The AI Table: Hi everyone, and welcome to the AI table. I am your host, Dea Adams, former White House Enterprise Data and AI strategist. and Fortune 500 AI advisor. I have a wonderful guest with us today. someone who I hope she doesn't mind if I also call a friend, Jenna Zhang. She is Chief Data and AI officer at University of St. Thomas. But I will let her tell you a little bit more about herself. Jenna, you wanna talk a little bit about your background?
Jena Zangs -U of St. Thomas: Yes, lovely to meet everybody and thank you so much for joining. And as as Dea said, I'm the Chief Data and AI officer at the University of St. Thomas in Minnesota, not the island, which I feel like most people always dive right into. And I am a builder with heart and really focus on being a grounding force of AI across education. I'm the first chief data and AI officer across any higher education in the Midwest, so it's been thrilling and never a dull moment, which is exactly how I'd rather be living life. And The state ambassador for Minnesota for the Global Women in AI Group. So I've been really expanding the community around Minnesota and bringing people together again, trying to be that grounding force across education and AI and data, which is my absolute passion. I have a master's degree in data science and have focused a lot of my efforts on data modernization and data strategy and now AI components and machine learning in the last decade. So it has been Every day does not feel like work. It feels like fun, which I just I love ever so much. So I'm so honored to be here and talking about this topic today, too.
The AI Table: That's a great spot to be in. I didn't realize you were the first chief data officer for a university in the Midwest. That's quite an accomplishment. So you're truly, truly on the front line. So I'm glad that we have
Jena Zangs -U of St. Thomas: Yes.
The AI Table: you here. You're gonna have some interesting insights for us. So let's go ahead and dig in. First question. a lot of execs are familiar with generative AI as a conversational tool. You know, you ask ChatGPT a question and it gives you text. But Agentic AI shifts us from asking to delegating. What does this shift mean for how we build data platforms?
Jena Zangs -U of St. Thomas: Yeah, I think this is such an interesting time because we've spent really the last decade trying to build data for humans to look at. And when I think of data, I think of structured data. And now we have to build data for machines to reason with and potentially act on. So that really changes the stakes. And it also puts so much more emphasis on the The data access across our platforms. In in educational space, we tend to have we are very comparable to a city. So we have 250 products across the university, which is a significant amount. And when you think of an agentic individual on behalf of an individual, you have to manage the data access to 250 different tools, which not only the access and what they might be able to do, but the ability to even understand and monitor them, right? so I I think the mind shift of what data is available, what we're tracking on individuals, and the quality of data will be even more important and something for teams to think about that haven't gotten there. And I'll talk a little bit about unstructured data later on, but that's just the next I I mean frontier for us too.
The AI Table: Agreed, agreed. a lot of companies are still thinking of data as tables sitting in a warehouse. How do you define a true data product in this new context with autonomous agents?
Jena Zangs -U of St. Thomas: I love this question because again we Create ample amounts of data every single day, and people think it is Excel, a spreadsheet with all you know numbers and things. But a data product isn't a table, it's really a trusted package of information for use. And when you think of the unstructured data, it's really 80% of our world right now. Thinking of pictures, sound, video, you know, PDFs, anything that isn't in that table format, that is what most of our world is actually utilizing. So a data product isn't going to be a simple table or view that we're using. to it's going to start to incorporate unstructured information such as pictures or those PDFs and things that organizations are not even really tapping into now but agentic work can really flourish. The the examples that I think about are I've worked with some manufacturing companies doing computer vision and if you had a a data product that not only had the orders you're listening through but pictures that were from the floor you could actually have an agent monitoring what is happening across the floor. And looking for deficiencies and stopping production on your behalf when there is some type of deficiency or if there's a malfunction. Something that all could be packaged in this governed infrastructure or governed data product that they could be utilizing. So it it really just expands beyond a simple table. But to get to that final data product, it it takes so much effort, right? I I mean the definitions of things, even defining what a customer is is one of my favorite questions to ask when I'm I'm working with different organizations and they all have different answers, right? And that is the crux of the problem. so
The AI Table: Hopeful
Jena Zangs -U of St. Thomas: you really have to have a streamlined governed, you know process to be able to get to this place where the agentic has less questions because you've already done the the pre-work to defining and making sure you're in alignment across the organization.
The AI Table: As as data professionals, you know, we understand that process, but sometimes the leaders that we work with, they do not. They just want the result. Do you have any tips for other leaders on how to be persuasive or how to approach that situation?
Jena Zangs -U of St. Thomas: Yeah. I I really like to come at it and showcase the downstream effects because if it comes down to ROI or financials or you know, human capital in terms of the time people are spending on the efforts they're doing and when you can showcase that the differences of you know, a few hours versus hundreds of hours on something that they're doing. I think that is where I found the most progress, being able to understand the translate what we're doing on the back end to really make change on the front end. And most people actually don't like talking about data. We're kind of a rare few. And there's actually a great book about it that like
The AI Table: Yes, agree, agree.
