1031. Insights: How OpenAI is shaping the future of financial services

0:00Retail investment in the UK is the lowest in the G7. According to the Bank of England, there is over£280 billion sitting in UK accounts earning no interest. Something has to change. Financial firms in the UK must look at making investing accessible, contextual and trusted through everyday platforms. That means bringing investment journeys to the point of need, alongside spending, saving and budgeting, and within platforms that already play a meaningful role in customers' lives. We dive into this and more in our latest report, Taking Advantage of the Embedding Investing Opportunity, produced in association with SECL.

0:42Download your copy today at alumnafest.com forward slash embedded hyphen investing.

1:03Hello, and welcome to another killer episode of Fintech Insider, where we cut through all the noise and get straight to the people who are actually shaping the future of finance. AI has certainly been somewhat of a buzzword in 2025, with everyone from fintechs like Revolut to global institutions like LSEG adopting it. But aside from who's using it today, we want to drill into the brains behind it and explore what's actually doing to reshape financial services. So, without further ado, I'm thrilled to welcome Matt Weaver, Head of Solutions for EMEA at OpenAI to today's show. Thanks for joining us, Matt.

1:37How are you doing? Thanks for having me. Yeah, I'm doing great, thank you. Awesome. Yeah, I'm super excited for our discussion today. So, a little bit of background about Matt and what we're going to be talking about today. So, Matt leads a specialist team dedicated to helping organizations from universities to startups to the world's largest enterprises unlock the full potential of OpenAI's models and products. With a deep background spanning AI, engineering, and financial services, Matt brings a value-focused, hands-on approach to solution design and technical advisory in this rapidly moving space.

2:05Under his leadership, Matt's team partners closely with customers to identify their most pressing challenges and develop secure, scalable deployments of ChatGPT and other open AI tech. They collaborate on everything from solutions design to managing complex security and compliance requirements, ensuring that businesses achieve meaningful results and sustain competitive advantage through AI. So it's quite a critical discussion as we sit here today. OpenAI is announcing major momentum across Europe with new customers, including Revolut, Alicabank, HG, EQT, and Premira, along with a fresh impact data from Zopa and Oak North.

2:40So as stated on stage at the FT Global Banking Summit on December 3rd, OpenAI is highlighting how generative AI is fast becoming the operating system for core financial services workflows, from forward direction and investment analysis to lending and customer support. On top of that, LSEG revealed its collaboration with OpenAI, rolling out ChatGPT Enterprise to 4 ,000 employees and building a market connector platform, that's MCP, that brings its financial data and news directly into ChatGPT. So today, Matt and I will explore what all this means, how AI is becoming the operating system for financial services, as financial institutions move beyond chatbots and AI experimentation, how OpenAI is helping Europe's financial sector adopt AI safely and responsibly, and how AI can strengthen Europe's competitiveness in financial services.

3:29So let's dive in. And we're going to start, Matt, with just a bit of a dive into your career and like what brought you to where you are today at OpenAI? Well, yeah, that's a big question. I guess a lot of things led me here. So I've been doing roles like this one for the last 10 years or so. And that's kind of sitting on the fault line between technology and how that actually gets applied inside of businesses. As you mentioned in the introduction there, my team today is guiding our enterprise customers through understanding the pace of everything that's happening in AI right now. And not only that, how to best apply it in their industry.

4:08So I started my career actually in engineering software, like physical engineering. So working with like companies like Rolls-Royce and Dyson and some of the F1 teams, helping them design their physical products. From there, I moved into kind of a more retail space. So working with application monitoring software, so kind of retail websites, making sure they don't crash and stay up and the checkout's all working smoothly. And then I spent the four years before OpenAI working inside a fintech startup. It's a classic Silicon Valley early stage Series B company. I was one of the first hires in Europe.

