Five Minutes with Sov.ai's Dr. Derek Snow

INTERVIEW

Derek Snow is an expert in AI, which he sometimes calls “Automation Intelligence.”

He has quite the pedigree:

• Ph.D. in finance from the University of Auckland 

• Visiting doctoral scholar at NYU's School of Engineering and Cambridge Judge Business School 

• Alumnus of:  

  • The Alan Turing Institute

  • The UK's Institute for Artificial Intelligence

  • The Oxford-Man Institute of Quantitative Finance at Oxford University 

• Currently teaching machine learning in finance at NYU

• Author of MLQuant newsletter 

Derek is also the founder of Sov.ai, a quant research platform for fundamental investors. Sov.ai provides investors with datasets and tools that can spot trends and patterns in financial markets, without requiring them to be expert programmers or data scientists themselves.

Derek, who’s a native of New Zealand, splits his time between NYC and Boston.

Key Takeaways

1. AI vs. Automation: Much of what is called “AI” in business is actually automation.

2. Slow AI Adoption in Finance: Despite the excitement around AI, its adoption in finance remains slow, especially at larger institutions due to data quality and regulatory challenges.

3. Promising AI Applications: AI shows promise in areas like risk management, algorithmic trading, credit risk assessment, and natural language processing for financial analysis.

4. AI Risks: Over-reliance on AI, regulatory scrutiny, and the potential for bias are key risks that financial professionals should remain cautious about.

5. Sov.ai’s Role: Sov.ai aims to make financial data more accessible for AI applications, offering pre-processed datasets that are easier for firms to use.

6. Future AI Trends: Agent-based language models and multimodal AI could significantly enhance financial analysis.
   
   This interview has been edited for clarity and length.

Why use "automation" instead of "AI" in business contexts?

When people talk about AI in business, I think we should really be using the word 'automation' most of the time. It's not that AI isn't real or impressive, but in many corporate contexts, what they're really talking about is automating processes that have been manual up until now. Companies could have and should have been doing this kind of automation for years already. It's not new. But they haven't, for various reasons. Now, suddenly, because of all the buzz around 'AI', some of these companies are finally getting serious about automation. It's like the AI hype has given them permission or motivation to do what they should have been doing all along. 

So when we discuss AI in business, I think it's often more accurate and helpful to frame it in terms of automation. This helps cut through some of the hype and focuses on the practical impacts and changes that are actually happening in most cases.

How is AI adoption progressing in the finance industry?

AI adoption in finance is still relatively slow, especially at larger institutions. Many firms are using AI for basic tasks like code generation, but more sophisticated implementations are limited. There's a lot of marketing hype around AI usage, but the reality often falls short.

Some prop trading firms and quantitative hedge funds are more advanced in their AI adoption, particularly those with a mandate for applying quantitative methods. However, even at firms actively hiring for AI-related roles, new employees often find limited opportunities to experiment or implement cutting-edge techniques.

The median age of portfolio managers, especially in the US, tends to be around 55, which may contribute to slower adoption. Additionally, regulatory requirements can limit the use of advanced AI models in certain areas like options pricing or credit decision-making.

The rapid pace of AI development makes it hard to decide when and how to invest in new technologies. Data quality and availability issues, especially for training sophisticated AI models, pose additional challenges. Compliance hurdles in adopting new technologies or data sources and the need for specialized talent to develop and implement AI solutions further complicate matters.

Many firms are in a "deer in the headlights" moment, recognizing AI's potential but unsure how to proceed effectively.

What are some promising applications of AI in finance?

One area that's particularly interesting is in risk management and fraud detection. We can use machine learning models to analyze vast amounts of transaction data and identify patterns that might indicate fraudulent activity or increased risk. This isn't entirely new, but the models are becoming more sophisticated and can process data much faster than traditional methods.

Another promising application is in algorithmic trading and portfolio management. We're seeing more firms use reinforcement learning models to optimize trading strategies and manage portfolios. The interesting thing here is that these models can potentially handle multiple aspects of the process - from risk management to execution - in a more integrated way than traditional approaches.

In credit risk assessment, machine learning models are showing promise in predicting defaults more accurately than traditional models. This could have significant implications for lending practices, potentially making credit more accessible to some while reducing risk for lenders.

There's also interesting work being done in natural language processing for financial analysis. We can use these models to quickly analyze vast amounts of textual data - from financial reports to news articles to social media posts - and extract relevant information for investment decisions. However, it's crucial to approach this carefully, as these models can also amplify biases or misinterpret information if they’re not properly designed and monitored.

In the back office, we're seeing AI being used to automate many routine tasks, from data entry to reconciliation. This isn't as flashy as some other applications, but it has the potential to significantly improve efficiency and reduce errors.

