Latest / Construction Technology Podcast / Spencer Thompson on Civil Precon, Geotech Risk, and AI Guardrails
Transcript
- Luigi La Corte: Instruction. Sound good? All right. Hello, everyone. Welcome to the construction technology podcast. I am one of your hosts, Luigi from Provision. We have Andrew, who's in a wood cabin. He's from Join. And we have we have Mike from BuildVishion. We're also joined by Spencer Thompson from Cardinal Civil Contracting. Spencer, how are you? Good morning. Spencer Thompson: doing well. Thank you for having me. Luigi La Corte: Yeah, of course, of course. we were chatting with Kyle from Hensel Phelps. He highly recommended we get you on. he said you have really good opinions about tools, productivity, how the industry's changing. Obviously if you're from a civil pre construction background, tell us a bit about yourself and and Cardinal. Spencer Thompson: Yeah. I I have a lot of opinions, whether they're good or not is subject or to change, I guess, depending on the viewer and listener. Luigi La Corte: They don't have to be good, they d they just have to be controversial. That's it. Spencer Thompson: I have a lot of those. yeah, so I'm I'm in my you know, I'm thirty seven years old. I've been in civil contracting for I mean, it seems like almost my whole life. My mom worked for an asphalt paving company. So I've been kind of like connected to it at some point. I started really young with like a seven dollar an hour job just filing papers into a filing cabinet. invoices and stuff. And over the years I learned job costing with like Sage Timberline way back in the day and payroll and really like an HR side of things, which was very interesting. I didn't really understand any of that working for a small business. and then I after college I ended up back in the industry and started learning takeoff tools and then estimating. And then you can kind of see the free flow of information with my experience with the payroll And job costing and like invoices and all that stuff. And then actually see it come to life when you start doing takeoffs and understanding like what all these things actually mean, you know, like the life cycle of the job. So yeah, so now I've I've been estimating for ten years, roughly. did a little bit of project management in between there. And yeah, I never want to leave this industry. I love it. It's awesome. Luigi La Corte: Why don't we go first into like what your day-to-day is like and and kind of the problems that you deal with? so like let's just start with the broad strokes, right? So civil pre-con, can you just disambiguate between buildings and kind of like vertical pre-con? Give us the what what a day in your life looks like and and and what that actually looks like relative to to building pre-con. Spencer Thompson: Sure. I'm gonna pick on all the building pre-con guys. I feel like their life is so easy inside the walls. What are these people doing? at Cardinal Civil Contracting, so we're a self-performed contractor, right? So we do site development in the southeast part of the United States. We're publicly traded. So there's a lot of estimators that work across the company, and we deal with a lot of pain points on the job. So we I I always like to think of civil contracting as like we're like the windshield on the car. Like we're the the first boots on the ground, we're going to discover a lot of the issues right off the bat before you even start putting footers in the ground to go upward. a lot of times like I'd say some of the big differences that we have between like the building pre-construction estimators and then the site development pre-construction estimators is the building stuff a lot of time is a lot of knowns, right? You have a lot of you know beams, roofing, HVAC system, MEP Lot of things are are pretty standardized across the board, depending on where you go. and then in civil contracting, when it comes to earthwork, it's different every job. It every the the the type of dirt, the weather conditions, you know, moisture content, plasticity, rock, all these things, job to job are so different and your approach has to be so different all the time. So typically that's a lot of times what I see. like big differences, but the workflow itself. Like when you actually like zoom out and look at like kind of the workflow is is almost identical. We all do takeoffs, we put it in our estimating software, we review the bid, we send it out to the client. I mean, it's basically the same thing. It's really just the pain points that that we encounter with our scopes of work that are just dramatically different. Luigi La Corte: So there's a lot more unknowns, not necessarily unknown unknowns, like you you know the unknowns. You just don't know without the borehole logs or the kind of soil samples how it's gonna impact horizontal construction. You're saying like that's the biggest difference between the two. Spencer Thompson: Yeah, typically we and and that is not even typically that is the biggest difference. We rely a lot on due diligence from our partners up front, you know, when they're purchasing the property, who they purchase it from, who was actually in charge of going out there and doing the boreholes, the bore logs, the geotechnical report information. You know, w even when the LiDAR was flown for the site, did they pull it from a GIS or they pulling this from an actual surveyor? All of these Luigi La Corte: Mm. Spencer Thompson: things change job to job. so trying to ha get all that information and collect it up front and then make those recommendations to our partners, like, hey guys, you know, we need to do some more boards here. We don't really have enough information to make reasonable assumptions. We can make assumptions, but are they reasonable? Is it really giving you what you want? Or, you know, what what are we solving for here? So most jobs that we interact with on the first pass, I feel like I have more questions. Right. And then as we we go further along into those bid sets, just like you're doing with anything else in this industry, whether it's vertical or horizontal, you know, the first pass is always I have a lot of questions. And then you start trying to hone that in and dial it in with them and be you know, really a true partner with them. Luigi La Corte: So the majority of the difference just risk transfer? Like, because to me it seems like look, as a as a lay person, it seems it seems simpler, right? Like, yes, you are self-performing. So you are you're going into a level of detail that let's say a general contractor who doesn't self-perform would, right? You're obviously you're counting everything because you're taking the risk. But let's say if you're doing like roadways, like a highway, there's there's quite a few structures there, but there's still gonna be less divisions than a GC. So is the is the biggest difference in this question? really just risk transfer because I just do not know that the the subsurface conditions. Spencer Thompson: Sure. That's one aspect of it. The I think a larger aspect of it is the fact that you have self-perform work, right? So we're dealing with thousands of assets that we own. Right. So whereas if I'm a a lot of these vertical GCs, they may do some self-perform work at like a foundational level or some concrete work, some footer work. We're literally doing everything. So from grading to you underground utilities, water paving. Rock blasting, retaining walls, we build it all. So it risk transfer like scope gaps, like we are the scope gap per people. There are no scope gaps with us. Like we are