In this episode, the SLC Bot build gets a full behind-the-scenes breakdown, a custom GPT designed to encode Serious Land Capital’s underwriting expertise, deal history across thousands of reviewed transactions, and Chris’s voice into a deployable AI tool. The project is modeled on Callan Faulkner’s CalChat framework and is being built entirely through Claude Cowork, with phase one (knowledge processing) completed a full week ahead of schedule in a single day. The bot is targeting 85-90% accuracy before external deployment, with a paid wrapper already selected and beta testers lined up.
Key Takeaways:
- Your Data Moat Is the Actual Product The chatbot’s quality ceiling is set entirely by the underlying data, and years of podcasts, deal notes, and internal documentation are what make this buildable in weeks rather than months.
- Cowork Outperformed the Template Cowork identified a 15% efficiency gap in CalChat’s knowledge base file format and corrected it before the build even started, improving on a system that took Callan’s team a year to stress-test.
- Phase One Collapsed Into One Day By feeding Cowork clear criteria, it filtered 300 podcast episodes down to 48 relevant ones autonomously, compressing over a week of anticipated manual work into a single session.
- Voice Calibration Is the Hard Part The testing rubric includes 15 questions across topic categories, a grading framework, and a Notion doc for logging outputs, because the bot must reflect nuanced underwriting judgment, not generic AI responses.
- Overhead Down, Output Up The SLC Bot is being built while cutting costs across the board, which makes it a direct proof-of-concept that AI-driven productivity can move inverse to headcount and burn.
Tune in to the full episode for the complete four-phase build breakdown and to learn how to apply the same framework to your own business.
(Podcast transcript below)
Welcome to Get Serious where at Serious Land Capital, have invested in over $6 million of vacant land deals with industry leading 41 % operating margins. So today I wanted to give a behind the scenes look as we are rapidly building out this, uh, serious land capital chat bot, um, or, know, in short title that SLC bot, um, which is, uh, basically
compiling all of the content we’ve ever put together, plus a ton of our internal documentation and learnings and really every single deal that we have reviewed over the course of us being funders in this space, representing thousands of deals. All of this is anonymized, of course, and effectively building out
at its base case, an internal tool so that, you know, myself, really anybody on the team, has access to all of our learnings and knowledge, and procedural, strategies and techniques that we follow as a company, primarily for the sake of underwriting deals. but also as it relates to just general, you know, market sentiment and analysis,
how we might approach our operations as well too. Those are a bit more secondary and we’ll probably build those in even more going forward here. But primarily focused on the core underwriting piece since that’s what most of our data relates to. And this was all based off of the
you know, successful CalChat plus custom GPT that Callum Faulkner and her on common business team have put together really over the past year, took a lot of iteration to get it right. But basically they had compiled, you know, all of, you know, Callum and her teams, you know, AI business building strategies.
and tactics and put that into a custom GPT really pay walled for folks who are part of her monthly mastermind group. But they really set the precedent on what you can do with current AI tools in relation to feeding in all the correct data and structuring it in such a way that it can be easily readable and
accurately displayed to an end user regardless of what they might be asking the chat pod. this has been on our docket for probably a few months, but we had a lot of other higher priority focuses to nail as a company prior to that, not the least of which was just
Organizing our core data because I keep repeating this, but you know, if you don’t structure and organize your underlying company data or personal data, whatever you’re trying to do with AI, like all the outputs and even the inputs are going to, to fall apart without that. So that really has to be step one. so that, know, what the update process is, where your AI is pulling certain information from, because that’s going to everything smoother and everything work.
much, much better when you actually try to build something out like this. And it’s been such an interesting exercise to go forward with this. Obviously we’re using Claude Cowork to build this out. Again, on a daily basis keeps improving and it’s just an unbelievable tool. You’ve heard me talk about this a number of times here. No surprise that we’re continuing to push it to its limits on a daily basis.
