This episode provides a development update on LandPricer explaining why the September launch missed deadline—comp search now works reliably but backend pricing calculations require sophisticated weight adjustments before release.
Key Takeaways:
- Comp Search Solved, Pricing Math Hasn’t The AI successfully finds relevant comparables even for funded deals with thorough review processes, but mathematical weighting produces inconsistent results—some properties price accurately within 10-15%, others miss significantly.
- PPA Falls Apart Below Two Acres Price-per-acre metrics become unreliable for smaller parcels (0.1 to 0.2 acres show zero price difference despite 2x PPA variation), requiring raw pricing calculations for sub-two-acre properties.
- Manual Comp Grading Removed User testing revealed comp selection and subsequent grading created redundant work without meaningfully improving pricing—simplified to front-end comp selection only with better user education.
Listen to the full episode for insights on solving national-scale land pricing with noisy multi-platform data.
(Podcast transcript below)
Welcome to Get Serious, wanted to provide you all an update on LandPricer. I know I sent out an email to those of you on the email list, if you haven’t signed up or interested in what we’re doing here and trying to build the most reliable land pricing tool on the market. You can sign up at landpricer.ai. you know, our game plan was again, after multiple delays to…
get the product released by end of September here. Obviously I’m recording this in October, so we missed that deadline. Critically, some of these engineering challenges have just been much trickier to sort out than we anticipated. And it’s not really related to the AI side, like we solved that months ago. Or at least, you know, we’re only kind of minor tweaks are needed here and there.
you know, when it comes to our AI image analysis, the much more difficult part was just getting the comp search to work, consistently and kind of reducing the, you know, length of time it took to both, gather the necessary comps, rank order them, as well as, scrape.
the underlying data to make a lot of our backend calculations. So we seemed to have found a solution for that now after a lot of trial and error. And then there was just a lot of back and forth related to, you know, bugs corresponding to the various price calculations on the backend part of
the app. And so we sorted out most of those and now it’s been more testing. Yeah, we have some UI UX features to sort out here, but it’s really now come back to pricing, right? You know, it’s called BAMP Pricers. You would expect the pricing to be very accurate. So we are continuing to test that routinely and it’s something I wish we could have started testing a lot earlier. It was just
You know, again, hindsight 2020, but it was really hard to separate the signal from the noise when we had so many earlier bug fixes and the CompSearch wasn’t working and everything. So like it was hard to get an accurate sense on how pricing was going to turn out before we solve some of those upstream issues. But now when I’m going through the pricing here, it’s just, it’s not quite.
where I would like and some that’s the thing is some of these, some of the pricing is like very accurate with how some of the mathematical calculations are detecting it and like we’re testing it against properties that have already like sold. So you already have firm data and what the anticipated price should be. And it’s impossible to be like a hundred percent accurate, right? Like, you you’re operating without a site visit, some other factors that can.
come into play such as, you know, Deed to easements or, you know, whether there’s leases on the property or anything like that. Um, but, know, based on aerial info, we should be able to get within a range of, you know, 10 ish, maybe 15 % plus or minus. Um, like that’s a pretty decent range and, something that we would take into account, uh, within, um, you know, our own manual underwriting, like we consider our
our manual underwriting a win if we’re within 10 % plus or minus of our anticipated sales price here. Because yes, sometimes you’re negotiating with a seller or a buyer and so forth and back and forth. I don’t know anybody who’s having just 100 % accuracy and just knowing exactly where prices are going to happen. It’s just not realistic to do that. we’re attempting to get within a banded
variance, and, confidence scale and, how we’re pricing the land. but right now some of the, you know, backend weights are just not, they’re, they’re, they’re going to need more finesse here. And so we’re trying to add in a whole bunch more data, other properties. We already know how they sold. and so then we can just kind of utilize in engineering route to.
uh, kind of train from the data and determine, okay, what’s the, the true mathematical waiting to, um, more accurately price the property. Um, what I will say going through it though, is like some parts we really nailed, um, like the comp search piece is super solid. Uh, like that’s it’s finding comps, even for properties that we funded. Um, and you know, we have extremely thorough.
uh, comp review process and, you know, comps are like the most important part of any, any piece of pricing land. Um, and, uh, uh, that, that is really, really solid feature, um, as it stands here. Uh, but then past that, you know, we’re, so cautious about, okay, how much time is this going to take from the user? And.
just in practice realizing, you know, there was a process of where you kind of have to, you’re presented with up to 20 comps to select from. And you don’t want to choose some that like have nothing to do with the underlying property. Cause sometimes you might just, you know, you’re looking at a residential property and you get a comp that shows up. That’s just commercial, a totally different price point and price per acre. So you don’t want to include that. So you have to do some initial.
