image of research and ai
From the Field

Unleashing the power of AI in investment research


How could asset managers use AI to boost and refine investment research? Host Jennifer Martin is joined by Jordan Vinarub, head of our Tech Development Center, and Jay Nogueira, Director of Equity Research.

Podcast Host

Jennifer Martin Global Equity Portfolio Specialist


Jordan Vinarub Head of Tech Development Center
Jay Nogueira Director of Equity Research
View Transcript

Jennifer Martin 

Welcome back to the Angle, the podcast where curious investors can gain an information edge on today's evolving market themes. In this season of the Angle, we are focusing on the exciting world of artificial intelligence, otherwise known as AI. I’m Jennifer Martin, a global equity portfolio specialist at T. Rowe Price Associates, and I will be your host as we learn more about this intriguing and rapidly evolving area of the global economy and financial markets. 

In previous episodes in this series, we've explained why there's so much excitement surrounding AI and the potentially transformative impact that it could have on global productivity and, ultimately, on the world economy and society in general. 

In today's episode, we are going to explore the potential usages of AI for the investment management industry. Joining me for this discussion are Jordan Vinarub, who leads our Tech Development Center, and Jay Nogueira, our Director of T. Rowe Price Associates Equity research. Jordan, Jay, – it's great to have you both on today's episode. 


Jay Nogueira 

It's great to be here, Jennifer. 


Jordan Vinarub 

Yeah, thanks so much, guys. 


Jennifer Martin 

So let's start by going back to the launch of ChatGPT in November 2022. Can it really be regarded as the watershed moment as so many people think it is?  

Jay Nogueira 

Sure. I absolutely think it can. And I think it's really analogous to the development of a lot of technologies that have impacted asset management and really every industry. And what happens is you have a long phase of kind of technological development, but then you have an event and it's usually a user interface, where your average person can now interact with the technology. So if you think about PCs, the Mac in 1984 was where you had the first time you had kind of the mouse and the graphical user interface. The Internet under development for a long period of time, but it was to me, in my opinion, the Netscape browser in 1994, which really made it easy for anyone to access it. 

iPhone, I think very similar with kind of the touch screen in 2007. And so similarly, AI has been in development since the 1950s. So, you know, just like these other technologies, it's been a long road. But ChatGPT was the first user interface where kind of your average person could have a just natural language conversation and extract information. 

And just like past technology cycles, when you have that kind of really compelling user interface, it catalyzes widespread adoption and a huge investment cycle which kind of builds on itself. And so I do think it's been it's really been a watershed event. 


Jennifer Martin 

Thank you, Jay. That was very useful. So let's pin this down a little more. What are we talking about in practice? How can large language models like ChatGPT be used in our industry? Jordan, do you want to talk a little more on this? 


Jordan Vinarub 

Sure, thanks, Jennifer. I think as Jay Nogueira mentioned, the arrival of ChatGPT is like the way to bring AI to the masses or that that killer application that makes it more accessible and usable to all people, really gets shown in our own asset management industry. We're in an industry where information is key to decision-making. The proprietary nature of our business is to make effective decisions on behalf of our clients, and the use of information effectively in that is essential. 

So large language models, which is even just one subset of artificial intelligence (or even one subset of natural language processing), allows us to consume more information, derive those insights, and then help make those better decisions. Now, at T. Rowe Price we like to think of the three C's with large language models, which we call consumption, characterization, and creation. So the consumption of information, the ability to take in information, summarize it and make better sense of it. 

I think there's a statistic somewhere out there that says, in the past couple of years there's been more data generated than in all of human history, and that large amount of data is hitting our decisionmakers of the firm every single day, be it our investment analysts, directors of research, portfolio managers, even folks in sales and marketing. Their ability to consume information and make sense of it, is gotten to the point where we need to augment them and help them. Consumption really is a key part of what we're doing. 

Characterization is the ability to derive insights from that information. The ability to pull the right information out of that is essential. And then finally, Creation. This would be where a system like ChatGPT can create emails, notes, even poems and songs. 

