Latest / RoofersCoffeeShop / AI Forges Ahead in Metal Construction
Transcript
- Karen Edwards: Hello everyone, good morning, and welcome to Coffee Conversations from Roofers Coffee Shop. My name is Karen Edwards. Today's conversation is sponsored by the Metal Construction Association. Thank you very much. We are going to be talking about AI forging ahead in metal construction. A little bit of housekeeping before we get started. this is being recorded and it will be available within 24 to 48 hours on Roofers Coffee Shop. So feel free to share with colleagues or those who you think might benefit from it who were not able to attend today. okay, the chat's open. We really want this to be a conversation and interactive. So feel free to drop your name in the chat. Let us know. Who you where you're tuning in from. and if you have questions throughout the conversation, go ahead and put them right in there and we'll try to address them as we go. All right. So let's get on with our guest introductions. Ken, please introduce yourself. Good morning, everybody. My name's Ken McLaughlin. I'm the national sales manager for Carlisle Architectural Metals, the Association's group. been in the industry for about 35 plus or minus years. I've been both sides of the fence. really excited and nervous about having this discussion today. So hopefully we get all kinds of fun topics out of it. Yeah, definitely, for sure. Okay. Ryan King, please introduce yourself. Everyone, Ryan King, from Sedan Machinery. been in the roofing industry since about twenty eleven, but more specific to metal and machines. since twenty nineteen. And yeah, whether it's business or or for fun, you know, AI has been a fun topic. So appreciate you having me on. Yeah, yeah. Thanks for being here. It's it's something that we talk about so much and it just changes so much. That's why we have to keep talking about it. So, Rob, please introduce yourself to everyone. Drop super. I am formerly from Eagleview, recently retired, but keeping one foot in the industry that I love so much, doing some independent consulting. But I've been in the industry since the late seventies, actually, as a teenager shingling, framing houses and doing all that kind of stuff. And I've worked for installers, sales organizations, shingle manufacturers, and spent The majority of the last 10 years of my career in technology. Yeah, yeah. it's so so many of folks that have spent so long in the industry retire, but never really retire, right? You always kind of come back. Just you're in, you're in. You can never escape. That's for sure. Okay. so just to get started, I I wanna yeah, say, you know, let's talk about where AI is already being used today that people may not even realize, right? I I you know you talk to to folks and they're like, well, I I'm not using any AI at all. But they probably are because it's being, you know, included in some of the tools they use every day. So Ryan, do you want to get us started with that? sure. to your point, y even Google search anymore is not the Google search from even a year ago. the Google search now, you you'll see, like you might see that first result or two, but then it's you'll see that AI summary. And that's in their case, their service is called Gemini. so I think even people who have resisted or you know, I'm not into it really. You're s everybody's almost by default using it to some extent. It's just really how deep you go down the rabbit hole from there. and you know, we'll obviously get into it more, but sometimes just be curious and ask ask it to do something that you would have otherwise done. It's not gonna be perfect at first, but yeah, we're i it's it's kinda everywhere now. You can't really fully avoid it. but it it's ever evolving, like you said. Yeah. So just for those who are on the webinar, if you're using AI, let us know what you're using and maybe how you're using it. Drop it in the chat. I mean, we've got the big three, right? That everybody thinks of Claude, chat, GPT, and co pilot. so you know, if you're using one of those, let us know. We want to hear from you. so Ken. Yes, please everybody. We were talking the other day and and we were like, which directions do we go? And we it so please everybody if you have questions in the chat, yeah. Because we we could go on tangents for hours. We did. We said like this conversation could go so off the rails. We want to really keep it focused on what's happening in the construction industry. So Ken, from your perspective as as a manufacturer I know you said you you work with the the contractors and you're out in the field a lot. So you how are you seeing, you know, AI play a role for contractors and even inside your own operation there at Carlisle? Yeah. So I I I mean we've saw the change and I think the other day we were having some discussions amongst our group and we were laughing about We used to have things called visualizers. that was this fancy software that we had to take a picture and implement something. And now we use AI to say apply this roof to that product and change the seam spacing and the color and everything else and within seconds it's there. So we're seeing the contractor base use it at that point. even to like the and I'll put myself in the old guy group of the AI thing that I'm the novice in the group here, but we have lots of guys that I I talk with them about it can make you sound a lot better in an email. It can make you sound a lot better in a letter just by saying, Hey, can you take this and can you modify this to make it sound softler and gentler or more professional and more firm or more whatever you want it to do? And that's where I see a lot of these guys starting to experiment. the manufacturing side with Carlisle, it's something I really can't speak to openly. So I just want to put that out ahead of time. So we're talking about that. That's a an entire entirely different group that are way smarter than I am. So but like how it's getting used within our industry, that's what I'm really starting to see a lot, is where people are using it for presentations, people are using it for, I mean, everyday emails to to get their point across in a very directed form. Yeah. So, Rob, I know that you during your time at Eagleview, Eagleview is a leader in property data, in the measurement space, in that how how are they? So, you know, to a contractor who is ordering an Eagleview report, maybe they're gonna install a metal roof