You can write the best cold email of your career and have it land in a folder nobody opens. Deliverability is not a copywriting topic and it is not interesting, but it is upstream of everything else: if your open rate is thirty percent, no amount of clever first lines will save the campaign, because seventy percent of your audience never saw the first line.
What follows is the infrastructure layer, in the order you should build it, with the numbers that matter. None of it takes more than an afternoon plus two weeks of waiting. The waiting is the part people skip, and skipping it is the most common way to destroy a domain in month one.
Never prospect from the domain that pays your bills
Buy a separate domain for outbound. Not a subdomain of your main site — a genuinely separate registration. The logic is containment: if a sending domain gets burned, you want to throw it away and buy another one for the price of a coffee, not watch your invoices and client threads start landing in spam.
Choose something that plainly resembles your real brand so a prospect who checks does not think they are being phished. If your company is at example.com, register example.co, get-example.com, or example-agency.com. Avoid novelty extensions with poor reputation; a boring country or generic extension is fine. Register for a year, decline every upsell, and confirm the registrar’s verification email or the domain gets suspended a month later. Some operators leave the public registration record visible rather than hidden, on the theory that transparency helps rather than hurts.
Point the new domain at your main site with a redirect. It costs nothing and it means a curious merchant who strips your email address down to the domain lands somewhere legitimate instead of on a parked page.
One mailbox per domain, and the arithmetic that follows
The recommendation that survives contact with reality is one mailbox per domain. Two or three is tolerable. Beyond that you are concentrating risk: three mailboxes at fifty sends a day is a hundred and fifty emails leaving one domain daily, which is a volume profile that looks nothing like a human business.
Use your real first name in the address, and use a real photo on the account. Merchants who are interested will look you up before replying. An address that reads like a person, attached to a face and a phone number, is the cheapest trust signal available. A generic role address attached to nothing is the opposite.
The arithmetic to internalize: one mailbox running at the ceiling handles about twenty-five new prospects a day, because the other half of your daily volume is follow-ups to people you contacted last week. That is roughly five hundred new prospects a month per mailbox. If you need more, you add domains and mailboxes. You never add volume per mailbox. This is the single rule that most distinguishes people whose outbound still works in month six.
SPF, then DKIM, then DMARC — in that order
Three DNS records, and the order matters because the third one validates the first two. SPF first: a text record declaring which servers may send on behalf of your domain. DKIM second: generated inside your mail provider’s admin console, unique per domain, published as a text record. DMARC third: a text record telling receiving servers what to do when the first two fail.
Then verify. Send a test to one of the free mail-testing services and confirm all three pass. Ignore the composite spam score they hand you at this stage — it will be full of complaints about a test message with no history. What you care about is three green checks. If they are not green, wait: DNS propagation can take up to twenty-four or forty-eight hours depending on the registrar, and re-checking twelve times in an hour will not speed it up.
One more record worth adding: a custom tracking subdomain, published as a CNAME pointing at your sending platform. Without it, your open and click tracking runs through a domain shared with every other customer of that platform, including the ones behaving badly. With it, your tracking reputation is your own.
Fourteen days of warmup, and never turn it off
A new domain has no reputation, which mail providers treat as a reason for suspicion rather than neutrality. Warmup fixes this by generating genuine-looking inbound and outbound traffic with replies, gradually establishing that the mailbox belongs to someone who has conversations. Fourteen days is the minimum before your first prospecting send. Two weeks is not a suggestion.
Start warmup around twenty messages a day with a daily increment of one and a reply rate near thirty percent, then step it up in week two. Leave it running permanently after that — it is what absorbs the reputational cost of a campaign that underperforms, and it costs you nothing except an inbox filter. Set that filter on day one: route the warmup traffic straight past your inbox, marked as read, never to spam. Otherwise you will be deleting a hundred meaningless messages a day and you will stop opening your own mailbox.
