³Ô¹Ï²»´òìÈ

Good afternoon, everyone. Welcome. We have an action packed hour, so I’m gonna kick things off. My name is Leah Vusich. I’m the Director of Product and Content Marketing here at ³Ô¹Ï²»´òìÈ. I’m I’m gonna be the host of our session today. We all know AI is the talk of the town. It’s everywhere right now. Contractors like yourselves are seeing new tools pop up almost daily. Some are useful. A lot of them are really useful, but some of them maybe not so much, especially for restoration. So the big question that that everyone is asking is how do you cut through the hype and figure out what’s actually worth investing your time, energy, resources into? What’s gonna drive real long term results for your business? So that’s what we’re here to talk to you about today. And to do it, we’ve brought together three of the best, most intelligent people that I know. Not just because they understand the theory behind AI, but because they’re here at ³Ô¹Ï²»´òìÈ every day, turning this theory into practical tools that are gonna make a difference in your lives. So you’re gonna hear from our CTO, Ronuk, our CPO, Jeff, and our director of product management, Shaun. So these three people live and breathe this stuff every day, and they have some really great insights to share. So before we jump in, I have a couple of really quick housekeeping notes. This webinar, like I said, is gonna run for about an hour. We’ll try to have you out of here right on time at two PM Eastern. Please enter your questions in the Q and A box. If you enter them in the chat, I have a really hard time keeping track of all of them. So there’s a separate Q and A window in Zoom. So if you click on that, add your questions. I can filter and sort from there. So make sure that you put those in there. We’re going to answer questions at the end as best as we can. And this session is being recorded. So if you know somebody who wasn’t able to join today or you wanna watch it back later, you will get an email with the recording by the end of this week. So just watch for your inbox, and you can watch it again or share it with your team. Okay. I promise I’m not gonna talk the whole time. I’m gonna get out of here really soon, but I wanna just do some really quick intros. So I mentioned our chief technology officer, Ronak, is here today. Ronak not only is our CTO, but he’s also the cofounder of ³Ô¹Ï²»´òìÈ. So he has spent the last thirteen years figuring out how technology can solve real world problems for the restoration industry. He is the driving force behind ³Ô¹Ï²»´òìÈ’s AI features, constantly experimenting with the latest tools and weaving AI best practices into our product. Interestingly enough, Ronak doesn’t just build software, he builds literally everything. He is currently building his own snowboard for the upcoming ski season or snowboard season, and I think that really sums up his passion for taking ideas and turning them into something that you can actually use. So welcome, Ronak. Thanks for joining us today. Next is Jeff McDowell, our chief product officer. Jeff has been shaping ³Ô¹Ï²»´òìÈ’s product strategy and design for the past four years. But before that, he spent five years as the COO of an AI research and development company, which means he knows what it takes to make AI not just cool, but actually useful. Jeff is also a huge advocate for mobile technology, ³Ô¹Ï²»´òìÈ has a mobile first mentality, and Jeff really lives and breathes that. He spent twenty years at a wireless carrier as well as at BlackBerry. So he’s seen how the right tools can really empower people in the field and on the go. And Jeff also loves to loves to talk. And so if you want to get some insider information out of Jeff, you can really butter him up by talking about dogs. And I’m sure he will spill a secret or two about our ³Ô¹Ï²»´òìÈ roadmap. So if you wanna get in touch with him, just share some facts about dog. Last but not least, we have Shaun Coghlan, our director of product management. Shaun leads our team that turns your customer feedback or our customer feedback into the product updates that you see coming from ³Ô¹Ï²»´òìÈ every day. With more than twenty years of experience working with emerging technologies, anything from telecom to drones to hearing aids, Shaun now brings his experience to ³Ô¹Ï²»´òìÈ to take new ideas and make them work in the real world for restorers. And when Shaun is not focused on product roadmaps, he’s putting he’s outside putting miles in training for an ultramarathon, which I think tells us everything we need to know about Shaun’s ability to commit to a goal and stay focused on the long game, and he brings that into his work as well. Okay. So welcome, guys. I’m excited to to have you here. This is an amazing group. Before we really dive into the heavier content, I just have a few quick fire questions for each of you to really help set the stage on this topic. So my first question is for you, Shaun. Why do you think AI is such a hot topic for contractors right now? Yeah. Thanks, Lynn. Hi, everyone. Know, why is AI such a hot topic? It’s the promises that it can can bring to restoration in almost any workflow that you’re trying to work on, which is trying to how do I do things faster with more efficiency so I can use my time for where I really need to devote it? So in the case of restoration, how can you focus on the customer and restoring the property as quickly as possible? How how do how can AI help you achieve that kind of reduce the time in your workflows so you can get to helping the customer returning that property to to pre loss condition. I think that’s why it’s really such a big topic in in restoration these days. Thanks, Shaun. Ronuk, my next question is for you. What’s one myth about AI that you’d like to clear up right from the get go? That AI here is here to replace humans. It’s not. AI is very good at producing something, but there’s no replacing expert human judgment, especially in a nuanced industry like restoration. The opportunity here is to get to that judgment faster with better information available to you, the expert. So I wanna clear that up right from the get go. Thanks, Ronuk. Jeff, last question before we kick things off is over to you. Why do you think ³Ô¹Ï²»´òìÈ is uniquely suited to lead this conversation around AI and restoration? Oh, I get the easy question. Thanks, Leah. Good to meet well, not me, but good to see everybody. Great to be here. So yeah. So why is ³Ô¹Ï²»´òìÈ suited? Because well, as everybody who’s used ³Ô¹Ï²»´òìÈ would know, and as Leah mentioned earlier, we really laser focused on field documentation, the collection of data using a mobile phone, which is pretty much a a tool that everybody already has in their pocket. We really wanna be able to get down to the truth of what happened on a property loss, by by collecting data in the field where it happens. And, you know, that mission was really to pursue getting paid your full bill, getting paid on time, getting paid quicker. But it just so happens that that exact same data, that exact same, photo documentation, video notes, moisture readings, all the dimensions, all the other things that you collect within ³Ô¹Ï²»´òìÈ is all the data needed to automate downstream processes and and and make massive get