Your AI Looks Smart. Your Data Disagrees.

AI can generate answers instantly. But that doesn’t mean they’re right.

Most enterprise AI doesn’t fail because of the model. It fails because the data underneath it doesn’t agree. Different teams, different definitions, different outcomes. It’s subtle, and it breaks trust fast.

In this episode, Paul Bruffett, VP of Data and Analytics at Jack in the Box, joins WWT’s Dan Moristro to talk about what it actually took to fix that. A multi-year modernization across core systems set the stage, but the real shift came from treating data as a product and building consistency into how the business defines and uses it.

They get into what Dan calls semantic debt, why generative AI makes it harder to ignore, and how modular data products, a modern data stack, and a real MLOps foundation helped turn AI from something interesting into something reliable.

If your AI works in demos but not in the business, this is probably why.

Learn more about this week's guests:

Paul Bruffett is a data and analytics leader with deep experience across cloud platforms, data engineering, and data science. He has designed and implemented enterprise data lakes, big data platforms, and scalable architectures on AWS, Azure, and GCP. Known for a hands-on approach, he focuses on building systems that support continuous deployment of data and AI workloads at scale.

Paul's top pick: Jack in the Box CTO Reveals AI Playbook

Dan Morrisroe is Managing Director of AI, Data, Analytics, and Management Consulting at World Wide Technology. He works with organizations to turn data and AI strategy into real operating capability, helping align technology, teams, and processes to drive measurable business outcomes at scale.

Dan's top pick: From Static Scorecards to Real-Time Retail Intelligence

The AI Proving Ground Podcast leverages the deep AI technical and business expertise from within World Wide Technology's one-of-a-kind AI Proving Ground, which provides unrivaled access to the world's leading AI technologies. This unique lab environment accelerates your ability to learn about, test, train and implement AI solutions. 

Learn more about WWT's AI Proving Ground.

The AI Proving Ground is a composable lab environment that features the latest high-performance infrastructure and reference architectures from the world's leading AI companies, such as NVIDIA, Cisco, Dell, F5, AMD, Intel and others.

Developed within our Advanced Technology Center (ATC), this one-of-a-kind lab environment empowers IT teams to evaluate and test AI infrastructure, software and solutions for efficacy, scalability and flexibility — all under one roof. The AI Proving Ground provides visibility into data flows across the entire development pipeline, enabling more informed decision-making while safeguarding production environments. 

Transcript

AI Falls Apart When Definitions Don’t Match

SPEAKER_01 0:00

The entire promise of AI is simple: faster answers, smarter recommendations, more autonomous decisions. But before AI can recommend, predict, or act, your organization has to agree on what the data means and how decisions get made from that. At Jack in the Box, the fast food chain we all know and love, that can mean knowing which menu item to surface to the right customer, how stores are performing, which promotions are working, or how to help operators act on issues faster. But the potential of AI quickly runs into an older operational problem before it can be realized. The data has to line up, the metrics have to mean the same thing across teams, and the business has to trust that answer. That's true in quick service restaurants like Jack in the Box, and it's just as true in healthcare, manufacturing, retail, or financial services. So on today's episode of the AI Improving Ground podcast, we're talking with Jack in the Box VP of Data and Analytics, Paul Bruffitt, and WWT Managing Director of AI and Data Analytics, Dan Moristro, about how you can make faster decisions and automate more work when the underlying data still has to be reconciled across systems, teams, and definitions. It's a foundational question facing nearly every industry now trying to turn AI from a demo into an operating capability. So let's jump in. Paul, Dan, welcome to the AI App Proven Ground Podcast. How are the two of you today?

SPEAKER_00 1:25

I'm not too bad. Thanks for having me.

What It Actually Took to Rebuild the Data Foundation

SPEAKER_01 1:27

Yeah, doing well. Happy to be on. Excellent. Excited for this conversation. Paul, I'm going to start with you. So Jack in the Box has been on a years-long transformation journey. You know, this is this has been underway, at least as I understand it, well before, you know, AI really exploded onto the scene. So, Paul, just catch us up on what that transformation has been and what the goals were. And maybe when did AI start to kind of enter that equation?

SPEAKER_00 1:51

Yeah, so our business, as you noted, embarked on a digital transformation by doing everything, replacing our ERP, point of sale, back of house systems, relaunching our mobile app, and as I came in to do, modernizing and updating our data and analytical systems. And so AI really runs across all those, right? A number of vendors are increasingly adding more AI services to those offerings that we've brought in, especially the best of breed offerings, which we've really focused on as we've gone through that modernization. But in the data and analytics space, we really saw an opportunity to build proprietary value-added systems to uplevel our guest experience and some of the other lines of our business that we can talk about.

SPEAKER_01 2:36

Yeah, and maybe talk to me about how that either set you up in a good spot to start to capitalize on some of this, you know, AI hype that crept in a few years ago. What role did, you know, was like, did that help you get your data state in order? Were you still doing that work? Maybe a little bit of that.

