About this talk
Wealth Wizards are making financial advice affordable and accessible to all. AI is a key enabler of this vision; it slashes the costs involved in delivering regulated advice to people. Simrun will explain where they’re up to so far, and how they plan to leverage the latest developments to create their next generation product. He'll also explore what the future holds as AI creates a personal adviser in each of our pockets.
I'm gonna talk about what we do, which is financial advice, and talk about where we're up to with AI. Then I'm gonna talk about what we're working on in kind of our AI group in Wealth Wizards and kind of initiatives we've got in the lab and things we're playing with. So Wealth Wizards, our mission is to help people take control of their finances, and the way we do that is by making financial advice affordable and accessible to everyone. So what do I mean by that? An example is you saved up all your life. You worked through your life into a pension part, and you getting to the point where you wanna retire, and you need to make sure that that money is gonna last you until you die. You don't wanna run out of money when you're 75 and you haven't got a job anymore. And this is a great time to get some financial advice, an expert, someone who's regulated, to look through your finances and tell you exactly which financial product you should buy, where you should put your money to make sure you're gonna have a comfortable retirement, but actually, for the vast majority of people in the UK, that's not possible right now because to receive financial advice, that is very costly, and actually, it's a service which is only offered to wealthy people because of the cost in doing that. So what we are trying to do at Wealth Wizards is automate advice to bring down that cost and make those kind of services available to people, so let's continue with the at retirement example. Why is that, why is it so expensive to give that advice? Well the first thing you have to do is what's called a fact find. So in order to give you some advice, I need to know everything about you. I need to know all about your family, yourself, your job, your pensions, any other assets you own, your mortgage, la-da-da-da-da, list goes on, so for that kind of advice I just told you about, it's probably about 200 pieces of information we need to give you a piece of advice that we can be sure is good, sound financial advice. Then we need to go out to the market of financial providers. We need to look at dozens and dozens of different types of products, like annuities, and draw downs. We're thinking about bonds and equities, na-da-da-da-da. We need to compare them, we need to come up with a solution that's gonna fit your specific circumstances from those 200 points, right? And the last thing we need to do is write that all up in document which says what we know about you, the things we've considered, all the different options we looked at, what we discounted, and what we're gonna recommend to you and put our name behind, and the regulator wants to see that as well because if anything goes wrong, they're gonna be the ones paying compensation in 20 years' time. So traditionally, that service is offered by a financial advice firm for a few grand. And a process like that might take a company, say, 15 hours to do, and usually, you'd have a financial adviser and he'd be backed by what I call paraplanner in the office who would do a lot of that legwork. So where we're up toa company is, we have a product live in the market. It's called Robo Paraplanner, and it automates two of the three steps. It automates doing all of the research, doing all of the analysis and writing up the document. So the bit that's left is this fact-finding bit, this talking to the customer and understanding them, and that takes two hours. So we've already cut the time it takes to do this advice by about tenfold, and that's the cost tenfold. So who thinks that's AI? No one? So we've literally, one person. So we've literally, and we've been going as a company for around six years now. Started off as three people. I was the 12th person two years ago. Now we're at 70. So it's kind of... We've literally automated the work of those paraplanners. Anytime we solve a problem, and especially, as myself, I'm designing how it works, so our product, currently that's live in the market doesn't use any Cisco learning. It's rule-based. We encourage, here's all the rules of financial advice, the rules of chess. Here's all the constraints that we impose. Search, we pull in a space of possible products, and say search for the solution in this space. So it's all rules-based, but it is AI. I think it's fair to say. So that's my first interesting point, and that's what we're up to now. Now I'm not gonna drop the mike and say, "Right, we're doing AI." That's the talk, 'cause you guys are a bunch of techies and you wanna hear, like, where's the cool tech here, so let's talk a bit about some of the kind of things that we're working on right now in the lab using machine learning, which is hot. So back to this fact find, all right? So if we're saying, "Okay, now, we've got 15 hours "to two hours," and it's now two hours of that person's time, which is, which you're paying for, so