Name: How AI is changing the future of banking
Uploaded: 2025-12-11T18:46:21.530Z
Duration: 30 min 8 s
Description: How AI is changing the future of banking

Transcript for "How AI is changing the future of banking": Hey. Thanks for joining us for this webinar about how AI is changing the future of banking. I'm Matt Weaver, head of solutions engineering at OpenAI, and I'm joined today by Elena Alfaro, the global head of AI adoption at BBVA. And we're gonna learn about how financial services organizations are successfully adopting AI across the board. And gonna tell the story of how BBVA's own adoption journey have enabled them to achieve great things already. Elena, maybe we could get started. We'd love for you to just tell us a bit about yourself and your role. Of course. Thank you very much. I am Elena, and I lead a team that is called, global AI adoption within BBVA. And our mission is quite simple. We're trying to make AI a tool for everyday usage, you know, in our among our employees. So they see it as something that is really useful, not just a side experiment. No? So we do several things. Of course, we deploy, TileDB Enterprise, but we also try to help our users, in the discovery of use cases, in the creation of custom DTPs, etcetera. We provide them with training. We also have a very, very dynamic community of practice, among many other, initiatives, so they really engage with us. No? And we are basing all this in in several principles that we we really like and we we stress a lot. One is trust, another one is, experimentation, and the last one is, human empowerment. Amazing. Well, it's been amazing to watch BBVA's own AI adoption journey. And we think there's some lessons here for the broader financial services industry. So today, we're gonna walk through some of those, and use some real examples and demos to show how AI AI is being adopted across financial services. Before we get started, just some quick background. I'm incredibly lucky to work with some of the world's leading financial institutions on how to deploy AI securely across their organization, and it's amazing to see the progress that's already being made across the industry. We work with over 90% of the Fortune 500 and over a million business customers. And in a regulated industry, let's start with the big question. What about compliance? What about data security? We'll never train on your business data on an enterprise plan, and we have all of the enterprise security features and controls that you need to be confident in deploying AI within your organization through tools like ChatGPT Enterprise, which we'll take a look at today. Now there are kind of three really successful patterns that we're seeing in AI adoption across financial services. I'm gonna start, and walk through each of these with a demo to learn a bit about how BBVA have put this into practice. And we'll start with empowering employees by enabling AI literacy for all. So, Elena, to get us started, we'd love to understand just a little bit before we jump into the demo. How is it that you kind of enabled ChatGPT access across the organization? How did you decide who should get access and how they should use it? Yeah. Well, first of all, we are a small team, that works here in Madrid, centrally, but we work with a network of what we call AI adoption champions in all the countries and in all the areas. No? So the first thing that we did was to provide them with a set of licenses. We decided to spread them among all the areas, in the VBA, not just the, you know, specific selected teams like engineering data operations. We decided that this should be tested by everybody. So we gave them a a number of licenses, so they decided who should be using them. We just advised them to do it with people that are really, you know, engaged with the the technology already that could dedicate the time, that could be, you know, these persons that are willing to help others. And and with this, we started to create this network, to spread, the usage of the different tools. Another one of the key features of ChatGPT Enterprise that's been used widely is custom GPTs. Maybe you could just describe to us what a custom GPT is, and then we'd love to see one in action. Yes. A custom EPT, I like to describe it as something that, is an application that can be created by somebody, anybody that is using TPD today, and that can can solve a specific problem or, opportunity that this person or this area so you just have to, provide it with a knowledge base on the topic that you want it to to be useful about, and then give it some instructions. And then you can interestingly, you're gonna start sharing it with other people that can just reuse. So you could see this as a personal software in a way, and this has been incredibly successful, and we have been creating thousands of GPTs. Now we have more than 4,000 active in BBVA. And, yeah, I think this is one of the best features of the of the whole, LGBT enterprise. Amazing. So we'd love to see one of these in action. I know that your colleague, Concha, is gonna join us to show us, Amelia, the GPT. Hello. Hey, Concha. Hi. Thanks for joining us. Thank you. So we're gonna take a look at your custom GPT called Amelia today. Can you tell us a little bit about, like, what is Emilia, and why did you create it? Emilia is an expert in content design, and we need to create it because we need, provide to teams, a guide, a help to to can write content and communications with a radical customer's perspective. So you have a lot of brand guidelines about how to speak to customers with a radical customer perspective at BBVA. How does this custom GPT help you to do that? How does it work? Well, Emilia, knows all that guidelines and understand that and, is, able to to connect all the, all the guidelines with each, between, other. And this is, the magic of of Emilia. Amazing. Let's take a look at a real example. Okay. So to get started, we can see on the left hand side the Emilia custom GPT. We'll select that. And we can see here some suggestions for what we could type pop up. Let's start by just saying