Industry
Capital Markets
Unlocking Exponential Future Growth in Capital Market Firms
Highlights
- Wealth managers have to analyze massive amounts of internal and external data to construct portfolios that align with their investors’ objectives. This data is fragmented and comes from disparate sources, making investment research extremely demanding.
- While wealth management firms have adopted traditional ML and AI tools and techniques, they frequently fall short in addressing the requirements of the wealth manager.
- Wealth management firms must deploy an army of AI agents to enhance the pace and quality of investment research, which in turn will help unlock alpha for their customers.
On this page
On this pageinpage
The wealth manager’s conundrum
Wealth management firms operate in an ever-changing environment characterized by increasing market complexity, heightened customer expectations, and intense regulatory scrutiny.
At the same time, investors constantly demand additional alpha and expect hyper-personalized financial advice, holistic life planning, and proactive risk mitigation.
At the core, a wealth manager’s value proposition is heavily dependent on the ability to gather and analyze massive data from various sources. Wealth managers have to swim through datasets of the following types:
- Investor (client) specific data: Existing financial positions, tax considerations, behavior profiles, investment goals and objectives.
- Structured financial market data: Technical and fundamental data on all valid asset classes from reliable sources such as Bloomberg, S&P Global (S&P) or Yahoo Finance (YFinance).
- Macroeconomic and geopolitical news and developments.
- ESG and regulatory related data.
- Social media commentary, market sentiment.
Understanding relationships and establishing correlations among the various parameters of data that the wealth manager has access to, form the foundation of investment decisions and asset class allocation. However, doing this at scale for multiple clients with varied preferences and goals is extremely demanding.
Though access to a variety of data has improved considerably over the past few years, the challenge of bringing together everything relevant into a single location in the desired format still persists. Fragmented data spread across service providers, custodians and market feeds is one such example. Gathering all the data in one place, normalizing it, cleaning it and making it consumable is error-prone, if done manually. Though data from certain sources is available in real-time, combining it with legacy data residing on the wealth manager’s system adds to risks of data leakage and misinterpretation which ultimately lead to a lower probability of achieving alpha.
Traditional machine learning (ML) and AI have made significant inroads into wealth management, financial advisory, and portfolio development in the past decade. However, they may not be adequate to overcome the challenges facing the wealth industry. That will require wealth management firms to adopt an ecosystem of AI agents with the ability to greatly enhance productivity and empower wealth managers to unlock alpha for their customers.
The need for AI
The rise of large language models (LLMs) has pivoted the wealth management industry to a different level.
An LLM, like GPT, can understand natural language input and generate human-like responses. If a wealth manager were to ask a general-purpose GPT such as ChatGPT to build a portfolio with low standard deviation (risk) and high returns, the model’s response will be based on its training data, which would likely be information scraped from the internet. But the wealth manager expects a response that considers the investor’s risk appetite, size and composition of the existing portfolio, goals and desires, and any other restrictions that the investor wants to include. In short, the wealth manager expects a response that is not constrained by knowledge and reasoning limitations.
AI spanning conventional ML models and data ingestion and wrangling models can absorb data, clean it, identify relationships, and generate context that forms the basis for investment research. ML models can also synthesize unrelated datasets, extract actionable insights, and identify opportunities in real-time. By leveraging AI in the form of generative or descriptive models, a wealth manager can design investment strategies that accurately reflect client preferences and market conditions.
Why AI agents?
Constructing a portfolio begins with understanding the investor’s perspective.
An investor’s perspective is multi-dimensional spanning risk appetite, current wealth, future investment goals, access to financial instruments, and other personal quirks. The subsequent step is to identify the asset classes that might suit the investor and then break them into sub-classes. For example, if US-based growth equities are identified as a suitable asset class for a certain investor, the wealth manager begins identifying investment options that are worthy of inclusion in the investor’s portfolio. The process is effort-intensive and begins with extracting insights from a multitude of documents and ends with generating a tear sheet that is a crisp summary of the business fundamentals of the shortlisted investment options. The tear sheet becomes a primary input to investment decision-making.
While LLMs have made life easier for wealth managers, their output or responses are based on the data used to train the LLM, which means that the responses are constrained by knowledge and reasoning limitations. In contrast, an AI agent has certain techniques at its disposal that can look up more recent information and build complex portfolios for a wealth manager’s customers. Simply put, an AI agent is a piece of code that is powered by an AI system—in this case an LLM, which can then deduce the workflow sequence involved in investment decision-making in complex scenarios with minimal human intervention.
The fundamental advantage of AI agents is that as the wealth manager continues to use the agentic framework to service investors’ needs, the underlying agents continuously learn and evolve from user inputs and expectations. What distinguishes an AI agent is its ability to retain past interactions in-memory and deduce future actions and output with a hyper-personalized touch. For example, for an investor on the verge of retirement, the AI agent will consider the current net worth, risk appetite, recent updates to the portfolio, macroeconomic scenario, and age and goals and build a tailored portfolio. A conventional LLM, on the other hand, would be limited by the dataset it has been trained on and may ignore crucial elements of the investor’s ecosystem such as recent portfolio updates and macroeconomic scenario necessary to build an optimal portfolio.