it. So it it's not like understanding value of assets is also like huge to us. Like fuel burn, telematics on machines, things like that. Like there's a whole nother piece to it that comes into play that a lot of vertical GCs don't necessarily have to deal with because they're taking that information from us being the tier one subcontractor. Right? They're they're just looking at us as sub plus markup. And so there's like Mike Powers: Yeah. Spencer Thompson: another layer below that that we're having to deal with. Mike Powers: So does does that mean that you all like so for like a general contractor is like a three percent margin fee structure, like in the best case? Are you all able to like charge a higher fee structure because you're absorbing that risk? And then do you like try to like de risk it downstream? Or is it very much like, hey, we're just gonna like let it rip and hope for the best? Spencer Thompson: Yeah, so I don't know how much I can get into that. but what I will say is Mike Powers: At a normal civil contractor in Mark don't worry, not you guys, but hey, hey, like if you were to look at a business of a civil a large cyber versus Turner, Turner has a one and a half percent fee structure, what's what's the average? Spencer Thompson: Yeah, so i it it it's gonna range based on market conditions, I would say. Typically you don't see a fee structure like that. That's when you get that that above us, right? So we're tier one and then the GCs Mike Powers: Yeah, yeah. Spencer Thompson: hold that model of a fee structure because they're looking at everything through the lens of sub plus markup, Mike Powers: Yeah. Spencer Thompson: whereas our fee structure is completely different. Yeah, so it flu it can fluctuate. Market market conditions, job type, risk, it all fluctuates. Mike Powers: Are you like open and honest with the owner about that? Being like, hey, like we have no idea what subgrade here, so we're gonna charge a premium. Just as an example, like so my first job out of college, I worked at the Eagle Stadium. we did renovations there. Turner had won the project and built the project in whatever, 2006, and these were renovations 10 years later. And we won the project in the idea that like we knew where bedrock was. I spent my entire winter digging micropiles in the parking lot trying to find bedrock. Because we had no idea where it was, but we like sold the owner on that idea. And I'm curious, like, how does that like that's been my only experience in the kind of the civil groundwork world? And I'm Spencer Thompson: Yeah. Mike Powers: curious, like, how is that consistent in market, or are you very much like, hey, we need like we we know exactly what we're going into, we know exactly how to structure this, we know exactly how to like kind of CYA. Spencer Thompson: Yeah, so what I would like to say it like this, just to make sure I word it very, very clearly, is chart pricing something based on risk is not necessarily the best model because you never know what you're gonna get, right? And we don't think that that's really a partnership. I would say that what we will do is clarify and quantify what we do know and then allocate a cost structure to the unknowns in case they do happen, right? But we're not just gonna say, hey this is what we know. And because there's a lot of unknowns, we're just gonna charge you more money for the knowns. Mike Powers: Mm-hmm. Spencer Thompson: It's hey, these are the knowns. We will give you the standard, you know, markup that we typically would expect on a project like this. But here's all these buckets. Like, think of it like levers you're pulling on a machine. We want to provide all the levers to you for the scenarios in which, in case they do happen, you understand that there is a value to that unknown. So you use the bedrock example. That would be a good one, right? So we could say, hey, you know, your boreholes only go down 10 feet, but we're actually cutting to 20 feet. So there's 10 foot underneath there that we actually don't know what it is. So here's three different options that you can look at that has a dollar value to it for that quantity. We can assume all 10 feet of it is rock. We can assume all 10 feet of it is dirt. We can assume half and half. How would you like us to proceed? Right. And try to put that back as like a good partnership here is like there's a lot of different Mike Powers: Yeah. Spencer Thompson: ways to skin the cat. What what risk are you willing to accept? You know? But just providing information Mike Powers: So I guess like Yeah. I guess that kind of leans into like, you know, what you're really passionate about is like leveraging AI. Do this. Like for example, we see customers that like we do a lot of things with like chiller selections. And with AI, you instead historically you make one chiller selection for an engineering firm. Well with AI, you can actually make a million or a trillion AI selections for chillers because and you can give the owner that kind of exposure. What are you how are you leveraging AI and like partnering with your owner customers to do these types of things and give them like 'cause I'm guessing there's a trillion iterations that you could kind of like walk through in that example. Spencer Thompson: Yeah, and and s okay, so specifically about the AI example, the I wouldn't say AI is being leveraged in in in that sense, but what it's really good at for what we can find value in it would just be organization of data that we've already calculated. So you know, we're not using AI and we I haven't came across an AI tool that can do a three-dimensional modeling that is anywhere close to the capabilities we have now. So I would say, hey, we spend all this time Collecting these quantities, we can actually pair AI up to look to reference a geotechnical report and say, hey, look, based on the geotechnical report and these quantities and these assumptions we've made, you know, give us like 50 options, you know, 20, the top Mike Powers: Yeah. Spencer Thompson: 10 most common things you could see. And then we would vet that, right? So we can read through it and just say, this one actually doesn't make any sense at all, right? Trust but verify. This one makes most sense based on where we are. You know, AI's just gonna look at it as a as a total, like everything it can see in a almost like a way I look at it is like a memory bank. Whereas maybe we're only specifically know like in this area of where we are in North Carolina, we know that out of these ten options, three of these are the most plausible from what we've seen, and the other six are not out of the ten. Right. So we can use it in those ways to help us out. Mike Powers: Interest. just one thing that was just popped in my head was like pulling all geotechnical reports for a market and just being like, what's the estimate on these? I don't know. That that's where my I just I think there's a lot of fascinating things that can happen, like where like before Spencer Thompson: Yeah. Mike Powers: it probably doesn't make sense for you to go look at 10,000 documents in that market to be like, hey, like what's the risk here? But with AI, you're like, hey, I can go ping Gemini's API, and then we'll actually just go read the 10,000 documents and it'll give you a list of here's the 10 risks that you can think about, and you're like, Here's two that are interesting to me. Spencer Thompson: Yeah, and and that's what's interesting about it. so I have used a few AI tools to assess some geotechnical reports, a lot of different actually geotechnical reports, but a few AI tools, a