In the Mobile Era, How Does AI Conduct Wealth Management?
-
The author, drawing from practical work experience, describes the stage characteristics and service pathways of AI wealth management in the mobile era. In the future, AI wealth management will gradually achieve open services integrated into daily life.
Readers are advised that this article is based on the practical application experience of work-related products and is not a technical primer. Any shortcomings are open to discussion and correction.
The personal vision of AI wealth management is not just about helping high-net-worth individuals preserve and grow their assets but also addressing users across different asset classes and life stages. It involves understanding users' short-, medium-, and long-term financial goals and risk preferences, combined with their overall situation and market conditions, to help them gain clarity and plan for their current and future wealth. The goal is to assist users in maintaining or improving their quality of life and achieving their financial objectives.
When conditions permit, AI will surpass human professionals in its breadth of knowledge and absolute advantage in processing vast amounts of information. Computational power will replace human labor, breaking barriers in service capacity and enabling personalized wealth management services for more ordinary users.
The mobile era signifies that AI wealth management services will no longer be confined to an office or phone calls. From customer acquisition to understanding users and delivering services, everything can be accomplished on users' portable devices, offering a personalized, companion-like service. Eventually, it may even integrate into the minutiae of users' lives, becoming a new way of living.
Currently, there are many AI services in the financial sector, such as banking, insurance, fund management, and securities. However, from an overall service perspective, the primary stage characteristics can be summarized as "starting with vertical dimensions in AI services to build the foundation of wealth management."
For accuracy in the following discussion, current wealth management services will be referred to as financial AI services.
Financial AI services on mobile devices can be broadly divided into three parts:
The diagram below outlines a service framework based on personal understanding, provided for reference and open to corrections and additions.
Based on this framework, I believe financial AI services in the mobile era are currently in the first stage: exploring core user profiles in vertical segments and building a service framework.
In plain terms, the key characteristic of this stage is: "I want to know who you are, but my services are limited."
Here, "want" refers to service providers using various methods to understand users, while "limited" describes the current scope and depth of AI services, which are still in the exploratory phase of addressing diverse user needs.
In the early stages, services are often provided in isolated points. Without systematic planning, this can lead to low user awareness of AI service functionalities and ambiguity about which needs can be addressed.
Thus, I believe the key at this stage is systematically cultivating user habits and earning their trust.
Personal experience: In practice, "want" and "limited" are not inherently restrictive. The critical factor is whether service providers have a holistic plan for serving different users, identifying effective entry points, and determining the extent of services. Such planning not only helps quickly test effective isolated services for user groups but also connects these services based on user needs, forming a cohesive AI service experience. This ensures that users receive relevant, in-depth services in appropriate scenarios, fostering habits and trust.
Figure 1: ZhongAn Insurance - ZhongAn Assistant; Figure 2: Ant Fortune - Smart Financial Assistant; Figure 3: Tonghuashun - Ask Cai; Figure 4: Qieman - Xiao Gu
Using the intelligent services of these four apps as examples, let’s briefly discuss the service characteristics of this stage:
The initial significance lies in user profiling as a prerequisite for personalized AI services.
While some leading companies have accumulated vast user data and applied certain dimensions, the primary use case currently involves leveraging strongly correlated data vectors for recommendations. However, addressing wealth management needs involves multi-dimensional variables, dynamic demands, and diverse service logics.
For instance, solving a user’s retirement asset allocation problem requires knowledge of their financial situation, investment goals, risk tolerance, desired retirement lifestyle, etc. Such private data still relies on user disclosure.
In the mobile era, I believe the greater significance of user profiling lies in clarifying the user positioning and service framework for AI services.
User profiling should not merely serve as a data foundation for solving specific problems but should also help understand "who my users are, what their needs might be, and which scenarios are suitable for service delivery." It’s not just about users articulating needs but also identifying users with needs and stimulating latent demands at the right time (additional insights are welcome).