diligence, just choosing your comps anyway. And then there was a separate page where you’re going to be grading subjectively the comps that you did select. But in practice, especially going through this, you know, personally, and again, we’re trying to be a primary customer of this application. It was kind of like doubling the work because we were spending so much time just choosing the comps appropriately off the bat.
and then grading the comps after with an even more thorough review. And so it wasn’t dramatically changing the impact on the final pricing. So like, okay, let’s just get rid of that for the time being. just really put kind of, know, the education and highest user error risk is just choosing the comps from the off the bat here. And that’s eventually something that we’re going to want to
just hide behind the app entirely and have AI just be able to select comps by itself. But there’s, a lot of noise within various data across the various listing platforms. So it’s still a pretty challenging engineering problem. But nevertheless, we can do enough to educate the user. like, OK, yeah, just avoid the most glaring.
examples of bad comps to choose from. And so then you can be more, you can get a more reliable price on the backend here. And yeah, kind of choosing the right comps is like the more important part in relation to that final pricing output. So we’re just trying to simplify and remove features that aren’t going to be critical for that.
kind of final pricing output here. And then also just figure out, what do we need to do to adjust?
the anticipated pricing to ensure that we’re maintaining accuracy there. And it’s like, it’s just, hard to do. it’s again, it’s a really, really hard problem. No one has sorted this out properly yet. So like, yeah, I can’t tell you how many times I’m just like sitting staring up the ceiling, staring at the ceiling, just thinking, man, this is just, there’s like almost every single day, there’s like another problem. guess I hadn’t had the foresight to detect
you know, before we had already started building into software form, it just, again, super, super complex problem when you’re trying to do this on a national scale across all different types of markets. And you have to separate again, so much noise, from, from the data. but, you know, that, that, that’s the challenge to do this, to do this properly. So, we’re getting closer every single day here.
And even when I kept thinking, okay, well, we can do some of these manual adjustments, maybe like a blanket 20 % decrease on the averaged comp PPA value. And that will be more realistic to what that true sales value is. But then we went and did a review of our last 18 months worth of.
deals that we had manually done reviews on and we kind of compared, okay, what was the average comp price breakers versus what we thought internally was going to be like our final estimate, as well as what that true sold PPA was. And our internal valuation, as well as the sold PPA was like, you know, pretty spot on. Again, we’re, we’re solid about understanding where things are going to sell most of the time. but the comp
PPA versus what we thought or what the true sold PPA was, it was kind of like all over the place. Like the blended average was like, comps were overestimating by like a little over 30 % of the exit PPA for the most part. But when you looked at, know, if you were measuring the standard deviation, like it was, it was so significant. So you can’t just use like a, a blanket, decrease in PPA just,
you know, from an anticipated sold price perspective, cause like our own data isn’t showing that. you just, have to be more sophisticated with trying to utilize some more of the clues that are, that are within the,
Comp scraping data that we are pulling so I’m confident we’ll be able to figure that out. It’s just You know, we just needed to adjust the math Probably to have a lighter hand on it because like right now it’s just overestimating the PPA on average here plus we have to account for again PPA just it gets destroyed below two acres It’s not an exact metric, but like it’s a pretty accurate
measurement like the lower the acreage goes the more PPA just falls apart because there’s like virtually never a difference in pricing from like a point one acre lot to a point two acre lot, but it would double a PPA. So like it doesn’t so we’re trying to adjust it. So it’s just raw pricing for those lower acreage properties as well. Which can just complicate the math a little bit. Not like crazy for backend, but you know, it’s something that
we do have to adjust to become more accurate there. So just wanted to share a bit of an update with you all on what we’re doing again. Super, super challenging problem. We’re just trying to sort this out the best we can here. like salt, you know, we have to solve the pricing piece because if that is not in line with.
you know, kind of the whole vision of the product. Like it’s kind of unusable regardless of what the other features are. So once we solve that, then it’s like, okay, we’re now going to be much more solid to actually getting this thing launched here outside of, some UI, UX adjustments. So I will keep you in the loop as we continue forward. But hopefully this gives you some more insight on what exactly we’re trying to do here.
and, and solve, and, know, understanding kind of the inherent complexity of all of the variables, associated with properly pricing land. Again, there’s a reason nobody has firmly solved this yet because it’s, super, super challenging, but we believe we’re on the cusp of, at least putting the best foot forward yet. I have like so many other ideas on how this product can be improved and
closer to perfection over the coming years and decades, but you got to start somewhere. With that in mind, SeriousLand.Capital for any of your funding needs, Land Daily Diligence Facebook groups, or cost for your land deals. I just talked about Land Pricer AI. With that in mind, I hope you subscribe and share this with anybody else who you think might be interested in what we have to share here.
Looking forward to next time. Take care everybody. Bye
