Join Heather Hopkins and Alasdair Walker at Handford Aitkenhead & Walker as they explore the transformative impact of AI on financial advisory firms. Discover how industry leaders are integrating AI into their practices and preparing for the future.

In Season 5, Episode 7, Heather and Alasdair are joined by Gary Abela, Co-Founder & CCO of Re:LoA. Gary introduces us to Re:LoA, sharing its purpose and impact, along with a companion video featuring an in-depth product demo.

Transcript — 

Disclaimer: This transcript was produced with the assistance of AI, so it may contain errors or inaccuracies.

Heather

Heather, Hey, welcome back to NextWealth insights. My name is Heather Hopkins, Founder and Managing Director of NextWealth. I’m joined by my co host, Alistair Walker, Managing Director of h and w and a Chartered Financial Planner. Hello, Alistair, Hi

Alistair 

Heather. Great to be back for another episode, along with the humans, we’ve had some AI podcast hosts in recent episodes, and I’m glad to be back in real time. Really excited to speak with today’s guest. I’ll let you introduce in a minute. But as always, before we dive in, I would like to thank our sponsors. They make this podcast and the AI Lab project possible. So thank you very much. Ssnc, fidelity, Aviva and Salesforce. Over to you. Heather, fantastic. So we’re

Heather 

doing another demo today, which is great. We’ve got Gary Abella, who’s the co founder and Chief Commercial Officer at re lloway, with us. Gary, welcome to the podcast.

Gary

Thank you very much. Pleasure to be here. Thank you for the invitation. Heather and Alistair

Heather 

Gary tell us a little bit about rioloa and yourself. We’ve done a companion video along with this podcast, and we’ll put a link in the show notes so that you can see a demo of realloa in action. But Gary, tell us a little bit about you and rioloa Absolutely.

Gary

So I’m one of those individuals that kind of sat on both sides of the fence, both having worked for about 15 years in financial advice space. Started my career at the Hartford selling unit linked investment bonds to advisors, before becoming the head of advisory sales for lead Mason, which is a large US asset manager, and then I left the industry for about 10 years wanting to focus and double down on technology, and specifically been running a AI boutique agency in Germany for the past 10 years, building very complex machine learning applications for large enterprises such as PwC, Mercedes Benz and BCG. So I bring both sides of that, complexities of large scale AI deployment, but then also marry that to the day to day struggles that financial advisors have to deal with. And so that’s really where my mission is, with real A is, how can we harness AI to streamline pain points ultimately help advisors and their admin teams and power planning teams focus on delivering better customer outcomes by having more relationships with their customers and do more sort of high end customer activity and reduce the mundane. So help me

Heather 

understand, because 10 years working in AI, you’re probably one of the only people who’s been working 10 years in AI, right? So that that’s pretty impressive. But how did you settle on the LOA? Specifically? Because working in Germany, outside of financial services, it’s such a niche area, I’m just wondering. Why the LOA? How did you how did you come across the idea?

Gary

Yeah, it’s a great question, Heather, and it’s, um, I’d love to spend a bit more time on this. So wanting to come back to the UK financial advisory space, one of the reason I hesitated by not coming back sooner was that for twofold, really, back then, large language models, chat, GPT, these types of, should we say, technologies weren’t readily available, and so it was very expensive to build machine learning algorithms specific for a client. PwC spent millions every year on building their own custom models, and so it didn’t make sense to come back to financial advice, because nobody has the appetite to spend that kind of money on tech development. Fast forward to about, you know, 18 months ago, chat GPT really blew up the AI space. I saw the opportunity to come back into the UK financial advice space and build my own product, not a product that I get paid to build for other people, but something that I can actually get behind. I’ve always wanted to come back to advice, because I think the advisors are unsung heroes in the industry, and the business model is inherently difficult to scale and do at scale because of the laborious amounts of processes and paperwork that we have to go through in this industry. So that’s where we landed on the LOA problem. We went through the entire advice journey with some early development partners from the client. Fact, find the first meeting transcripts of the meetings through to suitability letter writing all the way down to looking at the annual review and compliance file checks. The reason I landed in LOA and document processing was because of that experience I have in building AI solutions at scale. We wanted to find a repeatable process that is consistent and standard across the industry, that also is something that we like to call a gateway use case, something that actually already today doesn’t mean you have to change the way that you work, but simply have a tool that makes you do that faster and smarter. And so that’s why we focus on this. It’s objective. The answer is, I arrived. It or wrong. It’s not subjective. And I think in relation to AI adoption at scale, you have to find these use cases that really do, I think, capture the hearts and minds of an entire organization, and not just a few that are ready to adopt, because the only then can we start to think about the cultural changes that are necessary to really become more efficient and AI enabled as an as a wider organization and an industry. And so yes, it’s very narrow, but it’s very broad in its applicability, in that, you know, we extract information from documents in lots of different processes within our organizations and lots of different departments. And so I’m a specialist in document processing, and we will stay that way, because we’re not financial advisors. Financial Advisors will be the ones that help stitch these component utilities together to eventually automate more complex workflows. But from experience focusing on the low hanging fruit and doing that above 90% accuracy every single time is how I believe actually, clients and firms enterprises can start to adopt AI in a more pronounced way across their organizations. So

