With the growing adoption of AI across all areas in financial services, the realm of investment planning is no exception and emerges as one of the key focal points for companies planning to build innovative solutions.

Interestingly, over the past year, while large language models have taken center stage, the power of traditional AI has taken a back seat. Although these LLMs seem invaluable for automating certain administrative tasks such as report generation, the broader application and maturity of various other deep learning models within the financial investment sector are still on the horizon.

Current State

The integration of AI into investment planning still has significant room for growth. As it stands, consultations with financial advisors often include preselected potential investment strategies, along with evidence of associated historical returns.

There are two primary issues with this approach. Firstly, these strategies are preset, such that investors cannot form their unique optimal portfolio mixes based on specific preferences. For instance, if an investor wanted portfolio options that would safeguard against a simultaneous rise in the unemployment rate and oil prices, they wouldn’t have the necessary insight to determine the ideal asset mix for their portfolio.

Secondly, the duration of the presented historical returns is often capped at a 10-year period, overlooking some significant economic downturns. Therefore, it’s imperative to have sophisticated investment planning tools capable of dynamically constructing asset portfolios, enabling more personalized, robust and informed investment planning.

Demand For AI

Traditional methods of forecasting portfolio return through linear extrapolation over decades have become obsolete, and well-informed investors are readily embracing AI-powered tools to determine their portfolio strategy. According to a CFP Board survey released in August 2023, a staggering 31% of investors would be comfortable implementing AI’s financial advice without verifying it first, and another 21% would be comfortable to a certain degree after verification.

This shows that more than half of the investors are bullish on the use of AI to drive their investment decisions. Such sophisticated investors demand more than just assurance about long-term returns. They are interested in establishing the right hedges for potential economic downturns and wish to understand the possible outcomes when inflation resurfaces, or rates are drastically lowered.

AI can assist by identifying patterns and suggesting assets negatively correlated with the broader market. Although such correlations may change over time, especially during downturns, it is within an AI model’s capability to identify these shifts.

For financial advisors, these tools can assist in modeling various scenarios and conducting exhaustive analyses for their clients. For instance, an AI-powered investment planning tool can predict the impact of a simultaneous rise in the unemployment rate and commodity prices on portfolio returns.

There is ample evidence to suggest that machine learning methods often outperform standard statistical models in accurately forecasting portfolio returns, primarily due to machine learning’s ability to recognize complex relationships between variables. That being said, a number of recent events have demonstrated the substantial operational risks associated with using AI in forecasting tools for investment planning.

Algorithmic Bias

One of the key risks associated with AI-based investment planning tools is algorithmic bias. If the tool was trained on biased training data that led to significantly positive results in the past, it may present misleading projections to customers, thereby causing them to make investment decisions not aligned with their risk profiles.

The financial world has seen many cycles, including the dot-com bubble of 2001, the financial crisis of 2008, the oil crisis of 2015 and, most recently, the Covid-19 crisis of 2020. In these cycles, we witnessed asset values take a nosedive. During 2020 and 2021, ultra-low policy rates were observed, and inflation dominated the discourse in 2022 and the first half of 2023.

To accurately capture the varying correlation between different asset types, we need to account for several economic downturns with data spanning over several decades. Not surprisingly, if a model is only fed with data from a short horizon, it is likely to be biased and reflect spurious correlations existing only in those years.

Data Protection

Another key consideration would be the protection of customer data to ensure the use of data in AI systems is consistent with permitted rights and to maintain the confidentiality of personal information. AI tools, during their calibration, can process sensitive financial information, posing a risk of data breaches.

The right way to ensure data is adequately protected is to collect and use only the minimum amount of data necessary for the intended purpose and bolster the process further with data quality checks, anonymization or encryption.

Transparency

Lastly, developers of investment planning AI tools continue to grapple with these tools’ lack of transparency, making it challenging for financial advisors to help customers understand the basis for a certain portfolio strategy.

Transparency and trust become especially crucial elements when considering using AI-powered systems in the context of customer financial decision-making processes. Given that financial decisions often involve substantial stakes and carry the potential risk of monetary loss for customers, these complexities become significantly heightened.

One possible way to address transparency concerns is by building AI tools in a manner that can provide clear explanations for their decisions. In addition, such AI tools should be thoroughly documented, including key activities associated with data such as collection, preprocessing and training.

A Final Thought

Investment planning indeed presents an excellent use case for AI, offering tremendous potential for extensive adoption. However, it is imperative to apply appropriate governance controls to instill trust in the development and use of these AI-powered tools.

LeackStat 2024