財聯社4月5日訊（記者 趙昕睿）The topic of AI has continued to rise in popularity since the launch of DeepSeek. This boom has not only driven the recovery of the technology stock market, but also invisibly accelerated the penetration of AI technology into the financial sector, profoundly reshaping the financial business model, especially in investment banking, research and other businesses. Under the AI boom, domestic and foreign investment banks are keenly seizing potential opportunities and deploying AI applications one after another, hoping to seize development opportunities. Are the two different paths in terms of AI job progress, or are they distinctive?

近期，高盛CEO David Solomon在AI峰會上的一起言論引起高度關注，他表示，傳統IPO招股書往往需要6個投行人用兩周完成，但如今AI可以在幾分鐘內就完成95%的工作。據記者瞭解，事實上，去年 5 月外媒就曾報導，華爾街投行藉助 AI 分析師，幾秒就能完成投行分析師原本需數小時甚至整個週末才能完成的工作。

In this regard, the reporter interviewed some domestic and foreign investment banks to conduct in-depth research on the similarities and differences between the two markets in the application of AI investment banking from multiple dimensions.

Survey 1: What is the actual situation of the operation process of foreign AI investment banks? Why does AI avoid touching core data?

Why is AI able to complete the "5%" of the prospectus? A foreign investment banker revealed to reporters that 0% of the content in the prospectus, such as the company's industrial and commercial registration information, past financial report data, and industry public statistics, are all public information, and AI can be easily obtained and integrated. The remaining "0%", such as the management analysis in the prospectus, the issuer's equity situation, etc., still needs to be manually optimized and improved.

Although the high proportion of "95%" data is very impactful, the reporter learned that at present, AI only plays the role of an intelligent engine in foreign investment banks, and there are two reasons behind this.

Data security, as the red line of investment banking business, is naturally the primary consideration in the process of integrating AI and investment banking business. In order to prevent unauthorized access, use, disclosure, destruction, and tampering of data, foreign investment banks only allow AI access to public data. After all, most of the data that brokerage and investment banks come into contact with on a daily basis involves trade secrets and customer privacy, and the risks of allowing AI to access this kind of data are incalculable.

Another reason is that AI model training relies on public data, which makes it difficult for AI to accurately meet the needs of the private sector when it is applied in investment banking business, and the ideal matching state has not yet been achieved.

It can be seen that some foreign investment banks only allow AI to obtain public information based on data security considerations. But why haven't foreign investment banks allowed AI to touch the company's core data? The reporter's investigation found that the lack of "localized deployment" is the key sticking point.

The iterative upgrade of the internal systems of foreign investment banks lags behind, and the adaptability to AI deployment is not good, and the localized deployment is still in the promotion stage. In contrast, the localization deployment process of AI in domestic investment banks is significantly faster. In addition, according to foreign investment bankers, domestic brokerages can transmit documents through WeChat, while abroad, once such "private message" behavior is discovered, the personnel involved will be immediately fired.

It can be seen that the adaptability of the internal system of investment banks and AI, the difference in the management and control of business development systems, and the progress of localized deployment are the key factors that cause the difference between AI applications in domestic and foreign investment banks.

Survey 2: What are the common characteristics of domestic and foreign investment banks in AI operations?

AI-driven transformation in the field of investment banking has become a broad consensus among investment banks at home and abroad. In addition to the above factors, the reporter found that there are two core dimensions that coincide with the development of foreign AI investment banks by investigating the front-line dynamics of domestic brokerages.

When it comes to improving efficiency and optimizing processes, the benefits of AI are on display. By building an investment banking knowledge base, securities firms provide investment bankers with intelligent search engines to improve the efficiency of prospectus writing and reduce the error rate. Second, under regulatory constraints, domestic and foreign investment banks have taken data security as a guideline and built a strong line of defense for data privacy protection. However, due to factors such as regional regulations and business models, there may be differences in specific protective measures between the two sides.

In addition to the above two core dimensions, domestic and foreign investment banks have also shown consistency in the following aspects:

The first is to assist decision-making support: with the help of data analysis and model prediction, the project risk is accurately assessed, industry research is carried out in-depth, and market trends are predicted, providing a reliable basis for investment banking business decision-making.

The second is customer service optimization: improve customer experience and satisfaction through intelligent customer service, customer portrait and intelligent customer classification.

