Ragitha === Rajitha: [00:00:00] Now LLMs have come a long way, they can do all of that. Work for you, but you still need to have something that's the ground truth that Jeff: Mm-hmm. Rajitha: Act as your data, that then you can deploy these LLMs to create more structured data that can be then used for outcomes. Jeff: Welcome to Launch Pod, the show from Log Rocket, where we sit down with top product and digital leaders today we're talking with Rita Chopper, VP of product at Procore, a construction management solution, and one of the most successful examples of vertical SaaS out there today. In this episode, we discuss Haida and Procore got hesitant. Construction executives hungry for AI by showing real immediate value in processing unstructured data, improving efficiency, and taking friction out of core workflows. Rita's critical processes to ensure data quality, consistency, and collection in service of AI solutions that deliver on their promises from day one, and how you can leverage third party data to unlock the promises of AI just like Rashita did. So here's a conversation. Witha Chopper at Procore. Hey Raha, [00:01:00] welcome to the show. Thanks so much for coming on. Rajitha: Glad to be here, Jeff. Jeff: You know, You're at Procore now, which has been a fantastic run. But like ZoomInfo Intralinks, clear result home site insurance, like a whole. Run of good companies. You just had a really kind of fast ascension there. But the one thing I do want to dig into right off the bat, this always comes up and we kinda always say it to the point where we've stopped doing this, but I need to re introduce this section because of your background. But like you do have a kind of. Different background than most PMs to coming into product. Can you maybe walk us through that? 'cause I think it's really key to how you've made a differentiated career through product, Rajitha: Yeah. It's true. I always thought it was an interesting transition. To product. , So I started my career as a credit risk analyst in the credit card industry. That's my first foray into using predictive analytics. And then from there back in the day when you did analytics, it wasn't, a function, it was how you did business, Jeff: Mm-hmm. Rajitha: started, you analyzed the data, you came up with the solutions of a [00:02:00] strategy of how to improve the business. Or improve a part. Then you implemented those strategies, you read the results, and then you improved, right? It's Jeff: Yeah. Rajitha: loop and you did the entire thing. Unlike now where you have specialists to do each part of it, back in the day we did it all. So that, that lends itself very nicely to going off to different parts and specializing. From there. I actually worked for an insurance company several stops later where I started as an business analyst analyzing our call center and online funnel. Trying to see Dropbox where we can improve. I. Ended up , working with a vendor to create an email solution to send out This was way before like email marketing systems and even marketing automation systems today, where we all build our own, used our own solutions to do it. And then from there the company was then building online solutions online servicing platform. And that's when I got dropped into The project, asking to do it. [00:03:00] Again, didn't know what product management was. I figured, hey, I can analyze data, I can improve funnel. This is the same as that. So my first introduction working with engineers was I told them, it was somewhat later in the project phase. They've already built the bare bones. And I came in and I said, oh, let's analyze the data because this is what I know. And they said, what do you want to track? I said, everything. That's my learning that with engineers, you have to be very Jeff: Yeah. Rajitha: You have requirements, right? You have to write down with mockups of what you want. So yeah, lot, lots of learnings. But , that was my introduction to product management. Jeff: Yeah, they're a literal bunch, but I do have to love the poetic justice. Maybe it's the fact that I've worked at startups through, almost all of my career, but you always run into. Engineers throughout who kind of like, oh, I can do that function. I can do that function. But\ it's great to hear kinda the other way where you're like I did data. I can make the product better. , let's have that. Let's go collect some data. And I can help you guys make this better, so [00:04:00] let's go make the thing better. And, you accidentally fell into product management, which is now in today's world. Massively vital. Back then in the, oh 8, 0 9, 20 10 time period, I don't think it was quite the function it is now. It sounds like even then though, you were able to make it a much more business impactful role than maybe some companies had. Rajitha: Yeah. The, again, I Jeff: Yeah. Rajitha: to my analytic background. I say using data to improve anything is the key. Jeff: Yeah. Rajitha: getting to business outcomes, right? So that, that's my North star, that's what I fall back into. And I still truly believe, you can do product management, but again, product management is based on analytics and based on understanding customer problems and then trying to figure out how do you solve the problems and then test and learn approach, so it's Jeff: Yeah. Rajitha: far out. And it's and more and more it's evolving into being a very data-driven approach. So there's a lot of synergies there. Jeff: So actually this is not something we talked about previously, but I am curious a little bit to just