Alistair 

I hear two big messages there that I’m going to paraphrase. The first is that you found a problem to solve, rather than building a solution in search for a problem. And the second thing is that effectively, this could act as thin, thin end of the wedge, if you were trying to introduce a culture of AI in a business. Is that, is that a fair summary? That’s a

Gary

great summary, absolutely, and the culture of AI as a business is really important. So I’m going to use a really good example here. About eight months ago, going into firms and trying to do meetings with power planning and admin teams, there was a big journey you had to go through to help educate that it wasn’t about replacing the humans that are doing the job. And so you know, coming in and explaining that it’s a powerful tool that helps a human do what they do faster and actually focus on the things they’re more interested in, and take away the mundane, opened up the gates to actually people wanting to test and validate and utilize it. From that, actually, if you think about what we’re doing, and you’ve raised an interesting point, Anastasia, is we’re actually enabling you to actually develop a database about what your clients are invested in. And so often, I think data is misunderstood. We think we have lots of data. We don’t have lots of data as an industry. And very much, advisors don’t have much data in their company about their customers investments. And so we have that data stored in lots and lots and lots of different documents, whether that’s your own suitability letters, whether that’s provided documentation. And so having a utility here that takes unstructured data and gives it structure is the very basis of foundation for what you can then start to build more complex models around to do more complex workflows. And so yeah, we took a very, a very pragmatic and sensible approach to actually automation in this space, in the

Heather 

demo that I hope people will watch in the video, you talked a bit about how quickly the tool can extract unstructured information to create a structured file, and you were talking in terms of one to two minutes to extract the information, one to Two Minutes to process. It can then be uploaded to the CRM. So, can you share results on the impact that you’re having in businesses in terms of time savings, efficiency? You know, what are the what are the metrics that they’re using to measure success Absolutely?

Gary

So I think it varies depending on the size of firm, of course, but if I take your average advisory practice you’re probably doing in excess of, you know, 20 loath on a monthly basis, and this is for your standard size advisory firm. A pension consolidation piece of work will require you to have potentially two to three different policies that you have to read through. It can take a trained admin person anywhere between 20 to 40 minutes to read through one policy and extract the information that the Clara planner needs to have to effectively do the recommendation. And so we’re already taking 40 minutes of work down to under five minutes per policy. So you do the math on that, and we’re doing that at a price point that’s under four pounds per checklist. It’s aI priced correctly, and it’s saving hours of time on a weekly basis for advice practices. I

Alistair 

have got to say, speaking as the advisor in the room, so to speak, the promise here is great, right? You’ve got a you’ve got a problem that is a problem that can be solved, that you’re helping to solve. The secret sauce I think, for making this work in the long term has got to be CRM integration. And I guess I’m thinking, what is the long term goal for you guys? What are you comfortable in this lane? You just going to go, right? We’re going to integrate with it, with the CRM. So that’s all we’re going to do. Or are there bigger ideas to roll on from this point for you?

