Credit Unions Battle for the Right Data to Fight Fraud

This is What's Next in Payments, a PYMNTS podcast. Forward-looking insights from industry leaders on the trends and technologies reshaping payments and fintech. With 4,000 credit union partners under its belt, Velera knows its audience. Jeremiah Lotz, SVP, Enterprise Data and Experience Design at Velera, tells PYMNTS that a consortium approach to data collection and use can help credit unions fill the gaps in knowledge caused by pressures on government reporting agencies.

Hi, everyone. It's Hal Levy at Payments.com once again and once again for our usual conversation with a twist. We've got Jeremiah Lotz, who is Senior Vice President, Enterprise Data, Experience Designer at Velero. Welcome back. It's nice to have you. Hello. Thanks for having me. So today we're discussing a timely subject, which is that of data. Now, that's always timely. But as we all know, it's been in the news. Data is used, is misused possibly. The collection of it and the dissemination of it, well, that's under pressure and under scrutiny. Depending on where you look, you've got agencies dismantling, defunding. Sometimes there are statistical programs that are being tweaked. Historical theories don't capture what's really going on in real time in our economy. All of that presents a challenge for AI, which promises to have insight while you've got fresh doubts about bias, about provenance, about all manner of the ways in which AI can be used here. Now that's a mouthful, but basically the topic today is data, what's up? And we're going to discuss that a little bit now. And, um, The first question obviously is a high-level one. When you are looking at and using the core government statistics, and those statistics go dark, or they're stale, or they have to be reconsidered, what are the alternative sources that must be used and should be used to anchor risk, credit, and growth? How do you vet them?

Yeah, absolutely. That's a good question. As you mentioned, I mean, data is critical to nearly every decision a financial institution is making and access to that data and reliance on the accuracy of that data is super important. So seeing the shrinkage of this access to this data or this historical information can be problematic for financial institutions and specifically credit unions and community institutions, they're really disproportionately impacted by this because they rely on these data sets and these market insights. I think one of the advantages of being a financial institution in an environment where there is cooperation with organizations like CUSOs or fintechs who are able to assist is really how do you leverage consortium data to make up for some of this lack of other statistical information. So if you think about the access to first party transactional data that an FI uses or for Velera, we have access to transactional data on behalf of our financial institutions. And so with that, that gives us insight into billions of transactions. And then, of course, we have, as do institutions themselves, have access to bring in other information, maybe it's commercial data, consortium data, base data sets, to then start to augment that information. And of course, those other third-party data sets that you bring in, you've got to be careful. You've got to vet those for accuracy and bias and that sort of thing. But if in reality, if you can combine those two things together and specifically for a service organization, servicing multiple financial institutions, we have the advantage of being able to do that. It allows us when we couple that with our overall governance that we focus on, it allows us to really kind of make up for where we might be missing some of those statistical opportunities and really even gives us the opportunity to create some proprietary indexes that we might focus on to show trends or do forecasting with the financial institutions in our space. So certainly there is a miss in data opportunities as these government agencies are shutting down. But really, those of us who are able to bring in first party data and commercial data and make sure it's properly vetted and governed, we have the ability to kind of make up for that with our financial institutions. and give reliable signals from what's happening within our own networks and scale and that sort of thing to continue those trend capabilities for them.

When we think about the role of historical data, obviously, it has a function, but it no longer perhaps is, I don't want to say useful, but maybe it has to be reconsidered. I want to talk about historical data's role in all of it, but I also want to get a sense from you of what I might call additional data. We're talking here about geolocation. We're talking about behavioral data, data that had not been previously available. So in that context, how can those newer sources be used to revalue historical?

Yeah. So I think that's important and using things like geographical data and other trends. So real-time data is certainly going to win the moment in many cases, but historical data is going to win the strategy and you need both of them to compete. And so when you think about about true consumer type of impact and you think about fraud management. We want to use real-time data and real-time transactions to be able to understand what's happening in that moment for that consumer. But we also want to look back at historical or retro data and really the outcome of certain models where we've used that data to say when we see transactions occur like this or in places like this or in these patterns, here are trends. And that intelligent data then becomes something that we can say, okay, now when I look at this real-time data, what does that model or that intelligent data tell me I should do right now? And so in that case, both the real-time and the retro data is important. And really, even when you think about making strategic decisions as an organization, using that background information, coupling it with things like you mentioned, like geographical data or other trends. And really, again, it goes back to using historical data, running it through, smart, well-managed models to then give you something intelligent. And then you can start to look at, so what's happening in my real-time environment and what's the decision I need to make for the future? So what trends am I seeing? What real-time capabilities or transactions am I seeing? How does that correlate to previous trends? And then how does that predict future trends or future expectations? As a business leader, then I can use that information to really make strong business decisions And so you can see it happening and how it might change the consumer interaction at that moment in a fraud scenario or maybe helping them get the right information when it turns to the next product or solution for them. But then also, how does a business leader make a good decision in the future based on using both real-time and retro type of data? So really both have a place. It's a matter of using them both together intelligently, really. Yeah.

