Chapter 4: The Power of Proprietary AI Models and Expert Systems

In the rapidly evolving landscape of wealth management, artificial intelligence (AI) has emerged as a transformative force, it is beginning to reshape how financial services are delivered and experienced. While public Large Language Models (LLMs) from companies like OpenAI, Anthropic, Google, and Facebook have garnered significant attention, relying solely on these models presents limitations that can hinder a firm's ability to provide truly personalized, secure client service and advice.

This chapter explores the compelling reasons why developing and using proprietary AI models and highly trained expert systems is crucial for wealth management firms aiming to enhance accuracy, maintain data privacy, and achieve competitive differentiation.

As we delve into this topic, we'll examine the limitations of public LLMs, the benefits of custom AI models, the process of building proprietary systems, and the integration of expert systems. We'll also explore case studies of successful implementations, regulatory considerations, and the future outlook of AI in wealth management.

By the end of this chapter, you'll have an understanding of why proprietary AI is not just an advantage but a necessity for the future of wealth management.

Limitations of Public LLMs in Wealth Management

Public LLMs have revolutionized natural language processing by providing an unprecedented ability to understand and generate human-like text. However, these models are trained on vast datasets from the internet and other publicly available datasets, encompassing a broad spectrum of topics, many of which are unrelated to financial services. While beneficial for wide applicability, this generalist nature introduces several challenges when these models are applied in wealth management.

Lack of Domain-Specific Expertise

Public LLMs are not trained specifically on financial planning principles or wealth management principles. Their knowledge is broad but relatively shallow, although it is getting better rapidly. Currently, they lack the depth required to navigate complex financial planning topics, regulatory requirements, and personalized client needs.?

You would not want your general practitioner physician performing a complex neurosurgery.? You would look for a highly skilled neurosurgeon to do this work.? Think of a public LLM as a general practitioner and a proprietary AI Model as a highly trained neurosurgeon.?

This limitation becomes particularly apparent when dealing with nuanced financial scenarios. For instance, a public LLM might struggle to provide accurate advice on complex tax optimization strategies or estate planning for high-net-worth individuals across multiple jurisdictions.

The model's generic knowledge base simply isn't equipped to handle the intricacies of wealth management, tax management, cash flow management, and overall optimization across all of these areas. However, Public LLMs, especially trained GPTs, are getting better constantly. We can't predict when these core models will be good enough to handle more advanced scenarios, but it will likely happen before the end of the decade. ?

Risk of Hallucinations

One of the well-documented issues with public LLMs is their propensity to generate "hallucinations"—plausible-sounding but incorrect information. In financial services, where decisions are often made based on precise data, time-tested financial planning principles, and appropriate adherence to regulations, such errors can lead to significant consequences, including financial loss and regulatory transgressions.

Consider a scenario where a wealth manager relies on a public LLM to summarize recent changes in tax laws affecting estate planning. If the model generates inaccurate information, it could lead to flawed advice, potentially resulting in substantial financial implications for the client and reputational damage for the firm.

Data Privacy and Security Concerns

Public LLMs are typically hosted on external cloud platforms. While these platforms may have robust security measures, they still pose risks, particularly when dealing with sensitive client data and firm IP (Intellectual property). Regulatory requirements in financial services demand strict data privacy and security protocols, and using external AI platforms can introduce vulnerabilities.

In wealth management, client trust is one of the most valuable assets a firm has. Any perceived risk to data security or privacy can severely damage client relationships and the firm's reputation. This concern is particularly acute when dealing with high-net-worth individuals who require the utmost discretion, confidentiality, and security in the financial advisory firms they work with.

There are ways to access these public LLMs in a private cloud, which makes it more secure, but ensuring that it provides appropriate advice remains challenging.

Lack of Customization

Public LLMs are designed to be general-purpose tools, which means they cannot fully align with a firm's unique methodologies, investment strategies, and client service philosophies. This limitation can result in generic advice that fails to reflect the firm's distinctive approach to wealth management.

For instance, if a firm specializes in high-net-worth retirement planning with a specific methodology for tax-efficient withdrawals and estate preservation across multiple asset classes and jurisdictions the answers of a public LLM would not adhere to its methodology.

On the other hand, several public financial planning and retirement advice GPTs are already pretty good, but they wouldn't meet a firm's data security, privacy, and regulatory compliance requirements. These GPTs are getting better and better as the core models improve and the marketplace grows.

A public LLM may lack the depth needed to provide optimized, tax-sensitive recommendations that align with both the client’s current portfolio and projected changes in tax law.

Without a proprietary AI model trained on this specific strategy, the firm risks delivering generic guidance that fails to leverage the unique tax and estate planning options available to high-net-worth clients. A public LLM, even a highly custom GPT, doesn't do what you can with a modern portfolio management system like Orion, Black Diamond, Envestnet, Advyzon, and Addepar.

Regulatory Compliance Challenges

The financial services industry is heavily regulated, with complex and often region-specific rules governing various aspects of wealth management. Public LLMs, not specifically trained on these regulations, may struggle to provide consistently compliant advice across different jurisdictions. This article has five specific examples of what firms should do from a compliance perspective. Link to the article

It is not wise to leverage AI to provide advice that is not fully compliant with all data privacy, security, and regulatory requirements.

