03.07. Regulex AI Founder on the coming agentic wave

In this episode of the IA Talks AI, we speak to Karim Marchoud, the Founder and CEO of Regulex, on how AI is moving from an augmentation phase to an agentic phase.

Transcript

SPEAKER_01 0:10

Hello and welcome back to the IA Talks AI. My name's James King. It's my pleasure to once again be your host. Thank you for joining us again because today we have a lovely episode lined up for you. We are speaking with Karim Marshud, the founder and uh CEO of Regulex. Um so, Karim, thank you so much for coming in and joining us.

SPEAKER_00 0:33

Yeah, thank you so much for having me.

SPEAKER_01 0:35

Well, excellent. Well, maybe we could start by just learning a bit more about yourself. You know, you're you're a young entrepreneur and you've just in the in the recent past started your new venture. So maybe you could tell us the story behind that, like what led you to where you are today?

SPEAKER_00 0:50

Yeah, sure, definitely. So started off my journey at KPMG on their sponsored degree program. So we were studying computer science-based degree externally, and then we were rotating through their technology teams for the better part of four years. And for three years, I was specialized in their advanced analytics team. So deploying machine learning and data science principles to commercial and financial DD projects. And it was during that time that we worked on some really interesting solutions during COVID, like taking in non-traditional data sets like TomTom, open table bookings, cinema bookings, to really get a feel for how people within economies are moving and what we can decipher off the back of that. So real big data focus. And then coming out of that, I was then headhunted into the buy-side asset management. So I was working on engagements within Schroeder's, MG, Ruffa, and Phoenix. And it was during my time kind of iterating across portfolio management, engineering, and compliance that I saw the real problem with multimodal unstructured data and the amount of manual processes that were being done, especially around distribution, investment relations, and essentially the front of house where the data initially gets dropped into the organizations. So due to this frustration, I then thought, right, okay, I'm going to try and create infrastructure to address this problem. And that's when I created Regulex 1.0. And since then, I've now joined ANTLA in their founders' residency program, which I'm now in week six of.

SPEAKER_01 2:18

Can you tell us a bit more about that residence program? It sounds pretty interesting.

SPEAKER_00 2:22

Yeah, sure. So essentially it's 80 founders in a room trying to build businesses all at the same time. You have eight weeks, and then at the end of the eight weeks, you're trying to get investment from the team. And essentially during the time you're incubating the business, speaking to as many people as you can, and trying to get business partnerships as well.

SPEAKER_01 2:43

Okay. So that sounds a bit like Dragon's Den.

SPEAKER_00 2:46

It's a reality TV show. Yeah, 100%. I mean, we're we're in week six now, but it feels like months. It's a real pressure cooker, but it's great.

SPEAKER_01 2:55

Okay, good stuff. So there must be a great buzz amongst the you know 80 founders or Yeah, yeah, yeah.

SPEAKER_00 3:01

They've done a really good job at getting some great people in, especially all for like different disciplines as well. So it's great to obviously meet the team, but yeah, hopefully I'll leave some friends as well.

SPEAKER_01 3:10

Okay. So it's not like completely cut folk competitory there as well. Well, not yet. We're not at the end yet. Okay, good stuff. Okay, well let's come back to Regulex. So maybe you could tell us a bit more about what it is and uh maybe a bit more detail about you know what made you decide that you wanted to start it at this stage.

SPEAKER_00 3:31

Sure, definitely. So it started from the day-to-day roles I had across the asset managers I mentioned, and those manual tasks that we were doing within those teams, especially with unstructured data, reams of PDFs, words, Excels, PowerPoint. And I thought it'd be great if we had some form of infrastructure to try and harness this data, but then be able to inject that into the LLMs that we were using. So then the outputs would be correlated to the data that was actually in the database and sound like the kind of research that we were reading instead of hallucinations and the generic output that LLMs give. So that's what I started off with and focused it on manual workflows within distribution, investor relations, and kind of front-of-house teams. And since being at Antler, we're now starting to incubate a new product. So essentially, it's a genetic intelligence platform starting with distribution and wholesale teams. So essentially, the first version will be a product that ingests macro data, news feeds, portfolio holdings, fun navs, and analytics and CRM data to provide you with a morning brief and intelligence before you get to your desk, trying to understand what's happened in the world overnight, how does it affect your funds and essentially your clients and trying to give you that intelligence straight away.