get your take. I was listening to, akash Gupta in [00:05:00] his podcast today, and he had an author on, and they were talking about this idea of impact driven product management because it kinda seems like at some level, through maybe the growth of product, we veered into a world where there's , two types of PMs. There are people who view progress as shipping features and people who view progress as. Impact, right? Like they understand the business, they're are moving a metric forward or some sort of business goal. Maybe it's not revenue 'cause it could be innovation focused development or something because it sounds like you fall very much and you've always kinda historically fallen on the impact side, which is I think a better way of looking at it. How do you look at throughout, companies roles. What are the things you need to focus on to drive that business impact or to return for the business or to, make the business grow ultimately, which is, our goals here is in product, is to solve customer problems and to grow the business, Rajitha: Yeah, ~like I said,~ getting to business outcomes and Jeff: yeah. Rajitha: the outcomes and what are the outcomes you wanna drive, and then backing towards what metrics and That we need to move to get to those outcomes. Then what [00:06:00] solutions we need to deploy and develop to create that. So that's where the funnel me Come in so handy. , if you're trying to do progress folks through a funnel to adopt a product or use a feature you're looking at where are they dropping off, where are they stopping? Jeff: Yeah. Rajitha: of the funnel the important thing? , there's not enough visibility for that feature or. Solution on the product so people can't find it, or are they starting it, but then dropping off because it's, at some Jeff: Yeah. Rajitha: and so having that visibility into the funnel and then going to improve that part of the funnel. LEDs to improving those KPIs and then the business outcomes, right? Jeff: Yeah. Rajitha: either, Hey, company needs to increase revenue, but then company needs to increase revenue by more customers adopting a product, or more customers buying a product. But even if you just say on the revenue piece, if you don't focus on adoption, then you churn that customer. Jeff: Right. Rajitha: get to how are they using [00:07:00] it? What are the problems they need to solve? Is it even a compelling problem that the customer Jeff: Yeah. Rajitha: this feature? So you, if you put them all together, they're not discreet outcomes, right? They all converge into a single thing. And I think the part of a product manager to take those high level outcomes and bring them down to the solution level and then the problem and tie them both together. . Jeff: , I love how you just casually whipped out a kind of nested funnel view that I think explains, impact driven product management in such a concise way. It's, do you have the problem that is even worth solving? Is there the pain there to do that? If so, can you get people to buy, or adopt? The solution you're selling, whatever it may be, do they use it? Does it actually solve their problem? Do they get value out of it? And then, can you get them to keep using it? But if you can do those kind of four things, you have a successful product probably, but , it's not super hard at some level, it's not super [00:08:00] complex. It's just like those four things and they're four discreet funnels and they nest. And if you can do those things, you'll be successful, the devil's in the details, but I don't think I've ever heard anyone just casually rip out, a couple sentences and explain that so nicely. Great kickoff. I'm stoked here. This is clearly, I think the analytics background here is a positive thing. , if we take that kinda analytics background, so you basically at home site fell into product because you're like. Let's make this better. I have ideas. And they said, great. Put your money where your mouth is. You are in product now. You did start to run into problems down the line where you started to see data is really important. And we lived this at the same time period where. I don't know. The data practices were fantastic back in the aughts. And you would always run into kind of shortages of data or the piece you needed wasn't reliable or wasn't there? Maybe you can talk about this, but like how did that work on your end and what did that turn into from a problem and a solution standpoint? Rajitha: Yeah, so try to do this a few times Jeff: I. Rajitha: Career at home site. One is just understanding, segmentation, trying to do segmentation of our customers [00:09:00] coming in. Where are we successful? Where are we able to convert them? The acquisition funnel we had an online quote platform. And folks would start courts online and then, there's a multiple steps to getting them to actually buy the policy. So I tried to do segmentation there. Initially when I first got there, we were asking our customers all of the data that's required in the court platform, right? So there's, as you can imagine, when cust when you ask people, what is your roof made of, you know, what kind of a Jeff: Hmm. Is your home? Rajitha: Is it a Victorian or a colonial? I don't know about you, but most folks don't know that. And what are your shingles made of? What is your wall made of I learned a lot while I was at Jeff: Yeah. Rajitha: construction, but before that I did not know any of this. And our, typical customers going through this flow were, fell into the same bucket, so they would answer to the best of the