S11:E06 AI In RegTech: Better Reporting, RegOps and Automation

Join us for this special episode with Pierre Khemdoudi, CEO of Gentek AI. Together, we frame the big picture of how AI is changing - and improving -  regulatory reporting, supervision and data quality. We're going in-depth on how to deploy automation in your own RegOps, how to change the way you collaborate with counterparties, and how to remove friction points along the way. Don't miss this hands-on, hype-free session with Pierre Khemdoudi, a pioneer of automated reporting at ISH Markit and S&P.

Chapters

• 0:00 Meet The Crew And Guest
• 3:00 Framing AI For Regulatory Reporting
• 6:10 From Episodic Checks To Continuous Oversight
• 10:30 Regulators’ Cautious Embrace Of AI
• 13:40 Product Teams Adopt AI First
• 16:50 Old Rule-Based Stacks Vs New Agents
• 21:00 No-Code Validation And Context Memory
• 24:15 Fixing Recs: Learning From Acceptable Breaks
• 28:00 Supervisors Investing In AI Analytics
• 30:05 Pre-Submission Controls And Operating Model
• 33:20 Accountability, Dora, And Human Oversight
• 37:00 Transparent Agentic Workflows And Traceability
• 40:00 A Day In The Life With AI RegOps

Transcript

SPEAKER_03: 0:04

And welcome back to the Registrum Roundup. Yes, we are back and in the hot seat. It's February, and we have got a great episode for you today because we're talking all about AI. And not in vague abstract terms, but actually hands-on how AI is uh developing new tools in RegTech. And joining us for that, we have uh the Red Hot new AI startup. Well, it did launch last year, but Red Hot AI startup with an old friend of the show, Pierre Chem Doody, who you will remember, of course, back in 2021 from season three, episode 10, and episode 11, when we did a special on SFTLR. Yes, you remember SFTMR? Can you can anyone else remember SFTR? That was the hottest news. Then I'm really glad it's AI now, because that's way cooler. Okay, so we are in. Now, before we introduce PNET properly, of course, uh we are here with the virtual studio crews, and I'm delighted to welcome a new face to the show, uh, the uh very handsome. I'm gonna throw that one in there for those of you who haven't seen in the studio, the very handsome and good-looking James Bond of the show, our very own head of product management, David Inglesias. David, welcome aboard. Thank you. And with David, we have two old hands. I say that. It's that's the terrible way to introduce uh next guest who, like me, makes a white beard look young again. It is, of course, Mr. John Cunan, the man who puts the canary in the wharf and he looks after some Mary's axe, head of ResDR in the UK. John, welcome back.

SPEAKER_04: 1:34

Thanks, Andrew. It's great to be here. Looking forward to the discussion today.

SPEAKER_03: 1:38

I'm bringing some uh you know much-needed uh serenity and uh high brown academic insight uh to the virtual studio crew. It is, of course, the Pride of Spain herself, the head of institutional relations, Laura Rodriguez. Laura, good to have you back.

SPEAKER_02: 1:52

Hi, Andrew. Thank you very much. Happy to be here.

SPEAKER_03: 1:55

Now, as I said, we have Pierre Kim Dudie with us. Pierre, of course, you will remember from the show, but if you don't, then you've probably met him at some point. If you work in the world of securities, finance derivatives, repo, reporting. He has been around doing lots of interesting things. There you have a stint for about 10 years uh in securities finance at BMP Parabuff. And then Pierre joined Market, which became IHS Market. You will, of course, know them for a stint where he pioneered the development of their regulatory reporting product line there. And he is now the uh, of course, IHS went to SP uh with Pierre as well. And he's now struck out with his own new AI startup, Gentech AI, where he's the CEO. Pierre, welcome to the show.

SPEAKER_01: 2:43

Thank you, Andrew. Thank you. Thank you, the Reduce Team, for having me. You know, delighted to spend uh the next uh half an hour or so, you know, discussing about you know AI and the great things, you know, that are um ahead of the regulatory reporting market.

