1006. Insights: AI in banking - What it really means for the customer

0:14Welcome to Fintech Insider Insights from 11FS. I'm David Barton Grimley, and today we are cutting through the hype to talk about what AI is actually doing for financial services, especially from a customer perspective. So look, AI is absolutely everywhere right now. It's on your newsfeed, your boardroom, your LinkedIn inbox, all buzzing with chat GPT, generative everything, and promises of transforming the future of finance. But here's the thing, for all the noise, when you zoom in on the actual customer-facing use cases in financial services? There's not much to see, at least not yet, or at least not that we've seen.

0:52So the question today is, why not? Why aren't we seeing a Cambrian explosion of genuinely useful AI-powered products for customers? What's holding it back? What's working beneath the surface? And where might we go next? So let's start with what's real. AI is making a difference in a very few specific places. So customer service, for example, being maybe one of the biggest ones. We're going to talk about that today. Smart chatbots, virtual agents, better call routing, things like that. It's saving time, it's improving triage, sometimes even beating humans at speed. So that's all very well and true.

1:30And behind the scenes, AI is quietly automating processes, document checks, KYC, compliance, cutting down hours of manual work to minutes with fewer errors in theory, right? So that's where AI is working today. The big question is, how do we take it from here and make it truly change the customer experience? All of those bets from those big tech firms on valuations, where do they come from? Are they going to become real in the future with the transformation that they're promising? So let's get into it. And the three people who know what's real and what's not real when it comes to AI is our panel.

2:04So let's meet them. First up, we have a FinTech Insider debut for Oscar Barlow, Head of AI Advocacy at Starling. How are you, Oscar? welcome to the pod. Tell us a little bit up to what you've been up to. Yeah, hi. It's really great to be on here. So yeah, as head of AI advocacy at Starling, it's my role to ensure that Starling is an AI fluent business. So I'm a software engineer by background, and I bring that training, that expertise to ensuring that we've got the technology and we've also got the culture and we've also got the processes so that Starling can make maximal use of AI. That's amazing.

2:39It's great to have you on the panel, Oscar. And next we have a fintech insider debut for one of our very own rosie lee senior ux researcher and customer strategist at 11fs welcome to the show rosie this is your first podcast and you're actually in the studio with me that's awesome um tell us a little bit about your role at 11fs thank you um yes i'm rosie nice to meet um you guys uh i spend my time speaking with customers to understand what they're trying to achieve, what their pain points are, what's stopping them from reaching their goals. I do things like focus groups, interviews, to test new ideas with customers.

3:17And then I take those insights to the banks and the financial institutions that we work with to help them design better products and services that make customers' lives better. Awesome. Great. Welcome on board. And last but not least, a warm welcome back to Vivek Madlani, co-founder and CEO of Multiply AI. Welcome back Vivek. Tell us a little bit more about what you've been up to since we last spoke. Thanks David. Yeah, we, so at Multiply, we help firms launch scaled advice experiences, launch advice propositions, and essentially make advice more accessible to their customers. We've been doing it for quite a while.

3:58Now and since we last spoke, we've been building a lot of the latest developments, whether that be agentic AI or whatever else into our offering and working with more of a variety of firms in terms of helping them launch advice propositions. So lots of exciting stuff to get into. Yeah, amazing. Right. Now we have a panel of AI experts. Let's get into the discussion. So Vivek, I'm actually going to come to you first as a kind of a founder and a builder in this space. Do you also sometimes when you read LinkedIn, want to throw your laptop at the wall when you see the amount. There's just so much out there about how it's going to transform the world.

4:40And if you do also want to throw your laptop at the wall, what do you think is that disconnect between the hype and the reality? Well, I think there's a few different ways of looking at this. I think one thing to say is that I do believe AI is being used quite a lot when it comes to helping people with their money. it just might not be where we most expect to see it. Anecdotally looking at some research and data from the likes of ChatGPT, from Google with their Gemini product, it does feel as though they get a lot of inbound queries, questions, whatever it may be, about finances, around what people want to be doing with their finances, and just getting help through that particular channel.

