Cold Outbound Playbook

Cold outbound works for a startup when three things hold: a short list of companies that look like your best customers, a reason to write to each of them now, and a sending setup that lands in the inbox. After that it is arithmetic: work back from the customers you need to the people you must reach, write short emails about their problem, follow up across email and LinkedIn, and answer every reply the same day.

Founders tell me cold email is dead. Usually what died is the version they tried: a bought list, a pitch about the company, and thousands of sends from the main domain in the first week. In my view, outbound still works when it is narrow and honest, because a well-chosen prospect has the problem today and doesn’t know you exist.

It is harder than it was. In Gartner’s survey of 646 B2B buyers, fielded in August and September 2025, 67% said they prefer a rep-free buying experience^[1]. So a cold email has to earn the reply on its first line, and the meeting it asks for has to feel smaller than the effort of ignoring it.

This playbook is the system around the emails: the target, the list, the sending setup, the sequence, the replies and the review. How to write one good email is covered in Cold Email for Founders.

Outbound fits when your deal is big enough to pay for a person’s time and your buyers can be named before they raise a hand. If either is false, put the hours into inbound, partners or product-led growth instead. Check three things first: whether a first-year contract is worth a conversation, whether you can name the buyer’s job title and companies, and whether you know why they would buy this quarter.

Then plan from the target, not from a list. Divide the customers you need by your win rate from a held first meeting, then by your show rate, then by the share of positive replies that book a meeting, then by your positive replies per person contacted (Figure 1). Use your own rates where you have them. Where you don’t, write the assumption down, label it, and replace it after the first month.

Count positive replies per person contacted, not replies per email sent. They are different numbers: one person can get five emails, and a “no thanks” is a reply. Only the first is a planning rate.

A good list does more for replies than a good email. Build it in three layers: the companies that fit, the people inside them, and a reason to write to each one this month.

Companies. Pull your closed-won deals and look for what they share: industry, size, the tools they run, how they buy. Turn those into weighted signals and tier every target account A, B or C; the ICP and Account-Scoring Playbook walks through it. Send to A-tier first and C-tier never.

People. Pick the person who feels the problem every week, not the most senior title, and one backup in the same company to contact only if the first doesn’t answer. My rule of thumb, not a benchmark: two people per company at seed. More than that and colleagues compare notes about your sequence.

Triggers. A trigger is a public event that makes your problem more likely this quarter: a new leader in the buyer’s seat, a funding round, a job posting for the role your product helps, a system migration. Write the trigger and its date beside every account. If you can’t find one, the account waits.

Verify every address before it enters a sequence, because bounces hurt the sending domain, not just the campaign. My rule of thumb, not a benchmark: keep bounces under 2%, and treat catch-all addresses as unverified. Don’t buy lists of strangers; they come with no fit score, no trigger and stale addresses.

The best email does nothing in a spam folder. Set up the sending side two weeks before the first campaign, and check the rules for every country you send to.

Send cold email from a separate domain that looks like yours, such as examplecohq.com for example.com, with its own mailboxes. If cold sending hurts that domain’s reputation, your invoices, password resets and customer email still arrive. That is my practice, not a mailbox provider’s rule.

Authenticate, warm up and pace

Gmail has required SPF or DKIM from everyone who sends to Gmail accounts since February 1, 2024, and DMARC from senders of 5,000 or more messages a day^[2]. Yahoo requires SPF or DKIM from all senders, and SPF, DKIM and a DMARC policy of at least p=none from bulk senders^[3]. Set up all three on day one: it takes an hour, and the Email Authentication Checker shows what is missing.

Warm new mailboxes with a few normal emails a day for two to three weeks before any cold sending. My rule of thumb, not a benchmark: no more than about 30 cold emails per mailbox per day after that, and more mailboxes rather than more volume per mailbox. Google asks senders to keep the spam rate reported in Postmaster Tools below 0.10% and never reach 0.30%^[2]; Yahoo’s ceiling for bulk senders is 0.3%^[3].

