Transformative Al & Technology for Decisionmaking Part 2

Welcome to “HSDF THE PODCAST,” a collection of policy discussions on government technology and homeland security brought to you by the Homeland Security and Defense Forum 

In this episode we take you inside the shift from siloed spreadsheets to governed data products, and show how that foundation lets AI cut hours from planning and reporting without replacing human judgment. Along the way, we break down the hidden process gaps that look like “bad data,” the cultural resistance that slows sharing, and the simple moves—like quality trend monitoring and clear ownership—that unlock momentum.

 Featuring:

- CDR Jonathan White, Cloud and Data Branch Chief, U.S. Coast Guard 

- Carin Quiroga, Chief Data Officer, Immigration and Customs Enforcement

- Courtney Whelan-Stillmun, Principal Architect, Google Public Sector (moderator)

This discussion took place January 22nd, 2026, at HSDF’s Technology Innovation in Government Symposium

Follow  HSDF THE PODCAST and never miss latest insider talk on government technology, innovation, and security.  Visit the HSDF YouTube channel to view hours of insightful policy discussion.  For more information about the Homeland Security & Defense Forum (HSDF), visit hsdf.org.

Transcript

SPEAKER_01 0:01

So kind of going back to some of the discussion that we had around data specifically, AI decision making is only as strong as the data behind it. And we hit on this a little bit, but what have been some of the hardest challenges in ensuring data quality timelines and relevance across legacy systems?

Culture, Silos, And Timeliness

Process Gaps Masquerading As Bad Data

SPEAKER_02 0:20

I can take this one. So for the first part. So I would say it's interesting because I think a lot of the data problems we've had in the past are all still there. Culture, data silos, people sharing data and whatnot and holding onto it too tight, and the quality of data and timeliness. But where I love where the shift we're going with AI is it's helping push that forward. It's showing people that we need to do that. And so yes, data is still number one over everything, but it's an important, like AI's helped propel it forward. So I would say, like, for me, it's interesting because, you know, one minute, we're talking about the development team, and in my head, I was like on a week, like, oh my gosh, we need them to be more disciplined with their data practices and better quality. And then I talk to a development team and they're like, no, it's not that at all. And and you realize like you're kind of peeling layers of an onion. Like you think it's a data quality issue, and then sometimes it's a business process issue where it's like, no, we don't have bad data. It's just there's no process in the pipeline to tell us that the data has changed. Like TDY. I'm dealing a lot with HR data right now, if you didn't notice. But like if somebody goes on TDY, there's not really a mandatory requirement to make that. So it's hard to know where people are, what roles they are, are they capstone? And until you work and partner with the business to understand that and also educate them to tell them why it's important, why it's important, and let's let's help peel this onion together, right? So that they understand that we can make their job faster, we can make decisions as an organization better, and then meet them where they're at. So, like say, okay, you're having challenges, let me help you with the data, let's and get tools. Like, I will say that and flat out. Like, we need tools to also help us do that. And so get like quality tools so we can help them measure their quality, um, see how it's trending over time. If there's anonymities or we're seeing it's going down, let's work with them to identify why it's going down and to get it better. And then with that, we can help them to get their data better, in better state, implementing processes, and at the same time, we're getting more mission value for other groups that need to use it. And so I would say it's still like the culture data quality issue. And then I think tooling and having that environment. I mean, we both talked about um the data architecture and the fabric and having the tools and API gateways to make the data more readily available. Um, I know some of you in the past, like I've always talked about the data bodega, right? And it was supposed to be that data marketplace that you can get our data products, but and it was great and it was an idea, but because it was just like a nice idea and it was something, it helped us move forward, but it didn't help us get there. And so now, um, and I mentioned earlier, and you guys might have heard him earlier, Dustin Getz, our CIO, he's really um, really an advocate for data and implementing our data architecture of fabric. And so, with that, we're gonna also mandate and put in reviews, like we need to have these data products readily readily available. It's gonna be in everyone's review, making sure that we have the quality and it's not just the data, but it's in a fit to use. So it's clean, it's ready. Um, we have the permissioning and the things around it, and like like who the zero trust built in. So things like that I think are gonna help us enable, but it's been a challenge and it's been hard, but I I think it's so hopeful for us to get there.