Gary

Yeah. So I think from the get go, it’s integrations is the top priority for me. We can help alleviate the bottlenecks in the workflow. But actually the input and the output piece and the RE keying is actually a big part of that time save as well. So that connected proposition is paramount to my strategy for the next six months. So we’re already on in teleflow. We’ll be coming on to planner as of next week, x plan career time for advice, and the list will go on. There are two types of integrations, basic integrations of pulling information and pushing the information back. That’s great. I want to get to a point of re eliminating rekey here, and so actually automatically populating the fields within these CRM systems is actually, I think, the holy grail of that integration. I think that’s also important because advisors are using about 13 different tools today, on average. I think I got that from one of your reports, Heather, do we really need another plethora of different AI solutions to add to the mix. I don’t believe so, and so I see it as a utility that plugs in and connects different workflows already within a CRM system. You know, companies spend a lot of time in thinking about their workflows and how they use CRMs. Why can’t we be the glue that takes the missing information and populates the piece back in. So that’s, that’s my direction of travel there in relation to future work. Absolutely, we’re building a basis here now of understanding what else we could do, more complex sort of checklists in the market. I do see compliance file check reviews as an avenue I’d like to get into. It’s effectively a checklist of checklists. Checklists, understand you’ve got the documents that you need, and then within that, do you have the right information within those documents? And again, it will always be around document processing. We won’t move into very essay populated spaces of suitability, letter writing or meeting notes. I think a lot of those toolings are a dime a dozen, this is a core piece of utility that I think can sit across an organization.

Heather 

Yeah, and I think it’s, I think it’s really interesting your motivations and the ambition of the business is really helpful. And I want to try to understand the other side of the of the coin, the advice firms that you’re working with, what, what are they trying to achieve? So obviously, you know, everybody wants to run more of to run more efficient businesses, but you know, if they were to get back more time, what? What are they looking for? Ai, I mean, it’s helping a problem within a business, because there’s inefficiency, there’s risk, etc, etc. But you know, where I’m getting at is, are they looking to take on more clients? Are they looking to improve job satisfaction. Is it, you know what? What are the things that they’re trying to achieve in the businesses? What are you hearing?

Gary

Yeah, so I think it’s all of the above. I think ultimately the key thing is that, how can they scale an operation better than they currently do? I see, you know, for every advisor, on average, two to three types of admin support team that’s required. So when you buy a new advisor in if you’re a consolidator, how much extra additional support do you need to actually service that book of business? And it doesn’t scale exponentially. And so I think one way of doing that is technology enablement. And so I think, you know, operational efficiency is a key aspect to my play here, right? You want to drive operational efficiencies by making our highly paid humans do what only humans can do, build customer advocates, get better referrals by having a better customer experience and remove the time they spend on the mundane tasks that are, to be honest with you, below their pay grade. I think what that does mean is that you can then ultimately take more clients on and scale it with the same human tech stack that you already have. We don’t have enough humans in this industry to actually service the Ford at UK market. And so there is a definite need here that, you know, automation has to be a way of doing that, and it’s not to replace the people that we have is to do more with the people that we have and and service our customers better as a result.

Alistair

That’s really interesting. Thank you. I’m now going to drag us back down to earth as the resident pessimist on the show. It’s my job to ask you about all the things that are going wrong. So I think, I think a lot about the risks and the challenges with adoption of AI, and we had at our September event, Adrian Hopgood talking about type one and type two errors. And that’s really lodged in my brain. You know, this idea that it isn’t just success rate, but it’s what’s the impact of a false negative and what’s the impact of a false positive, and how does that affect the overall workflow? And I suspect you’ve, you’ve come up against this as you’ve been training the models and building them out. So it’d be really interesting to hear, like, what do you see the challenges there? What are you trying to put in place to mitigate those? Yeah, it’s

Gary

a great question, and it’s, it’s really nice to get a bit of a technical question in the industry, and so just to kind of take a step. Back, what is a false positive, right? And the false negative. So a false positive occurs when the system incorrectly identifies something that isn’t actually there. So for example, in our use case, or in our product, the AI, if it mistakenly extracts a piece of data like a policy number, and that policy number isn’t, in fact, in the document, it’s also another way of talking about hallucination, effectively, a false negative is a slightly different problem in that it happens when the system actually fails to pick up information that is there. And so both of these are problems for our workflow. You want to ensure that the information that you need is given end of the day. And so we there’s a number of different ways you can mitigate to it. We approach the world of machine learning and llms with a rag model. So what that is retrieval, augmentation, generation. It’s an approach by which you can ensure that the models actually have been trained on a pattern that you’ve learned, but also that it retrieves the right bit of information in the input section correctly. So that helps mitigate some of these hallucinations and some of these risks of false positives and false negatives. What I’d add to that Furthermore, is that we actually have a very sophisticated rig that sits underneath the application, and so we have the ability to benchmark rank and score many different large language models. We don’t just use one large language model, we have in excess of six to 12 that we could use for any different type of question. You know, large language models sometimes are great for certain things. And so understanding which ones to use when is important, and have the ability and the mechanism in place to validate and check that you’re correct every single time. And so that happens in the background, automatically, and so these confidence scores, the ranking system, is very important. And then the final piece I’ll add to this is just having the feedback loop, having that human in the center of this process where you can source the document, actually have the human eyeball it and validate that it’s, in fact, correct. That’s the final stop, should we say, of ensuring that you actually avoid these false positives and false