maybe 200 quid, 100 quid. What can we do about that? And I don't know if you guys have had an experience where you sat with a professional, and there's a laptop in between you, and they're like, "Right, question one, "da-da-da-da-da, question two, da-da-da-da-da, "question three, da-da-da-da-da." It kind of feels like you're talking to them, but really, you're just talking through them to a computer, and they're not even looking at you. That's kind of what it feels like right now to get, if you phone up Liverpool Victoria, who we pair up with this tool. If you phone them up, that's kind of what it feels like to get advice right now. People do it because if you get retirement, it is a really complex, thorny thing, and actually, there's a lot of value so people kind of find their way through it, but can we make it, can we shorten that process and make it a better the user experience? So, what we've been doing is we've been playing with natural language processing. I'm just kind of start, on this part, starting to get a feel for what's possible. So we built a few chat bots. I think we bought a Facebook chat bot. We looked at API.AI, which is a bit more powerful. We've looked at kind of Alexa and really just, this is kinda just like prototyping, so again, kind of in the analogy of Josh's, we're kinda justto thinking which way are we gonna go up here? But what we've done is we landed on this kind of vision for an AI, and the AI's name is Lydia. You can ask me why afterwards. And Lydia will listen to the conversation between the adviser and yourself, and just take notes. Simple as that. So now, rather than saying question one, , question two, and having that very linear journey, you can just have a natural conversation with your adviser. "Okay, tell me about your pensions. "Tell me about what your retirement wants to look like. "Tell me about your kids," and at the end of that conversation, magically, Lydia's taken all the notes. And so it's a much better user experience, and also we've cut the time down by a huge factor, so this is something that we think, this is something that we're working on currently, and we think is a kind of, as a next generation, it will land in the next couple of years. So that's one area, is can we keep produce, by looking at kind of the human-computer interaction, can we keep streamlining that process to be really much more like conversation? Now, traditionally, for example, if you got advice from a traditional adviser, as well as taking them 15 hours, it might take you several months, because they've got to meet you, get all the stuff, go away, do some research, come at you again maybe, go away. So again, if we're powering up advisers with these tools, they can literally do the whole thing in one session. Now, so say we solve that issue. We've partially solved that issue of per client, how much time does it take to give you this piece of advice, this valuable recommendation? But our mission is to help millions of people in the UK by giving millions of people advice, so we need to go further. And the next big challenge is, so we're a B2B business, as you might have guessed, which means that we're working with big financial services companies, small financial advice firms all over the UK, many in the city, and so we need to get them to adopt our technology, and they want to adopt our technology, but an issue is right now, it might take us a few months for us to meet a new financial advice firm, work with them to integrate our technology with that firm. And the reason is whilst good financial advice is well established by the regulator, by the FCA, it's established in the form of principles, so the FCA will say things like if a client has a family, you obviously need to take account of their dependence when you give them advice. But the regulator doesn't say, "Okay, if they have a wife "and two kids, then if they die, "and they're over the age of 70, "then the kids need to get 20%, "and 20%, and the wife just get 50%," and there's a lot of detail which is left to the adviser. Now that's called a house style or a house view. So every financial advice firm will have their own sort of house view. And one thing that is really slowing our growth down right now is that we've got this long queue of financial advisers who are like, we wanna cut our time by tenfold, but they can't really afford to pay us to configure by hand, to create the rules in our system that's gonna work inside their business. At the moment, it's well, you can have it now, but you have to have the vanilla version, and that does, it doesn't work. It's not tenable. So another proof of concept we've been working on in the lab, and this one a bit more advanced actually is all these financial advice firms, thankfully, have these huge archives of past cases because they have to keep them in cases of complaint 20 years down the line. And so what we've worked on is, we look at this huge archive of past cases, we mine that data into a kind of a cleaner form, a spreadsheet, effectively, and then we run a machine learning algorithm to learn the rules, learn the house style of that financial adviser without us having to do any, do this hand-coding that's got us off the ground, so this really is like then, machine learning, so the second