Why Smart GovCon Bidders Are Turning to AI with Alex Cohen

Alex Cohen  00:00
The trouble with the government is one or two little things can throw you off, and then you're underwater, and that's bad for the government, because now you're going to start trying to cut corners to make back the profit that you lost. So once in a while, of course, it's going to get it right, and it's not going to be super wrong, but $10,000 can be the difference between losing and making money on a track like that. And so you have to be super careful. You have to be super careful with pricing.
 
Intro 00:26
We choose to go to the moon in this decade and do the other things, not because they are easy, but because they are hard. Through our blood in your bonds, we crushed the Germans before they got here. You and I have a rendezvous with destiny.
 
Alex Cohen  00:49
First exposure to GovCon was more than 10 years ago. I when I graduated from college, I worked at a bunch of startups. I started my own company, and we raised VC money and set it on fire. And, you know, had I learned a lot I called my first MBA, but when I stopped doing that, I started working for a company called COVID, which still does supply chain intelligence, but at the time, they were doing market intelligence in the GOV space. And so I started doing sales there, and I learned, like, how to spell IDIQ, and it was, you know, fantastic opportunity. And that was kind of my exposure to acquisitions and markets and mapping and all this stuff. So they were competing, like, at the time, with begov and gov, when IQ and that type of thing. From there, I started working in education technology for a little while, and then I got looped back in because one of my customers, actually from there, called me up more like, Hey, you're perfect for you sold us this product. We want a contract. We need a PM, and we need them to start like, really soon, yeah. And it's to do ed tech, like, purchases. You're going to work in acquisitions for the Bureau of Indian educate, Indian education. And so I went to work in acquisitions as a, you know, contract specialist. Ended up becoming a acquisition program manager. I learned FAR part, you know, every FAR part, everything, yeah. And, you know, I had a team of 33 contract specialists that that reported to me. We managed a couple billion dollars of acquisitions. And so I thought it was fascinating. I really enjoyed the mission. I enjoyed helping, like, buy the goods and services to reinvent the Indian education system in America for Native American schools. And then I actually went and I got my MBA, and I started writing proposals at that point, because I sat on technical evaluation panels and read 1000s and 1000s of proposals and bids and helped make recommendations for award. Obviously, as a contract specialist, I couldn't make any determinations. But after, you know, I worked with a bunch of clients, we won, you know, we did like 40 pursuits in two years. We won a dozen of them, you know, more than $140 million worth of federal work.
 
Ryan Connell  03:17
It's a pretty good rate, right? I mean, yeah, win rate.
 
Alex Cohen  03:21
A decent P win, decent fee, win so, and then I was working, you know, they offered me a job, but since that was the world I came from, and I just got a shiny, expensive MBA, I was like, let me try and do something else. So I got a job at Google, and I was working at Google, and I had access to a bunch of these AI models and chatgpt came out, and I started feeding them the largest and most complex documents available that I knew about, which are government contracts, large idiqs, you know, those types, you know, PPAs and asking questions and like diving into it from a Kos, you know, contracting officer perspective, putting my 1102 hat on, and I saw That was a little fine tuning and a little training, your response correctness went up about 100x so just asking chat, GBT, here's a document respond versus giving it a bunch of examples at you know, having it have context from the far doing some training and some fine tuning and rag on those elements, you could get 100x better, better more informative, more correct answer. And I was like, this is a significant difference. My initial instinct was, the government should be using this to write, like write solicitations. I approached a number of government agencies about that. They were just not in the mindset at the time. I think that there's more interest in that now.
 
Ryan Connell  04:41
What year was this? Just like level set like.
 
Alex Cohen  04:43
This was 2020, January of 2023, okay, all right, yeah, so last year? Yeah, as crazy as it sounds, it was. This was early last year. And so what I did was I started using it for my proposal writing clients. I went back to them and said, Hey, I. Could I create a custom, customized AI to help you guys write proposals? And they had been asking me kind of constantly, like, oh, whenever you know, if you're ever free, like, let us know. We would love to use you again as a writer. And so that was, those were my first clients. So I kind of came up with the idea, like, I turned the government side idea around, and was like, I'll use it for gov cons. And so that kind of those became the constraints. I was like, I know gov cons. I know what they need small, medium sized businesses. I figured, like, the large guys are going to do this for themselves. So how do we create something that's going to work for small and mid size? And then obviously, like, this entire space is about trusts and security and compliance. So how do we make something that's highly secure? So it took a long time to get to the right stuff on that, and I can dive in there too, but that's kind of the that's been the journey.
 