Jena Zangs -U of St. Thomas: they would rather do their taxes than talk about data. So people don't want to think about it that they're actually talking about data. So I try and reformat it into that business process. And they don't need to worry about how we're going to make it efficient. They just need to know this is the outcome they're going to have. And so being really direct in that sense has been helpful and leaving a lot of the jargon out because they already trust you as the data lead. They don't need to know.
The AI Table: Yes, I I agree with that completely. and I'm just gonna tell on our profession a little bit here. I think a lot of us are in denial about that fact because we're so excited. We do we love data. This is you know, this is what we do and we want everyone else to love data, too. And sometimes we can come to those conversations and bring our jargon and I think It's counterproductive most of the time when we do that. I often say that in order to be a strong data and AI leader, you have to be multilingual, meaning that we have to be able to speak data, but then when we're at different tables, we have to be able to speak business so that people can understand us and not zone out when we're when we're in the meeting talking. So it's great that you accept that, understand that and are are navigating that with
Jena Zangs -U of St. Thomas: And I think to build on that even more that The new piece with AI has been how to na navigate the pressure that's coming in. Of course, C-level suites always have pressure, but this AI pressure is not only coming from C-suite, it is coming from media, it's coming from you know, s your performance, it's every angle you are getting. It feels like a middle school peer
The AI Table: Mm-hmm. Mm-hmm.
Jena Zangs -U of St. Thomas: pressure all the time. And so I think as a leader too, trying to figure out where do you pass that pressure on or where do you absorb it and and deflate it actually to showcase you know
The AI Table: Mm-hmm.
Jena Zangs -U of St. Thomas: it It's really not that big of a deal. We can figure these pieces out and so I think that's a a new dynamic that i has been very interesting in this position just to navigate and figure out
The AI Table: It's it's very unique, agreed. I've never seen anything like this in my career. It's very exciting times for folks like us. but I like what you said. that it's it it's really about exercising good judgment, for leaders in this moment because you know, you use the word deflating and I th I think that's a good term to use. That that's something that is important to really decipher the hype from real value and and whether you should go in a certain direction or not. It's it requires knowledge and it r requires strong leadership. so new topic. We have seen companies in the news recently who have spent gobs and gobs of money on tokens with with little ROI, you know, every every CFO's nightmare. Do you have thoughts on how data teams can turn raw data dumps into lean high density cheat sheets that can keep these AI costs under control?
Jena Zangs -U of St. Thomas: So if your tactic is to really go out and again just say, go to all my tools, find that information, the likelihood of that data being quality is is pretty minimal. And it may be quality from the information you're putting in, but if you're relying on anybody else, or if you're looking for more systematic maybe data that could help really improve, you're gonna find, like you said, you're gonna be wasting credits on things that, we didn't found it, but you just got charged for it. And so Reminding yourself of what key data points are actually important. So I I think too, when you just say go get everything from a tool, you're probably grabbing information you also don't even need for the thing you're trying to execute. So if you can continue to narrow in your agent on I want these specific data points, and you might have to be the validator, or you build an agent to be your validator and then start working that way. But from a data team perspective, you should know what the most important data is at your organization, and that's what you prioritize for your data. Products so something around people, right? How all the systems work together so that the agents can interchange. Then, for our specific, you know, we have a student focus, HR focused, finance, and and so and having those pieces available and already through the pipeline so that you can really push individuals to use agents towards those quality data products and eliminate the need to actually just go into the tool themselves. That is what the mode should move to. And A lot of us are needing to be collaborators with those business owners and the business owners have to have a lot more stake in the game too if they are allowing those
The AI Table: Absolutely.
Jena Zangs -U of St. Thomas: agents, whether they are are open to it or not, those agents are already going in there and doing that. We we are seeing that with students, right? They're they're pushing the limits just like they should be, which is one of my favorite parts, and showcasing how they can go in and grab all their homework that they're looking for, work on things and then put it back in there. I mean, I'm talking with schools who are are dealing with that right now. And so it's just an interesting time that we want to keep up with that innovation, but yet making sure we're being protective of the quality of information they're they're grabbing.
The AI Table: Great, great answer. so next up, connecting an AI agent to many different company databases usually requires building expensive custom pipelines for every single one. can you talk a little bit about some of the new universal plug and place standards that can help solve this and and connect to AI more seamlessly?
