Is Your Campus Data AI-Ready?

0:00:12.7 Speaker 1: Hello and welcome to Office Hours with EAB. Today, we tackle a familiar headache across higher ed: how to access critical data stuck in disconnected campus IT systems. Our guests share what’s finally working to bring it all together and where AI can accelerate progress without sacrificing privacy or trust. So give these folks a listen and enjoy.

0:00:41.7 Christine Doci: Hello and welcome to Office Hours with EAB. My name is Christine Doci. I’m a senior director here, and my job is to help university leaders evaluate technology solutions that will serve their colleges and universities, and to play nice with their people and their processes. If you work at a college or university, even if you’re not in IT, you probably have at least a vague sense of the struggle that higher ed is facing right now as we think about scattered data across a patchwork of different information systems managed by disparate departments. To be fair, colleges have made progress in this area in the past few years, but many still struggle with fragmented data and systems. And the good news is, AI has the potential to make this chore more manageable. The reality check, though, is that to take advantage of AI automation, you first need to find an effective data management platform in place that’s truly AI-ready. Joining me today on our podcast is a colleague of mine who understands all this stuff and who’s going to do his level best to help you understand it too. Lars, would you mind telling the folks who you are and what you do here at EAB?

0:01:47.2 Lars Waldo: Thanks for that intro, Christine. I’m Lars Waldo, and I’m a managing director here at EAB. I am currently the head of product on our Edify data management platform, and my background is actually in data science. So I’ve been doing AI and machine learning work here at EAB for a little over a decade, and now I work with our data management platform, helping to deal with all the things that you just talked about, silo busting, getting fragmented data all into one place so that it’s usable. And historically, when we thought about making data usable on campus, we were mostly talking about usability for humans. But now that we have this new generative AI technology that can use data the way that humans do in some ways, we’re now talking about usability both for humans and for AI. So I’m really excited for the conversation we’re gonna have today.

0:02:41.7 Christine Doci: I love that background. Just the fact that you’ve already been in this kind of machine learning world for a decade is a good reminder, right, to us lay people out here that AI didn’t just spring up overnight, and there is a deep bench of experts who’ve really been kind of in the trenches with AI for quite some time. So speaking of those experts, Lars, I read the paper that you and our research team kind of pulled together recently. It was a playbook for higher ed leaders to help them get their data infrastructure AI-ready. So let’s take a few steps back and level set here. What are some of the core data management challenges that you and the research team kind of unearthed in that? What are universities facing in terms of challenges they’ve been trying to solve over this past decade or so? And why does that feel so hard to grasp, so elusive right now?

0:03:38.3 Lars Waldo: Yeah, so I would say that the biggest data management challenge that universities have been dealing with over the past decade is data access and specifically the silo busting aspect of data access. Back in the sort of the aughts, there was a huge proliferation of data systems across campus where all of these core campus processes were able to have software solutions. And so campuses adopted them and got all these data systems up and running. Then, once campus leaders realized how much data was being generated on campus and the value of the insights that you could get from that data, they realized how much of a benefit it would be to have a data warehouse where you had a centralized location to access that data. But of course, getting access to all of those different systems is much easier said than done. It’s both a technical challenge where you need to actually mechanically get the data from systems A, B, and C altogether in one place. But you also need to be dealing with potentially different system owners and security risks as well. And so going around to campus to all of your data system owners and your IT department and making sure that everybody is on board with putting in the effort to take the data out of all these systems and put them in one location can be a pretty big challenge because you need to get everybody to work together and put in a lot of effort to get the data all in one place.

0:05:16.3 Christine Doci: Yeah, it’s so interesting. As you were talking about kind of this just proliferation of tech systems, I often talk with campus leaders that describe almost a fork in the road. Some campuses go for more of a best-in-breed approach to technology, right, where they’re getting a ton of point solutions, but those solutions and platforms are really operating at the top of the order. Then we’ve got another camp of leaders that go with more of a monolithic approach or the “one system to rule them all”. And we know there’s pros and cons to both approaches. What I heard from you there is you talked about those data owners on campus, and really stewarding data management and getting step one going with data governance. So I wanna go there and talk about what I often hear from colleges is this concept of, yeah, we wish we had data stewards, but sometimes we have data hoarders. The data is mine, and there’s like Gollum-esque kind of approach. Why do you think so many early data warehouse initiatives have stalled out? What are some of the factors there that play into that maybe lackluster performance in data warehousing?