SPEAKER_03: 2:58

It's a really good time, actually, to get you on the show because regulatory reporting is entering a new phase. I suppose everything's entering a new phase now in the world of financial services. As everyone is focused on automation and AI tools to match participants in GC pooling. That's happening. Uh we know that uh they're automating processes right the way across the cycle from the front office right the way through to the back. And uh we're seeing increasing use of AI to manage transactions and uh manage settlement efficiency, and those kinds of applications are only gonna keep on coming. And of course, regulatory reporting is a huge part of that. Uh now we know that firms are also experimenting with AI native and agentic workflows. We'll come to those in a second. Uh, and that's gonna fundamentally shift the way that uh reporting uh is monitored and controlled across Amir, Mifir, SFTR. Yes, it's back. Um, possibly an SFTR refit on the horizon. We might touch on that if we have time. And so, Pierre, having you here is great, but there are so many buzzwords, there's so many news stories, it's such a big topic. Can you frame the context of AI when it comes to uh supervision, risk-based supervision, data-driven oversight using automated tools? ESM has been very clear in reach and in its re ESMA has been very clear in its recent principles on risk-based supervision that regulators are moving towards continuous data-driven oversight. Now, this is super exciting if you're a bit of a nerd like me. But what does it mean, hands-on, for firms? Is real-time reporting just around the corner with AI uh and with Gentech, presumably?

SPEAKER_01: 4:50

Um yes and no. And I think um I'll start with um let me take a few steps back. We talk about AI, as as you said, you know, it's a buzzword and everybody's talking about it. But I would like to come back is what are we talking about? You know, what are we talking, you know, when we talk about AI, you know, in the regulatory reporting uh world, but more importantly, you know, what are the names, you know, and put like uh labels on names. So you have three main categories LLMs, large language model, that's the brain, you know, that's the reasoning engine. They are very good at interpreting text, documentation, and instruction information. That's your chat GPT that you use every day and so on and so forth. Then you have agents. Agents, they are the workers. They are actually, you know, brains to which you give arms, tools, and everything like that, guardrails, and so on and so forth. So they execute tasks. They actually change the move from reasoning to execution. And then, as you mentioned, you have agentic workflows. It's the orchestration. That's the framework, the AI operational framework that actually coordinates the work of agent. And these three steps are very important because we tend to mash them all together, but they do very distinct, they are very distinctive and they're very important to segregate, so to identify them. Now, coming back to your question, when it comes to um uh risk-based supervision and the impact of AI and so on and so forth, what really changes is the tempo of supervision. You know, we move from um um, I think we are moving from episodic, you know, after-the-fact reviews to something much closer to continuous oversight. You talk about real-time reporting. I think it's more um, I wouldn't use real-time, but much more um at the moment, continuous, you know, control framework and much more efficient control framework. Um, completeness, timeliness, and accuracy that you hear over and over, you know, when it comes to control frameworks, you know, uh, they are no longer the destination. They used to be. Now they're just you know the starting point, essentially. That's the entry criteria, you know, in a in a modern control of AI-driven control frameworks. Um, if you hear about a little bit about, you know, you have on this show, you know, discuss about the latest um um um kind of um um input from the regulators, you hear that they're not saying really um, you know, they're not trying to figure out whether firms have reported correctly, they're trying to figure out the quality of the control framework. Can you prove at any moment, you know, that you know, the number that came into existence, you know, you have the full traceability of that and you understand where it's coming from. Okay. That's a big shift when it comes to AI-enabled reporting. It means AI, you know, is not a black box, you know, cannot be a black box, um, and it cannot be just bold-on. You know, there is a proper, you know, um um um process that needs to be put in place in order to have observable, explainable, and defensible, you know, and by design, you know, AI system.

SPEAKER_03: 7:32

But Lara, I want to come to you here because of course you're the expert from the institutional relations side. Um, so you you'll have some idea, of course. How are regulators approaching AI and getting market participants to engage with solutions like Gentech, um, which may be one of the first, but we can probably predict there will be more AI-enabled reporting tools for different applications emerging into the marketplace over the next couple of years. Uh, what's what's the view of ESMA on uh and more broadly regulators when it comes to AI? Are they ready to embrace it?