hello. Okay. So my Spanish is a little bit rusty, so we can ask Emilia to speak in English for me. Cool. So the next thing we can do is I'm gonna ask for some help to improve a message that appears on BBVA's commercial website. And for this, I'm just gonna add a screenshot of the file. So here we're using the multimodal capabilities of the model. I don't just have to type in text. I can also provide an image, a screenshot, and get some help there. The model's gonna read the the copy on the page, and we can see it's gonna give me a revised version that's aligned with BBVA's verbal identity. This works so well because we've actually embedded all of the brand and communication guidelines into this custom GPT, so it automatically knows how we should revise this. The headline's been optimized. We've made it clearer and simpler, and we've improved the tone and alignment as well. Now we could take this one step further. I could say, could you give me some more options for the headline? I think it could be better. And we'll get a quick suggestion as to what could come next. So here's some alternative headlines that follow the small and smart style that's embedded inside of BBVA's brand guidelines and recommendations, and we can take it from here. Now this was a super simple example of how we any user can create a custom GPT to solve challenges in their day to day. We can also take a look at a much more complex example, such as this credit risk assistant GPT. This can be used for helping me do all kinds of different credit risk analysis as an analyst. In this example, I'm gonna calculate the value at risk number or VAR based on a given data set. You can see here I can give a relatively simple input using the attached CSV computer one day VAR for a $100,000,000 long British pounds to USD position via historical simulation and the 100,000 SIM Monte Carlo. I'm gonna upload the file. And behind the scenes, because this credit risk assistant has been built with all of the instructions on how to behave and how to perform these analyses, it means that I, as an end user, can put a relatively simple prompt in and just get the answer that I need. So we can see here that in just a few seconds, it's actually been able to do that computation and create these rich visualizations right here inside of chat. It gives a little bit of a description about the inputs that we used, the methods. This is super important for explainability in financial services. You don't just want an output that you can't understand. It's really important that we're able to see behind the scenes at exactly how this was calculated. We can see we get the one day VAR position, some diagnostics, and then a commentary in less than 250 words. So you can see here that by building this as a custom GPT and a system which is already embedded with all the instructions on how to do this analysis, I, as a user, could very quickly just get the answers that I need. And importantly, they're fully explainable because built into the custom GPT, we gave instructions that it should always explain it's working to the end user for full auditability and explainability. So it was great to see that example, of the Amelia GPT, which is for content creation and content writing. But you've given access to this technology to all of your employees, across different areas of the business. So it would be great to understand, how you're seeing it being used elsewhere. Sure. So there are so many examples that it's very difficult for me to pick just one. But, for example, in the legal space, we have several DPTs that are experts in a specific area of legal, like personal data, contracts. We have one that supports our lawyers when they get questions from the branches, so that is big. Then we also have in the areas like risk management, for instance. There are so many also. One of them is helping the risk analyst to analyze the ESG perspective of the companies that come for a loan, going through the annual memories, through the websites, and other documents, and they provide this summary of what is the approach to ESG, for instance, which is important for us to give them loans. Amazing. And how much time is that saving in that process? Well, we are measuring this, with experimentation, not only, because people feel that they are saving time. And so in many cases, we are seeing seventy, eighty, 90% of time reduction in these processes. Amazing. It's great to see what happens when you put this in the hands of everybody. Absolutely. So the second pattern that we're gonna take a look at is leadership as an accelerator to adoption. We're often seeing across financial services that where we see the highest adoption and value being realized from our technology is where leaders themselves are leading the way, and actually being power users and advocating for the use of AI across the organize station. So, Elena, you did something really interesting in the way that you initially deployed ChatGPT Enterprise across BBVA related to leadership. Can you tell us a bit about that? Yes. So we we started with some thousands of licenses, and we thought that it was very important that the leadership of BBVA got the got to be the first users of this technology because we think that they really need to understand how this technology can make and experiment them this in with their own hands. No? So we started monitoring how they were doing, and some of them were doing really well and others, well, maybe not that that good. No? So we decided to, design a specific training, so they could learn to use it and find it interesting for their own, daily jobs. No? So for instance, analyzing presentations, preparing for board meetings, KPIs analysis, also giving feedback to their teams. So after these sessions that we made, kind of mandatory, for all of them, more than 200 top leaders, including the CEO, went through the training and we follow-up on the metrics after that, and we saw them increasing their usage, quite a lot. Amazing. And it's great to see where you have leaders as a power user of AI themselves advocating for that and showing it their