SPEAKER_00 2:51

The good thing, you know, the exciting thing about modernizing all those systems is we got to work with much more, I mean, modern, right? Much more leading edge systems that were API first, that could do event-based integrations and had really rich information about their domain, be it the transactions from point of sale, be it supply chain and our ERP. But the challenge was as we replace all those systems, the data team obviously had to do a lot of work just to re-plumb everything and get us back to a good state, right? And so the reality is it opened a lot of opportunities for us, but it also created, frankly, a lot of work that the business doesn't really see as valuable, right? They already had reporting from the point of sale. We had to do a lot of work just to get back to where we were. Now we've done a lot of that and we've modernized those systems, and it's really allowed us to unlock, it's really allowed us to unlock new data products, analytical products, and algorithms. And the way that we've used that first is in our digital channels. So one of their key drivers or the impetus for some of this transformation was really our business, quick service restaurants, digital is one of the only, it's really the primary growth driver, right? If you look at a lot of our competitors and ourselves included, digital is really where the growth is for transactions for customers, and it has the value added of allowing us to know our customers better, which is really how we leveraged AI in our proprietary value-added algorithms, is to say, hey, look, now we understand our customers much better because of all this information that we're getting from these new systems, from this new digital channel that we invested heavily in. How can we tailor the experience better to them, both to make them want to come back, both to make them enjoy using that and hopefully drive some incremental sales and get some additional value because Jack in the Box, relatively uniquely in the space, has a huge menu, right? And we're constantly adding new products. We do breakfast all day, right? People have a lot of choice. Sometimes it's really hard to get the right things in front of the right folks. And that's one of the key use cases for us for AI and our digital channels and our journey.

How We Finally Proved AI Wasn’t Just a Demo

SPEAKER_02 4:50

Yeah, Jack in the Box really did an incredible job setting that foundation, everything Paul's talking about, the modernization, getting the data AI ready. Like we love to partner with companies like Paul and Doug and others who set that foundation so that you can go and start iterating quickly and really building value. You know, I think people had been thinking, oh, it's kind of the latest in AI, you don't need to get the data ready. Let's just build, let's build AI on top of raw, messy applications and just kind of pointed at things. And the reality is that it doesn't work well if you do that, right? And you need to have the foresight to kind of build the foundation. The other thing, though, that I'm seeing at Jack in the Box, but some of our other customers as well that are successful is building incrementally on the foundation, right? That still is more important than ever. So you're not building the whole foundation, right? I think Paul, you'll talk a little about it, call it on the kind of the data product side of things, right? But kind of building out a component, a full slice of the foundation that enables some AI and really kind of proves value to the business while you continue to build out. Um so Paul, I don't know, anything you'd add on how you're kind of slicing up that data foundation?

SPEAKER_00 5:52

You you absolutely touched on it, especially with the incremental value, right? Our business, pretty much no business, has appetite to just continue investing in transformation really for its own sake, right? That's that's largely run its course, and we don't really have the luxury of that. And so we had to demonstrate that value incrementally, right? And so the work that we did together, as you well know, is build that first version of the customer data product, really, because previously at Jackman Box, we didn't have a great view of the customer in our analytics environment. We had it in some of our marketing and digital tools, and those were great, especially tools like amplitude, which our business used to understand the funnel and some of our A-B testing, but it limited the kinds of deep analytics they could do for like more like longitudinal studies of customers and how behaviors are changing. But we couldn't build everything in one go, right? So we built a key portion of it together, and we built the ML ops required to build the algorithm, and we built the algorithm. And we just be built kind of MVPs of all those and sort of got the right to continue investing there after demonstrating that value, right, by driving some additional average order value, by driving some additional margin in the transactions. And then we came back and continue to enhance it. And I expect we'll continue to enhance it more or less indefinitely, right? Because the business is never done asking questions and getting value out of that information. But we definitely have to continue demonstrating that's good. We've been good stewards of that, of that, of the investment, frankly.

The Stack That Took AI From Idea to Reality

SPEAKER_01 7:15

Yeah, Paul, I mean, so uh what you're talking about here just it feels like a pretty forward strategy in terms of where I see a lot of other organizations um are at in terms of their data strategy. Maybe a a slight backup. What role did modernize infrastructure have in enabling much of what you're talking about? Was that was that a nice to have, or was it more of an integral piece of the puzzle here?

SPEAKER_00 7:38

Yeah, you could think about it in two different ways, right? And so I talked about that ecosystem of technologies around us, the point of sale, the back of house, the ERP, all of that stuff getting modernized. And they definitely unlocked a lot of capability for us to get additional information that really is going to continue driving new intelligence. And so I would say that was key, right? Uh, we're always, again, going on that journey. There are things that we're going to continue to modernize over the next several years. On the data side, we moved from a relatively static, what I would consider to be legacy stack of warehousing and really struggled to add new pipelines and definitely didn't have any ML infrastructure. And we went to the modern data stack, the Snowflake, DBT, Airflow, and used Amazon services with SageMaker for ML ops together. And I think that was really critical because both it allowed us to take on new volumes of data. Customer, the data, the telemetry we get from the mobile app, is quite a lot larger and much more voluminous than some of our other, most of our other data sets, to be honest. And so we definitely needed this the scale, right, of those technologies as well as the cost profile of them to take it on. But really, we didn't have an ML ops like lifecycle ourselves. We didn't have an ML ops tool chain. And so we uh we really needed those tools to unlock that. And the other piece of it I'll say is we also thought about the data differently. Dan alluded to data products. Previously, we'd really built a lot of our data to drive specific reports, which was fine for the time, but it limited our agility as new requests came in because we really had to go and tinker with everything, even if it wasn't necessarily connected to the data request at hand. If it was supply chain data they wanted to bring in, we still had to go regression test our point of sale data because they were intertwined. And so this modernization also allowed us to modernize how we thought about data as a product, as a construct, independent of the technologies.