lot of different geotechnical reports. And ultimately what I can gather from it is it it doesn't recommend anything that I would not have recommended. You know what I mean? It it didn't show me anything that I wouldn't have already noticed because It's it's only comparing the knowns and I know and we know the knowns where we're at in the region. You know what I mean? It's I don't know how else to explain it other than that. So like I Mike Powers: Yeah. Spencer Thompson: know on the south side of of this region that there's a lot of rock. It's like, okay, so every Geotech report I ever have just confirms what I already knew. So the recommendations are only pulling from things I already knew. Y you know what I mean? Mike Powers: Yeah. Luigi La Corte: Mm. Mike Powers: Is there any proprietary Andrew Zukoski: Are there? Mike Powers: data sets out there that like could be leveraged? Sorry, go ahead, Andrew. Andrew Zukoski: I was going to ask a very similar question, just like one of the one of the things we've observed in our business and talking to other people is right, AI makes it a lot easier to sort through just a lot of information that's out there. And I got I I don't know very much about the subsurface world, I'm sort of happy to say, because it seems really complicated. But are there other data sets out there that you can now access or that you would be able to at least point AI at that would tell you about like, you know, what's in this particular piece of dirt? Spencer Thompson: That's a good question. I'm sure one exists. I would feel more comfortable if I built my own based on the data that we have in our market. only because we can how how do I I want to say this like in the in the right way. So a lot of things that end up in a geotechnical report are derived based on recommendations that we have made to them. Does that make sense? So So the first pass that you're going to get at a geotechnical report is going to check the box in an acquisitional phase so that someone can purchase the property, right? Because they don't want to purchase a property and build a subdivision on it and to go through all these design fees if it can't even work. So whoever they're buying the property from, whether that's a farmer or another land developer or anything like that, they go out there and they meet the bare minimum requirements. Just put 20 boreholes Andrew Zukoski: Mm. Spencer Thompson: on 100 acres. Well, that doesn't do us any good, right? So We take that information and we say, Look, based on your first pass at a design, we can see the sewer, we can see the water, we can see where all these utility ties in, we can see your deep cuts. We recommend that you go out and do all of these things. They go Andrew Zukoski: Mm. Spencer Thompson: out, they spend the, you know, 30, 40 grand or whatever, they run some tests, some more tests that we recommend. Then that gets uploaded back into another geotechnical report. So if you were to have a like a a preset like database that it's pulling the geotechnical information from, which iteration would it be? Because the second iteration that I said we recommend, it wouldn't exist into some database unless we force it to exist because we are the contractor at that point. You y you know what I mean? Does that make sense? Like we we are owning that. Andrew Zukoski: And is the like what you ask for, is that like Yeah. Like is what's in that what you ask for, is that specific to the project? Like if it's like you ask for different things, if it's like a big retail super center versus you know, tracked homes or something like that. Like is it just like so specific to you Spencer Thompson: Yeah. Andrew Zukoski: and the project and the owner that no one else would be able to ask those questions? Spencer Thompson: So I would say yes and no. they may everyone can ask the question. It's just maybe they would select a different area to do something different with it. Like maybe they see something we don't see, right? That's possible. They may view the site, the way you move the dirt can vary from contractor to contractor, even from superintendent to superintendent, right? So we have so many meetings talking about the best way to means and methods to run this project before we even start it. So the the The first part of your question though about do we look for something different depending on what is actually going to be on the job? That's absolutely yes. So subdivisional requirements, they typically are surrounded by perimeter retaining walls to make the grades work. They have retaining walls surrounding the entire site. Commercial sites, because they're mostly a building pad, their grades are a lot of times different and they may not have as many walls because their sites their site footprint might be a little bit smaller. so Just the I I I talked about this on the advancing pre-construction podcast I did where we talked about wall backfill. And that's always a huge kick to us is wall backfill. where we at in the market, there's a very specific parameter that the dirt must meet to go inside a reinforced zone of a segmented block wall. Like it just has to happen. A lot of sites don't have that. So we always try to dig through the geotechnical report, understand where that dirt may exist on the site. And say we need these tests done in this area. Those tests can take up to like three or four weeks. So there's a complicated moving machine here where you may have to uncover that dirt to even test it. Well, you can't uncover the dirt to test it until you are awarded the project and you get out there. So there's a fine balance here of like risk, right? We were talking about risk earlier, is hey client, you know, do you want to wait on the project? Do you want to move forward with the project? How do you want to handle this? Do you want us to price, give you these levers and buckets to import it? How does that affect the balance of your job? There's a lot of variables within just a geotechnical report itself. It's really telling you what is on what is the dirt, how is it going to react, and how can you move it most efficiently? Andrew Zukoski: Feels like Luigi La Corte: we can talk quickly Spencer Thompson: I know it was a long winded answer. Luigi La Corte: about no no no that was a great answer maybe we can talk about the impact on the like the delivery model right so in in in horizontal jobs obviously you see like unit rate type jobs right give us your unit rates for moving you know x pounds of dirt or linear feet etc etc in Canada where where I'm based we're still seeing a lot of lump sum like full risk transfer models in horizontal And some of the the biggest challenging projects that we've had have been these horizontal jobs that are on a fixed cost, fixed schedule model. Eglinton Crosstown is very famous in that regard. Basically any LRT now is kind of doomed and cursed, for for a lot of reasons. What are you seeing as the evolution of the project delivery model in civil relative to buildings? Spencer Thompson: Yeah, that's a good question. So to elaborate on your first point, we actually in our markets that I've seen throughout the Southeast is we're operating, unless you're on DOT projects and public bids, those are a little bit different. So let's just talk private here. Those are all lump sum contracts, right? And they have unit prices built in a lot of times that that you know you can agree to, but overarchingly we assume the risk of the side of the lump sum level. So the contract's gonna be written like a lump sum, but we just provide unit rates to show very transparent