4:41And I guess we were an AI company there before everybody was an AI company. This is before ChatGPT. So writing lots of Python code and doing classical machine learning. And the thing we were solving there was extracting unstructured data out of documents, which financial institutions have at huge scale. Imagine if you're submitting documents for a loan application, someone has to transcribe all that data out of those or doing KYC and onboarding. What was really interesting about that experience for me and that startup before I came to this role was that I had a front row seat to the way that the entire industry was disrupted as this new wave of transformers based AI models, which is a new technology that came out around 2019, and then large language models.

5:24And the moment ChatGPT burst onto the scene just over three years ago, that kind of changed everything. And it became clear to me that the best way for me to have an impact and to continue working on these super interesting use cases inside of large enterprises and the financial services industry was to actually come on board at OpenAI. So we've been rapidly scaling the team out here. It's a super eclectic mix of tasks. Some weeks on stage talking about the way our products are used. Other times working closely with security teams to help them kind of unblock and deploy our technology. and really the thing that interests me most about this area.

6:01I think that the financial services industry can sometimes have an unfair reputation for being slow to adopt new technology. It's true that this space is regulated and that can sometimes slow things down or cause friction, but actually the opportunity inside of financial services is bigger than almost any other industry. It's one of our fastest growing segments in terms of the customers that we work with at OpenAI. and the enterprises that are doing this really well today, and I'm sure we'll talk about some of them, are seeing huge gains already. And what's most exciting there is that as this technology continues to improve, as the pace of AI continues to develop, we're going to fundamentally transform the way that this entire industry operates.

6:42Yeah, it's an exciting place to be in financial services for a change, because I've also been in the industry for very, very long and I agree there's so many interesting use cases. Your background, I think, just gives you this very interesting kind of helicopter view of how different types of institutions are approaching and adopting AI. Are there any kind of patterns that you've noticed, you know, fintech versus incumbent, government institution versus private, as to how they're approaching AI? Yeah, for sure. I would actually say what's been most interesting, and you're right, I get to kind of speak to, you know, hundreds of companies at a time around how they're thinking about and deploying AI.

7:22And I actually would say that the patterns that we see aren't necessarily fragmented by legacy or newer kind of challenger institution. We see examples such as BBVA, which is like one of the largest banks in Spain and Latin America, adopting our technology. They're just rolling out now ChatGPT Enterprise, the secure enterprise-grade version of our platform to all employees globally, which is like really exciting. And then at the same time, smaller organizations such as Zopa, more of a fintech player, are successfully deploying our technology both ChatGPT Enterprise and then also using our APIs to develop ways to increase the quality of their customer service interactions.

8:06But the patterns, so it's not really broken down by if you're an older, more legacy institution or a newer one, but there are these common patterns that we're seeing of the companies that are doing it really well. I think the components of an AI strategy for a financial institution that really work, you kind of have to tackle it from two angles, right? Bottoms up and tops down. That bottoms up component is really about employee AI literacy. We see that, you know, it doesn't matter how brilliantly written your AI strategy is, if it's not kind of just built into the foundation of the way that everyone inside your organization is doing work, if they're not intuitively able to understand the strengths and limitations of AI as it stands today, then we often see that you can get stuck.

8:49So it's really important that you just are deploying tools like a chatGPT enterprise so that everyone's familiar and able to use those. And then on the tops down component, then having like really strong buy-in from the board level of what are the big priorities, the bets that we want to make as an organization to really invest in truly transforming the way that our business operates, whether that's the process of lending out credit, whether that's the way that you interact with your customers through customer service, picking some of those really strategic things. And this is not something which is done on the side, you know, as like a side project, it kind of needs to be core to the overall strategy of the overall business, because there are companies out there that are moving really quickly.

9:32And so if what we see there is that, you know, when you have that tops down sponsorship from the senior executive level, those companies are the ones that are making great progress. Yeah, absolutely. And like, what would you say are the key lessons that you've learned, or maybe some of the risks that you see in getting financial institutions to move beyond experimentation? Because I think what you said that was key in that, like, it's not about doing this side of desk, it's about doing this core. So core means in production level, you know, maybe customer facing, maybe not customer facing, but certainly touching critical processes within the bank to actually make a change.

