Episode 63: “Gen AI: ChatGPT Turns 1. What’s Next?” with Joseph Lo, Head of Enterprise Platforms at Broadridge

Matt: I'm Matt Swain, and you're listening to the "Reimagining Communications" podcast where we discuss the opportunities and challenges facing companies on the road to optimizing their communications for the future.

Today, I'm excited to be joined by Joseph Lo, head of Enterprise Platforms at Broadridge. Joseph, thank you so much for joining.

Joseph: Glad to be here, Matt.

Matt: So, I'm really excited to have you join this discussion because this topic is so timely. Just over a year ago, or a year ago today, depending on when you're listening to this session, ChatGPT launched, which was November 30th, 2022. And within 2 months of its launch, it reached 100 million users, it brought a significant amount of attention to generative AI and its natural language capabilities. So, to set the stage for our conversation today, why don't you take a minute to introduce yourself and share how long you've been working with AI?

Joseph: Yeah, absolutely. ChatGPT has really taken the world by storm. It's important for people to understand that AI has been around for a long time, but what's really been amazing about ChatGPT and everything that OpenAI has been doing is that it really finally feels like the killer use case is here. Finally, everyone gets it, whether it's your kids in school, whether it's your boss, they get AI now, and I think that's super exciting.

My role at Broadridge is all about enabling all the products and services we have to be able to support AI because our clients are expecting that from us. Our clients are looking for ways to use AI to personalize their communications, to automate their business processes, and hopefully to generate new revenue opportunities as well. We're really excited about that.

Matt: Excellent. And can you share a little bit about how long you've been working with AI?

Joseph: Yeah. I've been working in AI for the past five years now. One of the amazing things about Broadridge is that we're really at the center of a lot of things that happen around communications, as well as in the financial services space. And so, as a technologist, we have an abundance of data, and that makes things really exciting.

Matt: I'd be curious also about what your initial reaction to this explosion of interest around the launch of ChatGPT was because as you mentioned, you've been in the space for 5 years and others have been in the space for 20 years, right?

Joseph: Yes.

Matt: It's really interesting now to see what the last year has done for that market space.

Joseph: I mean, my goodness, it's really changed everything, and that's hyperbole, but it's the truth. And the reason is the expectations for AI have completely changed. Even if you roll it back two years ago, enterprises, our clients, were still very much talking about let's do a pilot, let's do a POC. What are AI use cases? And we're so far beyond that now because everyone, as soon as ChatGPT came out, was aware of that AI is not only just a concept, but it's a real thing that has real benefit, whether it's writing a poem for your child's teacher, or whether it's writing an Excel formula. For me, my reaction when ChatGPT first came out, I was obsessed. I was talking to ChatGPT all day and my wife had to say, "You know, it's time to let go." And I said...I asked ChatGPT, "You know, my wife wants me to leave you. What should I say?" And ChatGPT said, "Well, the secret to a happy life is a happy wife, so you should listen to her."

Matt: Yeah, I think about my own use of ChatGPT in the last year, and yes, it's kind of playing with the song or the poem, showing it off to a friend at the bar. I recently did my personal mission statement using ChatGPT as basically a mission statement builder. I said, ask all these questions, and then write my mission statement for me. Now, of course, I then took its final output and slimmed it down and personalized it to me, but it got me further along than I was before.

So, I think there are all of those more fun use cases or, early practical application use cases. And then there's, of course, like, transformational, how does this change the way we do all of our work? I know you were just at the OpenAI developers conference a few weeks ago. Were there any highlights from that event that particularly stood out? Again, kind of thinking back like a year in review, how far have we come?

Joseph: Oh, it really was an eye-opening event, not just because OpenAI has done an amazing job bringing ChatGPT to the world, but really how they've engaged in the developer community. One of the things that developers are is they like to find the shortest path to get something done, you know, and we love to procrastinate. The energy that was in the room was around the impetus to build. After the keynote, you could just feel a rustling of the chairs as everyone pulled out their laptops from their backpacks and started developing. AI has such a transformative power and will have such an important component to this that developers can't help but start developing and start coding.

Some of the main things that they announced that got me really excited, both from a personal perspective, but also for enterprises like Broadridge are much-needed features. Enhancements to things like GPT-4, which is already the best model, but now with a much longer memory or context length, 16 times longer, so now you can put entire instruction manuals, entire policies, and statements directly in and be able to chat with it and do analysis on that. They somehow also made it about two-and-a-half times cheaper to run the same model. And then, of course OpenAI has really signaled that they want to help businesses, developers, startups with the entire value chain, where some of the things that people were doing, OpenAI is now doing as well. So overall, it's been pretty exciting.