scope. Like this is what we're charging you for every little widget on the job. I don't see it moving away from that. the reason I say I don't ever see it moving away from that to get to a pad delivery model for a contractor we're working for is because if it becomes more of a unit price model, is you they don't underst that that can fluctuate on the job. they they want to know up front, like this is my this is my my bucket of money based on the scope. If you miss something, you miss something. If you, if you hit something, you know, that's on you, right? You assume all the risk. This is my fixed cost moving into it. When you get into unit prices about, well, did we really install that? Like in a DOT, if you if you didn't do it, they're only going to pay you for what you did and they want the money back. And then if you go over it, they'll pay you more for the money that you go over it. And I don't I haven't met a general contractor in the building world that ever wants to go to that model. Cause that that's more change orders for them. And they're have they don't want to have to go back to the bank for anything. Luigi La Corte: Maybe in one way in which AI has been helping is helping parse through the these kind of mega documents that DOTs love to hand over, right? So in Canada, we have the Ministry of Transport, they have the Ministry of Transport design manual. This thing's gotta be like 20,000 pages, right? Contemplating every specific Spencer Thompson: Yeah, yeah. Luigi La Corte: edge case. There's like a billion contradictions. Spencer Thompson: Yeah. Luigi La Corte: I don't know if that's the specific use case that you focus on, but maybe give me the the kind of like the marquee use cases. yeah. In addition to what you've already said before, like where is it really making a difference now for you? Spencer Thompson: Yeah, so we we don't do a lot of public work from that sense. By the way, when you say ministries and stuff, I'm like, dude, what is this? Canada's like Harry Potter, man. We got the Ministry of Defense, the Minister of this. no, but the we we don't do a lot of public like major highway jobs like that, right? So that they're gonna have a ten thousand page spec book, which I'm familiar with, like airport jobs and stuff like that. We do a lot of private work. And with the private work, we follow municipality specifications a lot. So whatever the municipality is dictating or the geotechnical report is dictating is what we're gonna follow. And then the contract, whatever the contract says. Now, there are cases like data centers that we've done, Walmart distribution centers, things like that, they will come to the table with a spec book because when you get into larger builders like that, they want their own specifications followed. That is where AI can be leveraged heavily with specification reviews, pulling out scopes, things like that. I mean, AI helps us sort through that very quickly. Like very quickly. Luigi La Corte: And what about areas where let's say the marketing is overhyped? And maybe we can make this like its own little segment, right? So Kyle was very honest about where things are working and where things aren't working. Where are you seeing a lot of marketing hype, but it fails to translate into an actual good work product or output? You can talk categorically or you can talk about individual companies. It doesn't matter. You don't have to pull punches. I know your LinkedIn Spencer Thompson: No, that's a that's a great Luigi La Corte: on your LinkedIn you definitely don't pull punch, pull punches with with some people, which is great. Spencer Thompson: I've gotten in trouble a few times maybe on LinkedIn. no, I where okay, so let's start with like I'm a first principles guy. and I believe that Luigi La Corte: Mm-hmm. Spencer Thompson: I think me and Kyle share this like deeply, which is why we get along so well. Is we believe the takeoff is the basic of all truths on the job. If you break every issue down from finance to to operations and To estimating all the way back down to the first data point, it's gonna be your takeoff. And everything points back to it, like everything. And so I believe that your takeoff is your basic principle of it, that if you if you can solve the takeoff problem and standardize around your takeoff, everything downstream from that data point falls in line, right? So everything can point back to your takeoff. It should be where you live and die at. It's where you spend most of your time, where you learn the job, everything. I can go down a long rabbit hole with this. With that being said, that is the first place I see where in our industry it severely lacks is lacking hard. there are some really good two dimensional AI takeoff tools, and there are some really good that are not AI two dimensional takeoff tools. Where things do not exist is in a three dimensional space. There is no AI takeoff tool in a three dimensional space that I've discovered that's that's even getting close to increasing efficiency in a workflow. It's still antiquated, manual data entry. We're still doing a lot of that number punching. and and it doesn't seem like from where I'm sitting that it's getting much better over the next twelve months. You know, maybe in five years. Mike Powers: When you s when you say th yeah, when you say like three D world, you're like saying, Hey, I I'm on a I'm like walking a job site, that's being three D and like using cameras or whatever to capture that data, or are you seeing it like BIM? I'm guessing you're leaning more towards like the real world position of that doing Andrew Zukoski: Mm-hmm. Mike Powers: takeoffs? 'Cause like from BIM, like hey, if I'm if the guy runs thirty feet of duct work, like it it's it's in a model somewhere, it's in a database somewhere. Spencer Thompson: Yeah. Yeah, I guess I should clarify. I think everybody understands what I'm talking about. So when I say three-dimensional, I mean three-dimensional calculation. So underground utilities and earthwork, right? We need three-dimensional Mike Powers: Yeah. Spencer Thompson: modeling capabilities. We don't use BEM because that's vertical. So, Mike Powers: Mm-hmm. Spencer Thompson: you know, we're CAD people, right? So we need things that can process civil CAD files, which is, Mike Powers: Yeah. Spencer Thompson: you know, AgTech, Insight, TBC, some people use, right? But all of these tools, regardless of where you go in the market. It's still import CAD file, move data over. It's been the same workflow for years since I've been using them. Move data over, and then you manually have to type in the REM inverts, or maybe you can auto-assign them. You to type in the invert elevations, click and measure the pipe, and then it creates a profile for you. And then these softwares give you breakouts of your depths. And then you can base your production based on the depth of the pipe, right? But you have to do all of this information. All of this data entry comes from you. Whereas I'm looking at it on a s on a screen and I can quite literally still take to this day, I might have I have a scale ruler in here somewhere. I could go print out a 24 by 36 set of plans of scale with a scale ruler, and I could do a takeoff probably faster than what someone could do with the technology available to us sitting behind a computer. But the reason we want it done in these softwares is because it's trackable. drawings and sheets are not trackable. If I do it and it ends up in the trash can or I spill coffee on it, like who can find it, right? So this industry's has pivoted to software only takeoffs, which