and they want that report, there's a little bit of AI behind everything Eagle View's doing now. How much can you speak to that? and, you know, give us a some insight into how that helps? Yeah, and it's it's a great question. And it's one of the things that Eagleview has that really differentiates themselves from all the other technology platforms actually out there. AI requires data to process. So it can't just go into a vacuum and make something up. It actually requires some data set. If you're searching something online and you're you're trying to do some research on a vacation or whatever it is, it's going to go out and see all that open source data out there. And it's going to understand through the the LLM, the large learn language models, what you're trying to get out of it. And then it'll go and find those things. But A and AI is getting smarter and smarter. But really the the corollary to EagleView with that is that they've been flying airplanes and capturing imagery on structures across the United States where, you know, going getting close to 30 years now, 25 25 some years, and they've got image capture after image capture of the same structure all over the country. with that set of data, with that information and those images, they can create machine learning models and really get deep insights into structures. For instance, they could they know when because the materials changed or the roof color changed, they fly post-storms and all of the hurricanes that ringle through. They know when structures change and they can they can understand or the models that they're building through IAI can really understand how a structure has changed and matured over time. and it helps a variety of of partners in industries, you know whether it's the insurance industry, whether it's the contractors, whether it's manufacturers in the roofing industry, all really understand what the opportunities are as well. And I and I would say one of the key one of the key value positions there with Eagleview is they've also from the roof report. traditionally you'd you'd think of Evil View as a roof report business if you're in the roofing industry. but they've measured millions and millions and millions of homes and have the absolute ground truth on what that structure is from the outside, whether it's roofing, siding, windows, doors, all those things. So when they run the ML models on top of their data, they already have a training set built So the AI that they're using to get roof area estimates or to get some other kinds of attributes out of there i is highly accurate. it's it's more accurate than anything else because of the quality and and depth of imagery that they have. So it's really it's really where Eagleview's turning the corner in the industry and leveraging AI across their entire platform. Yeah. Yeah. so just a note from Fred. Thank you, Fred. Fred wants you to project your voice and speak up a little more. Rob, I guess you're a little quiet. but I I want to, we have a question that I want to answer. but before we do, I want Ryan, I would like you to speak because you made a really good point the other day about how AI needs context. And you kind of started to to go there, Rob, a little bit with the data, right? It's got all the data and and that's where it's getting it its info. But Ryan, can you talk about that context and how to be successful? Yeah, so one thing here to keep myself wrangled in and on track here is I even came up with five, starting to midpoint points and that's one of That's one of the first ones I have, and that is that context. And the simplest way to really think about it is it doesn't for this for this example, it doesn't matter if it's Gemini, Claude, Chat GBT. deep seek all these different models you hear right it doesn't really matter where I'm going with this but if you just start talking to it think of it as a new hire right it it it's a it's a very smart intern. The high SSAT score is brilliantly smart, but it doesn't know you at all. Okay. So you you almost want to when you start out really like start by giving it some information about you, your company, what you want it to know, right? And then as we go further down that into context, and this can mean several different things, but think of it as okay, I'm going to connect my cloud, my chat GPT, whatever, to a email folder or or we can even get more technical literal tie-ins, whether you use Gmail or Microsoft or other, where it can actually have what's called an API or better with AI called an MCP or model context protocol. All that means is just a a way of the AIs and the models to talk to each other, that's very efficient. That that's all that really is. so point being is it's it's not hard to figure out, right? If I ask it to do something where it just doesn't know anything about me, it's going to give you a very generic answer. If it has all this information that it can parse through, look up, it can be as advanced as as you know spreadsheets and PDFs and CAD drawings and and the sky's the limit, right? Or it could just be, hey, read through a few emails so you understand how I talk. Right. it i it it's really unlimited to where it goes. But without that context, just know that you're just gonna get a generic answer. even when it comes to if you wanted it to write an email to somebody, a customer, a colleague, whatever, right? If it doesn't know you and know how you write, it's gonna sound now granted, the models have gotten better, so it's a lot different than even six months, months ago. But if you just tell it to do something like that, it's gonna guess. it might assume. even a lot of these models, they'll they'll, you know, might find something it likes to it found and now it's gonna harp on it. So as you work with it too, you might even give it rules. like a big one that drove me nuts for a while when I was using Claude extensively was it loves hyphens. Every email, everything it wrote had hyphens, hyphens, hyphens, hyphens. And it it's not just me. A lot of people said that and drove me nuts. And so this is a little bit different from a context, more of a kind of a rule so you can set up for the AI. But I said no hyphens, unless it's a truly hyphenated word, stop using it. And you know, now it knows it's in its ground rules, that's going to be way different than when I was using it months ago. So the key to that, and I'll just kind of tie into the second point, but that is again