And any of those three areas of the three C's could help business people to improve their productivity and ultimately help. 


Jay Nogueira 

Yeah, I think Jordan summarized it really well. You know, and I’ve been in the industry 20 years, the amount of information available to investors has kind of far surpassed any individual’s ability to read all of it, a long time ago. And so that consumption element is critical. And so not only is it a productivity tool, which is incredibly helpful, but it makes sure that we're getting all of the key elements that are out there, separating the signal from 
a tremendous amount of noise. So making sure we're getting those critical investment insights is really important.  

And it's really further leveraging just all the information that we've already had available to us across the board. That characterization component is also important, which is again deriving those insights. So, when you think about some of the data source access to being able to derive more value from sources is going to be really important. On the creation side, that is going to enable, again, investors to communicate a lot more seamlessly, kind of create analyses and again, dramatically improve productivity. So big impact on all three within our group. 


Jordan Vinarub 

Jay, I think you bring up a really great point too. The amount of information inundating investment professionals. I know portfolio managers receive automated emails with summaries of relevant content every single day, such that they can't even read all the summaries that they're getting, let alone know which areas of research that they want to focus on a daily basis. 

And having the ability to augment that process is essential for them to get a handle on where the signal to noise is. 


Jennifer Martin 

Now my energy rises when I hear that we're going to have some real tools to leverage a lot of the research and really mining that for insights and pattern recognition in the future. So that is very exciting for industries like ours. So I guess moving on, T. Rowe Price has been working on these things for, for several years now (in fact, we quantified it as seven years, Jordan). Can you tell us a little bit more about T. Rowe Price’s AI journey so far? 


Jordan Vinarub 

Sure, Jennifer. So, the popularity of generative AI is relatively new, but we like to say AI is not new to T. Rowe Price. Since 2017, for seven plus years now, we've been on a journey to build new capabilities in data science, machine learning, predictive analytics to help our investors and other professionals harness the power of the data that we have in our business. 

And so really, we started seven years ago with the goal of building up those capabilities in-house, having our folks aligned with the business, and then ultimately trying to develop tools and solutions. We employ a strategy, we call it “intelligent augmentation”. So rather than automating the decision making, we're trying to bring the power of data and insights to the human decision maker in the existing business process that they have. 


Jordan Vinarub 

As we've been saying, human data is inundating our folks. How can we help them derive insights from that information in their daily tasks? Now, over the past seven years, we've been building, learning, exploring different areas. 

And we've even been able to build custom solutions that help share information and integrate the research process in a more broader way that ultimately helps advance the collective knowledge that we've got at the firm and harness that content that we're talking about. Over the past year, that was really where Jay and the directors of research stepped up to say, okay, How can we get a handle on generative AI? 

This seems like an existential moment for the firm - that we need to understand this, figure it out, see how we might apply it within our business rather than chasing the hype, but understand what can we glean from how that might help our business folks do their job. 

 And really, since then, we've done a lot of focus on how can we harness the power of our research content, how can we bring that to our business folks and then create a more seamless research environment or research ecosystem we're now calling ‘’ as a way to have AI powering the ecosystem of investments research. 


Jennifer Martin 

So that's a journey so far. Looking to the future, is it our expectation that AI will ultimately automate decision making or will it be used more as a tool to augment human decision making? 


Jay Nogueira 

Yeah, I think that's a really great question, Jennifer. I think it's really important to stress that we do not see decision making being automated. That is certainly not our current approach or what we'd expect in the future. I think you have to be very cautious on technologies and if you go back, there's a long history of you know, attempts at automating that that lead to very disastrous results. 

That is not at all our approach. So our approach – I think Jordan Vinarub touched on it a bit – is what we call intelligent augmentation. So if you think about the people we have who are kind of experts in their fields using technology to make them that much more effective and productive. So really just increasing the value of each of our associates dramatically. 

And you think about all of the past technologies have kind of been a slow build to that. Everything from if you think about the transition from, say, the seventies to the eighties nineties with PCs and using Excel versus literally people used to have models like pencil and paper, right? Huge productivity changes, huge capability changes. It went through the Internet in terms of our ability to access information. 

