Amanda Razani: Hello, I’m Amanda Razani with Techstrong.ai, and I’m excited to be here today with Roger Burkhardt. He is the chief technology officer for Broadridge as well as the AI officer. How are you doing today?
Roger Burkhardt: Very well. It’s good to be with you, Amanda.
Amanda Razani: Nice to have you on our show. Can you speak about Broadridge and what services do you provide?
Roger Burkhardt: Yeah, I’m glad to do that. Yeah, so think of Broadridge as kind of your grownup FinTech, right? We’re a 6 billion company. We use technology to serve the financial services industry. One part of our business is around investor communications, so we provide proxy and prospectuses and online digital communication channels. So we put out about 7 billion customer communications every year. And so a very important part of the democratization of information for finance. And then the other part of our business is around providing trading middle office and back office technology on a SaaS basis. So if you’re a bank or a broker dealer and you’re looking to trade inequities or fixed income or any one of probably 40 odd asset classes, you might use our platforms for the trading for risk management in the middle office and then for clearance and settlement and reporting. And so we service both the very largest institutions worldwide and then more tier two, tier three providers as well. And just to give you an idea, we process about $9 trillion worth of securities every day, and we’re over a hundred markets. So it is a really fun place to work, a lot of innovation going on, and we’re all about providing ongoing technology driven services to our clients.
Amanda Razani: Excellent. And so to that note, you brought up technology and innovation and Broadridge recently released a new product offering. Can you share a little bit more about that?
Roger Burkhardt: I’d be glad to do that. We have a whole portfolio of AI that we’ve been working on for a number of years here. And the product I’m going to share a little bit about with you is called BondGPT. And as you might gather from the name, it’s using some of the later AI technologies that have been popularized by ChatGPT. So these are large language models and the product is aimed for traders. So we went right into one of the most demanding set of users. If you’ve dealt with traders that they like to get information immediately, they’re constructively impatient, to put it nicely, and they make big decisions day in and day out. And we have a platform that enables traders to find liquidity and trade in the corporate bond market, which is a very large market in the US and elsewhere. And this product is providing pre-trade information to traders.
So let’s say you’re a bond trader and you want to find out which bonds might meet your needs and you want to do some sort of very complex question about which bonds with this yield and this maturity in the automotive sector are available, right? And what our clients are saying to us is, “We love your trading platform. Can you add some more pre-trade analytics?” So they use other vendor capabilities. You may have heard of vendors like Refinitiv or Bloomberg. But what they said to us is, “It’d be great if when we’re in your platform we can just quickly get answers to questions.” And so we’ve done a fair amount of interacting with our clients and what they were telling us was, we want something which is very small on the screen. They have a lot of screen real estate, but it’s very precious and we want a quick answer to our questions.
And so when these large language models exploded onto the scene, we said, let’s see if we could build a pre-trade analytics service where really the user interface is just a chat box. You don’t get much more than that, just a single line item where you can put in your question and say, “Which AAA bonds with at least 3% yield are available in the automotive sector?” And what was really striking about this technology is we were able to build a product in six weeks and not just a demo, but an actual working product integrated into our trading platform. And I’ve been around technology a long time, I’ve been doing AI since at least 20 years ago. I started at the New York Stock Exchange when we used AI for market surveillance there. And I can get used to the tools getting more powerful, but the ability for a very small team of people to build a product that would satisfy a very demanding set of users like bond traders in six weeks was really quite extraordinary. I think this is a very, very important technology.
Amanda Razani: It certainly is. And it’s amazing how fast this technology is advancing. It seems like just a year ago is when we really started hearing about GPT and it has just exploded into almost every sector. It seems there’s uses for it everywhere, but it sounds like this particular tool, it opened the doors to people. They didn’t have to have as many technical skills to be able to use this service right away.
Roger Burkhardt: Yeah, I mean, you asked a good question, which is, well, what did the users say? The traders were used to going to a market data provider and they would make selections on pull downs, and they would sort, and they would find the data and they would write a number down and go look somewhere else. And what they wanted was just to be able to come up with a query, a question like I talked about with you before. And what we were able to do was to create a system using these tools that would understand what they’re looking for. So use the large language model to understand what they’re looking for. And it would then on the fly write code to go and get the data they were looking for and bring it back to them and present it in terms of tables and graphs, the kind of thing that traders looking for, which is rather different from the standard text you might get from a ChatGPT just out of the box.
And we also checked along the way that we were complying with, in our answers, with the securities industry compliance rules. So we were able to deliver a system that could answer their questions in 10, 20 seconds. And the feedback from the users were, “Look, this could take me 10 minutes or half an hour if I had to hunt and peck around the screens I have today.” So they were really, really pleased with the ability to rapidly get the information they needed, and we then made it really easy for them just to click a button and say, “Okay, now I want to trade that bond.” Which for us, we’re looking to make our clients happy, we’re also looking for them to use the system to trade. So very, very positive feedback on that.
And it was tremendous to see how rapidly we could iterate the product by working with the users. So they would come up with industry lingo that they took for granted. Like they would say, “Show me all the on the run beer bonds.” Beer bonds, that is not a sector. If you look up market data, there isn’t a thing called beer bonds. But in industry they also about beer bonds, [inaudible 00:07:26] bonds, right? And so we had to understand their lingo and we found that we could create prompts. So the large language model would understand the industry lingo, such as I’ve described on the run being a bond market term. So yeah, we were very, very pleased with the speed at which we’re able to do this.
