We are excited to be bringing you podcasts featuring timely investment insights from our team of investment professionals. You can also find our podcasts on Amazon Music, Apple Podcasts, and Spotify as The Royce Exchange,

This episode has been edited for clarity.

Frank Gannon: Hello and welcome everyone. This is Francis Gannon, Co-Chief Investment Officer at Royce Investment Partners. Thanks for joining us. AI and how it's changing the world of investing is a topic we have spent an enormous amount of time on over the past several years here at Royce and its effects on the small cap asset class we find to be particularly interesting as well as its effects on the companies we invest in. The speed of change has been impressive as to have been the benefits. From our perspective, artificial intelligence is transforming the investment research process.

And as you will see, it is empowering analysts to identify trends and risks that might go unnoticed, unnoticed more effectively, freeing them, from our perspective at least, to spend more time on some higher valued insights.

Joining me in this conversation today are two members of our investment team that are on the forefront of how AI is becoming a critical part of our research capability and to be quite honest, a competitive edge. Joining me are Assistant Portfolio Manager Tim Hipskind and Senior Analyst Zach Weiss.

Guys, I think this is a fascinating and a huge topic. But Tim, perhaps you could start with a bit of a conversation around the evolution of AI and the theme here at Royce.

Tim Hipskind: The evolution has really been amazing to watch. So, what started out in I believe November 2022 of when ChatGPT kind of came into the zeitgeist and Zach and I started using it heavily just for a small cap analyst. It was very helpful at the time to just kind of have simple questions answered at whatever level of complexity you needed. We own a spintronic sensor company and to have some of the nuances of that explained to you was very helpful. That was kind of the beginning, just basic chat bot, ask a question, get an answer, before you might spend time on Google or call a sell side analyst. This allowed you to just get up to speed and learn things a lot quicker.

And it just seems like every three months since then, there's just been dramatic evolutions in what you've been able to do. The next big step was the reasoning models, which just allowed for a whole new level of, for lack of a better term, thinking by the robots. The depth of their output just increased dramatically. That opened up a lot of new use cases from a research standpoint. The other thing that came along at the same time as that reasoning model was the sourcing or the ability to search the Internet and provide sources back to you of what it was saying. Obviously, a big thing we'll get into is the hallucination risk. And early on there's a lot of questions around, hey, it'll say there's four Rs in strawberry or whatever, right?

The sourcing really changed a lot of that because now you can click through and say, OK, maybe it gave me a source from 2024 and I'm asking a question now and I don't want to put as much weight on that, but that was a big evolution. The buzzword, for the past year probably has been agents, and what will agents do? It's giving these AI models tools, and the ability to use these tools on a repetitive basis. It's really just been a funny couple of years where Zach and I sit next to each other, and we'll just hear the other, you know, make a noise somewhere between a laugh and a chortle when we see this new capability come out. And I want to hear Zach make that noise. I have to go over to his desk and ask him, “What have you learned? What are they capable of now?”

Frank: So, you keep hearing this concept of your digital brain. How has your digital brain changed?

Zach Weiss: Thanks for having us Frank. I think just overall the encouragement of the Royce management since the advent of AI came about encouraging us to use these tools and get smart and super grateful to work right next to a guy like Tim for the last 3 1/2 years and be able to spitball back and forth on how the stuff has been working. For me, it's kind of just like having a super smart sidekick that knows everything about everything on my side at all times. I have to be a little careful because in a lot of ways. AI can give me answers that I want to hear depending on how I ask the questions. So, a big part for me has been developing the way that I prompt the system to get certain answers.

I guess the digital brain question is, how it's impacted my part of the research process here at Royce, the breadth and depth that it allows me to go into on companies and industries. For example, we have management teams come into the office or meeting at conferences and historically the questions are, “So what do you guys do? How do you make money? Who are your customers?” And now I can ask all of those introductory background questions to AI and bypass the process of maybe speaking with industry experts, just really getting down to the most important factors that might move a stock over the long term and being equipped with really smart questions to ask management that I wouldn't have had before. It's kind of a can of worms type of question because there's really so much to how I've been using it here with respect to customized news alerts that I've set up or helping prepare for internal team meetings and preparing for questions that the portfolio managers might be asking of me on companies, helping generate new ideas via more of a qualitative screening process that want tools that weren't available to me before.