With using AI in the service of preventing fraud, there are a couple of things that pop up as always. One is how effective or efficient it has been in terms of spotting fraud. And maybe also we'll talk about credit underwriting too. So fine tuning as you go. And then on the other hand, you've got to be able to be reliant on the modeling so that It is not just good for the company, good for the FI, but good for the would-be borrower or customer. You've got to also be able to be confident in the reliability and the fact that you're in compliance. How do you manage to strike the balancing act that comes with both of

those? Yeah. So when it comes to reliability of data and reliability of AI models and all that, a lot of this points back to governance That's not a term or a word that people really love to talk about or plan about, but really, you can take data and AI and you can make it smart and you can make it do great things, but it can't just be smart. It literally has to be accountable for what it just did. How do you trace it back to what it did? Was it an ethical use of that data, an ethical use of a model? Really, that all boils down to putting governance in place to manage these frameworks. It's not that every single model requires a level of scrutiny that is a barrier to ultimately getting that to life because it's moving so quickly that if we have that type of issue in front of it where it takes a long time to get a model in place, then you're going to be behind in the game, right? But how ultimately do you have a framework in which you can trust, one, the data that you're building the AI off of, and then two, how are you building models in a that you know that they're ethical and that they're looking for fraud anomalies and that you are monitoring them for, you know, different biases and that sort of thing. And so when you are putting AI in place and have model management in place and model creation in place, you know, one of the underlying pins has to be, you know, what's a governing framework for those models and really how do you track and monitor them so that you can then insert, if necessary, a human into the process to then have the adjustment to make sure that if there's a tweet to the model that needs to occur or something in that regard that we're taking that action and keeping it safe. But there's no question that the reliability of AI with good data is absolutely strong. You just have to be able to hold it accountable and ultimately be able to audit the AI as well, just as you would human processes. And that's part of taking on the challenge of building and leveraging AI into the things that we do today.

Yeah.

A lot of what we've been talking about has been the melding of first-party and then third-party data, right? The idea of really getting a broad spectrum of information. Tell me a little bit about the ways in which specifically Velera is there to aid the construction of these models. I'd like to get a sense of how you might work to scrutinize third-party data and your own input. And then by extension, how you can help the FBI start to work toward their own internal processes to be improved and use AI.

Yeah. So, you know, I was talking about governance earlier, and that's one of the areas that is least attractive when we think about how to use data and how to use AI for all the amazing use cases that come into play. But it really is probably one of the most critical aspects of it is how do you put strong governance in place? And it doesn't have to be over-engineered and over-managed But I would say a simple practice, and of course, there has to be details behind this, but if you can't explain where it came from or how it's used or why it matters ultimately to the member, then you shouldn't use it. And so when you think about how we're going to bring data in, whether it's first-party data that we own or have direct access to in a transaction or bringing data in externally that we want to augment and use for very powerful predictive modeling that we see for future operations, opportunities or for real-time fraud or for AI to enhance an experience, you really have to be able to tie that back to where did that data come from? Can you tie it back from a lineage perspective? And also, how are you using it? Are you using anomalized data? Are you using real data or anonymized data? Or are you using data that maybe has sources that you're not super familiar with? And that's not something that you should be doing. And you don't want to mix data from, you know, financial institutions in a way that is, you know, going to jeopardize a consumer or jeopardize the financial institutions. So anonymizing that data and really ultimately being able to trace it back to the source of where it came from. And if you can't do that, then ultimately you shouldn't be using it as the kind of the punchline. But the reality is the things that you can do by bringing in third party data, especially in today's environment, when we have technology that's grown the way that it has, when we have data providers and we have modelers out there, that it's not always the raw data you need. It's really kind of the insight. What is an insight that someone externally has been able to create with a set of data that then if I take that insight and use it to power my model or to power other things, then my insights just became stronger because I was able to couple them together. And you can do that in a safe way where you're not necessarily having to bring in I jeopardize your environments with unprotected data or raw data. And that's important. But always being able to match it back to where did it come from? What is the use for is critical to that.

When we think about the future of payments, obviously, we've got open banking. We've got real-time payments. We've also got, as you've been mentioning, a consortium approach that really works here. What happens with that? What is the future? What is the immediacy? What is the consortium approach right now? What is it going to evolve into as we start to get ever faster with payments? Fraud detection and prevention comes. ever more urgent. And data, as we've been discussing, is coming in waves. So what is the consortium approach now? What will it be?

Yeah. Well, I would say my probably biggest piece of advice is not to be afraid of consortium data. Because if you consider the use cases that ultimately it can power, it's strong. And consumers know that we have their data. They know that we have access to their data. And we know that their data, we as a And so there's a general understanding that why can't these things talk to each other and make my experience better or make my experience more secure? And I think that's an important aspect of it is if we back up and we think about what the consumer is looking to experience, then there's really an opportunity and expectation to say, well, how do I let data talk to each other in ways that help ultimately the end consumer? And maybe it's not, you know, you don't always have to give up your most competitive assets. But how do you ultimately help a consumer have experiences across their digital life that they're looking to have? And I think active participation, as you mentioned, today is where we see active participation in fraud consortium is one of the strongest places. Again, we service over 4,000 financial institutions. And using that data helps us to protect hundreds of millions of dollars in potential fraud. And we might be helping a member Yeah, absolutely. How to do that is getting back to that securing data in a framework and in a technical framework that allows us to move data back and forth and allows us to protect what we would do with that consortium data and what information is in there. Again, I go back to trends versus true, all raw data capability in there. But I think the underlying message here is collaboration is essential to how to establish the best member experiences, both around protection as well as around other more positive opportunities. But creating trust and competitive protections are necessary. And that's kind of where the barriers are, I would say, mostly today. So looking at today's technology, looking at capabilities and things that we can do to share information in a consortium way that doesn't necessarily allow us or require us to expose unnecessary details behind it is where it's going to be critical. But again, it works. Shared data works when it's secure, when you can measure it, when you can track it back, and when you can provide some type of positive impact to everyone involved. And I think consumers expect us to use it in that way.

I love that. Shared data works. And I think that's a good tagline. And we'll leave it here for now. Thank you very much.

Thank you.

That's it for this episode of the PYMNTS podcast, the thinking behind the doing. Conversations with the leaders, transforming payments, commerce, and the digital economy. Be sure to follow us on Spotify and Apple podcasts. You can also catch every episode at PYMNTS.com/podcasts. Thanks for listening.