SPEAKER_02: 8:11

Very good question. Um and I would say yes, but they are being cautious at the same time uh because they see very clearly the opportunities that the AI have, but at the same time, they also see the risks that this can play. And uh Pierre was explaining them uh, you know, all very well. Um and they need to be ready, they need to have like a continuous upskilling uh on the the people that is working uh there to ensure that they have the capabilities to uh cover and to identify all the risks that are coming. Um and again, I think that one of the important keys will be cooperation. Uh it is they also need to learn how this goes, how this works and how the supervised entities are working with AI. So, for example, they have uh from the European Commission or and also coming from ESMA, they have been uh sending the industry questionnaires about AI, how we uh what is our relationship with the AI, how much uh um uh money uh entities are investing on this, um uh what frameworks we have. So I think they are in the in that process of learning because they absolutely see this as an opportunity, but they need to take into account the risks. And um so so yeah, I think it's uh a positive uh approach for the moment.

SPEAKER_04: 9:40

Is it not also a case of you know the proof of the puddings in the eating? And and I guess over the over the last 12 years, what we see here from from my perspective is potentially one of the biggest drivers for improved data quality. Uh, and I think you know that that ultimately will of course be uh hugely welcomed.

SPEAKER_03: 10:02

David, I'm gonna also bring you in here. Um is AI making an impact uh on your desk? I mean, there can't be a product manager working in financial services anywhere in the world right now who isn't looking at automation and smart workflows and those similar kinds of applications that the the Gentech are solving with the new Iris product. I mean, how what's your view on AI? When's it gonna start uh automating your day?

SPEAKER_00: 10:33

One of the team requirements that we said uh at the beginning of the year is that each of the user stories that we are creating, transforming the regulatory requirements into technical requirements, should be tried firstly to do with AI. And then there is uh the product team who review the content and trying to enrich that content. But for forcing the team to do that, we are trying to gain time, do until which point we can automize the different tasks that we are doing.

SPEAKER_03: 11:07

So it is right now in the day-to-day great time actually to come back to Pierre. You were involved in creating some of the early large-scale uh solutions for reporting S SFTR and MIR data. Uh, and that was a very sort of rule-based environment. It was, you know, driven by utility. And, you know, it was still in the days, of course, where market participants, some of them even still use spreadsheets to get that sort of data across. It was pre-the implementation uh of XML standards universally across the industry. It seems like those it's so long ago, it sounds almost like the days when your phone was nailed to the wall and didn't have a screen attached to it, right? It's almost unthinkable. But everyone, everyone's kind of upgraded their systems now. How big is the step to AI? Is it just coming up with a layer that speeds up that sort of rule-based utility-driven automation? Or does it involve uh a complete sort of ground-up rethink of how data flows from the systems internally of the market participant through to the trade repository?

