Ahead of the curve podcast

Artificial Intelligence and Tokenisation in Post Trade Solutions

  1. Overview
  2. More episodes

Overview

This episode of Ahead of the Curve offers a forward-looking perspective on how AI and tokenisation will redefine collateral management. Learn how Post Trade Solutions AI initiatives, from natural language search to predictive insights, are helping clients reduce manual tasks and focus on strategic risk management. Explore the intersection of AI and distributed ledger technology, and how tokenisation could enable real-time settlement, optimise collateral usage, and mitigate systemic risk. With regulatory considerations and client needs at the forefront, Post Trade Solutions is charting a path toward a more agile, intelligent, and secure post-trade ecosystem.

Listen to the podcast

Hello ton and welcome to another episode of lseg Post-Trade Solutions ahead of the curve. I have two really good, awesome speakers today with me, um, because we're gonna be speaking about two really awesome topics. Um, no, you know, no surprise, ai, uh, which is what everyone's been talking about, um, all over the place all the time, every day. Um, we're gonna be diving into that. Um, and really specifically on how AI is deployed, um, in L Ls X post-trade solutions business specifically to Acadia. Um, I'll caveat that, you know, across London Stock Exchange Group, um, AI is, is everywhere and prevalent, and there's a lot going on in that space. But for this particular conversation, we're gonna be talking about it, how we are deploying it in Acadia, how we're using it, and really kind of looking forward into the future. And really the other topic, um, that is kind of on everyone's minds in, you know, in kind of FinTech and in, uh, finance is tokenization. It's kind of, uh, not a new topic. It's something that we've been hearing about for quite a long time, goes along with DLT technology, but, uh, recently we've been hearing a lot from regulators, um, and from the industries, you know, in general around the use of tokenized assets, uh, for collateral, um, tokenized money market funds, um, and then how that's gonna work, um, and create efficiencies throughout, uh, the industry. So we want to kind of unpack that. We want to take a look at, um, you know, how that's being deployed as well. So we have Will Tomi with me, he's head a co-head of business development, and Jay Goldman here, who is kind of our AI expert, Acadia. He's been working on a lot of things, um, for quite a while now, I think. Mm-hmm. Yep. Um, we'll talk about the AI chat bot and so on and so forth. Um, so, you know, hopefully this is, you know, uh, a really good conversation and really kind of how the two might tie together. We'll kind of explore a little bit of that. Um, but, you know, I think we're gonna have a, a good conversation. Uh, Jake, I'm gonna, I'm gonna hand it off to you first. Um, can you just, before we kind of jump into like how the future's gonna look, can you just tell us a little bit about how, you know, Acadia has deployed ai, um, what you've worked on? And then we can kind of talk a little bit about, you know, where it's going. Mm-hmm. Yeah, yeah. Alright. Thank you. Thank you, John. Yeah. Um, so we've, we've chosen to tackle, uh, a, a few problems over the past kind of few months. We've been working on it for about a year, year and a half. Mm-hmm. Um, and that's of document searching, kind of a very classic, uh, 20, 24, 20 25 problem now of companies have a lot of files, they have a lot of, uh, documentation that clients want to see, and it's not always the easiest to search on it. Mm-hmm. So we've, we've decided to kind of tackle that one first. Um, and we've seen clients just kind of love it. So just a way easier to interact with our system. Um, so what we did is, is for our clients, instead of having to go through kind of an old documentation site, um, they can instead kind of interact with a, with a natural language. Um, and so we've kind of a little technical bits that we do from document ingestion to making sure that there's always a citation. Um, but ultimately we've found that that's, that's kind of the best easiest use case for clients to, to ease into this process. Um, uh, and, and we've, we've seen a lot of really great results with it. So, So, you know, when I think of ai, obviously, you know, I think of more like how things are autonomous, what you're talking about really is something that a person was doing kind of, you know, querying, um, clients, uh, issues or problems, and you've kind of trained this specific, um, agent to just answer those questions that, uh, maybe people have, were were answering or is it more of like a documentation search or is a combination of all those things mm-hmm. Because like, you know, really all comes down to is if, is creating efficiencies and, and kind of getting whatever friction is there in time consumption out of the process. Yeah. Is that kind of where Yeah, it's a little bit of both. Yeah. Okay. Um, we probably first started with just the simple searching of documentation. Mm-hmm. Uh, right. Someone wants to answer, you know, what does this field mean? Or what does this number, how is it calculated? Uh, and that's on page 405 of a a thousand, you know, page PDF. So that was just an easier way to, to find those answers. And then the first thing you mentioned around just answering common questions, we found that this as a result of having a language model that was trained on our documentation, it all of a sudden understood our products really well. So it could, it could answer these, these common questions that clients were coming to us with. Um, uh, and, and even kind of take it a step further where there were kind of new unique questions that had never seen, I know everybody's familiar with chat chui and things, it can kind of start to, to to understand, um, uh, what you ask about and what you might ask about. Uh, and that's kinda what we saw. We saw that instead of, you know, in our, in our land of collateral risk trade formatting, um, exposure calculation, uh, while it didn't, it didn't need to actually know anything about what the client is doing, their actual data, um, because it just knew our application so well. If someone had happened to ask, oh, how is this calculated? It can then actually apply that to exactly what they were talking about without ever needing access to any of their data, which is really important for us. We don't want to ever, and you