Ryan Connell  05:53
Yeah, no, that's incredible. So a lot in a short amount of time. I think the one of my weird takeaways from that whole thing was, who gets into gov con? Like, right out of college? Like, that's, that's like, I never heard of GovCon. When I was a senior, I was doing far different things that are outside the scope of this podcast. So what like, how do you even like, what was that entry? Like, your, your, your may graduation. Like, how did you even get into GovCon?
 
Alex Cohen  06:15
So, yeah, I mean, I didn't get into it right out of college. I had started a company, and when my company went under, we wound it down, and I was looking for a job, and I knew I wanted to do sales. I, you know, it's a skill. I believe in stacking skills, and so I was pretty good at it for my company, and I wanted to get better. So I looked at, you know, I'm from the DC area, you know, I'm a DC area kid, and so I was looking at companies, and there was a startup. It was series A they'd raised a bundle of money. It was bright and shiny. I didn't know what GovCon was. I was just like, oh, it's, you know, it's a startup. I like startups. It's a, you know, in the area, I won't have to commute very far. And so I took the job, and they were like, All right, we're gonna teach you GovCon 101, you have one week to learn how to spell RFI and, like, that was it, you know. So, that was my intro.
 
Ryan Connell  07:10
Got it? No, that helps. All right, so let's dive into, like, the AI and acquisition thing. I'm curious, because there's so many use cases, right? Like, there's, like, so many different elements. Is there something that you think is, like, the niche or you're focused on? Or like, where you think there's the biggest bang for your buck in terms of using artificial intelligence, like, what part of the acquisition cycle, or all of it, like, where are you focused?
 
Alex Cohen  07:33
Yeah, I think that we see on our website, pink team drafts is the key. I think that one of the hardest parts is staring at a blank page, because if you have written proposals, you get to this point where you're like, Oh man, I've read the solicitation however, many times, and I know that we've written a bunch of stuff about this before, and I can either go and collect it and copy paste it into a document, but that's not going to be the final document. I need to kind of do this thing where I stare at a blank page, come up with an outline, start populating sections, and just, you know, getting bullets on the page. AI is just extraordinarily good at that, especially if you have trained it. And that's where we focus is, how do we get a contractor from a quick read of solicitation to see if it makes sense for their business to I have the right number of pages, and I have roughly the right stuff on every page, which is like somewhere between pink and red team as quickly as possible. And so we've taken that process. I mean, you know, everyone who listens to your podcast probably knows roughly how long that takes in their mind that's, you know, a couple days a week, you know, maybe 10 days, depending on how many other things you have flying around. And we've taken that to taking a couple of hours, like, two hours or so,
 
Ryan Connell  08:54
Yeah, do you find any additional, like, more or less success with the AI generated proposals?
 
Alex Cohen  09:01
It's hard because a bunch of our customers are really small, so they're doing like, 15 million a year, you know, they're maybe 100 people, maybe a little bit more. And so they're used to doing like, you know, 1015, proposals a year. And so when they go to doing, you know, like 20 or 30 or 35 proposals a year. Obviously, they were winning like 80 or 90% because they were doing so much capture, because these are the opportunities, and they had people in their staff, and now they're kind of growing and expanding. So they're they go from, you know, 90% win rate to 50% win rate. So it's a huge drop, but their dollars go up significantly as well. A number of wins goes up significantly. Well, our larger customers don't see as like, almost any change whatsoever they get. They get a lower lift in terms of their volume. They usually increase by 20 or 25% the number of proposals they get out the door. Because once you're getting out like 40 proposals a year, you have an operation. And you have a team, you have staff, you like, this is a well known, well oiled machine, at least somewhat. I mean, it's never perfect, but so they'll see like a 20, 25% lift, and their P win won't go down very significantly. It won't go up very significantly either. You know, the limitations that we've found are much more around. You know, AI doesn't do a good job at pricing AI and resumes. We haven't, I think there's other companies that focus on this. We haven't done a hardcore focus on, like, of our bank of resumes, of everyone who's the best fit, who should we propose? That's something that humans still have, I think significant edge on is, like, who's going to be the right fit for this proposal, and why? Who are the key personnel we should propose?
 