generation style financial advice, and in the lab, we've managed to get this to kind of 70% accuracy, which is not bad, and actually, if we keep pushing this up to 80, 90%, then we start to get superhuman level of accuracy, so actually, most financial advisers wouldn't get financial advice to be correct like super correct, like 95% of the time. That's why they have hordes of people in compliance, checking the advice all the time. There's a really big challenge in this space for us still, before we can get something live, and that's a challenge with, actually, all machine learning techniques, and that's that we can't explain the neural network or whatever technique using, regression, can't explain why it's giving the advice. Really, the way in which advice is regulated right now is these documents, and if you, traditional piece of advice is literally a 20-page document that explains, as I said before, all the things we looked at, all the products we considered, what we know about you, what we've ruled out, what we think are good options, the option we're recommending, why recommending it. And so there's this fundamental issue with Cisco learning techniques that we still need to overcome before we can do this for real. There's hope. People are researching this, so in our proof of concept, we used a method called decision tree learning, whereby, instead of using a neural network, which is very popular at the moment, it's actually a pretty old school technique where the computer creates its own decision tree, and then you can at least see the steps it took, but then to make it effective, we need to use a hundred decision trees, and take an average, and then you lose that again, and so, these are the kind of models you get yourself into when you move from the academic while there's, these guys mentioned, into the business domain, where you've got people like the regulator asking you these questions. So that's kind of two of the projects that we're working on, and I say in the lab at the moment. We're kind of at the bottom of this mountain, looking up, thinking, taking one step at a time. The last thing I'll touch on is, is then, I guess, there's one point I forgot to make, which is that that proof of concept, where we got kind of 70%, not bad accuracy, took one data scientist a few months to do, and in parallel, we've got a whole team of engineers, financial guys, product managers, designers, working on the same problem in our traditional rules-based way, right? And that's, there's sort of, you've got a whole team versus one guy doing the same thing. If the one guy could get the thing to explain why it's done that, that would be a game-changer, so let me know if you have any ideas. Anyway, so what's the top of the mountain then? In financial services, we have this, there's this really, really big issue, and it's called the engagement issue, which everyone talks about in financial services, and I think AI's got a part to play in this. So here's how it goes. It goes kinda like why don't people engage with their pensions? Why don't people save for the future? Why aren't people engaging with this idea of retiring? People just bury their heads in the sand. I'm really, it comes down to the issue of people don't wake up in the morning and think like, "Right, today, I need to, "I wonder if I'm investing my pension "in a tax-efficient way, and I wonder "if the money in my pension is invested "in the right mix of bonds and equities, "so that it's not too risky, "but I'm gonna have a good return so I hit my goal." No one wakes up and thinks that, right? People wake up and they think things like, "Oh, shit, I'm a bit stressed out about my money. "I'm not really sure. "I don't really feel in control of it. "I need to kind of sort it out, "but I'm not really sure what I need to do," and actually, I think if there's one thing that human advisers are really good at and actually really enjoy, it's taking someone from that mode of just, I just feel stressed about my money, what do I do? Kind of having a cup of tea with them and just, and talking them through a really open-ended conversation, and getting them to a place where they're ready to act, so it might be, "Okay, whoa, whoa, calm down. "So have you got any kids? "Are they looked after if you die? "Okay, fine, let's look at your pension. "Let me just explain to you what a pension is. "Let me put this for you in layman's terms." So the financial services, you try to kind of educate people by throwing books and content at them, but actually, what I've is the most effective, is just someone sitting down and having a chat with someone, and then explaining to them in their terms. And so, again, there's a huge issue here because 50 million of us in the UK. We don't have an army of 50 million financial experts to go out and have that one-to-one chat, but at the top of the mountain, I kind of see an AI who could be that person for you, who you could have an open-ended chat with about your finances, and who could take you in the direction of the advice and help that you need. So I think that's eventually how we'll fulfil our mission of helping people take control of their finances. I think AI's a key enabler, and these are kind of the small steps that we're taking on the way. Thanks.