Trading Tomorrow - Navigating Trends in Capital Markets

The Rise of AI Agents in Capital Markets

• Numerix • Season 4 • Episode 36

Share episode

In this thought-provoking episode, fintech innovator and AI evangelist Peter Swain joins host Jim Jockle to unpack the rise of agentic AI and its sweeping impact on capital markets.

Swain shares a pragmatic roadmap for embracing this shift, starting not with grand-scale disruption but with small, high-impact wins—like automating supplier registration and managing calendar errors. These seemingly modest changes, he explains, can drive meaningful margin and set the stage for transformative growth.

Whether you're a financial professional navigating your future, a capital markets leader crafting an AI strategy, or simply curious about the future of work, this episode delivers candid truths and practical steps to stay ahead in the era of autonomous AI.

Jim Jockle:

Welcome to Trading Tomorrow Navigating Trends in Capital Markets the podcast where we deep dive into technologies reshaping the world of capital markets. I'm your host, jim Jockle, a veteran of the finance industry with a passion for the complexities of financial technologies and market trends. In each episode, we'll explore the cutting-edge trends, tools and strategies driving today's financial landscapes and paving the way for the future. With the finance industry at a pivotal point, influenced by groundbreaking innovations, it's more crucial than ever to understand how these technological advancements interact with market dynamics. Today, we're turning our focus to artificial intelligence. As we head deeper into 2025, ai is no longer a future disruptor. It's a present force. From generative AI to agentic AI, from predictive models to intelligent automation, we're seeing unprecedented transformation across capital markets.

Jim Jockle:

But which AI trends are truly poised to take off this year, and what will separate hype from lasting impact? To unpack this, I'm joined by Peter Swain. Peter is not just a fintech leader. He's a serial innovator, partner in next-gen lending platforms, former bank founder and now full-on AI evangelist. With a futurist mindset and a hands-on experience, he's at the epicenter of where financial systems meet machine intelligence. Peter welcome, thank you. Thank you for having me, so you know. Let's start. What was the pivot, what drove you from moving, you know, from banking to building AI, and what was the drivers behind that switch?

Guest:

Yeah, I think all entrepreneurs really are in the business of impact and legacy, and the money is hopefully the virtuous side effect. But when I had my children, it forced me to look at the world I was creating and the part I was playing in it, and looking at what can be the biggest shift for humanity as a whole. And whilst access to equitable finance is certainly on that list, I think that AI is both a massive opportunity and an existential threat for us as humans, that we need to wrap our heads around.

Jim Jockle:

So, you know, perhaps you know we can dive in a little bit now in terms of. You know, we started with kind of, you know, predictive, now to generative in terms of AI, and now we're in the let's call it the phase of agentic. You know, perhaps you can break down what agentic AI really is and why is it more than just a buzzword?

Guest:

Sure. So predictive and generative AI. And there's the third type, which is transforming. Transformer AI as well has been around for decades. It's the raw compute power and the attention we've put on it recently that has changed it. Where agentic is different from that is agentic is no longer about building new software. It's about building replacements for the humans using the software. So if you talk about 2023 and 2024, the focus was on building the next piece of shiny kit, whether that was chat, gpt or claude or mistral or more industry-specific applications. The focus now has been well, hang on a second. If we can get these things to understand, through vision and speech, what's in front of them, why can't we get them to navigate websites? And if we can get them to navigate websites, then do we need the human to use the AI to navigate the website? So agentic AI is essentially the premise of replacing the human operator with an AI operator.

Jim Jockle:

And so where is that taking us, though? In terms of you know, is this the new leap of replacement of the human We've already got to question three and we're already in dangerous murky waters.

Guest:

I mean, the simple answer is yes. The more long-winded answer is AI has yet to touch one of those jobs that I call the. When I grow up, I want to be jobs, so that when I grow up, I want to be an artist. When I grow up, I want to be a fireman. When I grow up, I want to be a princess. When I grow up, I want to be a fireman. When I grow up, I want to be a princess. When I grow up, I want to be a horse rider.

Guest:

Those jobs, the things that the human spirit feels compelled towards, where we find meaning, ai has very little part to play. The jobs that AI are taking are the jobs that, quite frankly, we never wanted. So the job of a customer service agent, the job of a data analyst, the job of a systems administrator, those kind of risk and compliance, those kind of jobs are the jobs that AI is taking, and what we do with that as a society and as a species is really up to us. But it certainly challenges the fundamentals of capitalism, as we know it.

