The Core Idea
Build an AI-powered system that finds undervalued properties — on-market and off-market — calculates the maximum viable offer for each one automatically, and sends Letters of Interest at scale.
Today, Kip's process is analog: manually scanning listings, evaluating deals sent by wholesalers, and running the numbers by hand on each property. Almost every step of that process is automatable. The system's job is to do the searching, screening, and math continuously, so the humans only look at properties that already clear the bar — and only write formal offers when a seller responds.
Why Current Tools Miss the Deals
MLS and Zillow saved searches are too generic. They filter on location, list price, square footage, and bed/bath count — but they miss the signals that actually indicate a lowball-friendly deal:
Price per square foot
This is the real key driver, not list price. No standard search tool lets you screen on it well, and it has to be judged relative to the neighborhood.
Lot size as a value input
Lot size materially affects property value, but search tools treat it as a checkbox filter instead of factoring it into the valuation.
Seller motivation signals
Phrases buried in listing descriptions — "motivated seller," "as-is," "estate sale," "needs TLC" — go completely undetected by standard searches.
Ownership context
Whether the property is held in a trust or estate, how long it's been owned, and whether it's free-and-clear — the strongest predictors of who accepts a low offer — aren't surfaced at all.
The seller profile that says yes: the classic lowball-friendly deal is an older owner who has moved into a care home, with adult children handling the sale who just want the house gone quickly. Long-time owners and free-and-clear owners have the equity room — and often the speed motivation — to accept below-market offers.
Deal Sourcing: Two Channels
Channel A — On-Market (MLS / Zillow)
A smarter search engine that pulls listings and scores each one on:
- Price per square foot as the primary filter, calibrated per neighborhood
- Lot size, factored into the valuation rather than used as a blunt filter
- Home size and bed/bath count as secondary criteria
- AI parsing of the listing description for motivated-seller language
- Ownership research on every listing: held in a trust or estate, long ownership duration, free-and-clear (no mortgage)
Channel B — Off-Market (Tax-Record Mining)
Direct outreach to owners who haven't listed:
- Pull county tax records filtered by target location, lot size, and home size
- Identify free-and-clear owners
- Identify second-home / absentee owners who rarely visit the property — the Carefree / Cave Creek-style second-home markets are the primary use case here
- Generate and send unsolicited offers or LOIs directly to the owner
This channel is live: browse the scored target list → — 1.4M Maricopa parcels scored on these exact signals, refreshed from the county's free data.
The Valuation Engine
This is where AI does the heavy lifting. The core math of every deal:
ARV (After Repair Value) is what the fully remodeled house would sell for. Example: if the neighborhood ARV is $650/sqft and a remodel runs $100/sqft, the target buy range is roughly $400–450 per square foot.
The nuance problem: ARV is hyper-local. $450/sqft is a great deal in Scottsdale Ranch and a terrible deal in Central Phoenix. Search criteria can't be metro-wide — every threshold has to be neighborhood-specific.
What the AI model does
- Estimates the max ARV price per square foot for each neighborhood, from comps of remodeled sales
- Back-calculates the maximum offer price per neighborhood
- Applies that neighborhood-specific threshold automatically to every property from either sourcing channel
- Flags only the properties that clear the bar
Kip's read: there is likely far more that can be offloaded to AI than either of us has thought of yet — this list is the floor, not the ceiling.
Automated Offer Generation (LOIs)
- LOIs are informal letters of interest, not binding offers — so they can be mass-generated and sent with low legal overhead.
- Precedent exists: Kip knows someone who built software that auto-generates LOIs and blasts them to listings — "we'd be willing to offer $X if your client is interested; let us know and we'll send a formal offer."
- Our version: the system auto-generates an LOI for every property that clears the valuation threshold, with the offer price calculated from the neighborhood ARV model.
- Formal offers only get written when a seller responds — eliminating all the time normally spent drafting offers that go nowhere.
System Architecture
Conceptually, three stages: everything flows in from sourcing, gets scored against neighborhood-specific thresholds, and only qualifying properties trigger action.
Sourcing
- MLS / Zillow feed
- Listing text (AI parse)
- County tax records
- Ownership data
Scoring
- $/sqft vs. neighborhood
- Lot-size adjustment
- ARV model per neighborhood
- Motivation score
Action
- Auto-generate LOI
- Send outreach
- Track responses
- Formal offer on reply
Roles
| Who | Brings |
|---|---|
| Kip | Deal evaluation expertise, wholesaler network, acquisition process knowledge, neighborhood ARV intuition |
| Austin | AI / automation build — search engine, listing parser, ARV model, LOI generator — and technical infrastructure |
Immediate Next Step
Austin shadows Kip for a few days. This is Kip's recommendation and the agreed starting point.