SPEAKER_02 9:28

Yeah, right, Paul. And I think that agility is only increasing, right? As more, you know, we start looking more forward thinking on data agents and more AI, kind of looking at the data and kind of, you know, the business wanting to ask infinite questions, then even say, can you help me monitor the data, right? That that agility Paul is talking about only expands, right? So you kind of need this product mindset on top of the data to prepare it, right? You can't do infinite preparation right without ROI, but you also need to do it in a way that's going to enable a series of questions, right? And this whole new mechanism of consuming that information, right? And so it's this kind of paradigm that, again, I think, you know, Paul and the Jack in the Box team is doing a great job trying to get ahead of.

SPEAKER_01 10:09

Yeah, Paul, I mean, you were alluding to it a little bit in your last answer, but what type of operational shift or mindset shift did you and your team have to make to take on that identity of you know data as a product? Data not only as a reporting artifact, but something that you could act on and and use more as an operational asset? Like, was that a big change that you had to to manage, or were teams easy to to identify with that?

SPEAKER_00 10:34

I think to some degree it varied, right? Customer, I think, was fairly straightforward because that was a completely new domain that we got to build together with Dan and his team from the ground up. I think for some of the domains that we had a lot more experience with, especially for transaction, especially for point of sale, we really had to think about them differently. And that took a little bit more time. And what I mean by that is we have a product ownership org, we have product owners who help us and our business users design and think about data. And we really had to collectively take a step back, both my team, the data team, the product owners, and the business team take a step back and think a little bit more deeply about what the data means, how we want to use it, where we want to go, the things we want to accomplish, as opposed to what a lot of folks, myself included, sometimes, are inclined to do is just say, hey, I just need this other field, right? Not necessarily about the why, not necessarily about the business outcomes, kind of in an aligned way, but just to say, hey, I got this field, I got this thing I'm trying to do, let's get it in there. And that really limits the way that we can that limits the reusability. And so by kind of taking a step back, and our business users were actually extremely excited about it. Some of our business users really thought all the way back about like, hey, how do we want to drive operational efficiencies? How do we want to run restaurants with data, right? What does it mean? What are the key success metrics, right? For example, they want to spend less time in the dashboards. That's great. I want them to spend less time in the dashboards, but get still get the same insights, if not more, right? And so some of those were the things that came out of them taking the step back because they stepped all the way back to what data means for operations, right? And for customer and for all these different domains. So I think that was that was great, but we definitely had to go on a journey. And frankly, we're still going through that, right, as we continue to learn and improve at this practice.

SPEAKER_02 12:12

Yeah, I think Paul said something really insightful in there too. Well, a lot of things. But one of the things was even though we're getting them out of dashboards, right, there still has to be really clear alignment on what the data means, right? What the KPI is, right? It doesn't mean that everyone can now have a different answer for same-store sales or operational metrics, right? If you don't kind of put that into the data and you kind of federate it and increase the agility, right? You're only going to get people more confused, take wrong business decisions, and then you know, kind of take steps backwards. So while you're getting them out of dashboards, building the data product in the right way kind of federates the consistency of metrics. And that's kind of not a not an easy process to do.

SPEAKER_00 12:50

Especially as we put generative AI on and allow folks to do increasing amounts of self-service, right, to get their questions answered, which is great. It's powerful, right? We don't need a dashboard for every question. Some of our sometimes our folks have a business question, they just want to ask it and get an answer and move on, right? And it's very inefficient for us to go and build a tableau data set and build a dashboard and roll it out. But to your point, Dan, if we have competing, if we have competing definitions that aren't obvious even to our business users, they definitely won't be obvious to generative artificial intelligence, and then we'll get confusing, conflicting answers. And really folks will lose confidence in the technology. It's just like with dashboarding, right? We get credibility and drips and we lose it in buckets, is what you always say, right? And so this will this is the case too. And so we definitely, it just makes that even more critical for us to get semantic information that's consistent, that's high quality, that's aligned with the business, for us to unlock this value that our business is very excited about some of the generative AI solutions that we've that we're uh prototyping with our partners. But we definitely want to make sure we get it right.

SPEAKER_01 14:01

Yeah, Paul, I mean, considering what what both you and Dan are talking about here, what are some of the clearest signs that you are looking at or that somebody out there listening in your same position could look at to know that you're doing it right? Are there certain outcomes that you're looking for? Any kind of KPIs that you're measuring against, or what what is telling you from you know the high-level view that you're you're on the right track?

SPEAKER_00 14:24

Yeah, that's a that's a good question. I mean, to some degree it varies, right? But to me, it's the cycle time for us to get new requests out. And so, you know, Dan, I think alluded to it as hey, we're not getting fewer requests. It's not like we've gone and we've done customer and now people are like, that's great. You told me everything I want to know about all of our customers, we're done here, you know, and we never will be. I I'm of the opinion, every organization I've been in, the more information you give the business, the more information they ask for, right? It just shows how much is possible and we just continue to kind of continue, excavate out, and then they say, Well, I now I want to know about this, and I want to know about that, and I want to know about the interplay of these things that I never even thought to ask about in the past. And so the demand actually increases as we give more data to the business, has been my experience pretty much everywhere. And so to me, the key metric becomes well, how fast can we service these new requests, right? We get a new request for a new data set, maybe new sources changed, maybe we need a new dashboard, maybe we need to democratize some information on board a team or a partner. We're increasingly working with partners that are just getting in and extracting value. I mean, you all are helping us with that as well, right? Of working with joint vendors and getting them to extract the information as well. That's mutually beneficial. And so the short of it is how fast can we get through that, right? Are we shrinking that time for most of our requests? Are we getting economies of scale, so to speak, where now that we've built this data set, we actually have to do less work? Because in the past that wasn't the case. In the past, it felt like every request, no matter how much work we had done, was still weeks to months, right? It didn't feel like we got any faster because we'd already done that work. But with data products, it should, right? With data products, we should not only be able to use folks that may not be as high cost, right? We don't need as many very experienced senior data engineers for some of those, but also we should get it out in days and weeks instead of weeks and months, is my hope. And so that's that's really what we try to continue to drive down responsibly, obviously. We still want to build high-quality solutions and software.

