CTRM Evolution: Adapting to new business dynamics, AI integration, future-proofing

SPEAKER A

Allow me to introduce our moderator, Chris Sass, who is, as you can see, the host and founder at Insider's Guide to Energy. We will be looking for questions from the audience. We will have a microphone if you can raise your hand, and I'll make sure I get that to you. So without further ado, Chris, over to you, sir.

SPEAKER B

Well, thank you all for coming to this session. I can't think of anything I'd rather do after lunch than talk about CTRM. It seems like a really exciting way to get your afternoon started. It's really cutting edge. I am Chris Sass. I am the host of Insider's guide to Energy. I'm also a technologist. I've started some energy tech companies. Last company I started was a company called Fidectis, where we did OTC settlement. So I work quite closely with the ETRm vendors in the ETRm space. It's somewhere where I spend quite a bit of my time during my day job. What we're going to do now is we're going to go down and meet the panel. They're each going to spend about two to three minutes telling yourselves who they are and kind of what's keeping them awake at night or what the trend they're seeing is in eTRm today. So let's start with that, and then we're gonna move into a conversation. When we start the conversation, I'm gonna ask you to start thinking about some of the trends you're thinking about or things that you would like to hear us as a panel talk about as well. So I'll ask for feedback about things that you're seeing in your environment. Feel free to add them if you don't. We've got plenty of examples up here, but we'd love to engage you at that point, and then we'll go through the questions that we've lined up for the panel. So without further ado, let's go with Vlad. Just go ahead and introduce yourself.

SPEAKER C

All right? Sure. Hi, is this all right? Can you guys hear me in the back? Yep. All right. So my name is Vlad Zieman. I work for a company called Traxxas. We're a metal trading company. I manage the information systems. Traxxas, we're a global metal trader located pretty much on all the continents and moving all kinds of physical material. My background in ctrm, so I've been. This has sort of been three decades for me in doing information systems for commodity trading, for physical commodity trading. So CTRM, you know, if it's a dirty word for anyone in the room or anything like that. It's very much been the bread and butter of what I've been involved in and what we've used pretty much everywhere I've worked. I started in energy, did some ags, did some. So oil, gas, power, ags, most recently metals, attracts, has kind of been across the spectrum and have probably either purchased, implemented, or inherited and had to support most of the CTRM systems that have been out there for the past few years. In terms of just what's trending, I think, and what I think would be interesting to talk about is just more and more demand for data, for information, more information coming at us, coming at us, CTRM users, and more information demanded from us, as, you know, whether it's regulators, stakeholders, whatever it is. So just dealing with all of that through our systems is kind of, you know, the big thing that's, that's impacting us today.

SPEAKER D

Hi, I'm Samir, Samir Soleja. I'm the founder of Molecule. It's a. We like to call ourselves the most advanced, technologically advanced ETRM system in the world. I founded the company back in 2012. Prior to that, I was in Sun Guard's consulting services unit, where we made custom software around energy. Went off to business school and came back and thought I could do a lot of things, and decided to start an ETRM company. I'll tell you how much fun that is. It's actually been a really fun ride. I've learned a lot, and I, you know, we spent a decade sort of crossing the chasm of features in order to be able to be useful to a broad variety of people who use CTRMs. Now, having crossed the chasm, you would think that we sort of have a moat with which to defend ourselves. But the truth is, technology is moving really, really fast. And I think the thing that keeps me up at night is, why does a CTRM matter? And what is it even going to look like ten years from now? How do we get there?

SPEAKER E

Hi, my name is Alex Zhukovsky. I work for a company called National Grid. This is international gas and power utility. Seven, 8 million customers, four, five, $6 billion in annual expenditures on commodity trading. And I am a director of energy portfolio risk management team. I started this job in 2008 and still in the same position, which means one of two things. Either I am good in what I am doing, or I am incapable of doing anything else. I'm particularly fascinated about automation within CTRM system, and laziness is a father of progress. So my best day in office, when I come and I see email that you have a potential problem. Automated email. So this is a day which will be interesting. So I'm happy to be on this panel to discuss the possible impact of AI on CTRM system. I think AI is a transformational technology. It will change the world as we know it, and CTRM also will be impacted. So to discuss how, it's very interesting also to hear from my colleagues, but also from audience as well. Thank you.

SPEAKER B

Awesome. So you'll see the panel by design. We have vendor users and a mix to give you an understanding of what we're seeing in the world and how we're seeing it. And what I want to do is start with the current challenges. Samir said, things are changing really quickly. I did an ETRM series on the podcast where we interviewed ten ETRM vendors, and it was probably about a year and a half, two years ago, and we're having to do it again because the world has changed that much, that it's interesting to go back and look at what's different between the vendors, but the changes are all predicated or driven by challenges. So what I want to do is just go down the line and see what are some of the challenges that are happening in your CTRM environment today that are driving change. Why are we doing things different? Why are we having this conversation? So start right down the line. Yeah, go ahead.

SPEAKER C

Sure. So I guess if I think back a little bit historically, so maybe what I'll start with, it's a little bit of what's changed and what hasn't changed. Right. Two sides to that question. So we actually, the first place I kind of worked in this space was in the nineties, actually, for a company. We tried to build our own. Well, we did build our own in house CTRM system. So it really resonates to me what Samir is saying about the challenges of that and the accelerating technology, because it's really kind of a situation where it's very difficult and increasingly difficult to keep up with that. And we set out at that time to sort of do the whole thing from start to finish, all the way through to the accounting systems or the logistics, the trade entry logistics, document generation, invoicing settlements, the whole thing. And it's a massive, massive task of, I think what's evolved for me, in my thinking, is really this kind of acceptance that CTRM is actually not your entire solution. Right. And today I would go out and I would never attempt to build an entire CTRM system because there are a lot of good software providers out there. With solutions. But also what I've seen as a buyer of those solutions is none of them are fully complete, and they can't be. They can never be, because the minute they are, there's some new challenge, there's some new issue that requires extensive development. So we've kind of taken the approach of buy the pieces you can and then build custom things yourself around that and integrate it all really well. Then you have to keep that dynamic and growing. So the challenge, I think the biggest challenge is to do that and to really keep up with all of that. I'll just throw out a simple example, and then I'll pass to my colleague. Sorry, if I'm going, you got to go with me. So here's an example, right? I think recent example is sort of carbon tracking. Anyone who's in the physical world has to deal with carbon emissions, starting to understand what your footprint is, track that, report that to various stakeholders, and realizing there's, like a huge absence of information. But doing that, that has to be combined with all your physical commodity flows, knowing where your materials came from before you even bought them. And there's this sort of dramatic increase in the amount of information you need about the origin of your materials, the path by which they traveled to before you even took ownership of them, that increasingly needs to be captured in your systems. We're finding that the only way to do that is really to think of CTRM as just one piece of a wider ecosystem with other software vendors, other solutions, and connecting that up.

SPEAKER B

So how does that work? Because traditional cTrms have kind of siloed your data, and you kind of have to extract it out, massage it, and do whatever you want to do with it and use it however you want. And every vendor wants to be the golden source of truth because you control the customer then. So how does it work in reality? How is that going for you, building those snap on parts?