0:06:37.0 Lars Waldo: Yeah, so I think that there are two big reasons that early data warehouse initiatives stall out. And now this is a case where every university’s data ecosystem is pretty different. There are a lot of options for every kind of system you might have. There are different SISs out there, LMS, CRM. There are a ton of different systems. And every implementation of every system is also different. The banner implementation at one school might be different from another. And so there are really, there are no two institutions that have the same data ecosystem, nor two institutions that have the same governance structure. And so every institution is very unique in terms of what they need to overcome in order to start data warehousing. But there are two pretty common themes that I think stall out these initiatives. The first is just the level of effort. Engineers are very notorious for underestimating the amount of work that it takes to build a complicated software system.

0:07:46.9 Christine Doci: Planning fallacy.

[laughter]

0:07:50.4 Lars Waldo: And so a lot of the time when you decide to build a data warehouse, you think about, okay, what are the steps that I need to take? What are the things that I need to do? And how long is this gonna take? And it just ends up being a lot more work than you expect. There are hurdles you need to overcome that you didn’t expect. Somebody finds out about the project a few months in and wants to change the scope of what you’re doing. And just, you might have some critical staff turnover during that time and have to backfill a role and get someone up to speed. There are just a lot of things that can happen to throw you a curveball while you’re trying to build this big system. That makes it more work than you expect. And that can be something that ends up turning into a quagmire and derailing the project. Another problem that you talked about is that silo busting from a human perspective, where you used the word data hoarders instead of data owners or data stewards. And that is really what it feels like a lot of the time. It feels like people are territorial about their data. They don’t want to grant access.

0:09:05.4 Lars Waldo: It feels to people outside of the ownership of a data system like the owner might be protecting their turf by not sharing data. But I don’t think that’s really what’s going on here because I actually don’t think that these data owners are as territorial and as hoarding as they may appear from the outside, because I think what it really is is a data governance and leadership challenge where to outsiders, it can feel like a data owner is hoarding their data. But if you don’t have good data governance structures in place that control who is allowed to access data, what are acceptable uses of data, if there aren’t clear guidelines here, then the data owner is just doing their job by being restrictive about data access. If there isn’t a well-understood decision-making structure about where data lives, who gets access, how do we control permissions, how do we document where the data is going. If you don’t have those things in place, then the responsible thing to do is to be careful about where you send the data and who you give access to the data to. And so, I think in many cases that the territoriality is actually a rational and responsible response to a lack of data governance.

0:10:41.4 Lars Waldo: And I think it can also be a leadership challenge where data warehousing and integrating data systems is such a technical issue. It just does not feel like a cabinet-level issue. It feels like something where, like, oh, why can’t the really technical software people in IT just get together and make this happen? Well, without the data governance in place, which needs to come from a pretty high level on campus, and without the leadership around picking what the solution is gonna be and deciding that everybody needs to move towards it together, it’s really hard to self-organize.

0:11:26.6 Christine Doci: Yeah, I heard a couple of themes there. So one was, you’re challenged to think about the symptom versus the cause, right? And sometimes what we might be able to observe and think it’s tied to more of that data hoarding or irrational control over data, that could actually just be a symptom of some other broken pieces in data governance, versus the thing that is causing data governance to break down. The other thing you said about the talent crunch and staff capacity, I think I wanna go there next, because that’s something I hear constantly with campus leaders thinking about not just how do I keep the talent that I have, but, “Hey, Christine, we’re dealing with rifts or reorgs. We’re really stressed about how to really keep our current talent engaged in this work.” And so I think about other industry podcasts, just to give a couple of quick shout-outs. I listened to one by HigherEdJobs. I also regularly listen to Educause’s The Integrative CIO. And there is such a capacity constraint that’s gripping higher ed right now that exacerbates some of the challenges you were talking about.

0:12:48.3 Christine Doci: Maybe let’s have you elaborate a bit, what other shortcomings and stumbling blocks do you see that are making this feel like such a drag and causing these knock-on effects as we try to move into 21st-century data management?

0:13:03.9 Lars Waldo: Yeah, so I have a few thoughts here. One is that the talent crunch is very real, and that can be a difficult problem to solve because there’s a lot of demand for technical work in higher ed. And there’s a set of people who work in that field, and it can just be hard to attract and retain them. But I think that this is actually a big area where, if you can get yourself into a virtuous cycle with AI, that can help relieve the talent crunch quite a lot. AI is really quite good at a lot of technical tasks, like writing code, writing SQL, explaining how to use, for example, cloud resources if your IT team manages AWS or Google Cloud Platform, or Microsoft Azure. AI can actually help you navigate the management of that pretty quickly. It’s like an expert that knows all the documentation, and it can write a lot of… Right now, I would say more basic code, but it can write it very quickly and effectively.

