Photo created by author using Dall·E 2
How wealth managers can adopt AI for future success.
The launch of the ChatGPT conversational chatbot prototype last November put artificial intelligence (AI) directly into the hands of individuals, wealth managers, and businesses for the first time. Within two months it had reportedly become the fastest-growing consumer application in history, with over 100 million monthly active users (UBS study). ChatGPT can write articles, complete homework, and brainstorm ideas, providing detailed responses on a variety of topics within seconds.
According to ChatGPT, “AI has already started to impact the wealth management industry and is expected to continue to do so in the future—particularly in automated trading, personalized investment advice, fraud detection, improved customer service, and enhanced portfolio management.”
This article explores the evolving needs of wealth management clients, provides an overview of AI and ChatGPT, and discusses the need for wealth managers and financial services domain experts to embrace AI in their offerings and services while understanding the potential risks to continue to thrive and serve their wealth customers.
New Wealth Management Customers Will Expect On-Demand Research and Services
A recent McKinsey study predicted that by 2030 up to 80 percent of new wealth management clients (the Gen X’s and Millennials) will require data-driven, hyper-personalized advice that can be delivered seamlessly, continuously, and effectively—afterall, they have grown accustomed to on-demand streaming (e.g., Netflix) and “one-click” purchasing (e.g., Amazon).
Meeting this customer expectation requires more than automation. Wealth managers who can leverage data and analytics in delivering services, cutting investment research costs and client reporting and acquisition time, will continue to thrive. Understanding how to leverage AI and tools like ChatGPT is existential.
ChatGPT and Other AI Tools Have Sparked Interest in Numerous Industries Including Wealth Management
Since the launch of OpenAI’s ChatGPT, a Large Language Model (LLM) designed to understand and respond to natural language (much like humans), industries from media to energy have been exploring how they can benefit from generative AI, that is, algorithms (such as ChatGPT) that can be used to generate content, including audio, code, images, text, simulations, and videos.
Should this concern the money manager? Yes.
According to ChatGPT,
“ChatGPT stands for ‘Conversational Generative Pretrained Transformer,’ while ‘GPT’ stands for ‘Generative Pretrained Transformer.’
The “secret sauce” of ChatGPT that makes it different from its predecessors is its size, training data, and the architecture of its neural network. ChatGPT is one of the largest language models ever created, with over 175 billion parameters. It was trained on a massive data set of text data from the internet, books, and other sources, which allows it to have a broad and diverse understanding of language.
The architecture of the neural network used in ChatGPT is based on the transformer model, which allows it to process and understand the relationships between words and phrases in a sentence. This architecture, combined with the vast amount of data used to train it, enables ChatGPT to generate highly natural and coherent language responses, making it feel like you’re having a conversation with a real human.
Furthermore, ChatGPT is highly adaptable, with the ability to perform “zero-shot learning” and “few-shot learning,” which allows it to generate responses to prompts it has never seen before and adapt to new tasks or domains with just a few examples of training data.”
While critics of ChatGPT have indicated the technology has been available for many years, this is the first time the technology has been available to the general public. Consequently, the launch of ChatGPT has produced an explosion of applications and investor interest in AI. Greylock said, “AI has become the enabling technology of our time; investing in FinTech increasingly means investing in AI.” With financial services estimated at around 25 percent of the global economy, there is an opportunity for financial services to lead the AI industrial revolution.
Financial Services Have Already Identified Multiple Opportunities to Benefit from AI
The digitalization of financial services has allowed us to apply for loans, transfer funds, buy insurance, trade stocks, and make payments from virtually anywhere using our computers or phones. This, in part driven by Fintech, has led to the generation of vast quantities of data – the raw material for AI.
Fundamentally, AI is capable of analyzing large volumes of data and identifying patterns and anomalies. It has typically been leveraged by the payments industry to detect fraud and potential money laundering activities. AI is also used by the securities regulators for market surveillance (e.g., SEC’s Consolidated Audit Trail) and by investment management firms to customize content for their customers through analysis of their historical investments.
