Cold email outbound engine

1. Write one sequence per ICP slice: a specific opener, one proof point, one ask.
2. Keep each email under 90 words and one idea long.
3. Set up the A/B on subject line and CTA.

Offer before copy. Settle the offer first. Most "the copy isn't working" problems are offer problems. A strong cold offer has:

- a specific outcome with a number and a timeframe;

- a believable mechanism;

- the risk on us, not them;

- near-zero effort to accept (a meeting before any value lands *is* effort);

- proof inside the message, not "happy to send a case study";

- them in the first line, not us.

Fix the weakest of these before writing a word.

First-touch rules. No links, images or attachments. Every merge field has a fallback. The opener only works for this recipient. Size the ask to the relationship: cold earns permission to send something, not a call. Each follow-up adds a new reason, never "just following up". Vary phrasing across the sequence, since identical strings at volume are a fingerprint.

Gate it. Score each first touch for pull (0–10) before it loads; under 7 gets rewritten, naming what is missing.

1. Load the sequence into Instantly across warmed inboxes.
2. Hold daily volume under the per-inbox cap so deliverability holds.
3. Watch bounce and spam rates daily; pause a variant that spikes.

Pre-launch check (the hour before go-live).

- The sending domain passes SPF, DKIM and DMARC, and click tracking runs on a custom domain. Any failure is a hold.

- The list is verified; drop anything the verifier is unsure about.

- Write the stop rules down before the first send. Default: pause a mailbox at 2% hard bounces or on any spam complaint, and alert the owner. Tune on your own data.

Warm-up ramp, per inbox. The clock runs on the inbox's age, not the campaign's: a handful of sends a day in week one, roughly doubling each week, at full volume around week four. Get more volume from more inboxes and domains, never by pushing one harder. Keep cold sending off the primary domain you can't afford to lose. Send on working days inside the recipient's working hours, and ramp across the first week rather than front-loading Monday.

First-week watch. Check bounces and replies daily per mailbox. Bounces spiking means the list is bad. Flat replies with healthy placement means the copy or offer missed. If both are healthy, scale up one step.

1. Triage every reply: interested, not now, wrong person, no.
2. Route the interested ones into HubSpot and book the call.
3. Ask a "wrong person" for the right one, it is the cheapest referral there is.
4. escalate anything that needs a founder to answer.

Also classify machine mail. Out-of-office keeps the lead in sequence, with the next touch moved to one working day after a *stated* return date. A hard bounce is suppressed for good. Neither counts as a reply.

Walk warm replies to the calendar. A positive reply is unfinished work, not a win. Every follow-up to an unbooked warm reply carries a new, concrete reason *and* a direct ask. "Let's talk" gets the booking link plus what you'll bring. A no-show is still hot: the same day, a fresh link and no blame. Close every reply with a dated next touch and its hook. Default cadence for a warm, unbooked reply: day 0, 2, 4, 7, then slow to nurture (tune on your data).

1. Compare reply and meeting rate by slice, not by sequence.
2. Retire slices that do not convert; widen the ones that do.
3. Feed the winners into next cycle's list build.

Diagnose before changing anything. Work in order and stop at the first break: delivery (bounces, complaints, placement), then engagement, then replies, then reply *quality* (positive vs negative vs not-now). Fix in this order too: market, offer, copy, list, volume. Quote the sample size with every rate. Change one variable at a time, and give each variant enough volume to read (default: a few hundred sends) before calling it.