Heather 

negatives? It’s really, really helpful, but when they’re assessing because, you know, you you’ve worked in this sector for 10 years, right for for the rest of us, what questions do you think that potential buyers of your services should be asking? What are the things that they’re not asking when they’re doing the, you know, the RFQ or RFP, when they’re going through their due diligence? What are the things they’re not asking you that they should be or the things that you’re your questions they’re asking that maybe they’re asking the wrong way or interpreting things a bit differently than they should? Yeah,

Gary

it’s a great question. Heather, I do an awful lot on trying to kind of educate around how your standard RFIs that you have are built around software. They’re not aligned to actual AI based products. And so we actually have our own standard due diligence document that we’ve we’ve completed, that we share with our customers as well. To go a bit deeper, but let me give you some of the pieces that I think irrelevant. So data security and privacy is paramount, and so in order to understand what data is being shared outside of the application, you have to ask the question, what models are you using? And if you are using the models, are they locally hosted within the application? It’s very easy today to build a chat GPT and to utilize chatgpt API. The reality is, you’re already in breach of your GDPR and data sensitivity rules which you’ve sent it out to a third party and an open public domain. So understanding the models that are being used and how they’ve orchestrated the models within the application and the infrastructure is important. I also think what I’m seeing more and more is people not fully asking the questions around third party tooling. Third party tooling is not a problem, per se. We use stripe for our account management and billing. You know, it’s no customer sensitive information within that from a client perspective. And so why would you want to build your own payment service tool when you can bring that in, and it’s much more effectively robust than I would ever build it in any case. But you know, think about if you are using OCR technology, that OCR, is it embedded within your application, or have you, by virtue of actually utilizing something to just scrape the information already been in breach of your GDPR. So you got the model, you’ve got the third party tooling. And another question that I think people don’t ask enough about is actually data anonymization. People say they anonymize the data in technical terms. There’s lots of different ways that you can anonymize data. How are people doing that? It can be a double edged sword, right? Poor anonymization techniques can still lead to data leaks. You can reverse engineer anonymizations quite easily from a technical and cybersecurity standpoint, and so for me, I don’t want to take any risks, and I don’t, I don’t believe anonymization is the is the answer here. I think it’s about building your. Solution with all of the necessary core components embedded within the software in a secure, scalable fashion on a UK server. Furthermore, I think actually larger firms that think about utilizing these technologies, the easiest option is ultimately to have a service that you can port directly into your own infrastructure and into your own sort of data, sort of security parameters. And with real away, we do work with larger enterprises and can port it as a Dockerized container inside their own cloud. And so that’s the whiter than white version of actually being able to utilize a solution like real away.

Alistair 

Thank you, Gary, that’s a really interesting point I would do wonder, actually, as more has written and talks about the sort of the centralized nature of the of the large language models where the people are going to start running your own server. I remember the days where we used to run our own server. You know, we might, we might be back our own AI models internally and refusing to use anyone else’s. Thank you that that’s been a really interesting kind of deep dive, and you did a really good job of explaining the technical concepts along the way as well, which is helpful. We’ve got a sort of final section now, really on rapid fire questions. So we asked same few questions to each guest. The first I’m going to ask you, and I’m going to change this rule slightly. So we’ve had a rule that says no meeting notes. Is an answer for how firms get started. But considering that you have, like, an easy to start with tool, I’m going to change the rule and say that you’re not allowed to recommend re LOA. So ah, that was exactly what I was going to do, other than real away, if a firm is interested in getting started of understanding or using or just getting to grips with the concepts of AI, what would be a great first step?

Gary

I think if I can’t use realloa, which would be my first step, I would just use a transcription service. Sorry, you said I can’t use a transcription service as an option. That’s

Heather 

all right. Meeting Notes are okay. We didn’t, we took that out of the rules. So you got to, you got to ski around them, and you mentioned Rio LA, which I’ve just done again for you. So there you go. And in terms of an application that’s made a big difference in your day to day, how are you using AI? What’s made a big difference in your day to day? Yeah,

Gary

so I like to think about AI with things that I don’t do really well myself. And so I use Canvas a lot for my marketing and development of my sort of digital assets. It’s got some great AI enabled tooling there. So yeah, Canvas is a great marketeers tool for content,

Alistair 

excellent. And we’re a podcast. We’re interested in podcasts. We like reading too. I’m hoping you might be interested in one or the other. So what would be your podcast or book recommendation around the topic of tech and AI, yeah.