massive benefit into automating, you know, things that other things that you have to do. So I think we’re just in a really good position given our our previous mission of providing truth and transparency to the claims process and and now using that data that we collect as part of an AI process that just gets a ton of, ton of efficiency. As Shaun said earlier, AI makes things faster. It makes things more efficient, and I think just ³Ô¹Ï²»´òìÈ’s in a in a great place to, to to provide for that. Awesome. Thanks, Jeff. I really I’m really excited about where this conversation is going today. So I wanna build on this and yeah, Jeff, you’re gonna I’m gonna kick things over to you to really set the stage for today’s conversation and help us make sense of the AI landscape and where the biggest opportunities for return on investment for Restorers are in in this space. So over to you, Jeff. Sounds good. Thanks again, Leah, and and thanks for hosting. So, I’ll I’ll talk a little bit about this slide in a little bit. I sorta wanted to start out with more of a story, before we get into some, you know, graphs. But, you know, I don’t think it’s just I’m not making a controversial statement. I’m certainly not making an insightful statement by saying AI is pretty hot right now. Everybody kinda knows that. It’s funny because it’s not it’s certainly not a new concept. It’s let’s go all the way back. I don’t wanna date myself, but going all the way back to the movie Space Odyssey two thousand and one, I think we’re introduced to a character called Hal who was artificial intelligent. I think that movie was made in the sixties. And and and, you know, that’s probably when when the the the term AI became known to people. Maybe it was a little bit, in that particular movie, kind of a scary thing as opposed to it is now, which is an extremely productive thing. But, nevertheless, it’s been around for a while. Lee also mentioned that before ³Ô¹Ï²»´òìÈ, I was with a company, a I r d company, and it started in two thousand and four. Did not become mainstream at all. In fact, two thousand and four, two thousand and five really was when mobile was coming. You know, BlackBerry was certainly popular, and that gave way for Android and and Apple for some time. So even though artificial intelligence has been around as a concept for quite some time, it really hasn’t become mainstream. And I think the reason for it is because it was extremely hard as I found out in my previous company. You know, we’re we’re it’s all about it was all about labeled data and all of these machine learning techniques to to eke out small amounts of value. And then around twenty nineteen, twenty twenty was when we had this brand new technology come around that I think everybody is familiar with at this point because I’m sure at some point you’ve used ChatGPT or Gemini or something like that. But that was the advent of LLMs. That was the the large language model, and that’s really what what changed everything. The ability to use natural language, as a as a query, and and re and get a response that was mostly natural language. I mean, everybody’s tried it. It sounds like a human’s talking to you for the most part. And it kinda made everything, unfortunately, for me, the my previous company kinda made all AI development that was happening before that almost irrelevant. I mean, luckily enough for me, I’d known our founder, Paul, for for many, many, many years. We went we went back to BlackBerry, and he was looking for somebody to help him out on the product side. So I had a nice easy transition, but it wasn’t so lucky for many other people that were doing AI research, and had their worlds changed when LLMs became very mainstream. But, you know, five years later, after their invention, they’re not slowing down. The innovation has just been incredible, and more and more you’ve got this human like, I don’t like to use the word sentient. It’s it’s a little too much for me, a little too science fiction for me, but you’re having a much more human like conversation with AI, and it is doing much more human like things when you when you when you ask it to. You know, the one thing about it, though, and this is where this graph will come in, and and you’ll see where I’m going in a minute, is I think, again, everybody’s had that experience of using ChatGPT, asking it a question, asking it to summarize an or, yeah, summarize an email or a document or help you rewrite an email. And, you know, you’ve done that, and and some of the results are kinda mind blowing, jaw dropping. I’ve had incredible conversations with ChatGPT. But even though it’s been sort of getting better over the years, I’m actually finding myself using it less, especially in a work context. And I didn’t really know why at first. I do know that you know, I would ask you to hey. I I gotta send this email to a to my boss or an important person, and I’d like to get the grammar perfect. And I’d go over to Gemini or ChatGPT, and it would do that for me. And but it was you know, I had to go into a in into a different app in order to make that happen. And I remember, at some point, me thinking that there’s a premium version of ChatGPT, and it was I think it was another twenty bucks, and it allowed you to upload photos documents and a bunch of other things that I need. And and I didn’t end up doing it because I just found it too too dis not disruptive. I just wasn’t using it as much. I just found switching back and forth between apps and also was more focused on me as an individual as it was the job that I was doing, understanding the product management world, understanding the design world. I could ask it individual questions that helped me, but it didn’t really help me and my job. It didn’t make my job better necessarily, or it didn’t make ³Ô¹Ï²»´òìÈ more efficient as a whole. And I didn’t end spending the twenty bucks because I’ve already got a hundred subscriptions to a bunch of other things as I’m sure everybody does that I’m not using, so I didn’t do it. But it did get me thinking, and and, you know, what I wanted to sort of set the stage for is is today’s seminar or or webinar is really to just give you an understanding of how ³Ô¹Ï²»´òìÈ is thinking about AI and how we’re going to be adopting it and and and weaving it into all of our products. But this is where things got interesting for me. We were we were using another product called Gong. I I don’t I don’t know if it’s a product that anybody would know about, but it’s a it’s a product that basically integrates with a Zoom or a Google Meet, and it records the call, and then it summarizes it afterwards. And our sales guys were using it for their sales calls, which was which was pretty interesting. But I I started using it for having conversations with customers, and I’m not particularly good at taking notes. I’m certainly not good at taking notes and having a a a a good conversation with the customer at the same time. I could do one or the other. I can’t do both. So when I started using Gong, I could I could have these great conversations with customers, understanding their problems, understanding how we could help, understanding how they use ³Ô¹Ï²»´òìÈ today, understanding what they don’t like about ³Ô¹Ï²»´òìÈ, understanding what they