teams what they're using it for in their own personal jobs. We see that driving adoption across the board. I thought we could take a quick look at a demo of how a leader might actually use ChatGBC to help hack their own job as an executive. So I'm actually inside OpenAI's Atlas web browser right now, which has ChatChBT built in. So as I'm navigating across the web, I can bring in assistance from a super smart AI at any point in time. Now in this example, I've been sent an interesting article from a colleague about agentic payments and agentic commerce. If I scroll down a little bit here, we can see there's a headline and a a bit of a summary at a glance. But I can actually bring ChatGPT with me onto this web page by clicking this ask ChatGPT button on the top right hand side. Now one of the first things I usually like to do is research the authors a little bit to understand their credentials and how they know so much about this topic. So I can select them by highlighting that text and then ask, tell me more about this author. And we'll see here, I get a quick answer with a bit of background and the professional profile of this person. They work at McKinsey. They've been there for quite a while by the looks of things. They've got expertise in technology, strategy, and architecture. So that tells me that they are a relevant person to be writing an article on this topic. Now the next thing is, you know, I could read this whole article, but I'm actually most interested in how this applies to me and my business in the region where I operate. So I can ask a question such as, what are the impacts and opportunities for for European banks? And we'll see here that in just a few seconds, I get a clear and structured view on how agentic commerce might impact, European banks. We've got this intermediation of traditional consumer touch points, a shift from card based rail dominance to protocol based dominance. And I'm gonna scroll down here a little bit further. We'll see that I'm gonna get some examples of how the regulation in my region might actually apply, because the EU has the AI act. Maybe that's something I need to consider, but there's loads of amazing opportunities as well to build some agent ready payment products. So this is great. I like this analysis and how it's been broken down in a way that's relevant to my business in my region. So maybe I actually wanna share this with some colleagues and write a memo to share with the board to get their input and feedback on how we can take advantage of this opportunity. So now I'm gonna say that I wanted to write a proposal, search for more information from other sources to supplement this and create a Google Doc with some insights, and then also comment in that document where I should review before sharing. And for this, I'm gonna enable the Atlas browser's agent mode, which is gonna give the agent the ability to use my web browser to search for other sources and even navigate tools like Google Docs or Office three six five applications. So we can see here that the agent is gonna, take control of my browser. I, as the user, of course, am always in control, and you can see down here, I have the option to take control at any time, and I can monitor and review everything that's happening. But I can also just hand off this task. It's a simple low risk research task, and so I'm gonna let the agent continue working. We're gonna move on and find a different task to work on in the meantime. Here, I'm gonna do some quick research into the data behind BBVA's green financing initiatives. So I can run a quick search query like this, and we can quickly call back all of the key data from the most recent annual report. We get some key targets and milestones, different allocations and amounts of the portfolio, and some information for the social bond allocation as well in this tabular format, which is nice. But tables maybe aren't always the most interactive way to digest information like this, so I'm gonna ask for an interactive dashboard. Here we can use the canvas feature, and I'm gonna turn on the thinking mode. And JWT will generate the code to create a beautiful visualization. We can see that this happened here, and we can just hit this preview button to see that dashboard. Awesome. This is a much more powerful way to view and interact with that data. I can see these pie charts where I can hover with my tool tip. We can see a side by side comparison of the two portfolios, and some eligible versus allocated and the allocation mix. So this is a much more interactive way compared to that text format for me to get quick insights as an executive. Now we can come back over to the Google document that's been completed while we were working on something else, and we can see here from the top that we've got this memo that was written for the board, the model word for around four minutes, and we can see it's got those market overview and trends, impacts on traditional banks, opportunities for our bank. And then as I asked, it's actually gone through and also added some comments because it's the agent is able to interact directly with my Google Docs, to add some areas where I might want to, double check the work before sharing with my colleagues. This means that I can hand off research work like this to my agent and come back and have a lot of the the heavy lifting done already that enables me to review this and then share it with my colleagues. A massive time saver for executives who are trying to get to insight super quickly. So let's move on to the the final pattern, which is kind of bringing your company's data into ChatGPT. I think most of us have used ChatGPT to search the web and to do research, but it becomes even more powerful in financial services when you're actually able to connect it to your company's valuable data and get real insights from within there. So, Elena, I know that, you've been experimenting with connectors in ChatGPT, and you have a real vision for where they're gonna take your next phase of deployment. Yes. Absolutely. So, of course, security, it's a it's a key aspect of the whole