1037. Insights: Shaping safe AI innovation - Live from the FCA’s Supercharged Sandbox Showcase

24:22Yeah, absolutely. So when the firm was first founded, it was pretty much out of that aspect of compliance can be difficult. It takes a long time to establish rules. So having a really easy rules builder and putting that back into the hands of the compliance officers was the ethos. And then over time, we spread out. We've had some really cool AI in the product. But when I joined five years, five and a half years ago, we started pushing out that and we've been iterating on that ever since, adding crazy ideas in as and when they make sense to do so. Yeah, and as those advancements in technology, I guess the ideas have got big.

25:00You know, 10 years ago, it all sounds like sci-fi fantasy and, you know, 10 years later, it's all possible, right? Yeah, absolutely. It's one of those weird things in AI that you always kind of overestimate where you'll be in the long term, but kind of underestimate where you'll be in the short term. And I think a lot of us have been really surprised by how quickly things have gone. I do remember in a previous role saying, it's like, oh, you know, give it time. And if you want to create an image, you'll just type it in and you'll get an image back. And I remember saying that at a conference and seeing a load of marketing people just, yeah, right.

25:33And that's where we are now. So I think all those wonderful things that we can imagine, they're starting to make sense. And, you know, there's still the guardrails and lots of great things that we need to put in place. but there's so much power and adaptability there that it's really going to make our job of financial crime compliance easier. And for me personally, that's a huge motivator is, you know, how can we detect and prevent all of the financial crime that's out there? Much more exciting to explain to your friends and family what you do for a job as well, right? Yeah, absolutely. 20 years ago, I worked in financial services.

26:06Now it's fintech. It sounds sexy, right? So ignore these two. Pretend they're not there. Okay, like, how's it been? Like, how's the experience of being in the sandbox? It's been great. We've had learnings. I think there's quite a few things that I think if I went back to when we applied, that I'd probably put in place in terms of what we needed to get done up front. And, you know, particularly sort of data, because we thought, okay, well, we've got this idea, let's apply and then if we succeed, we can go to customers, ask for data, maybe create our own. So when we started, we were a little bit on the back foot there and then we did actually break the sandbox by putting too much data in the wrong place.

26:52It's just a question of learning sort of where things went. Does it reply email? Thanks for that. To be honest, this is the first time I've heard of it. It shouldn't be broke by Sam. It was one of those things that, you know, we're used to developing locally, and we've got all our own resources, and then suddenly you're presented with an environment, and it's like, click here to launch something, this is how you transfer files. So we did, and then things wouldn't launch, and we had to work out why, and then we got more provision, we got wonderful GPUs. I mean, I'm loving the fact that I can go in, I push a button, and then I've got a terminal up, and it's like, great, I can transfer as much data as I want via an AWS bucket and transfer files locally and I can run things and I can go away and come back.

27:38And that's just been amazing, really able to do, let's say for those that were in my presentation, it's just some quite esoteric ideas where it comes to how do we detect different financial crime typologies. So that has just been tremendous and it's not something we'd have been able to do without the sandbox just because, as a scale-up company, it's been going 10 years, we have a lot of customers, we have a lot of demands on our time, and our internal resources are allocated. I mean, it's my job to allocate them quite well in advance in terms of what our roadmap is and what we're doing. So trying to think ahead as to what's going to go in the roadmap six to 12 months down the line and what research we need to do now, you know, the sandbox has just been an absolute catalyst for that.

28:21Yeah. I imagine everybody internally must have really liked that as well because it's, you know, all the really scary, fun ideas you put down on the roadmap, you were like, we're going to get to do those things now. So the team must have really enjoyed it. Yeah, I mean, particularly my team, sort of all the data scientists, it's like, we've got this wonderful resource and it's like, are there any other crazy ideas that fall under this umbrella that we can do at the same time? And then sort of the wider company, everyone I've spoken to is like, that's really cool, when can we talk about it? And it's like, well, we've got some results.

28:52Soon. Soon. Soon. When will them be now? soon. So yeah, it's just been really good and really positive to be involved in. Amazing. What's the experience of like standing up on stage in front of everybody? Like seven minutes is not a lot to like, you know, conveying like insanely complex topics. Like you've got to be pretty succinct, right? Yes. I think both myself and Johnny who presented with me, we'd have loved to have had an hour keynote to go through it at the depths. And if this was an academic conference, then we'd have pushed for that. But to get that down to who we are, problem statement, and the two projects that we did, because we didn't just do one, we did two parallel things, and then where we're going next and keep that like 30 seconds per slide was a challenge.