Ryan Connell  10:45
So I love this, right? So we get from a point, and I'm going to kind of play both sides of the equation, right? So get to a point where you have an AI generated proposal on behalf of a potentially AI generated solicitation, or human generated solicitation. We don't know. It feels this is Ryan's opinion, right? Like it feels like we're in a spot right now where the culture is accepting of AI generate proposals. But I don't know that. I feel like industry would be acceptance of accepting of AI generate proposal evaluation. And I'm curious what your thoughts are on that.
 
Alex Cohen  11:18
Yeah, I think it's really interesting. I think that there's, there's two layers to at least two layers to evaluation. I say a lot. I said this, I say this. I said this, my clients, before I say this to our customers. Now, proposals aren't read. They're scored because you're not you should be write it. You should be writing it to be scored a specific way. I think that the base layer of compliance is complicated enough as on its own. So writing a compliant proposal, sometimes the solicitation is going to be self contradictory. Sometimes it's going to be unclear, you know? Sometimes you just don't have enough space. They're trying to get, you know, fill, put 10 gallons at a five gallon sack. So you have to choose and make strategic choices. But I do think that AI can, I mean, we've seen, we've proven out. AI can do a very good job with evaluating compliance. Did you meet the shell statements? If so, where find the exact quotes prove it to me. Okay, great. In terms of actually evaluating a proposal, we have 11 Oh, twos on staff. You know, I I love contracting officers. I think that their brains are really interesting. I think their understanding of the FAR is super interesting. And so we've had them compare outputs of an AI trying to evaluate a proposal to what they would rank it. And there's, it's so subjective. I mean the adjectival ratings and where you'd put something, and why? How do you stress something? I mean thinking of the rubric that the government will use to evaluate a certain proposal is already highly speculative, and, you know, can give you a large leg up if you if you have information on that. So I think that some basic stuff, like compliance should, like, I've been part of those teams where you read a proposal, and on page 47 of 50. You find something that they were supposed to have signed and they didn't sign it, and then you How many hours did you waste reading those 47 pages? Like, you know, so like, if it can find that stuff and save the government massive time, awesome. But in terms of, you know, choosing the awardee, I'm not, I'm I agree with you, I'd say it's a little tenuous.
 
Ryan Connell  13:42
Yeah, it's interesting, right? So like we we've come to a conclusion as a society or community that we're okay with AI generating a requirement statement and solicitation and the proposal, but when it comes to that decision, we're not really there yet. So here's a we're gonna go down the whole acquisition cycle. So so in that proposal evaluation world, I've seen companies that kind of offer a myriad of things but but one is teaching the artificial intelligence or the or the knowledge base, effectively the like, starting with like, here's human like, here's human evaluators. Like, here's how they scored a proposal. I want you to think like human and, and here's some examples. And learn based on that. There's that, and then there's just, like, the vanilla compliance thing that you kind of talked about, of, like, did this match this, if not, tell me where it didn't right, or vice versa. So I was on a panel recently, and we and this came up in terms of that from the standpoint of bias, and we used AI, or the conversation was that AI can remove bias, because no longer is it me that might want to do business with one company. It's like, did this match this? How well did it match? So, like, it's interesting, because I thought that having human judgment, He. Evaluations was a benefit, and then now I'm like, Well, maybe it's not so I'm curious if you have any thoughts on that.
 
