Welcome to insiders in focus where the shield insiders like to share a little bit of information with you guys that maybe you don’t know. Not so scary. But really the purpose here is to not talk about things for about an hour and a half, because who has time for that, but to get really straight and to the point in about fifteen, twenty minutes and today’s topic, well, no topic has made more noise in recent days and months than artificial intelligence. It’s been well adapted in many industries for years. But with the rise of chat GPU and other products, we’re seeing massive adoption and integration of its use across every sector and even by private individuals. It’s been a hot topic in compliance at financial institutions for some time with strong opinions all around. So for this conversation, I’m excited to have some of the brightest minds in the industry share their perspectives on the market on ways to approach AI. And how to get meaningful value from AI at scale. Let me welcome Oliver Bradford. We’re gonna call him Ali just because we like him that much. Senior Director, strategic accounts at shield. I can tell you few people know the market and the people in it like Ollie. If you haven’t had a chat with him yet, I’d recommend you reach out to him on LinkedIn and just say hi or something. Shanit Levin, VP, Data Science at shield, brings with her not one, but two PhDs and a very shiny new cloud innovator of the year award from the cloud awards. So congratulations on that. Shamid is probably the smartest person I know and the smartest person that anyone I know has known so we’ll get a chance to hear some of her very smart things to say and hopefully we’ll be smart enough to understand what it is that she’s trying to teach. So we’ll we’ll get to that. I’m I’m very excited. So, Ali, let’s kick it off with you. There’s a lot of solutions out there that are marketed to the financial compliance sector as AI, machine learning, NLP, What are you hearing about the adoption of AI in the market? That’s a really good question, David. I guess it’s, I’d say first of all that there’s a huge amount of interest, you know, from the market, from banks, energy commodity trading firms, asset managers about AI and how it can be used and how it can benefit them. And there are a lot of vendors providing this in the marketplace at the moment. Overall, I’d say we’re we’re pretty pretty early on the adoption curve. Institutions in the financial sector started rolling this out probably two thousand sixteen, two thousand seventeen, and really then it was, strictly within the kind of tier one and two kind of global investment banks. And the they rolled out, both vendor solutions and and in house solutions. And really, it’s been kind of mixed results so far. There have been some very good results, but there’s also been some very, very long, you know, deployments we’re seeing, you know, quite a lot of skepticism from the people who initially roll it out to coming around to, you know, maybe relooking at at what they’re doing and improving things. And, yeah, things like chat chat chat GPT, you mentioned it. You know, I really feel like we’re kind of at a at a tipping point in terms of AI at the moment. This is what, you know, I hear from the market. And it’s kind of accelerated even even further the interest and people wanting those better results. And the the kind of whole compliance function, I guess, is right for automation and right for kind of AI being used because, you know, there’s there’s loads of manual processes so, yeah, that’s that’s what I’m kind of seeing and hearing at the moment. Yeah. Thank you, Ali. And I think you you speak to a very important point about how every organization is coming to compliance from a different perspective is coming to AI from a different perspective, and vendors come in with their own, like, this is how to do it, and that’s how to do it. And obviously, you know, I’m sitting here, we’re sitting here talking about, you know, coming from the shield perspective, but trying to just just take a step back from the vendor side of things it is a challenge of of how how do we as a as a financial institution get started with this? How do we adapt from where we’re starting from? I think I think that that’s You you you really, I think, nailed it over there. Show me, what are some of the limitations with some of these AI products that people need to know about and look out for? Okay. So maybe we’ll start from the evolution of AI in the financial industry. Yeah. This is a relatively traditional industry started originally from when the need came to or large amounts of data, and the humans will not sufficient anymore. The the initial solutions will what we call lexicon based, collecting manual collection of a a phrases or rules, to try to detect them in conversations and they highlight the risky conversations or the ones that should be manually reviewed. These solutions obviously did not always produce the best results, ongoing manual collection of examples is very tedious. And the the for a few years, the, are attempts to try to use AI in this industry. And the initial attempts of using classical machine learning, what we called on one hand, a promise to show good results and the ability to better generalize. On the other hand, in this refers to what Ollie was mentioning on the long deployment processes required a a vast amount of investment in data collection Emily, machine learning started, from being able to learn from examples. And the problem was to collect those examples out of large amounts of data and purposes. And it was the the fact that companies sat for months and sometimes even years trying to collect enough data to create their own models. So, this was a challenge and it created the