5:26Which, you know, we're probably kind of looking out for it, more looking at what the banks are doing or what life goes and what other firms are doing. I think on that side of things, in terms of looking at what is happening from within the finance industry, we do see quite a lot of innovation taking place. It's just when it comes to launching kind of AI customer experiences, given that they are kind of more probabilistic than they are deterministic, there's just a huge amount of testing that needs to go into place before they're launched into the hands of customers. And it does open up a lot of these firms into just a larger surface area of risk.

6:07And a lot of this stuff is quite new, right? So a lot of the kind of cutting edge ways of testing and ensuring robustness are kind of quite new. So the areas where firms are looking to kind of launch AI and get it kind of helping customers might be a bit more in the background, you know, where there are processes, where there's a human in the loop, where stuff can be double-checked. That is certainly what we're seeing a lot more at the moment rather than at the very kind of forefront, the kind of riskier sides of what firms may be doing with customers like giving advice, for example, right? So that's kind of what we're seeing.

6:49I think there is a lot there that is changing in terms of customer behaviors. So if you look at what customers are using digital experiences for, compared to perhaps what they were using them for two or three years ago, I think there is a trend happening in place right now. And it's not just, by the way, in finance, but I think it's in many different areas where they are using some of these AI tools a lot more for expert input and guidance and help, right? Whether that be nutrition, health, help with their money, legal, like the number of people who've come up to me recently saying, oh, you know, they had to draft a doc or whatever it may be, that they're using AI to kind of help them out with it, right?

7:37So I think it is being used. It is perhaps not being used as directly as we would expect it to be used. But from what I see working with firms, I think that change is coming. And I'm really excited to be powering it with a bunch of firms from our perspective. So it's off channel, I guess, as you say, right, Vivek? Oscar, come in. Yeah, sure. Yeah, I think something that just like sharpens this sort of disjoint is that those of us who are working with AI every day, it's so approachable, right? Like you can just get your hands on it and you can see what it can do. And it feels like when you're, you know, This is in the consumer tools, but also on the back end.

8:17It feels like the future is just stretching out in front of you, right? You can kind of see where everything's going to go. And then you take a step back for a moment. I think you put your finger on it, Vivek. You're doing your work with the AI. You've got this future envisaged in front of you. You're spraying your thoughts all over LinkedIn. And then we go, hold on a second. This is with five cases where a human is in the loop every single time, right? Talk about scaling that up into some customer-facing use case. It's a completely different animal. And so there is this little tension between what we can see and what we can feel when we've got our hands on it, and then when we're ready to put something in front of people, as you've alluded to in your opening.

8:59So we're not 100 % ready yet, or even close to 100 % ready yet, to let some of these things control our finances, which is, I guess, where the agentic component comes in. even though as you say Vivek people are absolutely using it off channel. I mean, I am, I mean, pretty much, you know, just asking it bloody everything about my life, which I probably shouldn't be. But I guess that's what a lot of customers are doing. Rosie, I'd love to bring you in to this conversation, because as you said in the intro, a lot of what you do is talking to customers and figuring out how they're using this stuff.

9:31I mean, what are you seeing people actually use AI for or maybe the kind of firms that are launching AI experiences? I want to echo what's been said already. And in a sense, we're seeing this kind of mismatch between what banks are wanting to develop and financial institutions are wanting to develop and what customers wanting to use. I feel like a lot of people don't know that AI has been around for 10 to 20 years. It's been working in the background. It's doing those things that customers don't necessarily see and use. And now that customers are starting to see it more on the front end and they're starting to engage with these tools and these chatbots more, they are starting to build an idea of what they want from this.

10:24So through research and the testing that we do, we're finding that customers are actually at the moment preferring AI tools that allow them to self-serve. So they're comfortable with automation. But when AI starts making decisions on their behalf, they feel like they're losing a bit of that control. So they're wanting to keep some of that control and use AI, but on their terms. They don't want the AI to take over completely and make the decisions for them. So we're seeing a bit of this mismatch between their attitudes towards AI and then the products and the services that banks and financial institutions are trying to offer.

