What I’m thinking about: A sense of awe and humility surrounding the capabilities of AI. After years of positioning ourselves as top-tier underwriters in the land industry, our AI models are now correcting my image analysis decisions more often than I care to admit.
We’re pushing hard to launch Land Pricer by late June after scrapping the first iteration earlier this year. The beta feedback was humbling – product wasn’t where it needed to be (too slow/clunky). Back to the drawing board, but this rebuild opened doors I hadn’t considered possible with today’s technology.
The breakthrough came when we shifted from requiring purely manual user inputs to letting large language models handle feature identification from aerial imagery. We’re hitting 95%+ accuracy on our 30+ property analysis questions, even using weaker, cost-efficient models instead of expensive reasoning models.
But here’s what keeps making my jaw drop: The AI is catching things I’m missing.
Last week, for a test case, I was reviewing a heavily wooded property, and looking for sheds/storage buildings. None were visible, so I marked it “no.” The model flagged it as “yes” – spotted something in the northwest corner hidden underneath a dense tree canopy. I zoomed in, studied the image more carefully. There it was: what appeared to be a shed or storage container that I’d completely overlooked.
(Here’s the aerial, so you can see how non-obvious this is.)
This isn’t isolated. Multiple times these past few weeks, the AI’s rationale forced me to re-examine images where I’d made judgments (that I thought were reliable). In several cases, the AI was right and I was wrong.
The reality check is sobering:
- AI is already routinely correcting (and educating) seasoned experts. If models start to outperform professional underwriters using cost-efficient processing, the capabilities gap will only widen.
- Speed of progress is exponential. What we’re seeing today with weaker models will pale compared to ~6-12 months from now when current top-tier (and upcoming) models become affordable at scale.
- Industry disruption timeline is shorter than expected. Roughly a year from now, I’m predicting that AI will be better at online due diligence than ANY real estate investor, land or otherwise.
We’ve built our reputation on world-class due diligence and aerial imagery analysis. Now I’m watching technology begin to surpass capabilities we’ve spent years perfecting. Can’t change the facts. Embrace change, and hold onto your hats, it’s going to be a wild ride.
HUGE CAVEAT: These AI models are only achieving such high fidelity and accurate results based on painstakingly-crafted and detailed prompts that I’ve spent disgusting amounts of hours on building and tweaking. This is also coupled with my CTO adjusting the backend image processing mechanics of the LLM to improve results further (alongside other “techy” optimizations).
Getting these types of results is certainly not simple OR easy (e.g. having AI reliably identify a primary buildable area of ANY land parcel ANYWHERE in the US).
That’s why we’re building the most efficient delivery system for these capabilities through Land Pricer. We’re building this software with Serious Land Capital in mind. I need the most reliable pricing tool on the market, and I won’t settle for anything less.
=====
Ready to access reliable, AI-powered property analysis? Sign up at LandPricer.ai for early access to our planned June launch.
Get the same analysis capabilities that are currently correcting world-class underwriters – available to every land investor.
P.S. Want the technical breakdown of exactly how these AI models are outperforming human analysis? Episode 144 of Get Serious walks through the specific prompts, accuracy rates, and cost structures we’re using. Raw data on the future of property underwriting. No fluff.