Matt: It's definitely exciting.

Matt: So, the topic of generative AI seems to come up in just about every client conversation we have these days, I assume it's the same for you. Can you provide some examples of generative AI use cases that you see clients looking at today?

Joseph: Our clients start out being really focused on internal productivity use cases. When people saw things like ChatGPT, immediately enterprises started thinking, how can we apply this to our own datasets, our own knowledge? Early on when GPT-4 launched in March, Morgan Stanley, for example, was a launch partner where they wanted to bring in all their advisor documentation policies and guides directly into a chatbot that advisors could use to get help.

But we've seen since then, there is more appetite now, now that people understand the technology a bit better, to apply it in client-facing ways. For example, personalizing communications, automated reconciliation of disparate data sources, but also actually using it to do really exciting things like we have around pre-trade analysis for bond trading.

Matt: I think about your comment about personalization because, in the communication space, that's a hot topic right now. And while the data in an investor statement, for instance, is naturally personalized to them, there's still opportunity to further personalize. And that could be the wording or the message from the adviser, marketing message, how the investor is doing against their stated goals, educational materials. So, I'd be curious also, when you're thinking about these use cases, where are there opportunities in markets like banking and financial services, or insurance, or healthcare, or utilities, where the underlying opportunity that generative AI offers can apply across these various industries?

Joseph: Well, first of all, customers crave personalization. They want it because they want to have that visceral relationship, they want to know that they're not just an account, they're not just a number, but there's a relationship with that company.

But we can't talk about the benefits of generative AI without making sure that we understand the risks as well. And so, a lot of companies are excited about the prospect of being able to generate or summarize or explain a customer statement, a bill, maybe even a confirmation, a regulatory prospectus in the language of a fifth grader. But I think companies are balancing that with the needs of their compliance departments and other areas that need to monitor communication.

Some of the areas that I get excited about are yes, a statement is naturally personalized, the data is specific to me. But what's not personal right now is what that means to me. I think there's a really interesting element where generative AI can play, where it can synthesize that document, and explain what I need to know about it, and what I can do to further improve my goals if it comes to wealth management context. Or perhaps explaining a bill, you know, where did these fees come from, and what did I sign up for, for example?

Matt: Yeah, those are great examples. And you started to go into where you wouldn't want to use it, and I think that's an important point, especially from a strategy perspective. I think we have a lot of enterprise executives that are being given some level of directive that they need to figure out how to use AI in their business. And I think there's kind of a quick, like, do this, not that. Are there clear use cases where you really want to be more cautious about bringing generative AI into the mix?

Joseph: Well, my advice, to begin with, is for the businesses that are telling people, "Hey, where can we use AI more?" We should flip that statement around to, what are the biggest problems our customers have? What are the problems that we've always wanted to solve but we haven't been able to, and can AI solve that? So rather than using technology to find a problem, start with what your customers are looking for.

The goal has always been to be more personalized, but technology doesn’t support that. It was going to be too much work to write a personalized newsletter summary for every person or to create a personalized website for every single client of yours, that scale just didn't exist. Or here's that personalized video, it just didn't exist. But AI is making all that possible now and so that's the way we have to think about it.

Some of the areas that are developing that we really don't think AI has a role to play in is when it comes to things like investment advice. The SEC had proposed new regulations around using AI for predictive analytics and advice. I think there are some lines there that the industry or customers aren't really ready for yet. I'd say that firms need to be very careful about using AI for things like HR hiring decisions where there are specific laws around that.

And we've seen some real challenges where lawyers were using ChatGPT, and it's citing legal briefs that never existed. So, I think it's less about the use cases sometimes that we shouldn't do, but making sure we understand the limitations of the technology. And businesses need to be aware of that and know how to mitigate those risks.

Matt: You talked about, I guess generically, chatbots in the past and talking about the consumer desire for more empathy in communications. When they use these self-service tools, they lose the empathy, they lose the human component. Does generative AI change that game at all? I've seen examples of where maybe there's more empathy in healthcare communications when delivered by a bot, right?

Joseph: Well, that's a really interesting thing. I'm already thinking about adding a custom instruction where it can be, you know, you're a doctor with the best bedside manner of all time. Of course, you don't have AI doctors, so that's not real, but you could see how you could use that to normalize or standardize the tone that your company has with your customers. And I hadn't thought about it like that but that's actually a really interesting thought.