SPEAKER C

Okay, I'll keep going and let my colleagues go afterwards because they may all have very different answers to this. But for us, it's been, there's a real need for our CTRM systems to be very open, to accept information from external sources, and we're finding good solutions out there that provide some of that information from different vendors. But they actually don't naturally feed into the CTRM system without a lot of work, without a lot of mapping, without a lot of reference data, without a lot of transformation. So, yeah, so it's been difficult, and I think this is kind of an opportunity for the CTRM vendors to. To maybe think about how to add value by making that task easier, because that integration task is really hard. But I think it's unavoidable.

SPEAKER B

So Samir is nodding yes. Yes, yes. So that must mean that as Nichiren vendor, he's thought through this problem and has some suggestions.

SPEAKER D

Well, yeah, sure. I mean, as at a vendor who thinks of itself as modern, these are challenges that we had to deal with some by necessity, where early on we didn't have support for all the commodities that people traded or all the instruments that people traded. And then later, as we started moving into larger and larger enterprise organizations, where there's all sorts of different things they needed to do with the data. So, I don't know, five or six years ago, we built a restful open API with complete documentation on it. And I guess we thought of it as an add on to the product that we had. Turns out it's a everybody's favorite feature, and it's essentially a product all of its own that is maintained, that is quality controlled in special ways and maintained on its own and versioned properly and things like that. We've recently just launched something else where we can automatically stream that data via push, via Kafka to a data lake as well, because we're finding that even restful APIs are not great for like time series analysis and things like that. So that was, this is not meant to be an advertisement, but like, that's how we solved the problem. We think it's a pretty good way to solve the problem. You know, going back to the question of what the challenges are. Right. That was the challenge we solved three or four years ago. The add on challenge to that is load. When people are pounding your APIs all day long for extremely dense bits of data, I think our standard extract is 200 columns wide and many customers 100,000 rows long. If the API gets pounded for that once a minute, that's a very painful thing to have to deal with. Load has been is the next challenge on top of that. And then I think coming from that, there's this sort of humbling realization as a CTRM vendor that the nature of the product is just not all that sexy. Like, it's a system of record. Even though you have fun building it and you have fun with your customers, coming to the conclusion that your job is precision. Four nines. Precision across 200 columns, across 100,000 rows a day per customer is sort of the second order issue. That comes from the large amounts of data. Large amounts of data come from electricity. Complicated data comes from renewables but the second order problem is that you've got to get precision right on enormous volumes of data. And I think that's where AI comes in.

SPEAKER E

Alex Hugo in my opinion, there is no such thing like ideal CTRM system, which can cover all aspects from physical to risk to accounting. And it's understandable because the business we're all in is very complex and has a lot of nuances. So if I think about CTRM system, which we can use, for example, first of all, the first requirement would be the ability to handle the physical commodity. Ability to move, ability to buy, to move, to schedule, to account for each molecule of gas or each electron of power. And it's a complex task. I'm not even talking about the risk manager as a risk manager, because the physical commodity is the key. And different systems, they are good in different commodities. And one of the requirements which I would ask CTRM to do is to give me tools to build my own logic to build the process around how we do our business. So it's almost like a Lego Tools, which you can combine in a clever way and customize the system to fit your needs. So, ability for system to expand and being able to handle new commodities, new business practices, new type of products. I think this is a key which makes this business unique. Because it's almost, you have to build something with your hands, day in and day out, as the World around US changes.

SPEAKER B

I've noticed a trend with all three of you speaking. You're talking singular CTRL. How many of you have one? SamIr?

SPEAKER F

No.

SPEAKER B

How many have one?

SPEAKER C

So we have one today. Sorry, we have one today. But previous places I've worked, we had a lot of them for different products. And it was an issue.

SPEAKER E

I said one. But when I talk about CTRM, it's part of the complex infrastructure. So we have a data management system, we have CTRM, we have a separate risk engine. We have interfaces with the ErpentAl and internally build programs which use APIs to get the data and make interpretation of the data. So while we have one, it's kind of like a data system rather than CTRM.

SPEAKER B

So turn it out before we move into the details. Are there other challenges that you in the audience are experiencing? So if you're in a role or you have that, anyone have any other challenges you want to share that we're going to cover here? Okay, not a problem. So I guess operationally, in efficiencies, you talked about the system and how you want it to be. Maybe we can go down the line and get a little bit of an expression of how that works. So I think you talked about the manual entry and some of the things that go there. Why don't we start with you and we'll work through.

SPEAKER C

Yeah. So in terms of sort of operational challenges, I think the biggest one we face with CTRM, and I've seen this before with, with different systems, this is not particular to any one system, is CTRMs are sort of like, they're kind of hungry little monsters, right? And they need a lot of information in order to produce something useful out. And they need that information input accurately, timely. You know, what am I talking about? I'm talking about, obviously, all the contract terms you have to capture them. They have to be accurate. Any estimated costs, for example, to get a good p and l, you have to estimate your in physical trading business, your storage costs, your transportation costs, your financing costs based on forward expected cash flows. So all of this stuff is a lot of information to capture very accurately. I find that the user experience often becomes, users feel like slaves to the cTRM. Like they're just, their job is just to keep feeding this hungry monster. And of course, that creates operational inefficiencies when people are spending a lot of time keeping the data up to date. I think that's something anyway. That's something we think about a lot. On the other hand, there are a lot of potential solutions out there to these problems, right? In fact, some of the software providers who are here today, presenting downstairs and also speaking at some of these other panels, have some interesting solutions to the problem. So, for example, you know, I know one company is helping with ingesting invoices, PDF invoices, and pulling out the relevant information to avoid people having to key that into the system, or at least to minimize the amount of manual keying, you know, very useful because it's a very high volume thing. We all get a lot of invoices in many different formats. Another example of what I'm talking about is tracking your physical movements, right? So we do a lot, we do some bulk shipping. We do a lot of container shipping. We need to know where our containers are at all times. For logistics people to be sort of, you know, doing that by receiving emails from our transport providers and keeping that information up to date is very difficult. But there are some good software vendors out there who are, who can give you real time information. So the key is to get that information into your CTRM system without a lot of effort.

SPEAKER B

But what's new about that, right? I mean, PDF and ingesting data that's been around for a while. It's perhaps, arguably, hopefully better with the technology. Right. So we're getting it in. I mean, even years ago, when I was at Ericsson, we managed shipping containers and we had RFid chips on them, and we were able to tell you where your shipment was at any point in time. So what's different in 2024 than throughout your career? Why is that a point, pain point today?

SPEAKER C

Well, so I'm not sure I agree that there were a lot of automated solutions for a lot of this stuff for a long time. It is, I'd say perhaps if you're a large, a large corporation that has a big presence or has its own shipping fleet or that sort of thing, then you have control over automating that. But if you're reliant on a lot of transportation providers or if you have a lot of external parties, I haven't seen historically that information coming in structured ways. It comes in many different formats, email, some of them are full text, some of them are PDF's. And so what's, what's, I think what's happening today is the technologies are now available to process a lot of that a lot better, that unstructured data being converted to structured data. And that's an opportunity, but you have to get it into your system, right. And that's the key to make that process easier.

