近期，关于生成式AI投资回报的讨论已成为产业界的核心议题 。当以大模型为代表的AI走向产业腹 地，一个关键挑战也随之浮现： 如何跨越其巨大的技术潜力与真实的商业价值之间的鸿沟？ 金融业是数字化转型的先锋。在人工智能+浪潮中，金融机构如何走好大模型落地的最后一公里？ 带着这一关切，腾讯研究院与毕马威企业咨询基于对金融机构的深度调研与前沿实践分析，联合撰写了 《2025金融业大模型应用报告》。报告的核心观点是，当前AI应用的关键，并非陷入"为了AI而AI"的技 术竞赛，而是要回归技术服务商业的本质——以投入产出比 （ ROI ） 为标尺，校准应用范式，优化落地 路径。 事实上，穿透喧嚣，一场由大模型驱动的、真正以ROI为导向的生产力革命，早已在金融业的头部机构 中悄然发生： 这远非零散试点或工具集成所能企及，它要求我们像建设工业时代的电网、信息时代的光缆一样，进行 系统性的规划与投入。这不仅是一场技术革命，更是一场涵盖数据基建、组织形态、信任机制乃至社会 伦理的全维度重构。 信审效率变革：一家领先大行将过去需要数小时甚至数天完成的复杂信贷审批报告分析压缩至3分 钟，准确率提升超15%； 投研能力破壁：一家头部券商 ...

Core Viewpoints - The rapid evolution of large models and intelligent agents is ushering in a new phase of intelligent upgrades across various aspects of the financial industry, including marketing, risk control, operations, compliance, and system support [2][3] - The upcoming AICon Global Artificial Intelligence Development and Application Conference will focus on innovative practices of large models in the financial sector, particularly in investment research, intelligent risk control, and compliance review [3] - The integration of large and small models is currently the main solution in the financial industry, as small models still play a crucial role in execution efficiency and problem-solving [3][10] Summary by Sections AI Project Evaluation - When evaluating an AI project, key considerations include identifying suitable application scenarios, verifying technical paths and implementation forms, and assessing ROI throughout the development and deployment process [5][6] - The focus should be on finding pain points in small scenarios and ensuring that the necessary conditions for end-to-end implementation are met [5] Application of Intelligent Agents - Intelligent agents are being utilized in various financial business scenarios, such as data insights, due diligence, and investment advisory, but face challenges due to the immaturity of foundational models and tools [3][7] - The combination of agents and large models is seen as beneficial, particularly in internal services, while external services require careful evaluation of compliance and ROI [6][7] Challenges in Implementation - Major challenges include the performance drop of large models when deployed locally, the high hardware costs associated with private deployment, and the difficulty for business personnel to accurately express requirements for workflow construction [26][27] - The sensitivity of large models to their operating environment poses significant challenges, as even minor changes can lead to inconsistent outputs [27][28] Future Directions - The future of intelligent agents in finance may involve the development of dynamic defense capabilities against AI-driven attacks and the establishment of an intelligent agent alliance for risk control across the industry [32][34] - There is a need for collaboration between traditional AI and large models to address specific financial scenarios, ensuring compliance and data quality while managing computational resources effectively [35][36]

Group 1 - The core theme of the live discussion is the application of large models in financial scenarios, questioning whether intelligent agents are a productivity tool or a false proposition [2][3]. - The live event features practitioners from banks, Tencent, and leading fintech institutions, focusing on the practical implementation of AI technology in finance [3][4]. - The discussion will cover various applications of large models in finance, including risk control, customer service, due diligence, and compliance [4][7]. Group 2 - Attendees will receive a resource package titled "Exploration of AI Applications and Trends in Finance," which includes technical solutions, application value, and practical experiences [7]. - The event aims to address challenges and solutions related to the use of large models in risk control, as well as new ideas and attempts in the "AI + Risk Control" domain [7]. - Participants will gain insights into the practical content and application results of financial risk control models, along with commercial considerations for decision-making [7].

Core Insights - The World Artificial Intelligence Conference (WAIC) highlights the increasing practicality of AI applications in the financial sector, with a focus on digital labor and the restructuring of business interactions [1][2] - Financial intelligent agents are evolving from single-point applications to core business scenarios, such as credit decision-making, indicating a shift towards scalable applications [1][2] - The financial industry is seen as an ideal sector for AI implementation due to its high digitalization and data density, with various financial sub-industries exploring intelligent agent applications [2] Group 1: AI Applications in Finance - Financial intelligent agents are expanding in both depth and breadth, with a significant increase in the number of application scenarios and improvements in business efficiency [2] - Major banks have implemented a technology architecture combining general models, specialized models, and intelligent agents, while leading securities firms have launched multiple AI application products [2] - The insurance sector is advancing the construction of intelligent claims systems, showcasing the diverse applications of AI across financial services [3][4] Group 2: Transformation of Business Interactions - The interaction model in financial services is transforming, exemplified by Shanghai Bank's launch of an AI mobile banking app that allows users to conduct transactions through conversational interfaces [5] - This shift from traditional menu navigation to "dialogue as a service" enhances user experience and personalizes financial services, particularly benefiting older customers [5] Group 3: Challenges in AI Implementation - Despite the advancements, challenges remain, including the "hallucination" issue of large models, which can lead to inaccuracies in instruction adherence [6] - The need for high-quality data sets for training specialized models is critical, requiring significant investment and long-term commitment [6][7] - Many financial institutions lack the engineering capabilities to integrate business needs, computational power, models, data, and knowledge bases effectively [7]

