1079. Insights: US firms leading the AI customer experience race

0:14Hello and welcome to FinTech Insider Insights. I'm Kate Moody, Customer Strategy Director here at 11FS. Today we're talking about one of the biggest themes shaping financial services right now, AI-powered customer experiences. Over the past few weeks, the 11Fest Pulse team has been researching and benchmarking some of the most interesting AI products and experiences emerging from the US market. Alongside that work, we've also spoken to a couple of the companies building them to better understand the thinking behind their approach. Throughout today's episode, we'll hear directly from those conversations before unpacking what they tell us about the future of AI in financial services.

0:50Joining me today are two colleagues from the 11Fest Pulse team who've been leading that research. First up, we have Joe Colchester, Head of Product. Joe, welcome back to FinTech Insider. Tell our listeners a little bit more about your role at 11FS Pulse. And also, I believe you've been chatting about this topic, not just recently with Inside Pulse, but also at Money2020, right? So lots of things to come with you today. Yeah, thank you so much for having me back. Yeah, I lead the product at 11FS Pulse, so manage the product team and the content team. And yes, I was recently speaking about AI and how best to design AI.

1:25at Money 2020. And a lot of the kind of themes that we'll be talking about today did come up. It was obviously a huge topic throughout the whole of the convention, but it's very exciting just talking about it today as well. Brilliant. Well, yeah, looking forward to picking your brains as we go through. And alongside Joe, we have Olivia Vashik, product manager and benchmark lead also at Lendfest Pulse. Olivia, welcome back to the show. Again, maybe remind our listeners a bit more about what it is precisely you do at Lendfest Pulse and maybe some things you've been seeing across the products you've been looking at.

1:52Sure. Thank you, Kate. So I'm a product manager and benchmarking me at Pulse. So what I do is I look under the hood of some of these AI products and experiences to see what's genuinely good design and what is maybe just AI dressed up for a little demo. And the thing that struck me was that we're adding AI means very different things at different companies. And actually the strongest experiences don't just add AI, but actually start with the question of, okay, how do we solve a problem for our customers? And maybe other thing is that the model that people are using is not really the differentiator, but rather the boundaries and the controls in place.

2:34Interesting. Well, yeah, definitely excited to explore that a bit more as we go through the conversation. So we've got a brilliant team assembled. Let's dive in. Financial services finds itself in the center of a hype to value cycle where brands are mobilizing to move beyond simply talking about the potential of AI and onto embedding it into their products and services. But as we know from the shift to digital, a transformation of that scale doesn't happen overnight. And that's exactly why AI isn't as prominent across the customer experience as your LinkedIn feed would have you believe. However, there are brands making considerable strides, particularly in the US, and they're going to be the focus of today's show.

3:10Joe and Olivia have crafted a list of their favorite experiences that embrace AI from across the pond, which we're going to go through today. So let's start with Robin Hood. We asked them what the core problem they were looking to solve for their customers was. Jordan Sinclair, Robinhood's UK president, answered our questions.

3:28For our UK customers, it was making sure that we met their UK needs. I think it's important when you expand our product started in the US and we've grown to over 27 million customers there. But that doesn't mean your product you can lift and drop and just hope they like it. Spending time with customers before we launched, understanding their pain points. I think in the UK, we heard loud and clear, customers were finding it incredibly difficult to just get easy access to the markets, not pay a ridiculous fee on some of the incumbent providers, but also get access to the market when the news is happening, 24-5.

4:00In the UK, the 2.30 is when the market's open in the US and nine o 'clock that closes. You're in the middle of your workday. A lot has happened on both sides of those times and the market moves. And for a long time, retail just didn't have access to those moves. And I think that's important. If you give customers access to the trade when the market's open, which these days is 24-5 and in time, I believe it will be 24-7, you end up giving retail access to what institutional has had for a long time. I think our AI tools like Cortex and giving more and more products available to customers so they can build a portfolio.

4:34That's the way you localize and make sure that your UK customers have their needs met. but also you bring the innovation that we've developed in the US that we know customers love and that a lot of customers keep asking us for. Yeah, super interesting to hear directly from Robinhood. I mean, Joe, you've examined a number of companies in this space. What stands out to you about how they're approaching solving customer problems? Well, I think we heard it just then and Olivia said it earlier as well. I think starting with the customer problem first, not just thinking how do we add AI because I'm hearing a lot of hype about it and there's lots of great technology out there.

5:13But really starting at the, where's the problem and how can we use these technologies to solve them? That's where the best products are observed, I think. I think we've seen examples across lots of different industries within finance as well. And insurance is another one, for example, Lemonade. Their chat interface is very, very tailored just to making how can we make claims faster and less painful? It's not just a generic chatbot. that is hoping to save money on customer support, it's very tailored and has a clear purpose that's rooted in customer value. Yeah, and I think that's really interesting.

