Outbound that doesn’t spam.

The core principle

Good outbound treats the recipient as a person you could meet at a conference. Bad outbound treats the recipient as a row in a spreadsheet. AI lets you do either at scale. The default, if you do not intervene, is bad.

The principle: never send a message you could not defend if the recipient replied “how did you get my email and why are you writing to me.” If the answer is “scraped from LinkedIn, blast template,” the answer is wrong. If the answer is “you published a blog post in March about our exact problem space, here is the specific thing I thought you would find useful,” the answer is right.

Before writing a single message

Is your ICP crisp?

You cannot personalize to an ICP you have not defined. If you are writing to “CTOs at Series A companies” you are going to lose to anyone writing to “CTOs at Series A companies that just posted a senior security role and are in healthcare or fintech.” Specificity is the prerequisite.

Are you legally clear?

Different jurisdictions, different rules. Summary (not legal advice; verify with counsel before you send):

• US (CAN-SPAM): B2B cold email generally allowed. Must identify sender, offer unsubscribe, honor it within 10 business days, physical mailing address in the footer.
• Canada (CASL): Strict. Express or implied consent required; implied consent includes a recent business relationship or a publicly posted email without a “no unsolicited” notice. Honor unsubscribe within 10 business days.
• EU (GDPR + PECR): Stricter. Cold B2B to personal-looking emails (john.smith@company.com) risks being personal data under GDPR. Role-based emails (contact@, sales@) are safer. Legitimate-interest basis is the usual argument; document your balancing test.
• UK: Similar to EU post-Brexit. PECR applies; GDPR equivalent in effect.
• Australia (Spam Act): Consent required. Inferred consent from “conspicuously published” work email acceptable for relevant business matters.

Hard rules that work everywhere: identify yourself, offer a real way to opt out, do not hide behind shell domains, respect opt-outs across all your domains, do not message the same person more than the sequence cadence allows.

Is your sending infrastructure healthy?

A technically illegal message on a pristine domain might land. A perfectly compliant message from a burnt domain will not. Before sending at volume:

• SPF, DKIM, DMARC set up properly. Not approximately - actually.
• Warm the domain for 2 to 4 weeks before high-volume sending. Start at 10 to 20 emails per day, ramp gradually.
• Dedicated sending domain (mail.yourcompany.com) so failures do not burn your primary domain.
• Monitor bounce rate, complaint rate, inbox placement. If any of those spike, stop.
• Personal-seeming mailboxes (firstname@yourcompany.com) outperform noreply@ for cold outreach.

The research pass (where AI earns its keep)

For every prospect, do a 5-minute research pass before writing a message. This is where Claude is most useful and where template senders skip entirely.

Inputs:

• Company: URL, recent news, funding, job postings, product pages
• Person: LinkedIn, Twitter / X, GitHub or other public profiles, any public writing

Ask Claude (or your pipeline) for:

• 2 to 3 specific things about this company or person that suggest the pain point your product solves
• 1 specific hook - a job post, a product launch, a public post, a change in the team - that is timely
• 1 reason this prospect might not be a fit, so you can rule out before sending

If the research cannot produce those three, the prospect is not ready to contact. Drop them, do not send anyway.

Message structure: the four-sentence email

A cold email that earns replies is roughly:

1. Hook. One sentence. Something specific about them, proving you did research. Not “I saw you on LinkedIn.”
2. Bridge. One sentence. The connection between what you noticed and the problem you solve.
3. Offer. One sentence. What you could do for them, with specificity. Not “we help companies like yours.”
4. Ask. One sentence. A concrete, small next step. Not “can we hop on a call.”

Total word count: 60 to 100. Subject line: 4 to 7 words, lowercase, specific. No exclamation marks. No “quick question.” No “[Name]” in the subject line.

Example (bad and good)

Bad (the kind of AI-generated outbound that gets your domain blacklisted):

Subject: Quick question about [Company]

Hi [First Name],

I hope this email finds you well. I came across [Company] and was impressed by your innovative work in the space. I wanted to reach out because we help forward-thinking companies like yours unlock their growth potential through our cutting-edge AI solutions.

Would you be open to a quick 15-minute chat this week?

Best, Sales Bot

Good (specific, researched, short, respectful):

Subject: your IAM job post at Cloudwrit

Priya - the senior security engineer req you posted two weeks ago lists “tenant isolation audit” as top-three. We just finished an audit exactly like that for another B2B SaaS at your ARR range.

Happy to share the redacted findings summary (12 pages, 20 minute read). No call needed unless useful.

NexcurAI

The good version is 60 words. It names a specific artifact (the job post), a specific problem (tenant isolation audit), a specific proof (we did one recently), and a specific small next step (read a 12-page summary). It does not ask for time. It does not mention “cutting-edge AI solutions.”

Sequence design

A sequence is 3 to 5 messages over 2 to 4 weeks. Any more and you are pestering.

Pattern:

• Day 1: Initial message (as above).
• Day 4: Bump. Two sentences. New small value ( “thought of you because of X published today” ) or a sharper ask.
• Day 10: Different angle. Same problem, different frame, or a different artifact.
• Day 17: Break-up. Honest closer. “I will not keep pinging - if the timing is wrong, no problem. Here is the summary in case it is useful later.”

Do not do 12-step sequences with “following up on my previous email” spam. It does not work and it hurts the brand.

Reply handling: this is the whole job

Most outbound programs fail at the reply step, not the send step.

Rules:

• Human replies within 4 business hours. Claude can draft; a human sends. Speed matters more than polish.
• Match the reply length to theirs. A one-line reply gets a one-line reply back. A paragraph earns a paragraph.
• Answer the actual question. Do not pivot to pitch. If they ask “how is this different from [competitor],” answer that precisely.
• Unsubscribe is always honored, never challenged. “Unsubscribe” or “not interested” gets a brief “understood, removing you now” and nothing else.

Metrics to watch

• Reply rate (positive + negative + neutral). Target: 8 to 15 percent for well-researched outbound. Below 5 percent suggests the research or targeting is off.
• Positive reply rate (interested, wants to learn more). Target: 2 to 5 percent. This is the real number.
• Meeting booked rate. Target: 0.5 to 2 percent of sent. Highly ICP-dependent.
• Complaint rate. Above 0.1 percent is a problem. Above 0.3 percent and providers start throttling you.
• Bounce rate. Above 2 percent suggests list hygiene issues. Clean the list.

When outbound is the wrong motion

• Consumer products. B2C outbound is almost always spam. Different playbook entirely.
• Under-baked positioning. Launch-stage companies should do founder-led conversations before automated outbound. Outbound amplifies whatever your positioning is - if it is weak, you amplify weakness.
• TAM under 200 accounts. At that size, do 1-to-1 account research and real relationship building. Automated outbound at that size is overkill.
• When you cannot answer replies fast. A program that generates replies and does not answer them is worse than no program. Staff first, send second.

When to talk to us

We run outbound programs for clients on marketing retainers. If you are trying to build this in-house, or want a second set of eyes on the sequences you already run, start a conversation.