It wasn't that long ago that I was, you know, literally getting hard documents off of computers. This is just a whole another step function because of what AI can do. So it's along that same line of if you think about how much more productive each one of our associates is than they were 30 years ago, how much more effective they can be. With AI and the tools available and all the information out there, we can really just make that a step function increase. 

So that, that’ll, that will be good for every facet like the day to day of our associates is going to be better. You're going to spend less time on kind of low value-added tasks. You're going to be spending more time on that kind of higher value-added, getting, really getting insights, being out there in the field. It won't be as frustrating to kind of access what you need to. So it's really almost making each of our associates kind of a super associate with, with the intelligence augmentation. 


Jordan Vinarub 

Yeah, I think that's a great point, Jay. And the best example I've heard is from portfolio manager Dom Rizzo saying, “Hey, I want to know what's everything we knew about Cisco in 1999.” And that took him eight hours over a weekend to find information, collect it in different places, read it, synthesize it, form his own opinion, and then ultimately take some notes around that.  

And so with the use of a large language model, the ability to quickly consume and summarize and give him information, and still allow him to read the source content, just as he normally would. Think about how much more time he can be focusing his energies on investment thesis that he's developing versus spending his time collecting information. 


Jay Nogueira 

Yeah, and Jordan I'm glad you, you brought up an example because I do think the productivity improvements here you know, it's kind of a general thing we talk about. It is, it is potentially just orders of magnitude. So just back to the information gathering, you know, in many cases, you know, we've kind of taken let's say a typical use case of a portfolio manager and all of the information they're trying to gather from different sources, you can replicate that. 

 And, you know, with natural language processing in terms of getting the summaries and what matters, you know, it's like a it can be a 10x change, you know, very, very dramatic. And so it's really exciting. 


Jordan Vinarub 

Yeah. Every time you get one of those wins, like 8 hours going down to 10 minutes or even less, the more of those you have throughout the day is the more productivity and the more benefit you're getting those folks applying to the process. 


Jennifer Martin 

Really helpful. Thank you. So I guess playing devil's advocate for a moment, is it possible that with all the excitement around AI, it has become overhyped to some extent? 

Jordan Vinarub 

Yeah, I would say the hype of generative AI has certainly far surpassed the current level of capability in the technology, and we've seen that, whether it's in certain vendor products or other experiments that we've done along the way over the past year and a half, and our focus on generative AI. Large language models are known to create hallucinations, which is basically providing answers that are factually incorrect and asserting them as correct. 

And so really requires, whether it's the human user of the technology, or the engineers who are working with it, to ensure that they're not basing any solutions on an expectation of complete correctness coming from it. And every week we're seeing new research on techniques for how to interact better with large language models through what's known as prompt engineering, or other techniques like retrieval-augmented generation, or other ways to make better sense of the answers coming out of it. 

But what we're finding is that, as long as everyone has the mindset that this is an emerging technology, that it's going to be improving through time, we can get the benefits even as the answer quality is on a scale of maturity where it's currently growing. It’s how we can ultimately build the real solution. The goal isn't to have an oracle that's got all the right answers. And the one good example we like to share is so we can ask a question, it can provide an answer. It can also provide the link to the source content. 

The answer might even be wrong currently, but that whole use case of asking the question and getting the answer and getting a link and then clicking on the source content to go read it, is so much faster than they would do it today, even without asking the question. So it's important that everyone understand the technology isn’t this perfect oracle yet, but there are lots of opportunities we have to improve people's productivity to give them the benefit while being cognizant of the risks involved. 


Jay Nogueira 

Yeah, I just add, and I think it's a famous quote that is very common. Technological innovation is almost always overrated in the short term, but almost always underrated in the long term. And so, if you go back to some of the technologies I mentioned, you know, when PCs kind of came out, your average person, the notion that everyone would have one of these on their desks, and is going to transform work. 

