Latest / Construction Technology Podcast / Construction Technology Podcast - Episode 014 - with Chris from Consigli - Advancing Preconstruction
Transcript
- Luigi La Corte: Hello everyone. Welcome to construction technology podcast. Happy Friday. Um, you're here with Luigi CEO provision. have my cohost Mike from build vision, Andrew from joint, we're joined by a special guest, Chris Murphy from Concealie. Chris, how are you? Very good. Thanks so much for joining. We are doing a advanced pre-construction, uh, wrap up debrief retro. And, uh, Chris was, was, uh, gracious enough to join us and give us his view of how the event went. Mike Powers: Goodness, we're back. Luigi La Corte: Okay, we're good. We're back. So let's go into the second question. So like from my perspective, I saw a couple of tools that we're scope generation, you know, like we're doing with scope agent now. What do you think about this category? And obviously don't pull any punches here. It doesn't, you know, I want to make sure that we're providing. Chris Murphy: Good, how you guys doing? Mike Powers: Isn't it consigli? Andrew Zukoski: Everyone's gonna do great except for the yellow company and their choice of yellow is what Mike Powers: Yeah, a lot of yellow. Luigi La Corte: Most of us were there, everyone except Mike, who couldn't make it. But we thought we would just do a retrospective of how the whole event went. Why don't I pass it off ⁓ to ⁓ Andrew just to tee us up on how the Concealie connection happened and then we can go from there. Mike Powers: What are your thoughts, Chris? What are your thoughts on the scope being Agent World? Chris Murphy: ⁓ I think the reasoning capability is real. ⁓ think the workflow understanding kinda still isn't there yet, but I think, you know, I don't think the scope, the scope gaps don't, they don't happen because people can't read the PDFs. I think it's just real estimators need to think a little bit more relationally. think ⁓ the only bad part about some of the stuff is, ⁓ Andrew Zukoski: Yes, Concealie, my favorite ⁓ Boston-based or outside of Boston-based general contractor. Yeah, so I we know a bunch of folks on the Concealie team and have for half a year. Chris and I connected at the conference. You and Rich gave a presentation on Monday. ⁓ Chris Murphy: is are the estimators gonna retain it? Are they gonna understand it? ⁓ You the output will look correct, is the, is it gonna be retained by the estimator? You know, can the... Andrew Zukoski: that everyone was talking to me about. You go to conference, you ask people about the sessions, and usually the answer is, well, I haven't been to any, or I didn't get a ton out of it. Or sort of like the baseline advancing pre-con. It's like, OK, people have the same problems we have, and there's a lot of validation that comes from that. But ⁓ by comparison, ⁓ multiple people were like, ⁓ we were really excited about the consigli presentation we saw in the workshop in the afternoon the first. Chris Murphy: Yeah, I can read the spec. The estimator remembers like a lawsuit from 2019. You know, it's like, can that connection be made? Or is it just gonna be? That's I think that, and I'm like rambling here, but I think that's the overarching issue with like AI is like, does it take away some of those missing pieces, those connections? And does it just automate stuff where the learning stops or like the, that's the, that's my only kind of, I guess, I don't know. Andrew Zukoski: day, the pre-conference day. And so I made sure to hunt down these new superstars of ⁓ pre-construction and data to hear a little bit about it. So that was how we connected. Luigi La Corte: you Chris Murphy: drawback with it. think it's incredible. The tools are incredible. I've taken a look at your tools. I think it saves a lot of time. I think the scope comparison tools are amazing. ⁓ So as long as we can keep continuing to teach the basics and make sure the estimators understand the work and understand what's going on, I think that's going to be the hardest lift for us internally as a GC. Yeah, actually, Andrew, I had no idea who you were, to be honest with you, because I'd never met you before. we talked briefly, and then you actually put me with one of your tech leads there. And he kind of brought me up to speed, like a 20-minute debrief to run over all the updates to join. Super impressed to go through that system, because I've used it about maybe a year and a half, two years ago. And it was really cool to see how much refinement, how much, obviously, money and time you guys have spent building that system up. It's incredible. Luigi La Corte: It's a common objection. mean, like the way people do it now is they flip these drawings, they print them out, they flip them, they'll highlight various scopes and they know that that goes to misc metals and this one goes to flashing and this one goes to ⁓ fireproofing or fire stopping, et cetera. There's a real concern if you abstract over that, whether anything is really learned by the estimator in as much depth. And I think it's a real concern. ⁓ It's fair. It's like, Andrew Zukoski: Thank you. We can end the podcast right here. Yeah, Mike and Luigi are going to be in an open revolt here in a second. Mike Powers: Is it commercial commercial ⁓ for. Chris Murphy: All right, see you guys later. Luigi La Corte: That's right. ⁓ Chris Murphy: Yes. Andrew Zukoski: Yeah, I'd love to hear about the presentation you gave, maybe a quick summary there. I've run through the materials, but I don't know if Mike and Luigi are familiar with it. Luigi La Corte: It's like the number one objection. like, I want my estimators to know the job. So ⁓ yeah, how do we balance that? Well, then how do you look at the trade off or the maybe at least the, go ahead. No, I didn't get a chance to see it, but I just heard good things from you, Andrew. Andrew Zukoski: Why is there? This is going to be a dumb question. And I think firms fall on both sides of this, but why is the right answer for the estimator to know it? Don't you want the people who are going to manage the project to understand this? Like, isn't that the most important person to understand how the scopes are divided? Chris Murphy: Yeah, did, I think, so Rich had like a two hour window at the conference and kind of brought me in at last minute to say like, can we get, can we get you in there for the last half hour or so to kind of review and kind of showcase what we're, what we're trying to build internally. And we started building this internal kind of smart hub system about maybe a month and a half, two months ago ⁓ when we actually set up our virtual machine environment. And ⁓ we had an enterprise agreement with, with Claude, Anthropics Claude set up. So we used Claude code to build everything out. And it was kind of a good time for us to, and we'd, been talking about this internally for Yes, ⁓ yes, but I think to buy it out properly and to understand the scope and to make sure things aren't missed and like, you're not double, triple, quadruple buying stuff out. Like that is like some of the basics. A lot of the estimators don't work in the field. My career started in the field. was a PE, APM. I built buildings from the ground up. I learned so much about what was going on underground, SOE and foundations systems. That stuff I never would have learned unless I was in the field on a job building it for three years. I think there's just a lot you can take from that. probably for about a year internally trying to trying to get gain support and trying to push