Imagine a world where artificial intelligence dictates your financial decisions; it might be just around the corner. Join host Jim Jockle of Numerix and expert Chandini Jain as they delve into AI/ML and how these technologies are reshaping the world of investing decisions. A former derivatives trader, Chandini is the CEO of Auquan, an AI innovator transforming vast, unstructured data into actionable intelligence for financial services customers.
Listen as Chandini and Jim discuss how AI/ML has helped companies avert disaster through actionable intelligence and what the future of this technology in the investing space could look like. Take advantage of this exciting episode.
Imagine a world where artificial intelligence dictates your financial decisions; it might be just around the corner. Join host Jim Jockle of Numerix and expert Chandini Jain as they delve into AI/ML and how these technologies are reshaping the world of investing decisions. A former derivatives trader, Chandini is the CEO of Auquan, an AI innovator transforming vast, unstructured data into actionable intelligence for financial services customers.
Listen as Chandini and Jim discuss how AI/ML has helped companies avert disaster through actionable intelligence and what the future of this technology in the investing space could look like. Take advantage of this exciting episode.
Welcome to Trading Tomorrow navigating trends in capital markets. I'm your host, jim Joggle. In my decade plus of working with Numeric's Global Leader in Capital Markets Risk Management Technology, I have launched our Thought Leadership Division, a place where insights, innovation and expertise converge, just like this podcast. Through my journey in the financial realm, I've had the privilege of witnessing firsthand how the capital markets landscape has transformed the complex dance of market trends, and innovative technology has redefined how the finance industry operates. With game-changing innovations just around the corner, we now stand at acrossroads, one where it is more crucial than ever to understand the interplay between these realms. That's what we do here. We talk about current and future processes and technologies you need to be aware of moving forward.
Speaker 1:As I've said before, ai seems to be the technology of the year, but one subfield of AI in particular has the potential to transform investing Machine learning. It's not just changing the game, it's rewriting the rulebook. Imagine a financial advisor who works tirelessly around the clock, analyzing vast amounts of data, identifying trends and making informed investment decisions. Well, that's precisely what machine learning is bringing to the table. In this episode, we'll explore how artificial intelligence and data-driven algorithms are changing the way we navigate the complex landscape of stocks, bonds and everything in between.
Speaker 1:And joining me to discuss machine learning and finance further is Shandhati Jain. She's the CEO of Okwon, an AI innovative transforming vast unstructured data into actionable intelligence for financial services customers such as UBS and Federated Hermes. The company was named the hottest fintech startup in Europe at the. Europas. Former derivatives trader, shandhati left the industry back in 2017 to start Okwon, where she uses cutting-edge machine learning and deep learning techniques to solve financial prediction problems. Shandhati has a master's in mechanical engineering and computational science from the University of Illinois. Welcome Thanks for joining us, shandhati Jain. Thank you for having me, robert Leonard. So let's just kick off the conversation with a few questions about Okwon. Where did you get the idea for this type of technology and what was the gap that you were trying to fill?
Speaker 2:Shandhati Jain. Yeah, so before Okwon, I used to work for a fund in Chicago as a trader. The company called Oktiver. They were one of the largest market makers in the world, actually, and we were big consumers of data in that fund and as we were looking towards my book, we basically subscribed to everything, anything that we thought could help us make better decisions Broker research, outside research, news.
Speaker 2:Obviously, information of the company was just closing and we very quickly got into a situation where we just had way more information that was coming on my desk than we had bandwidth in a day to read. I was reading all the time and all the analysts on my desk were reading all the time, and we still just couldn't get through everything. Almost to the point that it became a source of all this personal anxiety that I have a position on and I just don't know everything about it. And I know that that report came in and maybe if I knew, if I'd read that, I might have found something that I overlooked. So that's when we started thinking about this problem of how can you optimize your information flow. Basically, in this world of potentially unlimited data, unlimited bandwidth, limited drink sources, how can you make sure that you get to the information that is the most relevant, the most material for your use case quickly and cut through the noise. And that's how we came and we started thinking about building.
Speaker 1:Well, perhaps you can give us an example of a big event or situation that Oquan helped a client avoid.
Speaker 2:Yeah, there's a few, but I'll give you the one that we actually use the most often because it resonates quite a bit. There's a company called Teleperformance. It's the provider of content moderation services to companies like TikTok. Last year in November, the companies suffered a 30% drop in their stock price because they were accused of having poor working conditions or labor abuse. Now what's interesting is that all of Oquan's customers actually knew about that almost nine months before that happened, because in December of 2021, two former content moderators of TikTok sued the company for psychological trauma At the time, even though Teleperformance was not named anywhere.
