Kelly Kohlleffel (00:00)

Hi folks, welcome in to the Fivetran Data Podcast. I'm Kelly Kohlleffel, your host. On this show, we bring you insights from top experts across the data community. We’ll cover AI, machine learning, enterprise data, analytics and much more. Today, I'm pleased to be joined by Ilia Xheblati. Ilia is Director of Analytics Engineering at Northeastern University. 

Ilia is leading a multi-year initiative that's improving the reliability and accessibility of data at the university with nearly a decade of experience across various roles. Ilya plays a key part in helping Northeastern make data-driven decisions, especially in critical areas like admissions and enrollment. Ilia, it’s a pleasure to have you on the show today. Welcome in. 

Ilia Xheblati (00:49)

Thank you, Kelly. It's a pleasure to be here. I'm looking forward to our conversation.

Kelly Kohlleffel (00:53)

Same here. You're focused on a lot of key areas, accessibility, data governance, all types of data challenges. Can you give a quick overview for the audience on your work and some of the challenges that you're tackling right now?

Ilia Xheblati (01:07)

Yes, of course. So as you mentioned, I'm leading a multi-year project to transform the enrollment data analytics architecture here at Northeastern enrollment management team. So one of the key areas that I'm focused in, first of all, to modernize our tool stack, the three stack that goes well together that we're trying to adopt, that we're in the process of adopting, is Fivetran, Snowflake, and dbt. So that's the first goal that we have. 

In order for us to make the improvements that we want, it's imperative for us to have access to modern data stacks, right? And what we're focusing on right now is to improve data accessibility and also ensure that we have a robust data governance. Some of the challenges I would say are, again, like adopting new technologies is a challenge in and of itself. The other thing was a challenge is Northeastern is a big university. There is a lot of stakeholders, a lot of people that you need to bring in.

And then the other thing that I think is challenging is to make sure that everything, the solution that we're bringing into the table is scalable, right? Because sometimes people, when they think about, hey, we're going to be innovative, they do a proof of concept, they do a one-off, and then they discover it's not scalable, right? So that also has been a key challenge for our team to make sure that everything that we're doing is scalable.

Kelly Kohlleffel (02:26)

Yeah, I mean, when you start tackling business processes and trying to do more with data, when you're talking about admissions and enrollment, I mean, these are core flows within any university, and especially a university the size of Northeastern. I mean, I expect that, you know, getting those stakeholders on board, you have to have everything covered, not just the technology, but how is this going to affect their day-to-day work that they do, how am going to make this easier? How did you think about that? What was the process that you went about to get these key stakeholders on board across these different business units?

Ilia Xheblati (03:04)

Yeah, so when it comes to data, everybody wants data, right? So the first question is, do you want accurate data, right? Do you want to centralize the data? Just asking the questions to make sure that everybody understands, okay, what is it that we're trying to achieve, right? And before this, many people were frustrated because at the end of the day, if you don't have the right data, if the data is not centralized, everybody can feel it, right? People are given different reports. There's going to be different numbers. And nobody is going to be happy about it, right?

So the first approach is to make sure that you understand the problem that you're trying to solve. And in this case, it's having a centralized data place where we can have all our enrollment data for the analysis purposes. There is going to be the processes. The admissions processes are complicated. There's other tools that they're using for those. But the focus that I had was, okay, what are the benefits that we can provide on the data architecture side of things, right?

And the key aspect in there, the more of the frustration that people had was that there wasn't that one place where people could go like, let's say like, I understand that the data that I want is in this place, and I can trust it, right? And then for our cases, it was Snowflake, right? Because we have so many systems that we use at Northeastern and different people will go to their own system, will be like, the data in my system says this, the data my system says this, right?

Now we're saying, hey, let's go have that central repository where we have the data warehouse and make sure that everybody is using the same data.