Jena Zangs -U of St. Thomas: Yes, I I also love the we're really seeing the legacy data integration like you're talking about, the point to point custom scripts, which yes, AI can help make the scripts go faster, but really the overall model can't be that point-to-point anymore. And with this idea of the data product, it really turns into that hub and spoke model where that becomes your central place. And when you make one change to that data product, it can affect all the rest of those integrations instead of having to go to each script every single time. And so I think that allows you to not only put more emphasis on how important that data product is and have minimal changes needed when you're tracking it. But I love this idea that I could start actually going to my vendors and having my data product have an API endpoint that they actually have to come and grab my information. So in education, a lot of the times We're building a file, we're dropping it in into some SFTP site somewhere, and then they're utilizing that in the product. But now if they have agentic work going on, I could actually have them do the heavy lifting and relieve my team of having to manage that. And I worry about the data. and then they worry about actually making sure that the system operates. So I could see the shift in putting more responsibility onto the vendor versus just having it on our university. And we still would have all the logs of what they touch what they've used and so the the security piece is actually even more beneficial because we get that information versus just putting it in a file and not sure what happens after that.
The AI Table: Mm. Yeah, it it's been a real leap forward in that regard.
Jena Zangs -U of St. Thomas: Yeah.
The AI Table: So traditional data teams, they they run nightly batch jobs, but AI agents often need real time context to act safely. How do real time pipelines fit into agentic data architecture?
Jena Zangs -U of St. Thomas: Again, tying back to we produce so much data in our organizations, but really we're using probably 20 to 30% of the key metrics on a daily basis to actually act on. And so when you are focused on that type of data and making that available, that is what's gonna be most beneficial for you as you're acting on these agents. And yes, there are nightly scripts, but if you have those data products able to be better integrated throughout the day and maybe even more cost effective because you're only working on event change management rather than whole file droppage, right? That that can be more effective for. data team going forward and again getting out of that point to point and overnight I think the efficiency will allow us to have less of that because when things go wrong we the agent could have a backup agent that's helping them fix it at that same time. I mean this whole new web of of being available in live stream is something that we've always wanted more of but we're trying to be restrictive on on resources, right, to manage all that. So going forward, I do think that we could move away from more of that nightly batch as as the systems become more accepting of this type of infrastructure, right? many of us have legacy systems, which we're finding is the hardest part to integrate and still need some of those point to point and overnight things. But I do think over time as organizations continue to offer more of these advanced solutions and adopt agentic infrastructure that will see less of that nightly batch.
The AI Table: Yes, we're we're we're moving in that direction. data restructuring is expensive and it's quiet work. How do you convince a CFO or CEO to invest in agent ready data products when they just wanna see that front end shiny app?
Jena Zangs -U of St. Thomas: I'm sure you're probably feeling this too, but I love that the word governance is having its moment, like all over. I mean seeing it on a gigantic poster, I'm like, three years ago I couldn't get anyone to talk about this or even like want to participate, right?
The AI Table: No. It's so exciting. I haven't been this excited since we became data scientists. We were acknowledged
Jena Zangs -U of St. Thomas: Yes.
The AI Table: as I was like, yeah, I like that. give me that title. Yeah, we're we're moving up in the world.
Jena Zangs -U of St. Thomas: Yes.
The AI Table: They're they're speaking our language now. Yes, I am seeing that.
Jena Zangs -U of St. Thomas: Right. So that's that is one benefit of all of the media coming at us from all those angles is these keywords that we have been saying. They have they have listened to them and now are amplifying them. So I think that at the the C suite, they are hearing those and understand, okay, they must they have more value without having to understand how deep that actually goes. And having the you know, a data strategy that is intertwined with your AI strategy. One can't live without the other and when I tie it to the individuals and the operational pieces and you know, that I like to say at the university that the people are the heart, but the data is the heartbeat, right? You can't have anything going forward without that data piece. And so as we look at getting that shiny new app, if you put something, even a Vibecode prototype with some bad data, you could show them exactly the difference between bad data and good data, which I I love vibe coding right now for these quick little prototypes that get the real effect out there. but it's it's those small things that you can show. Nothing of this will work without the foundation and the governance and the people. It it it's just the the elements that we have known since we've been trained as a data scientist to to find and that the C suite is a lot more open to that now. so it's helpful.
The AI Table: Mm-hmm. Yeah.
Jena Zangs -U of St. Thomas: I I don't know if you've seen the same thing lately.
The AI Table: I I agree. This is really our moment to shine, I think. and I agree with you with respect to demoing. That's a that's a trick that I use all the time, because I find that non data nerds, they need something tangible that they can see that they can relate to that solves their problems, makes their job easier, helps them meet whatever goals and metrics that they're looking to meet. And once you can show that visibly an example of like this is why this should matter to you and then I I keep why it matters to me to myself. But then I show them this is why it matters to you and then we can all get on the same page. and I find that that really when you use real life examples and you have something that you can demonstrate or that other teams can play with, then it really elevates the value of the data team when you're able to do things like that. so that you can be seen as collaborators because sometimes I hate to say it, particularly when it comes to governance, historically we could get trapped in like what I like to call the data police state where everyone just thinks that we're there to say you did something bad, you're not doing your data right. you know, and we don't want to be seen like that. We want to be seen as data friends is how I like to be seen. I I'm your friend, I'm here to make life easier for you and and demos are a big part of that for me.