Tim: And if I could just add on this question of digital brain. We started out where you had an incredibly smart friend sitting next to you, and it was like Swiss cheese. There were some kind of holes in the knowledge, but you had this really smart person with you all the time that you could kind of bounce ideas off of.

Then we got to a point where we had to put in our own internal notes. That's when I really started having this thought of digital brain because I'm dumping all of my notes from meetings with management teams over years, and I can then talk back with that. Whereas before I would have gone to our research management system, gone through notes with the management teams and had to read those myself, whereas now I can really have a conversation with the notes and kind of with a past version of myself and see what questions was I asking two years ago? How have those been addressed? How have those changed? I think the evolution is, now I'm still pulling everything out of the system. I'm asking questions still to my notes. In the future the systems will know you so well that it'll say, “Hey, we've read the 8-Ks for these 200 companies, and based on everything we know you look for, these five companies are really what you should start looking into because they've said XYZ in their earnings call and 8-K.” I think as we go forward, it's going to be less pulling stuff out of the system and more of the system pushing things to you. So that's my view on the evolution of the digital brain.

Frank: But spend a second about your daily workflow and how that has changed, because I think that's kind of key to understanding how this is helping you. You've talked about running the smart person next to you, but how exactly are you doing that?

Zach: The day-to-day varies depending on whether we're at a conference or earnings or just kind of the general research pipeline that we're working on here day-to-day in the office. In each one of those aspects, there've been ways that we're kind of leveraging the tools. So, for example, if I'm going to a conference and there are 300 companies on the attendee list. Before I would have to manually go through each company attending the conference and say which one should I book a meeting with? Where now I can feed the list through AI and say, which of these companies are worth meeting based on our investment strategy? And then prepare for those meetings.

So, before maybe I only had the capacity for six meetings in the day, now I can do 12 and ask smarter questions based on the ability for AI to help me prepare for those. I think previously, if one of the portfolio managers prescribed me a stock to look into for a for a Strategy, it would be kind of maybe waiting a few days to before having booked a call with a sell side analyst or reading through the 10-K, and now I can kind of have a prescribed list of questions that I know that they would want to have answered and effectively just kind of speed up the process and allow me to go deeper and look at more things.

Earnings season always gets pretty busy around here and historically, we still do manually listen to each earnings call and take notes. But while I'm maybe listening and taking notes on a company that we own, I can have AI analyze the way that I'm prescribing it. All of the competitors and constituencies that might impact the company that we own in a portfolio to give us maybe a more holistic view.

Tim: One example I'll stick with is the earnings season motion. Obviously, it's one of our busiest times and this last earnings season I really tried to commit to see what can I throw at AI and see what sticks in terms of having the process evolve or work better.

To Zach's point, earnings calls are a huge one. So, before I would be typing in real time, trying to get the key points and to circulate those to the team and trying to analyze that in real time. Whereas now there are real time services that are transcribing the calls and then I can have that summarized in the way exactly that I like, and then send out to others on the team. That's really been beneficial. The other thing is just for companies we don't own in the portfolio, but earnings calls—if you take out the legal, all the pleasantries—if you filter all that down, you can cut out 30% of the call probably. I've just consumed probably 30% more earnings calls this earnings season because I'm able to summarize them in exactly the fashion I want. So, I'm getting to more calls maybe in large-cap or peers that otherwise would just fall through the cracks, so that's been a huge benefit.