capabilities. But that didn't always result in any personalization that we could do because it was incorrect. Jeff: Yeah. Rajitha: And that was surprising to me because I had come from industry like credit cards [00:10:00] where you could pick up a lot of different segmentations with a keystroke only because the data, they was very automated, like credit bureaus, there's a whole industry around getting the customer profile right. And. Aggregating your addresses, where you've lived in the past, like what credit lines you have open. So they that's a huge industry. Then the rest of the information is coming in from your point of sale systems, your payments processing so very curated automated data and that , lends itself very well to like predictive analytics and all, most AI functions in general. But as we got through evolving our Quoing platform where we started to use third party information to prefill some of those fields because there are com companies and in this aspect, core logic that we use, that was actually Jeff: Yeah. Rajitha: all of that stuff from title data. So we had that information once we started Jeff: Yeah. Rajitha: It was a lot easier to do the strategies. Jeff: Yeah. Rajitha: that's my [00:11:00] aha moment is wow, okay. Data is so important and it's not just any data. It should be standardized, normalized. It should be consistent. You need to use the same terms to mean the same things. Jeff: Yeah. Rajitha: when you're doing these methodologies and algorithms, there's no human to interpret it. So that's way, if you, at least back in the day when folks would talk about creating AI and analytics, they would say 80% is in the data prep and Jeff: Yeah. Rajitha: actually doing the algorithm. Now LLMs have come a long way, they can do all of that. Work for you, but you still need to have something that's the ground truth that Jeff: Mm-hmm. Rajitha: Act as your data, that then you can deploy these LLMs to create more structured data that can be then used for outcomes. Jeff: that's a pretty good kinda realization though, is. We need this data. We need it to be effective. I think starting in the kind of credit area is a fantastic place to start because they have built so much, just like practice and expertise, just [00:12:00] aggregating all this and being so sharp on, that's how they make their margin. That's how they stay profitable, is they have to be able to give the right terms to the right people and understand your risk profile. But. That same kind of grouping of data can often be used that same practice of, third party data, right? You said home site you could go buy details on homes and that would give you a much clearer picture about, what you need to do to ensure that person. And probably more accurate than what they would self-report, but also, maybe at some level in the funnel or adoption it's an advantage too 'cause you don't have this kind of friction of getting through this. It just works. And especially. Back in that day something just working online was like a mind blowing miracle. Rajitha: We've worked a lot on streamlining the number of questions you have to Jeff: yeah. Rajitha: because the less you ask, the more the chance of the user, the customer going through the flow. The more you know, we, we know that like survey, anything you do at all. The fewer the better to get from start to Jeff: Right. Right. And so, you know, why not get more accurate better conversion, happier [00:13:00] customers. It's like a win win win. I think you talked about too this kind of same thing occurred at clear result around like energy efficiency models too. That sounds a little bit different, but how did you address that piece there? Rajitha: Yeah. On um, the surface, it seems like insurance and energy efficiency have nothing in common, but I think the common factor there is really your home. You're talking about homes and home information. When I got a clear result, the ask there was to, again, be smart about it, what clear result did, and it used to be Jeff: Mm-hmm. Rajitha: Back when I worked, there , we were implementing energy efficiency programs on behalf of utilities, right? uh, , are managers spend a certain amount of their revenue on. energy efficiency measures, And then you have to show effectiveness. Again. Just how many customers did you reach, but how much energy efficiency did you gain to get that? You wanna go after . The homes that need the most energy efficiency work, that means the ones that are leaking energy. And then you want to also address the folks that are more inclined to [00:14:00] respond to these energy efficiency. Jeff: Yeah. Rajitha: Mailings I'm sure if you're like me, you get these mailings every day saying, Hey, mass save, since they're both in, Jeff: Yeah. Rajitha: we have this Mass Save program, Let us come you and give you a, an assessment of your home. But those solicitations you can send them to everybody, like direct mail. Maybe you get I don't know if you're lucky if you get 0.01% response rate Jeff: Yeah. Rajitha: you want to. rate which then leads to effectiveness is find the homes with the characteristics that if you take the home characteristics and you build a model with all the efficiency built in, you know how much energy they should use. And then , you get the energy used from the utility companies and then you compare the two. And the delta is really how much energy efficiency you can do by. Implementing the measures. So Jeff: Yeah. Rajitha: building science model that we were trying to build. And we had all the knowledge. CSG grew