Kelly Kohlleffel (04:36)

Ilia to that, you've got a tremendous amount of technologies out there that you can purchase. You could always build something as well. How did you look at, and you mentioned your stack, I think right now, Snowflake, DBT, Fivetran. How did you look at those build versus buy decisions and how did you make some of those calls internally at Northeastern as you were going through the process?

Ilia Xheblati (04:58)

So there is two. I want to separate because moving data is one thing and transforming is another thing, right? So when you move the data from source to the destination, there is going to be a required skill set, right? In this case, we're talking about the data engineer, right? I was doing some of this work myself and when we were trying to scale it, I was telling my team that, I mean, like the process is not rocket science, right? Like I can teach somebody else, they can do it.

But the problem is the time, right? The people don't have the time to do it. And then the other problem is not just building it, right? But there is like the maintenance that goes into it. There is like people changing the source system all the time, right? And then you need to make sure whatever you put on the backend you change it to remove the issues and everything. 

 And then we had two alternative processes. We have the Fivetran when we did the proof of concept.

And then we had the manual process, which was very similar. And then in this case, we could see that Fivetran was so much faster, especially for the building connectors, that you don't have to think about anything. You set it up, it takes five minutes, everything goes to Snowflake, and you're like, it looks like it's magic. when people started doing the QC and understand that the data was accurate, all the pre-transformation that Fivetrain is doing to standardize the data install flag was good. Then it was an easy decision, right? Because we didn't have a data engineer dedicated to do all of these things. And we couldn't scale if we didn't use a tool, right? So in each organization is different. But in our team, we didn't have support from the IT department because to get the support from the IT department, you need to go submit a ticket request and go through all of that process. We needed to do things faster, more reliably. And we wanted the team to also collaborate and work together on these processes.

Kelly Kohlleffel (06:48)

Are you seeing some measurable outcomes, business outcomes in taking this approach that you've been able to really sink your teeth in over the last few quarters?

Ilia Xheblati (06:58)

Yeah, so there is immediate outcomes that you can see. I'll share like an example, right? So we had, we worked with a vendor. They gave us research that they did in an Excel file. And they were like, okay, this is great. This is great data, right? But we need to have it in Snowflake. And I was like, this is one time request. It's not a big deal. I'll just do it. I don't need Fivetran to use this, right? And then I was going through the process. I was getting errors and then an hour into the process so many errors and I'm like, why am I doing this? Right? I already have Fivetran access. It's not a lot of data. It's not going to cost a lot of money. Right? So what am I trying to save in here? Because I thought I saving money, but I was spending time. Right? So if you think about it, I wasn't saving anything. Right? So then I go back to Snowflake, to Fivetran, and then I click sync. And I have the data in there. And then from there, I can take this transformation. But I don't have to worry about ingesting the data. Right? So that was a huge benefit.

A lot of people that have done similar work that I did can see that benefit. The more of the long term, what we've seen is to make sure that the pipeline downtime has been reduced and it's been reduced by 90%. We don't see, there is errors like that. There's that 10% not the tool. That's the problem. There's like things that go on, things that change, right? That we need to go and adjust. But the benefit that I've seen is not having to worry about moving the data, right? The data is in Snowflake. We can go in there with confidence. We have a sync time to know when was last synced and everything will work well, and everybody will be able to collaborate, right?

Kelly Kohlleffel (08:33)

What about when you start looking at, hey, I've got opportunities to innovate. I've got security requirements at a major university. Kind of balancing some of those out, security, privacy. I'm dealing with PII types of data. How are you handling the centralization aspect that you talked about, which you're getting a lot of value out of, balancing that with the security side?

Ilia Xheblati (09:00)

I work closely with the IT department because we have experts that work. I do have security certifications, but these things are, you know, evolve very fast, very complicated. So we have dedicated resources that help with this. But from my perspective, when it comes to the data, right, Snowflake is a very, a very good way of improving the security, right? You have role-based access control in there.

