AI can transform our industry—from investing to operations. We explore some of the questions asset owners and managers should be considering.
Artificial Intelligence (AI) is growing at a lightning pace—seemingly as fast as ChatGPT can respond to a user’s query. By rapidly penetrating every nook and cranny of the business world, AI is helping streamline operations, enhance client experiences and augment data-driven decisions.
For the investment industry, this capability has the potential to transform our roles through more informed investment decisions, better portfolio design, enhanced risk management and modernized operations. As AI continues to reshape the future, Andrew Chin, AB’s Head of Investment Solutions and Sciences, took the time to tackle five questions for asset managers and owners.
Q: AI is a transformative technology, but it will continue to evolve over time, with new applications and uses. So, it won’t all happen overnight. What’s the low-hanging fruit in the AI space today?
Andrew Chin, AB Head of Investment Solutions and Sciences: Right now, natural language processing (NLP), a branch of AI focused on understanding and interpreting textual and spoken data, is the lowest-hanging fruit in this arena. Our industry is inundated with an ocean of incoming data that are either text-based or that can be converted to text. For example, audio and video sources can be converted into text. Many of our jobs—whether you’re analyzing individual securities, refining a tactical asset allocation or seeking to learn more about what your clients need—revolve around synthesizing this data to extract insights and make informed decisions.
NLP is perfect for these tasks, because it’s able to ingest very large quantities of incoming data and synthesize them to suggest actions.
Investment professionals gather and comb through data from a wide variety of sources, seeking to uncover insights. For example, product and customer reviews on internet sites can shed light on businesses’ emerging opportunities and challenges. But not only is it impossible for humans to read through millions of reviews, but it will also be extremely difficult to discern themes and extract actionable insights from all that data. NLP-based tools can be extremely powerful in these tasks. They can be taught to analyze data more efficiently and effectively, covering dramatically more data, making fewer computational mistakes, and never taking breaks!
Q: To borrow a movie quote, “With great power comes great responsibility.” Can you share two challenges that AI practitioners might not be giving enough attention to?
Andrew Chin: Several challenges have been discussed at length. One of these is hallucinations—instances where an AI model produces incorrect information that is presented as fact. There are also numerous security issues for firms to work through. But I’d like to focus on a couple of other challenges that are important to financial applications but might not have received as much airtime.
In my view, the biggest challenge is the low signal-to-noise ratio in finance generally. Machine learning tools are extremely effective in situations where the problems are well-defined, and there are known features and models that can be trained to achieve a high level of accuracy—the investment world is not that. Here’s an example. In forecasting stock movements, about 40% to 50% of the overall volatility is idiosyncratic, so there likely aren’t features that can explain those movements. This situation makes it very difficult for machine learning models to achieve high accuracy in prediction problems.
Another challenge needing more attention is fine-tuning. In our AI work, we’ve found that models must be refined so that they’re effective within specific domains. Whether it’s investments, distribution or operations, each domain has its own parameters, relationships and nuances. Compare client outreach and compliance: the nature of the questions and required accuracy for models are very different. For compliance tasks, mistakes and errors can incur significant monetary and reputational costs, so more time is likely needed to fine-tune models for those use cases.
Even within investments, the ways we assess individual stocks, bonds or real estate investment trusts are very different, so it’s important to refine the model for those specific types of investments, with their inherent attributes and behaviors. This fine-tuning can make the difference between an effective investment model and one that produces ineffective outputs or decision recommendations. Given that both asset owners and managers are striving to deliver better outcomes, we have to get this right.
Q: There are obviously many uses for AI in the investment realm. What about business uses beyond the investment practice itself?
Andrew Chin: I can provide a couple of examples drawing on our own experiences. For instance, we’ve deployed AI within our distribution and operations functions. In the distribution world, we’ve leveraged AI to predict industry flows as an aid in prioritizing our sales efforts. These models are far from perfect, but we’ve found in practice that they provide useful perspectives that augment the decision-making process for our sales teams.
Within operational functions, such as compliance and risk management, AI tools can screen very large external and internal documents such as offering memorandums, marketing materials or fund prospectuses—uncovering potential risks and recommending actions. Given the large swath of text-based documents across asset-management organizations, NLP can be a powerful tool that empowers employees to be more effective and efficient across diverse functions.
Q: What are some of the considerations for investment organizations as they deploy data science and AI expertise within their organizational structures?
Andrew Chin: Most importantly, organizations should let investment controversies, operational inefficiencies or questions drive their use of data science. For shorthand, we like to say that “data science should always start with a question.” Though the shiny AI, machine learning and NLP tools are powerful, they may not be appropriate for all problems—it’s important to make sure that the underlying questions or projects require sophisticated techniques. This approach would also help organizations achieve practical solutions while they’re learning how to deploy and apply the new toolkits.
Also, domain expertise is critical, because adapting the wide variety of AI tools to tackle the investment realm requires a good understanding of the business challenges as well as the technical skills. People with advanced skills in the relevant area—whether it’s investing, operations or distribution—are critical in training models to operate in specific use cases, for prompt engineering and for fine-tuning. Organizations and models can fail spectacularly in rolling out AI if they don’t identify and apply the financial and task-specific skills in each of these three critical activities.
Q: Let’s close by looking at the AI world through a longer-term lens. If you look toward the horizon, how do you see the AI/organizational relationship evolving?
Andrew Chin: Firms will teach employees to use the new tools, and that will extend beyond the in-house data scientists and investment technology teams to the entire organization. As employees harness these tools to augment their specific roles, they’ll need to know how to use them appropriately.
That means understanding how to make better and faster decisions as well as when to override the automated algorithms using human insight and reasoning. AI and machine learning tools aren’t perfect, so it’s important to fully grasp their vulnerabilities and how employees can deploy these tools more effectively. We’ll collectively learn that we can’t just trust the new tools implicitly—we need to know how they make their decisions and when we need to intervene.
Along with that trend, we also expect to see improvements in interpretability. Humans will get better at understanding how these tools make their decisions. The entire notion of interpretability will become a more developed field—looking into the black box and making it less opaque. This development curve will need to bend upward in order for AI to be accepted more broadly, so we expect our industry will remain intently focused on this issue.
The Iron Man character can serve as inspiration here. What we mean by that is successfully integrating a technologically proficient human expert in a specific field, such as financial analysis, with an AI-powered assistant that can take in a massive amount of input, analyze it and suggest actions. Just like in the movies, humans can override that recommendation based on the circumstances and their own expertise and insights. We think this “Iron Person,” with augmented intelligence providing better and faster synthesis, is tomorrow’s winning combination for AI.
