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

Financial innovation in the AI era goes beyond supporting and applying finance to technology; more importantly, it redefines the financial paradigm through technological innovation. ICBC’s technology spending in 2025 totals as much as 28.588 billion yuan, the fifth consecutive year exceeding 20 billion yuan, with continued investment intensity that stays ahead of its peers

By | Deng Jia

Editor | Li Chenxi

Roll out more than 500 AI (artificial intelligence) application scenarios, a trillion-scale Token (vocabulary token) financial data set, and nearly 300 intelligent risk-control decision scenarios……

On March 27, Industrial and Commercial Bank of China (referred to as “ICBC” below) released its 2025 annual report, delivering an impressive AI application scorecard.

Since the beginning of the year, the “Lobster Fever” sparked by OpenClaw has pushed the AI wave to new heights. People’s expectations for AI have shifted from “answering questions” to “getting work done,” and the demand for technological deployment has never been more urgent.

This also raises a question: In the AI era, how can digital transformation be realized across all industries?

In his signed article, ICBC President Liu Jun pointed out that real digital transformation requires a complete reconfiguration of the underlying logic and methodology—achieving fundamental innovation from “improvement” to “remaking.”

Against this backdrop, Liu Jun believes that financial innovation in the AI era is not limited to finance supporting and applying technology. More importantly, it is about redefining the financial paradigm with technological innovation. Therefore, the future direction of finance is not only “financial innovation,” but “innovative finance,” along with more disruptive new methodologies and new system integration that come with it.

In practice, as the “vanguard goose” among the state-owned financial institutions, ICBC explores a new paradigm for AI empowering finance through deep advancement of the “Navigating AI+ Action Plan.”

Over the past year, ICBC seized major breakthroughs in generative AI, building an AI-native financial services ecosystem centered on intelligent agents, and promoting large-scale deployment of AI.

The annual report shows that ICBC has successfully rolled out more than 500 AI application scenarios across more than 30 business areas.

An industry insider said that ICBC not only overcame technical bottlenecks that prevent AI from moving from “technical concepts” to “real productive forces,” but also charted out a standardized AI-native transformation path for the banking industry, becoming a leading force in the industry’s digital and intelligent transformation.

Behind these frontier technology explorations lies strong performance support.

As of end-2025, ICBC Group’s total assets reached 5.3488 trillion yuan, becoming the first bank globally to break through the 5 trillion yuan threshold. In terms of performance, ICBC achieved operating income of 801.395 billion yuan in 2025, up 1.9% year over year; net profit was 370.766 billion yuan, up 1.0% year over year.

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

At the site of ICBC’s 2025 annual performance briefing, the bank’s management introduced that it upgraded the four-year “Digital ICBC (D-ICBC)” strategy to “Intelligent and Digital ICBC (AI-ICBC),” and will push it forward as a key focus for 2026.

“This upgrade marks that ICBC keeps pace with the times and advances together with the country, fully embracing artificial intelligence and striving to improve its intelligent and digital development level.” ICBC Vice President Zhao Guidu said, “For ICBC, intelligent and digital development is not a multiple-choice question; it is a required answer question. It is a strategic choice for us to seize the first-mover advantage and take the initiative.”

Intelligent and Digital Upgrade: From D-ICBC to AI-ICBC

In January 2026, ICBC launched the “Zhi Xiang +1.0” version of its mobile banking app, in which the upgraded “ICBC Xiao Zhi” has the ability to interact with customers via intelligent dialogue, becoming one of the few AI financial applications in the industry that is customer-facing.

According to ICBC, in the “Zhi Xiang +1.0” version, “ICBC Xiao Zhi” can understand customers’ needs using natural language, providing smart services covering scenarios such as wealth management consultation, business processing, and answering questions.

Since the debut of ChatGPT, generative AI applications that can have conversations with users have emerged one after another. However, due to the inherent “hallucination” flaws of large models, their use in the financial sector has been strictly limited.

To address this, ICBC independently built a full-stack AI security detection system featuring “risk quantification, attack simulation, and defense verification,” helping in two customer-facing scenarios for mobile banking intelligent customer service and branch customer assistants, and being the first among peers to pass a cybersecurity and informatization filing with the Cyberspace Administration.

“ICBC Xiao Zhi” as a customer-facing application is an industry-level breakthrough of financial large models, and also a milestone in ICBC’s exploration of upgrading the intelligence of financial services.

Taking the timeline back to early 2025, ICBC launched the “Navigating AI+ Action Plan,” announcing that it would build a new generation of enterprise-level business enablement model with the “ICBC Zhi Yong” large model at its core, fully promoting the intelligent upgrade of financial services, and exploring an ICBC paradigm for large model applications in the financial industry.

A person in charge related to “ICBC Zhi Yong,” in his signed article, disclosed that the bank explores and forms an “1+X” financial large model application paradigm and its supporting engineering solutions.

“The ‘1’ refers to a super intelligent agent. As the intelligent hub, it uses a modular and extensible multi-agent collaboration framework (MoA agent framework) to break down, plan, and execute complex financial tasks, with a success rate of over 90%. The ‘X’ is a specialized intelligent agent used for the intelligent hub to assemble and call, including knowledge retrieval, data analysis, document writing, intelligent search, and system API calls. In addition, knowledge Q&A satisfaction is 90%+, and the accuracy of dialogue-based queries is over 95%.”

According to industry observers, large models are the “brain” of AI, while intelligent agents are the “hands and feet” of AI—digital employees that execute instructions from the large models.

As the aforementioned person in charge said, under the “1+X” framework, ICBC can flexibly assemble functions like building blocks based on needs of different scenarios, enabling the rapid construction of AI applications. As a result, the bank can scale up the construction of intelligent agents and realize transformation of business in all directions, from empowering single scenarios to reshaping the entire business.

