AI Leadership Insights: Overcoming AI Data Challenges with Terren Peterson

Hello, and welcome to the latest edition of the Techstrong AI video series. I'm your host, Mike Bazar. Today we're with Terren Peterson, who's vice president of Data Engineering for Capital One.

And Dave, have a report out that kind of looks at how ready enterprises are for AI in general. Taran, welcome to the show. Hey, thanks for having me.

Walk us through some of the highlights of the report, but given your background, I'm always keen to know what surprised you in here. Yeah. Uh, so, so the report had some really interesting findings.

Uh, you know, some of it, uh, uh, ended up being around the importance of, of ai, uh, as, as it, uh, pertains to the larger business strategy of organizations. So I think that business leaders, uh, largely I think that the numbers were in the, the 80 percentile, uh, or, or in that range, that, that recognize how important AI and ml, uh, are to their broader business strategy. Uh, there's also some, uh, findings, uh, interviewed by tech practitioners.

Uh, and, and, and a lot of those findings were around the readiness of the data platforms that are gonna needed, uh, be needed to drive those, uh, AI and ML strategies. Uh, and, and there's some interesting data points that we're talking about, Hey, you know, what, what's the day in the life of a tech practitioner? Uh, and in particularly ones that are, uh, responsible for the data ecosystem?

And it highlighted that, that there's a lot of challenges that, that the practitioners have, uh, around data quality issues. And how much time is it that they're really spending, uh, wrestling with, uh, the data pipelines sort of addressing to the metadata management, some of the other data quality needs. Uh, that, that, you know, if you think about it, is that really where we want our individual's time being, uh, spent?

Or is it that we want the, uh, you know, the tech practitioners really to be spending time, you know, gaining insights on the customers, looking for new revenue opportunities, you know, figuring out, uh, how to go ahead and, and, and serve customers. Uh, and, and the, you know, the, what the, the survey highlighted, uh, was the, you know, the reality is that a lot of the data practitioners, uh, are spending a lot of time, uh, uh, with that, that that lovely term that we use sometimes called data wrangling. Does it surprise you that the struggle with data is so high?

Because I feel like we've been talking about data management for decades, and yet a lot of organizations maybe didn't quite give it the full attention it deserved, and now it's all coming home to roost in the age of ai. Yeah, I think that that's part of it. I think that that, you know, you and I have probably seen charts over the past, you know, three to five years showing, you know, this almost hockey stick explosion of data that most organizations are getting.

And, and, and I do think that, that there's a dynamic here that, you know, maybe we've been talking about it for a long time, uh, but, you know, with the, the explosion of data and all the new, you know, no matter what industry we look at, right? The, that, whether it's iot, whether it's about, you know, clickstream data there, there's new data sources coming in, uh, from all sorts of the enterprise. And I think that there, there's sort of a velocity question, you know, to me that, that jumps out that says, Hey, yes, you are, you are gaining more and more sources of data, you're getting more volume of data, you're getting more insight into the data.

Uh, but how much have you invested in sort of automation and, you know, strong data pipelines that can make sure that when it comes time to get the value out of the data, uh, that that, that the time can really be spent, uh, by the analysts and scientists really on, uh, gaining insight versus, you know, sort of going back and fixing things in the data pipeline or adding data quality checks. You know, it's, it's that, you know, high value activity, uh, that that, that we want our, our associates to be putting their time to. But I think that that what a lot of organizations, at least what I read into the study, are probably struggling with sort of that data velocity that they know it's important.

Uh, but, but hey, is it that they're, you know, have, have all the tools that they put in place platforms, are they keeping up with a scale at which data is growing at, uh, and, and when they don't keep up with that scale, you know, it's the time when they're, they're trying to realize value, which is when they, they spend sort of the catch up time. One of our organizations doing, well, at least the folks who get it, in terms of managing data. Are there certain best practices or principles that you've seen organizations start to embrace?

Yeah, you know, I can speak to what, what what we're doing at Capital One and where, where we're seeing a lot of value. You know, I think that, that, uh, one of the, the big pieces is in getting organized around your data, right? I think that the data ontology is as important as it's ever been.

You know, a lot of that gets into are you taking the time to model the data? You know, capital One in invests in, in dedicated resources for data product managers, and it's really the job of those data product managers to organize the data, to curate, to take a look at the data quality. Hey, are, are, are, are all the metadata fields that we need, you know, how, uh, how rich are the, the metadata, uh, around the different data sets and tables that we have?