There are free approximations if you are bootstrapping: put the address on your public profiles with notifications on, subscribe it to a few dozen real newsletters, have colleagues email it and reply. They are worse than a proper warmup service but far better than nothing. Use the fourteen days for the work that has to happen anyway — the filter stack, the offer, the sequence.
The ramp: your first 500 sends, day by day
Week one: twenty emails a day total, of which ten are new prospects and ten are follow-ups. Week two: thirty. Week three: forty. Week four onward: fifty, and stop there. Fifty per mailbox per day is the ceiling that keeps working. Push into the eighties and you are gambling with the account; people do get mailboxes suspended at that level.
Spread the sends across the whole working day rather than firing them in a block. An interval of roughly ten minutes plus a few minutes of randomization puts fifty messages across a seven-thirty to seven o’clock window, which is what a human doing outreach actually looks like. Set the timezone to your recipients’ timezone, not your tool’s default — sending European merchants mail at three in the morning is a reliable way to look automated.
Weekdays only. Weekend outreach reads as automation, gets buried under Monday morning, and buys you nothing. And enable stop-on-reply so that a merchant who answers on Tuesday does not receive your scheduled bump on Wednesday. Nothing kills a warm conversation faster than proof that you were not actually paying attention.
What actually gets you filtered
Links are the biggest lever. Every link in the body and the signature costs you something. Keep the body to one link at most, and preferably zero in the first message. Strip images and logos from your signature entirely — a name in bold, company, one link, a phone number. The phone number is what separates you from an offshore spam operation in the reader’s mind, and it costs nothing to include.
No attachments. If you have something to send, send a link and send it in a later message, not the first. Clean the HTML — most sending platforms have a button for this — because a message that arrives as a hand-typed note performs better than one wrapped in template markup. Skip bullet lists and heavy bolding for the same reason: they read as a newsletter, not as a person writing to another person.
On unsubscribe links there is a real trade-off. A visible unsubscribe link announces to the reader that they are a row in a database. The alternative many operators use is a plain postscript inviting the recipient to say if they would rather not be contacted, which produces a reply rather than a click — and replies are a positive signal, whereas unsubscribes are not. Whichever you choose, honour it instantly and remove the address from every list, not just the current campaign.
The numbers that tell you which thing is broken
Four numbers, and a strict order for reading them. Open rate below fifty percent means you have a delivery or subject problem and nothing else matters yet. Reply rate below five percent means the copy, the target or the offer is wrong. Positive replies — meetings and real interest — under three percent is bad, over five percent is working, over ten percent is very good. Bounce rate must stay under five percent.
Diagnose in that sequence and you will not waste weeks. Low opens: check authentication first, then change the subject line, then change the first line — because the first line shows in the inbox preview and a follow-up that opens with “just checking in on my last email” tells the reader everything they need to know before opening. Only when opens are healthy should you touch the body copy. Then, if replies stay flat with good opens, stop rewriting and go back to the filter stack. Flat replies with strong opens usually means you are talking to the wrong stores, not writing badly.
Run two to four campaigns of around a hundred stores in parallel and change exactly one variable between them. Subject in one test, first line in another, offer in a third. Changing two things at once gives you a result you cannot attribute, which means the test was free of information and you have to run it again.
Bounce, catch-alls and the cost of a dirty list
Bounces are the fastest way to damage a young domain, and they are entirely preventable. Anything above five percent is a problem; two to five is normal territory. If you are seeing fifteen or twenty percent, your source data is bad and no amount of warmup will compensate.
Ecommerce is friendlier here than most B2B verticals, because a store’s contact address is usually published on the site rather than guessed from a pattern. A contactEmail pulled from a store’s own contact page has a far better hit rate than a first-dot-last construction against a corporate domain. That is a structural advantage of selling to merchants — use it, and verify anything you did not get from the source.
On catch-all domains, the pragmatic answer is to include them at low volume. A meaningful share are valid, and at fifty sends a day the downside of a few silent failures is small. At ten thousand sends a day the calculation reverses — but if you are sending ten thousand a day, deliverability is not the thing you should be reading about.