And you know how this differs from what we were trying to do with land Pricer really over the past, you know, two plus years. First, the technology was just not even close to, you know, what was feasible. again, I’m still astonished by, you know, how much money I had invested to now inferior techniques for getting land Pricer off the ground. And, you know, inherently that business and application is a, uh,
more ambitious all-around project because it has to take into account, you know, live comp searching and determinations off of, numbers related to the market and, you know, aerial map reviews and so forth. like there’s a lot more inputs that require, you know, greater technical expertise that is outside of what cowork can do at the moment. But,
a kind of core component of landpricer that was also related to the offering was just like, okay, how do we translate over serious land capitals expertise and underwriting land deals to get the best price possible when somebody is trying to use the landpricer product? So we’re taking a chunk of what was going to be delivered there.
but just restricting it to a chat bot form, both because, again, the technological improvements just makes this way easier to build than in the past. And it’s just by nature, again, simpler in terms of its kind of final vision that we want to do and something that’s directly going to service our company.
that we can handle internally. doesn’t need to be a separate company like LandPricer did. And something that we can also deliver to the land investing community, similar to how Kellan and her team had built out the CalChat format. And so, as we were prepping to build this again, at its core, if we were gonna try to build this SLC bot that like,
much any question you could ask within the land investing space, how to get it to answer in an accurate fashion as if it was my voice, coming through to introduce the appropriate nuance and
you know, understand where to place guard rails or, you know, where to deliver more versus less info and so forth. like that’s very difficult in practice, right? think, you any of you can appreciate that, regardless of how often, or how it in depth, you’ve, you know, used AI thus far. And, you know, so.
And this is like maybe the greatest piece of this AI explosion in terms of capabilities to enhance productivity and so forth is just, again, how much it lowers the inertia factor for procrastination, which, yeah, I’ve never met an entrepreneur who didn’t struggle with that, at least sometimes.
throughout their, their careers, cause like by nature, the thing that we’re, should be working on the most, like usually we put off the most is cause it’s the hairiest problem and it’s going to move, move the most things forward. but with tools like cowork now, like you can just lower that inertia threshold so much lower.
because it’s just so much easier to at least start getting the initial steps done. And then you get the ball rolling here and just rolls and rolls and much faster than you ever might’ve anticipated. So it’s hard to fully convey that until you’ve experienced the magic of it yourself here. But like, again, our productivity and what we’ve been able to build over the past several weeks now, it’s just unlike anything in our entire.
you know, careers as land investors. And again, this is in the midst of, you know, chopping overhead all across the board and, um, you know, stretching every dollar here. So, you know, making cuts while productivity goes up, um, is, uh, you know, hard to do in, in practice. Um, but that is exactly what, um, what we’ve been able to do get given these technological investments. And so, you know, as I set up this cowork task,
Um, uh, again, I don’t want to. Understate the, um, difficulty in doing this well. Um, and you know, most critically is that we have so much underlying data to work with. Like we have hundreds of podcast episodes. so many newsletters, thousands and thousands of words written. have just.
enormous amounts of email correspondence going over due diligence and in-depth CRM, thousands of deals that we have notes upon, plus just additional internal documentation on our thought process, on how we approach things, how my voice is conveyed.
that we’ve utilized for other AI projects and so forth. Like we have so much that allows the build out for a tool like this to be an easier lift than if we were starting from scratch. Like if you’re trying to start from scratch here, yeah, like building the actual bot itself, like the custom GBT, like that’s no problem. But again, if you don’t have the underlying data and you have to build it all, it could take ages to properly do that, especially if you want to charge.
money for it too, which is our goal with the SLC bot. It’s not going to be a significant expense. We’ll have it serve more as a lead magnet for our core funding business. But because of how much info and know-how goes into it and a lot of our internal techniques that we’ve bled over and internal data, ultimately that’s
Yeah, something that, that, you know, we believe earns its own value proposition. So, you know, when I was starting out this process, like it was first telling coworker, okay, where is all that data located and, how, know, we initially, you know, envision structuring this SLC bot, which I kind of went over earlier in the podcast and our overall strategy and so forth. And that that’s really a route that Callan and her team had encouraged us to use like.
