How DoW Can Keep Mission-Critical Operations Running

Maintenance organizations already hold the information they need. Getting to it is the problem. That is the through-line of this GovLoop panel, recorded August 27, 2026, with Nina Khan of the U.S. Department of War, Thomas LaBatt of One Aligned, Dominique Luzeaux of NATO Allied Command Transformation, Dr. Mark Taylor, formerly of U.S. Special Operations Command, and Maddie Wolf, General Manager and Vice President of Operations at Legion Intelligence. Emily Jarvis of GovLoop moderates. The conversation covers where maintenance and facilities workflows break down today across the U.S. services and NATO, and what can be improved without replacing the systems already in place.

The panel treats fragmented data as the first constraint. Maintenance records sit in SharePoint, in FlankSpeed, in homegrown tools, and in handwritten fuel logs that someone retypes into a spreadsheet two weeks later. Panelists describe making that material AI-ready as a storage, format, and normalization problem before it is a model problem. A workforce problem runs alongside it. The institutional knowledge held by experienced maintainers is largely uncaptured, and turnover is outpacing any effort to write it down. Several speakers point to AI as the practical way to capture it, by letting the people who hold that knowledge talk into a system rather than document it. On legacy environments the panel is direct: telling an organization to replace its CMMS or ERP ends the conversation. Working on top of what is already deployed, including IBM Maximo, is the realistic path.

The second half moves past retrieval. Chat access to technical manuals on a technician tablet is treated as a starting point rather than a destination, with the more interesting work upstream. Optimizing reorder points, accounting for planned maintenance in inventory supply, scheduling jobs against parts that will actually be on hand, and pushing a likely diagnosis to a technician before they arrive on site. Panelists also cover spare parts demand forecasting, digital twins, and readiness degradation modeling, alongside the governance question of what guardrails and attribution look like when agentic workflows enter secure enclaves. On acquiring these capabilities, the recurring theme is requirements discipline and room to fail fast, with contract language that lets a program iterate rather than field something that reached end of life before it was delivered.

Legion Intelligence works in the environments this panel describes. The Facilities Maintenance Pack applies these workflows to maintenance and sustainment operations, and the MRO industry page covers the wider sustainment picture.

Transcript

This transcript was machine-generated from the session recording and lightly edited for readability. Names and technical terms may be rendered inaccurately; where a term was clearly mistranscribed, the correction appears in square brackets alongside the original.

Speakers

• Nina Khan, Strategic Operational Energy and AI Integration Lead, U.S. Department of War
• Thomas LaBatt, Chief Transformation Officer, One Aligned, Inc.
• Dominique Luzeaux, Transformation Champion and Special Advisor to the Supreme Allied Commander Transformation, NATO Allied Command Transformation
• Dr. Mark Taylor, Former Chief Technical Officer, U.S. Special Operations Command
• Maddie Wolf, General Manager and Vice President of Operations, Legion Intelligence
• Emily Jarvis, Senior Manager for Events, GovLoop (moderator)

Opening and Introductions

Emily Jarvis: Hello everyone, and welcome to our session, How the DOW [DoW] Can Keep Mission Critical Operations Running. By way of introduction, I'm Emily Jarvis, and I'm the Senior Manager for Events here at GovLoop. And today I have the added pleasure of also being your moderator, so thanks so much for joining me. And a special thank you to our sponsors at Legion for helping us put on this session today.

So let's get right into it, because behind every mission are the people, the systems, the maintenance, facilities, and infrastructure that really make it all work. But keeping those operations running is getting more complex, especially as defense organizations engage in aging assets, growing data, workforce transitions, and of course, higher expectations of readiness.

AI can be a tool that can help make information easier to find and connect teams to knowledge that they need and support faster, more informed decisions when something needs attention. But how? And how does it all work? Today our experts will explore how defense organizations are using AI to strengthen maintenance and sustainment operations, reduce downtime, and help keep teams' mission critical assets ready from the installation stages to the edge.

But before we get to all of that, we've got some housekeeping items to go over with you all. First things first, we have some great resources available for you on the console under that resources button. This is also where you can learn a little bit more about our speakers, so make sure you check that out.

If you need access to that closed captioning feature, just look for that Show Captions icon along the bottom right-hand side of your panel, and that's where you'll find that added accessibility. Of course, we hope you don't have any technical difficulties, but if you do, send a message to that Q&A tab and one of my GovLoop colleagues will get you all sorted out.

You can use that same tab to submit questions to our speakers, and we want your questions, so put them in early and often. You can also engage with your fellow attendees today using that chat feature. I know a lot of folks are already telling us where they're logging in from, so I really appreciate that.

I'm in Rockville, Maryland. Let me know where you're logging in from. All right. With all of that housekeeping out of the way, I want to introduce you to our speakers, and we've got quite the panel for you today. Joining us, we have Nina Khan, who was formerly, just recently left, the lead strategic operational energy and AI integrations with the US Department of War. Now she's an AI architect and automation strategist.