up in the building science, but we didn't know how to scale it. And then when I came on board, I said, Easy, we use CoreLogic. They [00:15:00] can give you the data on every home in any place. We called my contact at CoreLogic had the conversation. And got the data and it then after that building, the building science one for energy efficiency model was easy. And of course the response model as always, you're looking at your past mailings, looking at who responded, Jeff: Mm-hmm. To segment similar folks by that. So that's how it came in easy. So that's why I was like, that is I think the thread through my career is even though they're different industries. There's a lot of learnings you can take from one industry and apply to the other. For the most part I'd say 60, 70% can apply. And then there's the other 30% where I really enjoy 'cause I'm learning a new industry and new things. Completely new things. Yeah. I mean, If it was always just taking the same lesson and using it over again, it's a little boring. But one thing I think I find interesting is we're talking through this, is this idea that you need the data. The data was there, right? Like even without CoreLogic. Which, complete side note, but I [00:16:00] actually sat next to a person from CoreLogic the other day in Dallas. We had a dinner down there and just so ran him aside. One of the team, one of the product team from CoreLogic was literally at the event, and we had a great conversation about multi-family homes. But this threat of you need the day, but the day was there, like people's houses existed. You could look at them. You could go into, you could know what they're made of, what, energy efficient problems they probably were going to have. But it was a matter of how do you take raw information and distill it down into data And, we got better at it by, companies that CoreLogic exist and they parsed it and they gathered it and they correlated it and they normalized it. This is not to get ahead of ourselves, but one of the things I am really loving lately about. AI and how it has advanced is, sure. I don't have Jarvis yet who I can just talk to and is over my shoulder and he can book flights for me and whatever. But data normalization, I can take raw data that we gather ourselves or pull from somewhere that is un normalized and it can just fly through and normalize it really quickly or. For our product at Log Rocket, we can use visual models to actually [00:17:00] watch session replays and understand them and tell people, what are the few things that are going wrong in your digital experience that you really need to address. You don't have to anymore try to spend time watching or hope you find something or. Look for serendipity or basically only use to confirm. We can now watch everything for you and just tell you what's important. 'cause we can understand that. A couple years back you couldn't normalize this data. You couldn't use AI to watch things like that. And just this growth of AI as an assistant and helping us process the data that already existed all around us has been so helpful. Just there. digress. I don't wanna get too far off, but Rajitha: the point you make on and ai. Jeff: yeah. Rajitha: Even if you, few years ago, AI used to be this thing, this big thing, but most folks don't realize it's all around us. It's Jeff: Yeah. Rajitha: in our everyday tasks like. Maps. If you use maps, you've been using ai. If you are using any voice assistants, you've been using ai, Jeff: Yeah. Rajitha: are not new. What changed in the last few years with LLMs is it can take any [00:18:00] kind of inputs. Especially unstructured data, right? Like a Jeff: Mm-hmm. Rajitha: And drawings and freeform text fields and make those useful. That's where Jeff: Mm-hmm. Rajitha: Power is, and it already learned all of that based on everything that's available on the internet. Albert vr, some, copyright issues folks run into and what's been used and Jeff: Yeah. Rajitha: used, but for the most, taking what's publicly available and set the. Jeff: Right. Rajitha: Companies, a lot of companies are trying to do now is taking that and using that to very specific vertical, Jeff: Yeah. Rajitha: Even ZoomInfo that I worked at and that you, this third party data, they're also getting tremendous value from LLMs and generative ai, you know, as an added layer on top of the data that's already there. Jeff: Exactly. , the lead gen piece is an interesting one and just like how marketing has moved forward with that. But just almost every industry there is all this kind of unstructured, messy or somewhat accurate, or somewhat, it's not inaccurate. It's just like hard to [00:19:00] consume. That has been a huge move forward. It's just been the ability to leverage more of this in a way that you don't need a data scientist anymore. Or, I remember being an intern and literally my job was to go through, I mean, This dated myself a little bit more, but hundreds and hundreds of paper records to understand marriage patterns in Worcester County when I worked at a law firm. , but nowadays it's probably all digitized, but you could still feed that data in, into an LLM and just rip it out. What I took weeks to do would be done in minutes, Rajitha: And that's Jeff: which is great. Yeah. Rajitha: Going through those paper records and trying to Jeff: No, Rajitha: right? You would rather just scan those docs, get the insights, and then actually pour through and come up with Jeff: right. Rajitha: on what that was telling you. So I think that's how it's going to elevate our day to