And, and by, by staffing that, uh, you know, those data product manager roles, what that allows us to do is that by the time that somebody comes around to try to, you know, do some data consumption, right? And to try to go ahead and, uh, and come up with new models, uh, and come up with things that, that, that can gain insight by consuming the data, you know, the, the prep work has already been done, right? And that way when, when it comes time for that big AI project to, to come down the pike, you know, the, the, the, the data is actually ready and that the, the volume of time that the, the data scientists or the data analysts are gonna spend are really on the high value, uh, uh, uh, tasks and not on sort of playing catch up on a bunch of data quality rules, uh, that, that, that the data producers probably should have had in place.

You know, when the, the data streams actually were created, I feel like everybody walked around and said, data's the new oil, but then discovered we didn't have any refineries to make it into something usable. Do we? That's a great analogy that I, I like that.

So do we need to create a more of a platform mindset towards how we manage data and ultimately then expose that out to people who wanna use it to customize AI models or do whatever they need to do with it as an extension for, I don't know, retrieval, augmented generation, whatever they're doing? Yeah. Uh, oh.

Oh, I, I, I like the term that you use, you know, platform really resonates with me, uh, based on my experience at Capital One, because what it does is that it really highlights that like, look, this is a, this is a part of your organization that's gonna live on beyond any individual project. And, you know, and by investing in those, those data platforms and having a data, uh, platform, uh, culture, what that really does is that it paves the way for those projects that, you know, are gonna come down, uh, come down in, you know, in the 2025 planning. You know, you don't exactly know precisely at this point in the year that, that, that what are all the different ways you're gonna leverage your data, but by investing in the data platforms, you know, that, that that investment really allows you to be agile down the road.

So I agree, like data platforms is a, is a huge part of the strategy. Every organization is different, but I kind of feel like we went from irrational AI exuberance in 2023 to 2024, we're gonna experiment with, uh, all kinds of different things. And now going into 2025 are, are we trying to pick a few projects that are gonna be winners because, well, we can't do everything at once and AI's expensive.

Yeah. Oh, yeah. Yeah.

I, I do think that that, you know, if we're to try to, you know, chart the, uh, uh, the hype cycle, I, I, I do think that that there, there is that economic reality that every organization, uh, i i, is going to start to face. And, and that, that while there, there's some, uh, amazing results that, that different businesses are starting to show that you can get from AI and ml, uh, it, it is going to, to apply more and more pressure of that, hey, if you're gonna, uh, either, you know, work with a, a partner that has A-A-G-P-U farm and a cluster that's already built that you can consume, uh, for, for your project or, or whether or not you're gonna try to do something in house, like the, the expense is definitely there. And, and yes, it's been a, it's been a front page item for a couple of years now.

Uh, and, and I do think that there's, there's sort of a mounting pressure that says, Hey, where is it that, that you're, you know, whether it's you're, you're doing something internally for your organization around productivity improvements or it's something of where, hey, how is it that you can, you know, realize value for your customers? Uh, I, I do think that that, that, you know, 2025 a a key theme is going to be, uh, sort of, uh, show the, the return on investment. And, and I do think that there's a lot of, you know, there's a lot of, uh, projects that I see internally at Capital One where we're realizing a lot of value in it.

It's just that the, that there's gonna be more and more of that, that focus conversation. I think as we, as we head into, uh, 2025, The other thing that's become more apparent is that, um, historically if we manage data, it was very much a batching any kind of mindset. It seems like with ai, we absolutely have to figure out how to get the right data to the right place at the right time, because it's all much more event driven.

Yeah. And, and, and, and it's this sort of the rise of the real time enterprise, right? Of the, the, you know, the, the, sometimes it's, it's in the moment of the, you know, if you're trying to come up with a great offer for a customer with your, your mobile app, right?

It's, it's, it's, you want to, you wanna be able to get the, that information to the customer and that offer out to the customer, you know, at the time that matters, right? And not the, the fact of, hey, you know, three days later, go ahead and get that, uh, that, that email that, that came out from a batch process versus the, Hey, you know, as long as you're, you're doing the search or, or you're doing the chat on the, uh, the, the mobile app, you know, be able to, to, to do that in real time. And, and, and I agree, I think that, that the, the leaders in, in this are gonna be ones that can, uh, realize an, a event-driven architecture are ones that really understand the importance of, of real-time streaming, uh, and, and are ones that are starting to, uh, uh, migrate away from sort of the traditional batch monolith architectures.