Gary

So I love big picture piece around AI, having been studying it for the last 10 years. And so a great book is AI superpowers by Q fali. So really must read, and talks about the evolution of where AI is moving and who’s controlling it in the globe.

Heather 

Great. I’ve not heard of that one. Thank you. That’s a new one for me. We’ll put a link in the show notes to Canva, which I’m an avid user of as well, and to AI superpowers. That sounds great, right? Biggest AI fail mine, as listeners will know, was a cake. What is your big AI fail?

Gary

Mine’s probably a previous product that I launched and developed in the market. And so I built a product called stream AI in Germany, which was specifically for manufacturers. We got funded by the European Union, and made a number of mistakes. It wasn’t the right problem that was scalable. It needed high customization for every customer we went to. And so by virtue of that, it had died of death before it began. And secondly, we realized that actually the engineers didn’t want that to be automated. It was their core USP within their organization. And so think about the right problem and the right tool to the right problem, and if, in fact, the problem needs to be automated at all. That was a real lesson that I learned from that mistake.

Alistair 

Brilliant. Thank you. That’s a great that’s a great answer. I’m very, very honest as well. Thanks. Well, Gary has been really great having you. Thank you so much. Listeners will probably not know because of our great producer, but we’ve had lots of tech problems this episode, so thanks for bearing with us through those two Yeah. Thanks very much. Great to

Gary

have you. Thank you very much for your time. I really

Heather 

appreciate it. Thanks, Gary. That was a great demo on the video, and a great conversation with Gary, somebody who’s really knowledgeable about AI. I thought it was really interesting his comments about why the LOA process, because that was something I’d been curious about for a long time, because it’s so specific, and he’s not been in the UK financial advice market for a while. So So I thought that was really interesting, three things for me, and then I want to hear from you. Alistair, I thought the thing that’s that really stuck with me or still. Works with me after that conversation was was actually something that Gary said at the very end, and it was about his biggest AI fail, and it was the service that he launched in Germany that was the wrong product that wasn’t solving a problem. And I think that we always learn more from our mistakes than we do from our successes, and he’s obviously reflected on that, and it’s such a new category, a new sector, that I don’t think there’s many people who would have those scars of an AI solution that didn’t, didn’t have that product market fit. So I think that’s that actually is really reassuring to me about about the direction of the business, probably in the demo, he talked about how they built Ed solution to solve a repeatable process, and talked about it as a gateway use case, and you can see those learnings from that, that AI fail, that he talked about coming through there, but that repeatable process and making that something that you can get AI get people started In businesses. Yes.

Alistair 

Heather, I think coming in with a with a really clear idea of what problem you’re trying to solve has really helped the clarity of what Gary’s got to offer, right? And reloa, I felt that he was really focused on this 90% success rate, you know, like there’s, there’s quite a high bar for that, and I think that’s really helpful. Other things that he mentioned that I thought were quite, quite important was one around data sharing and sort of data sanctity, and this idea of sort of owning your data and knowing where it’s going, particularly because client data of this kind is so sensitive. So I thought Gary has some good things to say about that, but also the idea of empowering people in the in the businesses to not have to be doing work that’s kind of below their pay grades. I really like that idea, you know, because that’s the that’s the sales pitch. As a business owner, trying to get my team to engage, I need to explain to them, why, why this new piece of tech? Why this new thing? You know, I’m not trying to replace you. I’m just trying to help you do concentrate on the bits of work that you’re enjoying, you know, rather than reading from one document and transposing onto another. So I thought, I thought all of that was, was really interesting. It’s

Heather 

really interesting. It’s the consistent theme through the season. Is it? From, you know what Johnny was talking about, from Lyft to what Alan was talking about from advisory AI, I don’t really know what the bots from a notebook LM were talking about, but it’s always comes back to the people, right that you had to bring them on the journey. So listeners, check out re LOA and find out what Gary’s up to. Thank you to our sponsors, Aviva, Salesforce, ssnc and fidelity. Thank you to Artemis Irvin, who does a fantastic job sorting out all the tech glitches, and there were many, many today, and see you in a couple of weeks. Thanks to Alistair,

Alistair 

Thank you, Heather. See you soon. You