know, I'll say this right to the camera, we never want to train on our, uh, client information. Um, but there's a heck of a lot you can do once the ai, the language model knows how your products work. Okay. Um, so maybe an Important part to, to point to stress there, it's that it's our doc. I mean, we, we've built a lot of products, uh, for quite a long period of time. Mm-hmm. And so there's a lot of documentation we've generated to support those products both internally and externally. Right. So, um, this is using our own documentation that we've created for our products and training a model so that people can interact with that. Yeah. Yeah. And I, you know, I remember when you first started it, and honestly, you know, you talked about chat, GBT, obviously everybody knows what that is. Um, I think it just passed its three year anniversary. Hmm. Um, the first time I heard about chat, GBT was my daughter. She was a freshman in college. She had a friend who's like in coding. She's like, oh yeah, she's, he's using this really fun, cool thing to create code. And I was like, really skeptical. Now all of a sudden, everyone's using it, everybody knows about it, and it's become an agent for a lot of people. I think I just mentioned it to you before this podcast that like, I'm using it as like a personal assistant already. Mm-hmm. Um, and this is As you should, And we're just, and we're just scratching the surface, which I'm assuming we're, you know, the way we're deploying it at Acadia and post trade solutions, we're still, we're just scratching the surface, I think. So are, are there any things, like what, what are you excited about? Like what are the big, are they big problems or other problems that you think or you see AI helping to solve in our, in our world again? Mm-hmm. I'm caveating this because we are talking specifically about our business. Yeah. When I say our business, I mean specifically the Acadia business. I do think, you know, going in, we're, we're recording this in 2025, at the end of 2025, December, I think going into 2026, we're probably gonna have a much larger and broader conversation about how AI is deployed across all of post trade solutions. But, you know, what are some of the, what are some of the newer things or medium term things that you see coming besides, you know, this great stuff that we've done around documentation? Yeah. Um, so we, we've had about, um, at, at this point in time, about a a thousand interactions with it over the, we, we've been rolling it out in tears over since summer. Mm-hmm. You know, so large global banks used it. And then some, uh, hedge funds, multinationals, some pension funds have now accessed it. So we have a good bit of data for what people are, are using it for. Mm-hmm. Um, about 50 kind of unique firms globally. Um, and we're, we're excited 'cause it kind of really just solved the documentation piece, but in, in the, in the near term, we are working on some pretty exciting things I won't talk about. Um, but in the near term, uh, it's a lot of data representation. Yeah. So we take in a lot of information from, uh, our, our, our clients for, for trade data, for pricing sensitivity calculation and, and a few other things downstream. Um, and, uh, it's not always easiest to represent that data in the form it needs to be. And so, like I was just describing about having a client or having, having our language model understand the product and then it being able to extend itself a little bit outside of its realm because it, it doesn't actually need to know about the client's information. We've, we've, uh, seen it be really useful for having clients answer questions that instead of coming to us, um, they can actually figure out how to map fields, how to, how to solve data representation issues, um, much quicker than kind of, we could do it as helping from our, our, uh, client support team. Um, so that's probably that it's a boring thing, but it's a really exciting thing because clients can integrate into our system a lot, a lot easier, um, and solve, uh, downstream issues, exposure, collateral issues as a result of the key input issues that, that, uh, our, our system kind of points out. Yeah. You, I think you could argue that like at its foundation we take in data and send out data, you know, like we're kind of the, uh, we need to do that nimbly because unfortunately there's not a standardization of, of data in every single realm. Right. And, and, and certainly this helps take unstructured content mm-hmm. Helps structure that more into our data model equally on the output side of things. And it doesn't think a rocket scientist to think about, like, there's better ways that we could use language models to kind of provide insights and summarize things in, in a more, uh, you know, linguistic style representation of that content and, and, and, and less less of giving raw data to people and having them form their own opinions or find patterns and things of that nature. So, yeah. Yeah. I mean, um, you know, what, what strikes me in this whole space, right? Again, I'm, you know, I'm not an expert. Um, I'm, I'm as far an expert as far as I can Google things and use chat GPT. Um, but I, what I, That's, that's being an expert. Well, that's, That's arguable. Um, but I guess what strikes me, and we'll have a little bit of fun with this, but, you know, the, use the, the different words, you know, it started out with like just ai then general intelligence, a GI, and then you start hearing of l LMS language models. The new word is agentic, as far as I'm concerned. That's new. I just started hearing that. So, you know, what does that mean? I mean, I think, you know, from what, um, what I, I can glean is that agentic AI is, is kind of like what I mentioned before, like your agent, right? Mm-hmm. It's, it has memory, um, and it could really kind of dig in and, and be intuitive. Mm-hmm. Um, is that something that we're looking at down the road? Like, I think of like what we do as a business one, some of our core businesses, margin manager, IMEM, um, resolving disputes. Mm-hmm. Um, is that what we're, is that, I mean, without going into the special sauce again, we're, we're all, like, everyone kind of does similar things in FinTech. Um, and I think a lot of our competitors and people that we partner with especially are, I've been seeing them going into the space. I'm assuming we are. Um, but, you know, what are some of those things that might be exciting that we can talk about