going back to those MCPs and APIs, right? If it ties to your your CRM, your accounting software, your file or OneDrive, whatever it is, right? You now just gave this super genius information to make it smarter and then make you smarter. Now, i some people are a little uneasy about it, right? There's some people I that have all their financials, everything on it. Other people whoa, I I don't feel comfortable with that, right? And I'll first sidebar by saying, talk to IT, talk to your company, never, never go do stuff like that. That's that you could get in trouble, right? But overall, those connections and that context. is a really a big part. And with all that too, I'd say is start small. Don't don't start using AI and say, hey, build me a custom app for my roofing company. No, no, no, no. That's too advanced too quick, right? Start small. Have it write an email for you. Have it summarize a document you got. analyze the spreadsheet for me. Start there and you'll you'll grow. and if you have a question, ask it. Like that's always a funny one to me when on on forums and stuff. I'll see if people Hey, ha has anybody ever done this or how can I do this? And every response is almost always ask the AI. Ask it. It might not be perfect, but it's a great starting point. Yeah. And just because it couldn't do something six months ago doesn't mean it can't do it now. Like you said, it's it's changing so much. I totally can relate to the hyphen thing because we use Copilot a lot and it likes colons. So instead of hyphens, it would use colons. We're like, no colons. All right, so we have a question, John, at Sentin from Oregon Metal Roofing. Thank you for being on, John. how do I use how do you use AI during the metal roof install? They have a mobile on a trailer panel machine. And then also thoughts on discussing it during the metal quote process. I'm not sure if that means when putting your quote together using AI or just What what role can that play? So Ryan, I I I know you've been doing a lot of talking and I'm I want to hear from Ken and Rob as well. But because, you know, during the Metal Roof installant, they have a a panel machine on a trailer at the job site. Do you have any thoughts around that? So the my first thought is everybody's freaking out about AI taking jobs and everything else. Hey, we're in an industry that that's not gonna happen, at least not anytime soon, right? There's no robots going up. climb on the roof to to do installs, right? So during the job, you know, outside of, you know, maybe something where the crew can take pictures or upload what's going on, the progress, you know, that sort of thing, you know, I don't think there's a huge benefit with AI right there on the job site, besides like those little kind of, you know, update type things. Now, when it comes to bidding and estimating and taking priceless and and labor costs and even, factoring even the distance of the job and travel. Now we're talking, right? Now that's that's where AI can really help you sitting at a desk or looking at your phone. Where are my crews? you know, where how what's their progress? Let me see pictures. Like that's something AI can help with. But frankly, on the job, like I think that's where we're still roofers. And and and that's where I think we can feel a little more comfortable using AI to make us efficient on the back end, in that estimating side. But that we're still humans and and we're still putting the roof on. So I you know, not the best answer I could probably give, but I think that's just the reality of it. And AI's not taking that job from us. I think there's a place for it. and yes, John, there are roofing robots. It's renovate robotics, who we've had on podcasts and and before I know they are focusing on commercial right now. they tried residential and and they found out it was hard, right? And that you know, they they developed some solutions. but they didn't really they weren't really able to scale. but you also mentioned during the metal quote process. So Ryan, I think you captured it when, you know, you're working on the quote and saying, Hey, where are my crews? What is the distance? Like AI can help you with that. But Rob, I'm gonna point to you just being, you know, your Eagle View background. And how AI is helping be more accurate with metal quoting. Can you speak to that? I would say it it kind of ties into you know our the Evilview partnership with Sherman Williams and the Roofing Passport. so obviously Evilview is using the advanced AI to derive measurements and because of that the measurements have become highly accurate. And when you take that data and move it into a platform like the Roofing Passport, you can get takeoffs, material lists, cut lists, all those things, in a you know, down to a fraction of an inch, in in many instances. You know, what what we see is getting better from the air, from the imagery, consequently, the output from the estimating tools is getting better and better and better all the time. And so there's a there's a real good tie in to that on the bidding and estimating. as as the the software again continues to mature and as the the depth of data behind it validates those outputs, the outputs continue to get more and more refined. and we see that across not you know not just the metal side but across all all parts of the industry. Yeah. Yeah. So I you know, I worked at Eagle View years ago and back in like 2010, when the the the technology was very new, we would say, don't use it on metal. I mean, you can use it on metal, but always go verify. But it's gotten like like anything, the technology has gotten so good and so precise that you can. And that's what's key for, you know, accuracy. I also want to mention, I'm not sure if it comes if you guys can all see this in the chat because they're labeled questions. So I Steven, thank you, gave an example of AI on a job site. what hap like you could ask it, what happens if I remove this X bracing in this bay on a metal building? And that AI could tell you. But I always, you know, have that reservation of, okay, but is it, you know, it's good, but is it that good? Right. And how how trust but verify? Is that the the phrase that we use? Right. AI could say, it's gonna be just fine. And then, you know, disaster happens. Yeah, I mean to speak to that, I mean there's two ways that maybe that could work. And number one is, you know, somebody who's developed with or without AI, something