SPEAKER D

You know, Vlad, you mentioned something about ETRM systems being hungry, hungry little beasts, right? It rhymes a lot with something that we've heard about. Something that I see about etrms is that they're sort of needy. So occasionally we'll have a customer who maybe hasn't quite finished adoption yet. They'll log in at the beginning of the implementation process, they'll validate some stuff, and at the end they'll log back in and be like, yeah, this stuff's all wrong. You didn't look at it for 60 days and like, yeah, there's stuff drifted in it. Like, it is the way it works. I think there's a real challenge there, though, around making these things less needy, preventing the drift, increasing the accuracy, and, you know, again, not to beat a dead horse. Like, that's the sort of thing that AI tech is there for. I'm not sure how to use it yet in that way, but I think, I think that the people who really succeed will be, and who are able to reduce the costs around their etrms, and ctrms are the ones who are able to crack that and figure it out.

SPEAKER E

I think that challenges with CTRM are generate, at least for me, from two main sources. The first source is CTRM users, and we have front office, we have middle office, and we have back office. Front office. They like to do things in their Excel files and there is no way to force them out of these Excel files. So they don't like the fact of entering deals into the system or they okay actualizing volumes maybe, but similar situation with the back office. They not very usually technically savvy. They prefer to extract the data into or again, excel files do some reconciliation. So this manual process is somewhat complex and risky, and I hope actually that AI would simplify it to the extent which it would become seamless for people. This is a first problem, and the second problem is a CTRM implementation and upgrades, they're just inherently difficult. Anecdotally, everybody knows cases when CTRM implementations failed with some devastating results to whoever managed those implementations. And I'm saying that the easiest way to lose your job is to become a manager of CTRM implementation. Upgrades are sometimes as painful depending on how deep the changes are in how much regression testing you need to do. There are some tools to automate this process, but I found them not reliable, unfortunately. So these two challenges, they still remain manual labor, entering the deal and reconciling the deal, plus inherent complexity of CTRM implementation and upgrades.

SPEAKER B

So, but I think Vlad's point was there are some vendors trying to address the data part of that. Right? I mean, whether you go downstairs and talk to the vendor here, there's vendors trying to do that. What I hear you saying is it's not there yet, and people still like their spreadsheets and there's still plenty of PDF's in the world. So we're not getting XML files or anything useful quite yet that we can cut out the middleman.

SPEAKER E

I didn't find the software which would allow them to really seamlessly create those entries faster than they do it in the Excel files.

SPEAKER B

You believe that, but the traders don't.

SPEAKER E

Yes.

SPEAKER B

So empirically they can enter it into a piece of software quicker than managing Excel?

SPEAKER E

I hope so.

SPEAKER B

All right, so I guess strategically, if we're looking forward, we've got a bunch of problem statements, we've got a lot of things. How do we evolve this? Where do we go to improve?

SPEAKER D

So maybe just a little bit of behind the curtain on the economics of running a CTRM company. They're brutal. You've got a market with 100 vendors. There's a lot of price competition. If you go to any vendor's website, you'll see that? They claim to do everything because in order to get through an RFP, you have to claim to do everything. Whether you actually do or not is another challenge. And I mean, that's broken, I think, again. And then funding something like that is hard, right? So twelve years ago, if I were to have walked into a venture capital and fund and said, hey, you see this thing called open link? I want to build another one of these. I just need 75 million and we'll be good. Give me like five years would never have happened. And so these things are wide. These things claim to do everything. These things try to do everything. I think, again, from the vendor perspective, what would be amazing is some focus.

SPEAKER B

But isn't there a pivot taking place?
SPEAKER D

Right?

SPEAKER B

I mean, I look at the big bangers you mentioned, ion or fis. FIS was an example in my mind where they used to want to build everything, they wanted to have this walled garden, and they were going to do it all even though you knew that their settlement module wouldn't be as good as this module. That. And so they had all the checkboxes for the RFP. But about a year or so ago, they came back and said, you know what? We can't do everything. So we're going to take a best of breed approach. And you know what? We're going to partner with this guy. This guy, this guy. Isn't that happening or just not happening fast enough? Or are you guys seeing that in the industry? Because I see it when I talk to the vendors and I see users that say, just like you described, Alex described, an ecosystem or a system where you would tend to pick and choose and say, okay, well, this is going to be my risk module. This is going to be this and that. Are we at a point where we can build and deploy and operate a system like that yet? I think some people are trying, you.

SPEAKER D

Know, I would say in terms of people we talk to, maybe half are like that and the other half are still, well, I would like it to do all of these 613 things in this RFP. And if you can't, you know, we'll award it to the person who claims to do the 613.

SPEAKER B

So operationally they're saying, I want a single throat to choke.

SPEAKER D

Yeah.

SPEAKER B

I don't want vendors pointing at each other. Yeah, I want the best of everything. But you just deliver.

SPEAKER D

That's right. And I think that's a fair need because not all the vendors are at the same maturity in terms of technology or operational abilities, or maybe their implementation processes are very different and like you've got this ecosystem of, say, again, 100 different vendors at various points on various axes, and getting them all to work together is absolutely miserable.

SPEAKER C

Well, so if I could sort of maybe give the mirror image of that perspective. Right. So I think you're talking as a vendor with your frustrations with customer expectations, and I'm going to be a customer. I'm going to be a customer, right. So I think I, I'm not sure. I guess it's maybe my vantage point, but I don't necessarily see the preponderance of CTR, and vendors have suddenly had an aha. Moment and gone, let's be open and so on. I thought it was very interesting what you said, sameer, about when you first built an API a few years ago, you said, oh, this is kind of a cool sort of bolt on feature that we can maybe sell for extra. I don't know if you did or didn't, but then eventually it became sort of like fundamental to the system. Right. And I think, you know, I mean, anyone who is, you know, selling a system that doesn't have that openness built into it today is a tough sell for me personally. It has to be there. It's a fundamental feature of a ctrm. But I would go like a little bit further with that and say, you know, maybe I would love to see if our CTRM vendors weren't just selling us software, but we're selling us, you know, they'll say they're selling us solutions, right, but actually sell me a solution.

SPEAKER F

Right.

SPEAKER C

So the solution is, you know, I can bring in for you and maybe it's an additional service they charge for. Right. But I can bring in for you. If you put in all your bills of lading for all your, all your containers, then we will automatically update their positions daily or in real time, whatever it is. That's a service. We'll charge you x dollars per container per month, whatever it is, and we'll go out and we'll deal with the companies who provide that data and we'll just make it work in our system. Now you're moving away from that paradigm of we're just selling software to actually we're solving your problem. And I think that's kind of what I'd like to see CTR vendors doing here.

SPEAKER B

How about Alex on your side?