the the narrative that We should attempt we should attempt to try to build something internally the worst-case scenario. We spent a few thousand dollars and fail ⁓ You know, not all estimators will go through that, nor have most, but I think how do you help them learn the basics of what it is to be an estimator or what it used to be, I guess. I think that just, it's just whole different way to look at things, a different lens you have to put on and... ⁓ you well, there's plenty of solutions out there, but I think with us, it's, just such a large company and like every other GC in this industry, our data is pretty siloed. You grow so quickly and things work at a smaller level. And then before you know it, you're doing five plus billion in revenue and you're still using the same system architecture that you've had years ago. And we have really good information, really good data. It's just parsing through it, figuring out how to ingest it properly and then how to pull it up. I'm speaking more for the young kids coming out of school who are just like, know, tech forward, insanely fast response. Like everyone expects immediate responses now. So it's like when you pair a young estimator to an older estimator, it's like, it's crazy. But I think there's a lot of shared learning experiences that could be traded off there. And I think it's how do you, how do you have the gray hairs kind of teach and reinforce the basics of some of the job principles. ⁓ in a quick manner that's relevant to everybody that you can utilize in your day to day. So for us, we were like, let's attempt to try to build something internally where we can start at a basic level. Like let's just have a historical cost database. Let's have something where it kind of pulls in all of our recent budgets, all of our recent buy information, because we have a central purchasing group at Consigli, which is kind of unique. A lot of GCs, some do, some don't, but a lot use PXs and the PMs to kind of take care of that purchasing piece. Consigli has a central purchasing department. Andrew Zukoski: One of the things that jumped out to me, ⁓ looking at all the, we don't do a ton with scoping. ⁓ We hear about it from our customers, that's a big pain point, was the range of approaches where there's some very service forward offerings out there, and then there's some technology solution forward ones out there. And some, you go to them, you say, Chris Murphy: It helps us with the prequel stuff. It helps us not oversaturate a sub or have bias or preference. So it really helps with kind of keeping a balanced view on the market and making sure everyone's getting a piece of it and that we form great relationships with subs. But yeah, we started attempting to build a database to reference costs at a basic level, just updated stuff. within a few weeks, we had that up and running successfully. And we used like a small sandbox set of like 10 to 20 projects. Andrew Zukoski: I need scopes, specs and drawings, send them back, look for gaps, give me a first draft to work with. ⁓ And others, like provision, hey, we run this process, we want software to help with it, can we license the software and then we'll drive the software and run our own processes on it. Does the lake, does that change the, do those two camps like change the calculus on how you learn or? Chris Murphy: to start off with at one of our regional offices up north. It worked out great. And what we learned during that month or so was really the data is the most important thing. ⁓ If it's crap information in, it's crap information out. It's just, that's what we've kind of learned. so a lot of it was spent, a lot of the weeks were spent ⁓ parsing through the data, structuring it so it could be ⁓ referencing quickly, ⁓ pulled up, making sure that it was just clean data. So that was the biggest lift. And now that we've kind of figured that piece Andrew Zukoski: Any of this like, share, retain, I'll check. Chris Murphy: I think. I think it does. I think it's just a different way to ⁓ structure your staff and to understand. those handoffs and how, you know, when purchasing comes in and when the estimated cost stops. for us, it's that that's kind of a handoff that happens internally. Some GCs have PXs or PMs kind of work work from the start all the way through, you know, from pre con all the way to the inception of the project moving forward. So, you know, and I've worked for those companies as well. The smaller, the smaller all in one, you know, Dev GCs where it's like, you know, you're expected to kind of stay with the project before it's even a project two years before it is. And then you're fully up to speed. out. Now we're slowly adding more projects and kind of expanding the different capabilities of it beyond just cost modeling. Luigi La Corte: Can I double click on that? When you say the quality of the data, what does that look like specifically? Is that, I go into Excel, everyone classifies this line item differently. It's CSI here, it's uniform at here, it's not unified. Like super, super specifically, what does it mean? Mike Powers: Thanks. Chris Murphy: Yeah, CSI format, like picking a format, sticking to it. ⁓ And that was kind of like Rich's pitch for the first hour and a half of that workshop was really going through and kind of saying like, you have to pick a path and you have to standardize it and you have to stick to it. And the SOPs are really important to drive that home. So we have a lot of internal training. We have a learning development team. We have Consigli University at Consigli. So it's very easy to push down really good SOPs and reinforce that into our younger groups. ⁓ So that things just stay standardized. so thankfully we have that set up. and then you're actually buying out and scoping subs and making the phone calls. So I've been through that. don't know, I would prefer the process that we have because it's just, we're larger, we're dealing with more buyouts, more subs. I think that process works better when you have a central purchasing system. Now that I've been on this side and I've been at Conceivly for about three years now, it works better and it also helps cut down on bias and all the weird things you get when you start buying out trades and you start, if you're purchasing. Andrew Zukoski: I don't think it's biased if I give the work to the person who gives me the best sports tickets. That's objective. They were the best sports tickets, so they should get the job. Chris Murphy: which means the data that we have is pretty structured and within reason of, you we're not gonna spend months trying to reorient stuff. ⁓ So yeah, I think it's just sticking to a standard and really pushing that as the SOP across the board. Yeah, I mean, that's one way to handle things. I mean, but it's also like you guys see like, all right. Luigi La Corte: Good. Mike Powers: you Chris Murphy: What's their bonding capacity? Have we given them $20 million in work and they are only bonded for 10? know, it's like weird things like that. And I think you're going to get a lot less sub participation if the market knows that you only give to these two guys. You know, that's a big piece of it. And they're going to stop interacting with you. They're going to stop pricing stuff. They're not going to answer the phone. Like that's a big piece of it too. You want to have that network. I think that's a really, know, GCs don't, we know what we know, but the sub interaction, the feedback that comes from the market is the most Luigi La Corte: Practically, when you're