Cold outreach that gets replies

Cold outreach fails at three distinct points, and teams almost always try to fix the wrong one. It fails in the filter, before anyone sees it. It fails in the first line, before anyone reads it. And it fails at the ask, where a stranger is invited to give up thirty minutes of their week to someone they have no reason to trust. Rewriting the middle paragraph — the part everyone rewrites — fixes none of the three.

Point 1: does it arrive at all

This is unglamorous, mostly mechanical, and worth more than every copywriting change combined. Before sending anything:

1. Authenticate the sending domain — SPF, DKIM and DMARC all configured and passing. Unauthenticated mail is increasingly rejected outright rather than filtered.
2. Send cold outreach from a separate domain, not the one your customers and invoices use. If it goes wrong, it goes wrong somewhere that does not matter.
3. Warm the domain: start at a handful of sends a day and grow over weeks. A new domain sending 500 messages on day one is the textbook spam pattern.
4. Keep volume low per address. Dozens a day, not hundreds.
5. Clean the list. Bounces are the strongest negative signal there is; verify addresses before sending, not after.
6. Include a real, working way to opt out, and honour it immediately. This is a legal requirement in most markets and a reputation signal in all of them.

Formatting matters too, in ways that feel superstitious but are measurable: plain text outperforms heavily designed HTML, tracking pixels and shortened links both hurt, and one link is safer than three. Write it as you would write to a colleague, because that is what a filter is comparing it against.

Point 2: targeting, which is most of the work

The reason mass outreach performs badly is not that the copy is generic — it is that the list is. If a recipient has no current, visible reason to care, no wording will manufacture one.

A workable list is built from a trigger, not from a demographic. Something changed recently that makes your offer relevant now: they launched a store, they started hiring for the role your product replaces, they published something in your area, they moved onto a platform you integrate with, they left a review complaining about the thing you fix. Fifty recipients with a trigger will beat a thousand without, at every stage — deliverability, opens, replies and the quality of the conversations.

Point 3: the message

The subject line

Short, lower case, specific, and never clever. Three to five words that could plausibly be from a colleague: "question about your product pages". Anything resembling marketing — capitals, emoji, "quick win", "unlock" — is filtered by both the software and the human.

The first line

This is the whole email. It must prove, immediately, that you looked at them specifically — not "I loved your website" but a sentence that could not exist without having looked. "Your size guide is on the product pages but not on the category pages, so people filtering by size have to open each item." That is a first line. It also happens to be the value of the email.

Never open with who you are. Nobody has yet been given a reason to care.

The middle

Two or three sentences, maximum. What you noticed, what it costs them, what you do about it — with one concrete detail rather than a claim. If you have a comparable case, one clause is enough: "we did the same for a store with about 400 products; it took a morning."

The ask

The most common mistake in cold email is asking for a meeting. A meeting costs a stranger thirty minutes plus the risk of a sales pitch. Ask for a reply instead: "worth me sending the three pages I mean?" or "is this something you have already looked at?" A reply is cheap, and a conversation that has started can move to a call on its own.

One ask, one link at most, and no attachments. Sixty to ninety words in total. If it does not fit on a phone screen without scrolling, it is too long.

Follow-ups

Most replies come from the second or third message, and most damage comes from the fifth. Two follow-ups, three or four days apart, each shorter than the last and each adding something new — a relevant example, a second observation, a link to the thing you offered. Never send "just bumping this to the top of your inbox", which adds nothing and reads as pressure.

Then stop. A polite close ("I will leave it there — happy to help if it comes up later") preserves the option of contacting them in six months, which a sequence of nine emails does not.

What to measure

• Inbox placement, tested per provider. Everything downstream is meaningless without it.
• Reply rate, positive and negative separately. A negative reply is still a signal the targeting worked and the offer did not.
• Bounce rate. Above a couple of per cent means the list is bad and the sending reputation is being spent.
• Unsubscribes and complaints, which are the metric that decides whether you can send at all next quarter.
• Meetings and revenue, eventually — but not as the first optimisation target, because the leading indicators move first.

Open rate is worth less than it used to be — privacy features inflate it — so treat it as directional at best. The same rule applies across marketing metrics: prefer the number that cannot be faked by the measurement itself.

The honest test

Read your draft as the recipient. Would you reply? If the honest answer is "I would delete this", no template change fixes it — the problem is that you have nothing specific to say to this person, which is a targeting problem wearing a copywriting costume.

Frequently asked questions