Chuck Doerr: For those of you who don't know, this is actually our 26th FinJS conference. Maybe, maybe it's our first FinAI conference. I think we've had we've had two two talks officially about AI and nine that are touching on AI. And so we're here to talk about data, cloud, analytics, and success in the new age of generative AI and LLMs.

Chuck Doerr: We have an awesome panel here to cover some of these topics. And we're, we're to talk a little bit about sort of where data lives because as we all know you know, you kind of get out what you put in in terms of quality. And so having quality data, the best data in there is critically, critically important.

Chuck Doerr: So we're going to start with some introductions. We'll just go straight down the line, start with David. You know, name, title. Why are we here? What are we talking about?

David Jones: Hi I'm David Jones. I'm the CTO and one of the co founders of ipushpull. We're a data sharing and workflow automation platform that's used by a really broad range of financial markets, participants like banks, brokers, buy sides trading venues and so on.

David Jones: Really critically, I mean, echoing back to what Ned said in the previous panel, we, we try to deliver our service where the client already is. So that's integrations into client's own applications. So that could be ipushpull or an A. An open fit workspace or hooked into a chat platform or direct for API or into a desktop app like Excel.

David Jones: So we provide that kind of workflow and a kind of consistent data experience regardless of the application somebody's using.

Mohana Rajaram: Hello, everyone. My name is Mohana Rajaram. I, my title says that I'm the director for GoToMarket. What I actually do is I'm responsible for our financial product portfolio, industry solutions and strategy.

Mohana Rajaram: My remit is EMEA and UKI, and I'm responsible for bringing solutions to the market, as well as driving our product strategy in the region. And off late, since you may have, you may or may not have heard, at our Dreamforce event last year, We announced data cloud and then we have been talking about Einstein GPT and Einstein AI.

Mohana Rajaram: And then we are rebranding our CRM as the AI CRM. So I'm here to talk about what it means how we integrate data into our CRM, how we leverage the data we get in. You know, in the most trusted and secure way and how we can help deliver the Gen AI solutions and use cases to our customers.

Nick Pederson: Hi everyone.

Nick Pederson: I'm Nick Peterson. I work at NatWest. I'm the head of digital in our sales and trading business. So I look after all end to end automated trading and sales workflows on the business side. I'm here because we're experimenting like everybody else with generative AI tools. And we've had some successes and we've got some challenges and I'm keen to share where we're at.

Chuck Doerr: Awesome. Awesome. I'm Chuck Doerr. I'm OpenPIN's CIO and president. And now just we're going to pop into the conversation here. So, so Nick, we'll, we'll just go right back to you. It's been said that AI won't replace developers, but that developers using AI will replace the developers who don't.

Chuck Doerr: How do we feel about extending that analogy into the sales trading world?

Nick Pederson: It's, yes, it's a nice one. I think I'll start by, by maybe being quite crisp that AI in its general, you know, real general sense, i. e. machine learning. It's really not new in trading. It's not new in financial services more broadly.

Nick Pederson: On the trading side, if you cycle back 10 15 years there were certain asset classes that did have a fast number of traders, and that has diminished, largely thanks to algorithmic trading and traditional applications of machine learning. The asset classes where that's happened are the ones where there is readily available, rich, large data sets with which you can use to train models to construct prices.

Nick Pederson: So it's predominantly foreign exchange and equities, and then you've got asset classes like commodities and structured rate products where there really hasn't been that impact of machine learning in its traditional sense. So I think on the trading side, I'm pretty confident to say that traders will still exist.

Nick Pederson: Does generative AI and the intuitive style language conversations you can have with it? Change the way a trader does his job. I'm not sure. I certainly don't imagine a world where Contextual decision making around trading risk is done through generative AI applications I think that's a far far far far place away, but on the sales side I think what you might see is the same of ripple effects in the construction of pricing In trading that you you will have in sales.

Nick Pederson: So if you think about what a sales person's job is it's to interact with customers To summarize research, provide trade ideas, take cool notes, summarize requirements. All of those things are really ripe for the traditional use cases that we're seeing applied in generative AI. And that's some of the things we're experimenting in.

Nick Pederson: Is there a world where Salespeople don't exist? Probably not. And I think the main reason for that, and we'll touch on it later, I think, is actually going to be regulation. One of the benefits of traditional machine learning in construction of prices is explainability of models. When, when you have a trade idea sent to a customer and you, you can stand in front of a, a regulator and say, we believe this trade was appropriate because of X.

Nick Pederson: That's great. When, when you say, we believe this trade was appropriate because the model produced it, and we can't quite explain how the model came up with it, I think that's a big challenge that the industry's gonna, gonna have to grapple with. But, as, as everyone said, you know, augmenting what salespeople do.

Nick Pederson: I think it's going to be really, really fascinating and it comes down to having access to all of the data that you need, which is something that I think we're going to touch on in a bit.

Chuck Doerr: Fantastic. So, so maybe we'll go over to David for a quick second. What are some of the custom problems that you're working on to solve as it relates to sort of data LLM?