The third is to strengthen risk management: through the analysis of historical and real-time data, identify potential risks and give early warnings, and effectively control various risks in investment banking business.

Survey 3: Compared with foreign brokerages, what are the differences in the application positioning or progress of domestic investment banks?

In the process of AI operation, the commonality of AI application between domestic and foreign investment banks is certainly worth paying attention to, but the differentiated advantages are undoubtedly more interesting. Referring to the feedback from some brokerages, domestic investment banks are differentiated from foreign investment banks in terms of business scenarios, data focus, and technical ecological differences.

In terms of business territory, foreign investment banks focus on "globalization" expansion, and use AI more in complex scenarios such as global derivatives pricing and cross-border M&A valuation, mainly focusing on global data coverage, and AI tools need to be compatible with multiple languages. In contrast, domestic investment banks closely focus on China's capital market and further promote the "localization" strategy. While focusing on the domestic market, we actively carry out regional exploration and localized deployment.

From the perspective of technology ecosystem construction, foreign investment banks may be more inclined to purchase mature SaaS services rather than self-developed underlying models. Domestic investment banks tend to choose domestic alternatives, and have established a closer docking mechanism with domestic regulatory technology platforms. In the AI competition track, "localized deployment" has become a unique advantage of domestic investment banks.

According to a source from a brokerage investment bank, subject to domestic financial regulatory policies, data privacy and compliance are important considerations for brokerages to apply AI, and in order to ensure data security and compliance, brokerages choose to deploy AI tools locally. This allows domestic brokerages to effectively avoid data security risks while improving business efficiency with the help of AI technology, and better adapt to the local regulatory environment and business needs.

Combined with the actual operation cases of AI applications in some securities firms and investment banks, what is the stage of AI application promotion?

As the industry's first platform for the implementation of comprehensive solutions for the generation, review, extraction and search of investment banking models, GF Securities' "Investment Banking AI Wenquxing" platform has taken the lead in exploring the application practice of investment banking large models, and has explored a wealth of investment banking business scenarios to fully empower business execution, risk prevention and control, and operation management.

"Intelligent Q&A" can quickly answer investment banking business questions based on knowledge base documents, with an average accuracy rate of up to 85%, and supports traceability to the specific location of the original text, covering various types of documents such as investment banking laws and regulations.

"Smart Verification" uses large model technology to greatly improve the accuracy of document verification, the accuracy of prospectus verification is 50% higher than that of traditional AI verification, and the verification points are increased by 0% compared with traditional AI verification, while continuously optimizing audit rules to ensure the quality of information disclosure.

"Intelligent Generation" standardizes the process of template making and data filling according to specific business scenarios, reduces a lot of repetitive work, avoids the errors that may occur when writing documents manually, and realizes scenarios such as PPT generation and business manuscript assisted writing.

"Intelligent Extraction" supports the recognition of multi-modal images and complex tables, and automatically identifies chapters and more than 95 layout structures. Through the combination of large model technology and rule engine, the powerful understanding ability of large models is used to process complex semantic elements, with a character recognition accuracy of 0% and an accuracy of feature extraction exceeding 0%, breaking through the recognition limit of complex financial scenes.

In addition, the system can reuse the capabilities of the intelligent middle platform, provide AI tools such as format conversion, and support high-precision one-click conversion of multiple file formats such as JPG, DOC, DOCX, XLSX, PDF, etc., to meet the needs of various file formats.

As one of the earliest securities firms in China to apply AI technology to assist investment banking business, Industrial Securities has deeply integrated a large number of AI scenario technologies into its business links, and has landed a knowledge base of experts in the field of investment banking, among which the internal rules and regulations and regulatory inquiry database have been widely praised by front-line users. Compared with the traditional information retrieval method, the large model provides the functions of refining and summarizing and tracing the source of information, and the query and retrieval efficiency is improved by 200%. In the investment banking business, AI manuscript review tools are widely used to assist in verification, and various documents are checked more than 0 times a month on average.

At the same time, the company has simultaneously deployed the AI investment bank document intelligent writing function, which shortens the writing time of a single document from several days to less than 93 hours through automatic analysis of unstructured data and external data filling, and the data update completion rate is as high as 0%.

In addition, the established intelligent identification and review system for bank statements has become an essential part of the internal control review of investment banks, and has assisted in the verification of tens of thousands of flow documents and thousands of enterprises in recent years. The Smart Seal Review tool highlights missing and erroneous stamps. Since its launch, it has completed more than millions of stamp recognition tasks, and the accuracy rate of stamp sample recognition is more than 90%.