do like in about ³Ô¹Ï²»´òìÈ, you know, making suggestions on where we should go next and the type of features we should do. And Gong really allowed me to have those conversations because at the end of the conversation, I would get this email that would format the hour long meandering discussion in a hundred different directions, and it would tell me who was on the call with me. It would tell me the topics we discussed. It would talk about the different features of ³Ô¹Ï²»´òìÈ that we talked about. It would give me a list of their their wish list of things that we talked about that they wished ³Ô¹Ï²»´òìÈ did, and it would it it would then give me a list of action items that we had discussed. And all of a sudden, it just became a huge enabler because that was and then I would get in an email, which is where I spend half of my day anyway. So it was this very transparent, nondisruptive process that delivered a ton of value because I could just do what I was there to do, which is have a good conversation about product with a with a customer who who is very keen to tell us all about their ideas and their problems and other ways we can help them. And and I didn’t have to focus on, you know, taking notes or trying to understand what they were talking about and having to reiterate things because I knew all that would be captured in the summary. And that’s really the antithesis of the way we started thinking about how we can integrate AI into ³Ô¹Ï²»´òìÈ. This this little graph that is up on the screen right now just sort of talks about it just sort of illustrates what I’m talking about. If if if you look at the x axis long term business value, if you if you stay if you’re really just using general purpose AI, chatbots, or ChatGPT, which, again, I’ve spent hours to I’ve I’ve spent hours talking to ChatGPT, but for personal reasons, whether it’s medical, health questions, whatever. But I’m not getting a lot of long term business value out of them, but it’s general purpose AI that can help me as an individual. So I got a ton of value out of it, But I never really found it helpful for ³Ô¹Ï²»´òìÈ. And what I realized after using, one of, that Gong tool I told you about is in order to get real value, real long term business value, AI needs to be purpose built. Ronuk and Shaun are gonna go into a little more details of what that means, but just at a very high level, purpose built AI is is is basically, you know, AI that helps you do the specific task you’re trying to do in the way you already do that task, not be disruptive. It’s it’s disruptive technology in that it blew away anything that you know, any automations that happened before it, but it’s not interruptive. It’s not interrupting the way you do work. It’s not making you go to a different app. And, you know, that’s really the key. So while you’re while you’re evaluating I solutions AI solutions out there because, you know, as Leah mentioned, there’s just tons coming to market these days, Really ask yourself, have they have they properly integrated this workflow into what I already do what I already do on a day to day basis? Is it taking advantage of the data that I already collect in the field and and not making me do things in a specific new order that changes your SOPs? Is it just allowing me to go about my business of mitigating a loss and and documenting a loss so that I get paid and then taking all that information and using it very purposefully in automating all kinds of things that just really deserve to be automated because they they provide you a ton of value. They save time. They save money. They save resources. You know? I think Shaun’s probably gonna talk about the whole scoping and estimating process. I think a lot of people spend time sitting in a truck or sitting in an office, you know, trying to do all that paperwork. You know, really needs to be some automation there, and AI is perfect to do that. So, you know, when I just get thinking about how does ³Ô¹Ï²»´òìÈ or how is ³Ô¹Ï²»´òìÈ going to approach AI, I think about very much in the context of purpose built and then and then not being disruptive not being interruptive, sorry, to the workflows that you guys have already sort of put in place. And nobody wants to have to learn or do training and and learn something different. So that’s that’s really the philosophy that that we’re taking. That’s great. Thanks so much, Jeff. So if I can summarize, like Don, summarize what you just said, some some key takeaways. General purpose AI, they’re helpful and drive a lot of value for an individual. But when you look at something like purpose built AI, which I know Ronuk and Shaun are gonna get into, that’s where the real business transformation, the real ROI comes in. I love what you said about being we wanna be disruptive without being interrupted, so we want disruption without interruption. So we wanna transform those business processes without actually changing those business processes, which is a a interesting way to think about it, which I really like. Okay. So, Ronuk’s gonna build on that next. He, like I said before, is the the one leading the team that’s building and incorporating AI into into ³Ô¹Ï²»´òìÈ. So, Ronuk, I’m gonna hand it over to you, our technology AI professor. I don’t know if you have your elbow patches on today, but I’m gonna throw it over to you to share how your team is engineering AI into our product for big ROI, but also for trust and reliability. Cool. Thanks, Leah, and thanks, Jeff. Yeah. I’m gonna build on top of what Jeff was talking about with general purpose AI and purpose built AI, and I’m gonna do it with my T shirt on. No elbow patches. So here we go. So on this slide, it’s a very simple slide. But on the left, you have general purpose AI, which you should think of as ChatGPT. Super flexible, great at a lot of things, but can also go off track. Especially in restoration where accuracy is critical, that enter unpredictability is a risk. Not saying that it’s not a great tool or it doesn’t have use. It’s just something to be aware of. And on the other end, you have purpose built AI. So this is designed for very specific workflows with guardrails and checks in place. It’s harder to build, but it’s what delivers consistent, reliable results at scale. That’s where ³Ô¹Ï²»´òìÈ is focused. And I wanna use this time to contrast the two approaches in sort of the very specific comparisons and metrics that you should be, using when comparing, any of any any AI tool, and using it to apply to your business. So general purpose AI, is not restoration specific. It’s general purpose, and it’s optimized to be good at many things where purpose built AI is restoration specific and is engineered to give higher weights to the things that matter for restoration. So think of standards like the s five hundred, A purpose built AI tool for water mitigation would give higher weighting to those standards and those pieces of information than a general purpose AI would. I also wanna touch upon privacy and confidentiality. General purpose AI is designed for all sorts of data, especially the free tiers. You obviously wouldn’t have a contract if you’re using a tool for free. So, you know, who knows what the privacy and confidentiality there would be. But general purpose AI