Speaker 2:Because, the way our system works, we were able to catch that information. And two, we were able to identify, using an ecosystem mapping, that this information relates to Teleperformance. This was in December. Obviously, at that point, this was like a low risk information. It was more of a nice to know that this is happening.
Speaker 2:But then those lawsuits kept piling up and by March, a lot more content moderators of TikTok had sued the company. This entire time, teleperformance was not even named. But then in August, there was a major report that came out in Forbes that was talking about how these content moderators who worked for TikTok had to watch disturbing videos of child abuse. That is the first time Teleperformance was actually named. We heard all of our clients say this is actually a very high risk situation for Teleperformance. Then, a couple of months later, time Magazine did another expose this was specifically now aimed at Teleperformance saying that these content moderators work in Columbia, they're underpaid, they're monitored through video, they're made to watch these disturbing videos and post that.
Speaker 2:The Colombian government announced an investigation into Teleperformance. The stock price dropped. The company came out and said they are going to exit this part of the business. That entire revenue line got ridden off. What is interesting here is if you followed a traditional research workflow, you would have missed the entire early endorsement because there would have just been so much information that you were looking at. You wouldn't even have realized that this information could be material to you. Even if you realized it, let's say, in August, you might not have acted upon it in time, just because, again, that problem of just so much information being generated, the way our system works in one, collecting a lot of information for our customers that they would otherwise be doing manually, and processing it and cleaning it and filtering out the noise and surfacing for them one what is most material, also sometimes things that are just not obviously connected. That helps them avoid these kinds of situations.
Speaker 1:That's probably one of the best examples I've ever heard in these types of cases. That's actually incredible. Perhaps you can just give the listeners a sense of how this all comes together. You know, because there is so much noise, it were bombarded with it. How is that? Helping investors gather different insights, creating custom data sense? If we look at one specific instance, how long would this process take if you'd with or without your software?
Speaker 2:Yeah, in most cases, without our software, there would be a team of analysts doing this in-house, doing this manually. Basically, the analysts who work at investment banks, who work at asset managers, who work at private equity firms they are amazing, they're incredible. They're very smart people, highly compensated, very talented, very highly trained. Then they just end up spending so much of their time just collecting information. You talk to them and they'll say we just spent so much time scouring the internet trying to find data points which could be relevant to an investment decision that we're trying to make. A lot of time would just be spent that.
Speaker 2:I'm looking at this company. Let me try and identify if this company has any regulatory finds anywhere. Does this company have any on where you know suits that could affect the valuation of the company? Who are the suppliers for this company? Is there anything questionable with this company's supply chain? All of that work is done today manually.
Speaker 2:So at your time you were talking weeks to collect this kind of information, just because one, there is just so much information that is being published now that you have to look at. But also, second, what we're hearing from a lot of our customers. They're realizing, in a world that is increasingly interconnected. It is not enough to look at just the company anymore. You have to look at the entire company value chain. You have to look at what the company subsidiaries are doing. You have to look at where the company supply chain is. Where is the company sourcing its raw material? What are the company's customers? Are there any threats to the company's revenue based on any macro trends that might be happening in its customer base? And all of that then just adds to that information overload problem.
Speaker 1:So I guess one of the questions always is the data, the underlying data itself and different access points into the data. I mean, how is a platform like yours connecting into such a vast ecosystem?
Speaker 2:Yeah, so that's a question we get all the time. So you guys claim that you can surface so much information. How are you actually doing it if our internal analysts can't get through it? And that's really where the role of AI in automation comes in. If you think about a research workflow, it just starts with identifying where relevant information could exist, then collecting that information, bringing it in-house, cleaning that information, doing lots of joins across different datasets to say, actually, when you say Apple, then when you say Apple Inc and when you say Apple Limited, they're all talking about the same company, so all of this information is about the same company. That work doesn't have to be done manually. All of that work can be left to machines to do. You can automate the data Again. You can automate connecting to relevant datasets. You can automate bringing the information in-house and cleaning it. And then even the lower level cognitive work, which is just understanding basic understanding of all of these references are actually about the same company. All of these data points are actually relevant to a company. This data point mentions LinkedIn, but LinkedIn is a subsidiary of Microsoft, so this data point is relevant to Microsoft. You can also leverage AI to do that. And then what you need for these extremely trained analysts is just that higher value analysis, drawing conclusions, making decisions, making predictions, which is what they're actually good at. So that's really the principle that OCOIN system works on.
