Terry Gerton: Paul, let’s begin with you. How did r4 Technologies get connected with the Air Force’s Special Warfare Division and what problem did they ask you to solve?

Paul Breitenbach: Yeah, it’s great to be here, Terry. r4 is a commercial AI software company started by the founders of Priceline. Our technology in the commercial world drives decision advantage and predictive actions. And we were called on by the folks in the military saying, can this commercial off-the-shelf AI technology really help drive readiness, both in the talent management space by matching people to roles to organizations, and also in assets, in supply chain and predictive logistics? It’s all about driving military readiness to out-tech our adversaries. And so that’s how we got involved.

Terry Gerton: Priceline is something I think about when I’m trying to book a flight. How does that kind of technology connect to, say, the human resources work that you’re doing?

Paul Breitenbach: Yeah, so when you think about what made Priceline so magical, it’s data and math in real time to match supply and demand. And in the incredible military capability of trying to find recruits and then maximizing and optimizing the talent, it’s really the same thing about driving yield, about maximizing supply and demand of individuals and their capabilities and the roles that the military needs. So this is actually really an incredibly natural extension and another way to think of supply and demand. And what’s neat about the way new AI, when we’re deploying, is it not only maximizes the outcome for our military, but it also maximizes the outcome for the individual by really helping promote them into positions of which they are very, very likely to be outstanding fits for.

Terry Gerton: You talk about, I’m not asking you for your trade secrets, but you talk about r4 as a cross-enterprise AI tool. What does that mean in terms of how it works?

Paul Breitenbach: Yeah, so the thing that makes the new AI and what we do as a cross-enterprise AI capability, it’s all about pulling the vast silos of data together. When you think about every supply chain and every system like this, there are just hundreds of different data sources. And Ty can tell you about just the unbelievable complexity about trying to find people, matching them and all the information together. And so what makes cross-enterprise AI, the new AI, this predictive capability that we drive so different is its ability to pull all those data together from different silos. So you can see and make sense and drive actions that used to take months, sometimes years, and do it in minutes. That’s what the difference is of our new cross-enterprise AI technology.

Terry Gerton: So Tyler, let me go to you with that background. One of the things that r4 has done is kind of look at the recruiting pool for Air Force Special Operations or Special Warfare. You train the system on over a million profiles. How did you do that? What kind of data did you look at? And I guess more importantly, when you put it all together, how did you keep bias from entering into the calculation?

Tyler Zagurski: Yeah, thanks, it’s a great question. And so we worked very closely with, and we do this with all our DoD customers to really find data in silos, like Paul talked about. And I used to say when I was in the DoD for 30 years, like we are drowning in data, it’s everywhere. But your data is in your system, Paul’s data is in his system, mine’s in mine. It never comes together so that you can find insights, that you can do predictions or optimize anything. So we worked closely with customers and in this case looked at 1.5 million applicant data, thousands of data elements per person, and anonymized it. It was great work with a customer willing to work with us and anonymize data so that we can use it in ways that they can’t.

We don’t know who the humans are, but we can tell you an exquisite level of detail about those humans. More importantly, we can predict which people will exceed at what roles. Maybe more important than that, we can tell you why. Here are the factors and attributes that we’re seeing the model sees or has identified that would correlate with success.

Bias is always an important question in these sort of programs. I would say a couple things. One is we use an ensemble of different modeling techniques to try to account for that. The other thing I love about r4’s approach is there’s always a human in the loop. This is not automated actions like I just got orders from the service and I don’t even know where they came from. This is decision intelligence so that the people who serve in these functions and roles, the manpower planners, the recruiters, the people that do assignments, they are in the loop. This is just allowing them insights, predictions, optimizations that they can make better and quicker decisions over time, if that makes sense.

Terry Gerton: It does. And Paul, did you want to follow up on that?

Paul Breitenbach: Yeah, one of the things that, building on what Ty just said is so critical in eliminating bias, is using this huge disparate sources of data. Normally bias is created because you’re using a very limited amount of data because that’s, in the old world, the only thing you could do. So what’s so powerful about cross-enterprise new AI technology is by looking through hundreds of different data sources, you actually reduce bias by literally finding signals that are through all kinds of different data sources, and that’s one of the most powerful boosters to what makes this so different.

Tyler Zagurski: Yeah, and I think I tripped over a question you asked in that as well, is what kind of data were we looking at? And so, to that silo question, we were able to find data in various different systems, and it’s the same in most services, the people who do recruiting have their data, they have their system. The people who do training and education have their data, have their systems. The people who do assignments have their data, have their system. Well, it’s all about the same human, but it resides in different places and it never gets together to do anything.

We’re also always very careful to add external data, sort of the economics of who people are and where they come from, the demographics of where people live and who they are, becomes very important in what Paul talked about and just adding texture and richness to understanding, I would say, archetypes. Like, what’s the archetype of success I’m looking for? I like to think of it almost like a fingerprint. I found the fingerprint of the human I’m looking for. Now I can use the same model to find where they live.