somehow, but some frustration in the industry, trying to hold to hold on to, the more traditional list solution. They’re easier to adopt a quicker to change and maintain. We know that this industry is in the place changing regulations and changing use cases where the banks needs to constantly adapt providing a AI models that cannot be easily customized, is a challenge for those, institutes and, the so the first channel first attempts of using AI in this industry with mixed emotions. And, I think that this was the case until, about a year ago or or a few months ago, again, the sentiment in the industry, the same as the sentiment of in the entire, I think, modern world is changing, due to the visibility of charge and we’re at a new place right now. The technology is already, to some extent, available in the past one, two years, but now being exposed to the public. And now everybody knows what it can achieve and do. And we’ll end the facing breakthrough in the in the capabilities that we can sell. Thank you, Shlomit. The fact is that with the vast amount of data constantly being generated, we are seeing more and more adoption of AI there just doesn’t seem to be a good way to monitor all this data and identify risks at scale without without it. It just it just isn’t a way. So question for you, back to you, Shlomit, is are there better, more practical ways to leverage AI that really can make an impact on how firms identify risk what what are the ways to to take advantage of AI? Okay. So let’s go back again to the evolution of AI. As we said, initial attempts for classical machine learning and a lot of work in creating models that are being frozen. The new generation of AI, and the ones that also also serve models like ChatGPT are based on large language these models already learn a lot of human knowledge and a lot of knowledge of human language from huge amounts of purposes, even prior to being adopted to specific tasks. In the case of a compliance teams, we need to, solve the problem of identifying specific financial risks. We have a very specific target. It is a within domain expertise, and we need to to address it. But, when being able to use the more advanced language models, we can already start to train them from a a much higher, knowledgeable, more knowledgeable starting point. And quickly adapt to our own specific solutions. So this is this is technologies that, again, is already available already also available in some of the more advanced, compliance, technologies and, is starting to create a real breakthrough in this industry. The next phase of, using actual generative models is, creating a lot of interest in this, domain. And, especially in the ability to customize to specific questions or or to ongoing investigation of a specific, use case. And the this I think is something that we’re facing, as a changer, a deal changer in this industry. It will, swipe us all towards the future. Thank you so much for me. Yeah. There have been tons of advances. I mean, we’ve all we’ve all anybody who’s even, like, maybe heard of chatty with you knows the world is not the same. And what we can do is not the same. So, Ali, let me let me ask you, when it comes to the actual impact that these technologies have in a business Can you give us some details of what that impact is and what and what kind of approach, you know, that that firms can have in order to take advantage of this? Sure, David. Yeah. I mean, I’ll probably summarize three main points, I think. So the first, and this is, you know, a big, I guess time to value, impacts, which is really the the kind of overall deployment of these these types of solutions. So using the techniques, which kilometers mentioned, you know, we can get live deployments in, you know, three or four months. So, even two months getting, you know, data into the system and then sort of four months for the alerting side. And this is all through the kind of pre pre training of, models out of the out of the box behaviors, and the kind of inbuilt AI that we use. So you can really, you know, get those, get the alerting through to compliance quicker, and get get kind of value otherwise it might take, you know, up to two years sometimes. The second one really is around, you know, actually actually kind of finding more escalations. So it it’s one thing kind of using AI to to kind of reduce the false positives. I’ll talk about that in a second. But really, you know, what, what our clients want to be doing is is actually kind of finding instances which are interesting, worth escalating, you know, what we call true positives. And we can, you know, using techniques like semantic detection and and NLP, you know, we can really kind of accelerate that and increase that. And the last one, this is probably the, you know, the biggest, you know, impact. And that that’s all around, you know, the operational side and efficiency. So if you’re moving from a non AI based solution, even ones, you know, with AI, we see we see a lot more efficiency And, you know, often when we we talk to a kind of a new client that isn’t using AI, they’re maybe having false positive rates of you know, ninety nine percent of of the alerts that they see are just noise. It might be, you know, as officer replies, disclaimers, that sort of thing. And it’s and it’s just a a waste of everyone’s time, and money. And, you know, really, what what value people kind of adding in a process if ninety nine percent of what they look at is, is noise. And we we see that rate coming down by sort of ninety ninety five percent and what that allows you to do, these compliance teams is really kind of focus on, things which need us true positives and that sort of thing, and we can get alerting rates less than, I think we’re seeing less than one