setting up your historical cost database, does that mean ingesting every estimate, taking all the line items? So every tab, I'm assuming you do tab-based estimates in Excel. So like every subcontractor has a tab. Does that mean then taking those out and saying, hey, we're going to standardize it. We are going to do our, let's say, master format derivative ⁓ code. ⁓ Chris Murphy: Yeah. important thing. That's the most up to date. You can track trends. You can see where things are going. You understand some of the struggles. So it's like that piece is still super relevant and really important. And their relationships I think are definitely at the top there. Luigi La Corte: And is someone going in each one and saying, ⁓ actually, we put this in misc metals, but this really should be in rebar or this should actually be in structural steel. Like, do you have a team pouring over the data to make sure that it's aligned based on how Concealy does scopes? Let's move on to takeoffs. There's a lot of takeoff solutions out there. Your user Toggle ⁓ every day. Well, maybe not every day, but at least once a week I'm seeing a YC company that's coming out and launching a new takeoff tool, maybe for a specific trade. I saw one yesterday for concrete. ⁓ saw one that does full high rises. So they're going sector based. ⁓ What's your view? Is it saturated? Are there more problems to solve? And then what's your view of the landscape? Chris Murphy: Yeah, yeah, we do have a team that has been doing some data migration pieces. So that's been. ⁓ that's been underway for, I would say, six plus months. We are switching a couple systems internally, some on the accounting side of things too, that kind of goes hand in hand. So while we're doing that and switching over, we're parsing through that data and we're trying to basically clean it at the same time. And then we're really taking anything that's beyond, you're basically going from GMP level up for what we're pulling in as far as budgets. And you're talking about very detailed breakout budgets. use Sage to build out our estimates right now. Pretty archaic system. Andrew Zukoski: Hmm. Chris Murphy: ⁓ I don't know, think computer vision has gotten shockingly good. think workflow integration is still gonna be the hard part with this, but I think the takeoff itself is becoming less valuable than the decisions around that takeoff specifically. So I think it's like... But the detail is incredible and the breakouts all that stuff and as long as you're exporting it as a standard csv in a typical format we're able to kind of rip that out keep it standard break the tables out and ⁓ It just it's just start ingesting information. So And then when the next piece we're experimenting with now is like how do you take that to the buyout stage when you get a lump sum number from a Subcontractor for 20 million dollars and there's only two breakouts. How do you compare that to your internal estimates and all the detailed breakouts that you're you're I think it's empowering. think it helps our teams operate more efficiently. think there's time spent. Especially, Mike, you were talking about MEP and taking off. Those drawings specifically are intense and there's so much detail. And that is MEP, FP, if get into that world, it's like, that's 60 % plus of the job. It's the most important, arguably. And usually why jobs don't move is tied a lot to that. So I think being able to allow the estimators to not asking for, you want to know price of square foot or price per linear foot of something, how do you extrapolate that from lump sum numbers? That's kind of the more difficult part. We do have very detailed leveling sheets. Andrew Zukoski: Have you solved that? Sorry, go ahead. How do you do that? Chris Murphy: spend a week doing physical takeoffs ⁓ and then instead dive into the drawings and like I said pull up lessons learned or understand what other jobs were similar and why they worked or why they didn't or working on ⁓ just different pieces aspects of the job itself is incredibly valuable. I think AI for takeoffs is the best way to use it and we've had a lot of success in it with Toggle and we still use OST and we still use a couple others but like as we're sunsetting salmon Yeah, that's that's where Andrew Zukoski: I guess a mystery so many people wrestle with, right? And it always strikes me, like you build up like Sage and other estimating platforms, even like more recently developed ones, like you build out all this detail, but then you buy the job in like much larger lumps. so it's like, don't understand the numbers assigned to a linear foot of whatever are often appear to be a fiction, right? It might be sort of a fiction that has useful correspondence to the world. It's like, well, you don't buy a linear foot of whatever. Mike Powers: you ⁓ Chris Murphy: and focusing on others, could see the time savings. I think it's just, you're able to kind of empower the SMRs with better tools and they can focus on what makes more sense for the project or the team itself. I don't think it feels super crowded. I'm not super educated in the AI takeoff world, but, ⁓ so I don't know, can't comment on that. Yeah, I think that a lot of that is it falls back onto us. It's the scope sheets and the leveling sheets are really important. That's super critical. And we have very detailed leveling sheets. It's all in Excel. ⁓ There's obviously a lot of macros and VBA baked into it. So they're not just simple sheets, but ⁓ there is stuff there. I think it's just having the... ⁓ Mike Powers: you Luigi La Corte: What's your? Andrew Zukoski: I would just. Luigi La Corte: Yeah, that's fair. What's your view on? Chris Murphy: It's being structured enough where the estimators and the purchasers are really trying to break those things out. We have breakouts all over the place for all different types of bid packages. And it's really just trying to extract that information from the subs when you're on those calls or when you're doing those debriefs, descopes. That's the most important piece of it. There's no like magic recipe of like AI, you know, look at this lump sum number and extrapolate these 30 breakouts we're trying to fetch. I'm sorry, go ahead. Andrew Zukoski: Mike, you do this, I was just gonna ask one follow-up. Mike, you do this internally, like AI driven takeoffs, right? And one of the, and like, Chris, you could be like, Andrew, you're an idiot who's not from industry, this is totally wrong, but like one of the, we look at OST and like we've considered integrating with them and I looked at their like release log and there was a 10 year gap. Luigi La Corte: Yeah, go ahead. Andrew Zukoski: between bug fixes, not major releases. There was 10 years there where the solution was essentially finished. But I was like, well, everybody's using it and it works. So like, we don't, there's nothing left to do here. ⁓ And it seems like maybe AI changes that somehow, but it seems like doing takeoffs, you're both producing the takeoff, but then you're also, like if someone has manually done a takeoff, I'm pretty confident that they understand everything that's in the job. Mike Powers: Are you, for those calls with subs, are you like recording them so that you can like extract that data out, like using a granola or anything? It's just funny, cause like a sale, like a software sales team would a hundred percent do that and then feed that into the channel being like, hey, your ROI is seven because your team told me on this call six months ago that if every dollar saved, you get X. Chris