is great at the end of the day, but you're Luigi La Corte: Mm. Spencer Thompson: not really increasing speed in a lot of just just to get the measurements done. for especially for underground utilities. Mike Powers: Yeah. I think that like that's Spencer Thompson: Hope that made sense. Mike Powers: like the biggest del like so in the vertical world, like I would say the the first principal source of truth is actually like the model or however the engineer is designing the building or the architect design the building, not the takeoff, because like the fact that we like compress a 3D model into a two D model then go count grills and diffusers on a 2D document is like fascinating to me and it's frustrating that we actually do that. But in your world, because like you're really not like you're not creating a 3D dimension, you're not creating like a world model that like someone could be like, Hey, here's the rock. twenty feet down in this model, like go place your go place the roads on top of this rock model. Spencer Thompson: No, you're not cre you're not really generating models. You made for like one off exhibits to present to someone, but you're not like creating a model to save somewhere to reference later to provide to your client. That's yeah. That's why the BIM world is so interesting to me because I actually like the BIM world. I'm like, man, this is cool. I think this is kind of cool. you know, how do we figure this out in civil contracting, which we can have a whole separate conversation about real time cost capture and scanning and stuff? Like that's that's That's awesome to me. I'm trying to figure that out. Luigi La Corte: What about some specific tools? Mike Powers: Have you gone down the world of world models? Like have you started to look into like how some I mean, like Spencer Thompson: Ha ha Mike Powers: the new Astrodrop was pretty not even a world model, but what are your thoughts? Like I there's there's two different kinds of there's two kind of like there's LLMs and then these world models of like how do physics actually interact in the real environment. And it seems like the civil world probably should be on the other side of that lens. I think a lot of like mechanical systems are gonna be on the world model side, not the LLM side, because you have There's not an answer in mechanical systems. Like you could have the air from sixty five to seventy two, there's humidity, all these things. let's go down that rabbit hole. I that I that's like an old like a very technical question, but curious how you're thinking about those things. Spencer Thompson: Yeah, so I I I'm not too familiar with that specifically, but what what I would like to get to one day, and dream big with me here, is you know, la I was talking with I was at Hexagon Live, when was that, 2025, maybe 2024? And I spent a lot of time with Leica reps. And la I love Leica scanners. I think they're some of the best in the world. I think they do a lot of cool things in the geospatial system. And one thing if I remember correctly. Andrew Zukoski: So it was a it was a good dinner and party, yeah. Spencer Thompson: Yeah yeah. one thing I remember talking with somebody is they were going through the exercise of like creating a GIS of LIDAR of the entire world. Like that was like a project one of these companies was working on. And I said, Wow, that would be really cool if we could get our hands on that. So like let's just start there. Is I would love to have that type of detail across the United States alone. Like to be able to just pull up a a baseline of of where something exists and then be able to just start designing a site helping a client would be fantastic, right? The second piece to that would be, okay, so we we want to figure out, like I how about just me personally? We just leave Carnal. I like I would love to solve this is in this industry, is almost like a real-time cost capture. like we were talking about BEM a second ago, but The way I see AI really serving this industry, not just in pre-construction, but from an operation standpoint, is figuring out how to accurately tie systems together and pull data from a lot of different ways very quickly, integrating it and cycling it back through your bid process. So if I have a site, right, and I know what the lidar surface is of the site, like my beginning point. Well, we have grading models, right? That machines can see. It's a j it's a geospatial digital layer that machines can see. What's the job supposed to get to? And all the machines can see it. And it sits in a base station on site that I they all talk to. I would love to be able to figure out a way or some company figure out a way. There's some scratch in the surface, I think, is push costs down to some digital layer that sits over the site. So you have your LIDAR surface and then a digital layer of cost that kind of matches your grading model. Right, of what it's supposed to get to that overlaps it. And then by usage of scanner technologies, cameras, drones, whatever, all communicating together to understand when work was done. So, hey, work is done this day. Check that box off that this cost that you bid against is now done, right? Contracted price done versus what you bid it against. There's one mathematical formula. I would call that like phase one of my dream. Phase two of my dream would be okay, that's budgeted cost versus contracted price, right? Like what like what's the delta there? That doesn't really give you all the answers. That would just tell you like that you it's really telling you basically cost to complete, like, hey, as a project manager, I need to know that per the contract, this is what the dollars I have left to bill the client at the end of the month, right? So that's That's one piece. The second piece is like real-time cost capture. Telematics on your machines combined with some smart recognition software that can read machines, like the actual machine that's there, understand it pulls those costs from an ERP that we have and understands, hey, this John Deere 310 that's sitting on site with this model number, actually in our ERP costs this. And so we know that that machine right there, with based on its telematics running that long, is burning dollars, right? And and the laborer, you know, the guy sitting there, he costs this, the crew cost this, right? And so all these things come into play. Is someone needs to figure out someone needs to figure out like how do we capture that real time and compare that against the digital layer that that's on the site. And then understand that I'm sure it does. Yeah. Palantir, yeah. Mike Powers: There's a company that does it, it's just gonna cost you twenty five million dollars a year. It's what ca it's what Kavanaugh did. That's what Kavanaugh did pretty much. I that's what Kavanaugh's going towards with Palantir. Like it's just gonna cost you a shit ton of money. Spencer Thompson: Yeah. So so the Palanti like my understanding of the way Palantir's model works is interconnectivity, right? So like building connectors between data points that already exist. If you have the data, they can connect it. My thing is the data maybe exists. Like even Kavanaugh, I w maybe we should I would love to have conversations with those guys just to talk to But I would really even what I'm talking about, I don't think exists even with them because I don't know if they're getting even that granular. I think they're doing maybe the first phase of what I talked about, but the second phase is like a whole different ball game. Because how do you tell an excavator in real time with