We also have Thomas Labatt [LaBatt], who is the chief transformation officer at OneAligned Inc [One Aligned, Inc.], and a former Naval Sea Systems Command executive and director of industrial operation transformation. We have Dominique Luzak [Luzeaux], who is the transformation champion and special advisor to the Supreme Allied Commander Transformation for NATO Allied Command Transformation.

Dr. Mark Taylor, who's the former chief technical officer for US Special Operations Command, and the chief innovation officer at Sage Creek USA. And last but certainly not least, we have Maddie Wolf, who's the general manager and vice president of operations at Legion. Thank you all so much for being with us today.

And I want to get this conversation started. Maybe Thomas, maybe you can kick us off here, because for agencies who are responsible for keeping mission critical assets and facilities running, where do you see some of the biggest maintenance and sustainment challenges right now?

Access to Data Is the First Constraint

Thomas LaBatt: Very good. Excellent question, and I appreciate being here today. I really appreciate the invite.

The biggest challenge is, now, I come from the Navy background, naval aviation, ships. And it's about access to data, access to information. They're all about understanding the ship's condition, the visibility of that, understanding the flow and the learning.

That only happens when you have access to the right data. And when I say data it's not so much the MAS [MAF], the maintenance action plan, to have those things printed out. It's having access to unstructured data in a structured way so that you can actually do something with it. I like to call that AI-ready data.

Emily Jarvis: Okay. Mark, Dr. Taylor, I should say. Mark to his friends, probably. What do you see from your perspective as you're looking at these challenges when it relates to maintenance and sustainment? Where do you see some of the pain points popping up?

Dr. Mark Taylor: I think a big problem right now is the standardization of the data formats.

So it's a follow on. A, you gotta have good data. Garbage in equals garbage out. So are we collecting the data? Where is the data being collected? What format is the data? For years I've been in places where you'd have data scientists and engineers go to the customers and say, "Okay, tell me about your data."

And they're like, "I'm a war fighter. I don't know." So I think a lot of times you gotta look at, okay, what type of storage do you have? If you have a NAS, it's unstructured. If you have a SAN, it's structured. Okay, what are the apps that are working in those devices? Who runs those? So that you can try to get a feel for what is your data, and is it conditioned, and is it in a place or is there a gap that I need to fill in order to either do data normalization so that I can get to my data in a structured manner or some kind of way that is consistent?

And then what are the tools that I'm gonna use? And then, finally, without going too long, it gets down to the training. Do I have people that understand all of that stuff I just said, and is there a pathway, a bridge, on doing things the new way? So it starts with the storage. It then goes to the data, then it goes to a format and the standardization of the data.

And then do I have a bridge so that I can present that data to applications that can then give me insights so that I've got something that I can use the data? Am I doing prediction? Am I just doing categorization? Am I just doing a list of what I have? All of those things come into play.

Emily Jarvis: And when you list them out like that it does feel like a little bit of a daunting list there.

Nina, as we go around this round robin opening question, things obviously get harder to answer. But from your perspective, and I know you've just recently left the DOW [DoW], where do you see some of these challenges that you have experienced?

Nina Khan: Thank you for having me, Emily, first of all. But I wanted to start off with saying that I totally agree with my colleagues here. Data is a big thing, but coming and looking at it from the senior leadership's position at the Department of War, it's the retrieval of information that is a problem, and that, I feel, is a bigger problem that all of us are working with.

So in my position, one of the things that I did do was include and incorporate AI and try to get some sort of agents or agentic models into the data collection part, which is, of course, then going to lead to the garbage in, garbage out validation of data. It's going to lead to the cleanup of data, and then that data becomes information, and that is what you need at the drop of a hat, in a microsecond, so you can decide and see which way the forces should be led.

Maintenance as the Foundation of Readiness

Emily Jarvis: Makes a whole lot of sense.

All right, Maddie, you're obviously coming at this from a slightly different angle, but when you're going out and you're talking to your partners about this idea of operational readiness, what does that look like in practice for them, and why do you see maintenance as such a critical element in achieving those goals?

Maddie Wolf: In terms of maintenance being critical for operational readiness, you don't have the asset, it's not ready. It doesn't work, it's not ready. It's really as straightforward as that. I think though what I'm hearing from people is...

Emily Jarvis: I think we might have lost Maddie for a second, so we'll come back to her.

Maddie Wolf: I think a lot of the...

Emily Jarvis: I think she's having some connectivity issues. So we're gonna suss out what's going on with her feed, and we'll get back to her in just one second.

Welcome, Dominique. It's great to have you with us. How are you doing today?

Dominique Luzeaux: Very good, thank you. You are much welcome.

Emily Jarvis: Wonderful. All right, so we're laying the groundwork here, and obviously you are coming at us with a NATO perspective. Where do you see agencies responsible for keeping mission critical assets and facilities running, what are their biggest maintenance and sustainment challenges that you're seeing right now?

Dominique Luzeaux: So as NATO is 32 nations on two continents. Basically we have very fragmented data. So I also share what some of the previous panelists said.