Murphy: you No. No, no. That's a unique way to look at it. I mean, that could be. Most people in our industry don't like to be recorded. And I work specifically in private development. That's like the place where you just don't record. It's just not a common thing. You don't have an AI note taker. It's just off the cuff conversations, mostly driving some of this stuff. So that would be helpful to a certain extent. Andrew Zukoski: Right. It's sort of a, it's like a proof of work. Like the takeoff is proof of work of like a skilled person understanding the details of the job. And like, do you worry about that? Does it not come up? It turns out you can understand the job reviewing the like AI takeoff better than you can like flipping pages through the drawing set. Mike Powers: Yeah. Is it living in your email at all? Or are you Claude use the MCP of Microsoft Outlook to go read your emails and responses from subcontractors to get context? Chris Murphy: I think everyone learns a little differently. I think the way that some people learn a job intricately into the details and the weeds is changing. I don't know, you necessarily have to take off every linear foot of pipe in a job to understand that it's a VRF system. These are refrigerant lines. I don't know. ⁓ No, not yet. We use Copilot. That is our front facing security safe tool right now. Our virtual machine environment is using Anthropic system and we're slowly expanding that out to people internally as we build things out. But really we wanted to. Mike Powers: Yeah. Chris Murphy: I think that's in my opinion is a massive waste of time to have someone manually sitting there drawing and taking off stuff as has been done with OST for years. So the fact that now you can and the biggest thing with OST too and not to shit on them I think it's an amazing program in itself. Like it's just the gold standard. But the fact that you can't have multiple people working inside of it is also really frustrating. Like you can't that's the best thing about Toggle is you can have 10 people all in same takeoff doing different parts of the drawing really hammering it out. ⁓ And I think you could do incredible things with these frontier tools. And we wanted to keep it in a safe place where we could break stuff and experiment because we've already broken lots of things. So thankfully we can back it up. because it's a virtual machine environment, can increase ⁓ if we need more power, we need more storage. It's very easy. We looked at doing some hardware implementations if we wanted to host that locally, but we decided to go the VM route. ⁓ Mike Powers: Haha. Yeah. Chris Murphy: I don't know, there's a lot you can do with these tools, ⁓ so I would love for us to start providing more access to that at a corporate level. that synergy is really awesome and I think you're going to see a lot more of that in the future where people are going to asking for like hey I want to have multiple team members in this program we can all reference things we can all do pieces of work we can cut it up so I think that's incredibly ⁓ I don't know efficient and definitely moving in the direction that we would want to see. Mike Powers: Yeah, I mean, we're seeing a lot of like, ⁓ lot of our customers, have like an agent that plugs into their outlook inbox and it just parses all the data for them. So instead of ⁓ going in and seeing a hundred unread emails in the morning, it's literally just a summary of it. And they contextualize that knowledge graph to be like, ⁓ and you can start seeing trends. really fascinating. And it's, like, it is simplistic in the sense that you get like the way you can set it up. ⁓ so you can like protect information, make sure it's secure. ⁓ Chris Murphy: Yeah. Mike Powers: My, uh, my like contrarian position on this is it's like a problem that doesn't need to be solved or shouldn't be have to be solved. Like there's no reason why that owner creates a document set in a model. We can press it to a 2d process. We then send it to three general contractors who then send it to 10 different mechanical subcontractors and all 10 mechanical subcontractors do a takeoff. Those 10 mechanical subcontractors send to the reps. They do a takeoff when in reality, the quantities are already verified, validated by the engineering firm. you guys are safe, like with the Anthropic Enterprise Agreement, you guys are safe to make sure like none of the dev model, that none of that's training a model. But just to try to get, when you were like talking about like, hey, like we have all this, like kind of have this like, it was great data source. And then you're, most of the conversations are happening like via text or you're like walking up and talking to someone or someone on the phone or email. I mean, just like trying to get into that like intermediator is a challenge, I'm sure. the owner is now paying to do 17 take-offs, even though they already paid to do a take-off. So my position on AI take-off is it's a nice plug for what we have now. And I think in five years, maybe even two years, no one will even be talking about take-off anymore because the tooling that we have will be able to connect directly to the models, correctly to the BIM, correctly to the sources that need to be connected to. And ⁓ that's going to be, it's, Andrew Zukoski: layers. Chris Murphy: Yep. I think that's the future though. think less email, less tabs, less programs, less everything, more centralized. ⁓ That is kind of the goal of what our initiative is. Yep. We still exchange a lot of information via email, a lot of updates, lot of random, know, even the meetings that large corporate companies have are just, you know, I'm sure half those meetings could be cut if a dashboard could display that information a little bit more accurately. so, yeah, I'm totally on the same page because too much gets lost in translation and email. Andrew Zukoski: Someone's doing a lot of MEP on large projects. I can tell if the model is coming up as the source of truth for what's in the... Luigi La Corte: you Mike Powers: Yeah, so you guys are obviously scaling drastically. ⁓ Like what, I know we kind of chatted about this via text, but like what is like the goal? Like what's the outcome? Like you guys are obviously, you probably aren't trying to win more projects. I'm guessing you're trying to better projects or estimate better. Like what is like, what's the top line revenue driver for this initiative that you're trying to drive towards? It's, it's a mate like, so I'm telling you right now, um, we like in the last 30 days, we went from it, uh, Gemini, the whatever, a three, one, uh, three, one pro, um, not being able to read a grill diffuser that's overlaid on a duck work to reading a girl diffuser overlaid on a duck work out of the box. So Chris Murphy: I agree. more information, better information, ⁓ faster responses, faster turnaround time. It's kind of all the above. It's it's give better tools to the people who are already working at our company. It's like, instead of having to hire another hundred people to take care of a workflow, maybe there's just a refinement or an AI tool or an AI process that can kind of get layered in that could, again, cut some of those meetings out or provide the information you need on a relevant dashboard that is updated and not static. It's on email attachments. That's pretty cool. Mike Powers: Well, like, well, why would I pay for another software? Like it can. Chris Murphy: I think the biggest issue is just the risk trade off. That's the biggest piece of