Tools, Metrics, And Data Fabric

Mandating Data Products And Zero Trust

Force Design 2028 And Data As Asset

SPEAKER_00 3:22

Yes. Um, so last year the Coast Guard issued the Force Design 2028 Executive Action Plan and the summary uh document. I was uh a key writer in the on the technology side of that document. Um and so I was taking my two years of data fabric, data mesh, and and uh cloud and platform experience and just sort of dumping it into that document and saying, you know, here's how I really want to see what the if if you were to say new Coast Guard technology, this is what we want to ascribe to, right? We have three years, let's go. And the key, there's some key action items in there regarding data and AI, right? One is to use data as a foundational asset, an operational asset for all mission threads. Um, use data to enhance the Coast Guard experience, my experience, right? The journey through uh from commissioning or from enlistment all the way through retirement and beyond, right? Uh it's to use AI to shorten those chains, right? Be able to uh make the mission more engaging and exciting, all right? As a lot of what we do is pre- and post-mission work, right? You're writing a mission plan, you're collecting data, you're figuring out what to do, you're trying to convince someone to do something, right? These are not zero cost missions, they're actually quite deadly, right? Search and rescue is a deadly endeavor for the rescuers and searchers. So you have to you have to have a good foundation with which to say launch that helicopter or launch that aircraft, right? And then on the other side, you have a report. You have to document what happened, you have to uh show what you've used in terms of asset fuel or time or whatever it is, right? So what if you could put AI in both of those bookends so that it could help you shorten the time for planning and then making that decision on go no go, which hopefully saves more lives, right, in that in that mission context. And then on the other side, it's recording everything you've done in that mission because we're collecting this data continuously and feeding it into a centralized system. And then the report on the other end is really just a a validation, right? You're looking at it and say, Yep, totally got it, right? Almost like the transcript from a from a Teams meeting. It's always correct, and that's you gotta look at that. Right? We're not quite there yet where it's just like rubber stamp that, right? But could you say, I've just taken six hours off your plate and turned that into one hour? Now you can get back on mission that much faster, right? Or I need less people focused on a mission, I can broaden my mission space even further, right? So there's there's some knock-on effects to getting the data piece right and then using that in practice. Um and so what we focused on quite early, to Corinne's point, is personal readiness and HR data. That was what uh our office of data analytics did for a year. And I thankfully they did that, right? Because we didn't know Force Design 28 was coming at that point. Um, and so when we arrived, and readiness was a key topic area, and hiring 15,000 more coasties in three years is a key topic area, training them and doing all that work. How do you even start tracking and dealing with that? How do you know what your situation is? Well, we know we have the data products, we built it, we have the quality built in. They're actually building second order, third order products off of those, right? To really dive into the very specific aspects of that journey. And they're we're answering the hard questions we could never answer before because nobody put the legwork in to actually tease that information out of the source data. So it's really, really, really exciting. You know, you're starting to see the fruit bear from the tree, uh, which is sort of makes me really happy because I'm like, that told you so, right? Like those we're going to see three years ago and it's coming to fruition. It's really exciting to see that.

SPEAKER_01 7:09

That's awesome. That's incredibly exciting. Um, okay, so as chief data officers, you've seen the evolution for storing data to leveraging data to provide proactive insights and decision making at speed and scale. A lot of what you've just addressed already. But what lessons have you learned in deploying predictive and analytics across use cases within your organization? And can you share any best practices that you've picked up along the way?

AI At Mission Bookends

SPEAKER_02 7:36

Yeah, I think a lot of it, you know, for me goes back to data, and I hit on a lot of it, but it's like going back to the data quality, um, having the trust and having the relationships. Like those are our best practices, is like building those relationships with the bishop mission space, finding critical um wins that you'll have and operate critical like operation needs, but also ones that will be successful to help you evolve. I mean, in the past, like data was, you know, it was storage, it was there to use and do reporting and dashboarding. And now it's not only like learning about the quality, but like real time, right? We need to get timeliness. So, you know, and I'm not gonna say it's perfect because there's, you know, we still live in the world of spreadsheets and people giving here's our data, and it's in a Excel spreadsheet, and every week they're uploading it to a site, but we're really breaking through that and saying, okay, we need to get data from the source or from you know the service catalog where we can get the data. Um, because timely data is critical for those AI decision makings. Can't be the old static data that you know we used to do dashboards off of.

