Why advice firms should pay attention to AI compliance

(upbeat music) - Hello and welcome to the FT Advisor podcast. Each week we'll be joined by guests from the financial services world to discuss the most pressing industry issues. I'm Amy Austin, news editor at FT Advisor, and today I will be discussing artificial intelligence and how firms can ensure compliance when using this technology with Helena Wardle, founder and CEO at MoneyMeans, Dickon Johnston, CEO of Themis, and Joseph Twig, CEO of Avini. So welcome to you all, thank you for joining us today. So to kick off, maybe Joseph we can start with you. Could you maybe explain to us and the listeners and what is AI compliance and why is it maybe important to advisors? - Yeah, sure, so thanks for having me. The way to think about AI in compliance, there's actually multiple ways to think about it. So you've got the impact of AI on internal risk and compliance teams. So what does that mean for them? The way that they work, the way that they monitor compliance or to mission. You've also got to consider the impact of AI on the overall business. So how is AI being deployed by advisors, by para planners, by other people around the business? And then you've got to really think about the compliance of the tools themselves, right? So we all know that large language models are black boxes and come with a degree of greatness. What does that mean in terms of deploying AI for specific activities, especially regulated activities? And then you can't look anywhere in the context of AI without seeing AI agents being mentioned. And this transition to regentic AI is going to be a fundamental shift in the way we think about compliance. And it's going to be a very interesting space to watch going forward. - Sure, and Helena, why do you think advisors have to kind of watch this space? 'Cause they might just get AI into their firm and think, "Oh, you're done dusted, let's use it." - I think the biggest challenge I've seen from various sorts of conversations and conferences is that people actually need to really understand the limitations of the technology. So a cool part of that is knowing what it can do well and what it can't do well and shaping the business policies around that. It also informs the due diligence process if you do that. So a challenge that I've seen from struggle with is really understanding that when they use a technology partner for solutions within their team, that they really understand how is it going to work in terms of data ownership? What will happen if they want to change provider? Can they actually access the data or is there going to be any controls on it? Knowing and understanding if that data is going to be used to train the models that they are using. So is that business going to be using the assets that they're giving them in a way to inform how they're actually building it? And is that going to be used in a way that could disadvantage the business or the clients? And really just trying to get under the skin of it a lot more because I think when we use it and see the use cases of it, people can get quite excited and think, "Okay, this can solve some big problems in the business." But if we don't approach it with a real critical hat on by looking at, "Okay, how does that actually sit within how I'm using my customer data? How are we actually looking after our clients?" I think that is quite a big risk and I think it starts with really understanding what the tech does well and what it doesn't do well. And that really shapes the ability to ask good questions of the businesses that you work with and ability to just really make sure the technology suits what you're looking for. And I think that's quite a big challenge that there's a real lack of understanding of it other than saying, "Oh, it works and it does this." But actually the underlying black box, as Joseph called it, which is true, but there's a lot of guardrails and things that the businesses would use who built the tech and you need to get under the skin of that and really understand it, is my view. - Yeah, and I guess that's the thing with understanding it. That's when you can kind of know, is it following this rule? Is it doing what we want it to do? That's when the compliance really kind of comes into its own, isn't it? But Dickham, with Helena was mentioning there about kind of data, is data one of the big things when it comes to AI compliance? Is it kind of the data is what really needs to be made sure it's being looked after? - Well, thank you so much, Amy, personally for inviting me to join you on this podcast with the advisor, which is a privilege. I think you're right, and data is really, really important. And what AI allows us to do so effectively is actually assimilate millions and millions of data points from all over the world, plus look at sort of behavioral pattern analysis. And specifically in our field, that's a huge positive force in the fight against financial crime. So what it allows us to do is really detect those hidden patterns that criminals and organized crime groups are deliberately trying to obfuscate and hide behind. - Yeah, and do you think Helena as well, what kind of ethics are there that, you know, the advisors and, you know, even just normal people using AI kind of have to comply with when they're bringing these tools into their firms? - I think the core part of why advisors would be using technology like this is to create efficiencies to serve customers better or clients better. That's ultimately their aim. Whether that's to check whether the work is done correctly, whether it's to produce sensitivity letters, et cetera. So you have to view that in the same way as what kind of integrity would