10:11How are you seeing institutions sort of push beyond experimentation? Or what would you say are some of the principles that you're pushing for to get more production? Yeah, exactly. And to double click on that and expand on it, I think I heard from a customer recently that they felt like in the first half of the year, they were stuck in POC hell. You know, like lots of experiments happening, but not a lot of progress to production. I think what's really promising is that in 2025, we have seen these organizations move from POC to production for the first time. So your question is a good one, which is like, how are people doing that?

10:45What is it that's breaking beyond the POC stage? That's awesome. One of the first things, AI literacy is important. Executives and leaders actually leading from the front in their own personal use of AI is another key one just around overall adoption. And when we talk about going to production, I guess there's two categories, right? employee productivity overall as a topic, whether that's software developers using tools like OpenAI's Codex agent to accelerate their workflows or the use of a tool like ChatGPT connected into your enterprise data. Those are things that you can go live really quickly with because they're internal.

11:19And so the risk is lower. You're really just trying to accelerate people's productivity. When you're building externally facing experiences, for example, Revolut, one of our customers, have built their Rita assistant, like a Gen AI assistant on GPT-5, which is used in kind of customer support queries for end customers. So how on earth do you do that? In a regulated environment, how can you take an AI model and expose it to your customers? Well, it's actually, the good news here is that there is a solution and that is to build the right guardrails and evals. And I'll just unpack those two terms because in regulated industries, it's true in financial services, it's true in healthcare and life sciences and pharmaceuticals.

11:59you really need the ability to kind of measure the performance of the solution. And when you have a chatbot, you know, it's not like one plus one equals two. There's kind of a open natural language conversation. But an eval is basically a test. And you can write a few thousand of these that says, hey, based on the user's input request, what kind of response do we expect the model to return? And you can build those in an automated way. And a lot of the time that we spend, my team and the technical teams at OpenAI who partner closely with these large enterprises, we can provide expert advice guidance often will come in and like show you how to do how to do this directly which means that you can actually then measure with the percentage accuracy the performance of the solution and this enables you to go live with confidence but also satisfy the regulator that the solution that you've built is incredibly performant and so you'll often any advice anytime you speak to an open ai engineer giving you advice on this topic they'll say start with evals we often spend you know the first three months of a project writing the tests before we even start building the the agent, the AI agent.

12:57Because if you don't have the tests written up front, you can't possibly know how you're tracking as you're building that solution out. You kind of have to have that as the foundation. So, and then having guardrails in place. You know, sometimes guardrails could be something simple like a calculation. If I'm extracting data from someone's pay stub to validate their income for a loan, then I can add the numbers up and check that they compute correctly. So not everything has to be an AI LLM request. It could just be a traditional kind of calculation or something that's more what we call deterministic.

13:28You can just calculate it with logic. And then the other really powerful component of guardrails is often we'll have one large language model check the work of a previous one. In many critical banking processes today, you may have like a maker-checker process, right? If you're doing very large transactions, often one person will do the work and a second person will check the work of the first person before that transaction goes through. And we can do the same thing inside an LLM solution. So, you know, you might say, here's the output that we produce that we're thinking of sending back to the user.

14:01Here's the inputs we gave. And then you have the LLM kind of double check all of that work. And this enables you to maximize the accuracy, put the right guardrails in place. So these are just two examples, evals and guardrails of techniques that you have to kind of layer on top of the AI model itself to make it really robust and reliable inside those production environments. And we see this, you know, BBVA, also a bank, I mentioned them earlier, they have an AI assistant built into their mobile banking app called Blue. And Blue can query your transaction history, you can actually set up payments to send money to a friend using this AI chatbot.