it. I think a lot of the design teams are like, hey, all risk is with you guys. You own whatever pipe. We're not gonna tell you the totals of anything, right? And then we do the same thing internally. We hold our take off information and we give it to the subs and we say like, you guys own all this because you're taking it off. It's really a whole lot of just risk. It's hot potato. Mike Powers: Right. So the Right. Right. Chris Murphy: that you missed. It's really just trying to keep people in the loop and synced up as live as possible. And then if it's like a, if you're on a phone call with like a subcontractor or even a, you know, an owner or developer, you want to be able to look up like, all right, price of square foot for a metal panel assembly is X. We just purchased it earlier this morning. You know, like you can confidently say that to that person knowing that, Hey, this is going to be a cheaper alternative to that cementitious panel that you're looking at. Or that's that assembly. And here's why, like if you can turn around a Mike Powers: Yeah. So if think about like how if you ever go into a Mac store, they have they have all the computers, the MacBooks at a certain angle. And what that does is makes the user go in and like move it so they have ownership of it. One of the thesis that we're like trying to explore is that's the same thing that happened in takeoff. Because go back to Chris's point, you can't just have this output of like, hey, here's the output. It's this project is going to cost $10. Like the engineer, the estimator, somebody to know. Chris Murphy: Yeah. Mike Powers: what like the actual what's in the project or what's in the data. So what are things that technology can do to make people like feel ownership of that? So we've actually like thought of playing around with moving GRDs like three pixels off so that they would drag and drop it and put the GRD in the right location so they feel like they're actually validating where the data is. But it's like it's that Chris Murphy: ⁓ a set of comps the same day as opposed to taking two or three days or letting it slip a week. It's just cutting down the response rate, the time, and then I think that's gonna lead to a lot more opportunities and a lot better responses from owners and developers. Luigi La Corte: So obviously Concealy is very advanced relative to some other market participants I see. Other GCs aren't doing this. ⁓ I don't think they're putting up the resources to go into their historical cost data and normalize it so they can use it. I mean, it's very advanced. It's good data processes, like very, very, very prudent. ⁓ Then what's your perspective from advancing pre-con? Because you obviously saw a lot of tools that... Andrew Zukoski: It's gonna, guys, it's gonna be like sexual harassment prevention training. Like, everybody's gonna like sit through a presentation and have to answer a quiz about the project before they can proceed to buy it out. Mike Powers: ⁓ jeez, that went real dark, real fast. But it's something good, like again, my first job at Turner, like my first three months, they gave me the binders of the spec books. had to read it all. Chris Murphy: Painful. Mike Powers: And that's why I knew that we had this like double drainage system at the Eagles Stadium in Philadelphia, because that's like this, they had this weird layered system that's still leaked. Luigi La Corte: purport to do some, or at least provide some of the value that it sounds like you're experiencing, my product being one of them was Scope Agent, right? Normalizing scopes, helping you find gaps, et cetera, making that easy. So then what's, how do you look at those tools? Cause there were a lot of scoping tools there. I think there was at least four of them from what I saw. Maybe, maybe just bring this, you don't have to start with scope, but you know, having seen tools and having built some of your own software, give us your perspective about, you know, some of the vendors and how that went for you. Chris Murphy: I agree. I'm hoping that all of this leads to things taking less time. Overall, the whole process. Luigi La Corte: I think it will. It clearly will. I'm on the other side of Mike's argument. I don't think the models are very good at take offs at all. But Mike, let me ask you question. Based on what you have right now, let's say you were running a Duckwork company, you can choose one and only one resource to take off your projects. Do you use Gemini 3.1 or do you use a human? And you have to rely on their output. Which one do you choose? Chris Murphy: Yeah. We- It was actually my first time attending that conference and I thought it was really good, really well organized. Even the app was super clear, it easy to plan out your day. I loved how. Mike Powers: So I think. So you still have a human and the challenge with Duckwork is like the Duckwork is not even accurate on the drawings. So even if you have a perfect takeoff, like the first thing I learned as a mechanical estimator was that an engineering firm never does the, like at a drop ceiling building, they never do turnups and they never include the six inches through the floor. No matter what on any drawing, there's never that. So you always have to, every time you do a floor, he just, he had an Excel sheet and he's like, all right, we're gonna do, there's 13. Chris Murphy: I guess don't go to conferences too often. So I thought that was pretty streamlined. ⁓ Everyone had a UK accent, was a little different, but everyone was super nice. So I didn't expect that in Arizona. ⁓ we actually did a download internally yesterday. So all the Consigli team members that attended from our different regional offices, we had people from California in there and DC and Boston, Maine. And we all kind of came with different ⁓ Mike Powers: risers, there's 13 pipes in each riser, I'm gonna multiply by six inches every time. Like that's never gonna be solved by a So, that because the drawings are complete, but as we pull off of this idea that the drawings aren't actually the source of truth, then those things get solved because you are theoretically running the duck work through the risers already. ⁓ I think it's solving a problem that's not actually a real problem. That's my overarching take on it. Chris Murphy: different strategies, different things we were looking for. all involved in, we're all in pre-con, but we're all estimating. But we all kind of came there looking for different things. ⁓ And I think our honest take as a group was, it was a crowded floor. ⁓ I think we saw like seven companies reinventing the same feature within like 200 feet of each other. So I think most booths were variations of like the same three pitches, which were like the AI takeoff, AI bid leveling, and AI spec review. ⁓ Luigi La Corte: While we're potentially wrapping up here, what are the top lessons learned from yourself and the Constigli team? Chris, apart from you being a great poker player. Chris Murphy: But I the cool thing is like the capabilities are real now in a way that they really weren't two to three years ago. know, think was, was computer vision actually works now, which is incredible. And large language models are amazing and can read specs. And ⁓ I still think the operational gap is pretty big, but you know, there's a massive difference between, you know, Ken demo and Ken deploy. ⁓ So I think it's a lot less, it's still the hype, but it's now you're starting to see, they're starting to showcase some tools that are actually real and actually do some good