SPEAKER_01: 12:20

Very good question. So, first, you know, I do have a lot of gray hair and I'm definitely losing my hair, but it was not that long time ago. You know, we had XML standard. Actually, SFTR was, I think, the first one to use ISO 2022 to standard. But uh, but having said that, you're correct, you know. With uh I was involved, I was actually experiencing, you know, the the starting with HSC SFTR, and then after you know broading up to the G20 regimes, you know, the the build of the transaction reporting. It actually, matter of fact, you know, some of my uh team as you know is part of the Gen Tech story as well, you know, from back then. Um but when you come to SFTR, and in particular SFTR, you know, um it was built from scratch, you know, and we kind of leverage um what was at the time you know leading edge cloud and compute, you know, um technology. Um and I would call that what we build the second generation. You know, the first generation was mostly deployed technology, you know, um a lot of TNM, not really scalable, but it was kind of starting to industrialize the reporting. Second generation is what we build, the kind of the SaaS offering, you know, highly scalable, mutualized solution, you know, um, that were very focused on delivering the data fast and validated to the end point. You know, that was really the main goal, is, you know, um and in a and in a cost-efficient way. Um and I think you know these these are very strong solutions to start with, you know, in their world. But you're absolutely right. You know, there is a strong, you know, role-based um um and hard coding part in that tech stack, not just because, you know, because that's the tool that we had at the time, and we did, you know, try to uh improve on that, but the reality is is is there, you know, it's hard-coded. So when um you know um the logic is embedded directly into the system, you know, validation lifecycle, you know, handling, reconciliation checks, schema interpretation. And generally it works well in stable situation in stable conditions, but unfortunately, regulatory is a little bit is not stable, you know, and and they make it brittle. You know, every regulatory change requires quite a lot of involvement in terms of upgrading the systems, the data fields, and then and so on and so forth. Uh you ask about you know, uh, what do we uh what is the versatile existing, you know, how do we position ourselves? Um uh uh the other equally I want to bring another point. These platforms were optimized for surface to surface exceptions. So there is in all these you know vendors in that space middle where there's a lot of validation, you know, a lot of exceptions that are being thrown out, you know, in order to ensure that the data is from quality to the endpoint, you know. And that's another challenge, you know, um, um, in the process. You talk about IRIS. So let me talk about, you know, this is our product, you know, this is what we call it's it's iris stand for intelligent reporting and integrator integration um uh solutions. Uh what we do is we we're fundamentally different, no different. We uh iris is our you know, agency control framework um that investigates, reconciles, and resolves regulatory reporting exception in real time with full transparency and human oversight. So iris is very output-based, you know, not rule-driven. Um the objective is clear. You know, we understand the exception, we assess the regulatory relevance, depending on the regimes and everything like that, and resolve it correctly and process defensible evidence. Evidence is always very important. Um, the rules where before we were coding XSDs and validation, you know, they are not coded, they are part of the context. You know, I talk about agent having a memory and everything like that. That's the context. We feed that context. We don't hard code. And we combine the outcome to the agent and to the workflows, you know, in a lot of what they need to deliver. Each step needs to deliver, and each agent has a framework to hand over the tax to the to the task to the next agent. So, in some ways, if you think about change, if you think about change in terms of new regimes, you think about change in terms within a regime, but refit or rework or something like that, what we change is not we don't record everything, we actually change the context. So we bring new memory and we say, before good look like this, now good look like that. You know, so that's much faster. It's actually much less brittle as well. It got a complete understanding of the regulatory, the laws, the best practices, the intrinsic of you know each firm, and so on and so forth, and can adjust and adapt much faster than traditional system. Sorry for the long, long, long-winded explanation.

SPEAKER_04: 16:48

Would that be ingesting, for example, TR feedback reports and things like that?

SPEAKER_01: 16:53

We do ingest that, but uh, of course, you know, we do ingest you know um exception for many layers. But I think what's more extraordinary, and I think you know, I was showing that to a client yesterday, is what's extraordinary, is for take this SFTR and M here big spreadsheet, you know, validations where you have all these things and everything. We don't code that. We actually load the spreadsheet and we have created an interpreter of the spreadsheet for the agent to actually navigate the spreadsheet. So when we say validate a message, it looks at the validation rule and navigate the whole thing by itself and figure out what needs to what could look like based on the invalidation. No code. If tomorrow we change the spreadsheet, you know, or we bring another regime with a different format, it could be a PDF or anything like that. As long as we point and say that is the knowledge for this, you know, the agent will actually interpret that and use that as a as a tool, which is uh I'm sure you can sense that it is, I mean, for me, it's extraordinary because I remember, you know, about at the time when we coded, you know, all the schema validation, you know, it's thousands and thousands of rules that you actually can get, you know, implemented in a matter of an hour or so, you know, and it's it's it's it's revolutionary.

SPEAKER_03: 18:02

Um what about okay, and it's it's sort of elephant in the room time. Uh a big issue isn't always, of course, your own reporting, it's your counterparty's reporting and the the mismatches in in reconciliation. And obviously, it's those things like pricing differences or life cycle mismatches, missing fields, late reporting. Um you know, there's a raft of issues that Esmer highlights in their data quality reports. How does the AI tool within one firm uh improve reconciliation rates? Because obviously if you're automating your reporting, it's coming in earlier, it's coming in faster. There must be a lot of ways that then you can improve reconciliation off the back of having higher data quality.