SPEAKER E

I will sound like a broken record, but again, I want the solution which would handle the physical commodity, and I cannot build this. Just, it's tough. But if, let's say I have natural gas and this particularly CTRM handles all aspects of buying gears, scheduling gears, actualizing gears, capacity releases, asset management agreements, and then give me tools to build around this CTRM. The processing and maybe with AI, this building thing with programming capabilities would give me opportunity to create prototypes rather quickly, but embed the logic into this programming pieces which would work together. While the physical aspects of moving commodity and accounting for commodity will be provided by CTRM vendor. And it might be one solution for gas, another solution for oil. Somehow each vendor specializes in one or two commodities at a time.

SPEAKER D

Well, and I think that last bit is the key, right? There's sort of good behavior on the part of vendors where everybody says, look, this is what I do best, this is all I'm going to promise. This is everything. And hopefully people will buy from me because I'm always honest and I always achieve what I say I'm going to achieve. That's a lot to ask. I hope we can get people to do that, though.

SPEAKER B

So assuming we're not Greenfield, assuming I have infrastructure in place and I've got a legacy ETRM that's got all my data siloed deep down into some legacy architecture, how do I go forward? How do I future proof? How am I making investments today that are going to be reasonable in five years from today with the technology moving at the pace it is? So, strategically, what are you guys thinking to protect yourself? Right? What are you investing if you find that new one that does everything from nomination all the way through to every bit that you just described and gets, attracts every bit of oil or metals or whatever, and you buy that today, how is that going to be relevant in five years from now? That infrastructure that you're putting in place? Or maybe it's not relevant, maybe it's throwaway. What do you think?

SPEAKER E

I don't have an answer to that. And the reason is I'm from utility company. We are not allowed to change software vendors every two years. So when we buy the software, we usually go through depreciation cycle and it's seven or eight or nine years. So it's really problem. I don't have a good answer, but if my colleagues have all ears, I.

SPEAKER D

Mean, I think you just got to get your data out. I mean, we use internally like 100 different SaaS apps for various things from invoicing to whatever. We recently built an internal data lake to give us a 360 degree view of our customers, from what tickets they've put in to what we invoiced, to what they did in the app. And was there a problem with it, et cetera, it required us to build our own data lake. And I suspect for people who have ctrms and trading organizations, you got to do the same.

SPEAKER C

So I completely agree with that. The openness of the data is critical, and being able to get it out and get it out of a transactional database into denormalized reporting, data warehouse type of structures that you can do a lot with that data, that's just a must. Having said that, though, you know, this, I mean, you can't change your ctrm every two years, right? I would quit. Right. So I mean, it's, even if we didn't have like, you know, the depreciation issue, right. It's a massive, massive undertaking.

SPEAKER B

Sometimes it's hard to upgrade your legacy.

SPEAKER C

Each year for two years. Right. And so you just don't want to be doing that. Right. So for me, future proofing is a bit more of hopefully you've started out by selecting a vendor that you can have a much longer term relationship with, and then I think there has to be good partnering with you. So as a customer with a software provider, you know, kind of pushing them a little bit to continue to innovate the product, but in directions that are useful to you and that I believe as a customer would be useful to their other customers as well, not just exclusively to us. Don't build customizations for us because that's the best way to kill to choke your software vendor. Right. If you're asking for too much custom stuff and, you know, and at the same time challenge them, but at the same time introduce them to these kind of partnering opportunities. Right. So when I gave the example earlier of it would be great if our CTRM vendor couldn't give us a service of telling us where all our materials, it also is. I don't mean for them to go out and build a whole new set of software to do that. I mean, find, you know, connect up with other software providers, other data providers, partner with them to provide a joint solution where both parties profit from providing that solution to us as a customer, don't try to do it yourselves, please.

SPEAKER B

Because it's, it sounds like it goes back to Samir's point about staying in your swim lane, knowing what you do and what your competence is and excelling at that. Before we move to the audience, question, AI, you mentioned a little bit about it. What are some of your expectations for AI? Where do you think we'll start seeing it first when it comes to CTRM?

SPEAKER C

Go ahead. Well, so we've had a really good experience with AI. Think I mentioned the example earlier with taking incoming documents that are kind of like in a human readable form, so specifically invoices, and getting that very unstructured data, because you've got a really hundreds or thousands of different invoice formats with a lot of detail and sometimes tables and lists of shipping and storage details, and getting that into structured data where your trading system can ingest it and you can put it through to approval workflow. So that's one application of AI that we've seen that's actually working and it works well and we've had a really.

SPEAKER B

Good train those models. Did it take a lot of data to train the models to do that?

SPEAKER C

Well? So we work with our software provider on that. I think there are different models in which you can work. One is you just buy the software and do it yourself. We chose to kind of work with them and they do that piece for us, but it's quite quick, right. We have to give them sample, a bunch of sample formats and then they kind of provide that as a service, as an add on service to the software, the training. So anyway, so that's one example of something we've seen that's good. I don't know if we want to talk about sort of potential future things that we're going to get to.

SPEAKER B

We're going to run out of time.

SPEAKER C

So I'll leave it there.

SPEAKER E

I completely agree. Confirmations, invoice reconciliation, this is easy targets for AI, but on the long term, five, maybe ten years. I like to see the system where you don't click at all, you just talk and natural language processing would do the task. AI magically knows your data structure of the solution you have. It's capable to write the code which would change system behavior. Ad hoc reports. When you tell what you want to see. I want to see my five highest counterparties in terms of volumetric exposure, or mark to market exposure. Or please add another column to this report and it will be done on the fly for you. So this is kind of a fantasy world. It requires a lot of investment, but it has a potential to transform CTRM as we know it.

SPEAKER D

I don't really think it's fantasy. It's something that's very, I mean, it sounds eerily similar to some things that we've been skunk working, but like the tools are here today to do it and it is, especially if you've got access to your data, it's not that hard. Even power Bi will give you some of that right out of the box. You can chat with the darn thing and say, what's my p and l with my five highest counterparties?

SPEAKER E

Because it knows the data structure. Now the question is to teach AI your data structure.

SPEAKER D

That's right. That's right.

SPEAKER B

So as we come to time here, any final thoughts? I think we're pressing up as quarter after right when we're supposed to end here. We got a couple minutes. Okay, let's do. What I would like to do is ask the audience. You've heard our panel's thoughts. You've heard some of my thoughts. Any questions from the audience? Yes, please.

SPEAKER D

I don't know if I need a mic.

SPEAKER B

I don't know.

SPEAKER D

The magic box, the Pandora's box, basically, that Alex is describing. Sounds great, but how do you control that? Not everyone is creating then their own truth in the examples you gave.

SPEAKER B

So how is an organization to make.

SPEAKER D

Sure that it is still one source of truth?

SPEAKER E

This is a good question, from what I see now, AI is very good in simple tasks, but it's not as good for the complex questions. But the problem is it gives you the answer with the same degree of confidence, like simple answer, and it won't be the final truth, in my opinion. This will be a prototype, which you have to check and then move this prototype into production.

SPEAKER B

But isn't it a prompt? I think the AI can do a lot of what you're asking it to do. It's. The prompting is a chest. The question, in my experience. Right. So if I say I want this multifaceted question, if I don't define it, just like talking to an engineer, and if I don't tell the engineer exactly what I need them to do, there's no room for interpretation. So you get garbage out if you didn't do the prompt. Right. So I would think the AI could. I don't know that we answered his question on the source of truth, but I think it can solve the problem.