Jena Zangs -U of St. Thomas: I'm so glad that we aligned on that. And yes, the the word empowerment was our instead a few years, I don't know, five, six years ago when we started on the data governance, that we were advised try and say something like empowerment or something. So that's what we did, and and now governance is a big thing. So we've been deciding how to shift. But I love seeing the data talent as well being pushed to evolve along with this, right? It's no longer good
The AI Table: Mm.
Jena Zangs -U of St. Thomas: enough to just be a BI reporter. You need to elevate and understand some of this infrastructure and how these types of making this data available and and figuring out how to use AI for better quality too.
The AI Table: Yes, because truly we are ambassadors is is how I think of us. We are ambassadors to the other business units, helping the organization to operate better. yeah, so don't resist that folks if you're out there listening. It's it's a part of our jobs now. Get on board.
Jena Zangs -U of St. Thomas: Yes.
The AI Table: so another question, When it comes to AI, sometimes things fail. When an agent is querying data across different units, HR, finance, customer support, et cetera, to make decisions, how do you determine who owns the underlying data product or who gets blamed if the agent fails?
Jena Zangs -U of St. Thomas: Yeah, so we've been actually working through a pretty intensive racy on on this and and making sure who's accountable and responsible for this very scenario as we're working through our own lakehouse development and I'm finding that there is a we are actually putting in a role for a data product owner which could be different the data owner always owns the data so I think when I think of your scenario you got HR finance and customer experience so those would be your data owners but if the product has multiple things because again we we picked a hybrid a data mesh architecture so that the businesses could stay closer, but as those things start to blend, we're finding that the data product owner might be one of those three, and that is who becomes accountable for what is happening in there. That's their job is to manage and understand and make sure the data is good in there, while the data owner is to make sure that they are showcasing what should be used, how are we defining the final decision maker on what who can have access, those types of things, but the data product owner becomes accountable. Now If the infrastructure that you're using, some cloud-based infrastructure or tool set, then I see that that those errors coming out of IT or whoever is managing that infrastructure side. But ultimately if the data is bad or there's a failure in something that was uploaded from the product that is feeding these data products, then I think it becomes that data owner and that data product owner too. Is that resident? yeah, go ahead.
The AI Table: When it comes yes, when when it comes to determining who the data product owner should be, is that something you let folks volunteer for? Or are you saying, Hey, I think you you fit the bill for this?
Jena Zangs -U of St. Thomas: So we actually are working with the data owners to say who is the one that understands and is the best subject matter expert in your area to to navigate some of these pieces. Who will notice when something is off by a bit? And when I think about this too with the when you're asking about the agentic work too before, you know, you can notice errors when they're like huge drastic changes. But when it's like, let's say $30 of a difference on a budget, is someone truly gonna notice that or not? You know, and so that's the type of person we're looking for who has the technique technical acutely. to understand what they're looking at and question it. And so what we
The AI Table: yeah.
Jena Zangs -U of St. Thomas: found is that the the data owners actually work together and decide who that will be. And sometimes it is two people, which isn't a bad thing either, but typically it comes down to one.
The AI Table: Yes, that that's how that's what I've seen as well.
Jena Zangs -U of St. Thomas: Yeah. Great.
The AI Table: so my last question, any data or AI, pet peeves, things you wanna get off your chest, things you wanna stop seeing, things you wanna see more of, or just anything you wanna leave the audience with.
Jena Zangs -U of St. Thomas: So I guess I would say a pet beef right now is I want I don't want people to feel so pressured that they are behind. I am really tired of seeing that all the time, of this just anxious feeling that it's creating. We you don't need to move at any specific pace other than what works for you and your learning ability. And and so I just I would love to continue to put that message out there that you just move at the pace that you can do, you know, that you feel comfortable with. And it's okay if you feel like you should be using AI more or data. And there's there's always today. It's a gift. And then I think another thing that I would love to share is that any students or any anyone who's looking to evolve their job into this role, it is only opportunity right now. Hey, it's just especially for I see all these students in education and there's a lot of turmoil going on between, you know, when is AI used for critical thinking, when isn't it, all of this pieces, but for anyone, it's just it's a builder's playground, right? There are free credits everywhere. Everyone wants you to try it. And there's just
The AI Table: yeah.
Jena Zangs -U of St. Thomas: The idea of failing at something is just is in ridiculous in my mind that you could go out and try anything that you've that idea you've had for five years, it's there. And so
The AI Table: Yeah. Yeah.
Jena Zangs -U of St. Thomas: and I love that you can you can really build that and that we're questioning what type of data we should be capturing with people and so I would just I I want you to people to just go out and be that builder and not be afraid to try.
The AI Table: Yes, that's a really great point. It's it's a great time to create and to bring your ideas to life. AI makes that a lot easier. Thank you so much, Jenna. It was it was an absolute pleasure having you on.
