[Patent 1] How GenAI Transforms Wealth Management

The True Pain Point of Relationship Managers

While building AI products at the bank, I spent a lot of time chatting with Relationship Managers (RMs).

They don’t lack effort, nor do they lack an understanding of their clients.

What they lack is: the capability to rapidly synthesize all relevant information and provide persuasive advice right in front of the client.

The reality looks like this: A client says, “I have 5 million to adjust my portfolio.” The RM needs to accomplish the following in just a few minutes:

• Query the client’s current asset allocation.
• Understand the client’s risk appetite and history.
• Check current market trends and interest rate curves.
• Compare deliverable internal wealth products.
• Make personalized recommendations.

This is a multi-source information synthesis + personalized recommendation task, which is the perfect scenario for GenAI.

This insight was the origin of M670472 “Financial Investment Recommendation Generation System”.

---

System Architecture: An AI Assistant That “Knows the Business”

The system design revolves around the daily workflow of the RM:

Information Input Layer

• External Servers: Real-time market info, trends, product quotes.
• Internal Servers: Client profiles, risk assessment reports, client assets, available products.

AI Processing Layer

1. NLU Module: Parses the client’s financial investment needs (which can be unstructured natural language).
2. Multi-Source Synthesis Module: Synchronously pulls external market data and internal client data.
3. Generative AI Model (LLM): Integrates all inputs to generate an initial customized investment proposal.

Visual Output Layer 4. Interactive Asset Dashboard Generation Module: Displays multi-dimensional analysis of the client’s current allocation. 5. Investment Proposal Dashboard Generation Module: Provides custom recommendations with multi-scenario simulations. 6. Dynamic Visualization Module: Combines the above into a single cohesive functional financial dashboard.

Finally, the Continuous Learning Module optimizes the AI models based on user interaction feedback over time.

---

As a GenAI Product Manager: What Did This Project Teach Me?

1. “Personalization” requires foundational data architecture.

Many GenAI products claim to be personalized, but in reality, they just stuff a couple of user fields into the prompt.

True personalization requires a complete client data model: risk preferences, asset structures, historical behavior, interaction feedback… Designing this data architecture is more important than the AI model itself.

2. “Multi-scenario simulation” is a killer feature for AI.

Traditional RMs can usually only provide one recommendation at a time. This system can simultaneously output “Conservative,” “Balanced,” and “Aggressive” scenario recommendations, letting the client choose.

This is extremely hard for a human mind to do on the fly, but effortless for an LLM. Finding an asymmetrical advantage of AI is the core job of a GenAI PM.

3. The continuous learning loop is the source of long-term competitiveness.

Launching the product is just the beginning. The system learns from every RM interaction and client reaction, making the model more accurate over time. This learning flywheel is the hardest moat to replicate.

---

Product Design Advice for GenAI POs

If you’re designing financial AI products, my advice is:

Don’t start from the technology; start from the workflow of the RM (or any frontline staff).

Find where they spend the most time, make the most mistakes, and most require info-synthesis support—that is where GenAI can create the most value.

The best positioning for AI isn’t “replacing the RM”; it’s “equipping every RM with an omniscient assistant standing right beside them.”

---

M670472 Financial Investment Recommendation Generation System (Customized GenAI Financial Dashboard) | Grant Date: 2025/05/11 | Sole Inventor: Nils Liu

💬 Read more: 2025 Year in Review (English)