SPEAKER D

Well. So I think there's an art in combining said generative features with deterministic questions, and I think it'll be really interesting for the people who crack it. Again, I think the tools are there for that, where you're already asking the CTRM deterministic question. So you really just need to figure out how to make your request into one of the deterministic questions.

SPEAKER C

Yeah, I think that distinction actually between sort of, I would say between deterministic, which is the word that was in my mind as you said it, and versus, I don't know if probabilistic is the right idea. But AI kind of operates in that space of probabilistic. Right. Like sort of identifying patterns is great potential for that. Right. Or taking, taking somewhat fuzzy data and trying to make it, you know, trying to get some information out of it. But you still, in the end, you know, you've got to have a p and l number. You've got to know your p and l number is x point y, you know, and it's not some, you know, approximate range of things. And, you know, you've got submit your financials and so on. So there's always going to be kind of this need for ctrms to have accurate, precise numbers. Where I find it, you know, it's less obvious to me how the AI fits into that. But in terms of AI looking at your data and saying there's some trend here, there's some, like, likelihood that, you know, something, something funky is going on in this area. Look closer in there. It's sort of those areas, I think, where I think there's tremendous potential. Yeah. But I think there has to be one source of the deterministic truth. That's what I'm trying to get at. To your question, which is kind of your CTRM database, but there's a lot of. Well, what do you make of that data? How are you going to use it? How does it influence your decision making? That's based on the numbers, but AI generated, I guess, yeah.

SPEAKER B

So maybe one other question. Is there one last question in the room? Anybody? All right, well, I want to thank the panel for the conversation. It's been interesting. Thank you so much.

SPEAKER C

Thank you.

SPEAKER A

Fantastic. Thank you very much indeed to the panel there, to Alexander, to Vlad, to Samir, and of course, to Chris. Obviously, we touched there on part of AI and the future of CTRM. We're going to start our next panel in just a session in just a moment. I'm hoping that we have got Mark, Amir, and Katie here. If you could make your way to the stage. Thank you. And, ladies and gentlemen, there are more seats here at the front of the room, so please do make your way forward. And what's on? Get those. Should I be on focus? Hi. That was definitely. Let me give you that one for a moment. Thank you. Okay, ladies and gentlemen, if you'd like to take your seat, we will kick off with our next session focusing on artificial intelligence in action. Make sure we've got the right slide. Let me introduce our moderator for this session, Mark Fleming Williams from capital Fund management. Mark, over to you, sir. Thank you.

SPEAKER F

Thank you.

SPEAKER G

Very much so. Thank you very much for joining today for artificial intelligence in action, how trading companies are using AI to streamline manual processes, reduce risk, and increase revenue. My name is Mark Fleming Williams. I'm the head of data sourcing for CFM, which is a Paris based quantitative hedge fund. And it's my pleasure to be joined today by Amir Azhar of Quantum Horizon Capital and Katie Carter of Clear Docs. Welcome.

SPEAKER H

Thank you, Mark.

SPEAKER G

I think actually, why don't you both begin by introducing yourselves a little bit, talking a little bit about your roles and your organizations, and then we can go from there.

SPEAKER F

Katie, you want to go first?

SPEAKER C

Sure.

SPEAKER H

Thank you. So thank you for the introduction. Thank you for inviting me to this panel. My name is Katie Carter. I'm the VP of product management at Cleardocs. And Cleardocs is providing a platform for AI solutions and specifically building intelligent apps around the commodity trading lifecycle. So to kind of follow on from some of the conversations that we just heard, we're working, you know, in complement with CTRM systems to help build context and clarity around some of the data that they need to consume to do their roles.

SPEAKER F

Thank you. Works so well what we do at Quantum Horizon Capital. I can summarize it with two key words, crude oil and machine learning. So that's the VR commodity training advisor registered on the CFTC. And I can go through the details, but I would think that those two keywords should be at this point, fantastic.

SPEAKER G

So this is a panel all about AI, artificial intelligence and action. AI is a bit of a catch all phrase sometimes and can mean a lot of things. So I don't know how you guys feel about telling me what you mean by AI. When you say you use AI in your processes, what do you mean by AI?

SPEAKER H

Yes. So for clear docs, what we mean by AI is building and maintaining custom pre trained AI models, specifically using a variety of technologies that we've talked about, whether it be OCR, NLP, and in the future, generative AI, and marrying that together with process centric and industry centric automations and putting those two things together to deliver solutions. So we're using specific models around document digitization and bringing data to extracting them from those documents in the various formats. And that's really the specific use case for clear docs today and where we're.

SPEAKER F

Delivering solutions, I think that's a very tough question. So, I mean, for us, the situation is simpler than it. We are dealing with numbers mostly. So over 90% of the work is how the AI and the ML in particular can help us with the numbers, and then as the new features becoming available, how we utilize those to accelerate whatever we already are doing. So, as I mentioned, like AI is, as to Mark's point, is quite big. All I want to say, that's not a crystal ball that we are looking at and tell us the future. It needs a lot of work, constant and continuous work.

SPEAKER G

For you, Amir, when you hear AI, do you think machine learning is that?

SPEAKER F

I do think, first of all, I think go back to artificial neural network back in early 2000, and that's the first thing that comes to my mind. I differentiate between AI and generative AI. That's today's definition of AI, but it's a major breakthrough. But for what we do, it's not as fundamental compared to regular AI yet.

SPEAKER G

Okay. So when I think AI, I think data is going to be quite integral. So, again, setting the scene a little bit. What data? You've mentioned it briefly, Katie, already. But what data is going into the machine? What data is going. Going in and where is it coming from?

SPEAKER F

So we had a session a couple of hours ago about that. So in our case, our data is the fundamentals and how they go. They are fed to the machine at the moment that they are available. So that's the key component that has been changed compared to many years ago from a and M times, and that those fundamental data could be macroeconomic data or preparatory microeconomic data that are developed. I would like to also mention that there are some indicators that you build based on the data that are available. They are preparatory, and you feed those as well to your model as you have built. The trust on those indicators can go on and on.

SPEAKER H

Yeah, sure. So to me, it means unlocking the data that's present in documents, and not just documents, unstructured forms of communication. So along the commodity trading lifecycle, you have documents and communication. That's underpinning many different inflection points. And so AI is being used to unlock that data, normalize it, enrich it, and allow it to be processed downstream for better decision making. And that's really where I agree with you. When I think of AI, data is really the fundamental that is driving the automation that we've all been talking about for a long time. I think AI, with its access to data, helps drive that automation.

SPEAKER G

Okay, and so why. Okay, big, broad, wavy hand question. Why is AI such a game changer? That's what you said. AI is a game changer. Why do you not think it's a game changer?

