Demystifying Generative AI: A Primer for Asset & Wealth Management Professionals
Learn about our Generative AI corporate education services here.
ChatGPT and Generative AI (GenAI) is just the latest evolution in Artificial Intelligence (AI).
AI in finance is not new, particularly in the asset & wealth management sectors. It has already enabled firms to unlock untapped opportunities, improve operational efficiency, and enhance customer experiences. Generative AI is just the latest tool in the toolkit.
And that's just it. AI is simply a tool to achieve an objective. Just like the internal combustion engine, typewriter, or Microsoft Excel.
The challenge for leaders at asset and wealth management firms is to properly define the business objective that AI can be used to augment.
In this and in future posts, we will demystify AI and the emerging concept of Generative AI and provide asset & wealth management professionals with a comprehensive understanding of its applications, benefits, challenges, and practical considerations..
"As asset & wealth management embrace the power of AI, it is crucial for professionals to gain a deep understanding of its applications, including the emerging field of Generative AI. At Blu, we are committed to empowering professionals in these sectors with the knowledge and tools to stay ahead in this rapidly evolving landscape."
Fabrice Fischer, CEO, Blu Artificial Intelligence
Practical Definitions of AI and Generative AI
First, let us set down some non-technical definitions of AI and its components.
Artificial Intelligence is an umbrella term for a set of technologies that train computers to perform tasks that normally require human intelligence. Facial recognition cameras, self-driving cars and ChatGPT are all examples of AI tools.
Machine Learning is a major subset of AI. Machine Learning gives computers the ability to learn how to perform a certain task without explicit programming. At its simplest, machine learning models are trained as follows:
A model is fed a set of training data and uses a learning algorithm to identify patterns and relationships to understand the underlying data
The trained model can then take new data and classify it, make predictions based on it, or even recommend a course of action
Machine learning models are embedded in apps that we use every day. Netflix and YouTube use machine learning to recommend videos to us, based on its understanding of our unique preferences and the preferences of similar users.
In financial services, machine learning models learn from customer data to identify unique customer segments to enable better cross and upselling. Investors are increasingly relying on ML to predict macroeconomic risks, select investments, and even identify profitable trading strategies.
Deep Learning takes machine learning to the next level by processing more data and identifying more complex patterns.These neural network models, mimicking the structure of the human nervous system, take a variety of inputs, perform intensive calculations to understand the underlying data structure. Deep learning models then use this understanding to predict investment outcomes or identify fraudulent transactions, for instance.
Below is a simplified structure of an Autoencoder, a type of deep learning neural network. In finance, it is one possible way to detect fraudulent transactions in real-time.
The input layer is fed credit card transactions. The processing layers deconstructs and simplifies the transactions into their core components. Since the model would have been trained on old transaction data, it uses this understanding to then reconstruct this data to see if it matches the original transaction. If it does not, it is potentially fraudulent.
Generative AI is a subset of Deep Learning that also uses elements of machine learning. Generative AI models train on large amounts of data and use this training to generate new content.
ChatGPT and other Large Language Models (LLMs) are GenAI tools that can produce text in almost any format including summaries of complex information, client reports, emails, etc. There are many LLM models out there, including GPT-3.5 and GPT-4 from OpenAI. ChatGPT uses both of these models. Other LLMs include PaLM 2 from Google, which is the backbone of Bard (Google's answer to ChatGPT).
Generative AI Opportunities & Challenges in Asset & Wealth Management
Generative AI, particularly LLMs, can be applied across many asset and wealth management functions including portfolio management, financial research, and client services.
To take it a step further, Generative AI tools trained on internal company data can enable analysts and relationship managers to work faster.
These enterprise-grade LLMs (i.e. ChatGPTs) can take the following forms:
Firms pay OpenAI (for example) to use its ChatGPT language model and train it on internal company data (we are already seeing this)
Firm build their own LLMs (we will see more of this in the near future)
Generative AI for Portfolio Management
Machine learning is already heavily used by investment firms for portfolio construction, risk management, and rebalancing.
This typically involves simulating numerous portfolio allocations based on desired risk levels or expected returns. These simulated portfolios are subjected to stress tests and scenario analysis to evaluate their performance under various market conditions. Furthermore, risk management platforms such as Blackrock’s Aladdin analyze portfolio risk exposure in real-time. More than two thousand investment risk factors from interest rates, exchange rates, and other investment-specific risks are monitored daily.
Integrating generative AI or ChatGPT into portfolio construction can assist portfolio managers in identifying the optimal mix of assets for a specific investor. Generative AI tools can look at internal client data to create personalized portfolios that take into account client-specific risk & return criteria.
Generative AI for Financial Research
Financial research that is assisted by generative AI allows analysts to focus on high-value analysis tasks.
From an investment research context, ChatGPT could serve as an internal search engine for analysts to retrieve answers to specific queries from both internal and external financial & news sources.
ChatGPT can also summarize large volumes of financial data into easily digestible insights. This is a huge time saver for routine analysis of financial statements, corporate disclosures, and earnings call transcripts.
By automating these routine tasks, analysts can then focus on higher-level responsibilities such as interpreting results and formulating investment strategies. Exciting developments in the field include the upcoming release of BloombergGPT, which promises to enhance financial research capabilities even further.
Finally, generative AI tools can draft personalized research tailored to specific clients based on current portfolio allocation and individual financial goals.
In addition to generative AI, there are existing AI tools designed to automate financial analysis. Vendors such as AlphaSense and Dataminr are no-code platforms for sentiment analysis & information summarization. These platforms contribute to the evolving landscape of AI-powered financial research, enabling analysts to access comprehensive and valuable insights efficiently.
Generative AI for Client Services
An internal ChatGPT can draft personalized client reports, emails, and routine client communication. What’s more, you can specify that these drafts should be in your specific writing style or even the firm’s writing style.
Morgan Stanley's wealth management arm is already using OpenAI's GPT-4 to speed up the way that staff locate and access information.
Morgan Stanley's content library contains hundreds of thousands of PDF pages of insights spanning investment strategies, market research, commentary, and analyst insights. Previously, staff had to painstakingly navigate through large amounts of content to find specific answers.
The company recently built an internal chatbot powered by GPT-4's retrieval capabilities. This GPT-4 Chatbot quickly searches company content, effectively unlocking the combined knowledge of Morgan Stanley Wealth Management and parsing the insights into a more usable and actionable format. It's like having your most knowledgeable analysts on-call 24/7.
Challenge to Adoption: Can We Trust The Output?
A significant issue with LLMs such as ChatGPT is that they tend to output incorrect information or make things up (hallucinations).
Training GPT-4 or another LLM on internal company data does reduce the probability of these issues. After all, a model trained on company databases is less likely to produce something that is unrelated to the company and its operations.
And yet, what happens when the data itself is incorrect or outdated? This would be another case of garbage-in, garbage-out, despite the fancy Generative AI wrapper.
Concerns such as these are top of mind for firms that are implementing GenAI tools. Each firm will come up with their own mitigation strategies and rely on GenAI tools to the extent that they are comfortable with.
Takeaways for Business Leaders:
AI continues to reshape asset and wealth management . Professionals in these industries must equip themselves with a deep understanding of AI's potential, including the emerging field of Generative AI. By demystifying AI, exploring real-life examples, discussing the benefits and challenges, and highlighting its impact on these sectors, asset & wealth management will be empowered to make informed decisions, embrace AI as a catalyst for success, and navigate the ever-evolving landscape
Please reach out to us at email@example.com for a confidential chat about our corporate education and strategy consulting services related to AI and Generative AI. You can also learn more here.