Alex Cohen  15:06
AI is really interesting because it can also introduce its own bias. And the thing is, it's totally unintentional bias. I'll take a very innocuous example. They trained an AI to try and look at a mole or a or a freckle and see if it had cancer or not. And they gave it a bunch of freckles that were cancer and a bunch of freckles that weren't cancer. And then they started testing it on things that wasn't trained on. And what they found is that there was this unintentional thing that happened, where if in the image, there was a ruler next to the freckle. It ranked it as cancer 99.99% of the time, because all the training images had this little ruler to see how big it was. And so, you know, you can introduce bias that can be super get hard coded into the model depending on what you train it on. And you know, large companies like Google have faced this type of problem, you know, a lot with, you know, racial bias in their models and other things so, and they've tried to counteract this. It's definitely an open thing. I wouldn't say that, you know, I think that you need to have both. I think, like the Centaur approach of like, part AI, part human. Working together is definitely the way the industry is going to advance and move forward. So it's about giving the people the tools. I don't think that. I think we can work Work smarter together.
 
Ryan Connell  16:34
I like that center approach I wrote down. I'm going to use that you mentioned something earlier. So I'm kind of again, going down that acquisition cycle, kind of in the term of proposal valuation. But you said the AI is not really good with pricing. And I'm curious, if you want to unpack that a little bit.
 
Alex Cohen  16:49
Pricing is one of those things where you need to not only know about your own, you know, make choices internally about your own company of all right, who am I going to staff this with? How many people are they going to need? What's the specific solution? How long is it going to take? You need to make a bunch of those assumptions. And then you also need to think in a competitive landscape of, okay, there's subcontractor requirements. What are my you know, what's what's my wrap rate, what are my subs, rap rates? How much do I need to sub this out? And then you also need to think about, and that's kind of the partnership piece. And you also need to think about your competitors. Who's the incumbent? What subs do they have? How much do they charge? How much, you know, what was the ceiling? Did they come near it? There's so many factors that, you know, and AIs are not great with numbers. You have to kind of check their numeracy and make sure that they didn't kind of skip a step and do something slightly wrong that, I think, like the AI can create, can craft a great narrative if you have, like, a pricing narrative section of like, why? You know, if it's an RFI, like, why should this be firm fixed price? Why should it be cost plus, why should it be time, materials? Or if you have to do a narrative as part of your management plan, of like, well, here's the people we chose. And this is the, you know, these are the lcats labor categories that we wanted them at for, you know, this solution, it's awesome at that stuff, the backing into it. It's great at the coming from whole cloth. Of like, what should we price this at to win? I mean, people pay consultants price to win. Consultants that 10s of hundreds of 1000s of dollars for this stuff. If you could just ask an AI the question, you know, it's just just not there. I mean, maybe it will get there where it can run a lot of these workflows and do this complex analysis and give you a dashboard that'll help you. Think that's super fascinating. We've thought about that, but I worry that people think of it in more of a chat GPT type of way, where it's like, how much should we price this contract at beep? And like, it spits out an answer, and you're like, All right, cool. Well, let's price it that. And that's really not the way to do it in this industry. You'll you'll either lose a lot of money, or you won't win, because someone will have priced it more advantageously.
 
Ryan Connell  19:08
No, that's super interesting, because you're and Expectedly, your answer came from, like, the selling perspective about your actual realized risk and cost and all of that. And I was only thinking about it from the buying perspective, which is, like, how much is the solution worth? What's the value of it? I don't care what it costs you. I don't you know, like, because there are other companies, there's competition in the space all of that, right? I'll share a story. My My dad builds houses, general contractor, and he had a plan for, like, blueprints for a barn. And I dropped the two pager. It was, like, the floor plans, exterior, like rendering, and then, like, spec sheet, like, here's the material it's gonna get out of all that stuff into chat GPT. And just said, like, Hey, can you estimate this? It came back at, like, three, 360 1000 to $370,000 and I said that to my dad, and he's like, that is, like, spot on. What I'm going to charge this customer, and we were both kind of blown away, and so I've had an interest of pricing in my mind since that, but I've never dove in on the GovCon side, so maybe I'll proceed cautiously based on your recommendation.
 