Zach mentioned custom prompting: building out prompts specific to individual companies has been something that I've been working on this quarter. So, you can imagine the 10-Q comes out each quarter; you read it each quarter and obviously when you've owned a company for a few years, you can get through that pretty quickly. But now what I'm doing is, as I'm reading that 10-Q, the blackline version of that, I'm talking into my headset, which is going to ChatGPT saying, “Hey, footnote 3 says this, that's important; footnote 7 say this, that's important; footnote 8 says this,” and I then tell subsequently ChatGPT to summarize this and make instructions for reading that 10-Q. To Zach's point earlier, there's new ideas of thinking of ways you can try and use it. Companies you've owned in the past that you now have really good notes for reviewing their SEC filings, and you want to get up to speed quickly. It could just be, you know, something that's useful in that respect.

Frank: How has it changed your conversations with management teams?

Tim: Before, if I wanted to know the pricing of a specific commodity that's relevant to a company, a lot of the times, especially in small-cap companies, there's not a universal index I can go check that price at. So, you could do some channel checks. You could try to find other ways to find sources of information that are relevant, but it would often be challenging. With the reasoning models I mentioned earlier, really Chat GPT's version, the model O3, the ability to search the entire Internet and pull out data points that you would have had to spend weeks finding before has really been revolutionary.

For example, the pricing of lime, there is a PPI index for it, but it's generally kind of an opaque price, but it is used in a lot of municipal water departments, so there are public bids. For the first time I realized, oh, I can just have ChatGPT go pull a bunch of public bids for lime pricing, and I can even see the specific companies, the bidding, the pricing and the dates. You can think of a bunch of examples like that where the ability to just find things that before would have taken an incredible amount of time, or you just would never have found has increased.

I think we're kind of shocking managements with the level of depth you can go to these days.

Zach: I would echo Tim. This is probably one of the more fun aspects of where we see AI come up in our jobs, at least for me. For example, we have a company that a couple of quarters ago reported in their press release—they kind of buried in there that they won a massive one-time order in a different industry than they've historically participated in. And because of the privacy, I guess with respect to the customer, they were hesitant about how much they talked about the details on the call. And so, I fed the press release into AI and said, based on what you know about the company, who do you think is the customer? Tell me more about the industry applications and the future growth opportunities. I had a meeting with the CEO the day and I said, “OK, so you're now getting into pulse power applications. And he said, wait a second, how do you know that?” Because I did my homework in a much faster way than I would have been able to do before, it maybe gave me more insights on how the core competency of the company is set up to maybe address some new market opportunities, to come to them with the level of depth that I'm not able to get anywhere else from any analysts. To the question of how it impacts the day-to-day, I'd say the day-to-day—really the core of what we do as research analysts and portfolio managers—is predominantly the same, but the tools, the breadth and depth that we can do it with and the speed is the biggest game changer.

Frank: Let's turn to the small-cap asset class. It's the forgotten asset class of late. Everybody's focusing on the upper end from a market cap perspective. It's, I think, one of the last inefficient asset classes out there. We know there's very little analyst coverage on many of the companies that we spend our time on. How is AI helping you round the asset class and specifically what we focus our lives on around here?

Tim: That's been a huge benefit. To your point, Frank, some of the portfolios here have close to 50% of the companies that have no analyst coverage at all, and then you can imagine maybe they have one or two analysts covering some of the others. So, the ability to build sort of a digital analyst that can do a very lightweight version of coverage on these companies is something that's really interesting. And so, one of the things that I did initially through an application called Cursor, which is really just an amazing thing in its own right, but it allows someone like me who has very limited coding experience to build applications and actually get output from them. One of the first things I did was just build an 8-K summarizer. I took a test population of 10 companies that we theoretically could own that have no analyst coverage. And can I build something that summarizes the 8-K whenever they put one out. That was kind of an eye-opening experience because yes, you can, and it can do a really good job.

To build out a full-scale digital analyst is a little beyond the scope of my capabilities, given the other responsibilities I have. But that's one of the really great things about being a subsidiary of Franklin Templeton is, we have a lot of resources going into this, and this sort of digital analyst that can cover some of these small-cap companies or maybe bubble up ideas. The AI has read 200 filings from companies that aren't covered and says, “Tim, we know you like this specific type of company. These three companies look like they're at an inflection point based on these specific quotes out of the filing.” You can imagine it can really help speed up the research for uncovered companies.