here? Yeah. Yeah. You, you'd be correct. We're we, we are, um, I would say beta testing a lot of that stuff internally. Um, uh, to answer your first question, age agentic would, would basically imply it's kind of doing multiple steps. It's, it's not just a single, you know, you ask chat GPT how to, um, how to answer this thing or how to make a good pizza. It's, it's then, oh, you know, going to order the ingredients and going to do kind of multiple things, showing you the videos, how to do it. Um, so that multi-step process is something we're, we're, we're exploring with respect to things like exposure, reconciliation, kind of definitely in, um, those, those key applications you mentioned. Because, um, we wanna empower users that, that, uh, usually are, you know, they're given many different systems, many different things they want to access. Can we boil that down into an area that they, they don't need to look at a hundred different screens. Um, uh, can we, can we give them all, all that information just in one and then be a bit of a feedback loop? So kind of another term, not to throw another one at you, but reinforcement learning is kind of used in training language models For me anyway. Um, yeah. I'm ignorance, I'm sorry. It's the, it's the process of, of reinforcing kind of what the language model knows. Mm-hmm. And so in that process, if you can boil it down kind of the, just the summarization of issues on a given day, that's an easy thing. You're, you just, you dump it a bunch of data in a certain way, and then it's now given the client, you know, Hey, all, here's everything that's going on. Mm-hmm. Can you then get a little bit of a feedback loop for it to then start to take actions for that person? Can it then go and actually fix the thing for you? Um, and that's what we're really exploring. Can it, can it, can it do, uh, repetitive tasks for you start to learn and then be proactive? Um, and I think all of that, is this The same as being autonomous? To some extent, yeah. A little bit. We're not there yet. It's not, it's not, uh, anything like it, it's not said It and forget it. Yeah. I wanna fast forward to that. Yeah. Um, to that part, um, at some point. No, look, I, I understand that's, that's really, that's really exciting. I mean, on the margin manager side of things, I mean, any thoughts, um, on, on maybe like, you know, what, where are we heading? Is there a strategy? Is there something that we're, we're excited about? Or, you know, is it more of like a multi, multi-step process here that We're, I I think it's the, it's kind of the data and data. I think I mentioned before, I, I if you, if you write the history of what margin Manager is, it more or less is a defacto messaging standard for people to communicate amongst different collateral systems, right? And so, um, what that does require, however, is people to write effectively to our API structure commonly. Yeah. Now we also take in data in a variety of different forms, right? It's not just always our API, it's fixed messages. Mm-hmm. It's, um, you know, file based, uh, SFTP of files to us that we transform into our data model. So if you take all of that and, and there's other avenues of, of, of, of getting just data into the, the messaging network, um, I think we want to be less structured over time and be able to support that. Or even if it's the, the scenario where we're writing transformations of, uh, you know, their client's data model and representation of whatever it is versus how it needs to look within our, in our service. Mm-hmm. You know, we, we want that to be a much lighter touch human involvement process. I mean, in theory, you could make that more self-service for clients. Uh, and I think equally on the output of data, you know, now that you're running that, and when we mentioned about disputes and stuff, there's only so much content we might have in, in certain cases, depending upon the workflow that we're referring to. Um, but sometimes it doesn't take, take the interest statements workflow, the interest statements workflow is, is, uh, you know, should be institutions submitting in kind of a daily interest accrual over the month based off of a balance and an interest rate, so on and so forth. Um, the workflow, the value, the workflow, of course, it gets people to immediately agree or not agree and then show where you might be off. You have a different reference rate on a particular day. You have a different balance on a particular day. It's not rocket science. Right. But like, But that requires someone, But that requires someone to go look at it, Right. And researching. That's right. What rate I am, what That's right. Rate they're using. That's right. Do the math. And we can describe that back. I mean, I think like instead of having a human have to figure that out, there's no reason why we can't describe it back. Now in talking to, to, to some clients about these types of ideas, the interesting points raised like they, you know, because the, the, the margin manager network's very much kind of the glue in the middle. Um, most institutions are, you know, using their own collateral management system, whatever that is, whether it could be our collateral management system too much you, but they're using a collateral management system to connect into that gateway. Yeah. And that's what they live in. So if we produce outputs of data, then you still get back into the, well, what's the structure of that data? So that can be read into something else. Mm-hmm. Mm-hmm. And, and, and so I think that that becomes a little bit challenging, presumably, but, you know, you know, it, it's, there's no reason why we can't use it to help pattern data, explain data, produce some rationale, some insights, but it's like, I don't know if anyone wants to log in to something else to see that. It's like, then it's like, how do we get that injected into, into something And work in the background? That's right. Yeah. I mean, I think that's, that's the one thing that's exciting to me is, you know, errors that are repetitive and known things that people are doing, like humans are doing over and over again. And, you know, I know there's a lot of, you know, especially, you know, on the other side, our clients are, are going into root cause analysis and trying to figure out like, why is my data this way and trying to fix it. Um, you know, if, if, if this AI technology or any type of AI technology can, can kind of speed that up at some point, which it seems like that's