Here’s a certain prompt that they had basically reversed engineered from their CalChat to generalize an approach that any other business could use if they wanted to build out a similar type of CalChat-esque application, which is exactly what we’re doing with SLC chat.
Again given how much data that we had in some of the initial info that I input into co-work like it was able to fill out most of that on its own and then you I went over with a fine-tooth comb like you know, This voice doesn’t sound appropriate here. No, I wouldn’t use phrasing like this Or you know the approach here just doesn’t quite make sense like I get it like as I’m saying this sounds more general, but
Ultimately, again, cowork is so good at producing, um, you know, initial iterations to, okay, here’s the, like, anticipated project. You gave us a bunch of info, like, let’s try to break these down into next steps here and take a look at this. Are we on the right track? And then you, just review like a manager and like go back and forth, back and forth. And when I felt more comfortable with, um, you know, the approach.
that we were utilizing, who we were targeting, which data sources were more primary versus secondary and so forth, and making sure that cowork had access to all of them. Then it was time to, okay, let’s start building this out with this vision in mind. And because we have cowork attached to Notion, which again, I would say Notion is kind of the gold standard for information management.
Uh, especially visually, um, it’s almost hard to describe again, what, like all the capabilities of, of notion. can be a meeting, summarizer, a CRM, it can be a task manager, um, a journal, uh, you know, just all around note compilation tool. Plus it has AI in, in, in the back background and it can attach or, uh, connect to cowork, um, pretty seamlessly.
super, super powerful tool here. And, you know, the task management and the CRM side is like, you know, kind of primary to what we utilize it for. And so I had asked cowork, okay, let’s build out this process for building out this SLC bot and have it added directly into our home task manager and break out, you know, the subtasks or the subtest of the subtasks and assign dates to them based on like realistic.
delivery dates to move forward with now that we have an anticipated plan for what we want to do. And like also, okay, what’s the first iteration going to look like? When do we bring in some initial testers? If we wanted to turn this into a, you know, paywall product outside of our internal team, here’s a tool that we would choose and the approach vector that we would utilize for that.
and it, it just handled this brilliantly. and I, I don’t know if like some of you listening might get as much of a thrill, but like on a task manager for just like, you know, even for small tasks, like just checking, checking things off. Okay. We’re in progress there. We’re done there. like there’s always a, like a nice little dopamine hit. Maybe I’m the only one, but I think a lot of people can, can relate to that. I just like, okay, yeah, something’s accomplished. Something’s moving forward here.
So like we had spent all that initial time just to set the groundwork for it. And we had, you know, this SLC bot divided up into four phases. and I’ll just read them off here in front of me, like this phase one was just the knowledge processing. So it’s like, okay, how do we boil down all of the knowledge that we have into the correct knowledge base? That is the most concise and accurately formatted for a custom GPT or an LLM to be able to read.
so like this gets really technical quickly. but it’s absolutely critical for the function of the underlying application. like, th this is probably the stuff that not enough people appreciate in, order to how to structure your data so that the AIs can read it, and pull up, information quickly and accurately for, know, demonstration to the end user. And so very fortunate.
Again, we had some underlying knowledge-based documentation from CalChat and how they had spent a year testing different formats to make sure that their custom GPT was working appropriately. And so we were able to then link that into our cowork and say, okay, we know this format is stress tested and it works. How do we translate our data into a format like this?