Speaker 2:We connect to over 2 million sources of information online, and all of these are data sources that we think could have anything material, anything relevant from a financial perspective. So it's everything from company filings, regulatory filings, government databases on, let's say, health and safety fines, any financial fines, allegations of corruption, bribery, money laundering, international sanction lists, ngo reports, news, very local sources of news as well, cell-side research, industry research. So it's a lot of data which could be financially relevant. We bring it all in-house and we run it through our RAG AI system. Rag is basically retrieval of mental generation. It's a technology that is pioneered by Meta, which overcomes all of these shortcomings of generative AI that have prevented widespread adoption of financial services, and then, from running it through the RAG AI system, we're able to one create an underlying data store of information that is relevant to a company, but also clean out a lot of the noise, information that is not about the company, that is not financially material, that doesn't solve for a person's use case. And once we have that underlying store of information on the company, then when a user comes in, we understand that user's context. So we would be able to understand if you're logging in as a private equity analyst or if you're logging in as somebody who works in KYB, in an investment bank, because you will want very different types of information.
Speaker 2:The example that I give about that cleaning up of noise and customizing it to a use case is if you went in search for Nokia on Google right now, you would see everything from here's how my new Nokia 6600 does, to the Black Friday deals on Nokia, to Nokia lays of 5000 stuff. The only thing that is financially relevant is Nokia lays of 5000 stuff. So somebody from the financial world, you don't want to see. The other two you only want to see because otherwise I'm just creating information overload for you by connecting to lots of data sources. But then what you also want.
Speaker 2:For example, if you were searching for information on Microsoft and now we only left with financially relevant information, let's say you had two things. One said the cloud revenues of Microsoft, or some projection of cloud revenues of Microsoft or how they would do against AWS, and then the other piece of data was talking about the decarbonization strategy that Microsoft has in place. An equity research analyst cares about the first one, but a sustainability analyst cares about the second one. So to be able to be customized even further, so that a user sees information that is relevant, financially material, but also customized specifically to their use case, to save them loads of time.
Speaker 1:So, as a former trader, I'm sure you wish you had a back when you were a trader. How much of your and your team and your expert knowledge and experience is critical in the training of the AI to get to that level of appropriate information.
Speaker 2:No, it is super important.
Speaker 2:That is actually one of our very specific modes.
Speaker 2:In the business, you have a lot of generic AI systems that are able to do the basic filtering, but to be able to look at datasets and say this information is financially relevant and this information is not financially relevant, this information is relevant for this use case.
Speaker 2:This information is not relevant for this use case. Even simple things, like I talked about the connected ecosystem that we're able to connect information about suppliers, about subsidiaries, but let's say, we all know that TSMC is one of the suppliers to Apple, but knowing that anything about TSMC announces new sustainability initiatives is not relevant to Apple. But saying TSMC foresees a shortfall in chip production next quarter is relevant to Apple because that actually directly affects Apple. So there's a lot of these nuances when I say cleaning up noise, identifying information that is relevant, identifying information that is customized for use case, which started with, obviously, my expertise, but over time they have required my team's expertise, who are qualified and being able to answer some of these questions, and then also our customers expertise, who have helped us figure out what information is relevant, what information is not relevant, what information helped them make a good decision, what information is something they could not see, and all of that is baked into our system today. That's what makes it performant.
Speaker 1:So how are you looking to transform, investing?
Speaker 2:We want to make it more cost-efficient and we want to help our customers identify information that they otherwise wouldn't have access to, because it is coming from sources they wouldn't have looked in before or it's coming from connections that they wouldn't have known about before.
Speaker 1:So I would be remiss if I didn't ask a couple tough questions. But what are some of the pitfalls of using these types of technologies?
Speaker 2:So I mentioned retrieval, augmented generation, and that is something that we so when we started the company we started with, we started FIGERSBACK, so we started with the natural language processing because that's what everybody used to do then, and then our birds came out and language models came out, and then we progressed to language models and then large language models came out, and then we progressed to large language models and then everybody's seen the general shortcomings of generative AI. I think the fact that they have a standard out-of-the-box large language model has a training cut-off date, which means anything that happened beyond that date is not baked into the response that it is giving you, the fact that they tend to hallucinate and they tend to give you linguistically coherent but factually inaccurate responses. So I think, as a general sense, I would say that just being aware of the limitations of the model that you're deploying is extremely important. Otherwise, you might end up misleading your customers or giving them inaccurate information, which is an absolute non-on-financial services. For example, if you're developing risk AI co-pilot for risk manager to help them understand the risk of their portfolio investments, you have to make sure that the information is based on the most accurate data. It factors in the most recent macro changes, like interest rate hikes. It factors in any changes to regulation. It factors in any changes to company specific information. You have to make sure the information is actually accurate, it is coming from credible sources, right, otherwise they will not deploy it. Or they do deploy it in production. It could lead to significant losses. It could lead to regulatory fines.