where we're, where headed mm-hmm. Um, where it learns it's intuitive, it could, you know, figure things out and just say, okay, this problem will not exist anymore. Right. Right. And I think that that's the right way to think of it. The, yeah. The, the margin call itself. Like we've talking About spot, like we've been talking about Snap That's right. Snap, like forever. Right. That's, that's a huge problem. Still a problem. Yeah. Right. But the margin call itself is such an aggregated number, right? Like, and you think even for one institution to do their own calculation of a margin requirement, oftentimes they could have multiple trading or risk systems. Um, you know, all of those, the, the risk calculations that have to be performed overnight. Mm-hmm. You know, all of that's kind of getting fed into a collateral management system along with positions and, and, and, and, and, and, and instrument reference data and pricing on securities and, you know, understanding parts of the legal di. So like all of that kind of is a choke point. Um, and, and I think there's no reason why AI can't help pattern the repetitive nature of you. You continually fail or have a dispute or whatever as a result of something. I think the difficulty really ends up being the starting point is kinda the middle ground, the messaging part of it, but it's like, how do you then understand additional information that sits in other hubs and supplement that mm-hmm. And provide context and things of that nature. Yeah. Jake, what are some of the things that, like, I think I, I asked you before, but I want to kind of pinpoint like, what are you excited about going forward? Like what are some of the, what are some of the things that you see that you're gonna be excited about in next year and in the year after, um, you know, using this technology? Is there anything emerging that you see that's new? Um, are there any new like adjectives to explain different kinds of AI that I, I haven't heard of yet? Um, but is there, you know, is there anything that comes to mind? Yeah, I, I think there's probably two things. One is the, is the squashing of repetitive tasks. Mm-hmm. At least in the initial margin space first. Yep. Um, we are kind of the central point for a lot of really great information. Mm-hmm. And there's no reason we shouldn't be able to pinpoint at the lowest level. 'cause you know, the margin call is high, but we have all the low level detail to know what should be the issue or what's the, what's the likely cause. And I think the other thing is, um, around unifying, uh, and, and, and kind of sidestepping issues of the past of legacy technology, right? I mean, you were just talking about, um, a a margin call amount or, or even margin manager encompassing very various different event types of substitutions interest. Um, and instead of having to, to, as, from a developer point of view, instead of having to build all these different screens, unify them, yes, they might go to a collateral system, you know, no, they might not come into our ui, AI is almost a way to bring all that together in, in a, in a simpler way than having to think about all those things as being separate and unifying them. I'll give you another acronym. Uh, and, and MCP is like a certain protocol that, that people are using now to, uh, attach a language model into people's APIs. So you can think of that LSEG is, is, is, uh, has publicly done this with, um, a few companies for, for kind of their various DNA related businesses. But if we can think about that in the markets, in post trade space, creating a, a, a unified layer above all of our products, um, uh, we can, we can start to bring all this together a lot quicker than having to build separate things or, you know, luring someone away from their collateral system. It can just kind of be an appendage that, um, naturally could be anywhere. Wow. We Should probably Sounds cool. Probably mention like, all of these things are fairly complicated engineering things that are, that are that mm-hmm. A lot of people are putting a lot of time and effort in. And I think that, you know, we, we do not take lightly as a company being kind of a trusted partner to the large financial or the financial institutions on the planet. Mm-hmm. And, and I think we do this very thoughtfully. We do it with a lot of pe, you know, smart people thinking through how this should be engineered, how it should work. Mm-hmm. Uh, and yes, we're more kind of probably towards the start of the journey than the end of the journey, but, um, I think we will do this in a way that is, uh, beneficial to us, beneficial to our clients, and done in a, in a very trusted way. Yeah. Yeah. I mean, I think, um, it's exciting times for sure. And, you know, you mentioning, you know, across the enterprise of lseg, how it's all gonna bring a lot of the stuff together. And I think, you know, as we talk about our journey in post-trade solutions, you know, 2026 is very much one post-trade solutions. Like, if AI is gonna help, you know, harness that, that power and put it all together and help us do that, I think that's gonna be a huge win outside of all the, the specific fun things that we're looking at too. So, yeah, that's great. You gotta keep smart kids like Jake, Jake entertained here in terms of figuring out new things. It just put me in a little box, gimme some toys to play with. Keep Going. Yeah, keep going. Um, see How far we can take it. Yeah. Sounds great. It sounds great to me. All right. So, I mean, you know, we can probably spend more time on ai. We can kind of come back to a few things, but the other thing that I mentioned before around tokenization is another I, you know, I call these like, you know, pillars of, of what's, what's kind of like hot topics these days, topical things, um, that are really affecting, um, in a good way, I think. Um, but obviously there's, there's caveats to everything. Um, how, how something like tokenization is going to, you know, change our industry, um, how it's gonna change our business, the way we do business, how, how our clients are gonna change. Um, and I know it sounds like one thing, right? Mm-hmm. But it's not one thing, right? It's, it, it could be quite a few things. Um, but will I, like, I guess first before we kind of dive into the specifics, can you just give us like a generalization of what, when we, when I say tokenization, when we say tokenization, what do we mean? Yeah. The, the, the misnomer of course is people jump to