11:07Yeah, no, it's very interesting. And actually, I think that's quite relevant to Oscar, your world in Starling to some extent, because, you know, you guys have just launched this spend intelligence tool, which I know is powered by AI in various different ways. And, you know, there's generative in there as well, which is interesting. But I think what you're saying, Rosie, is that it's, I'm interested, but to an extent I need the ability to say no or to stop it or to have some degree of control. Are you seeing that as well, Oscar, in some of the early kind of responses to what you guys are doing?

11:40Well, what we're seeing is that it's useful, right? We can see that people are using it. I think that's great. I think that it's an important component for us in terms of bringing some more of these things. You sort of alluded to agentic use cases, for example. That's something we're certainly looking at. It's an important component for us in building the trust with the public, right? because I think it's worth remembering that sort of my perspective as someone who's deeply into this tech, thinking about this all day, I love it. I love thinking about the tech, but the public is not necessarily there with me.

12:16And certainly, as you were just saying, Rosie, there's a certain amount of trust that has to be earned from the public in order to be able to show, look, we know what we're doing with these things, and we can make your experience better. we can maybe even allow a bit of semi-autonomous behavior in here. But we're going to have to take it step by step so that you're coming on the journey with us. We don't just want to be like landing you with a load of AI in front of you and you're like, whoa, what's this? And are you going to take decisions for me? No, we've got to take you on the journey. We've really got to establish the trust.

12:50Yeah, that makes sense. Even though I guess in some ways people do have that available to them via just going on ChatGPT and uploading some documents and asking it to do stuff, you know, for me. And it is kind of interesting to see how long that is going to persist given some of the regulatory issues that come down the line with letting people do that. I mean, the whole territory is quite fraught, to be honest. Vivek, I want to pick up on your opening point about backend operations and, you know, how it is already been there and is incredibly important. I mean, from your world in advice, Like, how are you seeing some of these back-end, almost invisible customer automations actually make a change to the customer experience itself, even if they don't see it or perceive it?

13:38Well, I mean, we're seeing some pretty significant benefits come through to those core customer relationships themselves, just by virtue of financial advisors themselves and not having to do a whole slew, a whole raft of activities that they'd otherwise be doing. even things as simple as having to write notes during a meeting, having the conversation transcribed, having notes automatically taken, having all the kind of information that's been collected, updated in those core CRMs, all the way through to having follow-up emails, documents, key documents generated on the back of the meeting, makes such a huge difference to the effectiveness of the advice service, you know, just being present in the meeting all the way through to having follow-ups and the core documents in place, not weeks after the meeting, but days, if not hours.

14:32And that's really transforming a bunch of the firms that we're working with. I think beyond that, it is opening the eyes to advisors as to almost like what's possible. There's an advisor we're working with at the moment who they always had like, they had this kind of steady stream of customers coming through one of their channels that they just had to kind of turn down a lot of the leads coming through because they couldn't offer a service at a price point that made sense to those people coming through. But through working with us, they've actually made it possible. It's kind of ended up almost like an eighth of the price point of the original price point of their service, offering a proposition, albeit through all of these kind of automation and kind of productivity enhancements, and making that available.

15:23And that is a total game changer for some of these firms now, and opening up the service of advice to what we hope will be millions more customers. So it's really is having a huge impact not only in terms of the quality of the service itself, but also the kind of resulting price point that can be offered at. And in so doing that, the kind of volume of customers that can now be served. And in doing so, I suppose that opens all sorts of new business model potentials. Absolutely. If all of a sudden the price collapses, then pretty much anybody can launch something like this. And I guess that's the case for any industry.

15:59And I suppose why this conversation is so fascinating because of the tools that this empowers us to do. Correct. Oscar, I'd love to, you know, based on your background as an engineer, I'd love to get a little bit geeky here just for a minute and talk about the fine-tuning of these models. So whenever we talk about generative AI, we are relying on the veracity of the underlying foundation model to be able to provide that intelligence that is then serviced up to that customer. What are some of the challenges that you see in fine-tuning these models for understanding customer intent better? Or is it a question that they get it already?