percent of total communications. And, yeah, that’s all done through, AI, NLP and different sorts of AI techniques. And also, not sure I’ve got time to, to cover this, but also very important. And, you know, the the EU is is, is really hot on, kind of AI and how it’s used in transparency, and this we’ve seen this filtered down to to regulators as well. It’s knowing why why something is alerted. And, you know, we it’s it’s very important when when you see these alerts for the, the the person reviewing them and for internal audit teams, to be able to to know exactly what is alerted, and to be able to explain this internally, externally, and also record, you know, the, record this. And thank you, Ali. And I I mean, those are all really important. And I think what what ultimately can come down to for organization is are you focusing on risk? A lot of times, there’s just the activities you need to do to support a regime here, but But is the focus on what you need to do to support it or actually what you get out of it, which is focusing on risk? And what what does it mean for your you know, for for your review quality if the vast majority of what you’re looking at is really relevant anyways. What, like, what does that do for your sort of you know, get I I don’t know what the word is. I mean, it’s like going by road. Agreed. Yeah. I mean, it’s not a great way of managing this. Is it if you if you’re showing to regulator world, this is these are all processes. But ninety nine percent of what we look at is, isn’t really a nothing any risk. So Yeah. I thank you for sharing those out. Yeah. And that is very important. For an organization to be able to make that shift from, like, how are we managing all this work into actually, like, getting proactive, seeing what you need to see, analyzing it, getting those insights. So thank you very much. I wanna move to part of a conversation that you like to call five and five. We’re gonna try to get some quick questions and quick answers out the door. So guys take a quick deep breath and run through as fast as you can. So let me spoke about modern advancements in maturity and AI methodologies over the years. What are some of the historical innovations in AI that we saw before this new era? Okay. So again, the first attempts of the AI were to use classical machine learning, collect again a lot, a lot of examples in order to train, your own models. This, seem to be very tedious work and to some extent did not even converge. The challenge in the financial risk strays that we are trying to, detect a needle in a haystack. We were not trying to differentiate between black and white or, very clear, two classes of a domain. We’re trying to find the rare cases. And, in this case, the ability to collect large amounts of the examples, seem to be, really challenging. Another challenge of this industry was, in the early adoption of a language model as well. That, actually, financial talk is a very unique talk. It’s not straight English, straight, a lot of internal term terminology, a lot of the shortcuts, to the laymen, sometimes it’s, not understandable at all. So even the models that were trained on pure English or on a classical English, will not fully, fit to perform this, these tasks because the conversations are completely different. So, it became a challenge to create, both advanced technology with already leveraging a vast amounts of what we call unsupervised data, but that is fit to the financial domain. And, this is, where we are, today? Back to you, Shlomi. Where are we gonna be in, let’s say, five years. It feels like an eternity in this field, but maybe like the next couple of years, where do you see the next advancements coming from? Yes. So I think that, we cannot avoid, from adopting new or newer technologies as everybody is filling in their day to day or read about in the paper. The future is already here, and the ability to communicate with models is already accessible and feasible. So if in the past, we needed to be able to, really carefully curate the questions that we are trying to collect answers on. In the future, I think that the compliance teams will be able to do the room, work, more in a more communication way with, models. We are not there yet. It will take us a little bit, time to do the sleep, but, definitely, we’ll be there in a tour yeah, less than five years social. Thank you, Shlomid. Ali, let’s do a mythbuster here. Can you reflect on the AI solutions out there in the market? Are they worth the investment? It depends. I’d say, normally, for sure, we’ve talked about some of the advantages like you know, finding more true positives, reducing those those false positives and generally kind of improving the efficiency and effectiveness of of surveillance. I guess you gotta think about how how you manage it without AI, and really it isn’t as effective then. You know, I’ve talked to tons of, people in the market who are not using it, and they to kind of not create an operational issue. They scale back the Mexicans, to reduce the noise, you know, use samples, do random sampling, you know, manually listen to phone calls and that sort of thing. So it’s generally, that’s not a not a great approach to risk. And generally, regtech firms offering AI are gonna be kind of more future ready definitely worth talking to, but I’d say, you know, do your due due diligence and, there there are different scales of, of firms kind of having experience in this having implemented it. So, you know, you need to see more than just a a slideshow. And also it’s not just AI that kind of makes your surveillance, effective. You know, you need perhaps better search, better case management, better data management, and that sort of thing. So it’s kind of, you need to look at the the