SPEAKER_01 16:23

Well, Dan, a little bit of the same question here. You're working with organizations from a variety of industries here at WWT. Is is are the benefits and outcomes uh that Paul's talking about, is that does that translate into other industries? Or is there even more benefits to kind of you know using data as more of that operational effect?

SPEAKER_02 16:41

Absolutely. Yeah, everything Paul's kind of talking about is really a cross-industry problem, right? It's kind of how do you now operationalize data? I love how Paul was talking about, you know, it it should get incrementally faster, right? It should, you know, adding to and building on top of these data products should now go from months to weeks to days. And you know, the other thing that kind of we see good versus bad, and people are sitting out there thinking, how do I know if it's working or not? What are the new kind of concepts we're seeing often and have to overcome is kind of this idea of semantic debt, right? It's something now that's more and more common when we're trying to build data products and and and maybe things that haven't been going the right way, you know, is kind of going from technical debt to semantic debt. And semantic debt is when all the stuff that that Paul and I are talking about on how where does the data live? How does it need to be calculated, how do, you know, what are the key metrics that people align on, if that's embedded in different places. I think Paul, your team kind of has been talking about pushing it, right, pushing it all into one place. But if you have you know kind of dashboards that have some amount of the semantics, right, how should things be calculated? And then other things are in warehouses, and then more is in you know, kind of applications or AI, MCP servers, right? It can get pretty complicated. And if how the core of your organization, the data is organized and understood, the semantics, right, is lost in different places, it becomes really hard for the organization to move fast and ever to be talking about the same, same you know, metrics and get the same answers to the same questions.

SPEAKER_00 18:08

Yeah, no, that's that's absolutely right. And and as you noted, you know, I mean, frankly, we still struggle with that. Our everybody's data, it seems like I've I've never worked for or with an organization that said we're completely happy with the way our data is developed and modeled. We got data products, we're good. Everywhere I've I've worked and worked with has felt like, yeah, we got a lot of work to do, which is probably always the case. And I guess it'll keep me gainfully employed for the foreseeable future. You know, hopefully all of us gainfully employed for the foreseeably future. But, you know, the the challenge is for us, it's like, hey, there's so much domain knowledge that's required to make use of even some of our building block data sets, right? You all got in there and started working with the transactions, and you're like, why is it this way? You know, I'm you got to call up one of the data engineers that's worked there for 10 years, so like, oh yeah, this is it.

SPEAKER_02 18:53

And here's why this is seven columns that are all talking about revenue, right?

SPEAKER_00 18:56

It's you know, use product identifiers getting reused, which we don't love, but it is what it is, and I can't change it now, right? And so putting that in a semantic layer, I hope, helps everybody, right? It means because we have people asking that question all the time. We're constantly getting new folks that are analyzing some of those data sets. And it's it's in the places you talked about, it's in confluence and it's in some Word documents and some design deliverables. I'm hoping that as we get more, we're never going to compress it all in the semantic layer, right? It's just oh well, I say never, but not foreseeably. But I'm hoping that we can get enough in there that it'll be kind of that'll help 80% of the people 80% of the time, and it'll dramatically reduce how much onboarding and kind of that churning has to happen. But that's definitely still a journey, and that's you know, that's the vision.

Why Big Data Models Fail and Smaller Ones Win

SPEAKER_02 19:44

Yeah, like one of the examples Paul was saying in there too, right? It's all these different places that need access to it, right? If you picture like the AI recommendation algorithm that we build for the app, right, that needs to know the menu, right? And the menu, not as it is in the raw system, but all these jumbo jacks are the same thing, right? Paul. And so it's got to know how to kind of understand the menu so it can give a really smart recommendation to the app. But at the same time, the supply chain team needs to know like the the right understanding of the menu, right? And the marketing team. And so that semantic, you know, you can't just organize at one place, right? And even from kind of like these AI systems to an individual, right? We should they should all be pulling off that same kind of semantic data product.

SPEAKER_01 20:25

Yeah, Paul, I I've heard you talk, or at least in the some of the conversations we've had, just to prep for this conversation, so you've talked more about modular data products. Can you unpack a little bit more about you know what that means in relation to you know taking a more modern approach to data in the age of AI?

