AI driven wealth Management
The wealth management industry is having profit and fee pressures due to secular trends in the market that lead investors to move away from actively managed funds to index funds and other passive investments. Additionally, Fintech companies such as Wealth Simple are taking away business by offering easy to buy products with simple investment objectives for the lower end of the investor population. What can Banks, Brokerage Houses, and Wealth Management firms do in the context of these trends and competitive pressures? AI and Machine Learning can provide a way for these firms to move to more sophisticated financial planning advice with personalized, risk balanced portfolios. In this article, we cover how wealth management firms are and can leverage these techniques to differentiate themselves and to get back some of the revenue they may have lost.
The traditional wealth management industry consisting of banks, brokers and financial advisors who work with banks as custodians, and firms that support retail wealth management are facing several significant challenges. First, investor customers are leaving institutions at ever larger numbers (in some cases double the normal rate) and are churning through several different firms to find the best mix of net investment gains (investment gain minus fees). Loyalty to a bank or wealth management firm has reduced significantly forcing these firms to reduce their fees and to offer promotions to retain investors. Second, they are moving funds from actively managed funds to passively managed ETF funds to avoid the heavy fees associated with actively managed funds. Third, while “roboadvisors” have been the craze in the wealth management world, investors are finding that these roboadvisors cannot deal with tax planning, portfolio balancing, and other needs that even mass affluent customers have. This becomes an important issue as the baby boomers hit retirement age and minimum distributions force many investors to make sub-optimal decisions.
In this context, we believe that wealth management firms can improve loyalty, reduce fee attrition, and increase customer satisfaction by using AI and machine learning. Through new advances in deep learning techniques and natural language dialog approaches, wealth management firms can improve relationship management with their investor clients and help them with more frequently.
AI and Predictive analytics for deeper relationship building
As the population of investors has gotten more digital, they expect wealth management firms to anticipate their needs and to provide help for their needs. Two emerging areas are helping wealth management firms to offer deeper relationship management for investors. The first is predictive analytics, which helps wealth management firms anticipate changes in an investor’s life so that they can prospectively help the investor. Using predictive analytics, wealth management firms can predict changes in the life style of investors, predict the product to offer to increase retention and share of wallet, predict emotion and personality-based segmentations to target their marketing, and identify the reason and the timing for the reach to the investor. These prospective capabilities allow wealth management firms to identify who and when to reach out and can also give an indication of what the topic of conversation should be about and offers to make to the investor.
This, combined with the capabilities to help the investor make decisions, allows wealth management firms to increase loyalty of investors and to increase revenue.
AI and machine learning in helping decision making by investors
Two advances in AI and machine learning presents wealth management firms new capabilities. First, as was demonstrated by Google’s Alpha Go program, with the appropriate use of “search” techniques, AI programs can defeat human players when there are too many choices to be considered. Furthermore, machine learning techniques, such as deep learning, in combination with the appropriate search techniques are also demonstrating that computer programs can make decisions that are novel and not biased by human disadvantages such as rooting, aversion to loss, etc. The combination of these two advances allows wealth management firms to come up with portfolios and advice on specific actions with those portfolios by leveraging historical data.
Creating a portfolio or re-balancing a portfolio is a “search” problem where multiple objectives needs to be considered and an optimal decision needs to be made. For an investor, objectives can range from life style goals for pleasure and enjoyment, aspirations for security for self and family, control over one’s life and investment and contributions to the family and the community. Turning these high-level objectives into financial objectives, finding a portfolio of instruments that would achieve those objectives, and helping the investor make investment decisions (trades, withdrawals, deposits, incremental investments, etc.) covers a large search space in an uncertain world since the investor must make decisions for gaining benefits in the future. Furthermore, given implicit risk tolerance and behavioral decision-making approaches used by individuals, generating advice / recommendation for an individual investor needs to be personalized for that investor.
In the past, pieces parts of this puzzle were addressed with human intuition playing a very large role. Advisors would use their expertise in picking investment instruments and their knowledge of the investor to have a conversation about the type of products to consider in a portfolio and would advise the investor on when to sell / buy / hold elements of that portfolio. For mass-affluent and lower end of the high network individuals, the help provided by advisors was static and very infrequent.
AI and machine learning addresses problems with this high touch and intuition driven approach. AI algorithms can select between multiple options based on historical data and based on simulated performance of the portfolio in the future. They can consider far more options than can be done by a human advisor. Deep machine learning approaches can learn about investing styles and decision frameworks by mining past data to be able to create a portfolio that would fit with the style of the investor.
The learned decision framework, investment style, and portfolio options can then be used to have meaningful conversations in key moments with an investor. Whether that event is about the investor’s life (marriage, move, loss of a job) or trend towards dormancy prior to attrition, the advisor can use predicted events to have a “reason to call” and propose the appropriate solution using the “learned” decision framework and the portfolio options found through AI search.
Machine learning for decision framework in wealth management
Investors make decisions that can be improved with an appropriate decision framework. Human tendency to root on well-known factors can lead to concentrating positions in a few equities. Avoidance of loss behavior can lead to portfolio imbalance because a certain type of instrument has not performed well in the past for the investor. Wealth management firms have tried to address these issues by providing investors with access to simulation tools and periodic advice on portfolio balancing. However, given these simulation tools do not help create a decision framework, but provide point-in-time data, their adoption by investors has been very limited. More recent attempts at using roboadvisors to create plans that would address specific financial objectives have also not penetrated as deeply as expected because they too do not provide a framework for decision making where the investor may be interested in exploring alternatives.
However, a decision framework that allows investors to evaluate alternatives can have a significant impact in good decision making by investors. Deep machine learning can be used to create such a decision framework. The decision framework for investment needs to consider the following:
· Investment product categories to consider (e.g., equities, bonds, etc.)
· The specific options available within these categories (e.g., stock symbols, type of bond, funds, etc.)
· The position to hold in each of the options
Wealth management firms have access to past data on these decisions and performance of the investment given such decisions. Furthermore, they have access to whether the investor was satisfied or not satisfied with the performance given the decisions that s/he made through advisor notes and call center interaction data. Given this data, deep learning techniques that can work on significantly large number of variables can consider the three types of data over a period of time to create configuration of products, positions, and performance as available options that would satisfy an investor’s goal.
Selection of the appropriate configuration for recommendation then becomes a search problem that can be performed in two ways. To help make the investor decide that what s/he feels comfortable with, a similarity search to find options that are like the current options can be performed. The options along with their past performance can then be shown to the investor which will help clarify to the investor the biases the investor may have and will open the investor’s decision making.
A more aggressive option is to use search approach like AlphaGo where taking the current position of the investor in their portfolio as the starting position and exploring potential decisions the investor could make by using simulation to explore future decisions that the investor can make and the expected performance from those decisions. This “search” can result in a smaller set of decision recommendations that can be made by the investor.
In summary, there are multiple options available to wealth management firms to provide decision making assistance using machine learning techniques. The choice to be used depends on how aggressive a wealth management firm wants to be and the data available with the firm to learn patterns from its customer interactions.