0:14:26.3 Lars Waldo: And so I think that AI really has the potential to relieve the talent crunch quite a bit, but you need to get your data and systems into a place where you can actually start effectively leveraging AI in order to get into that virtuous cycle where, let’s say that you have an IT team that is just constantly underwater with requests for ad hoc data pulls. If they are constantly underwater with requests for ad hoc data pulls, then they’re not gonna have time to centralize the data and put an AI tool on top of it that can help people access the data more easily without needing to go to the IT team. But if they can just get ahead of the ball for long enough to get the data into a place that you can put AI tools on top of it, then all of a sudden it becomes much easier for a larger group of people on campus to just get direct access to that data without having to go through the IT team. That frees up their time for more building infrastructure and tools that people across campus can use, rather than acting as the infrastructure and tools themselves, because they’re just experts that know how to do it.

0:15:45.6 Christine Doci: Well, so to be fair, I’m hearing some, maybe some assumptions there, Lars, that you’re making it sound a little bit easier than it might be to get started on this journey. So…

0:15:56.5 Lars Waldo: It’s not easy.

0:15:58.6 Christine Doci: So let’s get tactical for our audience. What we’re talking about, it feels like this quantum leap forward into incorporating AI into efforts. What does this look like? What are some first steps or baby steps to get our institutions to, again, take that initial step in this journey to AI adoption and acceleration?

0:16:21.8 Lars Waldo: Yeah. Well, I really have two answers for that question. One is really a cultural answer, and then the other one is a more nuts-and-bolts technical answer. So I’ll start off with the cultural answer, and then I’ll talk a little bit about the nuts and bolts technical stuff. So I think the most important way to take a first step on the quantum leap to sort of mixed metaphors a little bit, I guess, is really…

0:16:56.5 Christine Doci: You said quantum, you said quagmire, let’s just keep it going. Why quit while you’re ahead?

0:17:01.9 Lars Waldo: Yeah. So I think that the most important first step to take is really build a culture and belief around AI being an important part of the future of technology on campus. I think people are in all kinds of different places personally in terms of how familiar they are with AI, how comfortable they are with AI, and where they see AI coming in the future. What I’ve noticed from myself, my colleagues, my personal friends around AI is that there tends to just be a light bulb moment for each person where they see AI do something new. And that thing, whatever it is, makes the light bulb go off. And they’re just like, “Oh my gosh, the world has changed.” I remember for me personally, it was the first time I saw AI write really good code, which was when GPT-4 came out. I think in March 2023 would have been when I first saw it write really, really good executable code. And it wasn’t perfect. It was just writing simple functions. Some of them didn’t work the first time. You had to feed the error back into the chat, and it would give you an improved version of that function that didn’t bug out on that particular error. But at that moment, I felt like I could see the future where I was like, “Oh my gosh, this needs a little bit of ironing out. But once it gets ironed out, and once you can plug it straight into your code base, this is gonna be able to do incredible things.”

0:18:46.3 Lars Waldo: I felt like I was there plowing a field behind an ox, like I’d done my entire life. I just saw a tractor drive by for the first time. It might be a little bit rickety. It might break down from time to time. But once you see that tractor drive by, you’re like, “Okay, this is the new world that we’re in.” And sure enough, now, companies like Google and Microsoft, who are some of the bigger, heavier, early AI adopters, they’re now saying that 20% to 30% of their new code is written by AI. And remember that this is only two years after GPT-4 came out, which was the first AI that was really good at writing code. Because we talk about AI so much, it can be easy to forget what a short period of time it has been since generative AI first came out. But nobody had heard of ChatGPT or AI outside of a small group in technical circles just three years ago. It was really in the fall of 2022 when ChatGPT 3.5 started raising some eyebrows. And then the spring of 2023, when GPT-4 really made substantial strides forward.