that is really understanding of how a metal, you know, that structure works. where the load transfers, all that sort of stuff. or building one your own that you, you know, teach and train and get that information in there. If you just take your chat GBT or Claude or whatever, take a picture and ask it, absolutely not under any circumstances. It probably would come back to say, No, I don't know. You have to talk to an engineer. Like you might have noticed if you ask legal type questions to an AI chatbot, even though there are now legal chatbots, but that's a whole different thing. But Point being is it's probably going to even deflect in the chat, right? I think even anthropic and and open AI and all them have even trained their models to, especially when they get into that safety or legal world, to know wow, you know, I'm not a counselor, I'm not a lawyer, I'm not an engineer, whatever. so in that specific example, I think yes, with a major asterisk that you would really have to either be a highly trained model on what you're doing. or working with some sort of AI software that's specific to that. But you just taking your app and doing that? I I would say absolutely not. Yeah. And like Steven commented, he's been training it for a few months. So that's absolutely what you have to do is train it and give it that data, give it that context that Ryan was talking about. Sean, thank you for your comment. AI does sometimes deliver correct information But it can struggle with things like part numbers and measurements. so you know, yeah. How do you balance accuracy and verification when using this for business? I I think it I don't know. Any thoughts, Ken? You look like you've got something else. Well, no, I I'm sitting here thinking, I think all the information that we put out is really good. But I think going back to the original question, how can it be used on a job site? So I was part of the original group of contractors and suppliers that have got involved with Sherwin Williams. And we all sat in a dark room and signed a piece of paper, said we weren't gonna say anything for years. and we sat and tried to talk through all the parts and pieces. And and originally, yeah, the dimensions weren't something that you would say, hey, I'm gonna go and order a panel to this length. today they're a lot closer than that. So I mean, that is an actual usage, you know, for a job site where you can do it. I also think of other things that, you know, coming from the contracting side. When the homeowner, the architect, the consultant walks up and says, Hey, listen, I don't really like the way this looks. What are the other options? In the past we used to do it on a piece of drywall or a two by four or the back of roofing felt or whatever it was. And now you can literally take an image and say, you know, change this to this dimension so we can look at this in this image. And I I think that's a for a contractor, that's a usage that they could use on site. And and the Estimates today I would agree a hundred percent that it is it's trust but verify still. But if I went out and if it was six thousand square feet, I believe the imagery in that is close enough that you're gonna be within that range. But it allows you to be a lot more productive when you can show up and say the panel length on this standing scene roof is twenty three feet four inches and all you're doing is verifying rather than trying to measure. so I I think there's huge advancements there. And that's usually a skilled person and that's the hardest thing. we all talk about, where's the benefit of AI? skilled labor is the hard thing for us to replace or to find. So if we can truly find uses for AI with that, there's a value to the contractor. Yeah, definitely. Steven put his lengthy chat GPT answer out there. which I love that it's source it's cited MBMA, right? The Metal Building Manufacturers Association. so you know, that that that's a very reputable source. So, you know, when in those cases, yeah. I mean, it was pretty pretty cautious. I want to talk about Sean's question about balancing accuracy and verification. I think we have to figure out our ourselves what to use it for. Simple thing example, we use Copilot at the coffee shops, right? Because we're a Microsoft driven, you know, everything, SharePoint, Outlook and whatnot. And it integrates really well. And I've I forget to ask it to do things. And we were at a MetalCon was last week. I saw all three of you gentlemen there at Metalcon and probably some of the folks on this webinar. and you know, you collect business cards and and you make contacts and we in the past, you know, we always like take pictures of the cards and send them to someone to enter that into my thing. So I need to follow up. And one of my colleagues just laid all out in a row and took a picture and put it in co pilot and said, make a spreadsheet. And Dang, if it didn't do that and had all the information accurate and we were able to send it right over and put it in our database so we remember to say, nice to meet you at the show. let's follow up. I think we we have to decide what if if it's taking me a long time to verify something, then maybe that AI tool isn't ready to be doing that yet. Or maybe I need to train it more or maybe I just need to not use it for that and try again in a few months. That's a perfect example, I think, Karen, too, of a repeatable task, right? It's very think of again, go back to that intern example. Hey, intern, here's ten business cards. I need you to put these in a spreadsheet or put them in the CRM, right? That's a very easy and and many people on here probably are familiar with skills. I think some of the different AIs call it different things, but you I'm just that's the one I know most. Skill, right? You go through it one time with the AI, now save this. save this as a skill. So the next time I say take all these business cards and and put them here or do that, whatever, I don't have to tell it what to do. I just say, put these in the spreadsheet. It knows. It knows where you can tell it where to save it. You can tell it where to go. You can even have it go right into a CRM or something going back to those MCPs and stuff. but that's a great starting point, right? Is is that was one of my others on my list. Skills. Take something you do that's repeatable and and train the AI on it essentially. It's that's where you can start saving time quick. Yeah. I want to