HomeArticle

The AI financial analyst is here, and both Hong Kong listed companies and family offices have made investments.

36氪的朋友们2026-11-17 08:23
AI is becoming a new friend in investors' contact lists.

In Hong Kong, AI is becoming a new contact added to investors' address books.

There is no need to download a new App. Simply scan the QR code and add a contact named ARTi, then you can inquire about stock market quotes, individual stock trends just like chatting, and even ask it to generate an investment research report. Different from ordinary chatbots, ARTi will record every judgment it makes on stocks, and verify it against the real market trend at a certain point in the future.

Behind this product is an AI startup incubated from the family office business system in Hong Kong.

In June this year, Delin Holdings (1709.HK), a Hong Kong-listed company, disclosed that ARTi, the AI investment research platform in which it made a strategic investment, has received 1.5 million US dollars of Pre-A round investment from C Capital, a well-known Hong Kong fund, with a pre-money valuation of 30 million US dollars. In July, ARTi was officially launched. In just 4 months, the product has accumulated 70,000 users. Recently, Eason Zhang, CEO of ARTi, revealed to us that the company has started its Series A financing.

At a time when AI Agents are springing up everywhere, financial investment is a scenario with relatively fierce competition. As more and more financial Agents are born, how many human analysts do investors still need?

Financial Agent Born Out of Hong Kong Family Offices

"The first version of ARTi was my own Personal Agent." The birth of ARTi originally stemmed from Eason Zhang's personal needs.

Eason Zhang used to be a strategic consulting partner at a top global consulting firm, serving financial institutions for a long time. After leaving in 2015, he founded a fintech enterprise which was later acquired. In the following ten years, he successively participated in investments in the technology sector.

At the beginning of this year, with the rapid development of AI Agent and AI coding tools, he began to try to build an investment research Agent for himself. Originally, he only wanted to use AI to handle daily financial research work, but later he provided the product to friends around him for trial and received good feedback.

In his view, what investors often face is not insufficient information, but information overload. After analysts of traditional research reports give bullish or bearish views, the market is still changing constantly, but the research conclusions may not be systematically tracked and reviewed. After a report is delivered, the research work often comes to an end. Therefore, he hopes to solve two problems: one is to use AI to process the massive amount of information that is too large for human beings to exhaust, and the other is to make every investment judgment verifiable.

Now users can also get investment analysis by opening ChatGPT, Claude or other general large models. Eason Zhang believes that the difference between general large models and professional financial Agents cannot be simply understood as the quality of answers, but lies in whether they have sufficient data sources, analysis frameworks suitable for financial scenarios, and verification mechanisms for judgment results. ARTi currently adopts the technical path of multi-model collaboration, combines the capabilities of different large models, and establishes a set of engineering systems in data retrieval, model scheduling, analysis process and result verification.

Everyone is talking about the hallucinations of large models. The ARTi team found that the hallucinations in finance have their own characteristics. In actual measurement, hallucinations may appear in the following forms: taking expired data as current data, misidentifying companies with the same name or listed in two places, etc.

In layman's terms, hallucinations are content generated by the model that is fluent and seemingly reasonable, but inconsistent with the facts. A 2025 paper by OpenAI researchers gave a simple explanation: both training and evaluation reward guessing instead of admitting uncertainty, just like when you encounter a question you can't answer in an exam, guessing an answer is better than leaving it blank.

In the financial field, there are at least four forms of such hallucinations. The first is fabricating numbers and sources: making up plausible growth rates, a plausible source of research reports, etc. The second is being confused about the date: taking expired data as current data, or quietly supplementing and fabricating relatively reasonable numbers when data is missing. The third is misidentifying the object: companies with the same name, companies listed on both A-share and Hong Kong stock markets. Once the identification is wrong, all subsequent analysis will be invalid. The fourth is that everything seems normal when an error occurs: interface timeout is described as "no news".

"Researchers found that the model is trained to prefer guessing over admitting that it does not know, and it is useless to tell it 'please be accurate'. So we don't expect it to make no mistakes, but build mechanisms outside: sources need receipts, data comes with timestamps, abstain if there is insufficient evidence, and check judgments against real market conditions when they expire. We don't dare say it will never make mistakes, but every mistake should be able to be found."

Most of these mechanisms are summed up after detours: chained retry for research reports, skipped processes are migrated, and an access control is added every time an accident occurs. Coupled with more than 16,000 targets, nearly 10 million pieces of data every day, with different calibers, currencies and trading calendars. "What users see is a contact, and behind it is a whole set of financial engineering."

In terms of team background, ARTi adopts the mode of combining financial professionals and AI technical personnel. Its CTO Steve has deep background in Silicon Valley technology entrepreneurship and AI technology, and is responsible for the overall technical architecture of the product. The team also has senior technical leaders in model algorithms. In addition, the chief scientist from the core AI laboratory in Silicon Valley serves as an expert consultant, and cooperates with the team in the directions of model self-evolution and financial data application.

The product soon received support from Delin Holdings. Delin Holdings is a comprehensive financial services group listed in Hong Kong, starting with family office and wealth management business, and gradually expanding to securities trading, investment banking, asset management and digital finance.

Delin said that it is optimistic about ARTi mainly based on two judgments.