0:20:05.2 Lars Waldo: That’s how recently these technologies really got there. And you’re already seeing such a large share of code being written by these big tech companies a mere two years later. It’s all about getting people to that point where they sort of see that future for themselves. And so if you’re someone who hasn’t had that light bulb moment, I really encourage you to just use some AI every day and just ask it to help you with whatever your most pressing problem is or whatever your boss’s most pressing problem. Try to encourage your colleagues to use it. There’s all kinds of things that you can use it for, and different people get value from it in different ways. It could be writing code. It could be reading and summarizing your emails. It could be helping you draft a memo that you’re trying to write. It’s a different thing for everybody. But once that light bulb kind of goes off with the person, it really turns them into a convert, and they’re ready to hit the gas on AI adoption. And so this ended up being a long answer to my first of two points. But I think that people and culture thing is the really critical element that you need to get through in order to take your first steps on the AI journey.

0:21:34.0 Christine Doci: Yeah, it’s such a good point. You talked about culture, and what I’m hearing is almost there needs to be this communal approach to AI, right? The more you see others experiencing it and benefiting from it, there are ripple effects that can spread across the organization. So I know you wanted to talk about the technical side then, and what some of those first steps look like. Maybe let’s take our audience there. When we think about what this could look like from a technical perspective, if we’re even talking about specific platforms and what those may look like or feel like, what do you see as some of the key differentiators between those effective tools and platforms that are gonna help us solve for data governance, AI adoption? We really want to move beyond that world of black box, and make sure that we are acknowledging and respecting the culture of each unique organization, but knowing they’re all unique with their integration, data access, cybersecurity needs. Talk to me about the tools themselves and the more technical side.

0:22:46.5 Lars Waldo: Yeah, so from the technical side, it can be summed up in a single sentence, which is that in order to work for you effectively and be customized to your institution, AI needs access to the right context. Now, context, when you’re talking about AI, is the information that you are feeding the language model, along with your prompt that gives it information about the question it’s trying to answer. So when we’re talking about a data warehouse and data management generally, context includes things like documentation of your data. So what tables do you have? What fields do you have? What are the meanings of these fields so that the AI knows if it’s trying to write a SQL statement, it knows what tables to reference, what fields to use. It’ll understand, is this a field that can be summed up and grouped by one of its key fields? Or is this, like, you would never want to ask for a sum of student IDs. That wouldn’t make sense. Even though they might just be numbers. And so that’s an example of what the right context means.

0:24:08.5 Lars Waldo: But context can mean a lot more than just the written documentation about the meaning of fields. It can also be examples of SQL that you’ve saved elsewhere in your data management platform that tells it what fields to join on or what tables go with what other tables. So I don’t want to get too technical, but the big picture is really that in order to integrate AI with your data, you need to provide a context. And that is mostly stuff like data governance and your transformation logic. And really the aspects of transparency that you want to have anyways, because your human users need them. That’s an interesting thing about AI is that since generative AI is more human-like than other code that you might have around campus, its needs are actually pretty similar to human needs out of a data management platform. And so that makes things like just transparency a really important aspect of a data management platform. It helps if you only have human users, because humans need to understand what does my transformation logic look like? Where’s my documentation on what data I have and what it means?

0:25:42.0 Lars Waldo: These are all important things for human users, but that’s the secret sauce for AI if you want AI to be helping you with data access. It needs that same transparency and governance as humans do.

0:25:57.3 Christine Doci: It’s a great point. I was thinking through some of the conversations I’ve been having recently. And I hear comments like AI is forcing us to ask better questions. To build on your concept of high context communication, I think our campus leaders are quickly figuring out, our documentation isn’t up to snuff, or it’s not serving our institution to have data living in so many different systems, silos, spreadsheets, hard drives. I think this data governance journey for a lot right now, and AI wrapped therein feels a lot like two steps forward, one step back, or sometimes even one step forward, two steps back. You talked about generative AI, though, and I wanna make sure that we are sharing a glimmer, a sparkle of hope in this, because this is hard work, but it’s worthwhile. How are you seeing some colleges move towards this implementation of platforms that leverage AI? Maybe a mix of that kind of culture and technical piece would be helpful for our audience to understand what does it look like to leverage generative AI?

0:27:13.3 Lars Waldo: Yeah, so the first thing I want to say is that we’re still really in the early days of AI adoption for use on something like a data management platform. We’re still, I would say, in a window of time where just using ChatGPT or Quad or one of these other AI models, just chatting with it to help you with your work, is still kind of an early adopter behavior. Now, we’re getting to the point where a lot of institutions have started socializing the notion that, like, “Hey, you know, AI can be valuable for a lot of the things that you do throughout the day, so maybe you should be using it.” I know that at EAB, we all have access to, all of our employees have access to AI tools to help with our jobs where appropriate, and we’re working on finding better and more specialized ones for some of the more technical tasks. And we see a similar dynamic playing out in the institutions that we work with, where two years ago, nobody was using AI at work, and there were a lot of concerns around things like privacy, security, what’s an appropriate thing to use generative AI for.