29:38But careful rehearsal and knowing your stuff inside out, you can get it done. Yeah, like power talking, yeah. I often get told on the podcast people listen to me at half speed because I speak too quickly. But in this environment, that would be an advantage, wouldn't it? Absolutely. So was your sort of expectations really high going into it? I guess, has the expectations been met or have they sort of surpassed? Or what was your wildest dreams of how this would run? I think, yeah, they work high going in. And I think we've met them, but we're at a different peak. So we went to climb one mountain.

30:13And I think, and we're not the only participant who said this, but we tried some things up front that just didn't quite work. and as we were going through trying these out, which again, we needed the resource to play around with and make sure that we could do it with enough data to prove the concept and then we switched our direction and came to different approaches. So we got where we wanted to be but via a different route. Yeah, so it's not, you know, these things are hard work, right? You put work in, you put the effort in to get to a result but, you know, the point you made of we tried some stuff and it didn't work, like that's why this exists, right?

30:49Yeah, absolutely. And then that's science as well. As you know, I'm a data scientist. The scientific approach is you make a hypothesis, you try stuff out, you test it, it fails, you try something else out, and you keep iterating on that. So that was a really important part of it. And having that outside of the commercial expectation of this has to be delivered to client in two weeks is really useful. Yeah. So I guess, you know, for everybody, like this is the end of the two days, you know, it's sort of and actually the end of this period, right? What's the big learnings here other than don't let Janet pull down your infrastructure?

31:25Yeah, that's definitely one I'm going to have to put on the list. There's going to be a note later. Yeah, yeah. Drop to you in it. And no, I think the lessons learned for us is I think we need to probably have a bit more time, to be honest. I don't know how you feel Janet like whether the three month period was long enough or whether we needed to extend that a little bit. That's a really tricky one because three months is it's a nice chunk of time I think three months going over the Christmas break probably less so because we did lose a lot of time with that but just if it's too long then you run the risk of not putting enough effort towards it whereas three months it's like no we've got to do this we've got these weekly check-ins and we've got to show progress each week whereas I think had it been four or six months we might not have been as tight on it yeah and so I think my only thing was like just losing those couple of weeks over Christmas was a bit painful absolutely so cancel Christmas next year yeah yeah I've got a hot light to Santa any other learnings what were you I mean I'm making a big presumption here that you're going to do this again or like that this would be something you'd want to continue because I guess from your perspective is the is the outcome been what you wanted it to be yeah absolutely I mean the outcome has been fantastic like seeing the the teams present I said there's at such a high quality um now the journey for them we're keeping the sandbox open for for a little bit longer um like for another month so they can continue that development that I was talking about before um and not just cutting it straight off so um seeing what they do next is the important thing for us is now watching those firms and seeing how they to go on to develop further and seeing some of these things come to life.

33:05I'm not silly enough to sit back and say they're all going to succeed, but to see a good majority of them go out to market and see them be productionized and being used elsewhere, that's a huge success for me. Or even just like being handed off into some of our other innovation services. Because we're still like at the early stage innovation space, right? It's the, okay, well, now you've developed a product. Now it's probably time to go to Reg Sandbox because you're starting to test with real consumers. So you're continuing the journey within the FCA lifecycle, but also outside of it. Yeah, that lifecycle, having something to get to viability and having something to get to something that will work in production with real data and take it first.

33:47I mean, are there ideas that have come out of this that you think you'll take into the real world, as it were? Oh, absolutely. Both of the things that we worked on, the next step is to take our somewhat, you know, sort of quick and dirty code and get that into a productionised format. You know, the efficient computing approach, we want to get that in product very quickly. I think the fluid dynamics approach that I presented, I think it's a little bit going to take a lot more work just with more data sets. And I think that'll go through the life cycle, as Matt just said, just working through all the different approaches so that we can go to sort of real clients, get some real data, check that it's doing what I think it's doing.