Speaker 2:So I think recognizing that for any such system to be worthy of financial services use case it needs to it 100% needs to be credible and accurate and timely is important, and then making sure that the technology that you use is actually able to fulfill those needs is necessary. So we started experimenting with large language models, but then we realized that they have these limitations so we can actually deploy them into production, and we waited until we found something that we thought was the answer to timely, credible and accurate. And that's where the whole retrieval of mentor generation comes in. You'll hear about it a lot, I'm sure in the next, in the upcoming years, everybody else starts to realize the challenges of generative AI for enterprise. That will alongwind. An answer to your question. I just get very excited when I talk about this.
Speaker 1:No, your passion is totally coming through and I love it because you're getting me excited about it, and so, as someone in capital markets technology, we're still seeing institutions who have trouble moving to the cloud and putting mission critical systems on public clouds and things of that nature. What are some of the barriers or blockers for financial institutions in terms of fully embracing these types of new technologies?
Speaker 2:No, I think. First I would say this from my side and, I think, anybody else who's selling technology it is good to recognize that we're not selling technology. Financial institutions don't buy on the promise of this is world changing technology, or this is like some really cool technology that's hyped. They buy on a value. They buy on a problem that you're solving for them. So when I am selling to my customers, I'm not selling them like large language models or generative AI or AI even. We're sending them a way to drive efficiency in their work rules. We're selling them a way to save time. We're selling them a way to find information faster than they have been able to. We're selling them a way to find information that they previously didn't have access to. And I think that probably is the first barrier in selling to financial institutions is recognizing that they're not buying technology. They're buying a solution to a problem that they had, to a workflow. And then it is obviously this everything that you've probably heard before these institutions. They're slow by design.
Speaker 2:I don't want my bank to be innovative. I want my bank to not fail. I want it to be safe, which means that there is a lot of control and there's a lot of testing before they will onboard a solution. Also good merit there are long procurement cycles because they have to make sure that everything, because they're regulated, they have to make sure that everything that they do is actually okay by the regulator. So that means as a startup, when you're selling to a financial institution, you have to be prepared to be one enterprise, ready to sell. You have to make sure you have the right protocols in place, you actually fulfill the enterprise requirements and say can you just have to be prepared for long sales cycle?
Speaker 1:Well, sadly we've got to the final question of this podcast. We call this segment here to the trend drop. It's like a Desert Island type question. So if you could track only one way to use AI and ML technology within the markets, what would it be and why?
Speaker 2:Yeah, I mean I would say it's the problem that we're trying to solve, because obviously that problem excites me. I mean, up until now we didn't have the ability to have natural language understanding and it's only in the last few years that we built that ability. But what that means is a lot of these knowledge intensive workflows in financial services, things like equity research, credit research, but also things like KYB onboarding a customer doing assessment of credit to decide if you want to give loan to a counterparty. These are inherently just time-consuming and cumbersome.
Speaker 2:I joke around with my friends but nobody ever grows up saying you know when I want to be? I want to be that guy that approves if a company should get a loan or not. That's just not fun to do. So to be able to provide an assist, an AI enabled assist, to a lot of people who have those jobs that helps them be more efficient in their work. That helps them automate a lot of the mundane, the time consuming, the cumbersome, the not fun part of the job and focus their time on the higher value analysis. The things which actually be as humans are good at drawing conclusions, making decisions. If I had to track one thing, that's what I would track how have we been able to free up people's time using AI?
Speaker 1:Well, let's just hope no loan processors are listening to the podcast. I love it. I love it. Well, you know, first and foremost, I want to thank you. I mean, this was illuminating. I think you know the whole goal for us is to spark ideas and talk with people who are bringing new and exciting things into this world and creating, and you certainly are doing that. So thank you so much for your time today. I really, really appreciate it.
Speaker 2:Now, thank you. Thank you for this opportunity. I really enjoyed chatting with you.
Speaker 1:Coming up next week. We'll talk about increased volatility, how it's changed the energy market and the way some companies trade. It's a can't miss. But first, if you enjoyed the podcast, make sure you hit the subscribe button, leave a comment, a like and check out our other episodes. Thanks for joining.