crypto, right? Right. And so it, it distributed ledger technology, obviously, DLT, uh, supports the underpinning technology, supports the crypto markets, um, in the collateral management space, people need to remember that people only take, you know, typically very liquid forms of collateral that are very deep markets that are highly traded, that are transparent, so on and so forth. Um, what does that mean typically means there's an awful lot of cash in the major currencies transferred for ca for collateral purposes as well as securities. And those securities are usually kind of government securities. Of course, they do branch out into, you know, mortgage backed securities and corporates and equities, so on and so forth. But if you were to kind of, you know, put them in order, you would have, um, you know, government securities, US treasuries being kind of a, a big one. Mm-hmm. So what tokenization is really referring to in that context, it's more nothing that is typically transferred for collateral management purposes today is natively a digital asset. Right. It's not something, it's not natively issued in terms of it being a, you know, something that's one, a distributed ledger, uh, technology or platform. Um, so tokenization by nature is referring to saying that if you took, we took a real world, real world example, if I took a US treasury and segregated that, so it's somehow protected. I don't own it, I don't put it, it's not kept in my account where I control it, but put it in to a, a second account for a service and then create a token. At that moment in time, you have a, a digital, uh, twin if you will, of what that asset is. Now, once we've done that, the ledger can, can effectively maintain who owns that asset initially. I would own that asset. 'cause I've, I've tokenized that asset and I, that token's been created and issued to me. If I wanted to transfer something to Jake, it could be a margin call that he issues to me and I agree to send to him, I can then pipe that if we mutually agree to transfer that token, uh, we could pipe that. And this is kind of where the extension of margin manager, we, where we see it or what, where our role could very well be is that today we mutually people mutually agree to transfer collateral on our platform. Well, same thing would happen. There's probably slight changes in some of the data model to describe it's a tokenized asset of some variety. Uh, but once we do that, I think we wanna, you know, there's some number of these ledgers that will exist. We don't know how many, of course, but, but there's certainly going be more than one. Mm-hmm. Uh, and, you know, we could get from mutual agreement, which we call a pledge accept in our workflow today, to transfer that as an instruction to change ownership. And that can instantaneously happen, uh, as a result of doing that. You read that back to the institutions and, you know, so you get from mm-hmm. Margin call mutual agreement to transfer something and settlement immediately thereafter. Um, that's the promise of it. Yeah. You know, the difficulty of it is, is a little bit around, you know, some of the legal context, you know, collateral's all about how do I ensure that I, that there's a legal framework in whatever jurisdiction we're referring to, and there's opinions and so on and so forth, that lead me to believe that if there was a default, uh, that I can get ownership or perfect interest in that collateral, um, as I would need to Right. Immediately. Yeah, yeah, sure. As part of a legal proceeding. Right. There has to be people that are all in agreement. I think that the, the, you know, We won't get into smart contracts Yet. No, no. I think smart contracts a slightly different story there. Slightly. Yeah. But in, in this, in this realm, um, what you, whatever we're doing is we're all legally entrusting a ledger to say who owns an asset at any moment in time. Mm-hmm. So if a default were to happen, everyone needs to feel comfortable that, you know, that moment in time is, is struck. That asset shouldn't be able to be transferred any further. Mm-hmm. Right. And, and ultimately I should have a claim on that asset. And there should be, you know, enough comfortability in terms of the, the legal framework or jurisdiction that that, that everyone's operating under to be able to kind of say, okay, yes, you, you own that asset and you'll be able to get that asset. 'cause you, the token itself isn't the asset, it's just the, you know. Right. It's the token digital twin or equivalent of the asset. I actually need to get the underlying asset that sits in a seg account at the end of the day. Um, so that's tokenization. Now, some people will even go as further to say, to get confusing here, that they don't wanna think of it as a token. They wanna think of it as a, as a, a, a, um, a digital way to transfer legal ownership of something. Mm-hmm. Which is, I dunno, maybe splitting hairs a little bit, right? Yeah. But, but like, I think tokens an easier way to think of, of what that is. And, and, and you have stable coins of course and other things, and there's a, there's an implicit trust that needs to be built as well in the market in terms of, um, that whatever is sitting there that the thing that isn't the natively digital thing. So if we're all agreeing to transfer on a ledger, some, you know, uh, some, some, some digital message effectively, um, that I, I entrust that whatever is the service is keeping a one for one. If that's the expectation, if there's one, you know, $1 for every kind of dollar equivalent of token, uh, or stablecoin or whatever it is that, that those assets need to be there. Right. Well, Bitcoin, yeah. Well, and then crypto of course is the angle of natively digital assets. We won't go there. Um, you know, it's interesting, as you were talking it, it, I, I started thinking about the intersection between ai since we're talking about AI and tokenization. Um, when you say like, again, in a per, in a, in a perfect tokenized digital asset world where you are on ledger, both, you know, everything is on ledger, ownership is immutable, you know exactly who owns it. But then, you know, in an event of default where, where like the rubber meets the road mm-hmm. This is, you know, pledging collateral is obviously important for risk management. And we all know, you know, anyone who does this for a living knows, knows why that's important. But also on the flip side of it, when there is a default mm-hmm. Um, there's gonna be