SPEAKER_01: 18:49

Um so there are a couple of ways. So first is um um, and I'll maybe I'll come back to that after I answer your question. I think I think that you're right. There is a there is a long list of challenges, and I think reconciliation in particular is is is a is a hard topic for sure, you know. Um so um think about it this way is um um I'm I'm I'm I'm kind of forcing my point, you know, I mean in some ways, you know. In some ways, reconciliation is always thought you know in a bit of binary, you know, maxed or match in some ways, you know. Um when you have an agent process, when you have an agency process in the middle that actually sees this information over and over, sees the resolution, you know, whether it's manual or automated, you know, over time, it learns through it. And most a lot of the challenges that come from reconciliation is not that we don't we disagree with the data. It's actually the data, there is timeliness, you know, issues. You know, the data doesn't reflect in the system at the same time. So there is a little bit of a delay there. You know, you have interpretation, you know, people book, you know, interpretive rules or or or or or the way they book transactions in a different way, you know. Um enrichment logic comes from two different sources, you know, and so on and so forth. You know, there are many um um many challenges where the the answer, and that's the complexity. It's not like I'm right and you're wrong. You know, that's not the outcome. It's much more subtle than that. Now, think about you know, having a network of agents that actually have memory about all the stuff that is happening, context about the subtilities, the subtle changes between that counterpart and that counterpart, knowledge about best practices, you know, and everything and that dictated by industry association, and are able to, at the point of decision, you know, when they look at the at the reconciliation, at the reconciliation break, and say, oh, I know that, you know, I know these guys do this way. We've done that, you know, 25 times before, you know, and I know because of this rule and I know because of these best practices, these guys, you know, are more in that category. It's not a break. It's actually an acceptable break. We shouldn't spend time fixing it. It's not a right or wrong, it's a different way of viewing the world, essentially. And I'm not saying we can solve everything, but you can actually, by bringing that context, you can actually reduce the burden quite extensively, you know, um, on the process. It's a path. It's not like you plug it and all problems solved. It's a path to get there. But it's something that you built upon time, you know, and upon time, you will see kind of these brags that are not really breaks, you know, being handled in automatically and therefore, and and therefore you will human, you know, reg-ups will only be exposed to really the thing that matters. The noise will be handled automatically.

SPEAKER_03: 21:35

It's a very interesting idea, actually, to think that learning patterns of predictable errors due to differences in the way the data is filed will in fact streamline that process massively. Rather than it just failing or not uh to reconcile, you'll actually have a tool to unlock why it's not reconciling and then actually work with your counterparties to to get your data more. Effectively aligned. On that front, Laura, that is a very exciting development. What about at the regulatory end? Are the regulators actually developing their own AI tools to improve things like reconciliation and what have you? Or is it considered that that's all something that has to be done by market participants?

SPEAKER_02: 22:20

Not at all. I truly believe that authorities should also take this step and go forward to include AI on their supervisory analytics to improve the way they analyze the data. I mean, from the the data strategy report that they that they envisage to have from the since this year that was updated until 2028, we see that this is the way forward. They are going through that path of investing and taking the opportunity to use the AI for their supervisory activities. So I really expect that there will be in a future, uh not very far away, um a shared tools, you know, methodologies and platforms that we can see from the European authorities or all authorities, the NCAs, where uh we can see how they are using the AF the AI for their um uh supervisory activities. It would be very interesting.

SPEAKER_04: 24:01

I was just wondering, we've been talking about in intelligent exception management, and I'm I'm I'm wondering to what extent as well this this can be applied uh in terms of pre-submission controls. So before it even anticipating it before it even gets to that stage.

SPEAKER_01: 24:21

I mean, we totally we work with uh some clients on that front. Um I usually say the the following is you know, what before you run, you know. Um I think is uh when you implement um um um automatic resolution, it's as if you were multiplying your your Regobs teams by 10, 50, right? You know, so it has an operational model impact. You know, it's not just a technology implementation, this is an operational shift, right? You know, so um when you do that on the pre-reporting with the pressure of reporting at any cost within the time window, it it creates a lot of you know uh constraint as a first implementation. So what we always say is let's work on the post. It might not be the most significant return on investment to start with, but you actually get you know um you get your operational model right. And then after moving to pre, there is nothing on our side and within you know um our network, you know, that that that can that that prevents you know from the pre. Apart from the fact that everything goes much faster and the pressure is much faster to get it right. So we're fast always, but you know, can the team cope with that, you know, to start with? Better to as a second step. But but totally. And and by the way, uh John, you you we do we do we cover obviously the regulatory port in vertical, but we do other vertical, you know, in other financial markets, in other markets, in the community markets, and everything and that everybody faces the same challenges. You know, we convey information fast, we validate information fast, but we still have 100% or 99.9% of humans facing with things that goes when give when things go sideways, you know, that are facing this exception. It's what we are doing is in the reg side, but as well as as well, uh everybody, everywhere else. Pre-post reporting settlement, you know, uh, or even in commodity markets is is about is about giving a chance for people to actually have a control, you know, um uh a good control framework, you know, um and have good visibility about you know um what goes what is being handled automatically, but as well, you know, what they need to pay to spend their time on and and cover 100% of the challenges and not just you know 75.