is designed to, you know, help with math homework and cookie recipes, and you don’t want that exact tool to also have confidential policyholder data or, you know, all the other restricted or private information that a restoration contractor might have access to. Purpose built knows about restoration specific workflows, and it knows about the higher duty of care and standards that we that our customers expect, from us from a privacy and confidentiality perspective. Wanna talk about critical judgment. General purpose tools, they tend to be sycophantic a little bit. They tend to be overly agreeable and not apply enough critical judgment to the task at hand. How many times have you, you know, said something to chat GPT and it goes, that’s a great suggestion. And, you know, and you challenge it again, and it totally changes its mind. You know, it those are examples where if you’re in a critical in in industry like restoration, it needs to apply critical judgment to the task at hand. It can’t just agree with the information that you’ve given it. And in the case of purpose built AI, if the purpose built AI resulted in a misleading response, that’s could would be considered a bug in the mislead in the purpose built AI. But in the general purpose AI, it’s just, oh, you know, shrug your shoulders. It didn’t quite do what you wanted it to do. Let’s move on. At best, you could give it a thumbs down, whereas a purposeful AI that resulted in a misleading or not critical response where it didn’t ask you follow-up questions in the right way or made an assumption or hallucinated some detail, that would be treated as a bug that we would then dig into and and fix. And then finally, I wanna touch upon limitations of the the general purpose AI paradigm. ChatGPT, again, you have to give it a request. You type in or you upload a picture or, you know, you do some user initiated query, and it completes with by giving you a response. And there’s only so much information you can pack into a single request to get a single response. And what we’re what we’re what we’ve learned and what we’ve seen in all our testing is that the more nuanced, the more complex a situation gets, and they can get quite complex. The amount of information that you can pack into a single request starts becoming limited, and the AI model will struggle to extract all of that complexity out of a single request to come back with a response. You can get around it by asking follow-up questions, but that’s where that criticality piece that I just talked about becomes relevant. Whereas in the purpose built case, we are not limited to a single response and request. And, like, some of the flows that we’ve been working at in ³Ô¹Ï²»´òìÈ, we have, you know, maybe nineteen to thirty times that we might invoke an LLM or AI tool to get extract a very specific piece of information for a very specific workflow and then combine that together into building out the overall response. So we can work around the limitations of the complex use cases that are exist in restoration and produce a high quality, reliable, robust output that avoids many of the pitfalls that these general purpose AI tools fall into. So taking a step back, overall, there’s clearly lots of differences between general purpose AI and purpose built. But and if you are keeping track, people talk a lot about things like prompt engineering, and there’s a lot of other words that get thrown around in AI. But if you are evaluating an AI tool for your business, really ask the vendor about evaluations. Ask them how they’ve evaluated the tool that they’re selling for the scenarios that you are trying going to use it in. Ask them about where it works really well and where it’s falling down and what are the failure modes and how they’ve worked around it and, you know, what are the percentages of failure mode. So this is the word to use is eval or evaluations, and so just ask them about their eval process. Even if, you know, you you feel unqualified to maybe take in the response, just having a vendor that thinks about evaluations is a is a good sign. It’s a green flag. And one that doesn’t would be a red flag. Okay. That’s enough theory. That’s just a sort of comparison of general purpose versus purpose built. But let’s talk about how ³Ô¹Ï²»´òìÈ is thinking about AI. If we can move to the next slide. There we go. Alright. So this is an hourglass. This is the ³Ô¹Ï²»´òìÈ hourglass without AI. Historically, we’ve got all you know, this is the the magic of ³Ô¹Ï²»´òìÈ. At the very top, you’ve got unstructured field data, and we have best in class field documentation workflows to really get out of your way and get you taking the pictures and videos and notes and readings in the field, offline capable, and organizing it. And in the middle, you have the ³Ô¹Ï²»´òìÈ report engine, which is where the magic happens, which is where you can tell that curated story to describe what happened in the loss for the audience that needs to hear it. And then at the very bottom of that hourglass, you have those defensible reports that can go and service whatever needs that report needs to do. It could be a twenty four hour report. It could be, you know, an an like, a wrap up report. It could have all of the media in there if you’re trying to tell the whole story of what you did. It can service the needs that that well, of a bunch of problems and really open up the the wide hourglass from a very narrow waist. And what I wanna talk about here is evolving the story with AI. And so if we go to the next slide here, the evolution of this narrow waste is where the ³Ô¹Ï²»´òìÈ AI processes that processes field data from something structured becomes immediately useful to the contractor and keeps the the the process going. And and, specifically, I wanna talk about scoping and the scope document. The unstructured field data can use AI to produce an AI generated scope. And in that narrow waist, the critical moment where we can keep the human in the loop, where the restorer with their can apply their expert judgment and take, all of their knowledge about restoration and the specifics of the property to adjust that AI generated scope to, you know, increase the accuracy, be more robust, apply their judgment, so on and so forth. With the report, that was kind of the end of the value chain. But with the structured scope as a foundation, we can then apply the AI to branch out into all kinds of additional value added uses, real time estimates, instant compliance, QA checks, work orders, crew scheduling, the whole nine yards. Right? And so we apply that same story from unstructured to narrow waist to lots of different outputs with AI really leveraging leveraging AI and human in the loop at the right places, to unlock tremendous amounts of value. And that’s the key. Building trustworthy AI takes serious engineering. We rigorous rigorously test the failure modes. We add safeguards. We build for real world messiness. We keep the human in the loop at the key point where their judgment matters. So that’s why you can trust ³Ô¹Ï²»´òìÈ to scale AI responsibly. Thank you, Ronuk, professor. That’s the super clear picture of how ³Ô¹Ï²»´òìÈ is really thinking pragmatically about incorporating AI and building it to be trustworthy. So we’re gonna