The retrieval of the data is very important. But for NATO, fragmentation is the first stopping show. And this is where we can use AI, because AI can help us connect the different maintenance records, the logistics data. It helps also capture some higher level knowledge.

And this is how we can try to have a consolidated operational picture.

Aging Infrastructure and Operational Tech Debt

Emily Jarvis: Wonderful. All right, Thomas, so I'm gonna come back to you here. I think everyone threw out some really tough challenges and maybe some opportunities and perhaps where AI can play a role. Obviously, we know a lot of organizations are managing an aging infrastructure where there's some siloed systems perhaps, there's some workforce turnover. Where do these factors create the biggest risks when it comes to that mission readiness piece?

Thomas LaBatt: So when I think of mission readiness it's a closed loop system. You design systems. You do the RCM analysis after you do the FEMECA [FMECA], the maintenance task analysis, and you come out with that maintenance plan, right?

So that's all your baseline engineering assumptions. And once we get out there and you're trying to sustain these systems, if you have all these siloed systems or you can't get access to the operational data or how those systems were designed and how they're performing real time, then you're not going to get that closed loop to be able to say we have to update that maintenance plan.

We gotta go back through the RCM analysis and make sure that it was a 1,000-hour inspection, now it really needs to be an 800-hour inspection. Because that drives the supply chain. So I think some of the aging infrastructure part of that is if you have on-prem servers and you can't get access, you've gotta be able to connect that data, whether you're using RPAs or however you can, to connect the data so you get it going across the organization, and then be able to have that closed loop analysis where you can do a Bayesian method.

You can look at the readiness model and actually be more predictive as to what the health of those systems are gonna be in the future as opposed to what it is today.

Emily Jarvis: Mark, building on what Thomas just said here, in your opening remarks you were talking about everything from the processes in place to the workforce component. From your perspective, where do you see some of these risks to mission readiness within this realm?

Dr. Mark Taylor: I guess in this whole conversation it becomes a time to value metric. Like, how soon am I going to be able to provide value once we define what success is? Am I going to be able to use the tools relative to what I need to accomplish going forward?

And then is my current environment going to enable me to do the things that are gonna make me successful in that outcome? Which is a kind of a long-winded way of saying we've got operational tech debt. You go to war with the army you have, right? So nobody here has ever stepped foot as the new CIO or CTO or leader in IT, gone to the data center and seen pristine, standardized everything.

We've got this hodgepodge of I've got some Dell and some IBM and some HP and some Cisco. You have this kludge, and so you have a legacy network that you have to figure out how to keep running, and then you have changing requirements that are coming up, in which case you have to be mission ready.

So then there's the balance on what is it that I'm doing today that is gonna enable me to continue to achieve the mission, and then define those gaps. What we're finding is through AI, we're able to give an extension to gear that probably would have tech debt or not be able, in and of itself, to provide the capabilities that we need or the insights.

What's an example? Some of the things that we're trying to do with Zero Trust to try to make sure that all of our legacy systems are able to provide security as well as the security of the data meet the high level mark for our 2027 Davidson window. Not all of the gear that you have that's legacy is capable of providing that level of insight.

So then you have to look at the management plane for those individual one-offs, and then put those controls there, and get the insights of things that manage those devices, because the device at the bottom end might not have the capability to go all the way to your reporting mechanism. So it's coming up with a tiered approach.

In a perfect world, you draw a line in the sand and you say, "This is how I'm gonna do all of my new operations, and this is how I'm gonna maintain my legacy." The last thing I'll say, which is the thing that we notice that brings back problems is, A, whenever you bring something new, are you doing regression testing to make sure when the new thing came out it didn't unleash an old problem that for the last two versions were gone but it came back?

Which then also means you then have to have a very strict method of decommissioning something that's been old and been decided that it will leave the network, and actually make sure it is off the network. What I see a lot in the government is they're like, "Ooh, this server is now no longer doing this thing. I'll just reuse it to do this other thing." And then now you're extending the life of your tech debt.

When Information Becomes the Bottleneck

Emily Jarvis: I love your ideal world. But as a parent I know that ideal world almost never exists, so I understand the reality that you're talking about on the ground there as well.

All right. Maddie, I think we have you back, but just by voice. Can you hear me okay, Maddie?

Maddie Wolf: I can hear you just fine. Can you hear me?

Emily Jarvis: Yes, I can. So maintenance teams, they're often struggling to find the right information when they need it. Where do you see some of these common barriers that can slow down decision-making or maybe contribute to some downtime?

Maddie Wolf: I think there's a lot of different places where essentially information becomes the bottleneck, right? And it's not just when you're on the job or on site actually fixing something. It happens long before there. So I think the thing that's been the most interesting to me is not just helping technicians find information, but actually pushing information to them rather than them having to ask for that information, right?

So think of before I send a technician out to fix a given problem or to diagnose something, can I as a supervisor or someone else give them, based on their data, what the best guess probably is? And in the past, a lot of that was basically some form of predictive maintenance saying, "Hey, there was this sensor. There was something over here that went wrong."