Jim Jockle:

That's really an interesting way to think about it, especially, you know, as someone who has a young daughter. I think leaning into that spirit of what I want to be is a really great way to think about and potentially differentiate yourself in this new age. So thank you for that. So you know, just bringing it back to capital markets a little bit, you know where's the disruption over the next few years. Is it trading? Is it risk? Is it compliance? Is it the post-trade back office? You know, where do you see disruption?

Guest:

Well, if I could tell you a quick story, it might help on this. I was speaking to a gentleman that ran a debt fund out of New York with 20 billion in debt out in the world, and we were talking about exactly this subject like what's your concerns, what's your worries? And his concern was that, essentially, if you're a debt fund, there is no differentiation. If you're going to lend $10 million and I'm going to lend $10 million, and if I lend it at 12.6 and you offer it at 12.3, then you're going to get the business and I'm not. There's very little value add that you can bring to that transaction to differentiate yourself from the competition. So historically, what would happen is you would say 12.3, and I'd find a slightly more efficient way to underwrite the risk and I would offer it at 12. I then become more successful than you and I then consume your fund inside my fund and all of the analysts, all of the compliance people, all the risk people, majoritively speaking, stay employed. Maybe you have a 10% cull as you bring those two companies together. So that was his concern. My concern was slightly different and I think he begrudgingly slash terrifyingly, that's not a word from a place of terror agreed with.

Guest:

The larger problem was as follows If 20% of his fund goes delinquent, his fund becomes upside down and because he has no upside as a debt fund, there's no upside, there's only potential downside. So if he has $20 billion out in the market and you're talking about the wide-scale disruption that AI can bring, what if half of his fund doesn't adopt or doesn't survive that transition? That would be 50% If you half that number. So if my prediction of 50% won't make that transition is correct, if you half that number to 25%, he's still underwater. And if you half that number again, 25% he's still underwater. And if you half that number again, he's still in a very difficult position.

Guest:

So, yes, it's going to disrupt trading desk. Yes, it's going to disrupt risk. Yes, it's going to disrupt compliance. It's going to make those wildly efficient. And in the space of money and the level that capital markets move money around, a point of a basis point is enough to make a huge difference to somebody's effectiveness. But the larger concern really is actually what happens in the rest of the capital market. So you've got, for example, the CEO of Anthropic, the producers of Claude, predicting that 95% of software will be written by AI by the end of 2025. And Microsoft recently announced that in Copilot and GitHub, their two software repositories, the number is currently around 50% to 55%. If you were to just displace the software engineers in North America, you would have a very big change in how money gets allocated in capital.

Jim Jockle:

You know it's fascinating to have this conversation and not to go off track, but just thinking of everything that is going on in the world geopolitically, in terms of changes and tariffs and whatnot. You know, having this kind of thinking and overlay, you know, can really take you down an entire rabbit hole that we're not going to go down. But and overlay, you know, can really take you down an entire rabbit hole that we're not going to go down, but you know. You know trust is one of the key things that always comes up. You know there's things that the AI can do better, but you know it's. We've seemed to move away from hallucinations, you know, as is being the, the being the basis of conversation 18 months ago, but you know there are trust barriers. So where are those trust barriers? What are the biggest ones, or governance challenges that firms are really going to need to solve before they move to kind of an agent-first mindset?

Guest:

Well, the first thing I'd say and it's a great question because I think the trust and transparency is a really big side of this governance is incredibly hard in this area because the the area moves so quickly that if you take a fortune 500 company, by the time governance and regulatories are drawn up and handed down throughout the company. Almost almost every time the thing that they just governed is different. So it's more a case of guidelines and ethos. But in terms of trust and transparency, the first thing I say to people consistently is trust. These conversations don't really happen well in a vacuum. So there's a lot of people talking about whether they should or they shouldn't, they can, they can't, they will, they won't. A better conversation is okay, we did. What results did we get? So the advice that we give people quite consistently is find something very small and look into the effects of agentic AI as an example.

Guest:

We just implemented inside my own company an AI agent whose sole job it is to register a new supplier, because we found out that registering a supplier can take us anywhere from 15 minutes to four hours. 15 minutes is we send the request, they send back the information, we upload it to QuickBooks. Thanks very much Four hours is. We send the request. They don't, uh, answer it. We then chase it up. They don't answer it. They finally reply. They don't mention whether it's a wire or an ach. They don't send in the w9. So it's this back and forth period. So we put an agent in place that we literally cc into an email and at that point it will reach out, it will ask the questions, it will do the follow-up, it will grade what it's got as a return to make sure it's appropriate and then send it back. Now that's, that's not a big win by any stretch of the imagination, but it's saving us eight hours a month. That's 96 hours a year. Call it 100 hours. Call it it a $50 charge out. That's saving us $5,000 a year and that's replacing one function of one job.