Jena Zangs -U of St. Thomas: Thank you. I'm just I I love the the show and I'm so thankful to be connected with you too.
The AI Table: Assam.
Jena Zangs -U of St. Thomas: Yes, lovely to meet everybody and thank you so much for joining. And as as Dea said, I'm the Chief Data and AI officer at the University of St. Thomas in Minnesota, not the island, which I feel like most people always dive right into. And I am a builder with heart and really focus on being a grounding force of AI across education. I'm the first chief data and AI officer across any higher education in the Midwest, so it's been thrilling and never a dull moment, which is exactly how I'd rather be living life. And The state ambassador for Minnesota for the Global Women in AI Group. So I've been really expanding the community around Minnesota and bringing people together again, trying to be that grounding force across education and AI and data, which is my absolute passion. I have a master's degree in data science and have focused a lot of my efforts on data modernization and data strategy and now AI components and machine learning in the last decade. So it has been Every day does not feel like work. It feels like fun, which I just I love ever so much. So I'm so honored to be here and talking about this topic today, too.
The AI Table: That's a great spot to be in. I didn't realize you were the first chief data officer for a university in the Midwest. That's quite an accomplishment. So you're truly, truly on the front line. So I'm glad that we have
Jena Zangs -U of St. Thomas: Yes.
The AI Table: you here. You're gonna have some interesting insights for us. So let's go ahead and dig in. First question. a lot of execs are familiar with generative AI as a conversational tool. You know, you ask ChatGPT a question and it gives you text. But Agentic AI shifts us from asking to delegating. What does this shift mean for how we build data platforms?
Jena Zangs -U of St. Thomas: Yeah, I think this is such an interesting time because we've spent really the last decade trying to build data for humans to look at. And when I think of data, I think of structured data. And now we have to build data for machines to reason with and potentially act on. So that really changes the stakes. And it also puts so much more emphasis on the The data access across our platforms. In in educational space, we tend to have we are very comparable to a city. So we have 250 products across the university, which is a significant amount. And when you think of an agentic individual on behalf of an individual, you have to manage the data access to 250 different tools, which not only the access and what they might be able to do, but the ability to even understand and monitor them, right? so I I think the mind shift of what data is available, what we're tracking on individuals, and the quality of data will be even more important and something for teams to think about that haven't gotten there. And I'll talk a little bit about unstructured data later on, but that's just the next I I mean frontier for us too.
The AI Table: Agreed, agreed. a lot of companies are still thinking of data as tables sitting in a warehouse. How do you define a true data product in this new context with autonomous agents?
Jena Zangs -U of St. Thomas: I love this question because again we Create ample amounts of data every single day, and people think it is Excel, a spreadsheet with all you know numbers and things. But a data product isn't a table, it's really a trusted package of information for use. And when you think of the unstructured data, it's really 80% of our world right now. Thinking of pictures, sound, video, you know, PDFs, anything that isn't in that table format, that is what most of our world is actually utilizing. So a data product isn't going to be a simple table or view that we're using. to it's going to start to incorporate unstructured information such as pictures or those PDFs and things that organizations are not even really tapping into now but agentic work can really flourish. The the examples that I think about are I've worked with some manufacturing companies doing computer vision and if you had a a data product that not only had the orders you're listening through but pictures that were from the floor you could actually have an agent monitoring what is happening across the floor. And looking for deficiencies and stopping production on your behalf when there is some type of deficiency or if there's a malfunction. Something that all could be packaged in this governed infrastructure or governed data product that they could be utilizing. So it it really just expands beyond a simple table. But to get to that final data product, it it takes so much effort, right? I I mean the definitions of things, even defining what a customer is is one of my favorite questions to ask when I'm I'm working with different organizations and they all have different answers, right? And that is the crux of the problem. so
The AI Table: Hopeful
Jena Zangs -U of St. Thomas: you really have to have a streamlined governed, you know process to be able to get to this place where the agentic has less questions because you've already done the the pre-work to defining and making sure you're in alignment across the organization.
The AI Table: As as data professionals, you know, we understand that process, but sometimes the leaders that we work with, they do not. They just want the result. Do you have any tips for other leaders on how to be persuasive or how to approach that situation?
Jena Zangs -U of St. Thomas: Yeah. I I really like to come at it and showcase the downstream effects because if it comes down to ROI or financials or you know, human capital in terms of the time people are spending on the efforts they're doing and when you can showcase that the differences of you know, a few hours versus hundreds of hours on something that they're doing. I think that is where I found the most progress, being able to understand the translate what we're doing on the back end to really make change on the front end. And most people actually don't like talking about data. We're kind of a rare few. And there's actually a great book about it that like
The AI Table: Yes, agree, agree.
