Capital One Taps Open-Source, Cloud, Big Data for Advantage in Banking
Are there any key technologies such as programming languages or frameworks that you have taken a particular interest in? I know you guys have a keen interest in Node.js for one. We have a large innovative organization and we're doing a little bit of everything. We've certainly had a lot of success with Node.js. All of our mobile applications we build natively. So for iOS, we build apps using Swift. We do native Android apps. There are just a lot of the modern technologies in use here. Our Web technologies are all Angular and React based. We want the best performance in our apps and we also want to attract the best talent. And they want to work in the most leading-edge ways. That's really important to us—being able to leverage those frameworks and do it on the public cloud. We have a great partnership with Amazon Web Services and being able to innovate on the public cloud. It allows us to move at a speed that enables us to innovate quickly. There's a whole data dimension to this as well. We heavily leverage open-source data tools to build our data solutions. If you want to get to real-time banking, you really need to move to a streaming data kind of platform. So we have a heavy implementation of Kafka for instance. We stream our data in real time so we are able to make our decisions in real time, whether we leverage things like Spark or Apache Apex or other tools that allow us to do stream processing those are critical technologies for us. Those are all predecessors to being able to apply more machine learning and artificial intelligence kinds of capabilities into how you make decisions.Are you doing anything with cognitive apps and AI? Where do you stand with that? Yes, we are. We're applying machine-learning technologies in a number of different areas. One of the areas we're focused on as a financial institution is cyber-security, not surprisingly. This is a place that is a great opportunity for the application of machine learning. It starts with building a comprehensive data infrastructure on what's going on in your environment in order to be able to build tools and have insights into things that might be anomalies. And then you can use machine learning as a way to discriminate normal behavior from abnormal behavior. So we've applied machine learning in that context. It helps us identify malware or anomalous behavior or other kinds of things that are indicators of threats in our environment. That's a really well-suited domain for leveraging machine learning. There are other places around fraud and risk decisions, the ability to apply it in interactions with customers where you move from one-off interactions with customers to real-time, dynamic interactions with customers. For example, we launched a product called Second Look. It is a feature for card customers where if you're in a restaurant and you leave an unusually large tip, we can send you a notification in real time and say we noticed you left a large tip; did you intend to do that? And that's kind of useful if you do it 24 hours after the fact, after processing that over night. It's incredibly useful if you're able to do that in the moment while you're still sitting there in the restaurant. And, similarly, if you double-swipe, we'll be able to pick that up in real time and let you know that this transaction was just double-swiped, was that intentional or not? Do you address this at the developer level? Yes, we typically leverage open-source machine-learning frameworks. We build those into our apps. We hesitate to use the black-box approach. We feel like it's important for us to have the machine-learning skill sets in our organization. So we have a particular focus on recruiting and developing those skill sets internally. What we're seeing come together on this is this whole world of open-source data capability—starting with Hadoop and the whole ecosystem around that. Think about the move to streaming data, and then stream processing tools like Spark and Apex and Flink and those that allow you to process streams of data in real-time, those are the frameworks where we're building in more machine-learning intelligence. So, for us, it is really coming out in the world of the data ecosystem and those sets of tools. That's where we're seeing the real potential to apply machine-learning capabilities as opposed to just software engineering. The data world is where it's really powerful. You don't sound like a bank. You sound like a born-on-the-cloud startup. We feel like that's how we have to operate. It starts with our CEO and our business leaders really believing that for us to win as a company in this industry where it's going, we have to operate like a technology company. It is a sincerely deeply held belief; it is not a lip-service thing. And if you start with that, you are compelled to figure out how the best technology companies are operating and you turn yourself into that. And you won't be able to get the best talent out there to come and work for you unless you're doing these things.
These are areas where banks are going to really struggle. They're wrapped up in their batch processes today. First of all, they have to move to real-time streaming data and be able to deal with that. And then the notion of applying more sophisticated modeling and machine-learning and artificial-intelligence tools on top of that streaming data so you can make better decisions and more interactive decisions with customers. That is a really hard transition for financial institutions to make.