stuff. Yeah, play more poker, obviously. ⁓ I think most, less than I would say most AI projects are secretly data projects. think... ⁓ Luigi La Corte: you Chris Murphy: Data cleanup is the actual project and that's something that we're living through right now and I think it's a really important piece for people to focus on. When I was at that conference we had people obviously stopping us immediately after the presentation and then throughout the show and it was a lot of like, we're trying to do something similar but we're failing or we're getting really frustrated with this or I had one guy, like a Midwest GC that was there and he was like, yeah, I've got a 26 year old working for me and he's using Claude Enterprise and he's figuring stuff out but it's really frustrating because Andrew Zukoski: Have you guys deployed any of those three categories? So like AI bid leveling, AI takeoffs, AI scope review? Chris Murphy: We use Toggle. So Toggle AI ⁓ is really good. ⁓ I think I tested that out like almost two years ago. We do a lot of this internally. I can say we'll do a lot of pilots ⁓ and we'll break off different groups and we'll kind of touch base again months later to kind of provide thoughts and feedback. And if it's a good pilot, we'll move it forward to another stage. But I think we do that pretty well. We're still trying to figure out how to like, a lot of people will be involved in multiple pilots and we don't even know it. And we're all like, ⁓ did you find this company? And you're like, yeah, I talked to him. ⁓ We just can't you know, we can't connect all the dots and stuff and I think I think ⁓ There's a lot of people that just like on the verge of giving up and it's the worst time to do that in my opinion like every week there is more capabilities being shipped out and more amazing people sharing their their information and their insights and their findings through github or x-ray or any of these social networks where people are just building and ⁓ shipping and proving and I think it's a really it's a really unique way to share information and and Kind of help all of us kind of grow make two months ago. So we're still trying to figure out that sink because we're just so large. ⁓ But I think that was a really successful onboarding of a program that's really ⁓ saved a lot of time. And we have a wide range of estimators. We have very experienced ones. We have very young ones coming out of school. And it seems to kind of hit all the bases of everybody. The older estimators are like, this is an AI until one of the chief estimators sits next to me. like, these tools aren't going to wow me until I'm not having to do door takeoffs anymore or door counts. And Toggle AI does that. It does all your counts. the right decisions ⁓ together. I think that's, it's almost like having a peer group or something because there was some, I'm trying to think of who I have, but Burns and McDonald. did a presentation there. They're a huge design firm, international, but these guys, ⁓ they've done what we're attempting to do now, and they already have a system in place, and they've had it for I think over a year now at this point. it's possible to do these things, make modules that make sense and to ship out agents internally in a marketplace that solve problems internally. It just takes a lot of time and effort, so I just think people need to focused and driven. summarizes everything, it calls out issues. ⁓ So the door tagging situation was like, hey, he put his hand up. He's like, all right, this thing's pretty impressive. I use OST for multiple decades. This thing's awesome. ⁓ So I think it's, and we have SMEs and power users inside of our company that are helping kind of form that SOP, the standards, the different. ⁓ Mike Powers: Brett's awesome. Chris Murphy: templating structures that you can have in Toggle. So that's helped a lot too. You need to have those people, the power users, the people who are always going to be behind it, do the trainings, provide the reinforcement. ⁓ That's really important. That's what Consilva does really, really good. and this isn't a hype cycle. AI is not going away anytime soon. So I think it's like learn the tools, understand how they work and what they're good at and what they're not good at, and then see how other companies can help tie in and kind of help accomplish some of the friction points. Luigi La Corte: Where is your threshold between build versus buy? You're not afraid of building internal processes. You have your own hub. I don't know exactly what the hub does, but I imagine it's the place where your cost, historical cost data lives. And maybe you plug that into various tools. How do you think about that now having built some software? And if someone's going to get started or if a GC is going to get started, is it your point of view now that they should get started at the data layer and first take a shot at normalizing estimates to be in a single format? Chris Murphy: ⁓ yeah, yeah, 100%. And that's the biggest lift. That could take months to years for some companies, depending on how large they are and how much garbage they've incurred or ingested. I don't know. I would definitely focus on that piece first before you start thinking that Gemini is gonna solve all your problems and you just gotta dump all your information in there and query it. Like that's not gonna do what you think it's gonna do. And there might be hallucinations and you're not gonna have the accurate data, which is the entire point of this entire thing, is to have accurate results. ⁓ Yeah, I don't, Yeah, we're building software, should say. And it works. I think the harder part is going to be that we haven't yet encountered is scaling. Scaling is going to be a big piece of it. ⁓ So thankfully, we have some really good close alignment and liaisons in our IT group. So we're experimenting on that now. We're starting to plug information in. We're starting to provide access on limited basis. we're going through that now. But I don't think this is like a pure build versus buy conversation. You don't want stuff being hallucinated and made up. Because it will be confident. It will tell you confidently something is absolutely wrong, but it will convince you that you are absolutely right. view it as like an understanding where like your differentiation actually lives as a construction company. So I think most GCs should absolutely buy the majority of their core software stack, but you know. Luigi La Corte: Yeah, hallucinations haven't really gone away. I mean, they have been minimized, but you still see them and you don't want to in certain cases. Chris Murphy: Mm-hmm. I think ⁓ accounting, PM, scheduling, document management, that type of stuff, those are like mature categories. I think where things get interesting is that layer touching all the institutional knowledge and decision making, which is kind of what we're experimenting with now. I think that that makes one GC really different from another and it usually isn't a software license. ⁓ So I think it's the operational knowledge, that judgment, the relationships, the historical experience stuff that's all embedded inside the company. And for us, it's just really siloed and it's just building that middle layer that can kind of reach in. Luigi La Corte: Great. Anything else that sticks out from the conference? You're going to go again next year? Chris Murphy: I think so. think I'd be more, yeah, mean things are happening fast. So there's another conference in Boston I think at the end of the year that we'll be