shift things over to Shaun. Shaun is working directly with customers day to day on how they’re using ³Ô¹Ï²»´òìÈ and the problems that we can solve and building building these theoretical concepts or AI tools into into real features that drive real value. So, Shaun, can you show us what this actually looks like in the workflows Restorers use every day, and where can can Restorers expect to start seeing value? Yeah. For sure. Thanks, Leah. I think Jeff and Ronuk really laid a nice foundation of kind of AI and how we’re thinking of it as a company and being able to apply it to our solution. And then my team and myself are taking kind of these core fundamental thoughts that we have and structures that we have and now overlaying that into our solution and how we can bring this to market. So if we just step back a minute and kinda ground ourselves and kind of the restoration workflow so as Leah said in the beginning, I’ve with ³Ô¹Ï²»´òìÈ for five years, and I basically distilled all my knowledge down into this little circle. And this is the general workflow of all restoration jobs. It’s a bit tongue in cheek. Every job is very, very different and has its own complexities. But, really, at the end of the day, it comes down to a job comes in, whether that’s through an insurance carrier, a private job, TPA, however you get that job, you go and meet the homeowner or property owner, and you get an assessment of, from their point of view, what has happened. And then your restoration job kicks in. So you need to go out there, do the initial assessment of the job site. While you’re doing that, which includes taking pictures, documenting notes, getting the dimensions of the workspace that was affected by, by the loss, You’re kind of, in your mind, generating kind of the work that needs to be done, which is kind of starting that scope of work where you kinda see in that second, phase of of of the circle. In parallel, you’re probably telling your team as your crew shows up on-site, okay. We gotta start doing a, b, and c, and you’re making notes of this. All the while, you’re collecting all this data and all this information. And with ³Ô¹Ï²»´òìÈ, as Ronuk said and as we’ve talked about, we provide the best tool to do all of that right in the field. And that’s kinda been our mantra that every you know, the real things happen in the field. And I think there’s been this saying that’s been going around, you don’t necessarily get paid for what you do. You get paid for what you’ve documented. And that’s why, like, our legacy as a company has been to provide the best tool to document the loss and then document the work that goes on. And this is where the circle repeats. Your crew, yourself are doing the work. You document what you’ve done. You may uncover something new. There is mold behind the wall we didn’t see before. Water creep crept from the kitchen into the bathroom. We didn’t see that before. You need to document that. You need to go back, modify, refine your scope, update your estimate, and continue. And that circle, that kind of messy circle keeps going and going until at the end of the day, the property’s been restored. You’ve done the service for the customer, and can you move on to, like, the end result of of the things? So ³Ô¹Ï²»´òìÈ has really been in there, and for the most part, really focused on providing solutions to how do you solve and get as much data as you can as quickly as you can in the field. As Jeff and Ronuk have been saying, as we evolve as a company, we wanna start adding more and more, technology, particularly AI. And I think the key thing that we’ve heard loud and clear from our customers and has really been kind of a a goal, a lighthouse for us, is the idea of disruption without interruption. So even base in ³Ô¹Ï²»´òìÈ is that, like, you guys don’t wanna have to document if you don’t want to. So but we’re providing a tool to make it easy. You don’t have to use a digital camera and a notepad and then download from a Dropbox and and hand type all these notes at you before. We provide the technical solution that allows you to do the things you were already doing in a very simple way. So we move on and kinda look back at what Ronuk just showed us from this hourglass of, you know, how is ³Ô¹Ï²»´òìÈ evolving? So going from that very simple, we produced really great structured reports to now with this new technology, like Jeff said about this kind of advent of LLMs that allows us to take advantage of these massive algorithms and our expertise and knowledge of restoration that Ronuk talked about to be able to take this unstructured information that you see in the field, a burst pipe under the sink, wet drywall, pruned hardwood floors, capture that with our app, which provides the context in that data. And, really, when using these, AI systems, technical systems, if the data has no context, you’re not gonna get a really good output. And that context is either because you’re manually having to add it in there or you have a tool like Encego that helps organize that information. This was in the kitchen, so I took the photo in the kitchen. And I can write a note about the cause of loss was the burst pipe under the kitchen, and it bled water for three days, ruining the hardwood floor, the covers, and it soaked into the bathroom, which was on the other side of the wall. So you can document all that. And now what we wanna do is how do we help take all of that information and transform it? And, again, the idea of how do we add disruption to your workflows without interrupting them? How do we make it easier and more efficient? There’s kind of two areas that we’re gonna focus on. The first one, if we go back to that circle, is documentation. So we’re not done there. There’s still more things ³Ô¹Ï²»´òìÈ can do to make documentation even easier and more effective and more efficient for you in the field. We’ve already added these things. I’ve already talked about we help you organize and categorize data, put photos in the right room, write notes. You know, we already have a great kind of advanced technology for doing dimensions with encircle floor plan, which is a combination of taking a video, leveraging AI machine learning, as well as well as photogrammetry, and then producing a highly accurate floor plan with the dimensions that with our integration into Xactimate, all of a sudden, you’re that way there into into having your estimate created because we create help create that sketch for you in a very short period of time. So we’re already we have already innovated using these technologies. Other areas that we’re looking for, where you can see AI playing a fact. And, you know, Jeff made this comment to me the other day, it really, it really stuck with me. It’s, you know, AI should do the things you expect it to do. So if you’re a contents company, you need to do packouts. You’re already having to document every item that you’re packing out, putting into a box, and then putting it onto your truck that goes to your vaults, goes to cleaning, or wherever it goes. Or if you’re, having to deal with total losses, you still need to document everything, describe it, price it, and then get that customer to kinda sign off on that and get the carrier to sign off on that as well. There’s a big