And I think the other thing we found was that we talk to a lot of clients who are sort of nervous about getting into this technology because they have certain worries about the risks that we hear about. So will it hallucinate, will it give me bad data? Is a classic risk. And what we’re able to do is just create an architecture, a patent whereby we ensured that we always went to data we knew we could rely on. So we used the large language model in this case ChatGPT to understand the intent. But we don’t ask ChatGPT to give us bond data. We’re not interested in data scraped off the web two years ago or three years ago.
But what we can do then is once we understand what the client’s looking for, we can then go to our carefully curated data that we have gathered from traders. We bought it, we collect it from public sources, so we knew a hundred percent the data was accurate, and we could also make sure that nothing leaked out of the system. So another concern that people have is, “Well, if I use ChatGPT, will they somehow know things about my business that I don’t want them to know?” And so we were able to address those risks. And so in many cases, we’re dealing with organizations with very, very large budgets, billions and billions of dollars of IT budget for some of our clients. And they’ve been struggling to navigate their way through the risks. And so one of the big things for me was the users were happy, but also the managers were happy that, “Oh, we’ve got the risks nailed. We know the data’s good, and we know we’re not going to have any hallucinations on the output.”
Amanda Razani: Definitely. In the trading space, I’m sure that’s even more important than ever that you can trust the data and the technology you’re using. And of course the speed and the efficiency as well is important in that trading space. So how do you foresee this type of technology, and AI and GPT in general, how do you see this technology entering the enterprise in other ways?
Roger Burkhardt: Yeah, I think what we’re seeing is we’re learning by doing. So we produce product in six weeks, and then we’ve spawned a whole bunch of other products that are going at breakneck speed also. But what we’re seeing is there are some patterns that you can use the technology for. So one pattern that we’ve generalized now is a patent where a user can explain English, what they’re looking for for certain type of data retrieval and analysis, and we can use this technology to trigger that analysis and get the results back. So that kind of natural language interface to complex data is something we’re now using for regulatory data. We’re using it for trading data. So how are my orders going kind of data. We’re using it for operational data, which trades have issues, which are the most important issues? What’s the economic value of my 10 worst issues in my post-trade world? So that’s a cluster of use cases that we are basically implementing through a shared pattern.
So we don’t want each team to have to invent the wheel for themselves. So as we find a way to apply AI, in this case, generative AI, we’re encapsulating that in a very easy to use pattern in our internal platform. And that platform has a lot of controls around it. So for example, we log everything that goes on, we stop people if they’re trying to share data they shouldn’t share. So that was one cluster. I’ll give you a second example if you’ve got time for it, which is of what I call knowledge management. So we, like many organizations have very complex software systems, very complex operational processes. The financial service industry is complex. So oftentimes if you can get someone a quick answer, “How does this work? Or how could I make this happen? Or which of our clients are configured in this way so that I can share best practice with others?” That’s enormously valuable.
And so in the same way that you might do research using ChatGPT into something that you’re personally interested in. So I do a lot of work around in my own time on community solar projects. And so I do a lot of research in what’s happening in that world, and I’ll use ChatGPT, basically it’s going out to the internet and getting things. In the same way, we are now building our own capabilities where our associates, whether they are in development or onboarding or professional services or even our clients themselves can get rapid answers to complex questions about our software, about industry processes. And in that case, like in the BondGPT example, we are not looking for ChatGPT to give us the answer, right? We’re looking to go to our own proprietary data. And I talked to quite a lot of other folks in the industry who were saying, I really need to interrogate my data. And that way also I can be sure there’s no hallucinations involved.
Amanda Razani: Absolutely. So for companies that are trying to jump on board this AI and harness it for their business, what tips do you have for them as far as where do they start?
Roger Burkhardt: Yeah, so I think we do believe that this is going to be just permeate all the work that we do. So we don’t see this as a passing phase. We don’t see this as something that just affects this category of work. We think it’s going to become just part of the way we all do business. In the same way that the internet became, we all experienced those early web browsers, ’95, and here we are today. It’s like in the water, it’s in the air, we just assume it’s there. We think AI is that kind of technology. So it’s not just for a high priesthood of data scientists. And so it’s important to have that mindset and to provide access to the technology in a way that encourages adoption, but does put some guardrails in terms of managing risks around it.
But I do think that it’s important to learn by doing. I would encourage people to get into the arena and manage the risks and having some sort of central team who can help manage the risks for the whole company probably makes sense for most organizations. But I don’t think it needs to be a super big COE. I don’t think the barriers to entry are very high. I think it’s a bit of thought, you can get stuck into this and learn by doing. And as we have done, think through the patterns of use cases that will fit your business and try and find a way to encourage them to be adopted very broadly across the organization. So we’re rolling out next week our own version of GPT called BroadGPT, which allows us any associate, it could be a salesperson, a marketing person, developer, someone who works with clients and implementing systems, any one of them can just go on there and use the system and we put a wrapper around it that helps them do that in a safe way. And we think all kinds of new ideas and new projects will come out of that by just providing access in a safe way to the technology.
Amanda Razani: Wonderful. If there’s one key tip or takeaway that you would like our audience to have today, what is that?
Roger Burkhardt: I think it’s to build in a quality into your patterns from the start. So generative AI is famously good at writing things, creating things. And what I’d like to say is it’s a wonderful way of creating the first draft, but you have to think through how you turn the first draft into the final product. How are you going to ensure that the quality of product is going to meet your needs? And in our case, I gave you the example of making sure it came from our own data, but more generally, I think it’s important to build into your plans. How are I going to make sure it’s correct and over time. And correct may have many different flavors to it. There are questions around bias that are very, very important, if you’re servicing the public. So just building quality in to the initial design I think is the key tip I would share.
Amanda Razani: Well, wonderful. Thank you so much for coming on our show and sharing your insights today.
Roger Burkhardt: Amanda, thanks for having the chance to chat with you. Much enjoyed it.