We trained a model on 4,000,000 cold emails. It learned what gets answered.
Three years of sending taught it what gets answered.
$89M+pipeline
$9.2Mclosed by clients
30 minto every reply
Worked with
$89M of pipeline
came out of these.
OPEN ANY CLIENT TO SEE WHO BACKS THEM
What came back
50 receipts.
30 REPLIES · 8 PAID LEADS · 7 EVENT REGISTRATIONS · 5 DECK OPENS
Senders and personal details blurred at source. Open any logo to see exactly what came back, labelled for what it is. Pick your industry to see yours.
Six builds
The engine is six machines.
Each one gets built for you.
- 00
The engine
A model trained on four million sent emails learned the science of cold email. Subject, length, the right ask per persona. It writes with it.
- Every lead researched
- 6+ variables, plus a discard list
- 4,000,000 email corpus
- A person approves
- 01
Deliverability
We build the sending stack rather than rent it. Mailboxes, custom SMTP and our own sending servers, sized to your volume.
- Mailboxes we build and warm
- Custom SMTP
- Our own sending servers
- Reserve capacity, rotated in
- 02
Intent signalling
An engine built around your ICP that watches for the moment a company becomes worth contacting.
- Job changes
- Funding and PR
- Tech stack changes
- Events and filings
- 03
Reply agent
An agent trained on your voice, your product and your customers. Every reply answered within 30 minutes.
- Trained on your writing
- 30 minute answer
- A person approves
- Also runs on your site
- 04
Method
Define the ICPs. Write many variants per ICP. Measure on reply, meeting and close rate. Scale only the one that wins.
- ICPs first, always
- Angles tested against each other
- Reply, meeting, close
- Volume behind the winner
- 05
Automations
What happens the second a reply lands. Interested leads reach your CRM and your team without anyone copying anything.
- Interested leads to CRM
- Slack with the thread
- Winning variant attributed
- No reply into LinkedIn
The machine, in public
Don’t take the model’s word for it. Use it.
Start
Send us the sequence
you are running.
We come back with a read on the message, the personas we would split it into, and the first angles we would test. No call needed for that.
The artifact
This is what lands in your inbox.
A read is a written teardown, not a pitch. The sequence below is a composite we built for this page; yours gets the same treatment, on the same working day.
THE SEQUENCE, AS SENT
SUBJECT: Quick question about {{company}} outbound
Hi {{firstName}} — noticed you just raised your Series A, congrats! Our AI-powered platform leverages cutting-edge machine learning to craft hyper-personalized emails at scale. We’ve helped hundreds of companies 3x their pipeline in 90 days. Would you be open to a quick 15-minute call next week? I promise it will be worth your time.
THE READ
KEEP
“Noticed you just raised” is the only researched fact in the email. The instinct is right. The congratulations is not — every other sender that week opened with it.
CUT
“our AI-powered platform leverages cutting-edge machine learning” The reader builds AI products. To them this is wallpaper, and wallpaper reads as template.
CUT
“would you be open to a quick 15-minute call next week?” The ask costs more than the email earned. First touch earns a reply, not a meeting.
CUT
“I promise it will be worth your time” A promise is not evidence. Cut it and the sentence above it gets stronger.
CUT
“we've helped hundreds of companies 3x their pipeline” Unverifiable claims price the whole email. One named, checkable fact beats three round numbers.
THE SPLIT
This is one email trying to reach two different people. The technical founder who signs cares that the machine is inspectable. The sales lead who evaluates cares whose pipeline it filled. Two personas, two sequences, different first lines — never one email straddling both.
FIRST ANGLES
Angle one: the specific thing that changes after a raise (headcount plan meets an empty pipeline). Angle two: name the tool they already tried and where it stalls. Both run as variants against each other, and reply rate decides — not us.
WHAT WE MEASURE
Every variant reports reply, meeting and close, per persona. Losers get cut with the reason stated. The winner earns volume. You see the same board we do.
— HUGO · SAME WORKING DAY · NO CALL REQUIRED