SPEAKER_00 20:40

Yeah, I mean, for uh us, and the the way I usually think about it, is kind of the two facets, right? The fairly straightforward, hey, we've got the raw, intermediate, processed, and then we've probably got the layers that are tailored to a specific use case, specific dashboard, whatever the case is, right? Where we we've then composed those pieces together into one view that's really however the report needs it. But then the other facet is hey, you know, historically, like I alluded to, we intermingle to say the product data is in with the transactional data, which is in with the store data, which is in with the financial hierarchy, right? And and the and the operational hierarchy, which is a key piece for us. What's the reporting structure? And we kind of trammed all those in together and we applied the business logic at that layer, right? And so that's what I was talking about, where it really limits in our historical warehouse, it really limits our agility because anytime the reporting hierarchy changes, we have to regression test all the other stuff, even though the transactional data didn't change at all, right? This the restaurant data didn't change at all. And so that made it take literally weeks or months, and we still introduce errors in the reports in a lot of cases where we do things like double counting, which were you know really hard to identify until we rolled out to our business users, and then they, you know, told us that the store sales were off. And so, in the in the new world, we've made those much more domain specific, like we were talking about. So for us. Domains to me are a little bit like microservices, where it's like, well, you end up shuffling them around and breaking them down and conglomerating them as you learn a little bit more. But the highest level, it's you know, customer, it's Thor, it's product, it's it is the reporting hierarchy, it's supply chain, right? It's those big chunks of our business that we've put into separate data products. And data product isn't necessarily a table, a physical table, but it's probably a couple of tables. But then those evolve together, right? We've got a product owner who's assigned to those, we've got a backlog for those. We've got key business stakeholders to define requirements for it, and it grows and obviously has to relate to other domains, but then we can come and compose them and they can move at different speeds right now. Customers moving very fast. For example, stores actually moving relatively fast, but but transaction is much slower now. We went through the point of sale, it's pretty well baked, right? But it links to those other two, but doesn't have to have a change anytime that we make a change to the store data product, right? Because they're a little bit at arm's length and they're modular with domain-driven design, is what we we would say it is, you know. So does that make sense?

SPEAKER_01 23:05

Yeah, no, absolutely. I mean, Dan, anything that you would add on top of that in terms of you know where he sees kind of that puck going?

SPEAKER_02 23:11

Yeah, no, I think it's a great overview of kind of the situation. Again, we're seeing this in a variety of our customers, and really kind of the the challenge, you know, kind of what we're hearing is the business, especially now, is there's more AI use cases, right? Everyone wants to go really fast. There's all these ideas on what could be done. The business is saying, I want to do these things and I want to do them really fast. And then they're kind of passing it over the wall. And IT saying, well, it's gonna take, I got to do all of these things. I got to organize, I got to get all these data, I got to do all these things. And that middle layer that Paul described is often what's missing, right? It's kind of that that product mindset that's more understanding exactly what the business is saying, right, in the context of this data and AI and technical environment in a way that can kind of build out the data product. So it's it's it's almost a new methodology that that brings together the translation of what the business wants to do, right? And everyone's getting more technical and has more ideas on what they could do and kind of the IT side that knows how do I get these things actually stood up, right? It's kind of that more modular way Paul Paul described it.

SPEAKER_00 24:12

Well, and and I think, you know, there tended to be, I feel like, two pathologies with our older way, both in addition to being slow and error prone, like I was talking about. But in a lot of cases, we'd go and we'd get what the business asked for. We'd bring it in, we'd model it, we'd publish it, and then the business say, Oh, I really want these other attributes. I mean, we we're working through that with our back of house system right now, right? There's other attributes where the business says, this is great, I'm using the report now, but I really need these other things that I didn't know to ask for before. And they all go, all right, we got to go grab those, we got to bring them in, we got to model them, you know. And so, but in a data product, we think about it. I mean, obviously it has to support the business requirements for reporting, right? But it's its own solution, it's its own, you know, product, right? And it evolves at its own pace that can be independent from the direct use cases for. And so the idea is we'll want to go grab all the data that we can into those raw layers. Maybe it doesn't make it all the way into the product, but we have it at our disposal, and the product can include things that maybe the business didn't know what to ask for. But because we thought about it kind of in its own, you know, way, we knew, hey, eventually I'm probably gonna want these other attributes, they're key KPIs or metrics that are being captured. I want to make those available. And so it helps us a little bit, you know, as we get that modeling right to accelerate as the business inevitably asks for new attributes. But also for us at Jack in the Box, it underpins the increasing self-service or data democratization because now we can feel confident people, our business users, our power users, can go in and make their own reports, they can make their own visualizations and answer their own questions directly on those data products because they're governed things that are canonical and that are going to evolve in a controlled way. And so we feel comfortable saying, hey, yeah, go compose those, right? You don't have to go use a tableau data source or whatever the case is. You can use those products that we use to compose it that are more powerful and are going to be safe, right? And give you the right answer in a controlled way that we know we can support the analysis you build, because that's really how we scale insights, right? Is by unlocking it for the business users directly using whatever tool they're comfortable with, not just dashboard, not just Excel.

SPEAKER_01 26:30

Well, Paul, maybe you know, you're talking about business users right now. How does this all extend to the end user, to customers walking in, you know, the front door of Jack in the Box as well? I mean, you have a very extensive uh menu, one of the most extensive in the industry. How does how does that change my experience as a as a somebody that's eating at Jack in the Box or ordering from Jack in the Box and what I experienced there?