One of the other things that AI has been really helpful with in terms of focusing on the small-cap asset class, and particularly at Royce, is, given that we have such a history of focusing on this asset class we've owned probably thousands of companies in the small-cap space. But obviously we don't all own them all at once. But we do have this repository of research information. And so with the security that we now have through some of these AI systems, you can take old research notes or just old emails of a PM saying, “Hey, I owned this in 2008, and here are my notes,” and you can feed that into AI and say, “Update the thought process here, go through the recent past decade of filings, and give me an update as to what the view was then and what the view is now.” So just given the nature of what we do focusing on small-cap, it's hugely beneficial.

Zach: Yeah, I might add to what Tim said. I think particularly with companies where there might be a lack of sell side coverage or just getting a general understanding: I mentioned this company receiving a large order from a customer. They had two sell side analysts covering the stock. One of them actually happened to be in our office the following week, and I asked him what he thought about this big order and he responded, “What order?” So to me that was a good example of the inefficiency that still exists in small-cap land.

Having tools that we have available now to sort of get ahead with respect to any type of insights that might lead to different investment outcomes, I think is really beneficial. There are companies that are in harder to understand industries, and if there's no sell side coverage, it can be hard to get smart on what exactly these companies do.

For example, we were looking at a company that makes ion implanters. They're a part of the semiconductor manufacturing process to make the chips more effective in certain instances. And I didn't understand what it did. So, I asked ChatGPT. I said, “Explain an ion implanter to a five-year-old,” and it said, “OK, imagine you're making a gigantic cookie. The ion implanter is a special sprinkle machine that makes specialized sprinkles depending on how you want the cookie to taste. And I said, “OK, that I can understand.”

It's amazing how many inefficiencies still exist in the small cap space. You can have companies where there's meaningful changes going on that the market is not necessarily catching on to yet, where with the breadth and depth that these tools allow us to get to, might help flag more opportunities than we would have been able to find before.

Frank: At the same time, though, there's enormous misconceptions about AI, right? And I think there's a fear at times about AI and how to implement it and how we should use it. Have you guys thought about that or how that's affecting your day-to-day?

Tim: It's definitely something we think about.

Frank: Let's be honest, AI is wrong at times.

Tim: Yeah, absolutely. And so, the biggest thing that has helped with that is the sourcing of information when you get the responses. You can oftentimes click the link to see where ChatGPT came up with this information? So, that's very helpful. The other thing that I've done is within the background of the AI tools, when I talk to them, I have it tag every statement it says to me with high, medium, or low confidence, which I've found helpful in understanding when is it freestyling a little bit versus when does it have something directly it can point to. So good prompting helps out a lot with that, but there is no getting around it. It's been described as a jagged edge of intelligence, so you have to be careful. It's obviously not something we're relying on wholeheartedly, but to Zach's point, it kind of gives you ammunition to get smart on things before you might go talk to a management team or then go do your diligence in, you know, all the ways that we do that. So, there are definitely risks, but the benefits are just so tremendous. You have to kind of find ways to work with the risks.

Zach: Some other misconceptions that come to mind for me. One in particular is people often talk about the job security that AI or the job risk that AI might pose and that we're all going to be out of work since AI has come into the forefront, just because I can do a lot more stuff, it definitely makes the job more fun. I was getting a haircut yesterday and the barber was kind of defending her job because it was protected from AI. I still think there's a lot of fear out there with respect to job security, and I just continue to explore and play around with the tools to the extent that they can be useful.

With respect to the hallucination risk, it definitely is a real one with AI and there are hallucinations that existed before AI. Tim and I could be in a meeting with the CEO and could walk out of the meeting having heard two different things. That's the importance of having a team dynamic. There have been situations where I've generated something that has helped influence my thoughts on a company, and I've sent it to our team. They'll point out a mistake that's in there and it's a dose of humility. The AI made a hallucination and I'm really glad the team member was able to point out the mistake. Fortunately, those types of checks and balances exist to help mitigate the downside with respect to the mistakes that AI can make. But I also think back to whether it's a sell side analyst who was wrong on a particular trend or a company was wrong. Management teams can be very optimistic with respect to business outlooks. Before AI came about, I think that's just part of the nature of the business is mistakes. Mistakes are going to be made and I think just the extent that we can have checks and balances to mitigate that is important.