go back to the the question I Stephen said he has to run. Thanks for being here. But about in the metal quote, I think one of the things that you could do, how can you use it post job, right? The job's done, the building's built, the roof's installed, whatever that is. How can AI help you understand maybe where you were, where you had some inefficiencies? waiting for someone to bite. it again it's it's the it's the depth of the data behind behind it and you know in in any CRM platform if you've captured that data and and you you identify shortages or material waste or or gaps in your process through that and document it inside of your platform, the AI tools can reach back into years and years of history and understand what what has been happening and give you insights on how to fine tune future outcomes, you know, based on based on the past information that's been going in there. So and I think that's one of the you know huge benefit. And it's an unta it's an untapped resource in many of the CRMs that are out there, many of the project management software platforms that are out there. They've got if you've been on it for 10 or 15 years, you've got probably thousands and thousands of customers and consumers using some of the AI tools that these platforms are building that allow you to understand your future customer and and where you can go to actually stay in touch with the customer, reach out to the customer, understand when they might need a new roof based on storms or weather or aging or, you know, the relocation or house sales. I mean, there's So many things that you can do inside of your own customer base, regardless of going out and trying to find new business. and I see it routinely in the contracting business, you know, especially in storm contractors, they're out trying to catch the new customer and completely ignoring their existing customers. And I think that's a really an untapped resource for a lot of contractors to dig into. with existing data. And I think that's I think the tools that are gonna come out are gonna have allow contractors to better predict the behavior of their existing customers down the road. and really contractors who are not just roofing contractors but exterior contractors, whether it's metal roofing or pole barns or siding windows, doors, all those things, you can understand the propensity of the buyer really to to buy additional services from you. And and and and you can automate those tasks to stay in touch with those customers. So you never lose touch. You spend a lot of money acquiring the the customer acquisition cost is very high in the roofing industry and or construction industry in particular. And once you get you don't want to lose them. And use those AI tools to really dig in and stay in touch with those customers. So Thanks, John, for your question. How can AI help your waste factor calculation? I know that was always way back in the day in early of Eagle View. It was waste. help help me figure out my waste. And and people kind of settle on this 10%, right? As kind of this a standard. But how could AI help your waste factor get down to 6%? Right. And is anyone actually doing that? what what information would you put in to an AI, what would it need to know in order to determine that waste? Ken, I think you might have some thoughts here. No, I I have thoughts, but I how the application, but I think Ryan and Rob are going to add icing to this cake. I a lot of the waste factors on any project is based off experience, right? You look at it, you know as a contractor that the roof's cut up, it's this slope, you're going to have these type of issues. and I think back to Ryan's teaching part and then how Rob had to do this with The Eagle View thing in the beginning, I think the discussions will come up. Cause this used to be one of the biggest shortcomings was you know, they put an average 10% waste in and it it didn't cover a lot of projects. And people said, well, the the system's flawed. Well, it isn't the system's flawed, it's giving you base information. And I think this is where you'd have to teach, and I'll let somebody else speak to that part about how you would teach your AI bot to be able to understand. by so many square feet and by such a slope and by the amount or degree that this roof is is cut up or section, you had to add a certain amount of waste. 'cause that that would be the challenge, right? Is like in a valley in a in a metal application, every piece you cut off is usually scrap. And depending on the slope, that gets affected mathematically because your drop keeps adding. So I I'll let one of the other gentlemen talk to that, but I think it's a great question and I think this would be a fantastic use of AI. I'd say, Karen, it goes back to that after the job point you made a little bit ago, right? I I think it's it's a circle, right? Hey, we estimated that this, this is how much waste we had. Okay, on the next job, let's learn from it, be better. Feed it back in. All right, let's estimate this job. And and I really think it's just now but instead of by doing a pen and paper or spreadsheet, and doing all these calculations, just give it data, right? The the a good AI, and it's to another question earlier I was gonna bring up, but is clean, good data. Go we've this phrase has been around way before AI, but garbage in, garbage out. Right? that's even more true with AI because as a human we can say, well that that's just not right. Yeah, that seems off. The computer's just no, it's in there. I'm gonna keep going with it. So that's my take is just that that recycling and keep the circle going, whether it's a good job, a bad job, or somewhere in between, feed it back, learn, feed it back, learn. And and that's how you get better. Just like a human. It th that's not even an AI thing. That that's just in general. Yeah. And Michael commented that you to really teach AI to better calculate waste, you would need to go back and give accurate results after the job is done. exactly here's what actually happened on this job and to continue doing that and then it's gonna learn. you know, the more the more you feed it, the more it learns. I I wonder anyone out there, Steven, he left already, but he named his chat GPT Edgar. Does anyone else name their their AI tools? I have not yet. If you use Grokbot, Grokbot's one that's commonly people name because it's the way it's set up, but I I haven't. I feel like I as as much as I've