34:31And then, yeah, absolutely, I'd love to see that in our product in the next six months. Yeah, I guess that's another big reason why having, you know, dev environments that really replicate what happens in production is really important. This isn't throwaway prototypes that you don't do anything with going forward. You can, you know, take these forwards and actually put them in the world. Yeah, and I think the work that not just has been done in the supercharged sandbox, but the digital sandbox, the regulatory sandbox, the innovation pathways, the FCA kind of pulls together. You go back and look at what the firms here have done, but you can kind of traverse backwards and look at what some of the previous firms have done.

35:10And there are firms who have taken part in the AI supercharge sandbox who are, this isn't the first thing they've done with the FCA. And you can go back and look at the ones who have done things with the digital sandbox. And you go on and go, wow, they've done some really, really awesome things commercially and got funding post that. And I think for me, being involved in some of the stuff that the FCA does, even in the small capacity comparatively that I get to be, that's where it gets really exciting, right? You've given them a springboard, but the value that the work that then that generates and the value that that generates into these companies and seeing some of them even create end-to-end products in three months as part of this, that's exciting.

35:57The next step is you've got to commercialize it at some point but there is a track record of that is very, very possible. I guess, back to you, I mean, I joke about that difficult second album vibes but it seems like you nailed it. Good job. What happened last time when I saw this happen here was the sort of ripple of the things that you did here kind of all around the world. Having the pleasure of working in far-flung places then everybody seems to have emulated the work that has happened do you do you foresee a similar thing happening in this space i imagine the how have you done that and what's this process like and i mean part of leading is sharing in that regard i guess yeah i mean absolutely we uh well we have been having those conversations already with a lot of international regulators um they're constantly asking okay how did you do that and i i do remember when i first started speaking to international regulators very much focused on the Tech Spirit Program, the Hackathon, so the first album, right?

36:52And now we're in the second album. It's like, oh, wow, how did you produce that? And so we still have those conversations. And one of the things I'm always stressing to them is, you know, you've got to have senior leadership sponsorship. And without that, it's going to be very difficult to get anything off the ground. So I really appreciate that from the FCA and giving us that space to create. But we are seeing that some other regulators internationally are starting to take this approach. They haven't got to the supercharged bit yet, but I think it's coming. Yeah. It's amazing how much of that is, you know, it's not can you share a handbook or, you know, the Jedi move you pulled or whatever.

37:31It's the sort of people that are involved in it, the sort of culture of how that works, isn't it? That's a difficult, often forgotten and very hard to replicate, right? Yeah, absolutely. And this is another thing I would stress for any regulator out there, if you're listening. There is something about having an innovation team. It's about having that creativity and being able to test and think and think outside of the box. And that's something, you know, we had a quick conversation about this before, about how do you make sure you're recruiting the right capabilities in that space? And as a regulator, that can be a difficult thing to do.

38:08Sure. To find that balance of like, I need somebody who can think about policy and I need somebody who can think creatively. and then bringing that together. And then you start to get like these great ideas and you also understand some of the parameters that you have to work within. So yeah, that's another thing that they need to take into account. Yeah. Stiff, not easy, but when you get it right, that combination of ingredients can do amazing things. But all right, enough questions for me probably. Like this is your opportunity from a crowd perspective. If you've got any questions for these guys, make it really difficult.

38:41Like, you know, if it's really difficult, then if you can make them sweat on stage slightly, that would be quite good. But what do you think? Have you got any questions that you want to turn over? There's a mic walking around. Difficult question for Janet. Hearing everything that you've heard over the last two days and thinking about our approach to regulation, what do you think about the rules and the framework? We have an approach that we've taken which is very open and conducive to innovation and growth. What's your view on the policy approach? I mean, for me, I think, as we said at the start of the day, the outcome's the important thing.

39:26As long as you get there in a way that's clear and auditable, especially when it comes to AI, that's the key thing. We don't want to have a situation where it's almost a computer says yes or no. We've got to have that clarity. And particularly as someone in the AI space, that level of explanation and audit is clear. But from a regulation point of view, that outcome-based approach is the only sensible way forward. Otherwise, you're going to end up almost over-regulating a wave behind where technology can allow you to go. and particularly from what I see with the people we're trying to detect, they're actively absorbing as much technology as they can get their hands on, so we need to do the same.

40:14Yeah. Bad guy, somewhat innovative, right? Yeah. Where is the mic? I know there's one at the back. Who are you and where are you from? Hi, my name is Denise Rudich. I am CEO of Rudich Advisory and founder of Productopedia. We follow the money so you don't have to to help you fight financial crime. This is not your seven minutes, by the way. Thank you. And I'm a huge fan, not only of the FCA's innovation team. This question is for Matt. Matt, one of the biggest challenges, as you mentioned, is what happens next in terms of embedding solutions, in terms of incubating solutions that have been developed.