There is other ways to flag sensitive data, but also we can make sure that we have compliance with security regulations like GDPR and FERPA. So it was, in a way, this kind of security was also one of the decisions that led us to say, “Okay, now we need to use Snowflake, right?” 

Because what happens is, people can go and download the CSV or people say, “Hey, can you share this data with me? Can you share that data with me?” If you don't have a framework, it's very hard to even know to audit, right? Whereas in Snowflake, right now, I can write where is I can know what data I'm sharing with who. If I need to delete data, I can delete them. Whatever historical data we have in there is in there. And everybody that needs to have access has access right now.

Kelly Kohlleffel (10:09)

Have you or your team tried out the connector SDK today?

Ilia Xheblati (10:14)

Yes. so we're getting the data through our main admission system through SFTP and batch updates once a day. And there was an increased demand on having real-time analytics. And one of the challenges, it's not on the Fivetran side because on the Fivetran side you can have syncs up to a minute, but on the the system that we're using, couldn't export more than four or five times a day. And they had a period of time where they could export the data to the SFTP server. So the only solution that we had was to use the Fivetran SDK. So when I was going through the documentation, it looks like it's a interface for developers, right, because you have more customizability, but you still are saving the time on the back end, because Fivetran is still going to figure out the things that you're used to when you build a connector. So when we had the proof of concept, we used a public API.

We went through the documentation. We were like, all right, so let's deploy this. And then there was a command to deploy it, and then it worked. And I was like, five minutes to deploy? I've never seen that before. And then the other thing that is good that Fivetran helped, the first connector, the Fivetran is offering to build it for free. So right now we're in the process working with a developer. And so far, it's been very successful. Things seem to be working great. 

Kelly Kohlleffel (11:35)

That's great. What, what about AI, ML, GenAI types of applications? Are you or your team involved on those, considering those, have anything in development right now or maybe moving towards production?

Ilia Xheblati (11:50)

Yes, So one of the things that we're working as a team, we have that we found as a team to where we can find value is to build an internal chatbot. 

There's like so many documentations. There are so many things happening. And I keep getting these questions like, hey, how can I do this? Or what's happening with this? Or what is that? Where is this? Where is that? Right? And I often laugh and I say, you need a guide about the guides right now. Because there are so many guides that you need a guide how to find those guides, right?

So what we're thinking is having access to an internal chatbot for a Q&A, even if they reduce like 80 % of the questions that I get to say, hey, here are the resources. Or for example, could be the, for example, in TBT, we have developer documentation, we have style guides and all of this. It's true, like somebody can go on Google and can search these things, right? But they will find so many different styles, so many different preferences, they'll find so many conflicts. Whereas in here it's like, hey, here's the resource that is internal to us to make sure that everybody's on the same page.

Kelly Kohlleffel (12:51)

I really like that. think that chat, that Q&A chatbot kind of that instant feedback and direction is something that, especially for organizations, if you haven't jumped into this space yet is a great place to start. You know, you talked about these foundational technologies. I can keep everything secure. can, I can take my private enterprise data or, or my, the way that I approach maybe the way that I style something, as you said, I can make that very specific to me when that answer comes back versus, you know, getting a list of 10 or 20 or a hundred things I've got to sift through. So I love that idea. 

Let me ask you this, Ilia, because I'd love just some personal perspectives from you. over the next 12 months or so, are there any key trends you have your  eye on that you see that are starting to emerge in this data space that maybe you're not even testing out right now, but you want to try to take advantage of or start testing out over the next 12 months?

Ilia Xheblati (13:54)

Yes, of course. Of course, like, GenAI is going to be the top priority, I think, for many years to come. But with that comes more of, okay, the privacy, right? You need to think about, if I'm using GenAI, what's the bias in there? Like, are we following the best practices when it comes to privacy, right? And then the other thing that I'm more interested on is the real-time data observability.

Because right now there is like so much data moving right and then now you're saying okay Fivetean is moving the data or like I trust Fivetran, but there's like so much things happening so much faster that we need a way to have more observability on our data. And I think that's going to be having like real-time data observability is going to be a key factor when it comes to 2025 or later.