used a a nerd with it, I like kinda refuse. I know. That's a little weird, right? With people making friends and, you know, it's their confident now. okay. I wanna focus on you know, make sure we we touch on as much as we can in the hour. The time goes by so fast when you're talking about a t a topic as interesting and as impactful as AI. And I wanna make sure we talk about sales and marketing, because it is going to play a huge role in helping contractors. And I'm gonna speak specifically for metal roofs, right? Because metal roofing is a much higher end sale. it's a longer lasting solution. And I'm gonna call on you, Rob, because I know you're you are doing a little consulting work in this area, but how can contractors leverage property data that exists to to be more precise in their marketing? It's it's it's a great question. And it's evolving super, super rapidly. And I'll and I'll point to a couple things. You know, Eagleview, for one, has developed a platform called Horizon and it basically layers the imagery and the data we already have on millions and millions of properties and then allows you to interact with it in an AI fashion. So with the you know a large language model can query the data inside of Eagleview, for instance, Tell me in you know, I live in eight hundred zero one two six. Tell me how many roofs are asphalt that are greater than six hundred twelve that have yeah, you know, haven't had a roof change in the last fourteen years. And it will take what somebody might have marketed to, you know, in the thousands and boil that down to the hundreds, you know, or of really good opportunities. And that's really that's That's where the the value comes. And I also work with reworked AI in this space as well. and they are taking advantage of the Eagle View property data and then layering in propensity scores or propensity to buy. And they're kind of their tagline is we find bad roofs with good homeowners. so they are it rather than canvassing or broadly marketing to entire zip codes or areas. they can understand through Eagleviews data what the profile of a house might be and then also understand who lives in that house and whether or not they would be willing or would consider having a metal roof installed. Obviously it's a higher end product. It's gonna cost more. in areas where, you know, where I live, we have an HOA, we can't have metal roofs. It's all assault here. But there's plenty of other areas where Cedar might be coming off. And because of insurance regulations, cedar can't go back on, or not regulations, but you they're just not underwriting homes with non-fire resistant roofs on there. and those are great applications. You could say, hey, find me every cedar roof that's fifteen years old in this zip code, that is near a roof that had metal on it, for instance, because evil view can see what cedar and and asphalt or metal and identify all those opportunities. And through through a tool like Reworked, it it works perfectly. It's exactly what they do. They can you you can find a house or you can install a metal roof, for instance, and then find all of those roofs in the geographical area, in relative proximity to the existing house that are key, not just structurally the right target, but from a buyer perspective the right target. And whether they have Whether they own the property, whether they have equity in their house to afford it, whether they've bought recently other homeown home improvement projects that would be, in the range or in the scale of a of a new metal re-roof. Those are things that are super powered. They're supercharged by AI now. and you couldn't do that otherwise. Manually, it would take you forever and it would take you driving down the streets and trying to understand this in a in a completely automated fashion. you can market specifically to those those really key opportunities, which is which is fantastic news for contractors. Yeah, definitely. Because it it, it used to be spray and pray. You you put the put it out to everyone and maybe I'll I'll get a few. but if you really know the data and the information and all the stuff that you just mentioned, it really does help for for targeting, for w knocking on doors, for doing mailers, for all of that stuff. replace those traditional methods either. It just enhances them. You know, instead of knocking on a hundred doors, you got somebody knocking on the twenty-two that fit the profile and you're you're hitting those, which means you can get the more houses that are more likely to convert to a sale and be much more efficient with you with your field sales team that's out there. Yeah, definitely. I'm going to I had muted you guys, Ken and Ryan, because your mics are so good. They're picking up every little chair creak and and whatnot. So before you talk, check your mic. But I want to circle back. Ryan, you had some thoughts on what Sean's point about you know, verify spending too much time verifying and then losing the time that you saved by using AI. Can you speak to that a little bit? Yeah, so I will say is deep as I've gone down this hole. A lot of times there is a two step forward, one step back. So let me first say that and just that, this is great. And then wait, what are you doing? This is going down a weird path. but one of my points I I wanted to make sure I say today was, you know, that trust but verify we we touched about, right? and I there's a cool word that Sean used. I think it was where is it whi information with incorrect information with confidence. And that's true. so I'll I'll circle back to that, but there was a few points that got me thinking on that that I think is just good advice for people. First of all, I'd be curious, what system and what model you're using. when I first really dug into to AI, I was I had a Google account and I, you know, think I even had a limited Gemini subscription with it. So I I I really played in Gemini. And, you know, Karen, we joked about this last week, I think, where it's like Gemini from from February and March was like MS DOS computing, you know, in the eighties and nineties to what we have today. I mean, and that's six not even six months ago, right? So where I'm going with that is there's a thing you might hear AI halluciniz hallucinizations. And it's still a real thing, but I will say that the later models, these newer models, are much better at it. they're much better at at at not having hallucinators hallucinizations. I'm