Readiness And HR Data Wins

Lessons From Predictive Analytics

SPEAKER_00 8:35

Uh yeah, so we we did the six months of analytics um bliss, I would say, right? It's it's it's wonderful. But you're taking this original product that somebody made at some point and did a lot of people trust it, right? And you you actually peel that onion layer back, and you're like, maybe you shouldn't have trusted that so much, right? There's some weird stuff going on here, right? And and so this one-to-one transference of what you had today to tomorrow really should not happen, right? You you really shouldn't take take it as your inspiration, but don't take it as your as your rulebook, right? And so what we we had to what we did is we built um we did a use case inventory around our our BI. Um we came up with about, I think it was about 40 use cases, and underneath those use cases were uh analytic products or data products that would support those use cases. And that's how we structured the the work breakdown. And then we hired um a bunch of contract teams to come in and and augment our our uh nascent uh data teams, right? So we're also I mean it's simultaneous to all of this happening, we're trying to build the future of what our data organization would look like, um, which is disparate data teams centered around uh communities or or um um pillars of of uh data ownership, right? Um so we we have I think 12 uh data domains, right? We are establishing data stewards and data owners, and underneath them are are your actual data scientists and engineers and people that that need to really do the work, and there's domain expertise required there. This is one of the problems, is that you can't hire out expertise, right? We can Coast Guard is pretty small. We can probably find some ex-coasties that might have some knowledge of this stuff and bring them back in, right? But the the challenge is that that might be aged knowledge, right? It might not be information that's that's prescient, right? It's right at the timeliness. So finding those domain experts and actually empowering them and tasking them, because they're probably doing something else, right? Tasking them to provide their their knowledge to the people doing the work is so hard, right? That's the hardest part about this whole journey. But once you unlock that door and do a couple, you do a couple of I know. Well, I'll just talk loud. So there you go. So you do it, you do a couple of the um of the sequences, right? You're going through the flow and you're delivering products, and then all of a sudden this excitement builds. And that's the phase we're at right now. So people are you, you know, you kind of have to like lead, you have to lead out front, do the work, fight off all the people that are that are sort of naysaying you, then deliver the product, stand back, let it, let the organization absorb it, right? Because it's new. Everybody hates new stuff, right? So change management is also key here, right? So you got to get ahead of that with communication, uh, with uh preparatory information, and then stand back, let the organization give you their feedback, and then attack it again, right? Again, this is a very software-centric viewpoint, right? Keep iterating that product, it's never done. I think we're on the third iteration of some of our analytics right now, right? We have everyone, we have a feedback button on the top right. I submitted some feedback myself. I was like, I don't like this, I couldn't figure out what the heck's going on. If I can't figure it out, that guy can't figure it out, right? Being really mean about it. And and but that's you know, that's part of this journey, right? It's it's to your point, it's not static, right? We need to be engaged in this activity. And and there's there's some secondary, I think that's the analytic front, right? There's some secondary aspects of this where, okay, that's a very analytical view. Some people don't like viewing dashboards or like really complicated, you know, uh table diagrams. Maybe we can distill that into what maybe be like a readiness hub or some form of a presentation that's more consumable to the 99% of people who just want the information that's relevant to them in a f in a way that they can consume it readily, and then maybe a link to the actual analytic for the people who want to dive deeper, right? So like I think there's this yin and yang uh give and take where you're building kind of both of those experiences at the same time. So I'm very excited about that. Um, we're uh continuing that journey.

Use Case Inventory And Domains

SPEAKER_02 12:43

And I also think another best practice is you know, we've kind of paused a little bit because of like the influx of all the work coming in, but our data literacy program. And I still think like we're starting that up again, but I still think that's key and critical to continue um the journey on that because I think just not only is it educating and teaching and and even partnering if you're not the chief AI officer and data officer, but like partnering with AI officer and doing AI literacy as well as a part as a partner, because I think it's so important for people to understand that not only is it teaching them and educating them on why it's important, but it gets people excited. And I always said, like, if I could do nothing else as CDO, I'm gonna like be annoyingly happy and excited about data because I think people always saw it as like it's just data. Like you should have heard the comments from all my friends when I'm like, should I be the chief data officer? They're like, You're gonna be one of those boring data people. But like it's for me, it's everything, it's like become my passion. And it is everything, and I think it's highlighted. And so I think for me, the data literacy program is key and important to continue um to go through and to continue to do. And also, again, I know I mentioned business partner, but another example I just want to say is like there was an area of like quality. This was return to work, and we were using building data, and we realized how not horrible, but how like all the opportunities we had to improve our stability data. And you know, it was interesting. So we did all this work and we cleaned this data and we made it better, and then like two months later, I'm like, wait, how did this get bad again? And then we didn't realize that there was somebody that was in the background, the business was like, hey, I can just manually update this in another system, and then I want you to come back and pull that data. And so that's how the data quality was coming back, is you know, so it's interesting because you you get to learn so much new things when you start like building those relationships and digging deeper.