SPEAKER_00 26:52

Yeah. I mean, there are a couple of ways, because really our business is hungry for insights, right? And so you can think about everything from how do promotions perform. Our reporting really helps our business users understand that, right? And plan future promotions. Which items are resonating with which different customer groups, data, you know, that that's that's what drives that. How are stores performing? How could they improve? Or what should they learn from other stores that are performing really well, right? And benchmarking them and help helping every store, you know, meet everybody's expectations each time, right? And deliver that guest experience. The the reporting environment that my team delivers and supports is really what helps the business users in operations, in marketing, in our digital guest experience understand those. That's obviously more of that descriptive reporting, right? They get those analytics, they make a business decision, and that's what flows through the customers. Today, though, we're again increasingly using recommendations and intelligence in our digital channels in order to tailor that experience. And so today, the most obvious, straightforward way is you go, you browse the menu, you add something to your bag, you get recommendations, those recommendations are tailored to you. Hopefully, they're extremely relevant. And you maybe you didn't know you needed a shake, but we knew you needed a shake. And so you decide to add that to your bag, right? And that's good for everybody, right? It's something that you didn't even know to browse the menu. Maybe you weren't aware. We brought that back, right? And so to me, that's something great that we also, our business is extremely excited about how we can bring that to the stores because increasingly that digital experience is available in things like kiosks, right? And so, how can we not only bring that, but how can we tailor the content even more so it's easy to get your order, you know, and and and get your food in a fast, consistent manner and get up.

SPEAKER_01 28:35

Yeah, Dan, I mean, what what is the lesson there for somebody listening that's maybe not in QSR or in the restaurant industry? What's the lesson for, say, a healthcare provider or for manufacturing or pick your industry?

SPEAKER_02 28:47

Yeah, absolutely. I think I think the key in there, Paul said, is kind of with the shake, right? You can kind of see how that goes to other industries. It's kind of, you know, it's almost in a way like the classic recommendation, but now it's pulling more and more information and getting more and more personalized, right? So, you know, one of the most exciting things, you know, on the Jack in the Box side, once you rolled out that recommendation, was there was a high number of customers, I think like 30% customers, right, Paul, that tried something they had never tried before. Right. And so that tells us that we kind of delighted them and excited them with the huge menu and all the awesome stuff that Jack in the Box provides that maybe they didn't even know was sold before. And so that kind of goes into healthcare and all the other industries where can you get ahead of what someone really, really wants, really needs, right? Obviously, there's you know a ton of applications you know, you often talk about on this podcast from the healthcare space, right? But it's just your mind's easy to roam of once you get the data together and you kind of understand a person and you have a mechanism to send it back to them, right? Here's did you know we offered this, right? Did you know you might need this? Right? All those sorts of things just become more readily available, but it you you know, you need this foundation and everything that that Paul's talking about.

SPEAKER_00 29:55

Well, and you know, using the data products that you're talking about, and now we kind of built this muscle to build intelligence from them, right? And learn from the information that we've got that we're curating. We've really built for us the ability to personalize the experience. It took the form of recommendations initially, because that's really, you know, I'd argue that's table stakes for a retailer with any kind of digital or e-commerce, cross-sell, upselling your digital channels, right? Right. But now we've built this muscle and the understanding of our customers, our products, our stores, the whole thing. Now we can personalize other aspects of the experience, right? Maybe the menu should be ordered in a different way for these folks. Maybe we should show different creative content that'll resonate more directly with them. Maybe eventually we should do target targeted timed offers. We already do targeted offers, but maybe they should be part of that experience right there and tailored to that exact guest, that exact time, right? The things they've got in their cart. So there we've we've opened up a huge new opportunity for our business to think about how to apply data and intelligence to the customer training, to the digital experience that I'm excited about. And I know our business is really excited to continue evolving over the next several years because it really is becoming another example of a first-class product for us.

Agentic AI Breaks Fast Without Context

SPEAKER_01 31:05

Yeah, Paul, I mean, even you know, as what you're talking about here continues to evolve, I mean, the world of AI also continues to evolve as we move into that next era, whether it's you know, agentic, physical AI, more autonomy, things like that, how does that change the game? And you know, where do data platforms or data strategy, where do we need to go to be able to accommodate for what we think we're gonna need in the future?

SPEAKER_00 31:27

Yeah, so for us, probably like a lot of businesses, a lot of our discussions for agentic, for generative AI or internal use cases, how can we use them to make folks more productive? I mean, already, like a lot of companies, we're using coding assistance because I mean, the progress there has been incredible, even just in the last year and a half. And so, you know, it's very straightforward to give those tools to our developers, including the developers on my team, the data engineers, to build software more quickly. But how can we help our business users automate their tasks and become more productive? The key is really twofold, right? So on the data side, it's really how can we curate that data? How can we build the right semantic models in order to give people accurate answers that they can understand and interpret and that we can feel good about recognizing, you know, that we've opened up a huge scope that maybe we really hadn't ever reasoned about being used in this way before, right? So that's one thing on our side. But the other one for folks to take agentic actions is how can we open up, I've talked about data products, right? But we have a whole suite of microservices that eventually, in some capacity, will probably be opened up for these business users to build agentic automation against, right? But that's not really something that they were conceived, these services when they were built. That's not what they were conceived for. They were conceived for other developers to build other software applications that consumed and wrote to them, right? And so it's a little bit of a new world, but we're really excited about it and we're evaluating options for, hey, how can we continue to roll out orchestration and ingenic automation, both in the data space for question answering, like I said, and building dashboards and reports and probably even authoring data to that kind of workflow and automation. And then what form does it take in the customer? I mean, I think that's a really interesting, I think that's really an interesting question. I don't feel like I've seen a lot of QSR retailers inject it into the customer experience yet, but it's obviously februal space, so time will tell.