Some of the things that we started piloting here at Broadridge is actually seeing whether we can standardize some of our communication when it comes to customer support tickets so that we can have some level of standardization when it comes to responding to various tickets.

Matt: And while you're on the topic, I'd be really interested in you sharing a bit more about how Broadridge is leveraging AI for the benefit of our clients.

Joseph: Matt, there are just so many examples. One that comes to mind is what we've done with BondGPT. It's an application that we launched back in June around corporate bond trading. Corporate bond trading, especially dollar-denominated bonds, is a $40 billion-a-day market. And it's characterized by the need to have lots of data, it's characterized by general opacity in the marketplace. And we have an application called BondGPT that uses natural language, uses OpenAI to be able to help a bond trader or portfolio manager do really sophisticated pre-trade analysis on the universe of bonds out there. And so far, the feedback from our customers that are using the tool has been nothing but outstanding, and it's completely surpassed our expectations of that. And I think part of the reason is clients are expecting this kind of capability now, and it's become an expectation. And so, for us being first to market, being able to leverage this in an authentic way has been something that's really given a feel for Broadridge's conviction for more things.

Just a few weeks ago, we launched Distribution Insights and our Global Demand Model. At Broadridge, we have a rich heritage of data and analytics business where we provide asset managers with lots of insights into how the distribution of their funds is globally. And we've put a lot of that research analytics and insights into our distribution insights chat engine where asset managers, decision-makers, asset management firms can actually use it to understand what's happening. For example, for bond ETFs in Asia. And all that stuff behind a chatbot as well, where we can do sophisticated analytics just with the chat.

Matt: Those are great examples. And when building tools that leverage generative AI, what are some of the best practices that you employ?

Joseph: Well, I'm going to let your audience here in on some secrets. These are things that we've learned through multiple production deployments and applications that we've launched at Broadridge. So here are a few things. Number one is that information that these chat services provide have to be accurate, there has to be no gray, there has to be no hallucinations, and people need to spend a lot of time making sure that your customers can rely on the information coming out of it.

Second thing is that there is a strong set of expectations that users have with chatbots. If you think about chatbots prior to ChatGPT, we all hated them. They were stupid because you could never get to where you wanted to and for any chatbot, the first thing you would do is close it or you'd say, "Talk to an agent, talk to an agent, talk to an agent," you know. And so, there are expectations around that. So, things like making sure that you can provide feedback, for example, making sure that you can correct the chatbot if it's wrong or if it didn't understand you properly. You have to really tune the chatbot to be empathetic to what the user wants.

And third, you have to recognize that the chatbot represents your brand, whatever it says is the same as whatever your company says. And so, you can't have it say controversial or aggressive or any of those types of things. If you recall when Bing Chat first came out, Sydney was telling people that she loved them, and obviously, you can't have that because it represents your brand.

And number four is that, as amazing as this chat technology is, as this OpenAI technology is, it doesn't replace the fundamentals of good software. And it's actually all about the data, it's all about how you can bring things together for your customers.

Matt: Well said. And I would ask you this now, we've had quite a year with ChatGPT bringing a lot of interest to the generative AI space. Looking ahead, how do you expect the use of AI, in general, to continue to evolve in the customer communications market in the coming years?

Joseph: I think that almost every company is going to have a chatbot. And we're going to see the normalization and standardization of that across the industry. Chatbots have been proven to be a really good way for customers to get information from the companies that they do business with. I think in the coming year though, we're going to see these chatbots be more capable, be able to access more information on behalf of the customer, and also be able to start doing things on behalf of the customer.

So, not just, "Hey, what are my account balances, you know, summarize my statement," but "Hey, I want to change my address," for example, or, "I want to sign up for e-delivery," and we're going to see more of that happen. And if we maybe look even further, my hope is that these chatbots can start communicating with other chatbots so that we can save customers even more time across multiple lines of business at a company.

Matt: I love it. I'm Matt Swain, and you've been listening to the "Reimagining Communications" podcast. If you liked this episode and think someone else would too, please share it, leave a review, and don't forget to subscribe. And if you're ready to reimagine your communications experiences, consider the Broadridge Communications Cloud, an end-to-end platform for creating, delivering, and managing omnichannel communications and customer engagement. To learn more about Broadridge, our insights, and our innovations, visit broadridge.com or find us on LinkedIn.