But things have really changed from that. And I think now, with AI, we're essentially able to look at historical data, other information, and say, "Hey, when you go, here are the things that you actually should look at. Here's the information that might help you."

Rather than sending people out and telling them to phone home or figure out the information themselves, we should be providing it to them first. And it's a possibility now. So that's what I've been more interested in, and I think it flips the problem on the head. Rather than them not being able to find it, why don't we help them solve the problem by giving them the information they need?

Institutional Knowledge and Workforce Turnover

Emily Jarvis: That makes total sense to me. Nina, what would you say in addition to what Maddie was just talking about, getting information to the folks that need it and when they need it? Where do you see those slowdowns happening?

Nina Khan: So the number one reason for this information black hole, I would say, is because most of the information, which I like to call institutional knowledge, whether it's the ships or whatever it is in the military that we're talking about, it could be tanks, it could be a whole bunch of different things.

But the problem is that institutional knowledge is not captured, and then we are losing people very quickly. The turnover is very fast. And even right now I believe the military has achieved its benchmark of hiring, but still they constantly need the welders and the ship workers just to understand how it works.

So it's not a matter of putting butts in the seat, but it's more a matter of having the knowledge, even at a lower level, to be able to make sure that the maintenance is done the way it is done. AI can help with that very easily, because all you need to do is create a program and just have whoever has the institutional knowledge talk into it. Just talk into it, and that's it.

Working Within Existing Environments

Emily Jarvis: All right. Dominique, from your perspective, you're in a position where you're interacting with a lot of different entities, and obviously agencies in all forms are looking to modernize, but replacing legacy systems isn't always realistic budgetarily, systematically, programmatically. What challenges do organizations face when they try to improve these operations while working within these existing environments? And I know Mark was hinting at this a little bit earlier. Where do you see some of those pain points popping up?

Dominique Luzeaux: So among the pain points, we have the fact that the data, the information is scattered across multiple systems. So this is a difficulty in order to have an operational visibility. Another pain point is, of course, the poor data quality, the lack of real time asset visibility. So it's very difficult to have a real time operational view and situation awareness.

And we also mentioned the turnover in the people, and this is really something we see in many nations. And this raises difficulties to find the relevant technical documentation or information. And this is actually, I totally agree with the previous speaker, this is where AI can help us because it can very much more easily find the documentation or the relevant technical information.

Of course, this has to be validated, but still it's one way to be able to cope with this turnover. And the final thing I would say is that because of the fragmented scattered information and data you lose time. There is a lot of time spent in searching for the right information rather than doing the core job, which is acting, taking decisions and acting.

Knowledge Management and What AI Changes

Emily Jarvis: Makes total sense. So since we have brought in the idea of AI here, let's drill down a little bit. And Mark, I'm gonna start with you. We've heard from Nina and Dominique here, and Maddie of course as well, some initial ideas of where AI can bring maintenance records, technical documentation, this institutional knowledge together a little bit faster. Where do you see it playing a role directly in bridging some of these gaps that we've highlighted so far?

Dr. Mark Taylor: I think that Nina really hit the nail on the head. I think the system that was big about 15 years ago that from my observation has been dying because it wasn't funded or it rolled out correctly, is knowledge management.

Emily Jarvis: Ah, yes.

Dr. Mark Taylor: And a chatbot is a good example. Obviously, AI, the ability to talk to it and ability to train an agentic model. But it comes down to a number of different things. But a simple one is just knowledge management. For years there were custom and bespoke systems that would capture information so that lesser skilled or younger or less seasoned professionals in the industry could gain insights and understanding so that they could become more rapidly efficient.

And I'm gonna use the administrative example versus like a combat example. But folks didn't roll out knowledge management correctly because, A, they didn't wanna pay for a couple of graybeards, people who are getting ready to retire, to spend all day for their last three to four years to talk into or capture or write a system.

And SharePoint ended up being what was normally used in most of the DOW [DoW], formerly DOD, to be that system, which, it's a good enough tool, but that's not its purpose, right? Now you can get a bot, sit it on top of an FAQ, and now it's gonna be able to answer your questions based off of what you just asked it.

Now, take that into 2027 with the frontier models and all of the foundational capabilities that we've got with AI, much less the fact of injecting RAG, but now you're utilizing agentic or large context RAG or other ways to get bigger different custom questions into a large language model.

Now you can essentially build your knowledge management system with a virtual bot, person, entity, whatever you wanna call it, and now use an LLM that was either custom, bespoke, or something that was already out there, and then inject the new questions and then frame that and use its same context.

And now you've essentially short-circuited, closed that gap, or back to the whole time to value metric on being able to create instant value, provide a system that is showing people how I gain insights into what we've already been doing. Now, if you can take that and do that with your legacy and your current system, now going forward, you follow the same logic and things just grow exponentially.