day and the work we Jeff: Yeah, Rajitha: than doing those tedious tasks, We are trying to bring into construction, frankly. Jeff: I was gonna say this is not, gonna replace. Humans in that not yet, at least, but it's not gonna replace humans. It's not gonna [00:20:00] steal jobs from people doing interesting things. It's. The tedious things that are just terrible to do. Or at Procore, right? I know there is a ton of work. We had work done here when we first moved in, and there's just a lot of work of measuring and understanding size and how many nails does that mean and how much, what kind of structure is it built from and what do you need to build on top of that? And there's a lot of work that can probably be quickly distilled out by just kinda a scan of the room, gather measurements, what you know, materials do you need, and a lot of these things that took time to calculate. It can probably be done honestly, more accurately and more and more quickly using some kind of AI there. Rajitha: Yes. That, that is the example you just described is a great product. There is a Jeff: Yeah. Rajitha: automated area takeoff. So Is when you look at drawings and, or in 3D drawings, there's a way you can measure the room and then you take off your windows and doors, And then you can calculate, how much. You need how much flooring, how much roof area, right? All of those. And they're one of the first AI products we released at Procore was automated area [00:21:00] takeoff. To do that, you can use computer vision to look at these drawings, extract out the measurements, and then take out, and then you can say, okay, this, and then decompose that into, how much concrete your, how many dry sheets of drywall and so on. So those are exactly the kind of things that AI can help with. And now with LLMs, , you can use it for many more. Jeff: Yeah. Rajitha: Construction is full of unstructured data and I say this every day as I talk to our customers, is this is truly construction's time for Jeff: Yeah. Rajitha: because LLMs in their current form uh, best with unstructured and multimodal which Jeff: Yeah. Rajitha: the sphere of construction is between your. Capturing your videos, your photos, any kind of drone imagery that is automated and all the drawings that are involved. And then the documents back and forth. Jeff: Yeah. Rajitha: submitted your specifications and any. Correspondence between all the different collaborators that work on the project, your [00:22:00] architects, your owners, your general contractors, your There's a lot of content that generates and most systems will have all of these as document repositories, but then you also have to tag them separately. And Jeff: Yeah. Rajitha: As we started to do is more tagging. But now , you can ask questions. You , no longer need to rely on a person who remembers where the folder structure and who knows what document went where, right? You no longer have to rely on a person, you can just ask the system, Jeff: Right. All the information out. It's interesting because I think people looked at this and said oh, these, the trades are gonna be safe from ai. And yes, I think they're gonna be safe from AI in that there's not going to be an AI coming anytime soon that's going to build. The building, right? I think we're a little bit ways off from robots building things and replacing construction jobs. But what they are gonna come do is help you organize it better and help the business side be more organized and help this industry, like you said, huge amounts of paper data. You can PDF it, but how do you query it? How do you understand it? There's always questions. Revisions, drafts, [00:23:00] multiple drafts. There's an engineer who checks the architect plans. But I can't imagine that this was an industry. That immediately dove in. And maybe I'm biasing myself 'cause my father-in-law owns a contracting company and I think he can still barely use his digital camera. But how did that go on that end, right? When we think of vertical AI in these. More kind of hands-on physical verticals that maybe aren't thought of as, charging forward on innovation. How did that go? 'cause you just went, yep, Procore, I'm going all in. I'm gonna go to construction, I'm gonna drive AI forward. So what was this process? How did you get these people on board or get to the point where this was something that they were adopting so heavily? Rajitha: Construction is, I know it gets a bad rap for being very hands-on, but it is hands-on. It is very complex. Like I just mentioned, you have so many collaborators in a typical Jeff: Yeah. Rajitha: if it's a sizeable, think of a building, either an office [00:24:00] building or a airport or a bridge. There's usually like thousands and thousands of Jeff: Yeah. Rajitha: working on the same project, right? So you can, I imagine the complexity of everything involved and the complexity involved in getting the information to the right person at the right time. Jeff: Right. Rajitha: There's a reason why folks are not distracted and jumping on to the latest cool technology. They're very focused on getting the job done on time, under budget. And then without any major incidents, right? Because safety incidents cost you a lot of money Jeff: Yeah. Yes, they do. Rajitha: that's the focus. So the way, aI can help Construction, you're right, is more on the efficiencies. How can it , bring them under budget and on time without as few incidents as possible and as few reworks as possible. Rework is Jeff: Yeah. Rajitha: and you have to go in and drip it all out and do