Alex Cohen  20:14
The thing is, so the way that it works is it's looking for examples in the past, so the more that it knows about this situation and the bet, the better. It has a sense of exactly what you're doing and what you know, like the fact that you have all the plans drawn up, and it's exactly to, you know, to scope or to spec. It'll, it's going to do a better job assuming that it has a bunch of examples of that type of thing, or it can break it down, you know, has and has the right frameworks. The The trouble with the government is one or two little things can throw you off, and then you're underwater, and that's bad for the government, because now you're going to start trying to cut corners to make back the profit that you so, you know, yeah, you know, once in a while, of course, it's going to get it right, and it's not going to be super wrong, but I don't know what the margins are on. You know, construction, your dad's construction business, but like, you know, $10,000 can be the difference between losing and making money on a track like that. And so you have to be super careful. You have to be super careful with pricing.
 
Ryan Connell  21:19
Yeah, that makes sense. Is there anything you're excited about in this space in terms of, like, what's next? I mean, you talked about how in 2023, there was, like, zero adoption to this. Here we are. I mean, I know it was, I guess, more than 12 months, but we're in the next calendar year only, and there is adoption to this. Like, prediction, like, what do you think? Like, what are you excited about for next year?
 
Alex Cohen  21:42
I think from the government's perspective, I'm super excited for them to start using AI more in the acquisition process. You know, I think that you know anyone who's followed the acquisition community in 11, oh, twos and kind of what's going on? Like, there's been a lot of talk, especially more in the defense space than Civ side, but it kind of flows down about the bathtub effect and not having enough people to help the government in the ways that that it needs to, I think that it will, like the real, really great contracting officers, understand that their job is a people job, not a paper pushing job that they need to help their customer figure out the most advantageous way to acquire the goods and services that they need to run, whatever the function of the government is, and so it's they need to call people up and they need to ask them questions. They need to figure stuff out. They shouldn't be sitting there trying to, like, scratch their head and wonder, like, what should go in this statement of work? Or, you know, have a stack of, like, 13 acquisition strategies that they you know, what's the acquisition plan for this, that and the other? And like, Okay, well, I have to write two IGC ease and four statements of work, rewrite those. And so I'm hoping that AI can get a lot of that off their plate to make the process more you know. So I'm excited for that that development, as you know, as a major part of it on my side of the fence, I think that the voice side of AI is really an untapped potential. And so I think that growing everyone's ability to use voice as an interactive mode with AI is going to change things a lot in the next year.
 
Ryan Connell  23:21
Yeah, I love that. And I wondered if that was me as a personality trait. But sounds like, at least there's two of us. Like, I wrote a blog the other day while walking from the Crystal City metro stop to my building, which is like three blocks, just by talking out loud and said, Hey, here's what I need. Like, here's, you know, right? I mean, it was a draft one, right? It wasn't like the final product, but you start to think about that as a easier, like, I can talk and share IDs so much faster than I can type and, yeah, I mean, 100% I mean, the just, here's what I need. I need a solicitation. Here's what I'm looking for. Here's the, you know, specific requirement. But, yeah, that makes sense. I love the multi modal aspect, yeah, yeah.
 
Alex Cohen  24:04
And I think that from the proposal writing side, there's, I mean, from both sides, the funny thing is, like acquisitions, is this like meeting in the middle of these, these actual two separate things, because the government side, you have the program office that they need. They're the one that needs the good or service. They need, the thing to be done, widgets to arrive at location x, set time, whatever. And then, from the proposal writing side, they also they don't know the specific details. They need to go and talk to SMEs and find out what's the specific solution and how we're going to put the Legos together to build a boat or whatever it is. So since so much of that job, like, you know, I don't know how you played with the notebook? Lm, the the.
 
Ryan Connell  24:46
We did a defense Mavericks episode with it.
 
Alex Cohen  24:48
Sweet, yeah. But I think, like, there's a huge opportunity to collect a bunch of information by doing a one sided notebook. Lm, like, type of interview. Do with these SMEs. And then you get there's all the data is floating in there, and you can use an AI to cool down, like, what are the most important things, and have that stuff be well represented. And so, you know, I think we're playing with a bunch of things like that for the SMEs on, on the that side of the, that side of the fence, yeah.
 