Know the four rulebooks

The rules depend on where the recipient is, not where you are (Figure 3). In every country, the same habits keep you on the right side: write as yourself, from a real company, with a postal address, an easy way to say no, and a do-not-contact list you honor the same day. The Marketing Compliance Checklist lists each regulator’s own page. This is orientation, not legal advice.

In my view, the first email carries the sequence: if it misses, the follow-ups rarely rescue it. Write it in four parts, in plain text, with no images (Figure 4). My rule of thumb, not a benchmark: under 80 words.

Open with the trigger, then state the problem it usually causes as a cost they would recognize. Add one piece of proof you can name: one customer, one result, true and checkable. If you have no proof yet, say what you are learning instead, and never invent it.

End with one small ask: a yes-or-no question about interest, not a 30-minute demo. Write the subject line last, in three or four lowercase words a colleague might use. Leave out your funding, your founding story and your feature list.

AI can draft the variations, but a person checks every claim before it goes out. A personalized line that gets a fact wrong about the prospect is worse than none. If you sell an AI product, check accuracy and savings claims with the Claims & Proof Checker first.

Follow-ups reach the people the first email caught on a bad day. My rule of thumb, not a benchmark: five emails over about two weeks, two to four days apart. Every touch brings a new reason to reply, such as a second problem, a piece of proof or a question, and the last one says you will stop, and you do.

LinkedIn works as a second channel when it carries a different kind of touch. View the profile, connect with a short note that doesn’t pitch, and send a message only once the person accepts. A prospect who has seen your face and a few of your posts reads your next email differently, which is why the Founder-Led LinkedIn Playbook comes first for most founders.

I do LinkedIn by hand and don’t use tools that send invitations or messages as me. Read LinkedIn’s User Agreement before you buy any tool that acts for your account. In my view, a founder’s profile is too valuable to risk on a shortcut.

A reply is the one part of outbound you can’t schedule, and the part that pays. Sort each reply into one of five types: interested, not now, wrong person, objection, or no. Give each type an owner and a deadline, and write them down (Figure 6).

Answer the question before you ask for the meeting. If they asked for the two-line summary, send it, and put the meeting request underneath. A “no” goes on the do-not-contact list the same day: US law allows 10 business days for an opt-out, but there is no reason to use them.

Hand off everything the seller needs: the trigger, the email they replied to and their exact words go into the meeting note, so the first call starts from their problem. Running a Discovery Call covers the call itself.

Outbound improves through small, logged changes, not a new tool every quarter. Name one owner, review every two weeks, and read the stages from the top, because a problem upstream makes every number below it look wrong (Figure 7).

Plan from your own rates, not published ones. The reply-rate figures that circulate come from sending-tool vendors’ own customers, count replies per email sent, and include every “no”, so they don’t match positive replies per person. In my view, they are not worth planning on.

Change one element per test, such as the trigger, the subject line or the ask, and split new prospects evenly between the versions. At outbound volumes most differences are too small to be sure of within two weeks; the A/B Test Significance Calculator shows how many people a test needs. Log every test, its sample, the result and the decision.

What changes by stage?

What mistakes should I avoid?

• Starting with a list instead of a target. Without the arithmetic you can’t tell whether 300 accounts is enough (Step 1).
• Buying a list of strangers. No fit, no trigger, stale addresses (Step 2).
• Sending from your main domain. One bad month can send your invoices to spam (Step 3).
• Pitching in the first line. The reader decides on line one, and it should be about them (Step 4).
• Follow-ups with nothing new. “Just checking in” gives no reason to reply (Step 5).
• Letting an interested reply wait until tomorrow. Answer within the hour (Step 6).
• Emailing someone after they said no. It breaks CAN-SPAM in the US and loses the person’s trust everywhere (Steps 3 and 6).
• Changing three things at once. You won’t know which one worked (Step 7).