Soochow Securities continues to take AI deployment and R&D as its main task at this stage, and has set the following quantitative goals through practical use.

In the project contracting stage, AI is used to screen potential customer projects and complete 100% intelligent due diligence for target customers.

In the project undertaking stage, the due diligence efficiency is improved by at least 30% through AI-assisted due diligence procedures and document structuring.

At present, the flow verification of Soochow Intelligent Bank has been deployed and launched, and the scope of use has been gradually expanded, and the documents of the declared exchange have been automatically reviewed and compared.

財信證券主要在提升文檔處理效率及數據提取與校驗效率等2方面作出量化目標。文件處理方面，已基於DeepSeek當地語系化部署在實現知識問答場景全新接入，上線試運行財信證券大模型知識庫2.0版本，面向公司內部員工，在制度解讀、知識檢索、文件審核、數據校驗、數據提取等方面更為高效。業務流程優化方面，則通過DeepSeek R1模型的本地部署，為客戶及員工提供“更快、更準、更廣”的服務體驗。

Faced with the problem of homogeneous competition in the domestic brokerage industry, Caixin Securities said that it pays more attention to building competitive advantages through the differentiation of data assets and application scenarios. For example, improve the performance of AI tools by having more access to high-quality data assets in the region.

Survey 4: In the deep integration of AI into investment banking operations, what practical measures have been formulated to protect relevant core data?

Data is like a company's development DNA, which is related to business operation and development. In the context of increasingly complex data security threats, what plans and measures have domestic investment banks made to build a strong line of defense for data security?

GF Securities is based on the principle of minimum authority for investment banking business data, and the system supports the configuration of hierarchical and hierarchical data permissions for users. The specific measures are as follows:

Project information isolation mechanism based on RAG solution: Open up the traditional business system to obtain user permission data, use a multi-channel recall strategy to control the improper flow of investment bank project data, and ensure that the data recalled to the large model is accessible to users.

Data classification and grading: Meticulous classification and hierarchical management of core data of investment banking business, and clarification of the sensitivity and use authority of different data.

Data access control: Establish a strict data access control mechanism, and only authorized personnel can access and use core data. Through identity authentication, permission management and other technical means, ensure the legitimacy and security of data access.

Data usage audit: Conduct real-time audit and monitoring of the use of core data, and record the access, use, and modification of data so that abnormal behaviors in data use can be discovered and handled in a timely manner.

At present, large models generally have out-of-the-box characteristics, and most models support users to upload documents and build their own knowledge base. Therefore, in the investment banking business, there may be a risk of accidental leakage of non-public information. In this regard, Industrial Securities provides dual guarantees through two levels: business compliance constraints and technical control. First, in accordance with external regulatory provisions, establish and improve the registration and management of sensitive personnel positions and the confidentiality management of undisclosed information for investment banking business, and documents involving sensitive data are not allowed to be disseminated. The second is to process sensitive data and files through localized deployment of large models to prevent the leakage of core secrets such as customer data.

In addition, when using the investment banking management system, it needs to be used by specific personnel within a specific data scope and process stage, ensuring the effective isolation of business data and work authority. At the company level, a permission audit system has also been established to ensure business and data security through dynamic management of personnel permissions, regular audits and other measures.

In terms of data security, IB strictly implements the Guidelines for Data Security Management and Protection in the Securities and Futures Industry to classify and grade business data. Establish 3 types of management, technology, and data contact guidelines according to the process domain, and establish data encryption protection isolation, backup, and review mechanisms. In particular, for the customer information involved in investment banking business, the risk of data leakage is reduced through technical means such as desensitization, generalization, and encryption.

Soochow Securities has continuously improved the integration of AI and investment banking business by building a complete data governance system, strengthening data security technical protection, strengthening data security awareness training, establishing a data security monitoring and emergency response mechanism, and deepening cooperation and exchanges with external parties.

In the AI application track of investment banks, data security is the core point. Foreign investors should prevent data leakage and other related risks, and restrict AI from only having access to public data. Domestic investment banks have created a "protective wall" for data security by "localizing the deployment" of AI, making it a prominent advantage in AI competition. At present, AI technology is iterating rapidly, and what new changes will be triggered in the field of investment banking in the future have become the focus of continuous attention of the market.