DealScanner
deal sourcing

How it works

DealScanner turns 236 scattered broker websites into one short morning briefing of businesses worth buying — filtered to your investment thesis. Here is the whole pipeline, end to end.

1
BrokersA curated list of 236 business-for-sale brokers is the raw supply. You manage it on the Brokers page — add, archive, or fix a source.
2
Daily scrapeEvery morning the engine visits each live broker, reads the new listings, and stores them once (deduped by URL) with their full text and financials.
3
Thesis filterYour thesis — keywords, size band, geography, exclusions — decides what qualifies. A cheap global filter drops the obvious misses (salons, restaurants) before any AI runs.
4
AI scoringWhere wording is fuzzy, a cheap model reads the listing and judges fit, with a reason. Answers are cached, so the same listing is never paid for twice.
5
Morning briefingYou get one email: the new deals that cleared your thesis in the last 24 hours, linked straight to the board. That email is also the engine's heartbeat.
6
You actOpen the Deals board, skim the qualifying deals, and shortlist the ones worth a call. Everything older is a keyword away on Search.

From noise to a shortlist

Most listings are not for you. Each stage narrows the pile so you only ever read the deals that fit — the rest stay searchable but out of your way.

236 brokers scanned
Thousands of listings stored
Global filter drops obvious misses
Your thesis: keywords · size · geography
A handful qualify each day

Coming soon: your votes sharpen it

Today a deal either clears the thesis or it doesn't — a binary keyword-and-size gate. Every yes / maybe / no you cast is saved with the deal's full context. That becomes the training set for an instinct model that learns the judgment calls behind your votes — ranking deals the way you would, not just matching words.

instinct model (learns)binary gate (fixed)votes collected →match quality →

Two principles that keep it cheap and honest

The database is the brain. Every listing is stored once, thesis-neutrally. Your settings are a lens applied when you read — so changing a keyword re-ranks the whole board instantly and for free, with no re-scrape.

Spend is capped and visible. Only the cheapest model is used, every AI call is metered against a daily cap, and the Spend page shows exactly what each day cost versus what it found.