a scanner on it, like, hey, I actually picked up this pipe, it recognizes the type of pipe, it sticks the pipe in the ground, and a software is going to understand that this stick of pipe is 24 inch RCP and it costs like $30 a foot, right? Like based on the invoice that I've received and the purchase order and that lives in my ERP. Right. So you need you need like some type of scanning technology that is really, really good. And I don't even know what that looks like. So it at advancing cons yeah, go ahead. Mike Powers: Is that like is that Andrew Zukoski: What? Mike Powers: like a value accretive to you guys though? Like like knowing like w one thing I've always like thought through is like everything like like reactive does isn't as valuable as preactivity, like or preemptin? Like like saying that like, hey, we're dropping the pipe today at this exact time, does it actually matter if it happens in real time or it happens that day of? Like Spencer Thompson: No, as long as yeah, that's a good question. I I thought about that too. As long as we understand when it was completed, right? We know when it began and when it completed, right? That's really what you're after. But ba but the problem that I'm talking about, like the since it doesn't exist, is based on exactly what you just said, is the way that cost is captured now in our industry is very reactive. It's not proactive. So it's so making financial decisions just based on the landscape. shouldn't be as there shouldn't be as much friction there. the the way that it cost capture works now, right, is everything's uploaded to an ERP. You use your job cost tracking tools. A foreman or a superintendent has to sit on site and actually say, Hey, I did this thing today, and it belongs to this cost code. Right? That's a garbage in, garbage out system from the get go. That's set up to be not like just be eighty percent accurate, I would say. Mike Powers: Yeah. Spencer Thompson: so y you know, date now you got data scientists involved, project engineers involved, and I would love to figure out a way and dream big with people like how can we automate that out completely? The data may exist in a lot of different areas, and maybe it doesn't. And what I found is a lot of it doesn't from a lot of people. anyways, it's it's just an interesting Andrew Zukoski: W Spencer Thompson: thought experiment that I've been just playing around with. Andrew Zukoski: What is the business impact of that? Is it we get to save on people going around and doing data entry on what was built? Or is that knowing what got installed today, what we can bill and then what that cost us? Like what else does that enable that hits the top line or the bottom line? Spencer Thompson: Yeah, that's a good question. So actually the the personnel to do that is probably not that big of a cost impact to your business. It's just more of a manpower headache that you have to deal with with a training standpoint and paying someone, right? That that aren't going to actually increase revenue of the company. They're just managing data. Andrew Zukoski: It's overhead. Spencer Thompson: So the the I yeah. So I think the second biggest, the second biggest, like really the biggest thing is true production. Right? Understanding true production, not what somebody's Telling you they did, right? But seeing what they did in time. Right? So we're humans, man. Every company I've ever worked at, you track a lot of dirt. There's a lot of different ways to track dirt. I've seen it done a lot of different ways. Drones, load counts, you know, guy with a ticker sitting there just counting things, right? you know, how many buckets did you scoop out with this excavator? And then you understand the bucket size cubic yard, you try to You know, you're getting the best you can, but that's not really telling you much other than I'm getting as close as I can based on a feeling that I have and what I saw. Right? It doesn't really understand when I'm digging dirt, it's actually loose and it goes into a truck that's loose. Then I go put it down and then I compact it. Right. So you're doing the best you can with the tools at your disposal. You're taking these softwares and trying to like back into your data in a lot of ways. and it's again, I I I think understanding your true production, you then can make better decisions back into estimating, right? Right. So you cycle that information back in to truly understand it. I think you also have a a piece of this that's resource allocation. You may look at a job and say, Hey, you know, we didn't hit productions because you know, we we I'm just gonna make something up. We blew our labor and equipment. costs out of the water for this job. And you just assume it's a production thing. But the reality is, is what if I just had a a few crews out there that I that I just pay more than the other crews? They may have just been putting in the same thing, but but do you have all these systems in place to even fully understand that? Like does a pro a project manager's not gonna know all of that information across seventeen hundred employees? That's not it's not their job. It's not what they're you know. So I I just a lot of different pain points I feel like you could solve in that. Like truly taking like the data that you're paying for to like monitor machines, fuel burn, all of these things, like who's actually out there and then cycling it back through into your estimating software. Luigi La Corte: Hm. And what what's Spencer Thompson: Anyways, that's a Luigi La Corte: the advantage what's the advantage you get at the end of the day once once that happens? Like what's gonna be the ROI that you're forecasting? Spencer Thompson: gosh, I don't know. Honestly, I don't know. I'm still in the early phases of like just trying to like like understand how this even flows, like understanding the pieces that would be involved in it. so I don't know the exact ROI, but I would I would say millions based on just a feeling I have of profit paid, right? So I like when it when it when a when something does go arise or like it you know you Maybe you just don't fully understand the thing is how do you fix the thing? You kind of just are like shooting a shotgun and like hoping it hits a few things to fix it. Right. And then you wait a year for the next job to close out, and then you analyze finances and you go, okay, which of these things did we shoot at this that worked? And then maybe after five years, maybe you hone it in pretty good. Right? Again, it's it's it's a complicated issue. I'm simplifying something that is very complicated, you know, just for conversational purposes. It is fun to talk about. Luigi La Corte: I'm surprised these these things aren't doable now with AI. but maybe we can talk a little bit about Cardinal's approach. So I see a lot of AI committees, people who are designated to explore products, options. and then I see you know like one to three ways to implement it, right? Like they will buy a solution, like let's say one of our one of our products. they'll try and build it in house or they'll bring in a consultant. I'm still kind of scratching my head on the the consultant piece, maybe except in the case of Palantir. but how does Cardinal think about that and and what's your approach? Spencer Thompson: Yeah, that's a good question. I will say from a company wide standpoint, we are very cautious, right? We we want it, but we are like extremely cautious. And not not because it doesn't yield great efficiency or we don't see the value in it. Really, what it comes down to is a security thing, is from an IT