Kelly Kohlleffel (14:41)

Can you describe or at least high level kind of a use case for that? Just be interested to hear what that flow is around that real-time data observability.

Ilia Xheblati (14:50)

Yeah, so for right now, we have data from the admission system. Then there is, we combine it with many other data in the system. Like we have student visas. We have the enrollments and applications, financial aid, all these other things. And just to have it in a way where we can see how everything is being transformed is very powerful right now. But also to make sure that everybody's on the same page because sometimes one of the challenges things, especially for the downstream analyst, is that, I don't know what you're transforming, right? So they need to have that kind of observability. But also it comes like you can have the SQL, which not everyone understands, but from the user's perspective, they also need to know where the data is coming from, right? You have this dashboard that you're getting the aggregates. But just to understand where things are coming from is very, it's good and powerful as well.

Kelly Kohlleffel (15:50)

Let me also ask you this, when you look at what you've done and what you and your teams have done together around transformations within your organization, within your institution, are there any qualities that you want to call out that you believe are crucial to guiding a team through this type of transformation?

Ilia Xheblati (16:11)

Yes, I would say setting a vision is very important. Having very good communication skills is very important because you're going to communicate with technical user, business users, and then other users, right, of users that have different lenses, different aspects. So you have to make sure that you know your audience and how to communicate with everyone to avoid any miscommunication, right? So that's very important. The other thing is things move so fast. So also adaptability is very important. 

And then again, you need to foster a collaboration environment where everybody is working together towards the vision that was set, right? Because that's the only way to move forward. Otherwise everybody will focus on this, I'm working on this project, this is mine, you are doing this, he's doing that. And then people are not working together very well.

The other thing that I would say is when you are in the data world, right? You usually don't think about the software development because you're like, like they're built on software. But if you think about it, right, in a way, you're also building a product, you're building a data product. So what I would say is to adopt the agile processes of the software development in the data ops. I know it's kind of the world that everybody's using right now in your data operations is very important because there's like so many good processes in the, that have proven to work in the software development, then you can bring back to the data operations.

Kelly Kohlleffel (17:39)

Great points. How often do you go back and maybe have to reset or kind of adapt that vision based on changes within your institution, changes with technology, changes with a processor, downstream requirements.

Ilia Xheblati (17:53)

Very often because there is unknown unknowns, right? Things that you don't know that come up and you just have to adjust. But again, so when I set the goals, or like the vision is not set in stone. It's more of like a living thing that you always go back. When you have the agile processes, right, it gives you the option for you to change things quickly, right? As long as you don't go too far off the vision, you're fine. But at end of the day, there's like so many things that you need to balance that you have to be adaptable.

Kelly Kohlleffel (18:24)

Yeah. And, and then you talked about collaboration and communication too. That feeds right in with that vision and that adaptability. And if you're, I think communicating and collaborating with those extended teams all the way through some of those adaptations that you have to make become a little bit easier. Everybody understands. And it's not a shock to the system, so to speak.

Ilia Xheblati (18:47)

Yep, for sure.

Kelly Kohlleffel (18:48)

Really good. It was a lot of fun talking to you as well. Great to see you again. I really, really appreciate you joining the show today.

Ilia Xheblati (18:56)

Thank you for having me. It was a pleasure talking to you.

Kelly Kohlleffel (18:58)

I'm going to do my best also to keep up with everything you're doing at Northeastern. I want to, I want to hear how that vision setting early in the year is going as we progress. 

For everybody who listened in today, thank you so much. We appreciate each one of you. We would encourage you to subscribe to the podcast, any of the major platforms, Spotify, Apple, Google. You can also find us on YouTube, if you want the video version of this. Visit us also at Fivetran.com/podcast.

Send us any feedback or comments at podcast@fivetran.com, and we'd love to hear from you. We'll see you soon. Take care.