Two Use-Cases of AI in Wealth Management – with Nirupam Sarkar of Fidelity Investments

The advent of big data and the promise of AI signal a revolution in efficiency for wealth managers. Wealth management is “all about the numbers,” and AI technology can uncover patterns and crunch numbers better, faster and more accurately than any human. Yet to look at AI in wealth management as one big calculator and data repository is to miss the mark.

AI use cases in wealth management are several, supporting advisory and non-advisory functions. Wealth managers, including financial advisors (FAs), use AI and machine learning to manage investment portfolios, ensure compliance management, lead generation, application processing, document management, and more.

Yet significant obstacles remain. Poor integration with legacy systems, lack of personalized solutions, outdated document management/processing workflows, insufficient predictive models, and other issues continue to hamper the wealth management sector. If the current state of AI in wealth management is proximal to that cited in a recent McKinsey report [PDF], one questions the feasibility of these sentiments. 

In a thoughtful discourse with Emerj CEO Daniel Faggella, Dr. Nirupam Sarkar, VP of Data Science at Fidelity Investments, takes us “under the hood” to provide valuable, actionable insight for leaders on the future of decision augmentation at the upper echelons of wealth management.

Our 26-minute also conversation explores how wealth managers can optimally leverage AI to drive value, focusing on two use cases:

  • Leveraging text summarization for discovering salient data: Analyzing and summarizing text with machine learning and natural language processing to help investors automate the research process, gather critical points from the larger text, and focus on essential insights in their presentations.
  • Categorizing mutual funds to filter investment presentation: Automating mutual fund discovery and classification with machine learning and data analytics, given a set of defined criteria.

Listen to the full episode below:

Guest: Nirupam Sarkar, VP of Data Science, Fidelity Investments

Expertise: Computer Vision, AI, Machine Learning, Image Processing

Brief Recognition: Before becoming Fidelity’s VP of Data Science, Nirupam was the Director of Data Science at the British business intelligence firm Informa. He has held several data-science-centric positions at Proctor and Gamble and Lexis Nexis. Dr. Sarkar holds a Ph.D. in Image Processing from the Indian Statistical Institute.

Leveraging Text Summarization for Discovering Salient Data

Text summarization (“summarization”) is precisely that — a technique that produces a summary of a longer text. According to Nirupam, this natural language processing-based method has a critical use case in wealth management: research.

Investors spend a great deal of time researching the markets, looking for stock tips, etcetera. A bevy of information from news outlets, specialized and mainstream, caters to investors and wealth managers. The challenge is the sheer amount of data that an investor must consume or — more accurately — sift through. 

Investors and wealth managers spend a disproportionate amount of time researching the markets by visiting numerous websites and scrolling through a bevy of mostly irrelevant information. They hope here to “find the gist of it,” says Nirupam, a time-consuming method of discovery.

Text summarization AI can assist investors in their research by collecting large amounts of data and extracting the most relevant details depending on specific interests. Only the salient points of information are then presented to the wealth manager. Nirupam also says these functions can be performed “on the fly.” We presume that Nirupam is referring to the “set it and forget it” configuration of this particular technology.

Nirupam sets the stage for “salient, point-based” summarization, using semiconductor investing as an illustration: “I’m interested in only semiconductors, nothing else. So I’ll be looking for AMD, Intel, NVIDIA, these [kinds] of [companies]. I probably won’t be interested in, say, Tesla. Maybe very little Apple, but I’ll be more interested in NVIDIA, AMD, or Intel.” 

The result is streamlined output data for the user that text summarization “will figure out what the [stock] tickers are talking about. It will figure out, out of 150 [tickers], three or four that the advisor may want to be involved in, letting us know only a few salient points.”

The business value of summarization AI in wealth management is implied here: “If you have to go through all [of information sources] at a 30-minute rate for each, it’s really time-consuming to read and summarize.”

Nirupam also discusses a number of points regarding the limitations of summarization AI.

The innate probabilistic nature of AI means that this technology works best when humans are kept in the loop. To connect the aggregate data and information together “is still a daunting task, [and] there is no good way.” However, Nirupam adds, “It can help alleviate some of the pain of doing it all manually.” 

In other words, summarization AI should not be viewed as a holistic solution for investor research; instead, the technology should be seen as a valuable tool to delegate specific research tasks.

Using Mutual Fund Categorization to Narrow Down Investments

Nirupam then discusses the use case of classifying mutual funds. He uses a conjectural example of an investment portfolio with a percentage split between equity and bond funds. In this example, the investor or wealth manager wants to drill down to potential investments based on a set of criteria. 

In this case, the manager is looking for “socially responsible” – as defined by industry standards – assets from which to invest potentially. “So there should not be any stock from alcohol, tobacco, cannabis, and gambling. So, a similar thing could be added in the keyword search ‘green energy fund,’ or something like that.”‘

Make no mistake, this is a business problem. For wealth managers, this means spending an exorbitant amount of time classifying mutual funds instead of prospecting, consulting with clients, or doing some other value-adding work. 

“It takes nearly 90 minutes for a human to classify a new fund, [with] all the criteria a client might ask,” states Nirupam.

For example, a simple keyword search can potentially result in tens of thousands of investment possibilities. Nirupam then discusses why classification is necessary in the case of a resultant dataset of 80,000 to 100,00 funds: “to pick those funds which exactly match those criteria? You have to somehow classify the funds depending on the underlying assets, this holding, and of course, like market cap and other things.”  

One can deduce that the time spend therein is excessive and costly. Nirupam suggests instead entering your search criteria into an AI platform and having it search for and gather the assets which fit a set of standards. 

The end result is a clean list of acceptable assets from which to choose, which saves time and money and may lead to better customer satisfaction and quick turnaround times.

Nirupam then elaborates on the business impact, stating that AI can reduce the time spent on fund classification by 30 times, from 90 to three minutes — a 3000% improvement.

He also discusses the potential labor savings, using a company he consults as an illustration. Nirupam says that at one firm, twenty people are working around the clock to update mutual fund data as newer funds are constantly being created. “They can solve all of the problems with only two people,” he asserts.

Similar Posts