standpoint. So we're trying to find the balance of what could be unleashed within guardrails, you know, so things don't get out of control. Right. And that's that's how we assess it right off the bat. That's just and and so I work a lot with our IT team, right, from a security compliance standpoint to make sure that softwares that we're vetting and things that we're doing and maybe things that we roll out, actually can can it be managed from a an IT corporate level across the entire company. Andrew Zukoski: Thank you. Spencer Thompson: I I haven't came across a scenario where we've needed a consultant, so to say, for anything. maybe they have, you know, but I I personally haven't. Luigi La Corte: Can we talk about the implication? Like Mike Powers: So you're gonna h you're gonna hire McKinsey to to to tell you how to make your business better? Luigi La Corte: Maybe they'll know. I don't know. McKinsey pretty smart. I know that you're don't don't necessarily agree. I mean, can we talk about the security security question for a second? So because I I I we we Andrew Zukoski: Yeah. Luigi La Corte: sometimes come up against security. it's very common, right? it's Andrew Zukoski: Yeah. Luigi La Corte: kind of like, you know, those like those bills that pass in the US where they say like, you know, stop stop baby murderers act, you know? It's like of course you agree Spencer Thompson: Yeah. Luigi La Corte: with it on the surface, and then you dig in, you're like, well, why is there something about you know funding, you know, some developing country in here? I don't Very, very bad, bad analogy here, but you kind of see what I mean, right? Like security on the face of it, of course, we need to we need to respect, we need to be make sure that the the business is secure. But I find that companies end up trading off innovation in favor of security. And I sometimes scratch my head, like now having built software, there is no software that's a hundred percent secure, right? And you saw a lot of kind of stated preference backlash around, I want to control my data. but like you look at the two paths that took, right? In the in Europe, where you have GDPR, you don't see a lot of innovation, not to say that's completely causal. Spencer Thompson: Yeah. Luigi La Corte: in America or in the Canada, you know, I get great ads. I love the ads that I get, and I'm happy that they use my data in that way. So I do wonder like, are people overreacting to the security questions? But maybe let me put it in your court, Spencer. What is the specific downside or the the kind of worst case scenario that you're imagining with security? And do you feel like that's a good trade off that you're making? Spencer Thompson: I understand completely what you're saying. For a software like Provision, right, wouldn't be probably that wouldn't be that big of a deal. You would just have to meet SOC compliance, right? Something like that. When I'm talking about AI security, I specifically mean someone getting on their computer and going into Chat GPT, just the website. Or clawed, right? If we unleash claw to the entire company. Like what what happens there, right? So let's let's just I'm gonna separate provision and like software companies from that, right? Like AI modeling and stuff like that. So you when I was at ACT, I listened to a lot of these companies talk about how they implemented AI across their company. And there was a lot of different approaches. There were some companies that took the approach of we're just gonna turn it on and give it to the field and because we can't stop them from using it. Right. And there were some companies that were like, we're not letting anybody use it. And basically what I do know is as a human being and you know, just interactions with just people outside, everyone's using it. they go home, they you know, they get on Gemini, they're on ChatGBT, they're on Groc, they're they're doing it all. So, and they're just logging in with their personal accounts. So, security from that standpoint becomes a major pain point where I know we do not want that to happen, and we are very cautious on just turning AI on for everyone within the company without understanding what it is we're turning on and how we can guardrail it. A perfect example would be Claude, right? Claude code. Claude Does not have a great way from my experience of putting some pretty decent parameters on it to kind of control some situations. And to build connectors within Claude, you have to add, if you're Microsoft users, you give it access to your Microsoft account. And that's a big no-no to us. Like we don't want to give it access to all of our enterprise accounts at a Microsoft level. And have this thing being used throughout the company in whichever way we want it to be used, and then it you know somebody misuse it. Right. So I can tell you right now, that's not happening over here. It can stunt innovation to your point, because in a in a way, because if you don't allow people to actually experience it and use it and play with it, how do you know what the value could be? Right. But until we get to the point with a security standpoint of of like there needs to be guardrails put on this thing. Like we just can't comfortably release this to everyone. We just can't. Like I'm not even comfortable doing that, like where I'm at. but software companies are different. Mike Powers: When you say when you're like when you say like security concerns, like so are we saying like, Hey, I'm in I'm a Microsoft instance, I'm using copilot, I don't want to turn on Claudic Copilot, or are you saying like, Hey, I'm gonna sign an enterprise agreement with Anthropic, I don't wanna connect Microsoft auth in through Microsoft to Anthropic? The second one? Spencer Thompson: That's it. Yeah, that's it. The second one. Yeah. We like the cop the copilot stuff sort. Mike Powers: What what's yeah, what's your big concern wait, I'm curious, like what's your biggest concern around that? Spencer Thompson: So I don't to my knowledge, again, so I'm not an IT guy, right? It's just in my conversations with them. So I I can't speak at a high, high level from an IT standpoint. But I think from their standpoint, from how it's explained to me, is that there there could be things that are done or th or things give it's given access to that they don't want giving it given access to. Like I don't know what the specifically those things are from an IT standpoint, but that's the approach that's being taken. And Mike Powers: Yeah, just like w I just want to like like I know just the I my my soapbox of leaning into L LMs. so like with your enterprise agreement with Claude or Anthropic, like none of that data is like used to train foundation models, right? Like that's your proprietary Spencer Thompson: Yeah. Mike Powers: information. they've done a really, really good job of like structuring proprietary data as an example. Like I have I do like my budgeting through Claude as a use case and I like paying out QuickBooks and RAM. No one else on our team can access any of those data points because the controls go back to the source of the document. So in a Microsoft Spencer Thompson: cool. Mike Powers: instance, if if there was a proprietary data source, like, hey, you had comp plans for everybody and an Excel sheet and only, you know, Spencer had access to that. Andrew, even in your Claudanthropic instance, couldn't like call that that that document 'cause it's only in your security control. I think Spencer Thompson: Yeah. Mike Powers: the industry has done a terrible job of this like Andrew Zukoski: Response. Mike