Titanium Media International Think Tank released the “2024 Artificial Intelligence Special Topic – Large Model Application Report in the Financial Industry” (hereinafter referred to as the “Report”). The “Report” observes the progress of financial institutions and technology companies in financial large models, and details the use of large models in banks. , insurance, and securities industries, and presents the use of current large models in different financial business scenarios through multiple cases including Shanghai Pudong Development Bank, Online Banking, Sunshine Insurance, Huafu Securities, etc., and understands the implementation of large financial models through cases Existing technical, institutional and other difficulties, explore the future application of large models in the financial industry, and how to use large models to improve the efficiency of the financial industry while preventing the risks that large models bring to the financial system.
Table of contents
1. Description of the current situation: The explosive growth of large financial models, and the enthusiasm for layout in the industry is high
(1) The number of large models has surged, and AI has brought trillions of incremental business value to the financial industry.
(2) Banks: Large state-owned banks attach great importance to the research and development of large-model technology systems, and joint-stock banks are committed to exploring implementation scenarios.
(3) Insurance: The financial industry format that is most suitable for large-scale models. Mature application scenarios have obvious results in cost reduction and efficiency improvement.
(4) Securities: The application of large models is still in its infancy and is mostly used in non-decision-making business scenarios.
2. Policy review: Central policy support and supervision are both equally important, while local policy encouragement is the main focus.
3. Application scenarios: Large models are widely used in financial scenarios, and the linkage between businesses is strong
(1) Intelligent customer service: one of the most mature scenarios for the implementation of large models
(2) Intelligent marketing: realize the transformation from simple product promotion to precise personalized marketing
(3) Intelligent operation: improve operational efficiency in daily work
(4) Intelligent risk control: from data-driven to model-driven
(5) Robo-investment advisory: the “crown” of large-model financial applications
(6) Automatic code generation: Tools based on large model code generation technology will become more intelligent
4. Key cases: successful cases of application of large models in the three major financial industries of banking, insurance and securities
(1) Shanghai Pudong Development Bank: Large model reshapes retail business scenarios
(2) Huafu Securities: Large model solves the problem of inefficiency in securities account opening
(3) Online Banking: Apply the capabilities of AI large models to industrial chain finance
(4) Sunshine Insurance: Self-developed the Sunshine Zhengyan GPT large model and built an evaluation system and method for the large insurance model
(5) Hundsun Electronics: Based on the large financial model LightGPT, it launched a series of “photon” application products
5. Challenges and dilemmas: The implementation of large models in the financial industry faces five major problems
(1) Technical threshold, personnel and cost issues in the development stage
(2) Data sources, data quality and privacy protection issues
(3) Problems with computing power and memory allocation during operation
(4) Issues of interpretability and trust at the customer level
(5) The “fear” problem among internal employees caused by development
6. Ways to break the situation: Exploring the application paths of large models in the financial industry
(1) Build a new paradigm of “large model + small model” integration application
(2) Strengthen cooperation, integrate resources, and reshape capabilities
(3) Integrate supervision into the model to upgrade supervision technology
(4) Elevate the governance of artificial intelligence to the level of “top leaders”
Currently, various financial institutions and technology companies at home and abroad are actively conducting research and development on large models and implementing them in fields such as intelligent customer service, intelligent marketing, and intelligent operations. From the perspective of the model’s R&D mode, it is mainly divided into self-research and cooperative R&D. Among them, large technology companies and financial technology companies mainly conduct self-research, while financial institutions mainly conduct cooperation. From the perspective of implementation scenarios, the current large financial models have implementation scenarios in banks, insurance, and securities, including intelligent marketing, intelligent operations, intelligent office, intelligent investment advisory, intelligent risk control, and intelligent research and development.
In order to demonstrate the current implementation of large models in financial institutions, this report separately sorts out the exploration of large model applications in the three major financial industries of banking, insurance, and securities.
Banks: Major state-owned banks attach great importance to the research and development of large-model technology systems, while joint-stock banks are committed to exploring implementation scenarios.
Leading banks have actively invested in the research and development of large model technology systems and are exploring the implementation of large models in multiple business areas; joint-stock banks have launched more active explorations in the application of large models, with more application scenarios. For example, Ping An Bank has integrated large models into The technology is used in auto finance AI car inspection scenarios; small and medium-sized banks are more likely to deploy in mature scenarios through bidding.
In the fields of smart customer service, smart office, smart operation, smart R&D, etc., large models have been widely used and have achieved good results in cost reduction and efficiency improvement. The Industrial and Commercial Bank of China has created smart assistants for 200,000 branch employees, which are now online. The efficiency of outlets has been greatly improved. In 2023, the volume of intelligent processing business in the operation field will be 320 million, an increase of 14% over 2022.
At present, large model applications are gradually developing from process-level applications to decision-making levels, but “disruptive” applications have not yet appeared. Compared with large domestic banks that tend to develop their own large financial models based on general large models, overseas banks usually implement applications based on well-known overseas large models such as ChatGPT, Llama, and Google Cloud.
Insurance: The financial industry that is most suitable for large-scale models. Mature application scenarios have obvious results in reducing costs and increasing efficiency.