portion of that workflow where there’s manual data entry. I take a picture of my water bottle. I gotta describe it, whether it’s something really simple because I’m doing a pack out that just says green water bottle, or I need more details because I have to price it and get more details for a total loss. There’s a lot in that workflow that AI can leverage. And we’re right in the center of because we’re already in that flow of capturing data with the photos, we can leverage technology to describe those things on your behalf. And whether that’s for simple descriptions or more advanced where you can get full complete descriptions that may tie directly into pricing. So this is a a workflow from a simple picture that with AI’s knowledge of the workflows of contents and our ability to write and customize kind of the AI purpose built AI, we can produce those kind of, features that you could maybe do if you went outside and use a general purpose AI. But then you’re leaving where all your data is, like Jeff talked about. You’re going in a chat TPT, uploading a bunch of photos, having to figure out your own prompt. And Circle’s gonna do that heavy lifting and just make it part of your workflow, and your team just continues to snap photos and pack those items. Another thing where it can really help is when you’re doing your initial damage assessment, you know, a lot of our customers record videos. We have, you know, thousands of hours. I think we had over three hundred, four hundred thousand videos in our platform last last year alone or two years ago, it’s only grown since then. A lot of good information in there, particularly the voice that describes what’s in those videos. So being able to create a summary of the audio from that video, and all of a sudden, you know, there’s really detailed note backed up by the hard artifact, which was that video, which then you can use in your report. You can use in your twenty four hour reports, your end circle PDF reports, or as I’ll talk about in a second, you can take that data that you’ve described the loss, and we’ll be able to pump that in and help you drive your scope faster faster. So there’s lots of new areas for us to get into in the field, kind of in the documentation side that will help kind of reduce the time. It needs to be done, but what can we add to kind of reduce some of the human time you need? And as Ronick said, just put you in there as a reviewer to make sure that, yes, this is the right thing. We have the right validation and verification. But end circle, because we know the industry and we know the standards, we can already put a lot of guardrails around those AI technologies to make sure it’s the right output. So when you’re doing your video and you talk about cat, it means it’s a category of loss, not, hey. There’s three cats in this house. So, you know, there may be three cats in that house, but we wanna make sure that you’re actually talking about the right thing. But we know that, and if you don’t provide that input into a general purpose AI, it may think you know, I guess if you’re at a veterinary, there’s gonna be cats everywhere. Anyways, you where I’m going. Next, which is I think the biggest thing and what Ronic alluded to really in that hourglass diagram is there is a lot of time, and our customers have talked about this, and we’ve had questions about it for a long time, that goes into how do I properly scope and create an estimate for this job. And as we all know, that scope is a living thing. As the work continues, it needs to be modified, particularly for large complex losses or older properties where water may spread everywhere. And you didn’t realize that when you first saw it, you open up the floor or open up the walls. All of a sudden, there’s more damage, and you need to adjust it. But by using nCircle and documenting the job properly, we’re primed with all the right context for this information with some of these new features that we’re coming out with to be able to take that and really garner it down because we understand the industry. We understand the industry standards, whether that’s IR CRC or others. We can help create that first pass of the scope based on that initial documentation. So here’s a narrative of what was gone because you did a video or you wrote a note that gave you the damage description. Here’s all the materials that were affected based on kind of photos that you’ve uploaded and notes that you’ve added to say, okay. In the kitchen, we’ve got hardwood floors. We got cupboards. We got countertops. It was a supply line. It bled through a dry drywall into the bathroom, which has linoleum floor and there’s a shower. All of those things, even if you don’t verbalize it because we have the data, whether that’s photos or others, we can contextualize that and produce kind of that first pass on the scope. Very detailed, and it starts covering things that, you as I said, when you’re doing your initial inspection, you’re already thinking of these things. So you have this cognitive load of trying to remember all these things if you don’t write them down. But because we understand the workflows, even though every job is very different, we can apply a lot of this standard logic and start creating scopes that match that. And then over time, as we integrate with our technology partners like Ferris, start producing kind of line item scopes once we as we work through that, which really can shrink down kind of those workflows. And really, again, add this disruption, but not interrupting what you really do because we’re really about transforming the data in the way you need it. And so moving from kind of this bigger circle and with end circle technology, and with the AIs that we come out, we can really, you know, shrink it down, particularly the amount of time needed to analyze and interrogate that data. And we’re gonna be able to produce outputs that are fully verified, that are validated against standards. And because we know the industry, you can trust that when you say this is a cat three loss, it’s not gonna say, oh, there were three cats at this house. It’s a fun joke, but, you know, that’s kind of where we’re going and where where we really want to kind of apply kind of this technology and make it even easier for you to work. So we’re gonna open it up to questions. We’ve had a number of questions come come into the into the q and a. And we also had a number of questions submitted in advance of the webinar. So we have lots of things to grill these guys about. Okay. So I’m going to just give me a second. Ronuk, the first question’s for you. Where does general purpose AI get its information, and where does purpose AI purpose built AI get its information from? And do they need to be connected in any way? Okay. I will ask more about what you mean by connection in a second, but I can answer the first part of that question. In that, like, at this point, like, where does it get its information from? These the companies that are, are that are, pushing the bleeding edge of these AI models, people like OpenAI, people like, Anthropic and Google and all these people, they are at they’re scraping essentially the whole Internet at this point. They are taking ginormous amounts of data to the point where they