Chatbots (also known as conversational AI) have been used in financial services to provide faster answers to customer queries and reduce call center volume and wait times; however, chatbots have historically been easily confused and unable to adapt their behavior in response to customer sentiment. One firm having some success in addressing these challenges is Netomi, which is using natural language processing (NLP) and machine learning to reportedly provide fully automated resolution to over half of customer requests (“auto pilot” mode), share proposed responses to customer requests with customer service agents (“copilot mode”), and provide intelligent analysis to complex customer requests.
In accounting and expense reporting, Ramp has reportedly transformed spend and expense management with its expense software and corporate card, saving $400 million for its 15,000 businesses since 2019. Ramp uses AI and machine learning to automate the matching process of a card transaction and payment receipt and help companies close their books on average eight times faster than before, generate timely financial reports, and automate bill payments.
In investment and wealth management, Hunter Labs Technologies is reportedly working to accelerate investment decisions by using AI to generate market intelligence, reducing the effort typically required by investment research analysts to aggregate information and perform their analysis.
In the financial markets, trading desks and risk managers have long used algorithms and machine learning for trading and risk management. However, it has been challenging for firms to use AI to predict market movements because AI training requires the use of historical data and a consistent set of signals, whereas a multitude of factors affect market prices inconsistently, resulting in an undesirably high “noise-to-signal” ratio.
The other problem in using AI to predict market movements is related to the limited investors’ data sets, unlike ChatGPT, which is trained with 175 billion parameters (about 100 trillion parameters in ChatGPT-4). Jon McAuliffe, the cofounder of the machine learning quant shop Voleon Capital Management, remarked, “We don’t have unlimited amounts of data to help us run models of unlimited size.”
Further opportunities for AI have been identified in the CFA Institute Handbook of Artificial Intelligence and Big Data Applications in Investments launched in March 2023 (Release I) and April 2023 (Release II). The publications provide guided tours of the current state of AI and outline real-world, battle-tested solutions and a suite of case studies from leading industry organizations, such as Man Group, Ping An, and Nvidia, on topics including machine learning and big data, chatbots, knowledge graphs, AI Infrastructure, and ESG analysis.
CFA Institute’s message to investment professionals is clear: a combination of AI and human intelligence is the winning formula for successful financial institutions in the future. As investment professionals, it is crucial to understand how AI and big data applications can best integrate into the investment process and operations.
AI Innovation Is Expected to Accelerate Money Management Research, Analysis, and Decision-Making
The arrival of ChatGPT from OpenAI resulted in a flurry of announcements from both large technology firms and financial services firms. Most notable was Bloomberg’s announcement of BloombergGPT, a 50 billion parameter internal large language model (using competitive technology to ChatGPT) that Bloomberg had trained using both publicly available general data sets and their proprietary financial data accumulated over decades.
“Thanks to the collection of financial documents Bloomberg has curated over four decades, we were able to carefully create a large and clean, domain-specific data set to train a LLM that is best suited for financial use cases,” explained Gideon Mann, Bloomberg’s head of Machine Learning Product and Research.
Bloomberg plans to integrate BloombergGPT into the Bloomberg terminal. One practical example is that users will be able to type in a few words and directly pull up what they want in Excel, skipping the more complex coding language BQL. According to Bloomberg, BloombergGPT improves on existing financial NLP tasks, such as sentiment analysis, named entity recognition, news classification, and question answering, among others.
By using generative AI such as ChatGPT or BloombergGPT, market analysts will be able to get better summaries of financial reports and better monitoring of news stories and earnings transcripts. Financial analysts can input the most recent data into ChatGPT and get reports on trends and forecasts quicker and more accurately. As a result, analysts can focus on evaluating insights and making investment decisions instead of spending hours researching information. For financial advisors and planners, internal marketing notes can be quickly summarized, and customer reports with targeted content can be generated more efficiently.
At my firm, Lumen Global Investments, for our digital investment solution we train data to identify the market’s risk appetite and allow users to construct their portfolios to align with their underlying risk profile. In the future, we anticipate ChatGPT can enhance how optimal portfolios can be generated based on a few prompts (natural language queries).
A Note of Caution Interacting with AI
“In the past, AI was hidden within layers of back-end infrastructure for streamlining logistics or automating content moderation. Now, applications like ChatGPT and the image-generator Midjourney have placed the technology directly into the hands of individuals and small businesses who are using the tools to see if they can automate laborious tasks or speed up creative processes. Some are driven by the thrill of being able to do things not previously possible; others by an existential push to master the nascent technology so they don’t fall behind.”