Making Existing Data Usable

Emily Jarvis: Makes sense. Thomas, where are some practical ways that you see agencies really start using AI to reduce downtime or improve asset availability or maybe make better use of perhaps some of these staffing challenges that we've talked about so far?

Thomas LaBatt: Asset availability is critical. That feeds right into readiness, right? And I appreciate that Mark just mentioned SharePoint. There's also the Department of the Navy has what's called FlankSpeed. And these are just locations that have dumps of data.

Dr. Mark Taylor: Yeah.

Thomas LaBatt: And so much of that data is just not accessible into any real AI-ready data to go process any information.

So we're doing a lot of work currently with the Naval Air Systems Command in trying to understand where their data is and how to take it out, and we have to redact data, we have to reclassify the data because it's not classified, that unstructured, to get it structured, to get it into a format that now when you look at the maintenance system and you go from the design of a system to the operations of the system and sustainment of the system, and looking across the three levels of maintenance, then we're connecting that data to be able to get into the asset management and be more predictive.

And more predictive helps with the supply chain. When you're able to be a predictive model and not a reactive, and when I say reactive, today you've got organizations out there that are just tracking parts and chasing parts because their supportability analysis that was done put a certain demand on the supply system.

But the reality is, the sustainment and the operations didn't support that original analysis, and therefore the supply is not keeping up because the demand wasn't projected the way it should be. So now if we can use that data and use that closed loop, and you mentioned a large language model, build these models where the sailor or the airman can get in there, they can speak to it, it'll go through the data and pull up where it's searching the maintenance records.

It's searching the data out of the SharePoint and FlankSpeed that's been all unstructured and put into the model. It's that type of data that I think AI is gonna be so powerful and it's gonna be such a key enabler for the maintenance community.

Emily Jarvis: FlankSpeed is also just such a great name.

Thomas LaBatt: Yeah, that's about all it is.

Beyond the Chatbot

Emily Jarvis: We'll give it that. Okay. It got a great name. Maddie, from your perspective when you're working with your DOW [DoW] partners, what are some ways that you can see AI helping to reduce some of that downtime? Maddie, are you still there?

Maddie Wolf: Yeah. Can you hear me?

Emily Jarvis: Now I can, yeah.

Maddie Wolf: All right. Apologies once again. I ironically just got a message from my internet service provider telling me that they had planned maintenance occurring, which felt perfect and unfortunate timing for this webinar.

So I think there's a few things. You can start with the simple, which is, for example, we have a chat instance that is on tablets for technicians where we are porting all of the manuals that they have stored in SharePoint, allowing them to chat against those manuals, get answers that are referenceable, citable, and use it to help troubleshoot.

That already has proven really effective. We found a 25% increase in capacity partners immediately once they started doing that. Now, I think what I would say though is, why stop there? I think chatbots are boring. And so let's think through what else it is that we can do that'll help readiness, right?

And so what about before we even get there, like I said? Can we improve inventory readiness or improve inventory supply? So using AI to optimize reorder points, to better understand lead times, to take into account planned maintenance in the inventory supply we have and ordering and thinking ahead.

Can we go ahead and make sure that we are scheduling jobs in a smarter way using AI? Can we understand which jobs, for example, won't be able to be completed because of inventory and not with the model? Can we get smarter cannibalization when it has to happen? There's so much more we can do beyond just a chatbot.

And so what I would really push people with is, of course, a chat for a technician on a tablet, yes, it's proven super effective, and there's things we can do there. Let's go upstream, and let's do things that are a little bit more interesting around prescriptive maintenance. Those to me are a bit more difficult sometimes to get going. But we've seen a lot of success with it, and to me, it's pushing the boundaries.

Guardrails, Retrieval, and Handwritten Data

Emily Jarvis: Wonderful. I wanna remind our audience, don't wait until the very end to ask your questions. Put them in early and often.

And Nina and Dominique, I wanna get both of your insights here too. So Nina, I'll start with you here. AI is here. It's coming in. It's going to be a factor. Where do you see it potentially having some opportunities to make better use of some of these limited resources?

Nina Khan: That's a great question, Emily. I love it. So I feel that you're right, AI is here, and it's here to stay, so there's no need to fight against it, but embrace it. Embrace it, but ensure that you have a zero trust environment to protect all your data. Make sure that when you are creating models or agents or creating an agentic framework, loops, whatever, there are guardrails. So again, that is non-negotiable in my opinion.

The problem that I see, as far as tying in our conversation about maintenance, it's not that the information is not there. The information exists. It's, again, the matter of retrieval. How do you retrieve all that? And this is something that I had inside the Pentagon, so constantly being asked, "Okay, what's the readiness level of this? How many such and such things do we have? What can be placed in one section versus another section?"

I'll give you a small little example. Several of the Navy vessels have an issue because the fuel data, how much they are using, how much it's being expended, is still handwritten.

Emily Jarvis: Oh.

Nina Khan: Yes. And it comes in every two weeks. So then somebody has to physically sit down and actually figure out what those numbers are in whoever's handwriting is, and I hope to God they're not like mine, because mine's horrible. Put it in an Excel sheet and then run the analysis. Such a simple little thing which can be taken over by AI. This is such a tedious process.

