12 月 10 日，艾瑞咨询（iResearch）正式发布《iResearch Vendor Insight：中国金融智能体发展研究与厂 商评估报告 (2025)》（以下简称《报告》）。《报告》明确2025年为金融智能体发展元年，但行业整体 仍处于初步探索期，96%的应用实践集中在概念验证（Proof of Concept, POC）、智能体平台部署及试 点运行阶段，仅4%进入敏捷实践期且多聚焦于职能运营类或非核心业务场景。 在此背景下，市场呈现出鲜明的客户分布特征：银行业以43%的项目数量占比成为绝对核心阵地，资产 管理类机构（含证券、基金、信托等）以27%位居第二，保险业则以15%位列第三。2025年，金融智能 体平台与应用解决方案的市场规模为9.5亿元，预计到2030年将飙升至193亿元。然而，高涨的市场预期 与尚不成熟的落地现状之间存在巨大张力，报告预警，约20%至25%的项目将面临效果不达预期甚至失 败的风险。 尽管项目数量在增长，但报告警示，从试点到有效落地之间存在显著鸿沟。预计当前阶段20%-25%的项 目将面临效果不达预期甚至失败的风险。失败原因主要归结于三大类：产品技术能力不足（如多步任务 ...

Core Viewpoint - The article discusses the transition of large models in the financial industry from pilot projects to mass production by 2025, driven by improved regulations, reduced computing costs, and the integration of large models into core business processes, ultimately enhancing competitive advantage [5][20]. Development Path - By 2025, the financial industry is expected to reach a turning point for large model implementation, with regulations and guidelines being established, and GPU rental prices significantly decreasing, making these models accessible to a wider range of institutions [5]. - The consensus among financial institutions has shifted from whether to adopt large models to how to implement them more efficiently and effectively, influenced by the maturation of regulatory frameworks, model capabilities, costs, and ecosystem development [5]. Benchmark Construction - The industry has lacked a rigorous evaluation system tailored to real business scenarios, which has led to the development of benchmarks that convert real business pain points into assessment frameworks, focusing on core capabilities such as numerical calculation and trend prediction [8][9]. - These benchmarks typically include thousands of bilingual samples and assess models across various tasks, ensuring that evaluations reflect real-world applications and capabilities [8]. Practical Applications - Large model technology is deeply integrated into core business scenarios such as investment advisory and research, transforming financial services and enhancing operational efficiency [11]. - Financial intelligent platforms have emerged, capable of supporting millions of daily active users, combining tools, services, and compliance to address core pain points in financial technology innovation [12]. Industry Empowerment - The integration of large models is expected to enhance the quality of investment advisory and research services, addressing inefficiencies and subjective biases inherent in traditional methods [17]. - Smaller financial institutions can leverage standardized services and solutions provided by large models to overcome technological barriers, allowing them to innovate without significant resource investment [19]. Future Outlook - The selection criteria for suppliers are evolving from mere technical delivery to strategic collaboration and demonstrable effectiveness, requiring suppliers to excel in accuracy, compliance, and innovative business model support [21]. - As large model applications continue to evolve, the industry is expected to move towards a more integrated ecosystem, fostering collaboration among regulators, institutions, and investors to build a secure and inclusive financial intelligence environment [24].

Core Insights - The core competitiveness of financial AI lies in the deep integration of data assets, real-world scenarios, and financial technology genes, leading to synergistic effects [1] - The company has launched its self-developed financial AI platform and various intelligent applications, including AI approval officers and AI decision assistants [1] - The company’s credit assessment products for small and micro enterprises face significant technical challenges, particularly in data processing and model risk identification [1] Data and Technology - The intelligent assessment module for small and micro enterprises covers 99% of such businesses, with an accuracy rate of 98% for macro and micro information, addressing the financing pain points of "data scarcity and difficult assessment" [2] - Solutions for improving data accuracy and stability include activating knowledge through a knowledge graph and injecting past successful experiences into the model [2] - Establishing an open and transparent assessment system is crucial for the industry [2] Future Development - The company has recently implemented two model development intelligent agents that work 24/7, significantly enhancing work efficiency and improving model performance by nearly 1% in one month [3] - Future directions include creating an end-to-end decision risk intelligent agent to automate the entire process from data input to risk judgment and decision output [3] - The company anticipates that in about two years, AI intelligent agents may appear more as "digital employees," deeply involved in various business operations of financial institutions [3]