One area I find particularly interesting is using machine learning for feature engineering in financial modeling. This involves using AI to identify relevant factors or 'features' in financial data that can improve predictive models. It's a bit more technical, but it has the potential to uncover new insights that human analysts might miss.

But I'd caution against viewing AI as a magic solution. In many cases, the most effective applications of AI in finance involve augmenting human expertise rather than replacing it entirely. The firms that are likely to benefit most are those that can effectively combine AI capabilities with human judgment and domain knowledge.

Can you talk about how you’ve leveraged this technology? 

I did a study on restaurant closures as part of my PhD research in financial machine learning. It was quite an interesting project where we used machine learning to predict restaurant closures within the next year, with about 70% accuracy.

We started with a massive data dump from Yelp - it was an academic dataset they released. What caught my eye was that they had an indicator for when a restaurant closed. They determined this based on a certain number of reviewers reporting the restaurant as closed.

Once we had this target variable - whether a restaurant closed or not - we could work backwards to see what factors might predict it. The dataset was incredibly rich. We had all the reviews ever left for each restaurant, ratings, photos, menu items, price changes, and even details like whether they allow smoking or dogs.

In the end, we developed about 300 different features to feed into our model. Some of the results were quite surprising. For instance, one of the biggest factors was the gender balance of the customers. If a higher percentage of women frequented the restaurant compared to men, the restaurant was more likely to succeed.

We also incorporated external data, like using Google Maps to analyze the competitive landscape around each restaurant. We looked at the distance to competitors and developed a 'competitor score' to gauge how competitive each neighborhood was.

What I found fascinating about this study was how it demonstrated the power of machine learning when you have a rich, diverse dataset. We were able to uncover patterns and relationships that might not be immediately obvious to a human analyst.

What potential risks or downsides should finance professionals be aware of regarding AI?

First, there's the risk of over-reliance on these systems. While AI can process vast amounts of data and identify patterns quickly, it's not infallible. We've seen cases where algorithmic trading systems have caused significant market disruptions due to unforeseen circumstances or errors in their programming. Finance professionals need to maintain a level of skepticism and always be ready to intervene.

There's also a significant regulatory risk. Many financial decisions require clear explanations and transparency, which can be challenging with complex AI systems. If you can't explain how your AI arrived at a particular decision, you might run afoul of regulators.

Another concern is the potential for AI to amplify existing biases in financial data and decision-making. If we're not careful, we could end up codifying and scaling up unfair practices.

What is Sov.ai and how does it fit into the AI landscape in finance?

Sov.ai is a data company I've founded that's focused on addressing a critical need in the AI and finance space. Essentially, we're working to make high-quality financial data more accessible and affordable for firms looking to implement AI solutions.

The idea came from recognizing that many financial institutions, especially smaller and medium-sized firms, are paying exorbitant amounts for data sets that are often publicly available. The problem is that this data isn't always in a format that's immediately usable for AI applications.

What we're doing at Sov.ai is aggregating and standardizing this public data, making it more readily usable for AI and machine learning models. We're not just providing raw data, but creating what you might call 'feature stores' - pre-processed data that's ready for use in AI models.

Our target audience is primarily data heads and people who've realized they're overpaying for data sets. We're offering a more cost-effective solution, but we've found that the biggest hurdle isn't actually the price point - it's the compliance process. Getting new data sources approved in financial firms can be a months-long process.

Many institutions prefer to stick with known, albeit more expensive, data providers rather than switch to newer, more cost-effective solutions. 

What developments in AI are you watching for in the context of LLMs and finance?

First, I'm watching for advancements in agent-based language models. These could significantly enhance how AI interacts with financial systems and data.

Second, I'm keen to see progress in incorporating multimodal data into LLMs. Currently, LLMs don't really handle things like Excel documents or complex financial reports well. Once they can effectively process and analyze these types of structured financial data alongside text, we'll see a step change in their capabilities for financial analysis and decision-making.

When we combine these agent-based, multimodal LLMs with the vast amount of financial data out there, that's when we'll likely see truly transformative applications in finance. But we're not quite there yet.

ICYMI
The last 5 “Five Minutes with” Interviews

• Bain’s Richard Lichtenstein on AI Adoption in Private Equity

• Snowflake’s Jonathan Regenstein on AI building novel datasets

• Skadden’s Dan Michael on the SEC’s AI stance

• Stardog’s Matt Lucas on Hallucination-Free AI

• Celent’s Monica Summerville on AI Adoption in capital markets

Thanks for reading!

Drop me a line if you have story ideas, research, or upcoming conferences to share. [email protected]