thing. Having SolidityPT Enterprise as a secure platform, it has been key key since the very beginning. Also encryption, data protection, data residency are key aspects. Once we are there, there are further steps that we can take. So we are now deploying, this idea of connectors. So the idea here is that we can bring data from different sources and also connect to applications, inside BBVA so we can start to speak to those data sources or applications. And this is gonna unlock a new, no, category of use cases. So for large financial institutions like BBVA, where you've got all of this data, often it's buried or locked away or hard to find, using connectors in chat GBT, you can really put it at the fingertips of every employee. Absolutely. We have platforms or tools that are broadly used, you know, like Drive, Gmail, etcetera. But then we have a lot of, again, no, domain applications and databases, where people, would find it much more easy to, use through a conversational interface instead of just clicking or finding stuff inside. Amazing. So maybe let's take a look at that conversational interface inside of ChatGPT. Okay. So I'm just gonna open a new ChatGPT window here, and We're gonna take a look at two examples of bringing your data into ChatGPT. The first one is gonna be external data sources, like market data. And secondly, we'll look at how to pull your internal data and insights into chat. So to begin with, I'm gonna connect up to our London Stock Exchange data connector. This enables me to pull live market data and insights and news from the London Stock Exchange. And we'll take a query like this. What's the cost of hedging €10,000,000 to USD exposure for three months versus six months? This is a query that's gonna require getting the current price for each of these scenarios. And what's really powerful here is that the connector has many tools available to get different kinds of data, and it's just gonna automatically decide and figure out which ones to call based on my query. And we can see here that in just a few seconds, we get the forward hedge costs for these, two different options. We can see the forward rate, so all all the key data, a very clear cost comparison, and then, the actual dollar difference. So super powerful way to pull data from trusted, third party data providers into ChatGPT to get quick insights in this super conversational way. Okay. The the second example we're gonna look at is around using your internal data. You'll see we have this really handy company knowledge button that enables me to get the benefit of all of my internal company knowledge. You'll see here I can connect up many different types of connectors from Google to to Microsoft ecosystem, and many more. And we continue to add more and more connectors over time. So we just wanna get the status of a recent project that my team's been working on. What's the status of our new green rewards project? Many of you may already be familiar with using Jack GBT to search the web, but this is a really powerful way to search all of your internal data to answer questions that are totally internal to your organization. So we can see here that in just a few seconds, we've searched across many different internal sources. I've got things across Google Docs and SharePoint in these results, and we can get a quick update on the status of the project. So overall, we're on track, but I can see here some things that are at risk or delayed. Critically, I can actually see all of the sources of this information and click into those to go and open that file in SharePoint if I want to. I can also understand a little bit about around who owns which component. So maybe now I want to draft a simple email to the team outlining what needs to get done this week. And here we go. For each of the different work streams, for each of the different owners, I've got very clear deliverables, all pulled from the insights that are spread around my internal knowledge bases and and drives. And finally, we've been considering investing in a neobank recently. So I just wanna find out what the status of that is. Okay. And I can see here that in just a few seconds, again, I can get quick insights from across many sources. I've got things in Google Drive. I've got Excel sheets, full of data, and get a quick understanding of where the data is showing that we're behind, whether this investment might be in trouble, and then where I might want to lean in as suggested next steps. So company knowledge and the ability to connect ChatGPT to all of your enterprise data really means that you've now put the knowledge of your entire organization at the fingertips of every employee, which can be incredibly powerful for productivity and for the entire organization. So in conclusion, I guess what we've seen here is some of the patterns that we see of successful adoption of AI across financial services. Empowering the teams closest to the work to actually use AI and solve their own problems. We saw with the immediate GPT how individual teams could just be empowered to go and solve that right away. Leadership as the adoption multiplier, actually making sure that executives are some of the first people that you enable and that they themselves become power users and drive that in their own organizations. And finally, kind of moving from an isolated AI, an AI model that can answer questions, to really connecting that into your enterprise data to maximize the value that employees can get. Elena, any final thoughts? Looking forward to what is coming next year. Very excited about, what we're gonna do with data connectors and especially the step after that, which will be, identity workflows. Amazing. So if you'd like to find out more, we know that adopting AI inside a large regulated institution is a journey. We've got loads of resources to help. We'll be sharing more more details, which you can access after this webinar. If you'd like guides, case studies, tools and templates for how you can do this inside your own financial institution. And, of course, we, the OpenAI team, are here to help. Thanks for watching.