Guest:

So we got to see did this work? How did it work? What are the pros, what are the cons? And, in terms of the cons, quite specifically, how do you address this with the other humans in your business? How do you talk to your team and express is this a threat to them? Is it not a threat to them? How should they respond to it? That was probably the biggest area of insight we gained. But you don't gain anything by talking about something in a vacuum. You only gain by actually applying it, seeing it and doing it. So I think the answer to trust is to stop having conversations in a vacuum and start playing playing in very small ways so that you can see what the actual results are.

Jim Jockle:

And, just out of curiosity, what was? What was the by-product of the conversation was? Were the affected individuals happy? Did they find more time? Or, you know, was there an inherent distrust of oh you, this has been automated for me. And what's the next thing coming? You know how did that play?

Guest:

out. Well, I was brutally honest, which most people that know me know that I am. So I told my entire team that AI is coming for your job and I told them I was actively looking to replace their jobs, which led to lots of open-mouthed looks back on the Zoom. But my second sentence was but it's not coming for your career. So that's the question and choice you all have to make. You know what we do, you know who we're in service of. You know our mission, you know our vision. So come back to me and tell me where you can help, tell me where you can add value to me and to my customers and to my community and to my experience, because the human is now a hundred times more expensive, probably, than the AI equivalent, which means you have to be adding a hundred times the value. I'm not saying that's impossible by any stretch of the imagination. I'm not saying that's impossible by any stretch of the imagination. I'm just saying that AI is coming for everybody's job. Everybody's job is at risk, but I don't believe anybody's career is at risk.

Guest:

I think that people have to make a conscious decision to understand where they can add true value to an organization, to an enterprise, and then double, triple and quadruple down value to an organization, to an enterprise, and then double, triple and quadruple down. So I don't know if my team found that loving or not, but I find in almost every single instance the truth is a better solution Because, as you said, they know that I'm trying to do it. Why wouldn't I try and do it? So they know it's happening. So if I pretend it's not, I don't think I get their trust. I don't think I get their respect.

Guest:

I think by saying, listen, we can't ignore the benefits that this thing can bring us. But you, as the humans in my company, have a great deal of understanding in how we work, who we work for, why we work, why do I wake up in the morning? And if you can leverage that versus leveraging the hours that you're doing with me, you are worth an infinite amount to both myself and my community. If you think you're just here to push a button very soon, you're worthless.

Jim Jockle:

Well, you know, peter, just as an aside and a reaction to that, that's a real demonstration of being a leader and I just wanted to call that out. Thank you for sharing that. So, coming back in terms of you know fintechs and the rest of the world. You know, so I know you're deeply embedded in fintech lending. You know how are agents transforming the full lending lifecycle. You know from origination underwriting to servicing collections. You know how is that evolving.

Guest:

Well, I can tell you. So we run a lending company in Latin America, amongst other things I'm on the board of that and one of the large AI providers who I won't name. But they came to us and said hey, we want your data and we'd like to do a sample project with you just to show you what we can do. What can we do to show them what we can do? And we went completions. Now, for us, a completion is when we have somebody that has made a loan application and the electricity bill is in the wrong address or their pay slip shows a different amount from their loan application, and they're like okay, how many have you got? And we said, uh, 32 000. Okay, and what does the call send to him? Like, well, there's 30 people and they average 20 a day, so we do about 600 a day. All right, okay, we'll do that. So they uh deployed an agentic voice-based ai. So this is actually calling people in Mexican, argentinian and Peruvian Spanish, so very different dialects. And after six hours, the company in question finished the project because they had no completions left to do. So the human equivalent does 600 a day with 20, 30 people in the call center. It's like 26 or something. That means that they do 18,000 a month and the company in question did 32,000 in the first day and the statistics were shocking. At the very first moment it was terrible, it was absolutely horrendous. It was really really bad. But it learned on every single call it did and it got better on every single call it did. And this is really where you start seeing the power of AI, because humans have ego. We have massive amounts of ego. So whenever we listen to advice or feedback, we always filter it through our own human experience, which is naturally defensive. The AI doesn't have that problem. It literally is looking for the patterns in its own data and its own performance to improve itself. So every time it's doing an action, it's getting better at the action it's doing. So that was one example getting better at the action it's doing so as one example. The other example was uh, back in full 23. We fed in every lending decision we'd made into just a high street llm um, large language model um.