might be running out of human produced data that’s meaningful to feed into these models. And, you know, at the bleeding edge, that’s that’s what these models under the hood are that’s the datasets that they’re operating on, and they’re just sucking up everything. They’re sucking up people’s blogs. They’re sucking up PDFs like the IRCRC standards, cookie recipes, you name it. They’re just shoving it all into these models. And then the general purpose AI, so things that you see as a product like ChatGPT, is using those underlying models to produce its outputs in a way that’s helpful as an assistant. And that’s really where you start seeing, you know, people talk about prompt engineering. Like, that’s where, like, part of building a good prompt is, like, giving the model a role that it’s following. So you you can get very different results from, hey. You are a assistant that, you know, helps people with their math problems versus you are an assistant that helps people find recipes. And then off you go, and that helps the model produce different outputs. The purpose built AIs are also using the same underlying models, but they’re prompting in a very different way. And sort of what I alluded to earlier, they’re they can’t the purpose built AIs are able to build workflows that engage the model at multiple points throughout that workflow to produce the final output. So if you take something as complex as given all these pictures and given a walk through of and a damage description in the restoration context, produce a scope, which is a list of work you know, a task list, a high level task list of stuff that needs to be done, broken up down by room with measurements and all of this other stuff. There might be, like, thirty different areas where a model might be invoked to extract specific pieces of information to produce that final scope. And that’s really the difference between general purpose and purpose built is that they’re working off the same data, which at this point is the size of the Internet, but the approach taken is quite different. And that’s what leads to that difference in output. Great. Thanks. Jeff, I know you jumped in and answered a question in the chat, but I’m gonna ask I’m gonna ask the question again so you can share with the whole class. And so Joe asked, are you training your own model or have you or have really good prompt engineering to produce the correct out output? If you wanna just share your answer again or or maybe expand on it. Yeah. It’s it’s just really just a a follow on or a similar answer to what Ronika was just providing. The whole advent of LLMs that really came to prominence about five years ago really disrupted the requirement to collect all kinds of labeled training data, which is what everybody was doing per beforehand. Proprietary data was the was sort of the the the ammunition or the the the, you know, the the goal that you were trying to mine to feed your models. And and LLM just made that obsolete because as Ronuk said, they’re basically just taking every piece of data they can find in the Internet and and processing it into these large language models. So so, yes, it does become, you know, a prompt engineering problem or or a prompt engineering solution, I guess, which is definitely the core of of what we’re doing. The difference between like, I’m I’m just trying to think on the fly of asking a question. So if you go to ChatGPT or some other app that has, like, a a prompt availability in it and you ask the question, you know, what what what what would the scope be of a of a a three room water loss that included a bathroom and a dining room and hardwood floors and things like that? It it could give you somewhat of an answer, but, you know, it might not pick out details in the photos or the videos properly, and there’s no eval. Like, Ronuk talked about eval. There’s no eval process that’s specific to our our particular use case. So it’s not trying to optimize for damage descriptions or damaged items. So we have this series of prompt, which would be you know, again, if I was to to ChatGPT, I’d just ask a single question. When we do it and some of the research that we’ve done, that one question I just asked about what would a scope be for a three room water loss and a that one water supply, that prompt could be I don’t know. How many how many lines long, Bronik? And and written in JSON, not not English language. Thousands. It could be three pages long. Yeah. Thousands is what I said. Thousands of lines long. Right? Because we wanna make sure that not getting any hallucinations, and this was the point that I made in the in the we’re not in an industry where you can submit a a scope or an estimate to an adjuster and have a mistake on it because it’ll it’ll erode trust if you do that. So we’ve already taken into consideration we’re shooting for a hundred percent accuracy. And in these models that are generally trained, one of the downsides is hallucination. Well, you gotta you gotta build a system that just completely weeds out hallucination if we wanna if you want a serious answer, if you want a serious high value piece of output. And that’s that’s the approach that we’re taking. That leads into another question that that came through around around this idea of evals or evaluating AI. So, Ronuk, I’m gonna come back to you. When a restorer is evaluating AI for restoration, so I would call that purpose built AI, something for design for restoration, what concrete proof points should the decision makers or the people who are looking to implement or purchase these these tools, what should they be asking of these vendors or demanding during a trial of these of these technologies? Yeah. So the there there’s a variety of ways you can approach this, right, like, from the very technical to the very high level. But at the very least, the questions that you should be asking are around what are the failure modes that this vendor has seen in whatever the workflow is. Like, where what are the pitfalls? What is the the rate of failure? Because anyone that says, you know, it’s absolutely zero percent, they need to be able to back that up with, with why they believe it’s very hard. Right? But anyone that doesn’t give you an answer is even a bigger red flag because it probably indicates that they haven’t they haven’t done the work to properly evaluate the product that they’re putting. And in that sense, you’re just being the beta tester there for them. And when you find a a scenario that the AI model has fallen down on and then you bring that up to the vendor, that might be the first time that they’ve seen that kind of failure mode. And that’s that’s such a bad situation to be in, because you’re holding the liability there, and the vendor is learning about a bug that they that they have never encountered before. And so that’s the kind of question to ask at the the procurement level. It’s like, what are the failure modes? What are the evals that you that this vendor has done? And, you know, what are the percentages and places where the the model falls down? On a similar similar vein, you talked about liability. So I’m going to go down the path of safety and security. It’s something that comes up a lot. Data security, data privacy, especially in the