Moreover, by analyzing services for different users, we can gather effective, high-value app behavior data, summarizing service behavior patterns and insights to make service formats more efficient, humanized, and aligned with user preferences.
Below is my personal approach to user profiling, explained within the framework:
Example:
AI services here resemble insurance sales in daily life. Sales agents develop multiple approaches for different users, continually testing ways to engage those with needs. Once user needs are stimulated, they are more willing to communicate to obtain "personalized" solutions.
After gathering user information, the sales agent follows a service logic to provide insurance plans, clarifying which needs and risks can be addressed. For instance, if a user later has children, the agent might congratulate them upon seeing a social media post and consider recommending an education insurance plan. They refine their service logic based on successes and failures. In the mobile era, much of this story is replicated in apps for AI wealth management.
Beyond the four examples above, many robo-advisory services rely on app homepage placements to attract users into closed services. I see two issues with this approach (if the homepage is the sole core service scenario, disregard this section):
1) Traffic Diversion Issues
Relying on homepage placements may seem like prioritizing the prime traffic spot, but what services do users expect from the homepage?
Most current AI services aim to attract users into closed services with catchy headlines. While immersive services theoretically yield higher conversion rates, many valuable services address needs that arise in specific contexts. Divorced from these contexts, services risk becoming clickbait, leaving users underwhelmed.
I believe timing is key—sometimes less is more. Repeating the same prompts can make users indifferent.
For post-diversion conversions, here’s a suggestion: present key conclusions and use multi-step interactions to directly collect user feedback. Experiments show this can yield ~15% feedback from new users and ~70% from diverted users (other effective methods are welcome). Operational experience also suggests each redirect may retain only 10% of users from the previous step.
If your conversational service still relies on redirects, it’s either due to high concurrency limitations or a need for redesign.
2) Personalization Issues
Homepage placements limit content types, and current service capabilities are constrained. The basic approach is to break down user needs into granular, independent solutions—one query type per semantic parsing for a small problem.
Thus, providers often showcase a few "best" services on the homepage, hoping users will click options or ask standard queries. This caters mainly to niche users who know their needs or are highly curious. Poor execution may alienate them, making it inefficient to identify which users crave which services in the current environment.
Here, we further explain the concept of 'service integration with scenarios.' This does not mean abandoning obvious, fixed entry points but rather attempting to provide users with the services they most likely need in the appropriate scenarios. Currently, users still rely on traditional service scenarios to meet their needs through apps. This approach not only helps users perceive the capabilities of AI services but also strengthens the entry points for AI services. More importantly, it fosters the habit of using 'intelligent services' to address user needs, conveying the concept of intelligent services: 'I am here, I am professional, and you can come to me first to solve any problem.'
In practical work, we have found that pushing timely and scenario-specific services to a small number of personalized users has a significant effect on reactivating users who already have experience with the service. It also helps cultivate habits among new users and stimulates their imagination about the scope of services, bringing many underutilized intelligent services into use, retention, and optimization.
In summary, AI services should have the ability to stimulate user demand. Often, users do not reject services; they simply do not know they exist or what they should expect in a given scenario.
This is also a current pain point for service providers: why it is essential not only to understand users but also to make users understand 'you.' I believe there are two reasons:
1) User Trust
Users' 'self-orientation' determines that they often interpret information based on their own perspectives.
For example, when the Tonghuashun AI robot outputs only conclusions, it often triggers opposition from users with their own logic. This is not about whether the AI or the user is right or wrong but rather that, in the absence of logical output from the AI, users may perceive the conclusions as 'full of flaws.'
Suppose the AI service could, as shown in the example, change its output to 'Breakthrough with high volume, strong bullish momentum breaking resistance.' Would such users then find the service valuable and seek more insights?
This also highlights one of the key meanings of enabling AI services to communicate with users: users themselves can help optimize the service.
2) Service Risks
Any intelligent service carries risks, especially in its current early stages, which is most evident in investment.