SPEAKER F

Well, I mean, I think of AI as like Internet. So it was a game changer. And, but I may speak openly here, I think there is an element of over expectation here that it's going to solve all of the problems that we are facing now. It will be more and more efficient as we progress. The pace of that is very different from what it was a decade ago. So all of those contribute to the point of view that this is a game changer. And then the question is that sense, can we really rely on everything that AI says, or is it going to be. There is a danger here. If I may dive into this topic, since you started the conversation, please. So right now, let's say, talking about our generative AI today, it's trained by the data that is available online, whatever is available on the web, web and perfect. And it's using those data to generate the answers that we are looking for, or at least the components that lead us to better answers. So it saves a lot of time. So that contributes to generation, being a game changer. So let's say you want to do something that before you needed a lawyer, you can at least get started with that to see what kind of language you still need a lawyer for the application, let's say. But then you get a better understanding for basically at no cost, which is amazing, similar. If you wanted to know about certain things that you had to hire a consultant before and pay tens to hundreds of thousands of dollars. Now you can do it literally in couple of hours. So from that angle, it's a game changer. But now the question is, so as these machines expand, get more kind of established and they produce more, so the component that they are producing now becomes part of the original database that they are getting trained. And then that could exponentially accelerate. That raises the question, so are the answers that you are getting are sourced from the original point or is this a macro level AI generated that is getting retrained and getting back to us? So that's some question mark in my mind. I don't have an answer for it.

SPEAKER G

Well, I think, I mean, I think it's a difficult question. Partly it's going back to that AI definition question is that again, what do we call AI? Because I saw a nice meme, you know the meme where they're taking the mask off the Scooby Doo villain, and they're saying, oh my God, you're just. And it was like, oh my God, you're just if. Or statements like, essentially it's all the same old stuff and now it's just reached a level where it's called AI. But in your business, Katie, how would you say AI has changed the game? Is there a moment pre AI and a moment post AI?

SPEAKER H

Yeah, absolutely. I mean, for me, the game changing statement is a little bit simpler. And I agree with you in separating sort of AI versus generative AI. We'll talk about not generative AI. And what it allows, you know, people in this industry to do is access their data in normalized way that allows them to analyze it. And what that means is you can streamline your decision making. So when you have data and you can then very quickly add context and meaning and controls around that data through automation, then you can make quicker decisions. And I think that's the game changers, is we all understand how important automations are in bringing our businesses forward and being able to scale our businesses. And having been kind of in and around technology and commodity trading for a long time, that's not a new sentiment at all. But what seems new now is the ability to deliver on it that's enabled through some of that digitization capabilities from AI.

SPEAKER G

Fantastic. So Amir, delving into more detail from a cost structure and resources perspective, how would you compare leveraging AI versus a traditional analyst based approach?

SPEAKER F

I guess that's easier to visualize. Think about it like in particular for crude, you need to analyze many, many factors all at the same time. So either you have a couple of very good analysts or you need a team to go through all of those elements and things that, I mean, they are not rocket science, but data that need to be incorporated when you think about it from that angle. So now you can definitely have a more optimized resource allocation when you are using AI ML component, particularly when you, it is combined with the API that in the previous panel that we're discussing. So I think that's the component that we are missing. It's like when you think about, a lot of people talk about shale revolution, shale was there, oil was there, everybody knew about it. The horizontal drilling was there too. But when they got combined together, it became the shale revolution revolution. And I think the digitization and the accessibility of data without human interference combined with AI is giving us that revolutionary moment that everybody talks about.

SPEAKER G

I saw a similar comment around how Uber when it emerged, it combined a whole load of technologies which all existed, but it was the first company which actually just did it all together. And actually it was a beautiful thing. Maybe Airbnb could be a similar kind of idea as well, just ways of combining a whole lot of things. But, and Katie, where do you see companies currently using AI in their trading operations from your perspective at Cleardocs?

SPEAKER H

Yeah. So it's really following the commodity trading lifecycle. So that can be pre trade finance and using some of the documents required to understand the commitments there. It can be post trade execution for clearing and reconciliation, and then moving on as these documents get passed through physical movements, actualization, inventory, reconciliation, and then finally into settlement. So we find that there's different inflection points that correspond pretty well to specific business use cases where there custom pre trained AI models can be leveraged together with, like I mentioned early, the industry centric process, centric features that really help to unlock the lower decision risk needs of the community and making quick decisions with good information.

SPEAKER G

Okay, Amir, going into the, this, going into AI, there's always coming back to the data. There is always kind of two levels to it. There's looking at the data and making sure it's clean and prepared and understanding the data, etcetera and all these things. And then there's the AI algorithms that you're going to impose upon it. Where would you put the percentage importance on those two processes? Is it 50 50 or is there, is one of them perhaps in terms of time spent or man hours?

SPEAKER F

Well, I would say 80 20 towards data? Yes, it is way more important because it's always about that, even before the AI, it's always about how you choose and how you analyze your data. Then there is a little bit of art into it. It's not just science. So when you have the good data, getting a good model is not very complicated.

SPEAKER G

And is it easy to find good data?

SPEAKER F

Well, you can make them better. So it's just, again, it's like an art. You know, the first time around, you may not draw the best drawing, but then as you practice, and if you have some sort of talent for it, they work better.

SPEAKER G

Okay, Katie, what challenges are typical in the implementation of AI for the use case that you described and how are they overcome?

SPEAKER H

Yeah, so I think one main challenge is data mapping. So connecting disparate, like naming conventions for data that you see amongst the industry. I think this is a potential future use case for generative AI, where data mapping can be solved by looking into that knowledge base that will be generated there and proposals can be made and predictive analytics can be used. So I think that's something that is on the horizon of having a pretty good solution for. But today what we see is still a challenge and a bit of a manual intervention that's required. It's not so a challenge, but something that we take pretty seriously is data validation and making sure that while you're using AI, you still have a human in the loop at various points. And I think that's like an important infrastructure piece to keep in mind when you're thinking about AI solutions is how do you ensure you have the right level of intervention, really making sure that the data is clean, making sure that wherever needed, you have system defined or user defined business logic, so that you can find inaccuracies. Like you mentioned at the very beginning, you know, the AI isn't going to be 100% accurate at all, and so we need to plan for that and really understand the limitations.

SPEAKER G

Okay, Amir, from a trading perspective, there is obviously competition in your world, and that competition has a fair bit in common. Potentially, they've got data sources in common, so it can be hard to differentiate there. Obviously, you've got your own special knowledge and secret sauce that you're, which is your own activity. But in terms of the AI, do you think adding the AI and whatever machine learning is that, is it hard to differentiate within it, or do you feel like everyone's doing the same thing out there?

SPEAKER F

It's very difficult to know for me, what everyone does, especially in my field, because no one talks about what exactly they do, because the nature of it is preparatory. So I would say that, yes, there is a competition, and there are ways to make sure that you are not falling behind. It's very difficult to know that who is completely ahead of the game, because, again, nothing is fully transparent on this. And this is something that, at the end, going back to the art and science component, so it's not pure science. Actually, I want to raise a question here, whether the audience think of AI as a machine that with a certain input, it's going to give a certain output, or is more like a creature, that with a certain input or behavior, they could have varied, their reaction could vary. Do you think which one would be the case if, if you wanted to.