a mechanism somewhere where like something or someone has to make that determination, right? It's an illegal document. This, these are the, you know, these are provisions of, of what a default looks like. And then I would have to get that collateral back or transferred back to me. We lived through oh eight, right? Mm-hmm. Um, I remember being on, on the phone, I'm not gonna mention the firm's name, trying to get collateral back. Mm-hmm. And this was the days before even ma margin manager was around, and it was nothing was even electronic, let alone on a deal, on a distributed ledger technology platform. Um, in, in a perfect world, you, I would think that, you know, and again, maybe I'm throwing AI into this as well, that something, or some, someone or some, you know, system is going to actually make that determination instantaneously and move it over. Now maybe I'm getting si science fiction, but It could be like a rules based approach plus A little bit, right? Maybe, maybe it's a rules based approach plus, Well, if you go down the, if you go down the smart contract route, which I do think is a, is a much more complicated scenario Because that to me is, that's the real value. It's like, if I don't have to wait, and I know for sure, like you said, the word trust here is important. Yeah. I know exactly where my collateral is at every moment of, of the journey, whether it's substitutions, maybe that's a separate ledger. But when the rubber meets the road, when, you know, when it really comes to, you know, an issue where I, I need to be protected and I want my collateral back, because you're in default, no offense, I'm in default. Maybe I'm giving it to you. Yeah. Um, that I know it's definitely coming back to me and it's coming back to me Quick. Yeah. I mean, I I Of course, otherwise then there's stick with the old version. There's the what if, I mean, right now as we sit here today, it still would be common practice for people to have to issue a notice of default. And that's a, you know, oftentimes a physical couriered type of letter to someone, there's people that are actually trying to digitize that type of framework. Right? Um, so you, you have to take the baby steps or you have to get everything into a more digital framework. I think in terms of what is an actual default, it's an interesting question. Today, there's somewhere in the magnitude of three to 4% of, of settlements that are for collateral purposes and for the collateral transfer mm-hmm. Are not made. And by no means are all of those deemed defaults. Mm-hmm. Because what's happening is some form of operational failure. You know, IIII didn't get Mr. A window, Mr. A window. Right. Settlement system failure. Right. Whatever. Right. There's, there's technically You're in default, but you're not Really fault. That's right. So, so there is a need certainly for that to be a, a, a much tighter understanding. Right. So if you really go, go back to oh eight, the firms that did best, so there's, you know, not plenty of firms failed mm-hmm. But plenty of firms ended up surviving, whether it be by government support or not. But in those scenarios, the firms that were least impacted by the major defaults were typically the ones that were the quickest to pull the trigger to default to the institution that defaulted. Right. So there's a speed of which that is commercially beneficial for anyone that's, you know, in that situation. Um, so like, you know, you, you think of, do we want that to be the practice? I mean, that's a, that's a, that's a, that's a very, well, I think these are incentive structure, these are of things that have to be considered right? Sure, sure. Mm-hmm. As this is rolling out, I mean, there's a lot of things we're not gonna talk about every single thing, and I don't want to harp on the default scenario, but I feel like not enough people are, I think everyone's kind of looking at the, you know, the liquidity improvements, the speed in which, you know, the atomic speed in which you can move collateral. But I think, I think that's necessary because it's, again, like it's, it's where are the problems today? You, you don't actually have thousands of defaults on a regular basis, but you have thousands of, of settlement failures on a regular basis. Right? True. And so those settlement failures have real world Negative person. Well, I mean, It's like the, the very extreme I, you know, I have, I have war wounds like you do. Yeah. So I always think about the worst case Scenario, but I, I do think took the, took the role of tokenization, obviously, is to try to speed up the system to allow people to be more optimal. Mm-hmm. In terms of, you know, if I wanted to deploy a more, um, you know, oftentimes I'm really trying to optimization, optimize the collateral that I have. I'm bound by the constraint of operational capability, you know, how quickly can I do what this optimization engine's telling me to do? And you're, you are gonna be constrained by the way settlement works today. So, you know, tokenization offers a future which allows people to, you know, hopefully reduce settlement fails, reduce the cost of those fails, improve liquidity, improve kind of how much you can optimize the book of collateral that you're managing on a particular point in time. All of that has value. The, the, the other, the downside of it is, you know, at least at the onset, there's not an obvious takeout of some existing system. So you're still, you know, custodians still exists. Mm-hmm. There's still, um, you know, there's still, so your Legacy infrastructure Still stays, will Stay in, in theory. And you're layering on now a new new aspect of, of how you might go from leg legacy infrastructure to, you know, morph that into a more digital infrastructure. Mm-hmm. Uh, and so I think what you're also seeing, of course, is certain governments are very much pursuing natively issued debt, you know, government bonds, but they're, they're digital assets themselves. That's a slightly different paradigm shift, right? Yeah. Because at that moment, you now need to build a more, um, you know, the, we commonly hear a central security depository. CSD is the thing that kind of exists for a particular market. Now you need to build a digital securities deposit, uh, uh, in some A DSD, right? So you need something that's the equivalent of that, but it's operating in, in, in a more digital and DLT manner. That, and, and the fact that governments are pursuing that, it's, you know, that's a