SPEAKER_01 4:46

Fascinating stuff. So this is currently in the developmental stage, this new platform?

SPEAKER_00 4:52

Yeah, so we're developing right now, we're still trying to speak to as many people as possible just to make sure we ratify the problem and the space. And we're essentially trying to get design partnerships now just to show its real power and build it out from there.

SPEAKER_01 5:05

Sure. So let's fast forward to when the platform's completely up and running, it's in the hands of your intended users. What's it going to either enable them to do that they can't do now, or or is it more of a it'll they'll be able to do it more efficiently and faster? Like tell us about the the benefits that you're trying to reach for here.

SPEAKER_00 5:26

Yeah, sure. So currently there are very expensive data sets that these teams buy, and they buy various various set versions of them. And they also have point solutions as well. So essentially what the first version will look like is collapsing and consolidating all of that data and then having a genetic layer on top where you're able to surface all of your intelligence that you can then narrate to your clients to really understand what it is that's affecting them in the world right now and how it affects their holdings and the fund they hold, and as such, what to say to them. So that's gonna help you sell more funds, but also keep redemptions down to a low degree. And then essentially in the next version, we're gonna try and bring in the agentic capabilities. So before you turn up at your desk, you can have agents conducting research, generating briefs, but then populating this data within your CRM. So you don't have to keep jumping around different data sets, different products, and you can all keep it in one space.

SPEAKER_01 6:22

Okay, no, it definitely sounds very exciting. Well, maybe we could just speak a bit more generally about innovation in the industry. So it'd be great to hear your perspective on how you perhaps see the industry changing over the coming years. You know, let's say over like a five to ten year time horizon, what can you see happening?

SPEAKER_00 6:45

Yeah, sure. So we start on that distribution point. I think the distribution side will drift more away from the relationship basis, but more intelligence driven. So I think relationship managers of 2028, 20, 2030 will uh care about not who do I know, but what do I know about them and why is it that matter now. So you can better help them invest and better help them manage their money. I think next is the agentic AI wave, not the co-pilot, not the co-pilot wave. So essentially, I think a lot of the AI that's being used right now is augmentation. So summarising documents, reviewing briefs, and like, you know, populating Excels, for example, running data analysis. I think what will shift is this agentic wave where you start to have the agents doing more of these tasks. So before you sit down in the morning, you're having specific agents with specific access to data sets going around and completing tasks before before people arrive during the day, or running autonomously during the day to conduct those tasks that you don't have time for. I think the next change will be data becomes the moat, not the product. So once firms create the infrastructure that allows their systems to talk to each other, they start to be able to harness this intelligence and it will become democratized, meaning that you don't need large research teams to augment and generate the research that was once required with a four to five-person team.

SPEAKER_01 8:12

And and that agentic wave, do you think, from your perspective, that the industry's ready for that now?

SPEAKER_00 8:19

It's a good question. Are they ready for it? I feel like the tech, you know, the capabilities and the technology is here. I think if we talk about are they ready for it? I think that has two sides to it. First is the cultural side, and secondly, the infrastructure. So we talk about the infrastructure. Essentially, you cannot use these capabilities if you don't have your data strategy and infrastructure in place. So have you got your systems talking to each other? Have you got different layers of data moving between each other that's super tight and is also accurate? Because if you try and layer AI on top of it, it will lead to equally good capabilities with good data, but it'll lead to serious problems if it's bad data. So first you need to get your infrastructure and then you can start laying on these capabilities. And then on the cultural side, I think the framework to think about it is it's going to accelerate your expertise, but not replace your judgment. So it will help you perform your job better and it will give you also more time to think about second-order effects.

SPEAKER_01 9:23

So in that case, how do you think the skill set of people working in the industry is going to have to adjust to that new situation?

