← All work Productized system · Cold outbound · 2026

Cold email that reads like one person wrote it.

A white-label outbound machine we built and run ourselves. It scrapes and cleans the list, researches every company, writes one true line per lead, sends from warmed inboxes and sorts the replies.

TypeProductized system, white-label
We didInfra, data, copy engine, ops
TimelineApr to Aug 2026
StackPython · Claude Haiku · Instantly
2,953unique leads processed
4niches, each with its own offer pack
7pipeline stages from raw CSV to send file
750emails sent from one warmed inbox
The short version

Fix the plumbing before the words.

The whole idea in 60 seconds, one product decision at a time.

01Pipes before copy

Nothing gets written until the inbox is trusted.

Sending runs on separate domains, never the main one. Every mailbox is authenticated, warmed for weeks and capped per day. Shown: the 10-mailbox fleet we're rolling out next.

SPF · DKIM · DMARC2 inboxes per domainWarm-up to 80+20 a day capTracking offPlain text
02The list gate

Guessed emails bounce. So we built a gate.

Free contact scrapers guess first.last@ addresses. Our first lists bounced up to 68%. Now every list has to pass a free triage, then a paid verifier, before it can load.

03One true line

The first line is about them, not us.

Claude reads each company's homepage, finds one tension, and writes an observation plus a matching bridge. A free QA pass rejects anything that sounds like a bot.

Offer packs per nicheObservation + bridge pairsTier: specific or fallbackEvery lead gets a line
04The handoff

Every reply gets a label and a next step.

All mailboxes feed one inbox. Each reply is sorted into one of five labels, and a human approves every answer before it goes. Automatic labelling is the next build.

InterestedObjectionTimingWrong personNot interested
05Under the hood

Seven stages, one send file.

Every stage that touches the network or a model saves progress every 10 leads, so a crash never costs a rerun.

1Clean

Drop bad rows, fix names, strip Inc. and Ltd., remove duplicate emails.

2Enrich

Industry label, campaign tag and a casual city name.

3Scrape

Each homepage read in parallel: title, meta, headline, first paragraphs.

4Summarise

Claude Haiku returns what they do, who for, category and one signal.

5Personalise + assemble

Line and bridge per lead, QA, then a 19-column file ready for the sender.

06Honest numbers

What it has actually done.

Counted from the pipeline's own files and the sender's logs. No reply-rate theatre.

Research + writing speed

100 leads researched and written in 5 minutes 36 seconds.

One logged run: 85 homepages read, 82 summarised, 77 specific lines and 23 fallbacks. Model cost for the batch: $0.065.

85 scraped82 summarised77 specific lines
Niches

4 lanes, 5 offer packs.

Canadian founders, franchise operators, restaurant groups, and local clinics plus home services.

founder_saasunitos_narank_radar_narank_radar_indialocal_ops_na
List gate

2,385 addresses triaged for free.

2,037 cleared for verification, 335 quarantined as risky and 13 thrown out.

SendRiskyDead
Sent so far

750 emails from one warmed inbox, capped at 35 to 50 a day.

A three-step sequence, plain text, tracking off. The next fleet is 10 mailboxes across five domains.

Day 0Personal line + bridgeDay 4Short bumpDay 9Last note
"Volume is easy. A clean list and one true sentence per lead are the whole game."
The rule the system is built around.

Want this machine pointed at your market?

We set up the domains, the list gate, the copy engine and the reply inbox, then run it with you. Start with a free 20-minute call.