5:50Olivia, what's your take? What's surprising you about how companies are deciding where AI actually is adding value and how to use it? Yeah, I think what's surprising is that first, more AI doesn't always mean that an experience is going to be somehow better. And also that more autonomous AI doesn't mean kind of better or higher value. And sometimes the most value is added actually when AI stays in the background and actually does simplify a task or helps user makes a better decision. So we can think of it on a spectrum, unlike the use of AI on a spectrum from informing. So that would be something like a chatbot answering your queries to advising, which is something that public had, or still has alpha AI, to actually enacting the action.

6:39And the more you move forward to the end of enacting an action, the more guardrails you need to actually prove that the human is still in control, which is maybe counterintuitive, but that surprised me. Yeah, and I think we're definitely going to spend some time today talking about guardrails for sure. And I think, Jo, I thought it was a really interesting point you're making about needing to start from the customer problem and not kind of just start from the technology and how to apply the technology. I mean, we've kind of been here before in the world of the move to digital, more basically, right?

7:12Like when people first started to kind of launch banking apps, we saw lots of people just sort of saying, how do we get everything into the app? Like it's got to go into the app. It's got to be available in the app. Do you think this is just the same problem but with AI? Or is AI increasing this problem or kind of making this technology versus customer problem gap bigger than usual? I think the gap is larger than usual. and I think the stakes are higher than usual, but it is a similar problem to what we've dealt with before. I'm not sure what you think about that, Olivia, but I think the stakes is higher and the expectations are perhaps higher, but it's kind of get the customer problem done before implementing the technology first, I think.

7:53Yeah, I agree. Well, there's so many incentives, right? And outside pressure is to implement AI across all sorts of businesses We're not kind of financial services. It's not the only industry that's experiencing it. I think we are a particular industry and that's rightfully so. We need to, we are and we need to be very closely regulated. And so that is giving a lot of constraint into what AI can do. But then also you end up in a weird situation where, okay, there are these constraints. So the AI cannot do that much, but let's stick it on anyway. And so there is, you know, a frustrating situation, which you're trying to ask this financial co-pilot that's meant to change your life.

8:31how much did I spend on groceries last month? And it doesn't understand your prompt.

8:37So, yeah, that I think I see as a problem. Yeah, no, absolutely. And obviously, you know, AI is still relatively like new on the scene, obviously, but like not completely new, but compared to other technologies, it's new. But do you think we've seen any learning still within the time that AI has been a factor for these teams? Like, are we seeing companies applying it differently now versus they were, say, you know, six, 12 months ago? I think that we're starting to see better interpretations of queries and questions. It's not so, you know, you might used to find out, did you mean this? And it's, no, I did not mean this.

9:13And it would send you down the wrong kind of path. And you have a kind of strange triage that had nothing to do with the original query. It's better at doing that. As we know, language models are very good at interpreting what you're saying, even if there's spelling mistakes, misuse of jargon, local dialect is very good at doing that. So I think that's been one of the biggest leaps that we've seen over the past year is getting better understanding what the user means. And that's very positive user experience. Yeah, Olivia, what do you reckon? Yes, I also like the approach of AI being in the background.

9:44So I'm not a huge fan. All right, these experiences can be good, but, you know, they're very conversational bots with its own personality. I think that can feel a little gimmicky. but there are some experiences like QIIME AI that just works in the background and kind of works through nudges you know, behavioral nudges, prompts that kind of surfaces relevant actions or advice based on kind of all the information and proprietary data that it gathers and I think that can, I see a lot of potential in that more than in the kind of conversational chatbot. It's interesting actually how even all over a year on from when these products have been introduced.

10:28Still the biggest request and concern is, can I off-ramp off of this chatbot? It's still the most important thing to users because they might just not be getting the results they needed. They might just not have any inclination to want to speak to a chatbot. They might much rather speak to a human. So it's winning some people over, but there's still a big void between, I think, what the product teams are pushing and what users actually want, particularly in the case of banks. Okay, no, I think that's super interesting. And probably a good moment for us to jump over to our next clip. And this is from Yannick Malling, the CEO of Public.

11:00And we asked them, what tasks are they intentionally not letting AI do and why? so the way we've designed ai on public specifically our agents is so that if you're building a trading strategy for example they will stick to what you agreed and not hallucinate and try to improvise in a way that meaningfully changes your entire trading strategy basically having this split between having inference in the co-pilot companion that you use to build your training strategy, but then combined with deterministic execution means that it becomes something that's reliable and transparent and that you can actually allocate real capital to.