Ryan Connell  25:17
I mean, even I don't know, like, as as a weird weekend hobby I've been playing with, Hey, Jen, introduced an avatar that can go to like they always had the avatar model where I could record myself talking for two minutes, and then they now have your avatar of you talking, your voice, your person, like your hands are moving, like they're moving now, but they just, I think last week or two weeks ago, pushed it to be live, to zoom so, like, I can send my avatar. I can teach it everything about defense Mavericks. I can teach about cdao and go attend the Zoom meeting. For me, it's not great, but it's, but it's like, it's a really cool beta test. And yeah, I mean, you can see, you can see the obvious benefits of things like that you mentioned earlier. We can dive into it, about the ATO, the cyber side of things, you see a lot of like AI tools that are pre CUI data, right? So, like defining a requirement, putting a solicitation out, right, writing, potentially writing your proposal, the market research. And then there almost is, like this screeching halt of Cape, like so many players are in what I just said, and then there's kind of a limited number of players that are in the I can generate a cui proposal. I can generate I can ingest your proposals and make recommendations. How is that process for you?
 
Alex Cohen  26:37
Yeah, it's super interesting. So we made a decision early on to basically customize an LLM and run it in a completely secure container so that it had no outside access to any other proposals, any other data, the open Internet. It's it's completely closed off. This has allowed us to, as opposed to, like, just building on il architecture, FedRAMP certified architecture. We actually have, uh, you know, our business model, each, each LLM is, is its own secure container. So I knew from writing proposals for a while in my career that eventually everyone is going to want that they're going to worry that somehow some of the knowledge would permeate from one model to another model to another model, and even though all these three customers are, you know, shouldn't have access to the same thing if they're using the same AI model. Ultimately, they're all going to have this mean reversion. It's all going to start sounding really similar, and the only way to get it to sound like your company or like your agency is to train a model. We built ours on the llama models, the open source models released by Facebook. We've kind of invested in that, in that community, and it's been awesome. But I think that, you know, the cmmc requirements just came out, so we can run these models, you know, in an environment that we are working on, getting them on a completely independent stack, so they could run on a laptop independently, that could let them run inside of a skiff. I think that would be huge. But I think that, you know, it's going to take a little while for all this to happen. I think that probably middle of next year, middle 2025, you'll start to see it becoming more more accepted and more ways to there's gonna be more avenues to do what you're doing, what you're talking about.
 
Ryan Connell  28:49
There's so many, it's like any boom, right? Like there's so many companies in this space, and they're, frankly, not all gonna make it forever, right? That's just general business, right? I'm curious, like, if you could give advice to the government, if that makes sense, in terms of, like, how do we experiment in a way where you're playing with all these different solutions every company wants to grow horizontally through that whole acquisition cycle and be able to provide that cradle to grave. It doesn't really make sense to buy instances of every company for their cradle to grave. So you're kind of like nitpicking like, Oh, this is neat from market research. Oh, this develops solicitations. Oh, this makes, you know, evaluate proposals. But none of those systems talk to each other, right, right?
 
Alex Cohen  29:43
But the government already has this problem. I mean, they have, you know, a bunch of agencies use SAP for this and prism for that, or PD squared for this, and, you know, like they, you know, so it's not, this isn't like a new problem that those integrations are going to come those are probably things by the end of next year, you'll start to see. Them so. But I think one of the big problems right now is that these big companies not to pick on on anyone in particular. But when I worked for the Bureau of Indian Affairs, we used prism, prism zone by unison. And you know, large companies just have harder, a harder time innovating. It's just the fundamental way. So the small companies are these little, agile, little ships and the big, you know, tankers of the world, you know, are they don't know how to fit AI in. They're like, oh, let's just add a chat bot on top of it. And, like, that's not going to solve the problem. So I think that those integrations are going to be built. It's going to take a little while these companies know they're probably going to end up acquiring some, you know, some of the better solutions out there. And, yeah, things are going to get easier, but it's going to be the wild west for another couple of years while, while that all that stuff happens. My advice for the government, your other part of your question would be, you know, in the meantime, while that's happening, you should totally experiment. You should be finding that's the way that the industry is going to do it. It's, you know, we run tests, we do experiments, we see what works, and not try and think about it. These as frameworks, and try and keep, you know, the maximum flexibility you can with your data. You know, I think that open source models are really great for that reason, because you don't have to commit to putting all of your data into, you know, one company's thing. If you have an open source model, it can go, you know, it's already made to be plug and play. But we, we really like the open source world. So, yeah, that's one thing. And then, you know, getting smart on it and understanding what you need to do because some of these capabilities. It's just you need the compute power to run it, and then you'll need people to build it on on top, you know, build specific applications on top of that. But getting the GPUs to actually run these types of AI solutions is the base level. So if you don't have a way of doing that, you're going to kind of run into the same problems over and over, regardless of how you try and acquire them.
 