bigger picture and the the kind of holistic, view and not just focus on AI. Yeah. Thanks. That’s a good call out, Ali. And and following up on that. What advice do you have for people who are in talks with vendors or thinking about reaching out to some of these vendors? You know, looking out for AI solutions, what questions should they be asking? So, I mean, Shlomit and I have, I’ve gone through this many, many times with, you know, with different, financial institutions, you know, different customers and so on. I’d I’d say the top ones are, you know, what technology powers you at AI, and what is the goal of using it? Why why do you want to use it? And then from there, you know, AI is really an umbrella term. It’s, it’s kind of like a a catchall for, you know, where were a futuristic vendor because we have AI, but but, you know, do they use, named entity recognition? Do they use large language models? Do they use NLP? And and if so, how is the the data kind of trained and labeled? What data sets are they using? Do we need to kind of label our own data, or do you label it? A big one really is around effectiveness. So you need to be able to kind of, for us, we need to prove how, you know, we measure the performance of the models, how precise they are, you know, quality accuracy, that sort of thing. And from the client side, you wanna know, also these things have, to measure the effectiveness. And I guess kind of seeing some real results from the field as well, you know, what what results do you see from your clients, what improvements can we see? And then finally, what is the roadmap? You know, it’s a fast changing space. Are you incorporating chat, GPT? You know, what what other sorts of AI you bringing into your to your solution in the future. That’s, also really important. So, yeah, it’s just a a a kind of a few things that we’ve been done along the way. Thank you, Ali. Show me how can regulators leverage this newer approach so that they can better understand the risk models that are being adopted by banks? I don’t think that, regulators limit the usage of AI at all, but, to the constraints that they give the financial institutions in terms of the model governance and the ability to explain how do you cover the risks to some extent, deteriorate some of the customers from shifting towards a new AI technologies that really need different methodologies of governance and evaluation. Again, the future is already here, and, we will all be, required to use the advanced technologies simply because they provide us a much better solution. And it will become a regulator requirement as well. I give I can give an example on the voice surveillance, regulations that only last year, the FCA have provided a a large fine for a big, a broker’s company that, did not use full evaluation of, the voice recording, simply because they said that they didn’t have a good enough transcription technology. This is no longer the case. The solutions that runs a AI based solutions are already here, and the regulator will eventually force you to use them because they do perform better. However, and this comes to the question that you asked early before, when you come to evaluate a vendor, you must make sure that this vendor has the ability to first cover the real issues that are being looked at by the regulator that they have the experts in the industry that enables the model to answer the correct questions and that they have the governance methodologies in place that will enable you to give the regulator the right reports and explanations on why your models are performing well and not only the most advanced technologies. So in essence, you will need both. Thank you, Shlomit. This sounds like the opposite advice I give my kids which is just because you can doesn’t mean you should. But when you get to this place of technology where you can, then, yeah, you’re gonna have to because otherwise, just not doing enough. What you’re able to do becomes what you need to do. That’s just that’s just a trajectory. Thank you. That’s very good insight. Just to wrap up here. You know, as data piles up faster and faster, I think it’s worth noting that some of the key benefits of AI that we talked about today are actually dependent on embracing a wider view of what we can do with the data generated by our organizations. And we see more and more firms understand the power of the data they’re sitting on, not just in terms of the potential risk sitting in there, which which isn’t important. But the opportunity available by unlocking real meeting behind that data. So just to tell you from the the shield perspective here at shield, we’re we’re always talking about reading between the lines. Like how a data platform can capture that data and enrich that data, make it available in more and more meaningful ways that can really unleash that value so you can read between the lines in these un con unstructured conversations and better understand what the intent is, what are they trying to say? What’s the context here? I think that’s why it’s becoming more and more of a natural conversation to have about data. What can my data tell me? What analysis can you run-in this data? What can this data reveal to me? We hear these questions all the time now, but a few years ago, this is more like aspirational, maybe. So compliance has traditionally been more of an an art than a science I think, but with the right environment for our data, we can actually wear the lines a bit and maybe bring more science into that art. So thank you, Shlomit and Ali for sharing your insights on AI and financial compliance. This was an awesome conversation. Let’s keep the conversation going, hit up our website, and let’s have a chat. This is shield Insiders in focus. Thank you for watching.