40:56Is there appetite for the development of something similar to what the Monetary Authority of Singapore has, which is they've set up a regulatory technology grant to effectively help financial institutions embed some of these solutions. Or maybe the next iteration is there a chance of working with implementation partners. Thank you. Next question. Yeah, that's a biggie. Yeah, I mean, I'm going to say short answer is no for the MAS model. I mean, definitely that jurisdiction works a bit different to how we operate as a Financial Conduct Authority. We look at other avenues to encourage that investment, and a lot of that is within what we're doing today and making sure that we're using that convenient power.

41:45Things may change. I'm not in, well, it's definitely above my pay grade in terms of whether those changes would happen. But we also work quite closely with MAS. So we are doing a lot of work and this is me on the team, it's more Simran, but working with the MES in terms of sharing the knowledge and sharing the different ideas and the firms between both jurisdictions. So we're kind of like working together in that space. Now, we're trying to attract some of those over here and they're probably trying to attract some of the UK over to there. So we are still working with them and they may receive some of that funding, but at this stage I'm not going to say any more than that.

42:29Competition is good though, as the regulator often says, right? Yeah, competition is good. It raises the bar for everybody, doesn't it? Which is awesome. And that's the whole point of the innovation, right? Like innovation here is explicitly to drive that competition up and that's a really clever approach. 100%, yeah. Has anybody got a mic at the moment?

42:52Hi, David. This question is for you. This is Michael Burns, soon to be head of AI at the Eclipse Foundation. We worked on Starling, Monzo and many other innovative challenger banks in the UK and so on. Do you see a second wave, third wave coming through which will be much less people, purely AI? Is this propositions that you already see in the market? I'm asking from a German perspective, we're looking at this here as the ones that are doing it before us. Do you guys want to have another one or anybody want to jump in? I'm happy to do it. I mean, there was a face of, no, you take that one. I don't know, I think the sort of decision of like the interface being anything other than a human for at least 1 % of that journey, I think is going to be an interesting challenge because while all of these things are fundamentally, you know, an amazing advancements in the experiences that can be delivered, then really I think there are always places for people in what that means.

43:51So I don't see Monzo or Starling abandoning beautiful interfaces that they've created or, you know, NatWest, Lloyds or Barclays for that matter, but definitely there'll be an evolution of what those things in terms of where it stands. I don't think there's any financial services business in the UK that doesn't use AI in some form. it's just being fun in the back office instead of the interfaces that actually touch customers. So we'll see, but you never know. I've been wrong before. Any other questions? Yep, more hands going up. I was worried nobody's bloody asked anything. Now look at you lot.

44:26Thank you. Great discussion. Thank you. Thank you all. I'm Marco. I'm the director of AI at Moneybox. My question is probably both for Matt and Ollie. First off, I'm a big fan of the FCA's outcome-based approach, long may it live. It did feel like there was a but coming. There is a but. There's outcomes that are more easily codified, if you want. For instance, in the use cases that Janet, you guys, presented today, it's a well-established AML financial crime space. We can evaluate it through the existing metrics. Now, I've seen some things yesterday and today that I raised my eyebrows and I thought were way beyond the regulatory boundary.

45:12So maybe I'm too conservative, right? But obviously, what's been put out there is now a signal to the market. Should we take it as a signal of that's kind of where we can play or how should we be thinking about some of the more consumer-facing stuff, let's say? It's a good question. I'll take this one, Ali. Yeah, go for it. But just to be really clear, like what you've seen over the last few days is that early stage innovation. So yeah, you might be testing the boundaries of regulatory comfort, but they're not in production yet. So that's when they start to go through our other services. So if you start going through innovation pathways, you're starting to navigate the regulator, and then there's going to be, okay, well, you might need to think about setting it around these parameters instead and same with the reg sandbox world so the idea of like being in that early in the stage innovation is to yeah okay go large and maybe you push the boundaries a bit and then we'll start to make sure we refine that if they go through those processes i mean i mean i think that's a really interesting point because actually if you think instruments are often regulated individually and actually often what ai can do is traverse those things to create interesting services and actually it can then cut across lots of different regulation in lots of different ways but it's kind of part of the point i guess isn't it is where are those boundaries and this is why you test in a sandbox because you can play with it you can evolve it and then actually that can inform future things as well so uh more for more sir hi my name is sir thomas hergnan from thailand solutions i have a question to matt actually on which stage to have to be as a founder to be part of the sandbox.