Ryan Connell  32:11
All right, I got it so, so one, one final question. I like pulling this right, because I think it's, I would say, an interesting meditation, but an interesting just like thing to think about. So, so we talked AI for solicitation, AI for proposal development, AI for proposal review, AI for price analysis, right? Like for market research, that whole chain, if we get that figured out then, then why? Why not just cut out the entire process? And I'll put in three words of my problem statement is, and someone will tell me who to do business with.
 
Alex Cohen  32:43
I think that acquisitions is about managing risk. Every 1102 I talk to, when I talk to them about AI stuff, and, you know, eventually it comes down to like, Yeah, well, why don't you just ask the AI, which is the right vendor to do it, and then just sole source it to them and they, you know, the AI can do the JNA, but it's, you know, I think that the industry also has a has trouble understanding this. Of like, you know, what people ask, like, Why did NASA soup say that you had to disclose if you used AI in your proposal? And what they're looking for is confidence. You know, this is an industry where confidence really, really matters, not just because the ko will, you know, get called down to buy to the HCA office and say, Why did you make this choice? This was a terrible, you know, not just because there's protests that slow things down, but because I truly believe in my heart that the work that the government does for the American people is super important, and you don't want the risk of having a non compliant vendor that is trouble that ends up spending a million bucks just to find out that they didn't have the capability to do it in the first place, and you then have to spend a year reacquiring the same stupid good that you should have just done right the first time. Like I think that that type of cost is a form of waste and abuse that you know. So if it could guarantee that you get out of that, but it can't, you know, the truth of the matter that I believe in my heart is that our system is not the best system, it's not the fastest system, and it's not the most fair system, but it is better than any other country the way that they buy goods and services on Earth. So it's engineered to be fairer, faster and better than almost anything else out there. Of course we can improve. That's not to say there's not room for improvement, but it's a balance of that fairness and speed that we have come to and so I think if AI can help us increase speed by 10% by 50% that's great, but not at the expense of making it less fair. So that's still going to be, I think that those are still, that's still the gaps in the brakes of the so that doesn't change.
 
Ryan Connell  34:55
Oh, it makes a lot of sense. I said last question. But one more. One more quick one based on. That I have personally seen an influx of oral presentations, so like, almost the acquisition strategy of, hey, give me your AI proposal, and I'll read them. I'll pick the top 1015, whatever. I'll invite you all to then give me a presentation or a lead, give me a demo, et cetera. Have you seen that? And that kind of cuts out the AI part of it, right?
 
Alex Cohen  35:20
I don't think it fully cuts it out. I think what it does is it better aligns the incentives. I'm a big proponent of this. I think the government should move more to it. I think the AI part of it will, will change and adapt. We, we can make, you know, AI presentations based on, if you give me a proposal, we can turn around a presentation version in less than an hour, so it doesn't, you know, they you can have an AI translate a to b, b to a, you know, any sort of way. It's a translation machine. But what I like about the presentations is you have to have someone on your team that's competent, that can answer questions, that can show up, that knows the space, that has the chops. And when we're talking about like, what's the confidence? You know, this is an industry where confidence matters. You want the government to be confident in your solution. Then you want them to choose someone who's going to complete the work correctly, on time, without issues, etc, that will give the government confidence. And I think that's a great selection methodology. So I think getting your proposal selected is one thing, and then getting someone to come in and talk to that exact solution is an awesome way for the government to to select, you know, within sub, select, within or down, select, or, you know, whatever it is.
 
Ryan Connell  36:32
Makes sense. Hey, Alex, appreciate you being on today. This was an awesome conversation. Probably could have kept going for another hour if we wanted to. Yeah, thanks for being here.
 
Alex Cohen  36:41
Yeah, absolutely my pleasure. Thanks for having me.