Reporter Jiao Yue trainee reporter Liang Aonan

In the wave of accelerating digital transformation in the financial industry, optimizing service efficiency, strengthening risk management and control, and innovating marketing models have become the core strategic directions of financial institutions.

Since the Spring Festival this year, the practical application of large models such as DeepSeek has attracted the attention of the industry, and its technical characteristics have been further highlighted with the adaptability of knowledge-intensive industries such as finance. As an important carrier for the implementation of large-scale model technology, AI agents are gradually penetrating into all aspects of financial business processes by virtue of their integrated application capabilities in multiple scenarios, providing technical support for the upgrading of traditional business formats.

A few days ago, Flying Tiger Interactive Technology (Beijing) Co., Ltd. (hereinafter referred to as "Flying Tiger") completed a comprehensive upgrade of the intelligent body solution of the financial model. This set of solutions developed for banks and other institutions has been put into use in more than 110 financial institutions across the country, building an intelligent service matrix covering remote banking and virtual business halls.

"The complexity of the financial industry dictates that AI applications cannot remain superficial. Our ultimate goal is to achieve full-dimensional intelligent empowerment for the banking business chain by building a deep vertical agent. Shi Haidong, founder and chairman of Flying Tiger, told the "Securities Daily" reporter that with the rise of large models and agents such as DeepSeek and Manus, the market has experienced a popularization stage of technical cognition. But what financial institutions really need is a "business expert" that can solve real problems, not a general-purpose tool.

Agents reconstruct a new foundation for financial scenarios

According to Research and Market, the market size of AI agents will grow from $471 million in 0 years to $0 billion in 0 years. AI agents can be applied to many fields such as healthcare, finance, education, and manufacturing.

"The large model is not only a technological revolution, but also the engine of industrial evolution in the new era." Shi Haidong said, "Just as engines in the era of the industrial revolution gave birth to major innovations such as trains and airplanes, large models also need various agents as 'adapters' to achieve value transformation. In the financial industry, this transformation is particularly urgent, and every business scenario is worth redoing with agents. ”

Since the beginning of this year, more than 60 banks in China have said that they have accessed DeepSeek, but it is mainly used for smart office, operation and maintenance research and development, and smart investment research, assisting in the completion of report and operation copy generation, code generation and completion, investment research briefing generation, etc.

Based on the technology accumulation in the field of AI large models, Feihu has also built "remote banking" and "virtual business halls" in many banks, giving the large model financial application base and building a vertical agent empowerment matrix of "marketing-risk control-operation".

"We usually use a variety of general-purpose agents, and vertical scenario agents need to be deeply integrated with the business. Knowledge-intensive industries such as the financial industry require in-depth thinking and accurate information, which is difficult to solve with a single large-scale model paradigm. The financial industry is a natural place for agents to land. Shi Haidong said.

It is reported that since the beginning of 2023, Feihu has launched the research and development of financial vertical scene agents, from "anthropomorphic interaction", agent operation monitoring, to adaptation to financial-grade security compliance, financial scene fences, and exploration of "multi-agent collaboration", "agent special toolbox", "agent scenario strategy library and decision chain" and other capabilities, and quickly landed a number of financial cases.

Giving birth to a new curve of industry increment

This year, Manus's application innovation breakthrough based on the performance upgrade of large models has sparked heated discussions on intelligent twins in the capital market. A number of brokerages said in the research report that when the technology dividend and business scenarios collide deeply, the application of intelligent twins is expected to enter the first year of volume in 2025 years.

Focusing on the financial sector, the operational transformation of banks for a large number of customers has become an important engine for individual business growth. At present, private domain customers mainly rely on activities to operate, and the small scale of the account manager team and the average number of management accounts make it difficult to reach customers with high frequency and interact with them in a timely manner.

Qiao Yanjun, vice president of Feihu, told the "Securities Daily" reporter that this year, the financial industry is looking forward to promoting its own transformation with intelligent twins. In order to guide the agent to select the appropriate strategy from the policy base, retrieve relevant information from the knowledge base, obtain the required tools from the tool base, and verify whether the final transformation is successful from the decision base, the policy base, decision base, knowledge base, and tool library need to be continuously accumulated and updated. The accumulated resources can be used as a reference for large models to adapt to specific vertical application scenarios.

据统计，在风控机器人检测中，某省农信消费贷款余额规模为2900亿元，客户数202万。在飞虎风控检测推广后，预估每年可减少违约损失为4.96亿元。

It can be seen that as one of the important industries for the implementation of the industrial side, intelligent twins are creating more and more incremental value in the financial field.

Talking about the work done by the agent on the basis of the large model, Shi Haidong told reporters that the first is the RAG (Retrieval Enhanced Generation) knowledge base and strategy library, which builds a special knowledge base in the financial field, so that the agent can automatically retrieve professional data such as compliance clauses and product parameters before generating a reply, so as to reduce the problem of "information illusion"; The strategy library is a methodology, similar to knowledge accumulation and inheritance. The second is Multi-Agents, which assign different roles to different agents. The third is professional tools, which use special tools to enhance the ability of agents in specific vertical scenarios.

记者了解到，随着智能体业务的发展，在飞虎总营收中的占比日渐提高。以往，虚拟营业厅业务带来的收入在飞虎的整体营收中，占八成至九成。2023年起，大模型智能体业务跟传统虚拟营业厅业务的收入各占一半。

In Flying Tiger's planning, the company will continue to deepen and verticalize the financial industry and empower the financial industry with intelligent products. At the same time, 2025 years will gradually open up channels, plan to expand technology to insurance, securities, education and other fields, and actively explore overseas market opportunities.

(Edited by Qiao Chuanchuan)