Frank: I'm fascinated by the speed at which it's occurring, I mean go back two years or three years ago to where we are today and what AI is able to do. It's fascinating and you can only imagine what it's going to be three years from now, let alone three months from now.

Tim, I mean you speak a lot about this, you know, the speed at which this is changing. We hear a lot about agents and things like that—how is that going to be changing the investment world, do you think?

Tim: Yes, it's a great question. In terms of the actual workflow, what seems like is happening more and more is that part of your job at least is managing what kind of information you give to AI and what pieces you're outsourcing to AI. We actually own a company where they had a new CEO come in who said, we're basically mandating people have to think about where does it make sense to offload rote tasks to AI and where does it make sense to have the high value things being done by their talented journalists and other team members. So, it's something you have to think about. I heard an analogy, if anyone is a poker fan, there are these poker players who've got ten monitors up. And sometimes I feel like that when I've got all these different windows open, and I'm launching some of these queries that now can take up to half an hour research on “What's the history of the volatility of the steel pricing at this specific company?” Or all these random questions you can have.

You hear a lot about managing agents and managing AI, and I think that's going to be a huge revolution as well. For example, if you pass on a company as an investment because you think estimates are too high right now, maybe you have a company that does have a good amount of sell side coverage, and the estimates are too high. So, you move on and you go about the rest of your job. Well, you can imagine that in the future you've got an agent that specifically monitors that or they monitor why you passed on investments. They're perpetually revisiting that and come to you. It could come to you and say, “Before you passed on this investment because you thought the sell side was 20% too high in their estimates, they've all come down now. Otherwise, you didn't have a problem with the investment. Maybe do you want to reconsider that?” I think the future might be a lot more pushing of things to you based on your organization, where it understands our sales process, it understands our research process, and it’s fully integrated.

One of the other things that I've come to appreciate, and this actually came from an AI conference I went to, is in meeting a lot of the people that are doing start-ups or working in this area. You're shaking a lot of hands and you're saying, “What's your background? How did you get into AI?” I heard a lot of, “I was in business process mining.” I didn't really know what that was. I had five of these conversations where five different people explained to me that it's where you basically map out every single thing that happens at a company. And the end result looks like they basically described a plate of spaghetti where there are all these things going in different ways.

For example, person X has this task they have to do and there are 100 people X. 70% of the people do it this way, and it takes 10 minutes to resolve, 30% of the people do it this way and it takes half an hour to resolve. And so, this has been this long field of study where they would take this business process mining, take it to the executives and say, hey, here's how we think you can alter your workflow. In terms of the future of work, we're seeing all these people who have naturally found themselves in the AI world. You hear a lot now about workflows and about how you have to really think about your job, what workflows do you break that up into and where does it make sense after you've mined those processes to tack on an LLM or AI as an addition to your workflow or where can you fully outsource things like summarization?

That's one of the things I've really been thinking about a lot in terms of the workflows and how you optimize them and how do you edit those for better outcomes.

Zach: I have one thing I could add to the future one other additional tool that hasn't really come to the forefront, yet I think it's kind of like a portfolio aid that can give PMs real time insights on weightings or bit buys or sells instead of me running down to a PMs office and saying hey there's insider buying at XYZ company to have a real time alert for them to incorporate into their decision making process. More recently, we have access to a more secured version of AI where we can be more comfortable with our proprietary data that we've amassed. It seems like there's a lot of low hanging fruit to maybe improve investment decisions or at least serve as another pair of eyes that that didn't exist before.

Frank: I appreciate your time today, both of you, Zach and Tim. I think it’s a fascinating topic, one that has become critical in our investment process, and that really gives us a competitive advantage especially in the small-cap asset class, so thank you. 

Zach: Thanks for having us.

Tim: Thanks.

Important Disclosure Information