And the other subtlety that comes to mind is that, um, a lot of the AI models are multimodal, and we're combining text and video and audio, and these are all different data types. So, um, or do we need to kind of manage that differently than we have in the past? 'cause I feel like, you know, historically, if we had a different data type, we put in a different silo and managed it in isolation, and now it feels like all these things need to come together.

Yeah. I, I think that a, a pressure that, that the data platforms as, as you highlighted before, a pressure that the data platforms are gonna come under is a how ready are they for unstructured data? Uh, you know, because if, if, if we think in terms of like, Hey, how do you make things real time and intelligent?

You know, it could be that the ways that that we start to interact with customers, uh, are gonna be visual. They're gonna be audio, they're gonna be things that, that, that it's a, a document that somebody scanned in, uh, and, and things that are, uh, that are in the unstructured world. Uh, but, but at the same time, there's gonna be the need to still do, do all the delivery, uh, in real time.

So, uh, uh, will be something that when data architects think of the landscape of, of how they, they build their platforms to support all the, the workloads, it's gonna be a matter of saying, Hey, handle both the structured and the unstructured. And as you say, it can't just end up being different silos, uh, because those silos that, that, that how you're managing the data, it's probably gonna lead to silos in other parts, including the, the user experience. One other thing that I hear from folks that, uh, maybe they kind of knew about but didn't fully appreciate is this whole art of metadata.

Do you think folks are kind of taking a minute to kind of think through, because if I get the metadata right, then the rest of it becomes, I don't wanna say simpler, but good things will follow. Oh, yeah, yeah. Metadata and, and, and just in general, without metadata, it's hard to have great context around decision making, right?

So whether we're talking about an automated process or, or a business process that that's human driven, right? Without, without context, uh, it, it's really hard to make good decisions. And, and I think organizations are gonna find out that, that if they haven't invested in metadata, you know, the, their, their models are gonna lack context and they're gonna, uh, lack a lot of the richness, uh, of insight that, that you're gonna want that model to be able to, to interpret.

So based on what you're seeing in the report, what's your best advice to folks? What's the kind of the one thing that leaps at you that says, Hey, we gotta get our proverbial acting gear around this? Yeah.

You, you know, uh, I think that, that it's easy to look at a report like this and the, the, to feel like it's daunting. You know, there's a, there's an adage that, that I've always appreciated of that, hey, the best time to plant a tree was 20 years ago, right? But, but, but, uh, the second best time to plant a tree is today.

And so I think that a, a takeaway when, when somebody can look at a daunting task of saying, well, gosh, you know, maybe is, uh, companies are starting to look at their 20, 25 plans and, and, you know, hear from the tech practitioners that, like, there, there's some data, data work that has to happen, right? You know, you mentioned metadata. It could be that, that a lot of the, the first step that's gonna need to happen is enriching the metadata that, you know, it could have been collected by the data producers over, you know, the 10 years that they built the lake.

Uh, but, but it, you know, if, if they haven't done it, you know, the, there is a pretty good, uh, roadmap, uh, for how you go ahead and build out, uh, data architecture. So, you know, go ahead and, you know, use, use the new year as the kickoff to go ahead and kick off the, you know, the metadata enrichment projects, you know, the, the, you know, changing the, the data culture. You know, we saw in the, the strategy talking so much of, of, of how important data culture is.

And, and I think that that, that that's something of where, hey, you know, if you didn't do it 20 years ago, you can do it today. And, and, uh, you know, what a, what a great time of the year to go ahead and, and, and kick off new initiatives of, of doing some of these things that, that we see are, are critical for, for good data culture. I think it folks have always appreciated the data challenge, but do you think the business side partly is understanding what's really the scope of this day?

Yeah, I think that that, that the experienced, uh, uh, uh, business professionals probably have, uh, because I think that as they've, as they've invested in technology and engineering initiatives, I think that they've probably seen, you know, what are the barriers or what are the hindrances that have come up? And I think that that more and more of the investment strategy, if, if, if they've invested in AI or ml, I'm sure that they've heard about, uh, how important data is and seen that that's so much of where the leverage is. So, so I do think that that's something that, and the survey did highlight that there's the understanding at the strategic level, uh, how important it is.

I think it's sort of that connecting the dots to execution, that, that some, you know, that there's still some work to do, Right? Folks, you heard it here, even in the age of ai, garbage in, still garbage out. So it all starts with the day.

Hey, Taryn, thanks for being on the show. Thanks, Mike. Take care.

Thank you all for watching the latest episode of the Textron AI video series. You can catch this episode and others on our website. Until then, we'll see you next time.