completely different paradigm shift. I think even from primary issuance in the primary market to secondary trading to then collateral management. All of that needs to be done in a digital framework. Um, and so that's less tokenization, but it's still the same underpinning technology. Mm-hmm. So I think the smart people that are out here trying to solve these problems aren't thinking of it as a, you know, today we have to settle in this way, and then you have tokenization that is slightly different, and then you might have digital assets that are something different. I think you can blend all of that together and think of it as a continuum of, you know, we are, you know, where we are today and is it's functional and it works. And, and then, you know, there were things that will unlock whether they be my money market funds and token tokenizing them initially, just by the nature of I have excess cash, I wanna, you know, sweep that into a money market fund. But once I do that, I can't really do much with it. I, I get a return on my cash, but I'm not able to deploy that and send it to, to Jake in terms of, you know, a margin requirement, even though it implicitly has value. Mm-hmm. If I tokenize it, of course, I could think about doing that in a much more nimble way. So, you know, I think the, the industry's pursuing things as it always does. How do you take, you know, some form of binding constraint, improve upon it, make it more efficient, you know, squeeze a little bit more financial juice out of, of what we're doing. And I think tokenization certainly has a place in, in trying to solve some of those problems. Any thoughts, Jake? Yeah. Okay. Yeah, a few. I, I think it's, I, I like the fact from a technological point of view, margin manager in this approach would, would be that that kind of, uh, digital ledger, that that is the true truth that everybody has to kind of play off against. And the reason I like that is 'cause early DLT technology did it in kind of a more distributed approach where, you know, you had different, uh, nodes in a network. Everybody had to agree on that. That's, that's commonly how a lot of digital ledger digital ledgers work, where, you know, chains cannot get added to the blockchain and all this stuff. But you run the risk if you, if you distribute it, let's say in our scenario, all three of us are trying to move a piece of collateral. All three of us would have to have our systems up and running for any of it to, to be approved. But if will you have a operational failure in your bank's, uh, technology, you know, fails today, now all of a sudden that piece of collateral cannot get moved. You know, because all of us don't agree. So it doesn't really work. But when you have Margin Manager as the source, I really like that. It's like a, it's like a trusted third party and, you know, we're obviously a little biased, but it, you know, it's up just a little all the time and you don't have to worry about someone kind of, uh, lagging the system behind. I'd take another, like, you're right, the, the, the, the, the duration of a day is an interesting thing on its own, right? Like, we're a global business and, and, you know, you see the, the, the, the split when you have kind of firms in Asia facing firms in the Americas and Right. The time zone problems and shifts that happen mm-hmm. You know, folding into that equation is a bunch of exchanges on the planet for a variety of commercial reasons, wanting to extend the days of which you can execute and clear trading so you can, you know, we're entering this world where you can build up risk in a more global time period. Mm-hmm. But the settlement rails that exist for whatever particular type of currency, so have a business day to them. Right. So for us treasuries, you can't settle them really until after three p or you have to be before 3:00 PM uh, for US dollar cash before 6:00 PM technically, probably even a little bit before then to fund correctly. So you have real cutoffs that exist. And, and I think if the world does stretch out the trading day in a more globalized way, which does by all counts seem to be happening, the, the aspect of intraday, margining has to start happening. Mm-hmm. And, and thankfully, margin Manager is an event based infrastructure right. There, there, it's not, it's not batching anything. Right. It's just firing off back mm-hmm. Today, literally today. And, and not that there's a huge amount of use to this, but today firms can issue intraday margin requirements mm-hmm. If they have legal rights to do so and everything else. So you can kind of see the world happening where, um, there will be a need to calculate risk much more dynamically because it, the trading day's not gonna be so boxed in as it is today. Uh, as a result of that, you'll be able to kind of off, you need to be able to have some settlement mechanism that isn't so constrained by the time periods that it is. Mm-hmm. Again, the, the, the digital networks might be able to solve those problems, and you can easily think of it. Well, of course, AI has a role, again, in terms of data, in data out understanding what's happening. Uh, you know, if I wanted to optimize, you know, risk in terms of like, how can you get some insights in, in something, some tool that's giving you those insights mm-hmm. Agentic solutions to be able to, you know, pick more of those steps and unfold them. Mm-hmm. So I think like we have, we sit in the middle again as like a, a mix of a trusted partner plus some, some products of which people use as more of a network solution. I think we just see this natural evolution. I think we're, we're prepared for that, obviously, and is kind of where we're trying to go. Um, it, it, our, our jobs effectively, all of us here on this table, it's like, read the room, right. Read, read the industry in terms of where are we going and what do our clients need us to do, and then try to flex in that direction. And, you know, much of what we're talking about today is pontificating. Yeah. Yeah. But there is an element of we talk to clients about these things and try to get from them what are they thinking and how are they gonna use this? And we see that, you know, obviously shifting how the industry works. Yeah. I mean, look, we're, we're always, we're always client focused, and I know we're, we're product led, but we're also, you know, always listening to our clients completely and making sure that, you know, we're responding to them. We could, we probably talk about