ChatGPT is impressive and will revamp ways of conducting business. However, it has its limitations that we must consider:
- No access to recent information: ChatGPT was trained with data prior to September 2021 and does not have access to recent data, so its understanding of current topics is lacking.
- Biased language: ChatGPT was trained with vast quantities of data from the internet, and without strong content moderation, it may generate responses that reflect biases toward certain types of languages and may generate inappropriate or offensive content. OpenAI wrote in a company blog last month that they are going to “provide clearer instructions to reviewers about potential pitfalls and challenges tied to bias as well as controversial figures and themes.”
- Inconsistent responses: While ChatGPT has access to vast amounts of text data, it does not have a unified understanding of the world it can use to guide its responses, which can become inconsistent with previous statements or with the context of the conversation. To address this, researchers are training models on more diverse and representative data, incorporating external knowledge sources, and developing methods for detecting and correcting inconsistencies in the model’s output.
- Lack of morals: Being a machine learning model that has been trained on a large data set of text without any explicit moral guidance or principles, ChatGPT’s answers can appear offensive, insensitive, or even harmful. While it lacks morals, ChatGPT readily dispenses moral advice with contradictory advice on the same moral issues, which ties back to the response inconsistency problem identified in 3.
- Misinformation: ChatGPT may sometimes generate responses that are misleading, factually incorrect, or promote misinformation. The New York Times reported when researchers asked ChatGPT to write responses based on false and misleading ideas, the bot complied about 80 percent of the time. This misinformation problem may result from the quality of the training data, which includes both accurate and inaccurate information; ChatGPT’s lack of context of the conversation or the world; or confirmation bias where the model may be biased toward generating responses that confirm the user’s preexisting beliefs or opinions.
- Leaks of confidential information. Many organizations have banned the use of ChatGPT to avoid employees inadvertently sharing confidential information outside of their organization. OpenAI has been transparent that conversations including user input, file uploads, and user feedback may all be reviewed by OpenAI’s trainers working to improve ChatGPT, and ChatGPT may also retain and use the conversations for further training. Employees have reportedly mistakenly shared highly sensitive and proprietary information such as transcripts from internal new product meetings for ChatGPT to generate meeting notes and also shared source code for new software for ChatGPT to correct errors in the software.
Therefore, ChatGPT should be used with caution and discretion and best utilized with a human expert. Ultimately, ChatGPT is a tool to assist humans in focusing on what they do best.
Concluding Thoughts—What’s Next for Investment Managers?
AI combined with human intelligence and domain expertise is the key to successful financial institutions in the future. As technology advances at an unbelievable pace, it is vital for financial managers to understand, adopt, and implement these tools effectively while considering their limitations.
There are three areas for transformation investment firms can focus on:
- Develop an AI Strategy: Understand how AI can enhance a company’s existing business model and develop an AI strategy to address current business pain points. This includes considering whether to “build” versus “buy” AI capabilities.
- Organize for Success: Adopt the CFA Institute framework –T-shaped teams concept–and build out three specific areas i.e. traditional investment function, technology (data science) function, and innovation function, to bridge the communication between investment and data science professionals for AI adoption to succeed.
- Harness the Power of Data and Develop a “Clean Data” Strategy: Successful implementation of AI is predicated on a large quantity of good quality data. That means, data must be collected, processed, transformed, and analyzed to harness its power. For those investment firms that want to build their own AI capability, they should evaluate their data and determine whether it is fit for purpose: for example, can firms leverage their internal data to better meet their clients’ goals and manage their portfolio risks?
Paul Tudor Jones told his team in 2015: “No man is better than a machine, and no machine is better than a man with a machine.” With the arrival of ChatGPT and the proliferation of use cases in AI, time has proven him right.
“We must dare to think ‘unthinkable’ thoughts. We must learn to explore all the options and possibilities that confront us in a complex and rapidly changing world.”
~ J. William Fulbright
Marianne O, CFA, Lumen Global Investments, San Francisco
With special thanks to Paul Chew, former PwC Partner, Digital Transformation Specialist, and Global Angel of 100WF for his kind comments and contribution to this article.