SPEAKER_01 33:42

Not necessarily trust in humans, but trust in the AI, trust in the data. If we're going to start handing over more of these autonomous actions or tasks and responsibilities to AI, how does that data foundation, what's the role of the data foundation in delivering that trust so that the adoption continue to go forward?

SPEAKER_02 34:00

Yeah, it's essential, right? I mean, I think Paul mentioned kind of in the in the near term, it's been exciting. Our teams have been able to use AI, right? Kind of, you know, AI native engineering with the data, right? Kind of data engineering and kind of even generating some metadata off of data is a really great application for Gen AI and LLMs on kind of, you know, describe this menu item, bring these things together, right? Give us contextual information that kind of builds out this data product, but then even getting ahead of it, where you know, you and and Paul were talking about is you have to trust the data to to unleash kind of the agents, right? And we're seeing some success, you know, in some some other industries as well, where you can start having the agent, you know, call it monitor the data, right? And then kind of reach out and tell someone this this action should be taken, right? Whether it's in a restaurant or whether it's at the corporate level. But again, you need to completely trust the information. And that goes back to semantic models, so kind of calculations are right, but even more so building in like business understanding, right? No one, no one's gonna want their time wasted by you know, kind of all the these pings or take this action that is completely unrelated to how they do their job. So you kind of have to build in that business reasoning layer as well of this is how the QSR industry works, right? And then even further within there, this is how Jack in the Box works, and this is how kind of the operating model. So I don't know, Paul, if that kind of builds on top of some of what you're saying, but you have to you have to build that in to kind of get to that next step.

SPEAKER_00 35:23

No, you're right, you're absolutely right. And and you, as you rightly pointed out, it's it's not the same necessarily or exactly as the semantic layer, which tells you about the data. This is about the why, right? And so a lot of our, I mean, our partners with Snowflake have come out with some really compelling new solutions that our business is excited about, especially our power users today. But the question that our business leaders have is say, hey, it's great for you to tell me that maybe the stores in Texas are under or overperforming expectations, but why, right? The why, how can it? That's that's information. You've given me information, but I need insights. And those are not the same thing. And you have to understand the business because, for example, whenever we started looking at some of the models initially, just naively, right? We said, Hey, why are these sales off? And they said, Oh, well, the labor is lower than it was last year, but that's that's that's a follow-on, right? If the sales are down, people schedule less labor, right? Or vice versa. But that less labor didn't drive the sales up or down, right? At least not in that kind of a time frame. And so it didn't really understand the causality there of those of those factors. But that's why we're starting to collect exactly what some of what you talked about, Dan, is hey, how do these KPIs relate just conceptually at a business level? What do we think are the things that influence these KPIs and how they relate? And starting to feed that in kind of as a background, right? And hopefully continuing to drive more of that intelligence about the business and about how the information relates to to get closer to insights because that is definitely the field that we're that we're working against that our business would be much more excited about.

SPEAKER_02 37:00

Yeah, I think it's a great example there on the labor and the sales, right? Again, you know, we're kind of seeing that across customers. I think Paul kind of called it out. You know, kind of Snowflake's Cortex environment is incredible. We're finding some places where customers are already relying on the agent to understand and then give actions that people are trusting. And it's really shifting the whole how does the business work, right? It's kind of redesigning the business process because but it but you have to do everything that Paul said, right? You know, at first when you first see it and it gives you a recommendation that's completely backwards, you think, oh wow, this is really far off. But in reality, it's just how you kind of how you provide it, the context, business, semantic, all the rest, and then you kind of see how how much it grows, and then you can start trusting it.

Why Data Access Without Alignment Backfires

SPEAKER_01 37:43

Yeah, Paul, I mean we're coming close on the bottom of the episode here. So I got one more question before we'll get into somewhat of the uh conclusion here. But I was listening to a podcast that you were on, I believe it was recently, or maybe it was an article that I had seen you uh quoted in. But you were talking about data democratization as a continuum and not necessarily a binary. Wondering if you can kind of just unpack kind of that idea. Is it does that fit in with what we're talking about here, like we've already touched on it, or is that a little bit of a new uh issue to bring up?

SPEAKER_00 38:10

No, it's it it's related to what I was talking about with the data products, right? Where I think about in a lot of cases, we we think about, hey, how do I empower the business users, right? How do I meet them where they are? And our business users have differing levels of technical proficiency, technical being IT technical, right? And and so some of my business users really they just want a dashboard. That's all they want, that's all they'll ever want, right? Show it to me in a dashboard, I don't care about anything else. Some of them will walk back and they'll say, Well, I'll connect that to Excel, right? I'll connect Tableau Data Source or whatever it is I want in Excel. That's great. And then they go wild with it. But then some of my folks, maybe they're willing to get in, they're willing to modify some queries or explore it in Snowflake. And then some of the folks, they're gonna write Snowflake, you know, SQL directly in Snowflake, they're gonna write Python, you know, everything. And they understand the data as well as anybody on my team does. And so we kind of have this continuum of users that are willing to engage in the data and the technical platforms, technology platforms at differing levels. And historically, a check in the box, we've really indexed kind of on the right, right? We built a lot of data, dashboards for a lot of people. You really couldn't dig too far down in the data set unless if you were really one of a few folks. And so it left a lot of people waiting on this cycle over here, right? We've got to get the data, we've got to curate the data, we've got to publish the data, and then you get it, and you can maybe write some queries, but mostly you're gonna use Excel or Tableau, right? And now that we've built these data products, now we've got the modern data stack and some of the tools that we're talking about, like generative AI built into it, we unlock a little bit more of the left-hand side and we feel good about it because it obviously comes with some governance considerations, right? I've done that at a lot of organizations where now people are building their own data sets. That's awesome, right? And I've seen those blow up and become enterprise scale reports. And then you turn around and it's like, well, that person, that business analyst in a business unit is not going to support that report that they just built that 500 people are using now. I hope they use the right data, right? And so then we have to talk a lot about that governance. But I've always personally found that's a good conversation to be having because those business users are doing things we never thought of because those business stakeholders never thought to ask us, right? But once they have that deep understanding of the data and the technologies and the business in a way that we really can't always bring together, if I'm being totally honest, right? They can build new solutions that nobody even understood to ask for before and then democratize that. And people are getting value out of data at a faster rate in a deeper way than previously. And so that's that's kind of what I mean by that. Yeah. Yeah.