0:28:40.3 Lars Waldo: A lot of the early adopters are starting to figure out answers to some of those questions and become more comfortable using AI in the regular work, but that is only the first step. The next step is to start leveraging AI directly with your technology, and there, we do see some of our partners starting to leverage AI in a data management platform context. We’re talking very, very early adopters here. There is still time to be an early adopter of AI in data management, but some of the really cool use cases that we’ve seen early on have been helping build your documentation out with generative AI. I’ve talked a lot about how AI needs documentation. Well, AI can help write that documentation for you. We’ve seen partners who have brought in a new data system into their data warehouse and documented 5- to 10,000 new data fields almost instantaneously with generative AI. And they go through and review the descriptions that the AI writes, and they find that 99 plus percent of those descriptions are spot on. And so then you have this awesome data dictionary that was written mostly by AI that then the AI can reference when it’s trying to build a SQL query or something for you.

0:30:13.0 Lars Waldo: And that’s the other big use case that we’ve started to see some of our really bleeding-edge partners do is use AI to write SQL that they would use for transformation of data, for building reports, just for ad hoc data pulls. Generative AI is really quite good at writing SQL as long as it has that right context. And so as our partners are building up better and better libraries of context and getting more comfortable with generative AI, the AI’s ability to help you with the actual data management infrastructure is getting stronger and stronger.

0:30:53.3 Christine Doci: Yeah, and as we wrap up here, I think what we’d summarize is with AI, you just have to take a page out of Nike’s book and just do it, right? Give it a whirl…

0:31:06.2 Lars Waldo: Absolutely.

0:31:06.4 Christine Doci: Give it a try. I love that as a homework item, as a next step, make it communal. Talk about the culture around data stewardship and data ownership on campus, and just do it. Give it a go. This is such an exciting time, I think, to be in the higher ed industry, and schools look to us to help them innovate and to help them think about not just today’s challenges and problems, but what aren’t we even thinking about around the corner in 5, 10, 15, 20 years? That to me is what’s really exciting about AI is helping our schools really supercharge their initiatives and bring bigger impact to their communities. Any final thoughts that you have, Lars? Any advice that we can offer our higher ed leaders when they’re ready to take those concrete next steps, they want to maximize campus data? Let’s end on a high note.

0:32:04.1 Lars Waldo: Yeah, so the high note is that I’m really optimistic about what this technology is gonna do for the future of data use and access on campus. I think that the future of data access on campus is you ask your AI a question, and it just gives you the answer, and you trust it because it’s proven itself to be reliable and accurate. Now, we’re not there today, but just think how far we’ve come in the last two years, and think forward two years from now. You’ll probably just be able to ask your AI a question about your data, and it’ll know what you have permission to access, what you don’t, take a look at the data for you, and give you the answer you’re looking for. So it’s gonna be like everybody has a data expert right there with them all the time, and for most users, the technical aspects of accessing data are gonna go away. And for the IT team, it’s gonna be more a challenge of making sure that you have that infrastructure in place to give everyone their AI assistant rather than be doing all these ad hoc data pulls.

0:33:16.5 Lars Waldo: And so if you want to move forward towards that future as an institution, like I said, that cultural component is so important, where if everybody on your campus can see that future of you just ask a question and get an answer, and believes that generative AI is gonna be able to do that, it’s gonna be able to get you there, then everybody is gonna want to go on that journey as fast as they can. And it’s gonna be easy to coordinate people and deal with all of the tricky organizational shepherding that we need to do if we want to actually implement new technologies on campus and do things across silos. And so that cultural thing is really the big important step. And if you’re ready to go even further, just talk to your IT team and try to get an AI tool on top of some of your data. And it might not work perfectly at first. They’ll probably need some ironing out, might need a little bit of work and tinkering to get it to give you that correct answer. But before you know it, we’ll be two years down the road, and you’ll just have that direct line access to your data. And I think that’s a really exciting future.

0:34:36.8 Christine Doci: I love it. All right. Get what you need is what I heard. So thank you so much, Lars, for your time. This was a pleasure. And again, thanks to everyone who tuned in to EAB Office Hours. Until next time. Appreciate y’all. [Music]