SPEAKER_03: 26:42

Well, this is one actually that sort of takes us towards another sort of interesting aspect. Because a lot of things to do with AI and and more broadly with new technologies uh hits the same speed bump. And, you know, to put it very bluntly, both the FCA and ESMA have been very explicit that, you know, it doesn't matter if you're using the technology, the firm remains fully accountable for what happens with AI or with third-party supplied applications. And we've seen, of course, with Dora, there's been a huge focus on improving robustness of services and uh making sure that there's resilience in the system, but ultimately the buck has to stop somewhere, and that is with the governance team at the market participant. David, how do you approach that sort of governance? Presumably there's always a human with the hand on the steering wheel.

SPEAKER_00: 27:34

And the accountability is is is very important because in the end, uh why why companies are reporting? They are reporting not because they want to report, they report because it's not as much NCAs will come and put a fine on the table. And this fine goes addressed to specific people. So companies are right now a bit conscious about AI in terms of accountability. Because when you have a problem in the old systems or the system that companies have been using for regulatory reporting on the past years, they have a way of argument of giving an argument why this data has been produced like that. Why what interpretation we have done in order to produce the data like that? When we speak about AI, we see very well the results that is produced, but we don't see very well which is behind AI. Or this means that to give argument to the stereo is more complicated. We use AI in order to produce our code, but after the code is produced, this is validated right now by human, and we have not arrived to the step that AI substitutes the human. So they have in order to accelerate the task, but a human should be there to validate and have accountability of what they are doing.

SPEAKER_04: 29:02

I mean I suppose one of one of the key points here, isn't it, is coming coming back to that kind of immutable evidence concept and having you know having demonstrable rationale uh for what you've done, which then ultimately needs to be approved by a human, right?

SPEAKER_01: 29:21

This is why I made you know the distinction at the beginning, LLM agents and World Energy Workflow. By design, you know, agents have a full traceability of their chain of thoughts, action, and everything. By design. It's not us who built it, it's by design, it's there for the taking. So if you want to equate the work of an agent to a human, you know, um um uh Regops, for example, you see as a system what has been the impact, the outcome. So what you talk about, you know, David, you see what what what the the outcome that is being looked for, you know, whether it's uh surfacing intelligence or whether it's modifying your code or anything like that. Okay. Um through an LLM you can see that. Through an agent, and you see not only that, but you see as well, you know, um the regular the regulatory context it applied, you know, the option it considered, you know, the action it took, and it's as if you were plugged into the brain of someone and says, No, I've seen my system what you have done. But with an agent, you see what was built a thought process to actually come to the conclusion. So if I were to argue, if the system is built correctly, agentique workflows, I'm talking agentique foreclosure, I'm not talking about cursor, I'm not talking about, you know, this kind of surface level tool, agentic workflows give you a better transparency that you can ever have at the moment, you know. And not because we're clever, it's by design, you know, it's there for the taking, you know. So if it's built correctly, you can see so much there. You can go to the trace, the full traceability all the way to the text, but understand what is the plan. You know, I make that decision because I saw in that text paragraph 2.5 through three, and this is there. And then after I look at the validation in that document, validation 325 code, you know, whatever, this is what I did. You have the full traceability. For me, it's remarkable. And this is really what we tell our client is is when we have that type of conversation, we say, look, we'll show you and we'll show them you know the whole thing. And the transparency is like never, never before, never seen before in systems.

SPEAKER_03: 31:27

Pierre, Gentech's written about the way AI is a sort of horizontal infrastructure that gets embedded right the way across the firm, front to back, from the sales side right the way through to uh back office. And it's not a standalone reporting tool. This isn't working in a vertical, it's a much smarter workflow that can actually join up, sync up your business in true sort of RegOps style. Um so give us some ideas about what's that gonna be like? What's a day in the life of an Amir or Mifir reporting team going to look like? How's it gonna change, more importantly, I guess, when uh AI tools get used industry-wide?