And we’re even doing things like zip code lookalikes. As we sort of predict probability by zip code, all 42,000 zip codes in the country, we can then turn around and see where success lies and then find the lookalike zip codes. And I don’t mean just size, shape, population. I mean, what rich attributes exist about some zip code that would lead me to believe that this zip code looks like that one. And so it’s really powerful to be able to identify drivers and counter drivers, factors of the fingerprint of who I’m looking for, and now let’s look at where those people are. And that’s kind of the approach.

Terry Gerton: I’m speaking with Paul Breitenbach, he’s the founder and CEO of r4 Technologies, and retired U.S. Marine Colonel Tyler Zagurski is r4’s Vice President of Talent and Workforce Solutions. So Paul, let me come back to you. I think I’m a little nervous about what Tyler just said. So when we think about applying AI to the future of recruiting, does it really replace human judgment? And how does it, I don’t know, keep a modicum of privacy in there?

Paul Breitenbach: Well, this is what makes new AI, predictive AI, so powerful. This is very different than traditional LLMs, where they just repeat things. What we’re talking about is a deep tech capability that can be deployed in classified environments. So all the data is protected at the same standard, the rest of the military standard.

But what I think is so cool about this is it literally creates more capability in recruiters. So this is about actually accelerating the recruiter’s ability to find and match people in roles and then help promote them. This is not about replacing those people. I think a lot of people talk about AI is going to be eliminating jobs. I actually think, we think the opposite, where this is going to the biggest boon of human productivity by helping match people who otherwise would never have known their skillset, might be highly relevant for an incredibly fulfilling, satisfying job.

And so by taking all together, by using the new AI, the cross-enterprise approach, like Tyler was saying, this is, I think, the biggest accelerant to employing more people to more fulfilling careers that we’ve ever seen in humanity.

Terry Gerton: Well, Tyler, let me come back to you because we talked about your 1 million, or 1.5 million, records, but you narrowed the pool of high potential candidates to 5,000 people. What happens now with those 5,000? Is the Air Force reaching out to them? Do they know that they’ve been selected?

Tyler Zagurski: Yeah, it’s a great question. And I would say the number 5,000 kind of varies depending on what model we’re looking at it through, we can sort of run predictions in a number of ways. But the point is, yes, we’re able to find needles in a haystack. These are exquisite individuals who have the aptitude, the potential, the fortitude, physical capability to exceed at something very rigorous.

The ways the services are looking at using this right now is sort of quantity and quality, like if I have folks in the pipeline, how can I match people to the right roles? And I would just add, for us, this isn’t just about one community and one service. This is applicable to any service. You could be looking for cyber warfare individuals. You could looking for people that will exceed in a certain aeronautics field or community. And so the system can function in a number of ways to help across communities and across services.

But essentially the approach is this, like in the pipeline, you have X number of people. The system can help you match them to roles. And the way I think about talent management is an end-to-end system, and Paul kind of hit on that earlier. If I can better match people to roles, they perform better, they’re more satisfied, more likely to retain. And I can avoid that enormous cost of finding a replacement for somebody who attrited or decided not to stay in the service.

And then the quality piece is important as well. And I’ll just give you an example. If I have to send, I’ll use a round number, I have to send 200 people to the next phase of this pipeline, I’ve got 600 available that are in development, I’ve gotten all kinds of data on them, we’ve been giving them tests, we’ve got just a good profile of who’s who, you can use this system to sort of optimize of the 500, who’s the 200 you’re going to send.

And then you start to immediately get at attrition rates. And if you think of the special warfare communities, that very tippy end of training, there’s anywhere from 50 to 80% attrition through these schools. If I can better shape who’s ready, who has the highest probability of succeeding, I can still continue to develop the other candidates, I can still continue to bring in candidates, but in this way I’m really optimizing the throughput, I’m lowering my attrition, I’m lowering my cost, and those are the kind of ways that this can be used.

Terry Gerton: Paul, let me kind of wrap this up with you. How will you measure the success of this project? And what are you thinking comes next?

Paul Breitenbach: This is all about readiness. It’s about out-teching the adversary. And for that, you need the right people in the right jobs. And then you need the right supply and logistics, the right readiness of the equipment — and I think this is to Ty’s point — this is all measured in outcomes now.

And this is, so first is getting the talent and the pipeline to work. And then the next part is getting the asset, the weapons systems behind it with a supply chain and logistics, so you have the right person in the right job, you have the right weapons system that they need, it has the right equipment and it’s ready to fight.

And this heightened state of readiness is really the big idea. That is what is coming next, a full total picture of, are we ready? And this is the out-teching of the adversary that makes this so compelling, because we are, we have the greatest military on earth, and if we can bring commercial off-the-shelf AI technology to help dramatically increase readiness across the entire ecosystem that we need in order to really promote national security, this is really the big idea.

And what Ty is talking about that’s so neat is about flipping the paradigm of a 50 to 80% failure rate to a 20% failure rate. I mean, that is game changing for everybody involved. And this is what I think makes this a really exciting time in this golden age of applying new AI capabilities in a highly secure way, is really what I think, is just we’re just at the beginning. And that’s what makes it so exciting.