you want from a team member in your team when you are looking to get them to help you with those kind of things as a real person. So if you had a para planner writing your suitability letters, you would want them to conduct the work in a way that meets your standards of your business and the integrity that you're looking to set. The one thing I think is really a helpful reframe of AI as a technology is to think of it as a new team member. AI is a new kind of labor. I really like that as a view. So if you were interviewing or bringing someone new into your team, you would have expectations of conduct. How do you want them to be? What kind of tone and professionalism would you expect from them in your firm? How would you want that to come out in their work? So I think if you look at it more from that angle, which I know is digressing a bit, that helps with the compliance because you would want certain integrity within a team member because we're working with people's money. And it's the same approach to take on this in terms of the standards and ethics of the solutions that you're doing. We have to recognize that it's basically technology that predicts language. So it's no different to a cash flow tool if you look at it from that simplistic way because all it does is it takes data and predict what is the next best response for that. And we have to think that that data that goes into it, the training methods, how it's actually built, what is the sort of structure of how they've approached it, will lead to the outcomes you get. So that's a really good starting point of understanding how was this technology built? How was that actually created? What data have you used? How have you sort of set the guard rails? How have you set the way to evaluate your model? How are you checking its work? How are you making sure it's correct? Those are kind of questions that if you work with a good firm, they should be able to answer. And that's sort of where I know the ethics question could be broader more in terms of the ethics of AI generally, but if you think about it as the ethics of a firm using technology for the outcomes they want to create for the clients, that's the sort of angle I would take on that question. - Yeah, and Joseph, with kind of the look of ethics and AI compliance, our firms are having this good AI compliance. Is it making it easier for clients to kind of get on board with AI, would you say? 'Cause I think we did go for a period where people were like, I'm not going anywhere near AI. I don't want it anywhere near my money. I don't want it looking at my data. Whereas now I feel like people are a lot more on board with AI. - Yeah, and I think that's just a general familiarity with the technology through retail, large-language models and hyperscalar models like ChatGPT. So there's a general familiarity and understanding of what the technology is now. And so there's less. People have a high propensity to use the tech. I think from an ethics perspective, it's quite an interesting one. I think in the era of co-pilots, where the financial advisor or the power planner retains the risk, effectively, the AI is just getting you to the end point faster. But all the risk is returned by the human, by the individual. And therefore, the focus on ethics within the models themselves, although important, is transitioned. The risk is transitioned to the individual. If we move beyond that into agents and to autonomous actions taken by AI, it's a very, very different landscape. Effectively, if you think about at retirement pensions and you think about these models having been trained on the internet, how do you ensure that the right bias, not just any bias, the right bias is encoded in those models when, for example, you're coming up with retirement solutions for women. Women may live a little bit longer than men. They may have different retirement solutions. So you've got to make sure that models, when they move into a decision-making environment, are reflecting the bias that you expect to see for the benefit of your clients and the right outcomes for your clients. - Yeah, and Dickon, with the kind of, how is this new technology? How is AI reshaping compliance? And what is maybe the fraud risk that advisors will face? - So great question. And maybe there's two parts to this answer. So first, one of our goals at Themis is to democratize due diligence. So we want everybody to perform due diligence on any new professional relationship, whether this is a potential client, supplier, third party, or investment. And if we all do that, if everybody does that every day at work and at home, it becomes almost impossible for criminals to hide in the shadow economy without being detected and reported into local law enforcement. Now, you asked specifically about fraud, and that's a very, very important threat for both advisors and consumers alike to be aware of. Earlier this month, we announced that Themis has acquired PASABI, a leading agentic AI fraud monitoring platform. And this is very much part of our wider strategy of bringing all of your anti-financial crime checks into one core management platform. But why is that so important? Well, sadly, fraud has become one of the most pervasive forms of financial crime. Imposing costs estimated around 4.02 trillion pounds on businesses, economies, and societies. And we mustn't forget that criminals are also using AI to further their own illicit activities. And fraud very much plays a catalytic role in the wider organized crime landscape, providing the initial funds for them that then require laundering through sophisticated networks and all griest crime groups. I think for your listeners, I'd really try and paint a picture here