attending. ⁓ It's a precon, it's a pre-con related one. It's a... Luigi La Corte: Which one is that? Chris Murphy: camera roll it is I'll have to share it with you guys. Andrew Zukoski: You're have to send us a referral code. None of us know about it. Go for free! Get three companies signed up. Chris Murphy: Yeah, it's... without disrupting our day to day and being able to service that. That's what we're really after. Mike Powers: Get a comp, yeah. Get a comp. Chris Murphy: It's actually a newer, I think this might be the first annual, I'm not quite sure. I'll send it to you guys, I'll circulate it so you have it, but I think it's, there's a couple of them kinda geared toward this. So I don't know, we'll have see. We have a lot of people from our company that went to this one last year. Mike Powers: you Andrew Zukoski: That's interesting. So it sounds like the advent of these ⁓ AI tools was a big enabler for building the hub. You guys are starting to look at sort of scaling it. Is there any impact there that you've seen? is it, I mean, basically, and I'm curious what Mike and Luigi see, like we work with number of general contractors. Periodically people build solutions for various reasons and they all. Chris Murphy: ⁓ It was a lot of hype and talking about what we're going to build and now we're starting to see some bits and pieces of it this year. So think it's an exciting time to kind of be in the industry and for you guys, exciting to see what you guys produce and how you further refine what you guys are producing. I think it's going to be a really cool future in construction. Andrew Zukoski: When they succeed, it's great. And when they fail, it's always in scaling. It's never in actually building the technology. It's always in implementation and rollout and ongoing support and buy-in from. Luigi La Corte: think so too. ⁓ Mike, Andrew, anything else you want to touch on before we wrap? Mike Powers: ⁓ great place to, great, great, way to wrap. Luigi La Corte: There's a big difference between production ready and demoable to your point, Andrew. That's it. Okay, I'll end the stream now. Thanks everyone for joining. Andrew Zukoski: Yeah, yeah. And so it seems like the needle on building has been moved a lot. Do you have any sense yet of whether or the needle has moved on adoption with some of new capabilities that are out there in the tech world? Chris Murphy: Is that to me Andrew? ⁓ Andrew Zukoski: Yeah. Chris Murphy: Yes and no. I don't think we want to reinvent Procore. think I'm going to answer your question. we don't want to reinvent Procore. think everyone has to use that to a certain extent in our industry. That's kind of the generic bare bones basic. ⁓ Provides that cross-operational. kind of touch point for the field and for the office. But I think we want to enhance how our people operate inside and across those systems. That's kind what we're really focused on. And I think the smartest firms will probably end up in a hybrid scenario, know, buying mature systems where it makes sense and then building strategically around the areas that actually differentiate them. ⁓ So I think And I don't think a chat bot or an LLM is going to be the save all be all for some of these groups. ⁓ But for us specifically, we have great costing information. It's just really tough to find it, ⁓ to earth it and to do it in a fast enough manner where it is still relevant in a conversation. then pairing that with like lessons learned, we have a great lessons learned database. have great SOPs on the job closes. We do like a full day debrief with each team of like, what were the pain points? the friction, what were the wins, what worked, what didn't? We log all that information, but like surfacing that and like bringing it up and tying it to a new pursuit or a new opportunity or a similar job right down the street, that's the tough part. And I don't think there's anything like a software stack. We can just go buy and like that just solves the problem. I think a lot of it is like us figuring out internally how best to do it. And the coolest part is we're not like selling this or marketing this to anybody. So we can kind of take our time and really figure out what works, what makes sense and what we should focus on. I don't know if that's your question, sorry. Andrew Zukoski: It's like the tool doesn't have to serve, it has to serve cansigli, but it doesn't have to serve beyond cansigli because the cost of building it has dropped so much. That's interesting. Michael, are you seeing more build versus buy? Like we always see like, I don't know, 10%. Chris Murphy: Yes. Luigi La Corte: Yeah, totally. Yeah, I'm seeing, I'm seeing small contractors take stabs at building and they're building. They are. ⁓ It's, ⁓ it's takes nothing to spin up, ⁓ an account on Claude and build some software where they run into issues is the maintenance and there ends up becoming a threshold that they cross where they spend a little bit more time than they anticipated on maintenance. And now it's an issue. So I can, I can build something trivially. Like, you know, we had a contractor client of ours who wanted to build a QR code scanner. for job site registration and they built it. But then as soon as they had some bugs, they're like, okay, I'm taking a step back because I'm not a software maintainer or I'm not a software builder. So let me just go hire someone. I don't have to worry about this. ⁓ When I see bigger enterprises, I haven't seen someone build software really, really well to the degree where they're deploying it across the whole organization because like it is hard. It is hard to scale it to make something production ready. to handle all the edge cases to make, know, to put two factor authentication to feel safe that it's not being used incorrectly. It is, it's hard, right? This is what software companies do. But I think the bar has become higher for software companies. ⁓ You can't just simply provide like a chat agent and expect to get mega scale. You have to provide real differentiated value. And I think that's good. Mike Powers: I think I have a, I think I know that the region is my position. think there's like only really one boat left and it's like the multi-party portion of it. Like if Chris and can see, can spin up a internal hub that their team uses. doesn't need to go connect to the subcontractors hub or the owners hub. Um, I think a lot of those are going to be spun up and it's going to get better and better. think that's kind of one challenge, um, for software companies. And I think one challenge for. contractors is like the token spend and how do you actually like capture that? sent, yeah, Chris, sent Andrew and Luigi are like, my forecasted Gemini spend through the end of the year. And we'll be close to like a hundred thousand dollars a month just in Gemini spend. And we're having this like internal struggle. We had a software engineer, excuse me, spin up a full takeoff tool. Chris Murphy: ⁓ my god. Mike Powers: in a month. His cursor belt was eight. Andrew Zukoski: Was that the month that you excluded from the trend because it was so high? Mike Powers: Yeah, yeah, he spent $1,800 in cursor tokens, spun up a full takeoff tool, fully active. We only specifically focus on mechanical electrical equipment, so it's not like tick counting doors. And I'm sitting there, like, all right, well, that was $2,000 spent. I want him to spend $2,000 every month, but I wanna make sure he's pointed in the right direction. because there's going to be this challenge with like not so good engineers