Benefits of Developing Custom AI Models

Given the limitations of public LLMs, developing proprietary AI models tailored to the specific needs of wealth management firms offers several compelling advantages:

• Enhanced Accuracy and Relevance: Trained on firm-specific data, including anonymized financial planning information, tax returns, estate plans, and client interaction histories, these models deliver precise recommendations aligned with individual client goals and regulatory compliance standards. This specialized training allows for highly tailored advice on optimizing cash flows, minimizing tax liabilities, structuring estates, and financial planning to meet client goals. For example, a proprietary model could analyze a client's entire financial portfolio, tax situation, and long-term goals to recommend a personalized investment strategy that balances risk, return, and tax efficiency.
• Competitive Differentiation: In the financial services industry, where trust and expertise are key differentiators, leveraging a proprietary AI model can set a firm apart from its competitors. A custom AI model allows a firm to offer services and solutions that cannot be easily replicated by others relying on off-the-shelf technology.? For instance, a wealth management firm could develop an AI-driven portfolio optimization tool that not only considers financial returns but also aligns with a client's goals and core values. This level of personalization and alignment with client values can be a powerful differentiator in attracting and retaining clients.
• Data Privacy and Security: Developing proprietary AI models allows firms to maintain control over their data. These models can be hosted on private clouds, ensuring that client data is protected in accordance with the highest security standards. This control over data is particularly crucial in wealth management, where client confidentiality is paramount. As one industry leader stated, "Our clients trust us with their most sensitive financial information. Using proprietary AI models allows us to leverage advanced technology while keeping that data within our secure ecosystem.

Building Proprietary AI Models: A Strategic Approach

Developing a proprietary AI model is a complex but rewarding process that involves several key stages. Each stage is crucial to building a model that is not only effective but also aligned with the firm's goals and client needs.

Data Collection and Preparation

The first step in building a proprietary AI model is data collection and preparation. In wealth management, data comes from a variety of sources, including:

• Client financial records
• CRM Data
• Transaction histories
• Client interactions and communications
• Financial Planning Data
• Portfolio Management Data
• Marketing preferences

The quality of the AI model depends heavily on the quality of the data it is trained on. Therefore, this stage involves not just gathering data but also cleaning it, structuring it, and ensuring its accuracy and relevance.

Model Selection and Training

Choosing the right AI model is crucial to building a proprietary AI model. The choice of model depends on the firm's specific needs, such as whether the goal is to improve client engagement, optimize investment strategies, or enhance compliance monitoring.

Common types of models used in wealth management include:

• Machine learning algorithms for predictive analytics
• Natural Language Processing (NLP) models for client communication analysis
• Deep learning models for complex pattern recognition in client data

The training process involves feeding the selected model with the prepared data and fine-tuning its parameters. This is often an iterative process, requiring multiple rounds of training and evaluation to achieve optimal performance.

Validation and Deployment

Before deploying the AI model in a live environment, it must be thoroughly validated to ensure its accuracy, reliability, and robustness. Validation involves testing the model on a separate set of data that was not used during training.

Key aspects of the validation process include:

• Accuracy testing: Ensuring the model's predictions or recommendations are correct and reliable.
• Stress testing: Evaluating the model's performance under various scenarios.
• Bias detection: Checking for any unintended biases in the model's outputs.
• Compliance checks: Verifying that the model's recommendations align with regulatory requirements.

Once validated, the model can be deployed. This often involves a phased approach, starting with a pilot program before full-scale implementation.

Continuous Learning and Improvement

AI models are not static; they must be continuously updated and improved to remain effective. This requires:

• Ongoing data collection to keep the model updated with the latest information.
• Regular retraining to incorporate new data and adapt to changing conditions.
• Monitoring the model's performance in real-world scenarios
• Gathering feedback from users (both clients and wealth managers) to identify areas for improvement.

Feedback loops should be integrated to identify areas for improvement, and regular updates are necessary to keep the model current with new data and changing market conditions.

This continuous improvement cycle ensures that the AI model remains a valuable tool for the wealth management firm, adapting to new challenges and opportunities as they arise.

The Role of Expert Systems in Wealth Management

While AI models, particularly those based on machine learning, have gained much of the spotlight, expert systems remain a powerful tool in the wealth management industry. Expert systems are a type of AI that relies on a knowledge base of human expertise to perform tasks that typically require human judgment.

What is an Expert System?

An expert system is designed to emulate the decision-making abilities of a human expert. It consists of two main components:

• A knowledge base that contains domain-specific information and rules.
• An inference engine that applies logical rules to the knowledge base to solve problems or make decisions.

Unlike machine learning models, which learn from data, expert systems apply pre-defined rules and logic to make decisions. This makes them particularly useful in areas of wealth management where there are established best practices or regulatory requirements.

Advantages of Expert Systems

Expert systems offer several advantages in wealth management:

• Consistency: They apply rules consistently, reducing the risk of human error.
• Transparency: The decision-making process of expert systems is typically easier to audit and explain

I will post the end of the chapter in a subsequent post.