And it's not just in the Navy, I've seen it in the Army, in some cases prior to that in the Air Force as well. So it's a matter of collection and retrieval of data. So when you look at it from that aspect, this is a problem I was working with inside the government and now it's my primary focus outside at Amna19. It's basically this, how can we make things quicker and easier?

Predictive Maintenance and Readiness Forecasting

Emily Jarvis: Wonderful. All right, Dominique, you have the hard job here because you've heard now four other best practices. But I know you can rise to the challenge here. Where do you see, from your purview, AI coming in to reduce downtime or improve that asset availability? We know we've got limited resources on the table, so where do you see AI playing a role?

Dominique Luzeaux: So in addition to what was already said in terms of predictive maintenance, we can apply for equipment health monitoring, but also more importantly as a spare parts demand forecasting, and readiness degradation forecasting.

So of course we also have the same issues that were just said before. Some of the things are still made manually. But let us imagine the processes will be better. What we can do is maybe use AI to deliver insights faster and maybe also to identify pattern humans may miss, in order to provide a better operational picture and readiness picture. So basically this is where we expect from AI not only to give us the right picture, but also maybe to give us the better insights in order to not just be reactive but be proactive in terms of maintenance.

Evaluating Whether a Solution Works

Emily Jarvis: Makes sense. All right, Mark, we got a great question in here from one of our attendees, and she was asking, sometimes we bring in a new solution and it can fail to meet goals, whether that's performance or maybe there's pushback from the users or they're not really using it. How can you evaluate whether or not a solution is effective in helping, in this case, obviously we're talking about maintenance, but what areas do you look at to evaluate a program's success?

Dr. Mark Taylor: Part of the problem in the Fed I've seen over 30 years is the requirements of what the widget or service or capability is supposed to do did not have good enough requirements built in so that then vendors or systems integrators when they were pitching the solution, you're trying to explain an architecture to a person who can't see or hear, through the art of interpretive dance.

So if you give me a soft requirement, then how am I supposed to give you a discrete solution?

Emily Jarvis: Yeah.

Dr. Mark Taylor: So what we try to do is a very rigorous requirements validation to make sure that when we asked industry for something, that they gave us what it is that we think that we wanted. The next part of that is making sure that you open it up and you don't work in a vacuum.

So depending on your organization, at SOCOM, we're global. We would make sure that we would get representatives from each of the different groups around the globe and across the different user communities. SEALs like radios, but they probably like radios that are different than Army. Why? Because the waveforms have to travel different ways to enact and work in the environment differently.

So you have to get different people who have different equities to look at what it is that you're trying to do, and then they have to test and evaluate and validate it. A, did it meet the requirement that was stipulated to industry to respond to? B, did it meet your requirement that maybe we didn't capture?

And then ultimately, you've got to have the capability to fail fast. You have to write language into your requirement out to industry or whomever that basically is, this is a trial period, a get out of jail free card, a time limit, and we will test it up to this point.

And then if we have proven or disproven value, that we can cut sling load on that and reiterate and go back to the drawing board. Because I have seen programs where literally a system was being delivered that was not only end of sale, it was end of life. It was end of support. So when we tried to provide the next new version that still met the requirements, the government customer would not take it because it did not meet the FAR requirements of what was in the contract.

So literally, we had to buy from secondary markets just to keep enough on hand to replace what was there. That's what I'm trying to say. You need to have language that allows you to iterate. I think the correct word that Tom used earlier was Bayesian. You have to be able to look at where you are with the new information you have and make a better decision over time.

Because where we are now is not where we were two years ago when our money that we planned for this year was designed. Oh, by the way, in two years, what's going on in the world? We've had a whole lot of changes in the world that we're still working on that we funded two years ago.

Modernizing Without a Full Overhaul

Emily Jarvis: It is a big undertaking. All right, we are almost out of time today, but I've got a couple more questions that we're gonna do in rapid fire format. So Thomas, I'm gonna start with you with this next one. Any examples that you have, and you don't have to get ultra-specific, we're not trying to break any security clearances or anything here. But any examples you can share about how modernizing maintenance or sustainment operations can happen without perhaps a complete technology overhaul?

Thomas LaBatt: Very good. I'll give a quick example of working in the submarine community. We have NNPI data, and what that means is Naval Nuclear Propulsion Information data, and it's highly classified.

We needed to be able to do some analysis on it, but the portion of that data that we did not need was the classified part of that data. We just needed the actual operational side of that maintenance data. So we were able to extract CUI data from that using some AI and being able to get the methodology of it so that we could take that into our operational gap that we're trying to solve and use some of the analysis to get more of a predictive model.

So that was just a quick, it was an operational problem that we had. It was a small gap, and how do we use AI to get the data into an environment that we can then use it?

Emily Jarvis: Makes sense. Meena [Nina], any examples you wanna share about modernizing, let's say, a part of the system without taking the whole thing out of commission entirely?