Guest:

And I asked it just, this was just an experiment in the very early days. I said tell me something that you can see, that there's no way I can see. And it said you've got a problem on Tuesday. I'm like what it's like? You have a problem on Tuesday. Your acceptance rates are 2% higher and your average, your API, is 2% lower. I'm like how much does that cost us? It's like $220,000. Wow, okay.

Guest:

So I called together the C-suite and said tell me about tuesday. They're like what do you mean? Tell me about tuesday. I'm like, just tell me on tuesday. And they're like it's the day after monday. I'm like you know, I know it's the day after monday, but tell me what is different about tuesday? And they're like what do you mean? I'm like I can't tell you why I'm asking, because if I tell you why I'm asking, we won't get the answer. Tell me what's different about Tuesday. We were in the room for about 35 minutes before somebody said well, the head of risk has their day off on Tuesday. Could that be what you're talking about? I'm like yeah, that's exactly what I'm talking about.

Guest:

So the ability for AI to take these huge data sets and find patterns is an unparalleled capability that no human being could ever possibly get near. So I think you'll see wins across every single piece of the lending lifecycle, and the thing about being in lending is it's a great business if you're good at assessing the person that sits in front of you, it's a terrible business if you're not very good at assessing the person that sits in front of you. It's a terrible business if you're not very good at assessing the person that sits in front of you. So that's really. The only thing that a lending company does well or doesn't do well is underwriting. The deployment is normalized across the board. It's commoditized. Sorry, collections is commoditized. The only secret source that a lending company really has is their underlying data and their algorithm of who they're going to lend and how they're going to lend and how they make that decision, and AI will find the bias that humans have put into that system pretty much straight away.

Jim Jockle:

Here we are 2025, in terms of a revolution of AI. Would you categorize this year as being a potential tipping point, especially around regulation? Will tighter AI governance stop innovation or is it going to be a catalyst to building trust at scale?

Guest:

This is such a fantastic question.

Guest:

If you'd asked me in December, I'd have said yes, 2025 will be a year where AI regulation gets tighter and it helps us level the playing field. But obviously, as we know, that one of President Trump's first executive orders was to repeal the safeguards around AI that Biden has started putting in place, which led to the UK then following suit and opting out of the EU AI Safety Act as well. So and I'm not making this political, it's just a statement of fact the guardrails have actually been reduced, not increased. Now, if we were to get political, we could say that's, you know, spurring a huge wave of investment inside the us at the moment. We saw the the project with softbank stargate, I believe it was for 500 billion dollars come through almost as soon as that the executive order was repealed. We've seen softbank actually now step up a week ago and say they're actually looking at a number closer to $2 trillion of investment inside the US and the AI ecosystem there. So I think we're actually going to see less regulation, not more, which, from a capital perspective, is fantastic.

Jim Jockle:

From a humanity perspective, it's somewhat debatable you've coined the phrase ai agents as, as colleagues. What does that look like in practice? And I even think back to you know recently, uh, that you know the ceo of salesforce at a conference. He was out out talking about how he has multiplied his, his workforce, and, and it's human and uh, and and digital at this point. So, the agents as colleagues, what does that look like in practice? And are capital markets firms ready to work with AI as a?

Guest:

peer. So what does it look like in practice? I think the example I gave you of registering a supplier is a perfect example of as long as the work gets done, then does it really matter how it gets done? You know, we have embedded ad agencies nowadays that aren't your employees. We have, you know, w2s, and we have contractors and permanent members of staff. The workforce, the makeup of a workforce, is now so fluid. Anyway, I'm not sure it's really a differentiator if it's a human or an AI. Now, are the capital market firms ready to work with AI is a fantastic question, I think. Well, they're both great questions, but that's the one that really stands out to me, and the example that I'm taken to is when I've you know, I've been in digital for pretty much all my life. I had the 30-second listing on Yahoo and I was one of the lead developers for the UK version of Yelp back in the 90s.

Guest:

So I remember when Facebook started doing these things called lookalike audiences, which is where you can advertise to people that look like other people. So you've got 1,000 customers and you can advertise to people that look like other people. So you've got a thousand customers and you can advertise people that look like those customers and at the time that was invented, a lot of people in the industry, myself included, thought it was unethical. We thought it was ethical to say, hey, if Jim or Emily or Cheryl or Bob or James hits our website, tagging them with Facebook and then advertising is okay. But if Jim lands on the website advertising to Stuart because he has the same behavioral traits as Jim, that's not okay. Now no one needs to understand that example, because what then happened was everybody started doing lookalike audiences because it was too effective to not do it. It is now a perfectly normalized behavior to use lookalike audiences. So my answer to are they ready? Is it doesn't really matter, because if your competition starts doing it and they start posting 12% IRR versus your 10 or 15, versus your 12 or 20, versus your 9 or 30 versus your 8, you're going to get on board with it.