Toward a decision-friendly AI framework
Clearly, a wealth manager has to analyze massive amounts of data from a plethora of sources before making an investment decision.
Given quick investment decisions are critical to taking advantage of alpha opportunities as they arise, firms must empower wealth managers through an army of AI agents to perform the various activities that underpin accurate investment decisions. To achieve this objective, firms must incorporate the following components to build a decision-friendly tear sheet:
- Prompt with the objective to generate a tear sheet that would drive investment decisions.
- Simple logic to break complex prompt instructions into smaller steps.
- Multi-agent framework comprising agents to carry out various tasks:
- Financial reporting agent to evaluate the financial statements of the company, identify the ratios that are relevant to investment analysis, and highlight anomalies.
- Equity research agent to examine the business ecosystem in which the company operates, perform competition analysis, and review and highlight risk factors that cannot be ignored.
- Stock market agent to analyze open, high, low, close (OHLC) data for the stock over a period, compare with the volume traded and compute the relevant moving average.
- Tools to gather information from external sources, extract information, and perform calculations:
- Web search tools to wade through the latest news and articles on the business domain in which the company operates.
- Programs for mathematical operations such as computing exponential moving average (EMA) and momentum indicators like relative strength indicator (RSI).
- RAG tool to retrieve information from regulatory filings and standardized financial and business data, the tool can be trained to understand business and financial contexts.
- Pre-trained GPT (LLM) for conventional market analysis, content retrieval, and response generation.
- Memory systems to store learnings and retrieve when necessary, learn user preferences, and structure responses that are customized to user’s desires.
- Communication system to generate responses in the form of a tear sheet that details performance and market analysis of the company that is of interest.
Figure 1: AI agent ecosystem to aid wealth managers in investment decisions
So, how do AI agents work in response to a user prompt? When a user enters a prompt (see Figure 1), the AI agent:
- Combines all the parameters such as datasets comprising pretrained LLM data, third party databases, external application programming interfaces (APIs), if any, with internal data to create a holistic picture of the comprehensive research performed.
- Selects the appropriate tool, refines the prompt, and sends the query forward.
- Receives and processes the response; updates the current state and memory.
- Refines and improves with continuous follow-ups.
Each database or API that the agent interacts with has a unique identifier associated with it. As the AI agent services more prompts, the internal memory of the agent updates the unique identifier associated with the external databases and APIs. The next time a prompt of a similar nature comes in, the response is much more precise and personalized since the internal memory contains information about the user’s preferences.
It is undeniable that AI agents have a lot to offer wealth managers as they combine the best of both LLMs and rigid programming languages, supplementing the response generation capabilities of LLMs with AI backed tools that enhance the quality of the output. But transparency and explainability challenges remain as LLMs are designed to be black boxes. Furthermore, the infrastructure costs associated with the external APIs which act as tools can be huge. In a world where resources are limited, prudent judgement on powering and augmenting LLMs with additional data or processes is paramount. The scale of benefits will become clear only when the adoption of AI agents matures. However, their advantages over conventional AI cannot be ignored. In our view, AI agents have the potential to unlock alpha, and wealth firms must experiment with the technology before embarking on large scale adoption.
Case-in-point
A globally renowned wealth management firm was spending significant time and effort on creating tear sheets from all the information they had to process, making it expensive and cumbersome.
The firm adopted a multi-agent framework (an ecosystem of agents that communicate with each other internally) with the capability to invoke tools from outside the ecosystem to create a tear sheet containing decision-friendly insights. With this tool, the firm realized the following key benefits:
- 25-30% reduction in time taken to generate the tear sheet and perform analysis involving tasks like gathering data, structuring the draft, running screening checks, and populating charts.
- 55-60% increase in the number of tear sheets produced per analyst per month.
- 30% reduction in the turnaround time from request for tear sheet creation to completing the analysis and distributing the research to clients.
- 18-20% reduction in operational costs due to lower FTE hours spent on manual data collection, formatting, compliance checks, and basic analysis.
In a nutshell
AI and AI agents are here to assist humans, in this case, wealth managers.
Agents have the potential to increase the sphere of influence of a wealth manager by reducing time spends on mundane tasks. As AI agents penetrate further into the wealth manager ecosystem, the productivity of wealth managers and analysts will witness a manifold increase accompanied by a significant rise in the speed of execution and the consistency with which the analysis is produced.
As AI systems become more deeply embedded into the research and advisory workflows, wealth managers will be able to operate with real-time insights resulting in faster decision cycles and hyper-personalized client communication. In the long term, competitive advantage in the wealth management space will be defined less by access to information and more by how effectively a wealth manager is able to orchestrate an AI agent to convert information into actionable investment decisions. Given this backdrop, wealth management firms would do well to take quick action in incorporating AI agents into their operations. While this will come with its own challenges, the rewards will far outweigh the effort.