Powers: pushing that message. But yeah, I'd honestly have the same concern. I'd bet I like to have different team members ask our claud instance questions to see if the data starts pulling out. and again, like there are like some security controls in the background to prevent those things from happening and like you can restrict it from doing certain things. we just take in the bet that like the upside outweighs the downside of any type of like security risk. Yeah. Luigi La Corte: Mm. Spencer Thompson: Yeah, no, I think that's great. And I think everybody, like to your point, I you know, everybody's gonna have to make this own decision for their company, and it's it changes so fast. Like, I'll be in Claude and then there'll be three updates in one day. And you're like, dude, like I can't even keep up with this thing. There's so many updates, it's so fast. so I think there's also some concerns there. But I appreciate you sharing that with me. I think that's awesome that you guys are using it. I think it's all I love when people I hear people using it and how they're using it. I think it's cool. Mike Powers: We're look we're giving seats to customers of Claude seats. So we're like we Spencer Thompson: cool. Mike Powers: we the way we have it set up is like, hey, like if you're expensive like, hey, I can't get Claude through my IT group of whatever. All right, well we'll give you access, we'll help you set up different things, we'll help you set up MCP connectors, we'll help you make sure everything is locked in because of the way we're like you're able to push down skills, as an example in their world of like, Spencer Thompson: Yeah. Yeah. Mike Powers: hey, and you know, we the example we use is like we have a design library skill for build vision across the team. So if you create a PowerPoint It only has built vision styling. No matter what you do. You can't get Spencer Thompson: that's cool. Mike Powers: away from that 'cause you can like automatically force that down. for our customers we're doing the same thing of like, Hey, here's how you structure data down, et cetera. But I don't know, it's it's Spencer Thompson: pretty cool. Mike Powers: interesting. Spencer Thompson: That is cool. I like that. I was I've been I I know I've talked to Kyle a lot about C Ps and stuff and I'm like been reading up on but I just get bogged out with so much other stuff. I'm like, I would love to dive down this rabbit hole. I know he talked about Z Z take off with their C P I got to play with it. It was pretty cool. Mike Powers: Yeah. I don't think you're ever going to go into a in you as a like leading edge person will never actually go into a software interface. You'll go into a claude and just do whatever you need to do. And then the agents, Spencer Thompson: Yeah. Mike Powers: the background, will obfuscate it away. And it won't matter if you're using Opus, if you're using Fable, if you're using Astra, like you won't even know. You're just gonna have an interface that's like, hey, yeah, that example, like you could you could prompt claw if you had all your nodes connected, you could prompt claw to be like, How many how much dirt did we remove yesterday on A B C project? And Spencer Thompson: Yeah. Mike Powers: it would just tell you. and I think that's what's fascinating. And then it's this swing, this pendulum swing of it like ten years ago, everyone was trying to get you off of Excel and email into their software. I think we're swinging drastically back where it's like, just live in your email. Live on your phone. Why can't you just get a text Spencer Thompson: So true. Mike Powers: and be like so it's there's it's really cool. I think the Spencer Thompson: That is cool. Mike Powers: companies that bite on it and like are gonna crush it, like If you can go to an owner and be Hey, we can do X, Y, and Z, we'll give you complete visibility into A B C and owners are gonna bite on it. so it's cool. Spencer Thompson: Yeah. Yeah. Yeah. AI's a lot of fun. I mean, I I I get to do a lot of stuff with AI, so the IT team is very forgiving of me, knowing what like what I do and they're a little more trusting Mike Powers: Yeah. Spencer Thompson: of me. So I do get to ha play with a lot of the tools, whereas other people may not in the company. And I've got to spin out some really cool products to help in the workflow just You know, I think that's where I find the most value in it right now is a it's a great indexing tool to like search through mass search documents, compare like previous things. But really, where I found the most value is we were talking about like antiquated, like outdated programs that we still use to this day, I would say outdated. that yield really good results. It just takes a long time to get there. And so it's like, hey, I get this one report over here, I get this other report over here. These programs don't integrate. Our estimating software doesn't integrate with our takeoff software. So it's like, how do we do this data transfer? And so I like to build what I call like, you know, like horizontal bridge programs. And so file conversion programs, drag and drop assemblies, linking things to our code books, you know, I I find so much value in that, just helping the team just be able to drag a takeoff into it. It builds their bid Luigi La Corte: Mm. Spencer Thompson: in like five seconds. And then it's c now Mike Powers: Yeah. Spencer Thompson: converted into a format or estimating software can just import and then it just maps to our code books. Whereas, you Mike Powers: Yeah. Spencer Thompson: know, most people are just clicking through the code book, building something, you know? That takes forever. It takes for that's that's what we have though. And you know, that's that's where I got in a lot of trouble, you know, arguing with HCSS, you know, and it it they're and they're all like that. innate, HCSS, BidWin, you know, I challenge every single one of them to, you know. Integrate with a takeoff software. Like please. Like give me a takeoff button that I can click and it just it's always mapped to it. Like I don't really Mike Powers: It's HC is H C S S even in the cloud yet? I thought they were still on prem. Spencer Thompson: I think they're moving everybody to the cloud currently, but I'm not a HCSS user anymore. You'd have to ask somebody's an HCSS user. it's okay. okay. Mike Powers: Gosh. Sorry, twenty twenty six. Ayayayay. Luigi La Corte: Maybe we'll we'll we'll we'll call it there. Spencer, this has Spencer Thompson: Yeah. Luigi La Corte: been fantastic. Thank you so much for lending your your your opinions and your expertise. It's been a very fun conversation. Spencer Thompson: well I appreciate it. Thank you guys for having me and wish you nothing but the best with your companies. I I love watching you guys and seeing what you're doing. Luigi La Corte: Yeah, thank you. Thank you for talking Andrew Zukoski: Thank you, Spencer. Luigi La Corte: to Nick about his his his triathlons and all that. I w he he appreciates it. Spencer Thompson: Yeah, I really like him. He's such a nice guy. And I hope I see you guys at Autodesk university. Luigi La Corte: Yes, we're hosting a poker game. You need to come. I'll s I'll shoot you an email with the with the details. Spencer Thompson: Okay, thank you. It was nice meeting you, Andrew Luigi La Corte: Okay. Spencer Thompson: and Mike. Pleasure. Mike Powers: Nice to meet as well, Spencer. Andrew Zukoski: Let's Luigi La Corte: All right, we'll Andrew Zukoski: meet you Spencer. Luigi La Corte: wrap there. Thanks, Spencer. Bye. Spencer Thompson: All right, bye-bye.