Nina Khan: Yes. And again, I'm going to go back to AI because that's the quickest and easiest way to get things done. We can take handwritten notes, handwritten fuel data, energy data, electricity data, water data, whatever you have, and just use an AI agent to transfer it into whichever form you want and then actually run the analysis for you.

So that is the best way of doing it. And given the fact that the department has access to genai.mil [GenAI.mil] and a lot of other capabilities that normally we would not have access to. And it's a great tool. You just have to learn it, and you have to teach what you want out of it. I think it should be mandatory, in my opinion, for everybody to start using it as much as possible.

Security, Trust, and Agentic Workflows in Secure Enclaves

Emily Jarvis: Dominique, we got a question into the Q&A portal that is talking about the security side of things. And obviously security is front and center to the DOW [DoW] and NATO operations as a whole. So the question says, "Introducing agentic workflows to secure enclaves can introduce the potential for vulnerabilities and scope creep." How do you think about that security, the ownership piece, the attribution? There's a lot of elements that go into this.

Dominique Luzeaux: Yes, it's a very good question. It goes back to trust. Trusting data, trusting models, trusting the implementation of the models on your own networks or systems.

So there is no one true one size fits all. You really have to look at all the different perspectives, and transparency of the models, of the data, are very important. And of course you might have new vulnerabilities, but it's always a trade-off, and it's like going to the cloud. Secure AI and GenAI, as we just saw or heard, can bring new insights. And you have to see the trade-off between the benefits and the risks.

Emily Jarvis: Yeah.

Dominique Luzeaux: And another thing where we could really rather easily change things is using also modeling simulation and AI together. Using the digital twins. And this is where maintenance and predictive maintenance can really gain something.

Meeting Organizations Where They Are

Emily Jarvis: Maddie, I wanna give you a chance here to talk about some examples. So from your perspective, working with your DOW [DoW] partners, where are you seeing some examples about how they're modernizing some of their maintenance and sustainment operations?

Maddie Wolf: I will say, I think if you go to anybody and tell them that they need to overhaul their CMMS or their ERP, the conversation sorta ends there, right? That is, as much as we sometimes would all love that, I think it's just super unrealistic. And so we are very used to working with folks who have systems that are quite old. Occasionally maybe older than myself, who's to say. And that's just the world that we live in, right?

And so I think that if you're thinking about modernizing any maintenance system that you have right now, I would be very skeptical of anyone who says, "Hey, you need to throw what you have out the window." It's just not true. Nina said this earlier a little bit, it's not like AI is the magic bullet, but it's definitely gonna help.

So we, for example, work with some of the older systems. We work with IBM Maximo as well, which is not an older system, but I know a lot of people are familiar with it in the maintenance space. Sitting on top of that, sitting on top of SharePoint, sitting on top of some homegrown system that shall not be named and is great. It's definitely great. That's how we're used to working.

And so what I would say is, find folks who are willing to work with the system that you have, because it's just the most realistic way to get things done. And what I also would say is, my advice for anyone when talking with folks is find people who are gonna just meet you where you're at. And I mean that not just your system, but people who are gonna come on site and come talk to you, and sit with your technicians and sit with your technician supervisors or anybody that, anyone that's gonna do a good job, that's actually worth your time, is gonna put in the effort too.

So I would say find people who are gonna meet you where they're at, and find people you can trust. If you're arguing over requirements that are needed for the work that you're gonna do, it's already over. You shouldn't be arguing about those things. You should be working with a partner who actually wants to work with you and will show up and just do what needs to be done regardless if it's in the requirement or not. I know that's a little bit of a hot take, and it's a little idealistic sometimes, and I get that, but we exist. We're out here. Find the partners who want to do right by you and you'll end up in a lot better place.

Looking Ahead: The Next Few Years

Emily Jarvis: We like a spicy take. All right, I have a final question for all of you, and Thomas, I'm gonna start with you. And this is a doozy, so I'm not gonna ask you to do it fully and completely. But looking ahead, how do you see potentially AI changing the way some defense organizations perhaps approach maintenance, sustainment, and readiness over the next couple of years? Can you predict the future for us?

Thomas LaBatt: Of course I can predict the future. All my years in the Department of Navy and now working in Department of War, it's about that closed loop system.

It's about using the war data platform that they have in place, which used to be called Jupiter. It's got the foundry and Databricks in there. Understanding those systems. I've got engineers that are in there every day. They understand Databricks. They understand how to use the technology. And coming up with those practical use cases that are really gonna drive readiness.

And I'm talking about the interdependencies between the system. Don't just come up with a hodgepodge of, here's one tactical thing to go do, but how's that connect, and trace it all the way up to the readiness and putting warheads on foreheads. It's gonna change. It is changing the future already, and if you're not on board you've got to get on board. You gotta get the right people to support you. And like I always say with my customers, it's a win for the government, it's a win for us, and it's a win for the military to go fight. It's changing the future.