According to ICBC management, the bank has already rolled out more than 500 AI application scenarios across more than 30 business areas. For example, in investment and trading, it has pushed the financial market intelligent quotation assistant at full scale; the trading intelligence rate reached 96%, and the number of trades increased by 50% year over year.

Scale deployment of intelligent agents cannot be achieved without a strong technical foundation. Since 2023, ICBC has successfully built a full-stack, independently controllable “ICBC Zhi Yong” large model technology system, which features four major characteristics: efficient computing power, model adaptability, rich data, and security and reliability.

Among them, the elastic computing power pool for large models, mainly powered by domestic computing resources, enables minute-level switching between training modes and inference modes; the trillion-scale Token financial data set provides ample “nourishment” for large model training, making ICBC the only bank selected in the national data bureau’s 2025 trusted data space innovation development pilot list.

Looking ahead, at the 2026 CPC Party building and business management work conference, ICBC proposed “upgrading to build an intelligent and digital ICBC and strengthening intelligent and digital momentum,” thereby upgrading the four-year “Digital ICBC” strategy to “Intelligent and Digital ICBC.”

At the performance briefing, Zhao Guidu said, “Intelligent and Digital ICBC” is an important component of ICBC’s “Fifteenth Five-Year Plan (for 2026–2030)” planning. To that end, ICBC has clarified the main construction approach of “one new initiative and three highs,” namely: intelligent-and-digital-driven new-quality productive forces, ICBC’s high-quality development, high-level security across the group, and high-efficiency governance through the integration of business and technology.

Risk Control Leap: From Preventing Risks to Improving Efficiency

In building “Intelligent and Digital ICBC,” risk control is undoubtedly one of the business scenarios that benefits the most.

In recent years, pressure on credit risk management has generally risen for two major segments in the banking industry: inclusive finance and personal loans. Yet ICBC achieved five consecutive years of declines in the non-performing loan ratio.

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

Wang Jingwu, ICBC Vice President, said that in recent years, ICBC has strengthened the foundations of risk control for inclusive loans through various approaches, including continuously improving the forward-looking capability of risk control and the precision of risk monitoring and early warning using intelligent and centralized methods.

In 2025, ICBC accelerated the building and promotion of an enterprise-level intelligent risk-control platform. It currently covers nearly 300 risk-control decision scenarios, pushing risk control work to shift from human-based prevention and control to technology-based defense and intelligent control.

According to a signed article by the aforementioned person in charge of “ICBC Zhi Yong,” in the area of risk management, the bank extracts credit risk evaluation chain-of-thought data for users such as account managers, review and approval personnel, risk officers, and risk managers. By strengthening training of the bank’s risk-control capability based on the underlying large model, it builds an enterprise-dedicated large model for intelligent credit risk assessment. The model achieves integrated analysis across multiple professional risk assessment dimensions, significantly improving the accuracy and recall rate of identifying risk customers, and it also automatically generates risk evaluation descriptions with strong readability and complete logic, supporting risk management decision-making.

In credit management, ICBC has built a multi-agent collaborative matrix covering the entire credit lifecycle. It focuses on four types of business scenarios—policy explanations, business marketing, customer analysis, and risk investigation and monitoring—and provides services from more than 20 intelligent agents, comprehensively improving the intelligent and digital service level of the credit system. Among them, the credit review intelligent agent can intelligently analyze customers’ financial data and identify financial embellishment; the efficiency of financial risk analysis has increased by 20%.

At the 2025 annual performance briefing, ICBC management disclosed that the bank’s研发 (R&D) credit intelligent assistant has already provided intelligent support for the full-process business-element information of more than 20,000 credit staff across the bank.

The annual report shows that the current “4E Center” (the risk view, measurement, early warning, and decision four enterprise-level service centers of the enterprise-level intelligent risk-control platform) has fully covered the head office and branches, driving the bank’s non-performing loan ratio to fall by 3 BP (basis points) from the end of the prior year.

It is worth noting that the application of new technologies such as AI not only improves a bank’s risk-control capabilities and efficiency, but also creates possibilities to break through bottlenecks in inclusive finance and reduce financing thresholds.

Take the poultry farming industry as an example. Under traditional credit models that rely on manual assessment, poultry farming has long faced the problems of hard-to-assess and hard-to-finance. In 2025, an ICBC local branch introduced AI image recognition technology and launched “White Duck e-loan 2.0,” using an AI system to automatically identify stock numbers and health conditions, and dynamically adjusting credit limits based on real-time monitoring data, enabling local white duck breeders to complete end-to-end online loan processing.

The annual report shows that in 2025, ICBC explored the application of satellite remote sensing in agriculture. By combining technologies such as the Internet of Things and biometric identification, it carried out collateral monitoring and post-loan management, further improving digital credit capabilities. As of end-2025, ICBC’s agricultural-related loan balance was 507.22 billion yuan, an increase of 67.61 billion yuan from the previous year.

In addition, in the technology finance arena, ICBC also provides credit support to technology enterprises lacking collateral through intelligent and digital financial products such as “Innovation Points e-loan” and “Tech Innovation Cloud ZhiShuo e-loan.” As of end-2025, ICBC’s technology loans balance reached 6 trillion yuan, with a growth rate of 19.9%.

“We will definitely comprehensively and effectively validate a series of new technologies currently in place on ICBC’s broad platform. Then we will focus on application—on what everyone needs—and on the voices of investors. We will carefully refine the technology with seriousness and dedication, and ultimately serve the real economy, and serve people’s aspirations for a better life.” Liu Jun said at the performance briefing. (Relevant data sources in the article: ICBC 2025 annual report)

Risk warning: Financial products involve risk; investing requires caution. Credit services must comply with regulatory requirements and banks’ risk-control standards.