16:43In fact, it's so incredible, it's more about the regulatory issues. Do you see any technical limitations around the ability of these models to provide the intelligence? Or is the issue less technical and more just adoption? Yeah. Okay, interesting question. So I think for us, it's not a case of the limitations of the foundational models. For us, it's more about being smart about how we operate these things. So for instance, you know, one of the things that we're doing in spending intelligence is that you're going to say, I want to know about my spending patterns in such and such a date, right?

17:23And you're extracting the date from the text that the customer has given you. Well, throwing that at a frontier model is just a waste. There's kind of no point. So in fact, what we want to do there is have patterns where we can say, okay, this is good for a fine-tuned model and we can fine-tune and we can get really good at data extraction in fine-tuning, for example. But this other thing needs to be delegated to a much more capable model. And patterns like that where we're able to mix and match our tools according to the demands that we're placing on the model. Also, truthfully, it saves us a bit of money.

17:59And then the other question that we want to think about here is it's better for the environment, right? Like if we're using a smaller, less capable model for tasks where it's more appropriate, well, then we're going to be burning through less tokens and we're going to be chewing up less watts in a data center. So the things are kind of all pointing in the right direction there as well, I would say. It makes us easier to operate, but it has these other benefits as well. So it's multimodal. There's all sorts of different foundation, like models that you're using and different techniques. It's not a one size fits all by any means.

18:32Rosie, I'd love to come to you on maybe some of the practical examples of maybe things that are working or not. Like we talked a lot about how people are using it and how they're perceiving it. So if you look at examples like what Starling is doing, have you seen any other examples out there that are interesting that you like or that customers like? Of how things are working? AI-powered experiences? Yes. There's an example from a bank called Bunk. I think they're in the Netherlands. And Bunk is doing a really great job of making their chatbot, specifically talking about a chatbot here, feel quite human-like and making the customer feel a bit more in control through some clever features that we're not seeing other banks do in their chatbots.

19:23And the first one is reactive and revolving prompts that change each time the customer opens the chatbot in response to their recent activity. So customers feel like the chatbot kind of understands their needs and they can reach answers faster. And the second is that it's encouraging customers to provide feedback. And you might think that customers might find that a bit annoying having to give that feedback. but it's actually helping customers feel like they're involved in shaping the AI and improving it. That's interesting because certainly something I've noticed when I'm talking to these models myself is I do give them feedback.

20:05I will tell ChatGPT quite regularly that it's screwing up and that it missed something and it'll come back and be like, oh, I'm so sorry. I'm terribly sorry. I should never have done that to you. It's almost as if someone dialed up the sycophancy of these models to make me feel good. But I think that's a very important point that people feel that kind of that human contact. I mean, I suppose the final question that I'd love to ask in this section is how do we avoid falling into a shiny object situation? And I think a chatbot is an example of potentially a shiny object. You have this thing, it's cool, it's got some intelligence behind you?

20:43You know, how do you how do you operationalize that in a way that people are going to continue to trust? I don't know if anyone has any, I'm just open it up to the floor. Yeah, I think so. One of the things that I think is kind of a strength of the way that styling operates is that there's a lot of empowerment in the tech team, basically. So that's kind of where spending intelligence came from, was inside the tech team. Of course, we've got a product organization as well. We're approaching these kinds of things systematically. But we've also got groups of people who are close to the tech, close to the data, close to the business problems that they're seeing every day and they know that need to be solved.

21:27And that means that it's quite possible for us to go from idea to prototype and then put things on the way to production quite smoothly. So to answer your question, how does this stop us doing shiny object syndrome? It means that the people who are doing the work are aware of the business problems and they're focusing on that. And they understand the tech, they understand what it's for, but the tech is not the thing. Solving the problems is the thing and they see that and they feel that. That's how it's working for us. Yeah, you have empowered teams. Yeah, that really resonates with me actually because within Multiplier as well, like very similar stories of engineers working on certain parts of our products and over time being aware of things they could use to like really improve our approach.