SPEAKER_00 9:32

Yeah. Yeah, that's a good question. So I think essentially people will have to have a good understanding of what data is available, an understanding of the investment management landscape and the strategies and the products, but also what is the makeup of these tools? Now, you don't necessarily have to be technical and be able to code, but having an understanding of how do they work so you can use them to your advantage. I think the downfalls of that, if you don't know how they work, you might set up agents, for example, and you won't realize that they can hallucinate and drift. So I think having an understanding of the makeup of them and then of the data and the products will really help you take advantage of them.

SPEAKER_01 10:11

Okay. Okay, so let's picture this. We've got this uh agentic wave coming towards us, uh, but these potential breakwaters in the form of architecture on one side and then culture on the other, as you say. So what do you think needs to happen to spur the industry forward to really sort of seize and maximize the potential benefits that are ahead of us?

SPEAKER_00 10:35

Yeah, I think essentially it comes from the top down. You have to have the senior leaders within your organization pushing this forward, but also at the start, addressing your AI strategy. Within a lot of large enterprises, each desk and division essentially operates as its own company. It has its own budget, it has its own tools, it has its own data warehouse in some instances as well. So, first coming from the top down, addressing what the AI strategy is. Like, how do we intend to use AI? What are the use cases? So we focusing first on investment research, or are we looking into operations and then how can we roll that out across different desks, as opposed to having the siloed budgets and areas whether we're using different tools? Because essentially that leads to various generated data that doesn't have any auditability or transparency. I think that's when you get into some real messes.

SPEAKER_01 11:26

Okay, so maybe switching to a slightly different topic. Something that's been discussed extensively recently is this issue of data sovereignty. It'd be great to get your take on that, please.

SPEAKER_00 11:38

Yeah, sure. So I think the I think the framing has to be thought about. It's not necessarily a compliance question, but more of a strategical advantage. So it's not just where is our data stored, but it's who has access to it and how are we going to make it produce results for us year after year. Investment managers especially have access to some really powerful data, so like client behaviour, uh, fund flows, portfolio holdings. And I think right now people don't understand all of the data that they have access to. And I think if they treat data sovereignty purely as a legal question, they may wake up in five years and realize that their data's been commoditized by someone else. The other side of data sovereignty is on fintechs and third parties being allowed in. I think organizations need to create pathways for fintechs to get in quicker. Now that's not lowering the security bar, but pathways can be virtual private clouds, so they exist within your four walls, or that could be synthetic environments where the data or your data holds the same statistical properties as the real data, so you can prove a product's value, or federated learning, where essentially you can pull on data from disparate data sources, and then means the underlying data isn't exposed and it's not centralized. So you still have that core security.

SPEAKER_01 13:05

Excellent, excellent. Okay, and and of course, uh Regulex was selected onto the IA Engine Sparks program recently. Can we hear about how that experience has been so far?

SPEAKER_00 13:18

Yeah, definitely. Sparks has been great. If you're trying to build a product in this space, one of the hardest things is getting in front of people. The kind of people and decision the decision makers you need to speak to aren't going to reply to a cold message on LinkedIn. So having those pathways to really get in front of them is really, really important. And essentially, Sparks gives that to you. So you get access to there's a panel of 30 to 40 people, and essentially they're the C-suite of the investment managers, but also kind of heads of AI, analytics, and data. So you can get in front of them, you can pitch your idea, you can ratify it, maybe try and even get some proof of concepts and really get your product driving in the right direction.

SPEAKER_01 13:59

Excellent. And maybe you could share with us something that you've learned in the past year that you wish you knew one year ago, just based on all of your experience and uh, you know, what you've gone through launching your business from from the ground up.

SPEAKER_00 14:14

Yeah, sure. I think don't lead with the technology. Like really understand what it what is the the precise problem you're trying to solve and then be able to articulate that and then showing the technology that you're gonna be using, not the other way around. And the second point also is having a real understanding of who you're actually pitching to. You know, the same pitch isn't gonna work for everyone. Some people have more technical abilities than others. Don't just presume that people aren't gonna understand it because you've been looking at the same content for a few months now.

SPEAKER_01 14:46

Sure. Excellent, excellent. And do you feel like you're you're gaining you know good traction at the moment with the with the new venture?