it again. Jeff: This is not software, right? Where maybe you mess up code and you just go rewrite it. This is you. You spent money on [00:25:00] physical inventory. Rajitha: That's right. And it costs time and money. Jeff: Yeah. Rajitha: and wastage. That's where you wanna focus information the first focus is really dissemination of information. How can you make that faster, Point of the revision changes? Can you easily look at what changed from the last version to this? That's a huge use case. Following up on people, did everyone get the right revision? Did they look at it? Do they look at the questions that are coming? There is this process called RFIs. It's called re request for information. And it starts even before the project starts where the owners will get out specifications and say, this is what we want you to build. And the general contractors come in and say, okay, I understand what you want me to, let me go get estimates from my trade folks and my specialty contractors who need to do all of this work. So they go in and in that process, when they're sharing all of these documents, the trades will look at certain parts of the specification. In some cases, the GCs will look at certain specifications, say, oh, when you [00:26:00] set this, did you mean A, B, or C? And They need before they can start to price it out. And give an estimate, right? So that process is called request for information. So , it can come from either the trades or it can come from the GCs and it goes, usually goes to the owners and architects who made the specifications and the designs. That process takes a lot of time, and there's a lot of streamlining that can be done. How you asked the questions, did they answer it right? Is there intent? Can you determine intent from how? It was written to answer the question. If it's ambiguous, then go ahead and request the information and then make sure that the information coming out is the right and it answers the question at the right time. That's just an example of things you can streamline and that's where. You can apply there. There's a tons of these processes. This is Jeff: right. Rajitha: but the end, you are walking, doing a site walkthrough. You're looking at things that are not built right? You call them observations, and then there's a follow up process for it. So there's different ways you do that. But can you [00:27:00] do, just take pictures as you're walking the site? You take the pictures and then the observations can write themselves, the pictures, and then the team that's actually walking through can just. Verify the information and then send it off. So to the kicking off the next workflow in the process. Jeff: I love that the first example was, not even before a bid was accepted, but during request for information, there's, massive ways to save time and efficiency already. So I guess looking at this seems like one of those things where I've heard again and again and something we're trying to practice here is. A lot of these tools don't work great if they're bolted on or if you're making someone change their flow or their work pattern or kind of do something really differently, they're just gonna keep doing the thing they were doing to make it really work. It just has to work. Looking at Procore, was that like a key part of driving kind of the construction industry to move forward is, there's no change management, there's no changes. Just it's going to fit into making these things you already do better, faster, stronger, smarter. Rajitha: That is the ideal scenario, right? [00:28:00] I Jeff: Yeah. Rajitha: say if you can make AI work at a consumer grade level like the maps, if it folks can just use maps without. Jeff: Yeah. Rajitha: Understanding how it works, that's a win and everybody's going to use it. I get this question all the time from my customer. They say, oh, the trades won't use ai. And I said they use smartphones and smartphone is filled with ai. Jeff: Yeah. Rajitha: for sure it's getting that consumer grade. But some in some ways, it's also building the trust. If Jeff: Yeah. Rajitha: some of the tedious tasks and they can see that it can happen. Without them big, handholding the process all the way through. That builds trust. Then they have you build a trust to then leverage it to go up the chain. Serving up more insights, serving up more automations and suggesting more automations. They're more likely to take that Don. To get there. Now let me be clear. I'm not saying Procore by any means, solved all the AI challenges for the industry. We are still way early in our journey. What I can say is they [00:29:00] are very much more open and excited about the potential of AI and. As much as I've been evangelizing and procore's been evangelizing and rest of my partners in the construction industry, all the new startups are evangelizing. I think really the change came from LLMs and Using LLMs in their personal life. How easy it's to use and get answers. Jeff: Yeah. Rajitha: The, some of them bringing that learnings into their work and seeing the value of what it can bring. And actually a funny story two years ago is when Nelms really burst into the Jeff: Yeah. Rajitha: We had our executive cab meeting in February and we were ranking, and we tend to do this. We are very customer driven and we'll bring our. our cab members and say, Hey, here's a few directions we are thinking about. Give us what you think is important to you. Not that we'll take that a hundred percent, but that's one more data point into how we create our strategy. So when we did