SPEAKER_01 40:38

And I mean, Dan, that seems like a as good of a mental framework for any organization to carry on with.

SPEAKER_02 40:43

Yeah, absolutely. And even, you know, internally, right? Within WWT, we always hear kind of everyone talking about it. We're kind of following all these practices as well to kind of make ourselves AI first and expose data to all these applications in the way that you know, kind of Paul's talking about. So absolutely a framework across.

The Hard Lessons We’d Carry Forward

SPEAKER_01 40:59

Yeah. All right. Well, we'll close out with this question. And Dan, I'll start with you, and then Paul, I'll ask you to kind of build on what Dan says in his answer. If we're talking to organizations out there, our listeners out there that, you know, they may maybe inside any number of industries, if they're building or starting to formulate a roadmap for how to tackle what we have in store for us over the next two to three years, what do they need to do to start? And, you know, as it relates to that, that data foundations, what are the two or three kind of key lessons, so to speak, that we've covered here today?

SPEAKER_02 41:28

Yeah, you know, I can I'll I'll start, right? I think the first one is on this kind of mindset shift to a product mindset, right? To kind of really thinking about data in a different way than you kind of historically have. And it's really this intersection, more translation than ever between what are the business goals and looking out into the future, what will the business goals be? Right. You don't need all the details, but kind of what generally do we need to be able to do and then kind of translate that to the technology side of things and kind of set your data up and kind of this kind of productize modular way with the right semantic model and business reasoning and all these things, right? It's not spend a ton of time and build the foundation, right? But it is, you know, think about it to some extent and then build components of the foundation as you're proving value and you're unlocking. And then I think you know, Paul said it great before, you're just reducing the time to outcome continuously as you go when you do it this way.

SPEAKER_00 42:21

No, I I we're obviously on the same page there, Dan, because I, you know, whenever we started our journey to Snowflake, my business could not care less about a data warehouse migration, right? I mean, if I told them that's what we're doing, they'd say, oh my God, you guys are always moving from one thing to the other, and I'm gonna get what I had before, right? I mean, like, and so they do not, there's no inherent value, I would say, in moving from one warehouse to another warehouse. And so we had to make sure we're incrementally delivering new capability. In this case, that was new data products, right? Hey, you know, we were on Snowflake, and now because we're on Snowflake, you have access to customer data that you didn't have before. Now you have access to generative AI tools that help you ask questions directly. Now you can build your own stream apps, talk about a governance nightmare, but it's a great unlock for some of our business stakeholders that are extremely technical and were previously capturing data in Excel and Smartsheets, right? And so we have some of these technical unlocks that we've laddered to business unlocks to say, now I've got deeper insights about customers. Now we've corrected and we've reduced our time to market for new attributes about point of sale and transactions. And one of the things I'm super excited about as we do this next phase is hey, we'll start delivering back of house data like labor, like inventories, much faster than we did previously. Folks on our East Coast have to wait until the afternoon. We're gonna finish that migration, we're gonna pull that up to the morning. That'll be a huge business win. People don't care about Snowflake or our previous data warehouse, they care about getting their data faster, more accurately, right? And so those are all the things that we've tied, exactly like Dan said. We couldn't take six months a year to do pieces of it, right? We've got to show those breadcrumbs along the way. And that's really what we lead with. We don't lead with we're moving to Snowflake, we lead with. We're giving you more customer insights and data democratization because we've done this migration, because of this investment that we've made. And that gets people excited about it.

SPEAKER_01 44:09

Yeah, no, absolutely. Uh well, Dan, Paul, fantastic conversation, lots of insights, and um, you know, great just conversation um all in all. Uh, Paul, thank you so much for for the partnership and for being here on the show today. Dan, you as well for making time to be on the AI Proving Ground Podcast. We'll have you on both here again soon and uh look forward to that. Perfect. Yeah, thanks for having me. Thanks for okay. Thanks to Paul and Dan for taking their time today. Maybe the clearest lesson from this conversation is this AI does not create operating leverage on its own. Trust does. And at Jack in the Box, that trust has to extend from customer data to store data to menu data to the metrics that the business uses every day. This episode of the AI Proving Ground Podcast was co-produced by Nas Baker, Kara Kuhn, and Megan Wood. Our audio and video engineer is John Knoblock. My name is Brian Felt. Thanks for listening. See you next time.