And it took a while to get there. And in many cases it was hard to envision that when it happened. Same thing with the Internet was a great example, that there was tremendous amount of hype in the early years, took, you know, 15 years to really build out the infrastructure, but it absolutely transformed, you know, our world as we can all see in hindsight. 

This is this is no different. And I think Jordan and I one of the big things is is keeping people patient to some extent because there are you know, the possibility of it is incredibly intriguing and I do think we get there. But the current capabilities are limited in a lot of ways, and it is going to take some time. So it's a it's a really good point, Jennifer. 


Jennifer Martin 

Incorporating AI into the investment process will in some respects amount to a step into the unknown. What kind of risks come with that and how can they be mitigated? 


Jay Nogueira 

Yeah, maybe I'll start and I think I think that's a really important area that we've been very cognizant of right from the beginning. You know, putting up kind of a governance committee, you have to be very, very thoughtful around this because there really are a long list of risks. It's a new technology and there are things that we have to be very careful around. 

I think a couple of points and I think Jordan might have highlighted this. The first thing is in terms of actually using this on a day-to-day basis, you have to be cognizant of it is still very, very error prone. We've tried to put some guardrails around that in terms of always sourcing kind of the original information where it came from so you can always go back and check the source document. Educating our colleagues on this, the shortcoming. So you have to be very, very careful on kind of data quality. 

If you put anything into one of the public chat bots, it's now out there in the ether. And so we had to make sure that we have all of our data protected internally. And it's one of the key reasons of building some of these capabilities in-house, and not relying on external parties. So they can, they can augment the process. 

I think you also have to be cognizant of there is a transparency element of large language models, very complex. These foundational models, how they come up with, you know, their answers. There's an opaqueness to it and you don't really have a full understanding of it. So again, that's it's one of the reasons, Jennifer, you asked before about decision making. We're pretty far away from that, and we don't see that on the horizon at all because you really there is an opaqueness to how, you know, how the internal process kind of really works. And maybe I'll say, Jordan, I'm sure you can think about some other risk. There's a long list. 


Jordan Vinarub 

Yeah, I think you said it in a really good way. Jay. That the technology itself introduces a set of risks, and it's important that we are mitigating those on behalf of our clients. The information security is essential, right. The questions are just as important as the answers. So if you're asking questions of a public large language model like a ChatGPT, then you're giving away aspects of business. So we need to have protection around that. 

 And then in terms of other aspects of technology, you're saying like transparency and the opaqueness, that opens the door to regulatory issues in certain areas where decisions are 

being made, like in the case of HR and recruiting. If there is bias introduced by a model, the firm is potentially liable for that. 

And that's why our strategy of intelligent augmentation is essential on that. We are not abdicating decision making to the AI, but actually leveraging the AI for insights along the way. And just like a human investment analyst can offer an insight to a PM and the PM can disagree, we as decision-makers can disagree with the insights that we get from AI, and that is a key mitigant against that. 

And I think the most important thing, and I think you said it Jay, was like, we all need to approach this with eyes wide open. It's not a perfect technology. Risks are introduced into the mix by using it, and we need to make sure that we're ultimately serving our clients in the best way. 


Jennifer Martin 

So before we wrap up, can you both just finish by giving us a sense of how big a deal you think AI will be for the investment management industry? 


Jay Nogueira 

I think it's going to be a very big deal for the industry and really has the potential to have a huge impact on really global business and global GDP. I think you know one simple framework, if you think about GDP growth, whether it's in a country or the world, it's kind of a labor growth times productivity growth. 

And, you know, and with a lot of developed countries not having, you know, fast growth in their labor force, if anything, there's declines in most of the developed economies. You think about the productivity component of that and the potential with AI. I think it's profound. And so, you know, it's really easy to get very excited and optimistic about the impact long term just on global GDP and then certainly on our industry as well. I think there's going to be things, there's clear line of sight on a lot of productivity gains we can make in a lot of kind of additional insights we can derive out of the information that we already have. 

But I think there's a whole host of benefits that we can't even envision right now. And this is often the case when you get secondary and tertiary, you know, derivations of an original technology, especially with something that grows exponentially like AI. So it's very exciting. 