SPEAKER_01: 32:09

The biggest change, you know, is uh I usually that's we use that phrase entirely. So uh but the biggest change in in that Regobs, you know, team life, you know, is essentially they're gonna start the day, you know, um where they used to end it. Um what I mean is uh it's it's it's a bold statement, right? You know, but follow me. You know, uh historically teams would spend you know the day fighting volumes, you know. That's their that's their stuff, you know, chasing breaks, reconciling files, gathering evidence, you know, and and really only at the end of the day, you know, uh they would get to understand what actually what actually mattered. You know, we have we had clients who talk to us, you know, but that's you know, uh they won't 100% you know, you know, so their day is like you know, kind of wakamo type of situation, and I could appear, you know, and everything like that. So so that's that's you know, that's that's we we agency system, you know, that that flips essentially because uh if you think about it, when Wake Up's teams, you know, log in in the morning, the agents would have already dealt with that noise, you know, that volume essentially. They would have investigated in the exception and results on, and when they wouldn't have resulted, they would have said, look, this is what I found, you know, I need your input for that, you know. Um I have an ambiguous, you know, kind of um regulatory interpretation where I still, you know, I need you know an SME point of view. Can you help me out figure it out? And it's not like I I dump onto your desk, you know, the the workload and then you deal with it. Is I need your input. But then once I get the input, when I get the ambiguous part you know clarified, I can redo my work. So if I need to modify an XML and resubmit it automatically, I can do that right away. So it's not just you know, you get 50% of the of the of the exception solved and you get 50%. You know, it's no no. Look at everything, I facilitate your work, you know, on a daily basis, and I need whenever I need, I get your input. And then after I simulate your input into the whole workflow. So the game is not like it's pretty six is pretty significant, you know. Um to me, you know, if I were to summarize, you know, is that means that Recdop's team, you know, start the day in position of insight and not backlog the way they are, you know, the way they are doing it right now. You know, it's it's really, and as I said before, it's not just you know an efficiency play, it's really a mindset shift essentially.

SPEAKER_03: 34:26

Uh, do you think the market is sort of looking to companies like Gentech and products like Iris uh to start reducing the cost of reporting and also the friction it causes? And could that be the answer that you know will actually allow different products to be reported in different ways? So we might see actually an increase in the number of fields that get reported and new things added into regimes, but it'll be less of a burden because of automation.

SPEAKER_01: 34:58

Uh okay. Um there are a couple of underlying you know topics in into your question is first is um, and I'm gonna answer in a pragmatic slash cynical way, you know. Um uh first is the call for evidence, as you said, five, seven years down the road, you know. Uh uh it's a long time. In my uh in my uh lifespan, you know, uh in my uh myopic view as a as a startup CEO, it's it's a long, long time, you know. So it's uh it's more of a philosophical question, you know, than anything else for me. Having said that, if I want to be truly cynical and pragmatic, simplification will be a change, and a change will need to be absorbed and changes work, you know. So so you might end up from double-sided to single-sided, you know, but there will be changes, you know, here and there. So they will, the industry will have to adapt there. I don't think, I'm not too sure. It doesn't feel like double-sided will be, even though asked by the industry will be dropped, it feels that it's it's pretty cornerstone part of the reporting, so that will still be there. But a change is a change, even a simplification. So that's a lot of work, you know, anyway. Um let me answer to your question a little bit differently. What if I were picturing a little bit, you know, what futures would look like? A couple of years down the road, people actually embrace, you know, AI automation when it comes to exception management, when it comes, whether it's our system or somebody else, you know, um, um, um, um, what it what it would look like. It will look like first, you know, you will have a continuous, you know, an intelligent control. This is what regulators ask, you know, continually monitoring, not just, you know, um monitoring some and remediating others, you know, at the end of the, at the end of, you know, the month or the quarter, you know, through very expensive remediation work. Um, efficiency will definitely improve, uh, but not in the simplistic sense of you know, fewer people doing more work. You know, the real gain will come from um eliminating delay, you know, exception will be identifying, investigated, and resolving near real time. We talk about that at the beginning. You know, I truly believe that you know um that control that real-time control framework will be there. And it's gonna be big, it's gonna be needed no matter what, you know, for reg reporting, for post-trait, for to support T plus one, T plus zero. I don't know. You know, like you know, there is a lot, there is there is a there is a push towards efficiency and speed. Uh transparency will be you know um much bigger because you know systems will have embedded built-in transparency, which is not necessarily the case right now. Um the risk management will be more predictive, you know, uh and won't be faster. But um, but at the same time, let me be clear as well, you know, because I can see, I can hear the detractors, you know, gonna say, yeah, AI, AI, but still human has a very big part to play. Absolutely. You know, um, there are clear boundaries between AI and human, you know, full autonomy, fully autonomous, you know, interpretation of ambiguous, you know, regulatory intent, you know, we're we're far from there. You know, human needs to be there. You know, what we can do is we can enhance the processes to make the decision where there is a human decision points, and there are many, to make them you know fast, to make them efficient, and to make them transferable to the system very quickly. So um so to me, you know, there is there is a lot of you know, a lot to be excited about. You know, there is transformation as well. I think embracing the technology is important, but as well doing it in a way open eyes uh is important. Education is super important. Um and I think generally speaking, um the rights that our you know RegOps clients, you know, for Regis or for ourselves, you know, are you know needs to be sorted because that's the key to that's the key to quality reporting, that's the key to cost management, you know, that's the key to to have you know good data and good reporting and have team, you know, that are that feel that they are, you know, they are not you know just you know, as I said, you know, um working them all, you know, on a daily basis, but actually, you know, truly um um um improving and truly you know monitoring you know the the reporting of the organization.