SPEAKER_00 14:26

I'm gonna take another, I'm gonna throw another lesson out there. Here we go. Because we got we got some time. So one of the problems with this goes back to transformation, right? When you're transforming, you're building things and you're killing things, right? Uh it's just the nature of the beast. Well, guess what happens when you start killing things, right? What you relied on as a source data solution no longer is your source data solution. Or you might have merged two disparate source data solutions into one, and you have to deal with the repercussions of that, right? You have to you have to build new workflows, new ETL pipelines. Which which one is the author actual authoritative source? That's the other thing that's right, these systems they grow, they're all over the place. And you're like, oh, I I go to this system for training data, and then you ask the next guy, I go to that system for training data, and you ask the next guy, no, that's the actual training data system. Guess what? They're all correct because they all had different views on different trainings, right? They were not unified. And so from a data aggregation standpoint, I actually needed to connect to all three of them, right? The original target was just connect to the one because that was the authoritative system. But you actually had to connect to all three and then aggregate that information into the true training readiness data set. Thankfully, we we helped collapse that a little bit. But the lesson learned here is like don't go into the data journey with the assumption that you know where your data is and what it looks like.

unknown 15:45

Yeah.

Iteration, Change Management, Feedback

SPEAKER_00 15:45

It is all over the place, it's constantly moving, and you have to be on it. Right? And that's when you then the there's a transference here of that that journey from what would be like the technologist side, because we're doing like the real like connection stuff, you gotta transfer that to the domain owners at some point, right? The domain owners are wholly responsible for that, the answer to that question. And if they can't answer that question, find a different domain owner because they're not doing their job. That's my opinion. And they really, really need to be in control of that situation, right? And it's the the the issue here is that you cross you're crossing organizational silo boundaries, right? Because what would typically be an IT concern is now all of a sudden three or four people's concern, right? You have a privacy, you have the data owner, you have the IT person, you have the infrastructure people, that's me, right? And you have to convince all these people that this is what you want to have happen. And that is in the government is incredibly difficult, right? So that's where the CDO comes in, a really strong CDO like Corinne would come in and set those standards, right, and really enforce that. Because that's that is how we can move this forward.

SPEAKER_01 16:51

No, absolutely. I mean, data discovery is the foundational component, and then understanding what your data standards are and enforcing those is really foundational to all the incredible things that you're able to do with AI and other technologies. And even just having an understanding of the data lineage and data quality is also incredibly important because when it goes back to what we were talking about earlier with with keeping AI accountable, you know, how do we trace that back? We trace it back to the data. And where did this data come from? It came from this ETL process, from this data source, et cetera. So having that lineage is incredibly important as well. And so all of this really is, you know, may not be as interesting to some people as all the fun AI stuff we like to talk about, but it is foundational and critical to making this stuff operational and actually useful. So it makes a lot of sense. So that's a big focus area. Um okay. So when it comes to workforce and culture, what cultural or workforce shifts are required for leaders and operators to trust AI-assisted decisions without abdicating responsibility, which we we touched on a little bit, but we'd like to it would be nice to get a deeper perspective on what you all are thinking about this as well.

From Dashboards To Readiness Hubs

SPEAKER_02 18:05

Yeah, I think for culture changes, you know, again with the literacy and things like that, you're helping change that and being excited, showing those use cases as success stories and then teaching the boundaries and what's good for, what's not, and just enforcing like for not changing the decision. It's not a replacement, it is an ally. AI is an ally to all of us and to make our jobs easier. And I think also like leadership, leading by example, um, being accountable for the decisions that are made and holding that accountability, but then also showing the use and using AI for their own decision making and then owning that decision. Um, a lot of it's we've already touched on the culture, but you know, that's the key ones for me.

Data Literacy And Excitement

SPEAKER_00 18:44

Yep. Um so as you can tell, I'm pretty enthusiastic about this stuff. So that's part of the culture, right? Bring the enthusiasm to the table. The bodega analogy is culture, right? You're building this momentum, right? Nothing's gonna happen if people don't believe that you can get it done, right? And you build trust in that by sh by not only leading the way, like you're building space, like you're building safe space there for other people to follow in for you, but you're also sh you're also showing them competence and uh d decision making. And that's that's something that is um really important in this journey because you it it can go sideways 500 different ways during this journey. And if you don't have a really strong uh focus on on that, those basic requirements of just enthusiasm and and uh grit and focus on the mission, it it probably won't ever come to fruition, right? There's just too many ways it can kind of peel off and and become just a you know a black hole type circus, right? Uh so I'm really excited because we've had such a great um great l um lineage of of leaders helping with this journey with enthusiasm and with focus and drive. And I think that's why we are where we are today. Right? It takes more than just the IT people to do this work, right?

SPEAKER_02 20:05

And I think also the culture, like, you know, it's been a struggle for me, and I've said this to my leadership of not having a seat at the table as being CDO, right? And not being able to constantly talk about data. And I think having a leader who believes in data and the importance of it in governance as much as you do, so they can help bring your stories to the table or bring you to the table and enforce that through the agency, I think that is also key to having culture culture. It's that executive buy in and support all the way to the top. So it's gotta start at the top, otherwise you will never be successful.

SPEAKER_01 20:33

Yeah, that that makes a lot of sense.