Guest:

The AI first future is really now inevitable. There is too much money from sovereign wealth firms. There is too much money from sovereign wealth firms. There is too much investment for this not to succeed. This is now a. This is a given. It's kind of like saying that green energy is going to replace the oil industry in any time soon. There's too much money in the oil industry for that to happen. You know, no matter what benefits we can get from renewables, no matter what investments being put behind it, the oil industry is a multi-trillion dollar industry. The growth of AI is now a multi-trillion dollar industry, so this is going to happen.

Guest:

The phrase that capital markets and the firms around that ecosystem need to decide is a real, simple question of are we going to be first or are we going to be second? Do we want the competitive advantage and rewards that come with being first that is combined with the risk of going first, or do we want the safety of coming second, negating some of our risk but losing some of our reward at the same time? That's the question that has to be asked. It's not shall we do this or shan't we do this? It's we're going to do this. When is the right time to do this?

Jim Jockle:

And, for lack of a better term, I think a lot of institutions are having that come to Jesus moment. If you will, you know what would you say is from your perspective, what is the playbook for for scaling ai in in 2025? I mean what, what? Where should someone start and what's something people should stop doing at this point?

Guest:

so I think there's we have two phases for this. The first phase is your, your, your own, uh, personal use of ai and I, up until about a year ago, I had a sticky on my desktop that said if it took longer than five minutes, I should have probably used AI. So I think we have to get rid of this sorry, this residual automation that we have of not using AI and going to like Google. You know I'm sure some of your listeners still have this, but I often get like a phone call from my dad or something and say how long does the drive from London to your house? Or something like that. I'm like Google Maps man. Why am I still answering these questions? Google, google it. I'm just going to Google it for you. Why do I need to do this? So I really don't want people to be that in the AI space of just not having the reflex, like the muscle memory, to say AI first, that's where we go first. So I think that's the personal side of things. Then, on the business side of things, as we said earlier, I think the real thing to do is just look for that low hanging fruit. Don't look for the biggest win to begin with. Don't look for the biggest disruption, look for the smallest win that's actually going to make a difference.

Guest:

The first one we did was registering a supplier. The second one I did was around anyone that used my calendar link. We noticed that often people get confused with time zones. So they'll book it at four o'clock in the morning, pacific, and then obviously they don't show up because they didn't mean to book at four o'clock pacific, um and? But for my time, four o'clock pacific is noon, so it's actually quite a busy time for me. So we put an agent in place that looks at the calendar appointment that just landed and we'll literally just email or text the person and say hey, jim, you just booked a call with peter. I know he's looking forward to speaking to you. This is sally of his AI assistants. Just wanted to check because it looks like it's four o'clock in the morning, your time.

Guest:

If that was deliberate, no, you don't need to do anything. If that wasn't deliberate and there was maybe a time zone confusion. This is the link to reschedule. We've managed to eliminate almost all of those no-shows from that, which were probably two a week, so 104 a year, where I'm sitting for 30 minutes waiting, or 15 minutes, if I'm being honest, waiting to see if the person's going to show up. So that saved me 25 hours a year and all of these small wins this 25 hours and the $5,000 for registering a supplier soon start adding up because I'm driving margin. We have a phrase at this end, which is revenue is vanity and profit is sanity. We can all drive big revenue numbers and they're fantastic. I love those. But the numbers I'm really interested in, the numbers that profit, they're the numbers that end up and take home for founders and take home for partners and C-suites. So always driving margin means finding those tiny little efficiencies that start compounding real quick.

Jim Jockle:

So sadly, we've made it to the last question of the podcast. We call it the trend drop. It's like a desert island question, and if you could only watch or track one trend in AI, what would it be? It would be the percentage of agents deployed versus the percentage of humans displaced. Wow Well, peter, fascinating conversation and, I would say, probably the most quotes and thought-provoking statements of any of our podcasts to date. I want to thank you so much for your time. This was really fantastic.

Guest:

Thank you, I really appreciate it, and thanks for having me Please.

Jim Jockle:

Anytime. Thanks so much for listening to today's episode and if you're enjoying Trading Tomorrow, navigating trends and capital markets, be sure to like, subscribe and share, and we'll see you next time. Transcription by CastingWords.