Working demo · built in n8n

Deal email in. Underwritten PASS/FAIL out.

Five broker emails went into this workflow. Five underwritten verdicts came out, each with the reason attached. The AI reads the email and explains the result — it never touches a calculation. Every number below came from the run.

2:12, no audio. You are watching the actual workflow run in n8n - every figure on screen is real output, not an illustration.

The pipeline

1 · IntakeDeal emails, one at a time. In production this is a Gmail trigger on deals@.

2 · Extract AIAddress, price, rehab, ARV, rent, taxes, insurance out of messy text. Told to return null rather than estimate.

3 · Validate no AIParses "75k" to 75000, checks required fields, flags ARV below price or rent too high for the price.

4 · Underwrite no AIAll-in cost, loan, cash to close, refi, cash left, NOI, debt service, cash flow, DSCR, CoC, ROI, cap rate, 70% rule.

5 · Buy box no AITen checks read from a Google Sheet. Each returns its own pass/fail line.

6 · Explain AIReceives the finished numbers and writes three sentences. Forbidden to recalculate.

What came out of the run

deal	price	rehab	ARV	rent	all-in	cash left	cash flow	DSCR	verdict
3812 W 130th St, Cleveland	$75,000	$25,000	$145,000	$1,595	$107,681	$1,106	$228/mo	1.31	PASS
2140 Larchmont Rd, Cleveland	$145,000	$55,000	$215,000	$1,850	$213,904	$55,879	-$17/mo	0.98	FAIL
9021 Sycamore Ln, Columbus	$118,000	—	$175,000	—	—	—	—	—	NEEDS DATA
77 Beacon Hill Dr, Pittsburgh	$88,000	$22,000	$155,000	$1,650	$118,644	$4,719	$236/mo	1.30	FAIL
4408 Grove Ave, Cincinnati	$82,000	$28,000	$165,000	$1,795	$118,204	$0	$288/mo	1.34	PASS

Why each one landed where it did. Cincinnati ranked first: all-in is 72% of ARV and the refinance pulls every dollar back out. Cleveland W 130th passed with $1,106 left in. Larchmont failed six checks — the all-in is 99% of ARV. Pittsburgh is the interesting one: the numbers work, but the ZIP is outside the target list and the price is $1,500 over the 70% rule, so it fails with exactly those two lines. Sycamore was never underwritten at all — rent and rehab were missing, so it returns NEEDS DATA instead of a metric built on a guess.

The buy box is configuration, not code

I changed one cell in the sheet — min_dscr from 1.25 to 1.35 — and re-ran. No workflow edit.

deal	DSCR	before	after	reason
3812 W 130th St, Cleveland	1.31	PASS	FAIL	DSCR 1.31 vs min 1.35
4408 Grove Ave, Cincinnati	1.34	PASS	FAIL	DSCR 1.34 vs min 1.35

Purchase price, rehab ceiling, ARV rules, rent-to-price, cash left, CoC, cap rate, LTV, target ZIPs and every financing assumption work the same way.

What this is — and what it isn't yet

This is the screening core: intake, extraction, validation, underwriting, buy box, ranking and explanation, running end to end. It is not yet pulling third-party property data — the deal numbers come from the email, which is exactly how a broker submission arrives. The next two steps are the ones worth researching before building: pointing intake at deals@, and replacing the broker's ARV and rent with an API-sourced figure carrying a confidence and a source. The underwriting engine is plain code and moves anywhere: n8n, a Python service, or a Lambda.