Um, and interestingly, cowork, even as it was going over this was actually making suggestions on top of what Callan’s team had already done. They’re like, Hey, they’re, they’re probably leaving like 15 % of bandwidth on the table here because of the, uh, file type that they’re utilizing for the knowledge base. And so we were able to fix that and now make our system even more efficient. Um, because again, these.
models are just able to pull from, you know, community reviews or where people have struggled with certain knowledge base. they understand how to do all of this. So then we can just make it much more efficient and build out the exact files that are needed for the core brain, if you will, of the, custom GPT, which again, if I had to do all of this from scratch, like if we didn’t spend.
well over a year now, just producing content that we’re kind of like for so long. And sometimes it feels like you’re banging your head on the wall. You exceed like lower views and all that, but it’s like, okay, all of this was also serving another purpose to discontinue to add to our knowledge base of what we have learned. And so that we can utilize this data in other, you know, potentially even more efficient or value add ways.
So we’re just repurposing things that we’ve already done before that have taken hours and hours, hundreds of hours, if not thousands, to reduce at this point, which is really the moat that we had built. Because again, if we were starting from nothing, can’t overstate that enough. It would have been so much harder to do this. Whereas like cowork, it’s labeling these certain days for, okay, we can get this.
task done by this day and like what we’ll basically knock out a sub task every single day roughly. And maybe we’ll finish up this project within a month or so and get it off live and you know, more, more thoroughly tested on the market. But because we had so much of the underlying data and cowork and just structure the data into the files that it needs because it knows the end goal. It basically didn’t need me to review.
virtually anything, it could do its own quality assurance along the way. And so like the initial phase one of this knowledge processing didn’t need me to determine any of this or like, you know, out of the 300 or so podcasts that we’ve done so far is like, okay, that’s a lot to parse through. Can we sort out maybe the top 50 that might be the most relevant for this particular, you know, initial version of the SLC bot. And so I gave it some guidelines on which podcasts
Or even topics that I thought would be better. And then it was able to search all 300 and determine, okay, yeah, we got down to like 48 or so that are going to fit the certain criteria I want to pull from, off the bat without me needing to go back in and select all the different podcasts or see what, what I talked about. again, it is such a remarkable tool when it comes to data processing, assuming you have that underlying data. and you’re avoiding all of that manual.
nonsense of having to go through and, uh, know, select thing, you know, that’s the drudgery of that. And I’ve done that in the past and it’s just, sucks. Um, no one wants to do that, but now these AI tools just allow you to just boom, automate it. And so I was able to get like, you know, well over a week ahead of the anticipated milestone schedule all within one day, because it’s like, okay, boom, we knocked out one. You get that little check mark, you get that dopamine hit and co-works. Okay. You want to move on to the next step?
It’s like, well, why not? Yeah. You, you just push as much as you can for stuff that I don’t need to actively review here. and then we can take a break for, you know, the, the end of day. So then I can come back fresh for the stuff that I’m actually going to need to iterate back and forth. and so we finished up that whole phase one, all the subtasks, and now we’re on this phase two, which is the custom instructions and the GPT build. So, and this largely was pretty simple, like cowork just produced, Hey,
Here are the exact files that you’re gonna upload to the custom GPT. This is exactly what you need to put in the custom instructions, do all of that, and then we’re gonna start testing. And we need to do this voice calibration. So, hey, here’s a list of 15 questions across all of these different topics that we can utilize to test out. the content appropriate? Is the format appropriate? Is the voice appropriate? And here’s a grading rubric to utilize. Here’s an area for you to input.
What the custom GPT is giving you and then your notes on how it can be improved. Are the guard rails in place? Is it giving away confidential information? Um, what, what are certain tags for what needs to be improved? And all of this was produced into another notion document that was super easy for me just to, you know, it’s, it’s in front of me right now on my other screen where I can just look at it and see, okay, we’re going to test out this question. I’m going to input here. So easy to utilize. And then, you know, I can just piecemeal go back to coworking like, okay,
Let’s address the issues for this one. Then we can retest. So that this is where it starts to get more difficult and grinding to deal with. But this is really the secret sauce, right? Because like similar to when we were doing land prices, like I was not going to release this publicly unless I was absolutely certain about the quality. so like similar for this SLC bot, need it to be like, we have built a reputation around nuance.
