related to this article is sourced from ICBC’s 2025 annual report.)

Risk Warning: Investment has risks, and investors should be cautious. Credit services must comply with regulatory requirements and bank risk control standards.