that, [00:30:00] AI came at the bottom of the stack. I think it was the ninth priority. Fast forward three months, by June, we are having customers say, Hey, remember you guys wanna talk about ai? Let's talk ai. What does Procore think about ai? What are you planning to do? What should we know? What should we do? So it, it's come a long way, but this, it's just we are scratching the surface. It's still Jeff: Yeah. Rajitha: There's so much we can do. Jeff: Like you said, . People didn't realize how deeply you'd been using it already, when you can type in, the kind of rough name of, a restaurant in your map and it just magically takes you to the right restaurant. Or all the kind of voice assistance, there's some level of AI built into those things for them to work. It just now more , at some level , the curtain was pulled back a little bit to say, Hey, there's ai. Then people started to, lose trust then. Gain trust, and this is gonna be one of those kinda yo-yos, I feel like, where it keeps going back and forth until it just becomes a part. And, construction seems to be a great spot for this to really work into it. So going deeper here, right? So they came back [00:31:00] finally a couple months later and said, all right, what are you doing at Procore? How do we use this? What's the path forward? Where should we be using it? There is this idea of AI at the core, AI at the edge, like how do you build things in and make intrinsic versus add feature on top that kind of is additive bolted on, functionality? How have you all thought of that and how have you all driven to, to drive value for your customers through both those means? Or is there one you're focusing on more than the other? Rajitha: When we first started AI at Procore, and this is even before my time, right? We did a lot of experimentation, or the teams did a lot of experimentation and mostly you start behind the scenes Jeff: Yeah. Rajitha: to start is search. How can you use algorithms to make search better? How can you get, understand the intent and what they're doing? How can you. Understanding what they're looking for, show them Content, right? That's where we deployed search. Jeff: Yep. Rajitha: learned from that. And then we, automated Area takeoff was a customer facing product built into our estimation, which is a newer area that Proco was going to back in the day. [00:32:00] Those are great because it's easier to build. It's a new product, new tech stack. Into it, but that's not where most of the usage of the product is coming from. So we found that's great. We can say we did it . But adoption is low, Depends on actually folks , getting the new product, using the new product, finding utility. then we pivoted to saying, let's use, AI in the most used. Products, which is more in the project execution, Setting up location hierarchies in the RFI tool and the submittals and the specifications. So that's where we started to deploy it in the core, that's part of how Procore does business, and You get a lot more adoption because everyone's already using this and you show them a new way. if you show them a way to do it without any new learnings. So when we released our co-pilot, which is, asking questions and getting answers from all of the rich trove of information and the deep. Document repository. We understandably released it on specifications, submittals, and RFIs, our [00:33:00] most used tools, Because that's where the usage is. And that's where you wanna go deep into the attachments and look at Jeff: Yeah. Rajitha: and surface them. And now with our agent strategy, so copilot is one of the strategies, and now we have agent strategy. The agents also be deploying in all of these mostly used tools, because that's where a lot of the pain points are. That's where our customers and users spend a lot of time. And we want to make AI actually a very core to Procore. So we are also trying to go just beyond the AI team building solutions to us enabling tools. All our solution teams Jeff: Yeah. Rajitha: to build their own agents because nobody knows best what our customer needs are. And the , um, problem statements are then our product managers who are very close to our users doing it right. I know enough, but I don't know everything, and I Jeff: Right. Rajitha: those, those detailed workflows that really will make the, all the difference in the adoption. Jeff: And there's so much, if you look at product completion or [00:34:00] product pro or project progress, you can look at from, having an agent kind of viewing this stuff constantly, just understanding hey, maybe no one realizes that a delay on. Some subcontractor of one day, everyone goes, ah, that's a one day, delay. But in reality, the AI is able to look at past experience and projects that have happened before and map out the whole dependency, path and say, this is actually gonna cost us maybe three weeks in delivery in reality because of this one thing. So let's fix that. It's interesting because kinda what you said is let's put it where people are using it and drive the most adoption. We similarly saw the same phenomenon where we built what we call issues, which is basically, the thing that watches sessions and understands, impact and is what's going on affecting the user journey is just kinda like I. A nothing. Is it really a JavaScript error? Is it really a rage click? Is someone really frustrated or are they just, is it a false positive? And being able to understand that we, aggregate it all, we put it into this thing called issues and we saw that, new customers used it [00:35:00] and there was adoption for sure, and people were very positive. But when we put it into. Our core session replay where we knew everyone was and everyone was using it day to day, and really surfaced those insights there to say, Hey, you can see how many times this happens, and it's a lot, and this is really severe and this is important. You should go look at this. 