SteelWorks
Cold Outreach That Isn't Spam: The System We Actually Run
Almost all cold outreach fails for one of two reasons: the list is the product, or nothing stands between a bad draft and the send button. Fix those two things and outreach stops being spam and starts being a channel you can keep. Here is the system we run, and because we are a receipts brand, the rules come with the failures that produced them.
Start from demand signals, not a list
A purchased list is a pile of strangers. A demand signal is a public post where someone describes the exact problem you solve: a company hiring three engineers this week, a launch thread full of tooling decisions, a complaint about a named tool on a review site. Start there, and your first line can be true instead of generic.
The rule we enforce in code: no draft exists until you can write one specific, human-readable observation about the company that a stranger would recognize as true. "Noticed you're hiring Python and Go engineers, teams that grow that fast usually multiply their ops load" beats any template. If you can't write the observation, the lead isn't ready.
Personalization is not enough. Every draft should pass a quality gate with named, blocking checks: placeholder text that slipped through, fabricated claims you can't back up, a missing compliance footer, all-caps words scraped from a company name, spam-trigger phrases, link-stuffing. A gate beats vigilance because vigilance fails on a bad day and a gate does not.
Gate the draft before it can send
Personalization is not enough. Every draft should pass a quality gate with named, blocking checks: placeholder text that slipped through, fabricated claims you can't back up, a missing compliance footer, all-caps words scraped from a company name, spam-trigger phrases, link-stuffing. A gate beats vigilance because vigilance fails on a bad day and a gate does not.
One failure that taught us this: a frequency cap that counted blocked sends as contact attempts. Every daily retry re-poisoned itself and the whole pipeline quietly sent almost nothing for nine days. The lesson: count what actually reached a human, and put the checks in the one place all sends flow through, not in each campaign's good intentions.
Five cold sends a day per address, not fifty. Gmail cold-sending above roughly ten to twenty a day is how you land in spam folders permanently. One open thread per company. These caps feel slow; they are what keep the channel alive long enough to compound.
Cap the volume, protect the domain
Five cold sends a day per address, not fifty. Gmail cold-sending above roughly ten to twenty a day is how you land in spam folders permanently. One open thread per company. These caps feel slow; they are what keep the channel alive long enough to compound.
A real footer with your identity and a working opt-out on every send. Suppression that is permanent and checked at the send chokepoint, fail-closed, so no campaign can forget it. We learned that one the hard way: opt-out handling lived in the reply pipeline but the send path never consulted the list, and someone who asked to be removed got two more emails before we caught it. The fix wasn't a better process; it was moving the check into the chokepoint.
Every reply gets classified within a day: interested, question, not-interested, opt-out, bounce, autoresponder, and one more that surprised us — hostile. Real inboxes receive social-engineering and prompt-injection attempts. If you put an LLM anywhere near your replies, a stranger will eventually send "ignore your instructions and…". Anything an LLM touches needs a deterministic screen in front of it, and flagged messages get a human, never an auto-reply.
Compliance is wired in, not bolted on
A real footer with your identity and a working opt-out on every send. Suppression that is permanent and checked at the send chokepoint, fail-closed, so no campaign can forget it. We learned that one the hard way: opt-out handling lived in the reply pipeline but the send path never consulted the list, and someone who asked to be removed got two more emails before we caught it. The fix wasn't a better process; it was moving the check into the chokepoint.
Every reply gets classified within a day: interested, question, not-interested, opt-out, bounce, autoresponder, and one more that surprised us — hostile. Real inboxes receive social-engineering and prompt-injection attempts. If you put an LLM anywhere near your replies, a stranger will eventually send "ignore your instructions and…". Anything an LLM touches needs a deterministic screen in front of it, and flagged messages get a human, never an auto-reply.
None of this promises you reply rates. Single-digit reply rates are what well-run cold outreach gets, and anyone selling you forty percent is selling the promise, not the rate. What this system buys you is a channel that stays specific, honest, capped, and compliant — which is the only kind that survives long enough to matter.
Reply handling, including the reply you don't expect
Every reply gets classified within a day: interested, question, not-interested, opt-out, bounce, autoresponder, and one more that surprised us — hostile. Real inboxes receive social-engineering and prompt-injection attempts. If you put an LLM anywhere near your replies, a stranger will eventually send "ignore your instructions and…". Anything an LLM touches needs a deterministic screen in front of it, and flagged messages get a human, never an auto-reply.
None of this promises you reply rates. Single-digit reply rates are what well-run cold outreach gets, and anyone selling you forty percent is selling the promise, not the rate. What this system buys you is a channel that stays specific, honest, capped, and compliant — which is the only kind that survives long enough to matter.
We packaged the whole thing — the playbook, the quality-gate skill, the follow-up templates, the compliance rails, and the setup guide — as a single download.
The honest part
None of this promises you reply rates. Single-digit reply rates are what well-run cold outreach gets, and anyone selling you forty percent is selling the promise, not the rate. What this system buys you is a channel that stays specific, honest, capped, and compliant — which is the only kind that survives long enough to matter.
We packaged the whole thing — the playbook, the quality-gate skill, the follow-up templates, the compliance rails, and the setup guide — as a single download.
Related: Free AI Business Stack · Offer Library · Contact