accurate underwriting and approaches to land investing. So if we are not getting that conveyed through this, you know, custom GPT solution, like, then we need to continue to refine it. But again, we reduce the inertia so much because the refining process is so much more simple because then like
I’m not the one who has to go update all the knowledge base files. Like I submit, you know, I have my whisper flow on, I just say, yeah, this is not really working. Here’s why, why, why, why, why throw it back to cowork. It takes all that feedback listed out. Okay. Yeah. We’re going to fix this file, this file here, make these adjustments. So this error won’t come back out. Okay. Let’s retest to see if it fixed it. So then you’re not stuck doing any of that manual, work on, on the backend.
so it’s again, much more, managerial to, to actually do this, way easier to just stick with the program and just make tons of product. Like if I was trying to do this again, like a year ago, I just can’t get over how much easier this technology is, is doing, it’s making it, to build out a chat bot like this. Like again, even a year ago in the pet, like this would have cost tens of thousands of dollars to produce this.
yeah, you gotta have some engineering help and some design and so forth. Now, cowork can handle all of that. It just needs the underlying data, which I had already produced, within, the company structure. And so like that’s phase two, that’s where we’re at right now. But like, we’re way ahead of schedule, with, with, with doing that, I can just knock out a few voice questions day by day. Refine, refine, fine. Okay. Once it’s internally good enough from my end, bring in my team member.
Um, and then we have some beta testers already lined up who wanted to help out with this. If you already hear this, podcast, feel free to reach out to me directly and say, want to test this, um, as well. And then we’re just going to daily iterate, make sure, okay, we’re getting all these tests in. What do we need to refine further? Um, and then once we’re pretty settled there, the knowledge base is solid. It’s answering questions, you know, right. 85 to 90 % of the time, roughly, then we can do external deployment.
We already have the paid solution, the wrapper to add all of this directly into like, it’s not difficult. Like that technology already exists. We already have our providers set up for that. And then we can just release it out into the wild. Like again, this has turned into like a potential several week, multi-month process to really nail down to all of a sudden, like you can build one of these apps and you know, a couple of weeks if you’re really, really pushing on it.
and refining like crazy, which we are. hopefully this gives you an idea on, you know, both what we’re doing internally, the advantages that we do have, but also a level playing field in terms of like building something similar. If you wanted to pursue a direction like this and just understanding how powerful a tool like cowork is,
to really take your business to the next level and code your way to basically anything you want. So if all of that is interesting, again, reach out if you want to test out this bot. And again, I referenced Kalon a number of times. She was the inspiration for why we even built this in the first place. And the current bot would not work nearly as good as it is right now without having her
her team system in place. again, I’ve already been testing this custom GPT and AI. I’m super perfectionist when it comes to having my voice translated over from an AI perspective. And like right away, there were already some answers like, really don’t need to change anything. And this is really early on in the testing process. Like this is getting better by the day, but within like the first three or four uses it was like, yeah, there’s like
maybe one sentence that needs to be slightly adjusted here. It is really nailing what I was hoping for. So the iteration process, I don’t think it’s going to be nearly as painful as I might’ve been anticipating prior to getting this ball rolling. So if you want to sign up for Callan’s Bootcamp, by the way, the effortless business bootcamp, see the information below in the show notes. It starts April 21st. So,
you know, running out of time if you haven’t signed up already. And by the way, if you sign up with my link, you get a free session with me going over Claude, cowork direct applications for your business. Certain techniques that I can just give you the exact copy paste skills. You can be off and running within five minutes. I’m doing some of these, you know, absolutely magical techniques for, your own business there. So please sign up.
If you’re at all interested in, uh, you know, taking your AI game to the next level, these are best from the best, um, the best of the best, uh, that, that are, uh, you know, progressing in that domain right now. Seriously and dot capital. If you’re looking for any deal funding, 50 K purchase price and up, uh, looking forward to next time to subscribe and share everybody take care now.