that just burn tokens to burn tokens. And we saw that Amazon just released a report that their internal team had token metrics that had hit every quarter and they were creating tools to just like burn through tokens to burn through tokens. So I mean, say that I'm curious, like, you know, as consistently looks at this, is it like, is there a number like, Hey, we're to go spend $100,000 on Anthropic tokens this year, or is it a million? You don't have to share numbers, or is it like, we're expecting to spend 10,000. And when does that? That's that when does that inflection happen? Because like me spending $100,000 a month on tokens where I can theoretically distribute this across multiple people will help your costs go down. Just curious how think about that. I know it's kind of a lopsided question. Chris Murphy: It's funny cause like we've been, we've been obviously trying to build stuff and we've had lots of what we're calling the modules, but like I just said. It's just a hub and and spoke kind of set up smart hub hosting it a custom type dashboard where you can have different metrics, pulling different things, have little cards basically with what module you want to be utilizing. ⁓ So some of the modules that we put together are active or under development right now. And man, you know, it is it like I've been working on one for almost four days now and I've, I've messed up a lot of things. Like, thankfully we're able, cause I'm not like a, don't have coding background. Like I'm teaching myself, I'm learning as Mike Powers: Yeah. Luigi La Corte: You Mike Powers: Hahaha Chris Murphy: I go it's been a I feel like I'm Mike Powers: Do you have a GitHub? Do you have a GitHub at least? All right, cool. Chris Murphy: Yeah, yeah, yeah, yeah. I mean, I'm not deploying anything on there, but I'm constantly on X and GitHub. I mean, there's just amazing things that people, that's the coolest part about this is people are just building and deploying and sharing. So there's a lot of lessons learned. It's all active. It's like, it's like a hyperscale. It's just, it's happening at such an exponential curve with regards to how this is being innovated. So I'm learning and then obviously, you know, like I've, I've interacted with the anthropic engineers on X, like they're on there all the time, Boris, Boris and friends. So it's like, these guys are like, it's just so cool. that this is like a live, this is happening now. ⁓ But yeah, like last night I was, I'm building one of the modules here and it's really just a, it's a process to scrape through documents. So like we've got project, you we have project folders, it's a project number tied with a title. Inside you have documents, have specs, you have code memos, have appendices and this process, it was a module for visual purposes to kind of demonstrate that internally, but it's a process that would happen behind the scenes of Luigi La Corte: Hmm. Chris Murphy: Basically it'll go through in a very specific order and it will scrape all the relevant information from that product from that project to kind of give us like a nice kickoff document internally where it's like, Hey, these are the top 70 fields. Everyone kind of comes into this project now and they understand like, all right, these are the, the, you know, the architect of record. Here's the, ⁓ the mechanical system. Here's how many floors. Here's what the constructability type is. Like very basic things that take forever to pull out that we never seem to kind of standardize and put into any of our systems where people will just fill out the bare minimum. and you never really have any good metrics to fall back on. that just building that out though, last night I was kind of, I was actually sending a message to my CIO saying, hey, you know, I'm currently ⁓ in development with this. You're gonna see some API costs rocketing up here over the next few days. So that's how we internally like, and we have limits set that are much less than what you previously stated, but it's only three of us in this team and we're basically, you know. Mike Powers: Hahaha. Luigi La Corte: Thank Mike Powers: Haha Andrew Zukoski: Ha ha! Chris Murphy: We understand when we're like, when we're deploying, when we're building and there's some heavy, coding going on and we give them a heads up like, Hey, you're going to see like a hundred something dollars hit over the next hour or two. Don't freak out. And like, that's the coolest part about this one process though, because when I started it, like a week or two ago, it was doing a full scan and it was doing really well, getting like 90 % plus of all those fields back from scraping documents. But it was costing like, Luigi La Corte: Don't freak out. Mike Powers: You Luigi La Corte: What documents are we talking about here, by the way? Are we talking about PDFs or we talking about Excel sheets? Chris Murphy: Yeah, PDFs. Yep. And it's going through it's kind of I have a very specific instruction set of like start with the basic stuff. Start with like hit the specs, hit the code memos, hit the appendices and then go into the drawings. But when you go into the drawings, only hit certain subsections of the drawings where you'll find your equipment schedules or where you'll find, you know, so it's really just kind of pointing it in the direction of like what to do next. So it doesn't just start scanning a thousand page, you know, ⁓ drawing set, right? Like that's the parsing cost of scanning all that stuff will be It'll take forever and cost a fortune. And that's kind of what I started with. It was like $100 plus to scan one project. And now I'm down to like 10 bucks, which is really good. So I'm actually asking, Claude in the background, like how much does that process cost? Can we improve it by doing this? And because we have hundreds of projects and we're going to want to grab all that information as new projects are formed. But if we can keep that cost down and for $10, we can scan a project folder and kind of give like a kickoff document to our team. And they're already hitting the ground running. Mike Powers: $15 a project. Chris Murphy: Plus it's great metrics to tie everything back to for comps and stuff. Like that's worth it in my opinion. Luigi La Corte: I want to go to the image that you shared with us last night via text. We were talking about the agenda for today. One of them was advancing pre-con. Chris generated this. This document's a lot to take in. Walk us through what's in here, Chris, and what your perspective is. What's the advancing pre-con expo going to look like in 10 years? Chris Murphy: Well, hopefully not like this. I don't think it'll look like this. ⁓ Yeah, no, that was a funny, honestly, I didn't know this was gonna be part of it, but yes, this is in a nuclear winter in the future, we might have outdoor markets and outdoor expos to kind of keep things simpler, but ⁓ I mean, it's crazy. I think one of you guys commented in the chat, like, you Andrew Zukoski: other problems if it does. Chris Murphy: A couple months ago, all these pieces of signage would have had like scrambled text and it would have been, you know, hallucinations of the AI. Now between, know, GPT, this is GPT produced this, but even NanoBanana2 and these are incredible image generation ⁓ tools. It's just, it's insane. So yeah, this is more or less like, ⁓ a very exaggerated version of what was going on at the the expo. It was just a lot of it's it's hype. AI is the new hype right now and it's tough to kind of get away from it and everyone is pushing some.