Insurance is undoubtedly the financial industry that is most suitable for the implementation of large models. Large models will reconstruct the sales, underwriting, claims, services and other aspects of insurance. Through sorting out the current insurance industry’s implementation of large models, the following three main points are Discover:
First of all, the large-scale models in the insurance industry have a wide range of implementation scenarios, and the most common application scenarios are agent sales assistants, marketing material generation, knowledge question and answer assistants, coding auxiliary tools, intelligent office assistants, and intelligent claims settlement. Secondly, in more mature application scenarios, the insurance industry has achieved good cost reduction and efficiency improvement effects. For example, Ping An Group disclosed in its 2023 annual report, “In 2023, it will intelligently guide customers to self-service insurance renewals worth 300.3 billion yuan. , a year-on-year increase of 13%. Ping An has the world’s leading and Asia’s largest backup service platform, with approximately 2.22 billion AI agent services; intelligent claims interception and loss reduction of 10.82 billion yuan, a year-on-year increase of 16.0%. “Finally, in terms of automatic review and verification. High-value scenarios such as insurance auxiliary support, policy summaries, and fraud rule detection are currently only attempted by a few insurance companies.
Securities: The application of large models is still in its infancy and is mostly used in non-decision-making business scenarios.
In investment banking, financial management, institutional trading and other fields, large models have rich application scenarios. However, the integration of large models and securities is still in its infancy, and most of them are non-decision-making links with simple business scenarios. For example, in the field of investment banking business, you can use the data correlation capabilities of large models to penetrate various indicators, identify causal relationships, and strengthen risk prevention and protection for the business; use the generation capabilities of large models to assist in the preparation of due diligence reports, prospectuses, etc.; In the field of wealth management, large models can learn the capabilities of account managers, researchers, and investment advisors to become customized wealth advisors for customers.
It can be judged from the characteristics and advantages of large models that the application prospects of large models in financial business are very broad. From the front desk to the middle and back office business lines, the application potential of large models has room and potential to be tapped. Large models can help Financial institutions explore market segments, optimize financial service levels, improve internal management efficiency, reduce operating costs, and improve risk control systems.
Challenges and dilemmas: The implementation of large models in the financial industry faces five major problems
Due to the characteristics of the banking industry itself, both the government level and the industry regulatory level have high requirements for the compliance, security, accuracy, and credibility of bank data. Therefore, large model technologies used in the banking industry are all research and development Mature product. From the perspective of the entire large model industry, some industry insiders believe that the implementation of large models in the banking industry is slightly behind the overall industry.
Sun Hongjun of iSoftStone said that they initially predicted that the financial industry might be the first to use large-scale models. However, judging from the final connection with customers, the financial industry is not as fast as the legal, recruitment and other industries.
Large models in the banking industry are facing implementation difficulties. Therefore, for banks that have not yet developed large models at this stage, the first step is to list the current difficulties of large models in banks and think about how to solve these “shackles” instead of blindly following the trend. Large model.
The current problems that restrict the development of large-scale banking models mainly come from five aspects, namely, technical threshold, personnel and cost issues in the development stage, data sources, data quality and privacy protection issues, and computing power and storage allocation issues during operation. , issues such as interpretability and trust at the customer level, and “fear” among internal employees after development.
The way to break the game: Exploring the application paths of large models in the financial industry
Large models sometimes generate content that seems reasonable and practical but cannot be applied to actual situations. This is called the “illusion” problem, and is also ridiculed as “read chaos”. This is one of the main dilemmas in the implementation of large models. Banking business scenarios are different from consumer-level dialogue scenarios provided by large commercial dialogue models such as ChatGPT. Once reasoning errors and hallucinations occur in large models, it will affect the customer experience at least and the security of the financial system at worst, so a cautious attitude needs to be adopted.
In response to such problems, many experts have proposed a solution of “large model + small model”. Wei Wei, professor of management at Peking University HSBC Business School, and his collaborator Huang Hongwu, CTO of Shenzhen Sipu Technology Co., Ltd., proposed a new digital financial technology that “combines generative large models, discriminative small models and AI Agent systems.” paradigm. In this architecture, generative large models are combined with discriminative small models, and the Agent system is introduced to provide more accurate and comprehensive data processing and decision support functions. Generative large models can create high-quality, highly relevant content, while discriminative small models are responsible for identifying the accuracy and practicality of the content, ensuring that the generated information is consistent with real-world data and situations and can be traced to its source.
In addition, large models are an area with extremely high capital thresholds. Whether in terms of talent investment or computing power requirements, it is a “money-burning” project. Even large financial institutions need to strengthen external cooperation to continue to advance. . Therefore, only by strengthening cooperation within the same industry and across industries and integrating resources from all parties can we reshape the ability to develop and implement large models.
Zhang Haiyan, general manager of Beijing National Fintech Certification Center Co., Ltd., believes that on the one hand, large models require large computing power. Currently, the domestic computing power market is facing a shortage of computing power supply, insufficient domestic AI ecology, computer room network construction, etc. Financial institutions need to deepen cooperation with Cooperation between all parties in the industry will jointly promote the solution of large-scale computing power deployment and application challenges. On the other hand, large models require large cooperation. The banking industry needs to accelerate the exploration and introduction of large model technology strategies and practices common in the industry, and enhance the capabilities of large models by promoting the application practice of large model algorithms in the banking industry, thereby improving the service of large models to the financial industry. ability. On this basis, to improve innovation capabilities and deepen the application of large models in the banking industry, it is necessary to explore and form a set of high-standard, low-threshold banking financial large model application models for the banking industry.(This article was first published on Titanium Media APP, author｜Li Jingying, editor｜Liu Yangxue)
For more macro research information, please follow the official account of Titanium Media International Think Tank; enter[Financial Large Model]in the dialog box to receive the full text of the blockbuster report “2024 Artificial Intelligence Special Topic – Financial Industry Large Model Application Report”.
For more exciting content, follow Titanium Media’s WeChat ID (ID: taimeiti), or download Titanium Media App