Jena Zangs -U of St. Thomas: they would rather do their taxes than talk about data. So people don't want to think about it that they're actually talking about data. So I try and reformat it into that business process. And they don't need to worry about how we're going to make it efficient. They just need to know this is the outcome they're going to have. And so being really direct in that sense has been helpful and leaving a lot of the jargon out because they already trust you as the data lead. They don't need to know.
The AI Table: Yes, I I agree with that completely. and I'm just gonna tell on our profession a little bit here. I think a lot of us are in denial about that fact because we're so excited. We do we love data. This is you know, this is what we do and we want everyone else to love data, too. And sometimes we can come to those conversations and bring our jargon and I think It's counterproductive most of the time when we do that. I often say that in order to be a strong data and AI leader, you have to be multilingual, meaning that we have to be able to speak data, but then when we're at different tables, we have to be able to speak business so that people can understand us and not zone out when we're when we're in the meeting talking. So it's great that you accept that, understand that and are are navigating that with
Jena Zangs -U of St. Thomas: And I think to build on that even more that The new piece with AI has been how to na navigate the pressure that's coming in. Of course, C-level suites always have pressure, but this AI pressure is not only coming from C-suite, it is coming from media, it's coming from you know, s your performance, it's every angle you are getting. It feels like a middle school peer
The AI Table: Mm-hmm. Mm-hmm.
Jena Zangs -U of St. Thomas: pressure all the time. And so I think as a leader too, trying to figure out where do you pass that pressure on or where do you absorb it and and deflate it actually to showcase you know
The AI Table: Mm-hmm.
Jena Zangs -U of St. Thomas: it It's really not that big of a deal. We can figure these pieces out and so I think that's a a new dynamic that i has been very interesting in this position just to navigate and figure out
The AI Table: It's it's very unique, agreed. I've never seen anything like this in my career. It's very exciting times for folks like us. but I like what you said. that it's it it's really about exercising good judgment, for leaders in this moment because you know, you use the word deflating and I th I think that's a good term to use. That that's something that is important to really decipher the hype from real value and and whether you should go in a certain direction or not. It's it requires knowledge and it r requires strong leadership. so new topic. We have seen companies in the news recently who have spent gobs and gobs of money on tokens with with little ROI, you know, every every CFO's nightmare. Do you have thoughts on how data teams can turn raw data dumps into lean high density cheat sheets that can keep these AI costs under control?
Jena Zangs -U of St. Thomas: So if your tactic is to really go out and again just say, go to all my tools, find that information, the likelihood of that data being quality is is pretty minimal. And it may be quality from the information you're putting in, but if you're relying on anybody else, or if you're looking for more systematic maybe data that could help really improve, you're gonna find, like you said, you're gonna be wasting credits on things that, we didn't found it, but you just got charged for it. And so Reminding yourself of what key data points are actually important. So I I think too, when you just say go get everything from a tool, you're probably grabbing information you also don't even need for the thing you're trying to execute. So if you can continue to narrow in your agent on I want these specific data points, and you might have to be the validator, or you build an agent to be your validator and then start working that way. But from a data team perspective, you should know what the most important data is at your organization, and that's what you prioritize for your data. Products so something around people, right? How all the systems work together so that the agents can interchange. Then, for our specific, you know, we have a student focus, HR focused, finance, and and so and having those pieces available and already through the pipeline so that you can really push individuals to use agents towards those quality data products and eliminate the need to actually just go into the tool themselves. That is what the mode should move to. And A lot of us are needing to be collaborators with those business owners and the business owners have to have a lot more stake in the game too if they are allowing those
The AI Table: Absolutely.
Jena Zangs -U of St. Thomas: agents, whether they are are open to it or not, those agents are already going in there and doing that. We we are seeing that with students, right? They're they're pushing the limits just like they should be, which is one of my favorite parts, and showcasing how they can go in and grab all their homework that they're looking for, work on things and then put it back in there. I mean, I'm talking with schools who are are dealing with that right now. And so it's just an interesting time that we want to keep up with that innovation, but yet making sure we're being protective of the quality of information they're they're grabbing.
The AI Table: Great, great answer. so next up, connecting an AI agent to many different company databases usually requires building expensive custom pipelines for every single one. can you talk a little bit about some of the new universal plug and place standards that can help solve this and and connect to AI more seamlessly?
Jena Zangs -U of St. Thomas: Yes, I I also love the we're really seeing the legacy data integration like you're talking about, the point to point custom scripts, which yes, AI can help make the scripts go faster, but really the overall model can't be that point-to-point anymore. And with this idea of the data product, it really turns into that hub and spoke model where that becomes your central place. And when you make one change to that data product, it can affect all the rest of those integrations instead of having to go to each script every single time. And so I think that allows you to not only put more emphasis on how important that data product is and have minimal changes needed when you're tracking it. But I love this idea that I could start actually going to my vendors and having my data product have an API endpoint that they actually have to come and grab my information. So in education, a lot of the times We're building a file, we're dropping it in into some SFTP site somewhere, and then they're utilizing that in the product. But now if they have agentic work going on, I could actually have them do the heavy lifting and relieve my team of having to manage that. And I worry about the data. and then they worry about actually making sure that the system operates. So I could see the shift in putting more responsibility onto the vendor versus just having it on our university. And we still would have all the logs of what they touch what they've used and so the the security piece is actually even more beneficial because we get that information versus just putting it in a file and not sure what happens after that.