For instance, in intelligent investment portfolio recommendations, the more users you interact with, the more you realize that many users harbor the false notion of 'zero-risk, short-term wealth.' Their stated risk tolerance often differs vastly from their actual risk tolerance.
If an investment portfolio experiences a decline, users may quickly lose confidence and blame the 'intelligence' rather than understanding market fluctuations, their causes, or the actual impact on their returns. Therefore, helping users understand why a solution is proposed, along with its risks and benefits, is equally crucial.
The following figures illustrate a user's post-investment review of ZhaoShi MoJie ZhiTou (taken from Zhihu) and 'Qieman's' introduction to risks and returns in asset allocation. These are merely illustrative examples and will not be elaborated further.
In conclusion, for services at this stage, I believe three points are critical:
Due to these three points, the biggest limitation at this stage emerges—people, the internet + vertical experts.
Given the high cost of deep learning and the need for high-quality, rich core service data, the initial approach to problem-solving often relies on framework matching, with limited logic and targets. Thus, personalized services at this stage still largely depend on human input, requiring experts to judge and label what constitutes high-quality service data. When combined with machine learning algorithms, there is also a need for initial positioning and target expectations for vector dimensions, modeling, and results within the service framework.
Therefore, the core determinants of service outcomes at this stage are the provider's professionalism, the effectiveness of the service approach, and the speed of service refinement.
Taking investment information recommendations as an example from my work:
Those unfamiliar with investment may not see the issue here. My view is that making recommendations in the investment field is inherently challenging due to the high uncertainty and vast volume of content in the market. Filtering out suitable investment information for users is difficult, but this also highlights the user pain points.
In this example, I believe a fundamental problem arises—content tags are directly applied to the investment recommendation system, but these tags do not reach the level of understanding user needs in the investment context.
For this user, their content needs can be summarized as 'content with a clear short-term positive impact on stock prices.' The recommendation algorithm should first identify such users based on their characteristics, summarize their investment philosophy (this step can be complementary), and then learn from their reading preferences, stock behavior, and market performance to identify the content and features they prefer. This can then be linked to their investment philosophy to filter out content that aligns with their philosophy from the vast pool of information. Conversion data can further help determine which content satisfies which users.
Recommendations dominated by content tags often have little correlation with market performance, leading users to grow increasingly disappointed with the service (as reflected in extensive user feedback: 'I know you're recommending, but I find it useless.' This is just an example; actual user investment logic and recommendation matching are more complex).
Without detailed case studies, I would like to hypothesize the form of second-stage services based on the above discussion and explore the fundamental differences from the first two stages.
Let’s simulate a service scenario:
User profile: 28 years old, owns a car, plans to buy a house, married, considering having children.
Based on this, we can discuss the wealth management needs this user might have in their life and how their lifestyle might change with AI wealth management.
The figure below outlines a simple framework for four major wealth management needs for reference.
Red represents the user's goals as understood or inferred by AI wealth management, black represents the dimensions AI wealth management may need to consider and master, and blue represents the final decision conclusions provided to the user.
Combining the four major needs extracted above, what might the user's wealth outcomes look like over the next ten years?
These are not random examples; each scenario is based on real-life experiences of people around me. I believe many of you may have heard similar stories.
With AI wealth management, my brief expectations might include:
As described above, I believe the essence of the second stage is to deepen and refine vertical services and integrate the logic of past vertical services. The user's overall wealth situation, future needs, and solutions can be visualized and presented, offering personalized, companion-like services across daily, weekly, monthly, yearly, five-year, and ten-year timelines.
At this stage, AI wealth management services may resemble 'open-ended services integrated into life,' not only helping users manage major asset classes but also acting like the all-knowing life butler seen in sci-fi movies—suggesting the right clothes for the season, vacation destinations, or home renovation materials. Life would no longer require deliberate attention to trivial matters or the pain of unplanned expenses, allowing users to focus on self-improvement or enjoying life.
This article is limited by my personal work experience and understanding. I welcome readers to provide feedback, correct shortcomings, and share their thoughts.