14:38And that's only possible because they've done a really amazing job at building these guardrails and evals into the end-to-end development process, which takes them from POC all the way to production. That makes total sense. And I guess those evals and guardrails are also then how they can manage their internal audit and risk reporting and reporting to the regulator and all that. It's just amazing to see how an LLM can in some ways mirror an internal audit process. You would have somebody checking this anyway. So instead of a somebody, you can have another LLM doing that, which is just super interesting you say that.

15:10Yeah, exactly. And I think sometimes like a mistake that's easy to make is to think, well, I've just got this model and I'm going to ask the model a question and it has one shot to get it right. And if it makes a single mistake, then the whole thing isn't going to work. None of our internal processes today work like that. Often for important decisions or for critical processes, you have a few people collaborating together to check each other's work and make sure that everything's really high quality and we can create that same construct inside of these agentic AI applications. And this is where we partner deeply to give that advice and work closely with customers to kind of guide them through that process.

15:46Yeah. Let's talk a little bit about BBVA, because I think it's actually very interesting. You're doing the chatbot, but I understand there's a whole bunch of stuff you're planning with them. Just tell us a little bit about that. Yeah, really exciting. We recently announced this new strategic phase of our partnership together. I spent a lot of time traveling back and forth to Madrid this year, working closely with them on this. And I think what's exciting there and the scale of their ambition is about deploying ChatGPT Enterprise to all employees. They really see this as kind of the super assistant that's going to assist everybody to be more productive.

16:20Some important things there, ChatGPT Enterprise, from a business context, we never train on any of the data. It's a secure enterprise-grade version of the tool. You can also connect it into your enterprise systems, right? So think about connecting into SharePoint, connecting into Google Drive, actually adding the intelligence of your entire organization into a tool like ChatGPT and not just relying on kind of searching the web. So you can have a really like enterprise context enriched experience. But then beyond that, you know, whether that's across customer service and kind of the external facing applications like this blue assistant and the mobile banking app, which they built, where we're going to enhance that much further.

16:59But then also internal processes. When you think about how to empower bankers who are very often interfacing with clients, helping them do account research, provide the best possible advice, preparing for meetings, following up after meetings when they have those external ones. So that's like a key component. Back office operations, I mentioned much earlier in my career, I was working on kind of automating all the paper that flows through a bank. And there's lots of things we're going to be doing with BBVA there to streamline some of those back office processes, which can ultimately just shorten approval timelines, shorten client onboarding, give a much better experience for BBVA's customers.

17:35And then we're working with them deeply on kind of expanding how their software developers can be productive. I think many banks today, and this probably resonates with many of your listeners, banks kind of are software companies in some ways. There's a lot of internal systems to build and maintain. There's also a lot of legacy knocking around. Banks were adopting technical systems back in the 70s, 80s, 90s. And so anyone who's worked with an IBM mainframe and some of the legacy COBOL code there will know that it's a pretty challenging project to migrate a lot of that code. And so we're supporting them through some of those things as well.

18:14And then finally, in this kind of multidisciplinary approach across the entire business, the other one is around managing risk. How can you actually use AI to make better risk decisions? because ultimately many financial institutions, the business is about offering credit, which helps to stimulate small businesses by lending them money to stimulate the overall economy. So there's a lot of good that can be done by shortening credit approval timelines and actually making higher quality risk-based decisions, which is good for the bank's own balance sheet and book of risk as well. Yeah, there's just so much value potential there, isn't there?

18:50I'm just curious to know from your perspective if there are any surprises maybe since you started, any sort of surprising successes or failures or things that, you know, were kind of an assumption that may be obvious about how people adopt AI versus what you've seen. I'm just curious to know if there's anything there. Yeah, absolutely. I think 18 months ago or so when I joined OpenAI, there was still a narrative around kind of AI being quite expensive, like an extra cost. And that was true, actually, at the time. I'll give you one of my favorite statistics. So when we first launched GPT-4, if you were to consume it via our API, it would cost around$60 per million tokens.