this forever. I know from on the tokenization side of things, there is a, um, you know, from an industry side of things, like, you know, there's a lot of concerns, but there's a lot, a lot of positive responses, especially going back to the regulators. You know, I just read ista, um, uh, response to the CFTC. You know, they point out regulatory guardrails, making sure that, um, you know, any of this new technology that comes out, um, doesn't surpass that or stays within those guidelines. Um, so there's a lot, there's so much more to unpack here. Um, I like to ask this question, and I'm gonna ask both of you in terms of either one, whether it's AI tokenization, those two topics that we're talking about today. Mm-hmm. If you had a crystal ball, um, what is, what would you, what do you think you're gonna see, or what would you hope to see in either tokenization AI or maybe the intersection between the two in the next year or two? Like what do you, what do you want to see? What are you hoping to see? I Would, I would hope to see, uh, collateral and port rec teams basically supercharged. Mm-hmm. Um, not that their teams need to get any smaller, but they're able to do more with, with kind of the existing people they have on that team. Um, and they can focus on other things besides maybe daily grunt work of, you know, making sure someone has done some very mundane task to actually fixing the things that might, might have been outstanding for years or maybe even a decade at this point. Um, that, that our system could tell you based on historicals or, or things like that. So I would say it's, yeah, it's, it's supercharging teams that, uh, maybe don't get a lot of love too, you know, you don't get a lot of new technology. Good. So that's great. Jake, we didn't prepare, by the way, anyone listen to this, I didn't prepare Jake for that off the cuff question, and you got a little bit of time to think. I Might, I might pile on top of what he said. I mean, I think of this as if we stick in the nucleus of collateral management mm-hmm. That when things aren't working, you have either a dispute or you have a fail. Mm-hmm. Right? And as a result of, of what you really want your collateral management function doing is understanding the risk associated to these things. Is the fail an actual liquidity event that could be a default, is a dispute something that I really need to care about? Unfortunately, those things are exceptionally clouded today by operational noise and ops. People spend an awful lot of their time getting a process to run. Mm-hmm. Just trying to get to a point without really kind of understanding what's driving problems. And so I think we do want the supercharging you, right. Like, I always kind of equate it to being, you want your ops people to look more like risk managers. Right. And, and I think they can't today because they're having to focus on the operational stuff. Just making something go from step one to step two to step three to step four. And they're clouded by all of the things that can go wrong in today's, you know, technical and, and operational processes. And if we can clean that up and use these technologies to clean it up, I think you can start to get to the point where the exceptions become real things that you need to care about, right? Mm-hmm. That, that could very well be a much easier way to isolate that there is some form of credit risk, like event that I need to care about. And that should be the purpose of collateral Management. Yeah. Well, I think, um, we're gonna, you guys would come back next year around this time to talk about these two topics again. Sure. Um, we're gonna play this one back and see where we are in a year from now. I'm, I'm really excited by a lot of this stuff. I, if I had a crystal ball, I, I would like to see like if the evolution to more autonomous ai, the agentic ai, and seeing how, you know, smart, it becomes where it's solving some of these problems for our clients that we're talking about today. Keep That genie in a bottle. Yeah. Well, you know, I'm just kidding. I get it. I get it. But at the same time, I think it's inevitable if I'm using my own opinion. Um, but I do think there is a, a level of, it could be helpful, and like you said, maybe humans who are doing some of these things could pivot to more, you know, value added functions mm-hmm. Where you're more, like you said, you're, you're looking at risk, liquidity, capital, things that are really like, you know, hitting your p and l rather than getting your head down into solving like mundane, everyday problems that that persist constantly. Yeah. So hopefully maybe we'll see that. So we'll come back, we'll come back in about a year from now. Sure. We'll, we'll play this back and then we'll talk about it. Um, and then we'll be, maybe we'll publish them both together again in 2026 so that people can see how wrong we were. Yeah. No, hopefully we're a little bit right. Um, gentlemen, thank you so much for, for joining us today, and like I said, we'll, we'll be back and do this again. Thank you again for joining us today. I hope you found this one extremely interesting. I know I did. Um, you can find our, uh, ahead of the Curve podcast on all your favorite streaming services, Spotify, YouTube, and on uh, lseg.com. Thank you for joining us, and we'll see you again soon. Thank you.

Also available on

More episodes

View all episodes
Scott O’Malia and John Pucciarelli

Unpacking the agenda with Scott O'Malia - Part 4

John Pucciarelli sits down with ISDA CEO Scott O’Malia for an in-depth discussion on the evolving landscape of the derivatives markets. Scott outlines the impact of recent geopolitical developments and explains how derivatives play a critical role in managing market risk.
Listen to episode Opens in a new tab

Risk Reimagined: The Rise of Open Source in Finance

This episode dives into the evolving regulatory and modelling landscape in the UK for traded risk. Scott Sobolewski speaks with Xabier Anduaga and Joey O’Brien about trends in counterparty credit risk, model validation and the growing adoption of the Open Source Risk Engine (ORE).
Listen to episode Opens in a new tab

Navigating Complexity: FMS, Risk Management & the Power of Open Source

In this compelling episode of Ahead of the Curve, Roland Stamm is joined by Maeve Gear from FMS-SG to explore the intricate world of winding down complex financial portfolios in the wake of the global financial crisis.
Listen to episode Opens in a new tab