11:55Olivia, what do you make of Publix approach? I'm glad to hear it. And I agree. I think those trust mechanisms need to be transparent. And these guardrails kind of human in the loop, the fact that the decisions that the AI is taking is very clearly limited. It has a kind of a kale switch or a pulse switch. It's entirely transparent to the user. It's super important because a constraint buried in the, you know, terms and services might protect the company legally, but doesn't build the trust that is required. So yeah, very, very glad to hear it. And I feel like most of the strongest AI experiences that we have seen are about, still about support rather than replacement.

12:46So that human eyes still have the kind of final say in what the AI does. Joe, I can see you nodding. Yeah, no, I totally agree i think we've all experienced um uh talking to language models where they hallucinate they give the wrong answers they pivot and change their minds about something they get the maths wrong so when you confine that to to the financial services space the stakes are so high you really need to nail that and very glad to see that these chatbots do keep it restricted to what they can say and they let you know what they can and can't say because that can be confusing if they don't and they tend to not get things wrong because the guide rails are there because it only takes one very bad example to lose all trust in these bots as it's pretty new technology and maybe something that could go viral so a lot of brand damage could happen there as well so the guide rails are important for the user but also protecting the brand I think.

13:42Yeah, and obviously we've touched on Robinhood, we've touched on Public. I mean, Joe, which other companies are you kind of keeping an eye on? Which ones do you think are really approaching this in the right way? In terms of kind of moving more into everyday banking, we're seeing a lot of good stuff now, even in the US as well. So Chime was mentioned earlier, we just gave a Pulse Pick of the Week to Albert. Albert is called Genius AI. They're very good at doing kind of detailed spending analysis, can respond to queries about how the users should be spending their money and what things they can actually plan financial events like holidays and weddings and kind of thing.

14:20There's also an agentic layer, which we're starting to see. It's quite light touch, but it also allows the user full control over that side of things. Yeah, I hadn't even really thought about agentic AI organizing a wedding. That's like both amazing and terrifying at the same time. Olivia, who have you got your eye on? Who are you watching? Intuit, and they're kind of tax filing AI help, which again, it helps. but the final say always there's a human envelope that has a kind of final say it doesn't just file your taxes by itself and it's also interesting kind of the architecture behind it so from what i've seen they um what's publicly available is that they build a lot of kind of architecture behind the model so they monitor risk and fraud and monitor the incidence of hallucinations and they also have a kind of internal database of the AI interactions so that you get the same kind of experience and the same base across their different tools.

15:22So that's, you know, it's not flashy, but I feel like that's a really good example and a good direction to go in. Yeah, I mean, from a customer perspective, I still find this really interesting in that I completely agree. It's obviously, it's really important to strike that balance between, you know, like AI doing some of the heavy lifting, doing some of the kind of the real analysis, and then the kind of the end customer, the end user, being able to kind of make that final decision. But I mean, certainly what we've seen in lots of other spaces, not just in financial services, but elsewhere, right, is that when you give someone a summary, do they actually read and analyze all the things that are in that summary?

15:58Are they kind of going to look behind the kind of the highest level of information and actually check that all those assumptions are correct? Or are they just going to say like, you've set out these five things, I'm going to go say yes. And if something then goes wrong or they have a poor outcome, if they've taken the end action, again, like where that responsibility lies. Again, the example which comes to me is probably not entirely right, but the example which comes to mind is like big long terms and conditions documents, right? Like, you know, people are signing up to legal documents, but they're not actually in most instances reading the full list, right?

16:35Right. Joe, what do you think? Do you think that we've got the right balance here? Like, are we actually setting customers up to make decisions where they are fully responsible for the outcomes? Or is there kind of a bit of a gray space here? I think largely people are treading quite carefully, which is a good thing. We did an open mic recently, Pulse open mic night, and the consensus was we're not ready for agentic AI. So there has been very few big steps, at least on the customer side, where we're seeing that and people aren't fully ready for it and do require that sign-off, that sign-off we have been seeing.

17:12There are things on the horizon we heard from Robin Hurd and they've just introduced an agentic layer where you can get whatever AI you happen to be using, whatever chatbot to speak to the, to actually set up an account, an agentic account and give it instructions. You still have to sign off on the traits, but it does encourage a lot more trading and potentially a lot more irresponsible trading because it's so hard to get programmatic trading right and who to say that, you know, a new amateur is going to be any good at it. So I guess that kind of thing can give way to ill-advised behavior, but the responsibility is still given back to the user.

17:55Yeah, well, I mean, on that note, So we did actually speak to Robin Hood last week. They announced, as you say, some global accelerations, including agentic trading. And we asked them precisely, given the risks around customers delegating investment decisions to AI agents, what were the biggest concerns they had to work through before launching this and which product guardrails were non-negotiable? So let's hear from them now.