I'm just gonna give up on that one. but a few things you can do to help with that. First of all, the confidence part, right? you can tell it in your instructions or even just during a chat, hey, I want you to always be brutally honest with me. I want you to not sugarcoat, I want you to be concise, right? Like tell it how you want it to talk to you. a thing that can cause issues is long chats or long conversations with the same chat, whatever you want to call it. The longer you're in there, the more it it gets loaded with tons and tons of information. And there may be contradicts contradictions in there that you've talked to it about before. And I can go on and on. So I'm trying to, keep it a little more concise myself. So I'll give my own advice. But you want to after a a number of prompts, and I don't have a number for you, but a lot of different ones have a thing called a takeout or a summarize this chat, right? And then start over with a new chat, but load that summarization in. And what it does is it just takes out again, just it could be between your input and then the output of the AI, thousands and thousands or tens of thousands of lines. And every time you hit enter, every time you hit send, it's funny. It rereads everything in that chat. So not only could it give you that that mistakes or errors, but it's also chewing up your usage. And usage and models and all that's a whole nother topic. We don't have time for today. But the just with that c the the big things I want to take away there is i is do not keep going in the same chat over and over and over and over and over again is one. two, pick the model and whether it's Chat GBT, Claude, Microsoft, which by the way, Copilot actually uses Claude and and Chat GBT models. So it's people say, Copilot stinks. I'm like, Well, you're using not Microsoft's model, but that's a Again, another side topic. But at the end of the day, it's just you you you do have to check it, no doubt. You do have to make sure you're getting good things. But if you kind of do some these best practices of not having huge prompts, of having those guardrails in place, you will get better results. And again, the difference from using like Opus five five, which from anthropic is a very solid model to early Gemini is massive. And having said all that, Claude and ChatGBT have been going back and forth. And now all of a sudden Google's back in play with their new Argon, it's called, that is now right on their level after being, funny enough, Google being the laughing stock of AI for a while now. But it's going to keep happening. so there's a new one over here, there's a new one over there. They're just gonna keep leapfrogging and and models getting better. But that's just a thing I wanted everybody to know there that you you can at least help control some of these issues by model choice, reducing your chat size, having those guardrails or rules in place. Yeah. So thoughts on using multiple. Like I know some the you hear, well, this one is better at graphics. This one is better at text. This one is better. we've really tried to stick with just Copilot. But we understand that there may be some things that we want to do that another model is better for. So any thoughts, guidelines on what is best practice? Should you have four accounts? Should they be talking to each other? Yeah, I have a take on that. But even again, even within I don't think like for example, I I primarily use Anthropic, but I do dip my toe in the other ones as well. But even within Anthropic, they currently have four kind of key models. The the top two are Fable and Opus, then you have Sonnet, and then you have Haiku. Okay. Haiku is very good at grunt work. Like it's my go search the web and find me data. And Rob to a similar example, my father runs a business where he does a lot of buried oil tank removal and abandonments in Long Island, New York, right? People switch to gas or electric and they have to do something about this oil tank. So I had haiku, because it's very cheap to use and just quick, search where where are the most buried oil tanks. And it gave me this awesome 12 page report that would have taken me a week to compile and make it pretty. And I just had it go do its thing. So in that case, that haiku, their lowest model, was great at that because I didn't want it analyzing or anything. I just wanted you to go out and search and go nuts. But then when I want to code an app or do a video or something like that, yeah, I'm going over to Ape Opus or Fable. And then Sonnet's that good workhorse in the middle. It runs a lot of my daily skills, a lot of emails where it's I need decent performance, but I don't need to, use all my usage or get a big bill from anthropic for going nuts. So To your point, Karen, yeah, there are people that are gonna even go bouncing out of Claude to Anthropic to I'm sorry, ChatGPT and all these different ones, but most of the services now within their own world have, you know, two to four models that you can kind of choose from to get what you need based on the task. And again, what did I say earlier? If you're not sure, ask, hey, I wanna do a video of this. Which model should I use? Like, you know, yeah. If you're not sure, ask it. Ask it. Yeah. Okay. So I've heard a couple times about usage limits and bills and costs. So that we could probably spend a whole but just Ryan, you've got so much knowledge there, and I want to make sure that you you really do. I mean, and I know you've like figured this out mostly on your own, but if someone does subscribe to a premium level of of whatever AI platform they decide they want to use. Do they need to be aware of usage and overcharges? Yes and no. for the most part, if you're on subscriptions with most of these, most of them have for like a weekly limit, and then like a five hour limit. So you can't just burn through your whole week or month, you know, in a in a short amount of time. that's where again you that model choice does matter. if you have an unlimited budget, sure, go the top end model on the highest thinking and all that stuff will be great. But it will chew up your usage. And they measure it in tokens, and I'm not gonna go way down that path, but basically it just measures what's sent, what's output, and all that. But I learned this lesson when I was using Gemini. I was using it, going back and forth. It's like, you you you you can't use it again until twelve forty seven. I'm like, what