David Jones: Yeah, sure. It's, we're using AI in a slightly different way to a lot of the examples that people have talked about tonight, which is. They've been talking about using AI to interrogate or mine data that they've already got. The kind of question that our customers are asking us is, how can we use AI large language models and, you know, fine tuned models to take the, the structured data, the unstructured data that we're getting sent to us across a whole range of different channels and actually ingest that in a structured format into our platforms, whether that's trading platforms, pricing platforms.

David Jones: And so on. I Mean, the, the background is that our customers use our, our platform to automate workflows with their counterparties. So these, these are, these are workflows between participants in the financial markets. So it could be banks and brokers. It could be sell size and buy sides. In these OTC markets a lot of this workflow is not automated.

David Jones: So people are getting price information or trading information sent through chat or through email, even inside spreadsheets. We've built a very flexible data sharing and workflow automation platform that you can plug into all these different applications and and and chat platforms and can transform and map data into standardized structures and feed into these platforms.

David Jones: But there's a limit to how you can get, how far you can get with traditional parsing or traditional mapping techniques. I mean, our customers are living in a world of unstructured data. I mean, you can imagine if you're a, if you're a broker and you're getting orders sent via chat from your bank counterparties.

David Jones: They're all going to be using different formats. If you're a buy side trader you might be getting access sent to you from 10 different banks across different chat platforms, across Excel. Everyone's talking about the same thing, but everybody's using using different kind of formats and different ways of expressing it.

David Jones: So, how do you actually solve this, this challenge? And this is what we've done, so we've embedded a whole sort of range of different AI based techniques for automating the parsing of this complex, unstructured data. So it's everything from using LLMs so, you can now take unstructured data from a chat platform, feed it up into an LLAM and use a technique like an entity extraction to pull the structured data out from these unstructured chat messages.

David Jones: thEre's also you can also use fine tune models, we're using fine tune models, and using technical NER to extract, to do the same kind of thing, and that's a lot sort of faster. I think when you're looking at workflows where you're looking to apply AI to a workflow, you've really got to consider a whole load of different things about how, what performance you need, what model's going to be the most appropriate and how you can hook it into your platform.

David Jones: And so that's what we've been trying to do to solve these client problems.

Chuck Doerr: Very cool. So, Mohana. Salesforce sits in a really unique spot in the market. Right? So as the as the whole industry has been investing heavily in their, you know, data consolidation strategies, right? In parallel, they're also mass adopting SAS platforms and putting super important data like customer data on it.

Chuck Doerr: You know, what challenges are you seeing in this new era of, you know, sort of data and cloud and, You know,

Mohana Rajaram: tO start that, I need to kind of go back 24 years, right? So 24 years ago, Salesforce was the pioneer in building a SAS based customer relationship management platform, which involved putting customer data on the cloud.

Mohana Rajaram: Doing that with the, with the related security layers and security protocols and having its own hyper force infrastructure. So it's not new to us. We, our customers are, you know understand that and we have proven successful in terms of maintaining that data integrity, residency and all the data, data privacy compliance rules.

Mohana Rajaram: The last 10 years Einstein has been our predictive analytical modeling tool that we have been using. And as a part of our product innovation, we have always had a level of automation and then a level of AI that's been inbuilt into, into the processes that we build as a part of our customer onboarding, customer servicing, advisor management, or you know, insurance claims, for example.

Mohana Rajaram: And what we have just spiraled now into is the whole Gen AI based. Given that, we you know, when we talk about the best use of Gen AI, we heard a lot of use cases today, right? What we thought about is look at the business problem you're going to solve and when we work back to see what a platform capability Needs to be there to support those business problems.

Mohana Rajaram: It's it's starting at the customer You know, it's trying to basically shift the paradigm of offerings from a product based Use case to a customer base use case. So if you keep customer at the center, what we have is a CRM Data platform. We have a lot of customer data, customer information, and then we have all the basic automation and the predictive modeling that's been built and adhering to that.

Mohana Rajaram: What we then went and built is our whole data cloud, which Yeah, as a, which has a canonical model that takes all this information, not just within the salesforce suite of products, but can talk to external LLMs. It can talk to unstructured data sources. It can talk to different you know you know, your own specific models and put in a lot of data and be able to provide that dynamic prompt, which is grounded in a whole lot of context, which is, which is, which is the, which is the important thing.

Mohana Rajaram: We all can write prompts, we all could you know, educate and train, but providing the prompt with so much of contextual information to provide the right output is something that we have been working on, right? Now, when we go back and we talk to our customers, to provide this level of information, the first thing is data.

Mohana Rajaram: It's data and data. And A lot of our customers have undertaken data transformation, cloud transformation as a part of their, their transformation agendas, given that Gen AI is on every CEO's agenda ever since this year. And what that means is, it's providing a foundation. Now, we've seen different approaches based on different challenges.

Mohana Rajaram: The first is, With large institutions, there's a lot of siloed data platforms, legacy platforms missing data. A data which you you know, which is, which, and there's a lot of time which is spent in analyzing, cleansing, figuring out what we need to achieve a business you know, particular outcome. And then we have other organizations which are probably focusing on a specific proof of concept, a specific use case, a specific area of business.

