Financial intelligent agent, standing in the wilderness of business models
What are the most concerned issues when the financial industry engages in AI currently? Last month, I attended the 2025 Bund Summit and intensively listened to the passionate discussions among industry insiders from both home and abroad for three days. The strongest impression I got was that what everyone cares about most is not the technology, but the value.
“How much productivity can digital employees improve?” “How to calculate the ROI of agents?” “Who will be responsible if something goes wrong?” These are the frequently asked questions by financial industry executives, CIOs, and CTOs.
The financial agent is currently in a typical Wilderness period: similar to the “Exodus”, it has taken the first step in technological exploration but has not yet reached the promised land of value realization and can only walk in the wilderness.
The biggest obstacle across the wilderness is the lack of a market - recognized business model.
There is still great uncertainty in terms of value measurement and return on investment for financial AI and agents. For financial institutions, the inability to clearly define the specific business growth or cost savings brought by AI makes senior decision - makers hesitant. Therefore, the most urgent proposition in the industry at present is to find the real business value of AI.
Based on my observations at the Bund Summit, let's delve into the following questions: Where lies the real value of financial AI and agents? What traps are devouring the investments? And what actions can help financial institutions find the promised land flowing with technological value?
01 Entering the Wilderness: Agents from Imagination to Action
Last year, the discussions on AI at the Bund Summit were mainly about imagination, while this year's focus has completely shifted to actions and effects in practical applications.
To summarize the industry trends in the past year, “taking action first” has become the consensus among financial and insurance practitioners.
This consensus was triggered by the anxiety about the popularization of basic large - language models. A financial IT practitioner said bluntly that the rapid popularization of basic large - language models such as DeepSeek at the beginning of the year made many financial professionals anxious: “With DeepSeek, will there be no place for financial institutions or Fintech companies?”
Under this anxiety, from the end of last year to the beginning of this year, most financial institutions started AI R & D: the IT department and the business department worked together. Even if the clear value was not visible for the time being, running through the technical link and “taking action first” was still the top choice.
Wu Lianfeng, the vice - president and chief analyst of IDC China, shared a set of data at the “AI Bank Summit Forum” of the Bund Summit, which confirmed this trend: The proportion of all industries globally that have “done nothing” in generative AI is almost 0%.
Industrial Bank is a typical example. According to Tang Jiacai, the chief information officer of the bank, in 2021, Industrial Bank regarded digital transformation as a battle for survival. At the beginning of this year, it promoted the “Artificial Intelligence +” initiative across the bank. The most iconic measure was to include intelligent applications in the assessment indicators of each business department, prompting the business side to actively think and find high - value scenarios. As of now, the number of agents in the head and branch offices of Industrial Bank has reached 630, and the number is still growing.
However, behind the actions lies the collective confusion of the entire industry.
When it comes to the measurement standard of the value of AI applications, Tang Jiacai admitted: “This year, we didn't get entangled in the quantification of absolute value. We paid more attention to the number of applications, hoping to encourage the enterprise to embrace AI first.”
This reality reflects a key turning point: in the financial circle in 2025, hardly anyone doubts the potential of AI and agents anymore. However, how to transform this potential into measurable business value has become the core issue hanging over the industry.
02 The Puzzle in the Wilderness: The Absent Business Model
In the past year, AI has blossomed across the entire financial business, from marketing, risk control, operation, approval, customer service, AI mobile banking to operation and maintenance and R & D on the technical side, penetrating the entire business process. However, these actions are more about value paving, and the real business model has not yet taken shape.
The absent business model and the unverifiable business value bring great uncertainty to the financial industry in the wilderness period. The reasons behind this are extremely complex.
The most primary limitation is the unclear responsibility attribution of AI agents, which makes it difficult to implement many businesses and thus unable to generate value.
When agents start to undertake core businesses such as credit approval, compliance review, and financial advice, they are no longer just auxiliary tools but digital employees. However, problems follow: Who are their “superiors and subordinates”? In case of compliance issues, should we turn to the digital technology department or the business department? Take compliance as an example. The AI compliance officer is developed by the technical team but serves the compliance function. Who should bear the risk? The unclear responsibility attribution makes financial institutions reluctant to deeply embed agents in the core process, further hindering the formation of the business model.
Even if embedded, there are breakpoints in the business process for agents, making it difficult to calculate the overall value and effectiveness.
Many practitioners mentioned that currently, agents have great potential in business, but their applications are shallow. Many are built in a scattered way, like building blocks. Each performs excellently in its niche, but there are a large number of breakpoints when piecing them together.
Unlike traditional software that can be priced according to functional modules, the value of agents is reflected in the final business results, such as the growth of financial management scale, the improvement of customer activity, and the optimization of approval efficiency. This transformation makes it difficult to disassemble, attribute, and price the intermediate process, so it's hard to clarify the contribution of the block - like agents.