SPEAKER G

Well, I will speak for the audience, but no, the. So the, it seems, and it depends what we're talking about, but it seems, for example, with chat GPT, that if you put, if you order the words of your query differently, then the outcome can be different, even if the question is the same. So that makes it. It's an imperfect machine. I would put. Well, I mean, should we put a shit? Do you want a show of hands for who thinks machine? Who thinks creature?

SPEAKER F

Yeah, let's raise hands. Whoever thinks that's a machine.

SPEAKER G

AI and machine learning is machine versus who thinks it's a creature. So if you think it's a machine, raise your hand, and if you think AI is a creature, I'd say that's 50 50. And if you think it's neither, or you don't want to play.

SPEAKER F

Okay, so, I mean, I can share my perspective on this in a more detailed way, I think to your point that you mentioned, I simple case, you put in different order, you get a different answer, you can put in theory, you can put in the same order, and if you are a millisecond later, you may get a different answer. It's like that the way, that's the way the intelligence part of it has to work. It shouldn't be dependent like pure if functions. But then, on the other hand, goes back to the amount of data that it is trained for and the nature of that. So, so it's complicated from that angle. That's why it makes it kind of more difficult to fully replicate from. It's like an artist. Three artists may design, draw the same thing, but they have, their paintings may look different, not completely, but will look different and has different values.

SPEAKER H

Yeah, I think I agree with what you said. I think it reinforces the human in the loop need for when you're looking at AI solutions, specifically around prompt management, for example, and making sure that the queries are well thought out. The data source is the right source of data that you want to be training on, and that's an iterative task always. And so it's really important that I think in order to do that, you think of both solutions together.

SPEAKER G

I think, I mean, we're at risk of discovering that we're all actually machines if we continue along this route, because we are changing our answers as we receive new data. So if that makes us.

SPEAKER F

Exactly.

SPEAKER G

Are they creatures or are we machines? I'm not sure. But Katie, what makes an intelligent app and how are they used today?

SPEAKER H

Okay, so I touched on this a little bit earlier, but an intelligent app would be combining the AI models with industry specific training, with that industry specific process knowledge that enables automation. So together, the AI is going to be able to extract data that it understands because it's industry specific. And then that data can then be analyzed in a very specific way or automated, whether it be through reconciliation or workflow management. That doesn't require AI, but in order for it to work requires good data. And the combination of those two promotes better decision making and unlocks that intelligence. Really, I think each on their own is lacking. So from our perspective, an intelligent app is the combination of those things. And again, they really follow the trade life cycle in the solutions that we provide, because we are very commodity trading centric as an organization. So specific apps would be trade confirmation, trade finance, specifically around documentary letters of credit, inventory and movement reconciliations, potentially future solutions around carbon accounting. And really that can be from front office, mid office, back office, regulatory. Each of those business processes would require an app as the way we define it.

SPEAKER G

Okay, thank you, Amir. Is there such a thing as too much information in machine learning training? And how could that affect the revenue and profitability?

SPEAKER F

Well, I can answer the first part of the question, the second part of the question first. So when you, I don't want to use the word waste, but let's say a version of that, a lot of efforts that are done in the wrong direction basically are wasting your resources. So it affects your cost and reduces your revenue. That's. But then the question gets tricky. What is too much information? How do we know it's too much? How do you know that you are landing to your localized solution rather than a global solution? Think of it from when you are training your model. The curve has to get to the minimum error if you are thinking of it from a mathematical perspective. And then sometimes if you have certain amount of data that are skewed, you get trapped in a solution that is not the global minimum, but just the local minimum. So how do you avoid that? That's a question that, again, goes back to how you do your trial and error. So I would say that there could be a potential that you are over training your model the same way you can overtrain your dog, or your children or yourself, then you go to a gym. So where that point would be, it's very difficult to theoretically say exactly where it is. So I would go with the try and error process.

SPEAKER G

And how would you discover that you, you would. How would you discover that you had succeeded?

SPEAKER F

In our case, it's simpler because we are dealing with, it's kind of a binary situation. Either you make money on your trades.

SPEAKER G

Or not go out of business.

SPEAKER F

But in a case like your case might be a bit more complicated because it's many layers that you have to evaluate what happens. So it depends on the application.

SPEAKER G

Okay, Katie, so coming over to generative AI, which is obviously taken the world by storm, in the last year or so, year or two, where do you see generative AI affecting your business as it continues to, the wave continues to spread.

SPEAKER H

Yeah. So we see functional requests all the time that aren't currently served by our AI models, that would require generative AI technology to solve for. So that would be specifically around documents that are long form text, very unstructured, and would require sort of a secondary analysis or formula to produce the output that you would like from it. So that would be the simplest future use case for generative AI would be to augment and enrich the digitization capabilities. But beyond that, we're looking at more expanding into use cases that weren't possible before. So comparing documents for commercial terms and risk analysis that are not the economic terms, where it's very easy to compare price between two sources or three sources, looking more at the nuance of the language to say, flagging it as a risk. Perhaps if GT's and Cs have changed and your counterparty didn't notify you, do you want your system to notify you that there's a meaningful difference in the language here that you want to take a closer look at? You can't read every document in its entirety itself for every single trade that you do. So that would be another example. Content generation, document generation, being able to do some ad hoc analysis, asking questions, not just typing questions, but asking questions. There's different modalities in which the prompts can be delivered and then being specific about what you would like returned, and then being able to really do that on the fly, on demand with some flexibility. Those are all areas that we see a lot of exciting new future use cases for.

SPEAKER G

Lovely and Amir, for you on the generative AI side, have you found use cases for it yet? And I'm probably, I don't know, we're presumably talking about OpenAI here. And do you trust it? Can you trust it on a kind of rigorous way in your processes?

SPEAKER F

Again, the second part of the question is simpler. No, you cannot trust it, because even, I mean, I tried with chat GPT in particular at the very beginning, so I asked a very specific question, and it was kind of a historical question that I love history, and the answer was wrong. And then I told this is wrong, and this is the answer. And then second time around, it repeated the same answer. So it's, let's say, who was the king for XYZ location at that time? That was the question. And then it kept giving us after third or fourth time chat GPT responded to me that, you are right, that was the king of that location at that time. So, which makes me think that, okay, so do you really can fully trust it? So the answer is obviously no, it.

SPEAKER G

Works the other way as well. I understand if you ask it what two plus two is, and it says four. And you say, no, it's five. The next time you ask, then it'll.

SPEAKER F

That's the risk you are running, because then you can have another bot producing wrong information and feed it to the other bot. So that's, I mean, that goes back to the whole risk that we might have here. But I think I missed this. The first part.

SPEAKER G

No, the first part was so it sounds like you're not using it in your processes because you don't trust it.

SPEAKER F

We are considering it from content production perspective, and we are considering it how to automate some of the work from analysis perspective. You want to see what happens. You can look on the graphics and then see the charts and stuff all at the same time. But what if I want to produce a paragraph that instead of an analyst writing it, we can have the first iteration produced by, which is relatively easy, I would say, from technical perspective, and then you have to have someone review it. So that's how I would envision utilizing it. And this is not far in the future, in the. It's kind of near future that we are working and we are working on.

SPEAKER G

Okay, thank you.