SPEAKER_00 14:52

Yeah, definitely. I think Sparks really slingshot it forward. Antler's really helping right now. They're two incredibly good names to have, and we're we're being able to get in front of some great people. We've had two proof of concepts already, and now we're getting quite close to some design partnerships for the uh new product. So, yeah, there's a lot more work to do, and there's gonna be some hurdles that come up. But for now, yeah, I'm happy.

SPEAKER_01 15:17

Fantastic, fantastic. And what haven't I asked you that I should have asked you that we need to talk about? It's a hard question.

SPEAKER_00 15:24

Maybe the UK's position in all of this. Fantastic.

SPEAKER_01 15:27

Tell us about the UK's position in all of this.

SPEAKER_00 15:30

I don't know, but okay, I need to think of an answer now. Um the UK, okay. So recently the UK has been opening up its doors to the likes of Anthropic and OpenAI. Really recently, I think Anthropic's just got a new office. So the ability to get those companies as organizations and those capabilities into the economy is great. I think culturally, there's definitely still quite a lot of skepticism regarding AI and culturally, I think, especially in the UK, that can lead to stagnation. So I think, again to your question earlier, how you know how do organizations actually drive these capabilities forward is taking more risks. In America, for example, they are very focused on getting these capabilities into organizations and allowing people to fail and to break things and for people's heads not to be on the chopping board if they get some third party in and it uh doesn't work out. I know the industry uh regarding the data as quite severe if something does go wrong. So you need those guardrails. But I think, yeah, definitely that cultural shift of allowing these capabilities to come inside organizations and testing them out. Sure.

SPEAKER_01 16:39

But because you do start to hear more, especially nowadays, companies perhaps you know, coming away from the idea of, okay, let's just, you know, try things out, let's experiment. And the discussion becomes more about, okay, well, what's the return on investment? Can we measure the time saving? Can we measure the efficiency gains? So are you almost suggesting that maybe that's slightly premature and we need to, you know, embrace the discovery phase for a bit longer at least?

SPEAKER_00 17:08

I think from what I've experienced and the founders I've been working with is that organizations sometimes don't necessarily know what those use cases are. They want to get these capabilities in because they want to start using them, but maybe aren't exactly sure in some cases how they work and what they're going to be used for. And I see a lot of companies falling into POC paralysis. So they get them in and then it doesn't really go into production because I don't think that the AI strategy has uh been set in stone. So I think that once that is figured out, there are specific use cases that apply across the whole firm. It's not just sided off to different desks and departments, then we can better set ROIs and KPIs for these products because then we know what the expected outcome is.

SPEAKER_01 17:56

Okay, no, it definitely makes sense. Well, maybe we can finish with just a couple of quick questions. So, first of all, what's one area of AI that you believe is overhyped?

SPEAKER_00 18:09

The copilot wave is completely overhyped. I think it's within most asset and wealth managers, and yeah, I'm not really a fan of co-pilot. I think we need to get away from this augmentation and really start to get in the real like capabilities of a genetic AI and conducting those autonomous tasks.

SPEAKER_01 18:28

Okay. And one area of AI that's probably underappreciated, underhyped, that could do with a bit more of a pop?

SPEAKER_00 18:36

It's definitely gaining a lot of traction now, but definitely voice, uh voice agents. Uh, voice AI, the ability to interact with an AI within a room, say, I think has serious capabilities, especially in this industry. And I don't think it's it's barely being used really.

SPEAKER_01 18:53

So what does that mean? Like Alexa for investment managers?

SPEAKER_00 18:57

Exactly. Something to like ratify your ideas, store your assumptions, uh, listen to your conversations, and essentially it's that aggregation of those logs and those conversations that can then help you better remember what you were talking about, tying it in with real data and real time and having something actually converse of you instead of you having to like prompt it into a product.

SPEAKER_01 19:17

Okay, okay. And m maybe that's gonna be something you build in the future then.

SPEAKER_00 19:21

Yeah, phase two.

SPEAKER_01 19:22

Phase two. Okay. Well, I think that's a great point on which to end. So, Krim, thank you so much for for coming in. Really appreciate your time. And thank you very much to everyone for listening. I hope you enjoyed that. Please do join us again as we'll have more episodes on the way. Thank you.