SPEAKER_03: 38:59

Okay, that is sadly we are out of time for this episode. Yes, the pace of change doesn't uh slow down, and neither does the pace, the workload or the diary uh for the CEO of a hot new AI startup. You probably could have guessed that. Actually, so I am going to end here with uh Pierre for now. But I'm gonna say, Pierre, you will come back on the show this time next year so we can have an update on what's changed, which of course will probably be everything as AI rolls out.

SPEAKER_01: 39:29

Thanks, Andrew. Thanks, uh, thanks, Brian, thanks Lauren, thanks, Davinino, for for having me uh fantastic discussions.

SPEAKER_03: 39:36

Thank you, Pierre Chem Duty from Gentech AI, and do go and check out their website and the new product IRS. And the new product IS, because it is absolutely 100% all the things I've been predicting uh for uh the last five years and has never arrived. Finally. Finally, it's here. Uh that's great. And I say that as someone who always makes the same joke, and I think you might get this one, Pierre, is that you know you're getting old if you can remember going to a conference where they weren't talking about DLT and blockchain.

SPEAKER_01: 40:07

Very fun indeed. Thanks, Andrew. Thank you, guys.

SPEAKER_03: 40:11

Thanks, Pierre. And of course, thanks to the virtual studio crew who are in no particular order the Bride of Spain, herself the head of institutional relations, Lara Rodriguez.

SPEAKER_02: 40:21

Bye, Andrew, bye everyone, and thank you, Maspier, for these insights. This has been uh a great job.

SPEAKER_03: 40:27

And of course, on his debut performance, the Spanish James Bond, there in his Aston Martin, it is, of course, uh David Inglesias. Thank you, thank you very much. And finally, the man who puts the canary in the wharf and uh looks after Mary's axe is none other than the soon-to-be augmented by cybernetic technologies, no doubt, Mr. John Curden.

SPEAKER_04: 40:48

Thanks, Andrew. Thanks, everyone. And Pierre, great to talk to you again, as always. Thank you.

SPEAKER_03: 40:52

Don't forget to join us on our LinkedIn page. That is LinkedIn.com slash company slash registr-tr where you can network with Pierre and you can connect with Gentech, and you can connect with John and Laura and David, and of course myself and Manuel, the producer. Do get in touch. If you'd like to suggest a topic for the show, or if you'd like to be a guest, let us know. And in the meantime, from everyone here at Register and at the Six Group, uh have a good month, have a safe month, and we'll see you next month. Bye-bye. And remember, you've been listening to the Register Roundup, which is brought to you by Register TR, a member of the Sixth Group, and features members of the Regis TR team and special guests expressing their personal opinions, not the opinions of Register as an organization. There's no representation made as to the accuracy or completeness of information in this podcast, and neither should it be taken as any legal, tax, or other professional advice.