Tracking Zoning Decisions with an LLM
Using AI to monitor city land use and zoning changes
Every month, city councils and planning committees make thousands of land use decisions that impact the built world. From small code changes to spot rezonings to new master plans, what cities allow is constantly shifting. And the vast majority of these decisions fly under the radar of most real estate developers and investors, even when land use changes unlock new opportunities.
While zoning and land use decisions are technically accessible to the general public, they’re often locked in lengthy PDFs or meeting transcripts on hard-to-navigate local government websites. Unlocking, interpreting, and synthesizing this information is a great use case for large language models (LLMs), which can be fine-tuned on large zoning and land use data sets to track and interpret code changes.
Over the past few months, we at Thesis Driven have supported ReZone AI, a project that aims to do exactly this. ReZone is currently tracking zoning and land use changes across 65 cities nationwide and will track over 250 by the end of the year. You can try ReZone for free for 30 days here and use the code “ThesisDriven” to get 25% off.
I’ve asked Daniel Heller, the CEO of ReZone, to share a bit about the technology, his process, and ReZone’s path forward to documenting all land use decisions nationwide.
—Brad Hargreaves
The Problem
"Understanding the implications of changing legislation led me to maximize [...] opportunities" —Sam Zell (Chapter 12, Am I Being Too Subtle?).
Awareness of land use and zoning changes is crucial for real estate operators and investors. Sometimes legislation can aid your strategy, sometimes it can decimate it.
Currently, there are no good, inexpensive ways for real estate operators to track changing land use rules and rezonings—particularly if those operators work across multiple jurisdictions. While large investors retain public affairs or law firms to track local legislative activity, and others regularly attend city meetings or comb through decisions with dozens of pages of legalese, most simply attempt to follow (dwindling) local news coverage and hope for the best.
In short, tracking local decisions that impact real estate is difficult.
City decisions take months and sometimes years to prepare. Developers, planning committees, citizens, and mayors all have different agendas, requirements, and expectations. Everybody adds their two cents between the process of applications, reviews, drafts, re-drafts, and final approvals. Finding relevant information is like finding a needle in a haystack.
For the 65 largest cities in the US, we process an average of 12,652,584 words per week. And that’s new decisions only (excluding referenced documents). It would take an average human ~824 hours of non-stop reading to “process” the same amount of information.
There’s simply too much unstructured data. Of all the proposed applications of AI in real estate, this one is straightforward: a well-trained and calibrated model that can cut through the noise and deliver the most relevant decisions summarized and when they’re needed.
The LLM’s Role
At ReZone, we use a part-human, part-AI-driven approach, with support from LLMs. Any time a new decision is approved, we follow a multi-step analysis process:
The biggest challenge in interpreting new rules lies in the inconsistent zoning codes across different cities. C-3 in Phoenix is not the same as C-3 in Los Angeles. Every city / state requires particular attention to understand the relative importance of specific documents. For example, Charlotte takes their zoning code very seriously, whereas Houston doesn’t even have zoning!
So actually understanding the meaning of a land use change requires analyzing the change with an LLM familiar with each jurisdiction’s base zoning code itself. This is a work in progress today as we focus primarily on adding more cities to our roster and building out ReZone’s library of zoning code changes.
ReZone Today
ReZone today is tracking all zoning and land use changes in the 65 largest cities in the US—including most major jurisdictions in Texas and Arizona, and we expect to cover over 250 cities by the end of the year.
Not all zoning and land use changes are the same. A spot rezoning on a particular parcel, for instance, has far less relevance to the average developer or investor than a major code change or area rezoning. So we place each decision into one of eleven categories:
Spot Rezoning
Area Rezoning
Economic Development Incentives
Land Use Planning
Infrastructure Development
Project Amendments
Final Plat
Annexation
Affordable Housing
City Properties
Zoning Code Modification
While our large language model is constantly improving, we’re still primarily focused on identifying and assessing the relevance of legislative changes, not interpreting them. That is, we help our users identify zoning changes that might be relevant to their businesses; investors should still read through the actual legal sources themselves—which are linked from every ReZone notification—and consult a land use attorney before making any actual investment decisions.
The Path Forward
We’re adding new cities on a daily basis, and we plan to fully track 250 separate jurisdictions by the end of the year.
We’re also looking to help personalize the information we provide—and solve some of the discovery problems given the volume of information—by creating our own AI agent that tailors recommendations to users’ specific investment thesis, buy box, and search criteria, matching each subscriber with the most relevant regulatory decisions.
On the discovery front, we will also allow for parcel-specific zoning insights. For example, as a subscriber, you will be able to set a two-mile radius around existing properties and get all decisions that might impact your properties right into your inbox.
Want to try out ReZone? Start your free 30-day trial here, and use the code ‘ThesisDriven’ for a 25% discount.
—Daniel Heller
Daniel can be reached at daniel@re-zone.ai
 | A guest post by
|
Brilliant application of AI - The model reminds me of companies like Quorum, LexisNexis, etc!