about financial crime. We're not talking about well-dressed, respectable looking cards on the French Riviera à la Dirty Rotten Scandals. We're talking about highly organized crime groups who employ many hundreds of the finest analysts who are continuously looking at emerging technologies and employing AI themselves to help them scam people and businesses. These operations can operate at really an industrial scale, leveraging advanced digital technologies to generate vast levels of illicit profits. And so a good example in March, Interpol uncovered an investigation that saw victims from at least 66 countries who'd been trafficked into online scam centers. And while Southeast Asia remains the epicenter of these operations, authorities are now reporting a growing number of scam compounds emerging in the Middle East, West Africa, and Central America as well. And all of that leads to a very dramatic rise in the use of deep fakes, identity frauds, and so on. And so it's really important that both advisors and consumers are using AI to sort of help you with your compliance and staying one step ahead of the fraudsters and protecting your clients' supplies and other stakeholders from these threats. - Yeah. And Helena, did you have something to add there? - Yeah, I just think something that will help paint the picture of what Dickens is talking about, which I think is really important. There's a one-hour YouTube called the AI Dilemma, it's in 2023, so it's a little bit out of date, but it will scare the daylight side of people in terms of what the technology can do from a fraud perspective, which I think is really important. And I've shared it to people beyond industry to watch it, just to be aware of what the technology can do, 'cause I think it's really critical. - Yeah, and I think as it goes back to what you said before, it's just understanding AI's and it to kind of, you know, understand exactly what it can do good and bad. I think that's both. Yes, Dickon, did you have something to add as well? - I would say exactly right, Helena, but it's not all doom and gloom. So if we can stay one step ahead of the criminals, the advantages for advisors, companies, for using AI, particularly in the compliance field are huge. And what it does is it really empowers you with the knowledge and data so that you can onboard, onboard many more of the right type of clients really, really quickly and with total confidence as well. - And to kind of round up today, I just wanted to go around to each of you and maybe we can start with you, Dickon, seeing as we are on you right now. What could we maybe see happen with AI compliance in the future? - It's a very good question. I actually think the future is already here. So artificial intelligence was first coined in 1955 by the father of AI, John McCarthy. And it really entered into the public sphere through the work of pioneers like Alan Turing, Herbert Simon and Alan Newell. We saw in the 1980s the birth of machine learning, 90s neural networks. In 1997, IBM's Deep Blue defeated the world chess champion, Gary Kasparov. And fast forward to 2011 where we saw AI really entering into our homes with Siri and then later Alexa in 2014. But for me, the most extraordinary development was the sort of release and public widespread use of generative AI and large language models in 2021. Obviously open AI launched at GBT in November, 2022. But by July 25, they had 18 billion messages being sent each week by over 700 million users. So about 10% of the global population was already using generative AI. And so I think there really is a world free chat GBT and post chat GBT. But I think what's really interesting is what happens next. So we're all used to using generalist models. But the problem with these is the prompts people give can be buried and that means the answers can also be mixed. So now is the time for really deep specialist LLNs coming into play focused on specific aspects of compliance. So for example, thanks to a UK government grant from Innovate UK, we at Themis spent the last two years researching and developing our AI agents to sense, think and act like real human investigators. And earlier this month, we launched our sort of boldest innovation yet, the Themis AI Investigator. And this is really the first specialist LLN powered platform designed specifically to tackle financial crime. So that anyone, even those without the compliance background can generate an enhanced due diligence report in minutes rather than weeks. And what's important here is this isn't generic AI. This is a specialist model trained by real investigators using real conviction data and global typologies. And the reaction we've had from the market has been phenomenal. I think the market has really recognized the need to have access to so much data quickly and with confidence to really make sure that there are no unwanted or hidden connections to organized crime or criminality within your investor network, your client network, your supply chains and sort of wider third parties as well. So again, I think what the specialized tools allow you to do is really sort of turn compliance from a sort of quite lengthy sort of human process involved set of checks into something that is quick and easy to do, meaning that companies can onboard many, many more of the right type of clients quickly and with total confidence. Does that make sense? - Yeah, I think it is just what we're probably gonna see in the future is just this better use of AI and with it comes compliance. But I think we'll just have to wait and see with this space. But yeah, but thank you all for joining us today and tune in next week where we will discuss other goings on in the industry. (upbeat music)