Core Insights - The year 2025 is anticipated to be the breakthrough year for AI Agents, marking a transition from cutting-edge technology to practical applications [1] - AI Agents are expected to enhance productivity by directly impacting core production scenarios, enabling businesses to achieve cost efficiency and higher productivity [1][3] - The financial industry is entering its own era of AI Agents, with leading fintech companies like Ant Group and Qifu Technology launching financial AI products [2] Financial AI Agents - Financial AI Agents are defined as autonomous AI entities capable of perceiving their financial environment, reasoning, decision-making, and executing complex financial tasks [7] - Unlike traditional automation tools, which require predefined rules and processes, AI Agents can operate independently, adapting to various situations and continuously learning from their experiences [11][12] - The capabilities of AI Agents include end-to-end automation, real-time response to environmental changes, intelligent planning, and continuous self-optimization [16][17][19] Productivity Revolution - The emergence of financial AI Agents is seen as a catalyst for a significant productivity revolution within the financial sector, moving from peripheral applications to core business functions [21] - Financial AI Agents can break down process barriers, enabling comprehensive automation and enhancing service delivery to underserved populations [20][22] - The integration of AI Agents into financial services is expected to lower operational costs and improve service accessibility, thereby transforming the financial landscape [20][31] Challenges and Opportunities - Financial institutions face challenges such as data silos, high personnel costs, and the need for personalized services, which AI Agents can help mitigate [27][30] - The deployment of AI technology requires significant investment, with initial costs often exceeding millions, but the potential for quantifiable and sustainable value growth is promising [29][31] - The current state of financial AI development includes both single-agent and multi-agent systems, allowing institutions to gradually adopt AI solutions without overhauling existing frameworks [32] Strategic Implementation - Successful implementation of AI Agents in financial institutions is linked to direct involvement from top management, particularly CEOs, to drive financial performance improvements [35] - The transition from digitalization to a new paradigm in finance necessitates strategic restructuring, organizational change, and cultural transformation [35]

Core Viewpoint - The application of large models in the financial industry is transitioning from an exploratory phase to a practical phase, becoming a necessity rather than an option [2][3]. Group 1: AI Integration in Financial Institutions - Financial institutions are increasingly integrating large models into their core business processes, moving beyond auxiliary tools [2]. - The current trend shows that AI applications in finance are shifting from customer service to core business areas such as wealth management and insurance claims [3]. - The year is being referred to as the "Agent Year," indicating a significant evolution in AI capabilities from digital assistants to digital employees [3]. Group 2: Challenges in AI Implementation - Financial institutions face challenges with large models, including a lack of understanding of financial contexts and concerns about data safety and compliance [3][4]. - There is a need for a specialized financial model rather than generic models, which are often seen as inadequate for the complexities of the financial sector [4]. Group 3: Successful AI Implementation Factors - Successful implementation of financial AI requires a specialized financial model, a responsive knowledge base, and the ability to facilitate business analysis and decision-making [4]. - Ensuring safety, compliance, and professionalism in financial models is crucial for creating effective financial intelligent agents [4]. Group 4: Pathways for AI Deployment - Ant Group has identified four pathways for AI deployment in financial institutions: building a model platform, creating AI-native mobile banking services, applying models in business scenarios, and prioritizing model deployment as a key project [5]. - The company offers flexible service models, including private deployment, SaaS subscriptions, and performance-based billing [5]. Group 5: Collaboration and Innovation - Ant Group plans to launch over a hundred intelligent agent solutions across various financial sectors, including wealth management and risk control [6]. - The integration of AI into business processes is seen as a strategic opportunity for financial institutions to drive organizational upgrades [6]. Group 6: Future of Financial AI - The development of financial AI is viewed as a long-term process requiring continuous iteration and improvement [11]. - Ant Group is working on creating independent financial models to bridge the gap between generic models and the specific needs of financial institutions [19]. Group 7: Data Security and Knowledge Management - Data security concerns are addressed through methods such as data anonymization and hybrid model deployment [17]. - The importance of a unified knowledge base is emphasized, as fragmented knowledge can hinder the effectiveness of AI applications in finance [18]. Group 8: Ecosystem Collaboration - Ant Group is merging its AI and cloud services to enhance product interoperability and address the challenges faced by financial institutions [20]. - The company aims to provide a comprehensive AI product system that considers both technical and business aspects of AI implementation [20].