Jordan Vinarub 

Yeah, I share the excitement and optimism as to where this can go. I mean, we've seen we've seen examples of like with GitHub copilot, a software engineer taking a one hour task and doing it in 5 minutes. And so seeing like the real benefits of a productivity lift and how you might have that person be that much more impactful and deliver that much more. 

And this isn't just about the asset management industry, right? This is any industry that deals with information and needing to make sense of it. We need to, you know, manage against the risks as well. And I think the firms that understand how best to use it and understand how best to manage against the risks effectively will be the ones that can, can navigate this rapidly changing environment. But the optimism is certainly there where we think it can go. 


Jennifer Martin 

Jordan, Jay, thank you so much for your time today. We really appreciate your insights. 


Jay Nogueira 

It was great, it was great talking to you Jennifer. 


Jordan Vinarub 

Yeah, thanks for the dialogue. 


Jennifer Martin 

If I were to summarize today's discussion, the takeaways for me are AI is an augmented technology allowing for higher productivity and better decision making. And most importantly, you have to have a great mindset for emerging technology and appreciate that it improves over time.  

But Jay and Jordan also spoke about how T. Rowe Price’s approach is one of intelligent augmentation. Rather than automate decision making T. Rowe Price is looking to empower its decision makers with additional data and insights. Thank you for listening to The Angle. We look forward to your company on future episodes. 

You can find more information on Artificial intelligence on our website. Please rate us and subscribe wherever you get your podcasts. 



This podcast episode was recorded in April 2024. This podcast is for general information and educational purposes only, and outside the United States is intended for investment professional use only. It does not constitute a distribution, offer, invitation, recommendation, or solicitation to sell or buy any securities in any jurisdiction, or to conduct any particular investment activity. 

This podcast does not provide investment advice or recommendations, nor is it intended to serve as the primary basis for an investment decision. Prospective investors are recommended to seek independent legal, financial and tax advice before making any investment decision. 

The views contained herein are those of the speakers as of the date of the recording and are subject to change without notice. These views may differ from those of other T. Rowe Price companies and/or associates. Information is based upon sources we consider to be reliable; we do not, however, guarantee accuracy. 

Investing in technology stocks entails specific risks, including the potential for wide variations in performance and usually wide price swings, up and down. Technology companies can be affected by, among other things, intense competition, government regulation, earnings disappointments, dependency on patent protection and rapid obsolescence of products and services due to technological innovations or changing consumer preferences. 

Growth stocks are subject to the volatility inherent in common stock investing, and their share price may fluctuate more than that of income-oriented stocks. Diversification cannot assure a profit or protect against loss in a declining market. There is no guarantee that any forecasts made will come to pass. 

This podcast is copyright 2024 by T. Rowe Price. 


Subscribe to never miss an episode


This podcast episode was recorded in April 2024.

This podcast is for general information and educational purposes only, and outside the United States is intended for investment professional use only.

It does not constitute a distribution, offer, invitation, recommendation, or solicitation to sell or buy any securities in any jurisdiction, or to conduct any particular investment activity.

This podcast does not provide investment advice or recommendations, nor is it intended to serve as the primary basis for an investment decision. Prospective investors are recommended to seek independent legal, financial and tax advice before making any investment decision. 

The views contained herein are those of the speakers as of the date of the recording and are subject to change without notice. These views may differ from those of other T. Rowe Price companies and/or associates. Information is based upon sources we consider to be reliable; we do not, however, guarantee accuracy. 

Investing in technology stocks entails specific risks, including the potential for wide variations in performance and usually wide price swings, up and down. Technology companies can be affected by, among other things, intense competition, government regulation, earnings disappointments, dependency on patent protection and rapid obsolescence of products and services due to technological innovations or changing consumer preferences. 

Growth stocks are subject to the volatility inherent in common stock investing, and their share price may fluctuate more than that of income-oriented stocks. Diversification cannot assure a profit or protect against loss in a declining market.

There is no guarantee that any forecasts made will come to pass. 

This podcast is copyright 2024 by T. Rowe Price.