22:15And I think actually one of the great things about a lot of the tools out there and almost like the developer community is to build out an MVP can be quite quick. Now, I know what my engineers, I'm sure Oscar will second this is building up, building out an MVP is maybe 5-10 % of really the challenge. Like to get into production is really quite difficult beyond that and takes a lot of rigor and everything else. But like, just by virtue of building out that MVP, you can get a really good sense in quite a visceral way as to whether you're building something that solves that customer problem. And even beyond that does it in a delightful manner.

22:56Like whether you feel that real core product value. And it almost feels magical in a similar kind of way to when people first use ChatGPT or whatever it is. And like a lot of our innovation, a lot of the things we're proudest of have been driven by our, like from within our product and engineering teams, being aware of what we're trying to build out and what we're trying to solve. and that's really cool to see that happening organically from within the team. Yeah, that's awesome. Yeah, and just to sort of develop the thought a little bit further so earlier we were talking about how cheap execution becomes with AIs.

23:41It's just so easy now to get stuff done. What that means is that you really need that empowerment because you're no longer trying to figure out what to do to as great a degree right you're now figuring out okay like i've i've got this thing and i can make it happen am i able to move it forward are there structures in the organization that given that i'm able to do this work that are going to enable me to bring that value towards my customers that's such an interesting point oscar and isn't that always one of the hardest things in in an organization is is just to get everybody to connect and if you have the time to do that then my gosh, what can we actually achieve?

24:22And I guess the larger the organization, the harder it is in many ways, particularly when we talk about banks. Brilliant. On that note, we're going to take a quick break here, but don't go anywhere as in the final part of this podcast, we're going to be looking ahead at exploring what's holding AI back, but also where the real breakthroughs might come next. Stay tuned.

24:49about me. I get a lot of people trying to impersonate me online. Fake profiles, scam emails, the lot. And a big part of that comes from data brokers, hundreds of them quietly collecting and selling your personal information. Your phone number, email, home address, job title, all out there and all fueling identity theft, scam calls, and spam. If you've ever searched your own name online, hands up, who hasn't? You'll know how exposed you really are. That's why we've partnered with Incogni. They contact 230 plus data brokers and tell them to delete your information properly and legally under GDPR and CCPA.

25:26Doing it yourself would take hundreds of hours. Incogni automates the whole thing and keeps working with repeat removal requests if your data reappears. I tried it and within days saw brokers deleting my data. You can even protect your family members too. Fintech Insider listeners get 60 % off an annual plan. Just head to incogni.com slash fintechinsider and use code fintechinsider. And yes, it's risk-free with a 30-day money-back guarantee. You'll find the link in the description. All right, welcome back, everybody. In the first part of this podcast, we were talking about, is it just really shiny?

26:07You know, what are some of the problems that we're seeing and some of the barriers to adoption and perceptions. I'd just love to continue that thread a little bit and talk about the risk landscape to some extent. So, you know, Oscar, I'd love to get your point of view on this about, you know, what have we seen so far come up in risks? We knew this was risky. We're now a couple of years into experimenting and playing with this technology. What are you seeing emerge? And, And, you know, how is maybe government coming in to support? Yeah, sure. Yeah. So I think the thing that I'm seeing is that it's not necessarily the case that whole new categories of risk or issue are emerging.

26:47And rather that things that we already knew about are being amped up. So like a great example of this is in the realm of cybersecurity. security. We were concerned, for example, that people would use AI to develop new forms of attack or coordinate massive botnets and that would be some kind of new form of attack or it would amp up the traffic or something. That's actually not what we're seeing. What we're seeing is that people are using AIs to craft really great phishing emails and send more of them. So these are problems that we already knew about, but we just have more of them right it's not a new category it's an old category and a higher volume and i think that's also just to sort of type to answer the second part of your question what's going on in the regulatory space as well is that there's existing legislation for example in data privacy in relation to ai right it's not the case necessarily that that we need new regulation that specifically covers ai because we have it here in data privacy and we have it here in consumer duty and we have it here in model risk as well all of these things already address aspects of uh of what's going on with ai it's more trying to figure out how on earth do you remain compliant whilst deploying some of some of these these things i mean vivek what do you what do you think yeah i i tend to agree with oscar i feel like um the fca are like i i'd say they're almost in the front foot with a stuff.