On the one hand, there is real demand for financial investment research. Users need to continuously obtain market information, conduct analysis and form judgments. This kind of recurring demand can support AI products to accumulate usage data, and then improve model performance.

Secondly, ARTi has chosen a relatively lightweight product route. It did not invest a lot of money to develop an independent App from the very beginning, nor did it directly enter the securities trading execution link. Instead, it chose to connect the Agent to existing communication platforms such as WhatsApp. After adding the contact, users can inquire about market quotes, gold prices or the investment value of a certain stock just like chatting. This product form of "light entry" not only lowers the user threshold, but also allows the team to iterate faster and control product development and operation costs.

In this way, the starting point of ARTi's entrepreneurship is different from that of ordinary AI application startups. It not only has financial scenarios, but also can carry out product testing with the help of existing customer resources.

Who is Willing to Pay for AI Analysts?

Financial investment is a business built on trust. For ARTi, Hong Kong is not only the starting point of its entrepreneurship, but also an important support for its business model.

As an international financial center, Hong Kong has a mature financial market, regulatory system and licensed institution ecosystem, and connects the mainland and international capital markets at the same time. For investors who focus on Hong Kong stocks, US stocks and A-shares, there is real demand for cross-market information acquisition and investment research. In recent years, Hong Kong has also accelerated the layout of fintech. The "Fintech 2030" launched by the Hong Kong Monetary Authority lists artificial intelligence as one of the strategic directions, and the first five-year plan of the SAR government also proposes to promote the deep integration of AI and industries.

ARTi hopes to seize this opportunity to combine the AI technical capabilities from Silicon Valley with the existing financial business resources in Hong Kong. The wealth management business, financial institution cooperation network and related licensed resources of its strategic investor Delin Holdings provide the foundation for it to reach professional investors.

At present, ARTi's users are mainly divided into two categories. One category is Hong Kong university students, overseas Chinese and new workplace beginners who are just starting to get in touch with investment. They lack systematic investment research experience and hope to use AI to understand the market and learn investment methods. The other category is high-net-worth investors, mainly from Delin's wealth management business, as well as ARTi's cooperation network with other family offices and financial institutions.

In Eason Zhang's view, the two types of users correspond to different commercial values. Young investors have high acceptance of new technologies, and can help the team quickly find the shortcomings of the product through a large number of real questions and usage feedback, and also provide richer real scenarios for judgment verification. High-net-worth investors have clearer investment research needs and payment capabilities, and are also the core customers that ARTi most hopes to attract at this stage.

He admitted that the financial market never lacks people who are willing to pay for investment information. Whether it is purchasing professional databases, subscribing to research reports, or seeking investment advisory services, high-quality investment research always has its market. However, to make investors willing to switch from traditional services to new tools, reliable and professional analysis is indispensable. Although many securities firms and securities trading platforms have now accessed AI assistants, most of them still regard it as an additional function of their existing businesses. Simple information aggregation and financial data listing are far from meeting the requirements of professional investors, who are in greater need of research services with clear analysis frameworks, continuous tracking capabilities and verifiable historical judgments.

This is exactly the market that ARTi hopes to enter. At present, the company mainly provides products as value-added services to family offices and high-net-worth clients, and has not officially launched large-scale charging, but has received feedback on willingness to pay from some users.

On the individual user side, ARTi plans to adopt a subscription model, setting up free and different levels of paid packages. According to the current design of the team, the paid tiers include 20 US dollars, 100 US dollars and 200 US dollars per month, and users can use investment research services and virtual trading functions through corresponding points.

Eason Zhang plans to win more than 50,000 core users with strong payment ability in Hong Kong. He admitted that this is not easy, but for a startup that takes professional investment research services as its main selling point, finding enough high-value customers is more important than simply pursuing the scale of registered users.

At the same time, the institutional market may bring greater commercial space.

Family offices, securities firms, asset management institutions and research institutions themselves need to undertake a large number of information collection, financial analysis and investment judgment work. If ARTi can embed investment research capabilities into the existing work processes of financial institutions through APIs and other methods, it does not need to build a completely independent financial service system from scratch. This also means that ARTi does not necessarily have to become another Futu or Tiger Brokers.

The logic for financial institutions to purchase AI services lies in, on the one hand, reducing repetitive labor by automating basic research work; on the other hand, using continuous tracking and judgment verification mechanisms to provide investment managers with more systematic decision-making references.

However, the requirements for AI in the financial field are much higher than those for general information services. Even if AI can achieve high judgment accuracy at a certain stage, it does not mean that it can continue to achieve the same performance in different market cycles. Accurate historical judgments do not necessarily mean that users can obtain investment returns.

Eason Zhang maintains relatively restrained expectations on this. According to his vision, ARTi is currently closer to a junior analyst who has received good financial education. In the future, he hopes that this system can have the research capability of a financial analyst with more than 5 years of work experience, including information collection, financial analysis, valuation modeling, and relatively independent judgment on industries and enterprises.

"It is not yet a investment master. We still leave the final judgment to human beings, and it only provides continuously traceable suggestions."

In Eason Zhang's view, after AI enters the financial industry, it is expected to improve the overall efficiency of investment research. When the threshold for obtaining professional information is gradually lowered, investors' ability to use AI, experience accumulation and independent judgment may become a new watershed.

This article is from the WeChat official account "ChinaVenture", author: Wei Xianghui, published by 36Kr with authorization.