The AI Table: Mm. Yeah, it it's been a real leap forward in that regard.
Jena Zangs -U of St. Thomas: Yeah.
The AI Table: So traditional data teams, they they run nightly batch jobs, but AI agents often need real time context to act safely. How do real time pipelines fit into agentic data architecture?
Jena Zangs -U of St. Thomas: Again, tying back to we produce so much data in our organizations, but really we're using probably 20 to 30% of the key metrics on a daily basis to actually act on. And so when you are focused on that type of data and making that available, that is what's gonna be most beneficial for you as you're acting on these agents. And yes, there are nightly scripts, but if you have those data products able to be better integrated throughout the day and maybe even more cost effective because you're only working on event change management rather than whole file droppage, right? That that can be more effective for. data team going forward and again getting out of that point to point and overnight I think the efficiency will allow us to have less of that because when things go wrong we the agent could have a backup agent that's helping them fix it at that same time. I mean this whole new web of of being available in live stream is something that we've always wanted more of but we're trying to be restrictive on on resources, right, to manage all that. So going forward, I do think that we could move away from more of that nightly batch as as the systems become more accepting of this type of infrastructure, right? many of us have legacy systems, which we're finding is the hardest part to integrate and still need some of those point to point and overnight things. But I do think over time as organizations continue to offer more of these advanced solutions and adopt agentic infrastructure that will see less of that nightly batch.
The AI Table: Yes, we're we're we're moving in that direction. data restructuring is expensive and it's quiet work. How do you convince a CFO or CEO to invest in agent ready data products when they just wanna see that front end shiny app?
Jena Zangs -U of St. Thomas: I'm sure you're probably feeling this too, but I love that the word governance is having its moment, like all over. I mean seeing it on a gigantic poster, I'm like, three years ago I couldn't get anyone to talk about this or even like want to participate, right?
The AI Table: No. It's so exciting. I haven't been this excited since we became data scientists. We were acknowledged
Jena Zangs -U of St. Thomas: Yes.
The AI Table: as I was like, yeah, I like that. give me that title. Yeah, we're we're moving up in the world.
Jena Zangs -U of St. Thomas: Yes.
The AI Table: They're they're speaking our language now. Yes, I am seeing that.
Jena Zangs -U of St. Thomas: Right. So that's that is one benefit of all of the media coming at us from all those angles is these keywords that we have been saying. They have they have listened to them and now are amplifying them. So I think that at the the C suite, they are hearing those and understand, okay, they must they have more value without having to understand how deep that actually goes. And having the you know, a data strategy that is intertwined with your AI strategy. One can't live without the other and when I tie it to the individuals and the operational pieces and you know, that I like to say at the university that the people are the heart, but the data is the heartbeat, right? You can't have anything going forward without that data piece. And so as we look at getting that shiny new app, if you put something, even a Vibecode prototype with some bad data, you could show them exactly the difference between bad data and good data, which I I love vibe coding right now for these quick little prototypes that get the real effect out there. but it's it's those small things that you can show. Nothing of this will work without the foundation and the governance and the people. It it it's just the the elements that we have known since we've been trained as a data scientist to to find and that the C suite is a lot more open to that now. so it's helpful.
The AI Table: Mm-hmm. Yeah.
Jena Zangs -U of St. Thomas: I I don't know if you've seen the same thing lately.
The AI Table: I I agree. This is really our moment to shine, I think. and I agree with you with respect to demoing. That's a that's a trick that I use all the time, because I find that non data nerds, they need something tangible that they can see that they can relate to that solves their problems, makes their job easier, helps them meet whatever goals and metrics that they're looking to meet. And once you can show that visibly an example of like this is why this should matter to you and then I I keep why it matters to me to myself. But then I show them this is why it matters to you and then we can all get on the same page. and I find that that really when you use real life examples and you have something that you can demonstrate or that other teams can play with, then it really elevates the value of the data team when you're able to do things like that. so that you can be seen as collaborators because sometimes I hate to say it, particularly when it comes to governance, historically we could get trapped in like what I like to call the data police state where everyone just thinks that we're there to say you did something bad, you're not doing your data right. you know, and we don't want to be seen like that. We want to be seen as data friends is how I like to be seen. I I'm your friend, I'm here to make life easier for you and and demos are a big part of that for me.