restoration space with property claims and highly sensitive personal information. So how safe is entrusting AI with your company’s data? So for a restoration contractor, if they’re uploading company information or personal information of their clients, we know that AI needs data to be useful, and it needs that, like, rich contextual information. But where where do these legal issues or privacy informations fall in some of these solutions? Yeah. So I touched upon this briefly in my segment earlier, but, like, if you don’t have a contract with the AI vendor, that’s an immediate, like, flaw. They they could be doing anything with your data. The contract, I would say, for a restoration specific context needs to have some ability to say, the data that you provide, the AI model, will be kept private to your organization. It will not be, used in training cycles, that might, you know, leak, into other people’s context. And, certainly, all the tools that we build at ³Ô¹Ï²»´òìÈ, we we are able to provide those guarantees. Like, there that I guess, like, the nuclear doomsday scenario here is, like, your data somehow making it into some other organization’s context and then the AI model hallucinating a detail that came from your private data and leaking that out there. Policyholder names, sensitive information like that, so on and so forth. So, taking like, to summarize, it’s like, a contract in place, really evaluate that contract from a a a privacy lens similar to what someone might do for, like, a Dropbox or a Google Drive or what have you. It’s like it’s at that similar level of unique confidentiality and privacy to be in place. It’s not just enough to show value. It needs to it it needs to have the safeguards and guardrails in place to prevent your data from leaking out. Shaun, do you have anything to add to that or good? No. I I I think I think Rana covered it all. Again, yeah, we’re not in the business of training models anymore, so we’re not that that was absolutely a pretty major concern. But any data used in the context of any workflows will stay within your workflow. We’re SOC two compliant. That means something that that has a very serious security connotation of PII and data leakage and things like that. So, no, it’s it’s a it’s a priority for us. So, Jeff, I think the one of the last questions is this may be the last question. We’ll just have to watch the timing. This is a it’s a really good question. What makes ³Ô¹Ï²»´òìÈ different from other restoration AI platforms? So the person asking this question is an AI developer, and a lot of restoration AIs are popping up, which we talked about earlier. Xactimate, DrSketch, for example. It feel like every everybody is in the AI development space now. So what and I think this is just a great opportunity to recap the last sixty minutes and what is making end circle different in what we’re doing in in our approach to AI compared to what what other people might be doing. Okay. Let me I think I can handle this sort of generally, but then I’m gonna flip it over to Ronic to maybe talk about a few things specifically. So just generally, somebody who was it? Carmen. Oh, sorry. Cameron. Cameron was probably a bit disappointed that we didn’t, you know, get right down to the road map and when we’re delivering all of the stuff. As much as we would love to, and and at one point in time, a few months ago, we were hoping that by this time, we’d be able to be a little bit more explicit about what the road map look like. We wanna get this stuff right, and we do not think because of the because the opportunity for AI to completely optimize or even eliminate some busy work and workflows is is so great. We wanna get it right. And it would have been very easy. These LLM models have made it very easy to throw a prompt or some sort of lightweight AI tool into your app. And if we really wanted to, we could have done it a year ago, at the very least, when we started thinking about this stuff, or two years ago, we started thinking about this stuff. And we didn’t do that because, you know, the at first, there was a thought like, let’s say we get into scoping. There was a thought, well, if we get eighty percent of a scope rate, that’s pretty good. That means the the the you only have twenty percent to complete the scope. Well, thing is is that you don’t know what eighty percent we got right or what twenty percent you still have to fix or revise or add on. So we understood pretty early that this wasn’t a again, like, personal AI. If you just get close, if if you know, you don’t have to diagnose a disease, but if you if you’re able to provide some useful you know, take some vitamin c and vitamin d and and take the day off work, you know, that’s that’s a that’s a good enough answer. It might not be precise, but it’s good enough. In our world, you know, good enough isn’t good enough. We wanna be as precise as possible, and then it takes a a much deeper level and a much more serious approach to engineering. And and that’s why it’s taken us a little longer, to be honest, in terms of releasing these things. We we don’t wanna get it wrong. So in terms of the approach that we’re taking, we’re just taking a very serious approach where we’ve set a really high goal of the value that we’re gonna be delivering. So that’s on a high level. Ronuk, I don’t know if is there anything that you can sort of provide around our secret sauce just around the way the nodes work together for for evaluation purposes or for accuracy purposes? Is there some sort of metaphor you can provide? I don’t mean to be on the spot. I was gonna take it in maybe a slightly different direction. I was gonna steal also from Shaun’s, like, dictionary a little bit and talk about, you know, disruption without interruption. But, like, just if I had to summarize all of the things I could say really quickly, it would be just the amount of thoughtfulness that goes into building an AI product for this industry. It’s not just about the prompting and the workflow and the user experience. It’s also things like privacy and failure modes and evaluations and all of the other things that need to all come together into a product that you can incorporate into your SOPs without having to think too hard about when it’s going to fall down, without having to, you know, watch its back all the time and and and have to worry about deploying it. And so that thoughtfulness is what takes a lot of energy. It’s absolutely where we’re thinking, and where we’re spending our time, and that should reflect in the products we put out there. Okay. I know we are we are at time now, so thank you. I appreciate it. If I could sum that up, I would say, ³Ô¹Ï²»´òìÈ is a highly trusted brand amongst restoration contractors. Our customers trust us to build products that help them in the field and and to get that right. And so I think that that that trust we take very seriously. And and to reiterate what Jeff and Ronuk said is we don’t wanna we don’t want to get it wrong because our customers expect us to get it right. So that’s the end of the webinar for today. So thank you again for joining. Thank you so much for spending your time with us today. Thank you, Ronuk, Jeff, and Shaun for sharing your knowledge. Thank you again, and have a great day.