the heck do you mean? and so that you know, was like, okay, What do I need to do here? And that even goes back to the some of the stuff I was saying earlier. you sometimes want to say, All right, wow, we've been chatting a while, condense this and then and then let's start a new chat because it's chewing up my usage. very complex answer if anybody has more questions on it. But yes, there's gonna be different limits from free plans to twenty dollar plans to I think c chat GBT lost a launched a five hundred dollar plan recently. So You can go nuts, but I'm gonna s just take a guess that probably ninety-five plus percent of people are gonna be fine with a twenty or even hundred dollar plan. It's really when you get into heavy video or heavy coding of things that it's gonna chew it up in a big way. When you're having it help you write emails or summarize something, they they've gotten pretty efficient at those types of tasks. It reminds me when cell phones first c rolled out and we all had minutes. You only had so many minutes. and now minutes? What? Everybody's unlimited. So maybe we'll get there one day. I know we well we've only got like nine minutes left, which is crazy. but I I wanna ask the question of all of you and even those on the webinar, if you have any thoughts on this, is there a way, you know, one of the things with that we look at as being kind Of a struggle in the industry is training, right? Is our labor force. So is there a way that AI can help preserve decades of industry knowledge as our workforce ages and retires and then somehow be used to train the future workforce? I don't know the answer to that, but you what are your thoughts, gentlemen? And Fred, good point. Nobody calls anybody. I don't know. I don't know. that's what we we try. Ken, as being you know, part of Carlisle Architectural Metals training is a big part of what you guys do. Yeah, it it's the biggest challenge that we have is how do we pass on, you know, the tribal knowledge from one generation to the other. And you see it in every business, not just roofing. So it's it's the challenge. And I I think this comes back to the points that we've talked about over and over today about how do we get that information from the people that have done some maybe historical stuff that doesn't exist as often anymore and get it into something like this so it can be shared for the future. a lot of it's hands on and I'm surprised we haven't the great thing is we probably don't have those people on this call today. Because they're going to be like, it'll never replace me because they can't do what I do because I'm a I'm an artist, I'm a which is the truth. but there's a way to to record that data so that somebody in the future can go look for it, right? And and be able to use it. How do we do that? That's the challenge. So and I think Carlisle's stance on that is we've been working on it for the last 10 years. I've been part of a bunch of different groups over the time where we try to get the different management groups together to figure out how do we bring the next group of people into this. application and how do we not make them understand the old way but make it available I think is the key. Yeah. Well taking this coming from the the asphalt shingle side of the business, for years, I always thought it was funny that I would go out as a manufacturer and go to a job site where somebody had improperly installed our product. And I explain to them that on every bundle wrapper the instructions are there, whether it's nailing pattern, whether it's offsets, whether it's valley applications. And they're like, listen, Sonny, I've been doing this for thirty years. I'm like, hey pal, you've been doing it wrong for thirty years. And you've got s you've got eighty seven copies of the application instructions sitting there like on a palette for you. I think the the best thing we can do for the industry from a technical and training perspective is is use Use AI to dig into all of the manufacturers requirements. You know, the technical requirements, whether they're coming from, you know, the Metal Contractors Association, NRCA, all the asphalt manufacturers, and be able to distill that down into the right right way. And and I would say even it would be awesome to be able to take a picture of a if you're sitting on a roof. And you're installing some flashing detail or some kind of thing that, architects love the rain valleys into sidewalls. it looks great, but doesn't waterproof very well. And if you can take a picture of something like that and then have it run in run into, the database of of application instructions and and details and provide that answer to a crew on the roof, that would be not obviously not only a training thing, but it would it would support the the installation. Cruise and reinforce the right way to do it. that's ultimately the manufacturers need their products to be installed according to the manufacturer's instructions. Warranties matter, insurance matters, all the all those things I think would be a really good direction to take training in the industry. Yeah. Agreed. That's one thing AI's really good at at you know, looking at those that technical information, those installation instructions and and kinda helping put it in layman's terms, as all the technical people are going, no, don't don't do that. but yeah. Anything to add there, Ryan? Pretty solid. I mean no, I don't really have much to add. I'll shut up. I've said too much. Nice. All right. I am going to bring our slides back up here and thank you to the Metal Construction Association for sponsoring this coffee conversation. It is a topic that we have been talking about for months and we will continue to be talking about for months because it is changing so fast and all the time. And you know, we we wanna stay on top of it and we want to hear from you guys. Thanks for being all the attendees. Thanks for being here. Thanks for sharing what you're doing, what you're using. And we hope that you took some valuable information away today from this conversation. And thank you, Ken, Rob, and Ryan for being here. I really appreciate your you sharing your wisdom and knowledge as well. Truly our pleasure. Thanks, Karen. Yeah. And hopefully we'll see you guys on a future coffee conversations. We have them on Thursday mornings, always at ten AM Eastern, seven AM Pacific. thanks for being here today, and we'll see you all on the next one. Bye.