46:59So can I build the minimum viable product with you guys or do I have to be a bit further on my journey here? Do you mean to stand here on this stage? Are you saying like to get access to the Supercharged Sandbox to develop your product and then get out there? So I'm on concept phase at the moment. But do I have to be already like a product or like, you know, something to demonstrate? So to be accepted into the sandbox? Yeah, exactly. Yeah, at what point? Can it be white paper or near production? Yeah, I mean, we have seen applications that sort of vary. We saw some very much early stage whiteboarding, yes, and we would still look at those and we still consider them.

47:43Because part of the idea is that we're giving you access to the data and the infrastructure that you might not necessarily have. So conceptually, absolutely, if that's where you're at and that's where your head is at, then I would apply. We did actually see some other solutions that were far more developed. And they didn't necessarily come in to us. We were like, well, you've already got data, you've already got infrastructure, what do you need to test the environment that we've built? And that's where we would consider putting them into AI live testing. So they've already done something, then build something, or so how do they get some of that, I guess, that regulatory comfort around their models?

48:19So we would still signpost it to other areas, depending on where you are in that journey. But if you're at early concept stage, yeah, that's what a Supercharged Sandbox is for. All right, we've probably got time for one more. Gentleman in the middle. Yes. Just increasingly making it difficulty for you to get marks to make. Sorry. Thank you. I won't take time to congratulate FCA on a wonderful job done. My question is for Janet. For, I mean, obviously, there were those who are more early stages companies who probably can focus completely here. But for someone like you, who also is continuously running your own projects at your company and to still dedicate this kind of concentrated time here, how did you manage that?

49:05Personally, with difficulty. It was very easy for me to carve out time for my team and manage projects around them. But as a team leader, all the meetings that I've been on, I found that it became an evenings, early hours of the morning thing for me personally, which is not ideal. But I think going back to what I said earlier about really preparing for the amount of time, I think I would have probably removed myself from it just knowing that I wouldn't have the time, although it was my idea I really wanted to try, so I couldn't do it. But it was difficult. So I think anyone looking at this needs to really consider how much time they can donate of their teams and their resources, because to do it well, you really do have to put the time in.

49:56And I would suggest sleep is important as well. If we've learned nothing, sleep is important. That's good. All right. We don't have any time for any more, I'm afraid. I'm sure you have more questions. I'm sure these guys will be mingling upstairs later on. So feel free to come up and throw your question out. I'm in person. Sorry, guys, I just volunteered you for that one. But we are going to have to wrap up today's discussion, though. And a massive thank you to all of the panelists for breaking this down. Super important topic. And it sounds like real amazing developments that have happened over the space of, you know, Christmas and the months that go before that.

50:31But thank you so much for joining me. Where can people learn a little bit more about you and all the good stuff that you're doing? Janet, let's start with you. So, the napier.ai website is very easy to find, but I'm also very easy to find on LinkedIn. I believe I am the only Janet Basterman in the world, so you should be able to find me very easily. That's a bold claim right there. I'm going to Google it later. Oli, where can people find me? What about you? Yeah, so either nightwonder.com for the company and then for myself, Oli Platt on LinkedIn, but spelled O-L-I. So it's a little bit easier to find me, but not that easy.

51:02Very good. Matt, where can people learn more? F-C-A-I-L-A-B. Let's go. F-C-A-I-L-A-B. And what about you personally? Oh, personally. I'm imagining you. You're going to start getting DMs on LinkedIn of next time's applications. Already happening. So yeah, you can find me on LinkedIn and that's probably the best place to find me actually. Very good. As for me, yeah, lurking on LinkedIn most of the time these days, so feel free to reach out. If you like what you heard, then please follow the podcast and make it your favorite platform. Wherever you get your podcasts, you'll find us there. If you want to join the conversation, you can find us on pretty much every social media channel at this stage.

51:37Just search for 11FS or Fintech Insider. If you really want to, and we do get emails, it's weird. Email us at podcasts at 11fs.com. Thank you so much for listening, everybody. Goodbye.