18:17First and foremost, it's important to give customers the types of tools that institutional customers had for a long time. algorithmic trading or using air models to support trading strategies has been around for a long time, but not for retail. But there's a combination of human touch, human oversight, along with the agent. And that's the way we think about this in terms of benefits, but also protection and making it safe. So we allow customers to connect to an MCP of their model of choice, but we make sure that the account that the agent has access to is segregated. The customer can determine what the agent can and can't do.

18:51the amount that's in there where they can trade or just read only, can turn it off at any point in time. I think those are the types of controls that are important as customers kind of learn to use these types of tools. And I think that is that human oversight element where you can get notifications, you can request to approve, all those types of things where you're not just giving it off hands-free and then seeing the balance at the end of the day. And there might be a customer who chooses to go that route, but I think what's important is those controls and then customer support at the same time where customers perhaps got started on the journey, but maybe not comfortable where they are, wants to learn more, offering 24-7 phone chat, email support.

19:34I think that's an important component as well where we may be in a genetic world, but there are still customers who need support and thinking about that customer who's kind of a prolific clawed an MCP user and knows exactly what to do and has built some cool models to the customer who maybe just got started on Google Gemini and doesn't really know where to go. So I think you have to think about both customer sets.

19:59Jay, what did you think of that? Yeah, I think I do broadly agree with that. I mean, I think it's good that it's a separate agentic account. That's the first really good thing that they've got in there where, you know, it's not tapping into someone's potentially quite large main investment account. And that confirmation of purchase at every step, that human layer is so, so crucial at this stage, and probably prevents people who are building it in-house, building something really rash, a magenta AI that can go out there with all the permissions set to the max and just can just potentially do something quite disastrous.

20:34So in a way, it's good that they're building that environment, it's a safer environment. But it's just that question, does it actually encourage positive trading and investment behavior. I don't necessarily know because it opens up quite rash decisions to be made. But it's certainly offering the tools and offering the tools in a sensible way. It's just, it's a lot for people, amateur investors to be dealing with, I think. Olivia, what's your take? I think I agree as well. The positives I see is maybe taking the kind of emotion out of day trading, which I think can a lot of times feel a lot like gambling.

21:15So if this is a tool that can kind of take that impulsiveness and rush decision-making out of investing, that's a good thing. I think in terms of responsibility, this is all a legal and all sorts of ways gray area. I think it's also important here is the interpretation of intent. So this is an agentic AI that's based on the natural language processing. Intent is really hard to figure out. I mean, we see it in everyday interpersonal relationships, and it's easy to tell agentic AI in very precise trading terms what is it that you want to do. But if you are a novice investor and you maybe don't, you know, you're not that clued in and you kind of say something that gets misinterpreted, then that's real money at stake.

22:05and I feel like the interpretation of intent has not been figured out in AI just across the board. So now I'd be kind of interested to see what is in place in those financial services models to kind of guard against it or how to address it. But yeah, time will tell. It's really, really hard to predict which way this is going to go. For sure, for sure. Okay, so far we've explored how some of the most innovative fintechs in the US are thinking about AI from identifying the right customer problem to putting the right guardrails in place. But building something is one thing, building something that customers actually want to use is another.

22:46After the break, we'll look at what makes an AI experience truly stand out and whether great AI products are really about the technology at all.

22:59Welcome back to Fintech Insider Insights. In the first half, we explored how companies are approaching AI product design, from solving real customer problems to deciding where AI should and shouldn't play a role. Now let's look at what separates a useful AI feature from a genuinely great AI experience. We've got another soundbite from Public. We asked them, how is what you're offering different from ChatGPT or other popular models? How has it been personalized? The biggest challenge is not waiting around for a better model or even implementing AI. It's how do you get people to trust AI with their dollars, with their money, with their hard-earned investment capital.

23:38And in order to get there, we had to come up with a pretty novel concept of marrying the idea of inference and intelligence with deterministic execution. architect it in a way and build in a user experience that is seamless and makes it just easy for you to go to your AI and say, I need help with this. And then it just goes and does it without hallucinating, without improvising too much on its own so that you lose trust in the thing. Olivia, I think this is a question I hear time and time again, right? about what do banks and fintechs need to do to stand out from Claude, from Chat, GPT. What do you think of how public have approached it?

24:34I think it's not the model, it's the governance and architecture around it because every one of these companies is broadly drawing on the same or very similar kind of models. So what actually differs is the proprietary data that they can feed it, how much governments they build around it, and then how much they have guarded it from doing something dumb or damaging at scale. And it's interesting to hear them talk about kind of eliminating hallucinations because I don't think that's been done before, right? Like hallucinations is just, it kind of comes part and parcel with using AI. You can't really get rid of them.