Jena Zangs -U of St. Thomas: I'm so glad that we aligned on that. And yes, the the word empowerment was our instead a few years, I don't know, five, six years ago when we started on the data governance, that we were advised try and say something like empowerment or something. So that's what we did, and and now governance is a big thing. So we've been deciding how to shift. But I love seeing the data talent as well being pushed to evolve along with this, right? It's no longer good
The AI Table: Mm.
Jena Zangs -U of St. Thomas: enough to just be a BI reporter. You need to elevate and understand some of this infrastructure and how these types of making this data available and and figuring out how to use AI for better quality too.
The AI Table: Yes, because truly we are ambassadors is is how I think of us. We are ambassadors to the other business units, helping the organization to operate better. yeah, so don't resist that folks if you're out there listening. It's it's a part of our jobs now. Get on board.
Jena Zangs -U of St. Thomas: Yes.
The AI Table: so another question, When it comes to AI, sometimes things fail. When an agent is querying data across different units, HR, finance, customer support, et cetera, to make decisions, how do you determine who owns the underlying data product or who gets blamed if the agent fails?
Jena Zangs -U of St. Thomas: Yeah, so we've been actually working through a pretty intensive racy on on this and and making sure who's accountable and responsible for this very scenario as we're working through our own lakehouse development and I'm finding that there is a we are actually putting in a role for a data product owner which could be different the data owner always owns the data so I think when I think of your scenario you got HR finance and customer experience so those would be your data owners but if the product has multiple things because again we we picked a hybrid a data mesh architecture so that the businesses could stay closer, but as those things start to blend, we're finding that the data product owner might be one of those three, and that is who becomes accountable for what is happening in there. That's their job is to manage and understand and make sure the data is good in there, while the data owner is to make sure that they are showcasing what should be used, how are we defining the final decision maker on what who can have access, those types of things, but the data product owner becomes accountable. Now If the infrastructure that you're using, some cloud-based infrastructure or tool set, then I see that that those errors coming out of IT or whoever is managing that infrastructure side. But ultimately if the data is bad or there's a failure in something that was uploaded from the product that is feeding these data products, then I think it becomes that data owner and that data product owner too. Is that resident? yeah, go ahead.
The AI Table: When it comes yes, when when it comes to determining who the data product owner should be, is that something you let folks volunteer for? Or are you saying, Hey, I think you you fit the bill for this?
Jena Zangs -U of St. Thomas: So we actually are working with the data owners to say who is the one that understands and is the best subject matter expert in your area to to navigate some of these pieces. Who will notice when something is off by a bit? And when I think about this too with the when you're asking about the agentic work too before, you know, you can notice errors when they're like huge drastic changes. But when it's like, let's say $30 of a difference on a budget, is someone truly gonna notice that or not? You know, and so that's the type of person we're looking for who has the technique technical acutely. to understand what they're looking at and question it. And so what we
The AI Table: yeah.
Jena Zangs -U of St. Thomas: found is that the the data owners actually work together and decide who that will be. And sometimes it is two people, which isn't a bad thing either, but typically it comes down to one.
The AI Table: Yes, that that's how that's what I've seen as well.
Jena Zangs -U of St. Thomas: Yeah. Great.
The AI Table: so my last question, any data or AI, pet peeves, things you wanna get off your chest, things you wanna stop seeing, things you wanna see more of, or just anything you wanna leave the audience with.
Jena Zangs -U of St. Thomas: So I guess I would say a pet beef right now is I want I don't want people to feel so pressured that they are behind. I am really tired of seeing that all the time, of this just anxious feeling that it's creating. We you don't need to move at any specific pace other than what works for you and your learning ability. And and so I just I would love to continue to put that message out there that you just move at the pace that you can do, you know, that you feel comfortable with. And it's okay if you feel like you should be using AI more or data. And there's there's always today. It's a gift. And then I think another thing that I would love to share is that any students or any anyone who's looking to evolve their job into this role, it is only opportunity right now. Hey, it's just especially for I see all these students in education and there's a lot of turmoil going on between, you know, when is AI used for critical thinking, when isn't it, all of this pieces, but for anyone, it's just it's a builder's playground, right? There are free credits everywhere. Everyone wants you to try it. And there's just
The AI Table: yeah.
Jena Zangs -U of St. Thomas: The idea of failing at something is just is in ridiculous in my mind that you could go out and try anything that you've that idea you've had for five years, it's there. And so
The AI Table: Yeah. Yeah.
Jena Zangs -U of St. Thomas: and I love that you can you can really build that and that we're questioning what type of data we should be capturing with people and so I would just I I want you to people to just go out and be that builder and not be afraid to try.
The AI Table: Yes, that's a really great point. It's it's a great time to create and to bring your ideas to life. AI makes that a lot easier. Thank you so much, Jenna. It was it was an absolute pleasure having you on.
Jena Zangs -U of St. Thomas: Thank you. I'm just I I love the the show and I'm so thankful to be connected with you too.
The AI Table: Assam.