Mohana Rajaram: Now. The kind of challenges we see in both of this is one is missing data, where there's a great Gen AI use case to, you know, to build out synthetic data where, you know, they can help you know, complement that particular data set. The other challenges we see is from a data residency and from a data compliance point of view, it's trying to you know, have an infrastructure alignment.

Mohana Rajaram: So what's important is that. Data AI strategy should be a part of an enterprise strategy and organization. It can't just be a siloed strategy of I'm going to transform my data or get on a digital transformation. We have seen it work best, or we have seen it come through various life cycles in different customers, where it's a part of a larger strategy and AI and Gen AI is used.

Mohana Rajaram: As a compliment to other machine learning models and other AI tools that you have. So the output of one is feeding the input to another, rather than looking at it in siloed. So, so that's kind of a bit of the challenges we see and the advice that we, you know, try and work with our customers. And as a platform company, what we take back is how can we make That data assimilation contextualizing that information and be able to provide that user experience and user interface to help our customers leverage technology to, to drive the outcome.

Chuck Doerr: That's great insight. Nick, maybe quickly hit on like who, who do we get to do all this work? Like, is it, are we looking to our quants? Are we looking to our data scientists? Like you know. Who does the work?

Nick Pederson: Yeah, so I, I mean I think as a general rule, you, you gotta look to someone who's got a machine learning background of some description.

Nick Pederson: So we, you know, we, we look specifically to our quantitative analytics teams. And then, funnily enough, they're now working very closely with data scientists in other parts of NatWest, which before our quantitative trading teams would sit very siloed within the sales and trading business. And not really interact with, for example, the retail bank or the wealth part of our business.

Nick Pederson: So there's a lot of discussion happening there. I think financial services are a few decades, I don't know, a few years behind big tech in that we don't recruit specific machine learning experts for a broad variety of use cases. And I think that's something that needs to change. A small team of quantitative trading experts are pining on generative AI use cases in different parts of NatWest.

Nick Pederson: It's great for now, but I don't think it's something that's particularly scalable in the next 5 or 10 years. And to add to Mahana's point on the data strategy, so the use cases that everyone wants to see from generative AI I think the organizations that will do well will capture on that wave of enthusiasm whilst actually twisting it to be, it becoming a data cleansing exercise.

Nick Pederson: Because if anything I've learned in the experiences we've had the last year on use cases, they can be incredibly good if they're narrow. Once you start to scale them and require other pockets of data, it becomes an absolute nightmare and you have a multi year transformation just to find that data.

Nick Pederson: And that's some of the, the challenges that the quantitative team have been, have been working on. So, to answer your question Chuck, I think Machine learning experts are the go to and the financial services industry needs to start hiring more of these people and take them away from big tech.

Chuck Doerr: So I'm going to go ahead and re acknowledge that we're standing between this group of people and drinks and networking and our pixel performers are going to be downstairs so you can check out their outfits in there.

Chuck Doerr: So we'll wrap it up with some closing thoughts. Maybe start with David and we just come down the line. A parting thought for the audience. Well, I

David Jones: think that Gen AI is, is, is really revolutionizing the user experience that, that people are having when they're interacting with computer systems.

David Jones: You know, I think that workflows implemented using using Gen AI and, and APIs are, are going to be kind of the new apps. Workflows are almost, workflows are the new apps. So you can't really ignore it. So I think that It's critical to make sure that you keep humans in the loop in this process, because AI is still not really particularly explainable and isn't always completely robust, so you need to make sure that you've always got a human there to check your work.

David Jones: But essentially I think the really key thing is to sort of to embrace it and really think big, but start small to make sure that you're, you're, you're keeping people on board as you're, as you're starting your AI journey.

Mohana Rajaram: Yeah. But to add to David's point is, gen AI is here to stay. We all have to learn, adapt, and see how we can inculcate it into our, into our workflows and into our day to day jobs that needs to be done, right?

Mohana Rajaram: Having said that, it's, it's a, it, it, we need to have a mindset of continuous learning and continuous adaptation, both from a human and a machine learning point of view, so that we are able to, You know train the machine and eventually the the end goal is autonomy, but we have it's going to take a long time to get But in the meanwhile, it's, it's, it's continuously learning, continuously adapting, continuous ensuring that every activity that, you know, you would undertake as a part of your daily, daily jobs to be done is, is, there's a, there's a level of you know, I would say streamline it, look at it from how you could eventually automate it and Be able to, you know, upskill yourselves to, to do a more high value transaction.

David Jones: I

Nick Pederson: think I'll I'll touch again on the, on the data piece. So, you know, capitalize on the Gen AI wave of enthusiasm, but actually twist it and turn it into a really good data strategy. Because the use cases and the benefits of Gen AI. will only come with access to huge pools of data. That would be my comment.

Chuck Doerr: We did it. Ten seconds to spare. On time. Big thanks to the panel here. Nick Mohana, and David. Good to see you.