SPEAKER H

I would just add to what you said in the sort of the trust perspective, and that we wouldn't use an open source model in solutions for commodity trading, because it really doesn't make sense. It would have to be private and secure and specific to your data in order for it to be more trustworthy.

SPEAKER G

Sure. Okay, Katie, how are some of your clients currently using AI?

SPEAKER H

So our clients really specifically use AI for the document digitization capabilities that we provide. And like I said, documents could be anything from trader recaps to PDF's to handwritten documents, truck tickets, inventory statements. It really runs the gamut. But what we are able to do is then digitize, normalize, enrich and validate that data for onward consumption. And that's how they use AI today.

SPEAKER G

Fantastic. So, Amir, you are in a world where you're having to compete to hire people in AI, and they're very expensive, and there's not enough or there's not that many specialists yet. There are a few. How do you find, where do you look to hire for this kind of expertise? What are challenges you overcome?

SPEAKER F

So, yeah, that's something that I guess everybody at some point realizes that the skill set requires is so broad that it's hard to find one person that I would say one particular kind of skill set that you can say, oh, this is going to cover it. So the way I try to tackle the situation kind of divided to skill set on the coding and programming versus the skill set and subject matter experts. So you want to have a good mix of those kind of skills. In a way, it's important what not to use in your AI system as well as how much it's important what you utilize and use. So, and for that kind of approach, you need to know someone that knows that market, that commodity in and out. That kind of information is very important. You may design very sophisticated models, spend a lot of time trying to train it, while if you reduce the layers or if you made it simpler and use the right data, you could have gotten to the solution months earlier. So I would say subject matter expertise is the key here.

SPEAKER G

But they need some technical skills, don't they?

SPEAKER F

Yes, you can. The point is that technical skill set, the coding side. Yes. You need to bridge those two together. So somehow you need to have people in place to be able to navigate these languages. And that's how, I would say, optimized.

SPEAKER G

And so do you see the subject matter experts, in your case in the oil industry, all rushing off to do coding courses and get their technical skills to be useful? Or how will this problem be filled?

SPEAKER F

So the question is that is it better for subject matter experts to learn coding or is it better for the coders to learn subject matter expertise?

SPEAKER G

Well, it sounds like you would prefer. Well, yeah. Well, which.

SPEAKER F

It'S a very difficult question because a lot of subject matter experts, they have decades of experience, and learning the latest coding would be a challenge, especially because the coding is a very rapidly changing skills.

SPEAKER H

I would just validate what he's saying with our experiences that the subject matter expertise is a real differentiator, especially when we're talking about data language normalization, it really makes a big difference in how quickly we can deliver solutions. So I think it's a point well made and one that we definitely experience in our organization.

SPEAKER G

We need both. But there's a lot of talk at the moment about people saying that learning to code is the equivalent of learning journalism in the late nineties, because actually generative AI is going to kill the need to code. So perhaps, I don't know. So careful if you start learning to code right now.

SPEAKER F

Well, I've heard that, like, I have friends and like, even, like some scientists say that what they used to do, it's, now it's done by chat, GPT, even the, the free version. But I don't agree with that completely, because, yes, if, yes, you want to draw a circle, you had to write a code before. Now chat writes that for you, but the question is that now that your mind is free up from writing the code for drawing a circle, then what else is it that you can utilize your brain for? So it opens up a new horizon that the scientist's brain can work on and not be preoccupied with it.

SPEAKER G

Fantastic. So we've got a couple of minutes left. I don't know if there are any questions from the floor. Yes, here comes the microphone.

SPEAKER F

If you look left, what would be your suggestions for a freed up brain to focus on once I have chat.

SPEAKER E

GPT write my code?

SPEAKER F

Well, let's say if you're on the technical side, you can focus more on the subject matter expertise side. So that's going to give you an edge to communicate with kind of well established subject matter experts. Still, at the end, a lot of these are languages and keywords. Or sometimes you don't need a code, or you just need to find the right source of data. That could be way simpler than you can even think from where you are sitting. So I don't underestimate the value of subject matter expertise. That's what I'm trying to say.

SPEAKER H

Yeah. And I think it, you know, if it's well applied, then you are removing manual processes and you can do more, you know, thought leadership, you can do more cross training. Certainly there's new roles that are opening up as a result of AI technology. So if you're interested there, that could be a way to reapply those resources. And I think as you know, companies look at their future goals, and if growth is a part of their future goals, then having resources unlocked to then apply to that scale, that future scale is a great opportunity that could be offered.

SPEAKER F

Thank you.

SPEAKER G

Any more questions? Yes, there's one that's spaced around to keep you fit.

SPEAKER I

Thank you very much. Very interesting conversation. A question to katie. You have mentioned a couple times about data normalization. Data needs to be normalized. Normalization was more applicable to machine learning and less to artificial intelligence. Artificial intelligence can work with normalized, not normalized, data. Can you explain a little bit more about what do you mean the data needs to be normalized? And the second question, what you can offer to the customer, customer traders, more than chat, GPT or copilot, for example, so many features which we discuss, you discuss as described already, part of the copilot, Microsoft, what specific your product is offering for the customer which copilot doesn't have. Thank you.

SPEAKER H

Okay, thanks. I'll tackle the first one first. So when I talked about the challenges around data mapping. I think I raised it as a challenge because it's not something that we have today machine learning around. And therefore it's still a manual process when we're talking about implementing solutions here. And there's a vast variety in the way that certain key data words are described across the industry, and everybody knows the acronyms and everyone uses them differently. And so I think in the future there will be generative AI tools that will predict that it's seen this variety before and it can automatically map the differences together. But I really raised it as a point to call out, and it's a current challenge, it's a current problem to solve that's not today unlocked in the solutions that we're providing. And to your second question, really, because as a software provider, we have these adaptive workspaces that deliver intelligent apps that integrate data from all the different sources that you require, whether that's be third party documentation, whether that be your internal systems of record. And like I mentioned at the beginning, we really don't replace those systems of record, but we rather complement them. I don't think those tools and those applications for the commodity trading lifecycle are available today. With copilot, I think being able to consume your data, digitize your data, and then apply your business workflows to it is the differentiator.

SPEAKER G

There was one more question over here.

SPEAKER E

The topic says that AI has a capability to reduce risk. In the area of reduced risk, can you give an example? So we're talking about operational risk, risks like deal entry, for example, or risk of your trading strategy would produce negative results.

SPEAKER H

So I think for both of us, it's different answers, probably on both sides of the coin. So specifically for us, the risk that's being reduced is decision risk. So as the data becomes enriched with context, clarity, control, you can make better decisions with it, and that lowers decision risk. Additionally, there's other applications of AI that may give you the ability to predict risks or look at patterns, to highlight anomalies, or even, like I mentioned in the future, compare languages to look for nuance in words that meaningfully change commercial terms. Those are the types of risks that we would be looking at in our organization. I think it's a little bit different than yours.

SPEAKER F

Yeah, I mean, ours is very binary. It's basically reduction of cost. So it's not at this stage, super complicated. So you have a better way of optimizing your cost structure.

SPEAKER G

Okay, I think we're possibly out of time. So thank you very much both for joining today.

SPEAKER H

Thank you.