19:32Doesn't really matter what tokens mean in this context, but about$60 per million something. Today, for the same level of intelligence, for our smartest model, GPT-5, it's 99 % cheaper. It's just a few cents for the same amount of intelligence. And that trend has been pretty consistent. uh over the last couple of years of like the cost of intelligence falling as the models themselves get even smarter and what that means in practice for business leaders is that often you know you'll be people will be in a planning period now for how 2026 uh is going to unfold and where investments are going to be made and so even if a use case looks a little bit out of reach from a cost perspective if ai is like maybe not quite cheap enough today to make that viable um i recommend like still start building and start experimenting because in six months from now and nine months from now the cost will come down and all of a sudden a whole new wave of use cases becomes accessible um so like the cost coming down like to me was surprising i don't i think many didn't expect it and some still have this idea that ai is really expensive but the costs have just been falling dramatically whilst at the same time the pace of model intelligence except increasing is accelerating.

20:44We launched GPT-5 back in August. A month ago, we launched GPT-5.1. And then just one month later, we launched GPT-5.2. So you're going from like two or three months between model releases to one month between model releases. And I think you can expect that to accelerate even more into next year. And these are trends that it's kind of hard for our brains to kind of grasp this exponential. But we've seen this now continuing over a period of a couple of years, and you can kind of bet on that moving forward as you plan your investments into next year. There's something, I think that's really fascinating.

21:24There's also something in there about the kind of long tail legacy of, shall we say, slightly older models and how those get a hell of a lot cheaper over time. Because one of the things that I've picked up on as well is that intelligence in many ways is very important, but actually there are a lot of back office processes where maybe you could argue that the existing models today or maybe some of the older models are kind of okay, right? And so if that price sort of just drops so much, then the business cases to implement those things across some of the back office process has just become a lot clearer as well.

21:57I mean, it's a horrible analogy, but it's a little bit like you can have, I don't know, like an iPhone, but also you can use a feature phone for some things as well. And people still sell feature phones and people still buy feature phones. There might be something about the tech in the fullness of time that remains. It's not always about the kind of the thing that is like approaching AGI or AGI, but it might be about these other things as well that kind of came up along the way. Yeah, and at OpenAI, we provide like a full suite of models. I think most people are familiar with the ones that they see inside of ChatGPT that people are using every day.

22:30And, you know, we now have over 800 million people every week using ChatGPT in the consumer space, which is like 10 % of the world's population. so that for many people, AI is chat GBT. But in the background, we have, you know, GPT-5, GPT-5 Mini, GPT-5 Nano. So we launched these kind of like smaller, cheaper, faster models because many use cases, you know, they don't need that frontier level of intelligence. We now have, you know, over a million business customers who are building on our technology and millions of developers worldwide kind of building their own AI experiences on top. And many of them are using kind of some of the smaller models for those faster use cases where it makes sense.

23:11Yeah, that's awesome. All right, on that note, that wraps up part one. So in part one, we laid the situation. So what's going on right now? We talked about Matt's background. We talked about how AI is being adopted across Europe and we set the stage for the future. So we're going to shift into what's next for the sector coming right up. Hey folks, David Breer here, CEO of 11FS. Here's something you might not know about 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.

23:49Your 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. Doing 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.

24:29FinTech 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.

24:47Welcome back. Now, in this section, we're going to focus on AI's future in financial services and how banks, fintechs, and insurers could be leveraging it strategically into the future. So to kick this off, I'd like to talk a little bit about the concept of AI becoming an operating system, right? Sort of leveling up a little bit beyond that we're using it as a tool for productivity or a tool for a specific use case, but into that broader, you know, view of how do you operate a financial institution around this new technology? This is a really great question. And I think I was actually speaking to a very senior banker from one of the world's biggest banks yesterday about this exact question.