'cause it happened to a lot of people and it really impacted a lot of people. All of a sudden we saw adoption of this feature just shoot up incredibly fast because. We put it into the natural workflow of how people were using these things and we made their natural workflow just more effective versus having to train them to do something different or to look at a process differently. it looks like Procore is doing some really interesting things around AI and really able to kind of move it forward, understand how. These kind of gains in efficiency and gains in understanding can move that industry forward. Love to, just before we sign off, have you been able to distill any kind of meta lessons from this around how to look at AI and building solutions there? Maybe vertical or non vertical industries? [00:36:00] Just that people can start to use the frameworks to, like how to start to look at ways they could start to build this stuff in. Rajitha: Yeah. It's your most egregious problems. First, your most adopted products, but also what problems are you trying to solve in your most used workflows and how can you take, use AI to take friction out of those? And if you can use AI to take the friction out of those processes, it just improves. Your users and customers life tremendously. Jeff: Yeah. Rajitha: you take that big problem statement and then break it into manageable chunks. Like in our things we talk about epics and sprints, right? What can In the smallest increments, but at the same time? enough value that the, a customer, you can put it in a customer's hands that they can find value out of and give you feedback whether it's working, it's going exactly the way you want, or you missed a nuance. To earlier point Jeff: Yeah. Rajitha: about finding those problems and something works or not, it's typically when something doesn't work, it's not always, your strategy is wrong. [00:37:00] It's mostly the way you implement it or someone features working that's wrong. And if you can get an understanding of what that is, I think you can just make almost anything wrong unless you didn't start it with a solid problem to begin with, right? It's all in how you solve the problem rather than should you be solving this problem. Jeff: It goes back to, at some level a framing question of, and this is something I separately, something I've found AI very helpful in is you don't just have to solve the problem. You have to solve it in a way that people understand that it's solved and that it's gonna give them value and. That's actually one thing I found incredibly helpful. Just outside of, any of the products, how anyone could use it going forward to build product is, talking to a ton of users is great, but I found I can download it into a couple programs and they will help me triangulate how this group of people are. Talking about that problem and thinking about that problem and distill down, what is that core essence? So not just can you solve the problem, but you can solve it [00:38:00] in the way that they think about it being a problem and therefore you're not, redoing their workflow or their frame of reference to solve it. You're solving it from their point of view, which is key to these core things. Rajitha: Job shadowing or day in life off is a great way to solve for that. You see Jeff: Yeah, Rajitha: work through that. So you Jeff: I. Rajitha: at the desk with them and looking over their shoulder and noticing how they do it. Jeff: Yeah. Rajitha: we've done that in a few times and it's eye-opening, right? Jeff: Right. Rajitha: This is a problem and we won't do X, Y, and Z. Everyone comes in with our own biases. How we think they do that is very different from how they actually do it. So seeing, being there, putting yourself in the position and tackling the role the way they do it, I Us a lots of insights on what the Jeff: Yeah. Rajitha: solve for. Jeff: So in the end it comes full circle back to you need the right data, you need the right context on the data. You need it normalized in a way you can use it, whether it be going out and, job shadowing or whether it be, how do you gather what millions of houses are [00:39:00] made out of. Awesome. This has been a real blast. I really appreciate you coming on. I would keep going about this ad nauseum probably, but you have stuff to do. Procore needs to keep AIing. So I can't keep you here all day as much as I'd love to, if listeners have questions or thoughts or think they could, be of assistance here for you, what's the best, way to get in touch? Is it LinkedIn I found for a lot of people? Or is there another way that you know is easier for you? Rajitha: I'm on LinkedIn Jeff: Yeah, Rajitha: reach out on LinkedIn. Yeah, find Jeff: LinkedIn is fantastic. It's such a good resource for product people. Yeah, same. Awesome. Thank you so much for coming on. This has been a real blast. I hope we can stay in touch and I'd love to hear more about what Procore is doing as you guys continue to develop there. So thank you for taking the time today. Rajitha: Yeah. Thanks for having me. So excited to be here and having this conversation. love the opportunity. Jeff: Talk to you soon. Thank you. Rajitha: Bye-bye.