Emily Jarvis: All right. Dominique, peer into the future for me. Where do you see AI playing a role in maintenance, sustainment, readiness over the next couple years?

Dominique Luzeaux: I would say the main shift would be from reactive maintenance to predictive readiness.

Maddie Wolf: Yeah.

Dominique Luzeaux: This is model and simulation, the twins, digital twins, and all the models, predictive analytics. And what is important, in order to be able to do that, we have to be data-driven.

So this means that actually we need to have the data, we need to share the data, to exchange the data, and it goes back also to the basic connectivity infrastructure and the data exchange layer. And as a final comment, I would say that yes, we can expect a lot of AI, but AI will not replace maintainers. It will actually empower them. It will help them having the right insight at the right time, and this is what I would expect from all of these new technologies.

Emily Jarvis: Wonderful. All right, Nina, you're up. Peer into the future for me. What do you see here?

Nina Khan: So I'm gonna agree with my colleagues. Predictive maintenance, which has already started in some areas, is the way to go. Maintenance, I would hope, moves out of the rear and into operational planning, just as AI needs to be in operational planning. Energy has now become a part of operational planning. That is very important. That's the way I hope it's going to go.

One other thing is, five years from now, I don't think the question will be, "Is it broken?" It'll be, "How much life does this thing have left?" That's a different question, and we can't answer it today because that data doesn't exist or we haven't been capturing it. But it doesn't mean we can't. We can.

Emily Jarvis: That is quite the interesting reframing. I will definitely keep that in mind. All right, Mark, what do you see? This gets harder to predict new futures here, but what do you see for us?

Dr. Mark Taylor: So I agree with what's been said. I am thinking that we are gonna see more AI being used to help train the people, so bridging AI back to the humans.

A vast majority of government employees are eligible for retirement soon, right? So you're gonna have the largest group of retirees that the country's seen since we've had the Social Security system, to the best of my knowledge. So the baby boomers are aging out. Hopefully Gen X, yay, will be able to retire soon.

But the issue, what does that mean? That means that the younger workers, the folks who are coming into the workforce, are still gonna have to have access to that historical knowledge. So whether it be something like using an AI-enabled tool that is connected with the predictive models and stuff that's saying, "Okay, we've got this many more hours of fight left in this widget on this plane." I'm just making that up.

How about that same thing, training or being a resource to help better inform the team, so that now I can aggressively prioritize my team based off of those insights that I got. I've got 12 more hours of fight left in this system. Now me, as a junior manager of worker bees, I can now better direct my team on where to go.

And maybe along the way, I can get that interactive training on that widget, or a refresher. So I'm hoping it can do that part. And that's the part that I see. The how that we actually start doing that is we have to have a conversation about the roles and the permissions of the AI so that we have responsible, and I believe Nina said it, responsible AI that is doing things that are within bounds.

Emily Jarvis: Makes sense. All right, Maddie, you get to have our final peer into the future here. What do you see on the horizon as AI becomes more central to these maintenance operations?

Maddie Wolf: I think the question, once again, and I recognize I'm always pushing the boundaries a little bit. My question is always, how can you get the people who really know what they're talking about when it comes to maintenance, the people who have been doing this for 20, 30, 40, 50 years, how can you get both those people and the most cutting edge machine learning type engineers together and put them in a room?

That's when things are gonna get interesting to me, and that's where I hope that we're going. I think a lot of times, I worked for a startup. I actually grew up in Silicon Valley. A lot of times we work in a bubble. It does not work to work in a bubble. And so I think what it is, is making sure that we are taking the people who understand what AI can do beyond just the GPT-esque type things it's doing now, but really pushing boundaries.

And then also have the people with the maintenance expertise and fit them together and really start going for it. I think that if we work in two different bubbles, we're gonna do stuff. It's gonna be cool. It might move some things. It might make predictive analytics better. It might somewhat improve mean time to repair, or maybe your first time fix rate goes down. But unless we really start collaborating, in my opinion it's not gonna get interesting.

So that's what I'm always on the lookout for. Who are the people who are down for the ride, who wanna sit down and do the process mapping that's boring and really push their boundaries. That's what excites me, because then we can start actually having practical application rather than just talking about stuff up in the clouds that may or may not happen. Let's do something interesting. Let's make some moves. That's what I'm here for, hopefully finding folks who are also interested in doing it and making a real change.

Closing

Emily Jarvis: I love that. And that sounds like the most fun summer camp actually. Maybe that's the future. We'll see.

But unfortunately we are out of time for our session today. I wanna thank our wonderful speakers for sharing their insights with us. Just a quick reminder that there are some resources available for you to download, so make sure you check those out so you don't miss out on a single learning opportunity.

We'll also be sending you a link to the on-demand version of this session tomorrow, so check that out in your inboxes. But on behalf of GovLoop and Legion, I hope you enjoyed today's session. And as always, we really appreciate and value the work that you do, so thanks for taking the time out of your busy schedules to join us this afternoon.

Have a great rest of your day, and we'll see you in our next training.

Dominique Luzeaux: Thank you.

‍