Why are there breakpoints between agents and business? Yu Bin, the vice - president of Ant Digital Technology, pointed out sharply: lack of professionalism.
The core requirement of finance is professional logic, including rigor, compliance, and security, providing professional services without sacrificing user experience. Currently, most agents still lack this professional depth. If an AI financial advisor cannot provide professional consultation like a real person, it naturally cannot be charged at the value of an expert.
The field of AI agents has the most tolerant users, enterprises, and investors. The requirements for product completion in enterprise - level services are extremely strict. In the past, products with a functional maturity of only 30% - 60% had no chance of being implemented. Although AI agents only have technical indicators such as “token consumption”, the industry has shown rare patience. However, this tolerance is not infinite. Especially in the high - compliance and low - tolerance financial field, the market's requirements for delivery completion are rising rapidly. The window period of market tolerance for financial AI agents is narrowing.
Yu Bin, the vice - president of Ant Digital Technology, felt this deeply. He said bluntly at the forum: “Now, every time I communicate with partners, the core topic I must talk about is what specific business effects AI can bring.”
It can be said that if AI agents cannot deliver clear value for a long time, the market's patience will be exhausted soon.
03 The Road of Action: The Collective Trek across the Wilderness
Entering the wilderness is the real - life portrayal of the development of financial AI: the direction is clear, but the value path is still being explored.
It should be noted that the temporary lack of a business model and clear business value is not a reason to deny the technological direction of AI. As Wu Lianfeng, the vice - president and chief analyst of IDC China, judged, it will take at least ten years for financial AI to complete the full evolution path from “assistant → advisor → agent as application”.
That is to say, the next ten years will still be a critical window period for building core capabilities and defining value standards.
To truly get out of the wilderness, financial agents must complete the migration from technological feasibility to commercial credibility. And this migration cannot be achieved alone; it is a collective trek based on the division of ecological niches.
In the decade - long cycle of financial digital transformation, with the rapid increase in computing power and limited resources, redundant construction will only intensify internal strife. At the Bund Summit 2025, we also found out how different ecological niches are acting at present:
Focusing on Vertical Scenarios: From Generalization to Specialization
The practice of China Pacific Property Insurance is quite referential: in the first half of this year, the CIO of the company sent out questionnaires to national institutions to collect scenario suggestions and collected hundreds of suggestions within two weeks. As a large - scale financial institution, it is impossible to AI - enable all systems in the short term. The realistic path is to proceed step by step and finally focus on “high - frequency, massive, and rigid” high - value areas.
Without specialization, it is impossible to enter the core business; without core business, there is no talk of a business model.
2. Strengthening Technical Support: From Tool Delivery to Result Delivery.
The implementation of professional capabilities cannot be separated from technical support and requires model innovation. In the past, technology companies were like arms dealers, selling ammunition. However, the value demands of financial institutions require technology companies to be responsible for victory and results. In this case, technology service providers also need to change their roles.
Take Ant Digital Technology as an example. It proposed the RaaS (Results as a Service) model in the industry: instead of charging by API calls or SaaS subscriptions, it shares profits according to the actual business increment. This mechanism ties risks and rewards together, forcing technology providers to deeply embed in the business process to ensure that AI truly generates value.
Coexisting in Groups: Cracking the Risk of the Big Fish Eating the Small Fish
A bank president said bluntly, “As the president of the bank, I must ensure the core competitiveness and differentiated features of the institution. If the large - language model we use is exactly the same as that of other banks, where are our features? What's the value of me as the president of the enterprise? Now, AI has become the core of the bank's future. It must fit our development direction and adapt to our business features.”
However, not all banks have the technical support for customized AI. Small and medium - sized financial institutions need AI empowerment the most but lack the talent, funds, and data foundation the most. Therefore, the smaller the institution, the more it needs to cooperate with technology companies. Otherwise, large banks will swallow small banks, which violates the basic logic of financial risk dispersion.
The symbiotic structure where infrastructure is provided by giants and vertical scenarios are deeply explored by professional players may become the mainstream in the future.
In addition, the academic community is providing methodological support to promote the integration of technology and business. At this Bund Summit, the industry's first enterprise AI application maturity model was released. Professor Liu Shaoxuan and his team from the Antai College of Economics and Management at Shanghai Jiao Tong University systematically answered the three major questions of “Is it worth doing? Can it be done? Can it be sustained?”, covering nine key evaluations such as data governance, process adaptation, and ROI calculation, providing an operational action framework for institutions and avoiding blind investment.
Looking back at history, the human ethnic group that left Egypt and walked in the wilderness forged laws, organizations, and new civilizations while facing challenges. Today, financial AI is also in the stage of reconstructing the value system.
Searching for the business model and business value in the wilderness is a collective trek where each plays its part and coexists in the ecosystem to reach the promised land of digital and intelligent finance.
This article is from the WeChat official account “Brain Pole Body” (ID: unity007), written by Tibetan Fox, and is published by 36Kr with authorization.