Find the few listings worth a wholesaler’s time without reading every listing on the market.
The clientA real-estate wholesaler in Dallas–Fort Worth.
He was reading a couple of hundred listings a day by hand, photos included, to work out valuations.
Hours a day, and buying windows like a price cut or a re-list were being missed in a spreadsheet.
Before
- 1Scroll hundreds of listings
- 2Hunt for words like as-is, estate, must sell
- 3Eyeball photos to guess condition
- 4Work out value by hand
- 5Miss the price cut that happened yesterday
After
- 1Every listing is scored 0 to 100 for seller motivation
- 2Each score comes with its reasons in plain English
- 3Only listings that clear the bar get a photo review
- 4After-repair value and rehab estimate are produced and checked against his buy box
- 5He reviews 20 to 30 a day
The score, live
This is the real scoring function, running in your browser.
- Math
- 0
- Motivation
- 0
- Timing
- 0
The function returned no reasons for this listing.
The ranked list
Strike now, warming, cold.
| Address | Score | Tier |
|---|---|---|
| 4417 Larkmoor Ln | 92 | Strike now |
| 208 Quillfeather Ct | 81 | Strike now |
| 1530 Emberwick Dr | 64 | Warming |
| 77 Hollowpine Way | 52 | Warming |
| 3902 Saltmeadow St | 23 | Cold |
Illustrative data. Names are fictional.
AI, only where it’s subjective
Photos in, after-repair value and rehab line items out.
- Listing clears the bar
- →Vision model reviews up to 8 photos
- →After-repair value + rehab estimate
- →Fits the buy box?
- Listing data (saved searches + public listings API)
- →Deterministic scoring function
- →Vision review for the shortlist
- →Kanban board
Deterministic first, AI second.
A formula for the ranking
Asking a model to rate every listing
Nuance: a model might catch a motivated seller the formula misses
The score is free to run, repeatable, and auditable. That’s a precision-over-recall choice.
- 01Multi-tenant row security before a second user ever touches it
- 02Measure recall: sample low scorers to see what the formula misses
What this means for you
- 01A formula does the ranking, so the score is free to run, repeatable, and auditable.
- 02Every number explains itself. That’s what an auditor, or a bank’s risk team, asks for.
- 03AI is spent only where a formula can’t judge: condition from photos, for the shortlist, once per price.
- 04I learned the customer’s business well enough to encode his buy box. That’s the discovery half of forward deployed work.
→Why not just use AI for the score?
Cost, consistency and auditability. A wholesaler needs to trust why a house is ranked first.
→What does the formula miss?
Context. A strange remark that only reads as desperate to a human. I accepted missing a few.
→Is it multi-user?
Not yet. It’s single-operator, and I’d fix the row security before adding a second.