How to write cold emails that book meetings

The four-part structure we use

Every cold email we ship has the same skeleton. The first line shows the sender looked at this company. The second names one problem that company probably has. The third states what the reader gets. The last line asks a question that takes one word to answer.

Proof is optional. When we include it, it is one line with one number from a comparable client, placed after the offer. A second proof point starts to read like a brochure, and the call is a better place for it.

The rest of this playbook covers each part in order, then follow-up timing and how to test changes without fooling yourself. If you want the surrounding system first (domains, inboxes, list, sequencing), read the cold email playbook for B2B.

First lines that earn the second sentence

Gong Labs analyzed more than 30,000 prospecting emails from more than 250 companies. It reported that referencing a company-specific topic when writing to directors and above "triples prospecting email reply rates", and that one-to-one personalization more than doubles the reply rate for non-managerial buyers, with a smaller lift of about 50 percent for directors and above (Gong Labs). These are correlations inside Gong's own dataset, so read them as direction. In practice that points to a line about the company for a head of sales and a line about the person's own work for an individual contributor.

A good first line is a fact the sender can source and the reader can check in five seconds. A job posting, a new office, a funding round, a product launch, a change in leadership. These are the same events we use as triggers when we build lists, which is why the signal-based outbound playbook and this one share a data source.

We apply one test to every first line: if it would still be true after swapping in a different company name, delete it. "I came across your company and was impressed" fails. "You have three open SDR roles and no listed sales ops hire" passes, as long as the roles are really on the careers page.

When a segment has no usable signal, write the first line at segment level and say so honestly. "Most heads of sales at 50-person SaaS companies we talk to are hiring faster than they can train" is true of a group and claims nothing personal.

Offers and asks

The offer is what the reader receives for replying. "15 minutes to discuss" is a request for the reader's time, which a stranger has no reason to give. A deliverable gives them a reason: a sample, a teardown, a short list, a quote.

Two of our cases show the pattern. For Fema Tape the offer was a sample, and it sat in every message. We contacted 850 firms in month one and got 32 positive replies against a target of 15 (Fema Tape case study). A physical product makes that offer easy. For a service, the equivalent is a one-page teardown of the prospect's current setup, or ten accounts they might want to call first.

The ask depends on how much trust the email has built, which in a first email is very little.

A calendar request in the first email asks a stranger for the most expensive thing you could ask for, so we hold it back until someone has said yes to something smaller. What happens after that yes is covered in handle replies and book the call.

Set expectations before you send. Our planning benchmarks are 2 to 3 percent of sends replying, 15 to 50 percent of replies being positive, and 30 to 50 percent of positive replies booking. On 1,000 sends that works out to roughly 20 to 30 replies, 3 to 15 positive ones, and somewhere between 1 and 7 meetings. The range is wide because list quality and offer strength move it a lot.

Length and subject lines

Length matters less than most advice suggests. All vendor figures in this article were checked on 18 October 2026. Hunter analyzed 34 million cold emails sent through its sequencing tool between 2022 and 2024, after removing very small campaigns and the extremes of reply rate. Its author reports that emails under 100 words averaged a 3.9 percent reply rate against 3.7 percent for longer ones, and warns that the gap is not spectacular and that replies are not the same as positive replies (Hunter).

The pattern is noisy, with no steady slope. We write between 50 and 100 words because that is enough room for a first line, a problem, an offer and an ask, and short enough to read on a phone without scrolling. If an email needs 200 words, the offer is probably unclear.

For subject lines, Woodpecker's data from its platform (campaigns in the last 90 days of 2025, page updated 5 August 2026) says subjects of 20 to 29 characters had the best reply rate in both the US and non-US populations it examined, and that reply rates peaked at three to four words (Woodpecker). The same page calls open rate the noisier metric and says reply rate is the number that matters.

Our subject lines are short, lowercase and plain, and they match what the email says, such as "sdr hiring at [Company]". A subject that promises more than the body delivers invites spam reports, and Google tells bulk senders to keep the spam rate reported in Postmaster Tools below 0.10 percent and to avoid ever reaching 0.30 percent (Google). Those rules formally apply at 5,000 messages a day, but the targets are a sensible ceiling at any volume. More on that in the deliverability playbook.

Follow-up timing

Most replies come fast. Smartlead's dataset of more than 850 million emails, limited to senders with at least 500 sent, found 93.9 percent of replies arrive within 72 hours of an email, so a gap of three days or more gives each message time to land before the next one goes out (Smartlead).

The same dataset shows the return on each additional email falling: 1.18 percent reply rate on email one, 0.73 percent on email two, 0.56 percent on email three, and 0.33 percent on email seven. It recommends planning for three or four emails in total. Woodpecker reports a similar decline (1.0 percent on email one, 0.5 percent on email four) but also finds that in two-email campaigns the follow-up drew a 1.4 percent reply rate against 1.2 percent for the first email.

Send on business days in the prospect's time zone, and stop the sequence the moment anyone replies, positive or not. Route every reply to a person before any automated response goes out.

How to A/B test with sample sizes that mean something

Cold email tests fail because reply rates are low, and low rates need large samples. At a 2 percent baseline, a campaign of 200 emails produces about four replies. A variant that gets six replies has not won, because two extra replies are within chance.

The method we use has five rules. Change one variable per test. Split the same segment at random, from the same inboxes, on the same days. Decide the sample size before sending and do not stop early when one arm pulls ahead. Judge on positive replies and booked calls, not opens. Run the winner as the new control and test the next variable.

For the sample size, a two-sided test at 5 percent significance and 80 percent power gives these figures by the standard normal approximation for two proportions. This is our arithmetic, not a published benchmark, and calculators such as Evan Miller's take the same four inputs (baseline rate, minimum detectable effect, significance, power) and may return somewhat different numbers depending on their defaults.

The consequence is an order of tests. Segment and offer changes produce big swings, so test them first. First-line style and the ask come next. Subject lines come last, because a one-point lift on a 2 percent baseline needs roughly 7,600 sends in total. At 1,000 sends a day that is under eight days; at 100 a day it is 76 days.

Volume sets the limit. A run like the Fema Tape one, 850 firms in the first month, leaves about 425 per arm, below even the 3 to 6 percent row above. At that size you can test a new offer against an old one and expect to see a large difference, but not a new subject line. At the other end, the AI feedback SaaS in our work sent 1,000+ emails a day across multiple inboxes and booked 40+ meetings a month, which is the volume where careful testing becomes practical. The infrastructure that supports that is described on our cold email systems page.

Two worked examples

Both emails below are illustrations written for this article. They name no real company, and the details in brackets are placeholders to replace with facts you have verified.

Example 1: first email, signal-based

Subject: sdr hiring at [Company]

Hi [First name],

[Company] has [three] open SDR roles on its careers page.

We build and test the target list and the first 200 emails before new reps start, so they begin on a segment that already gets replies.

I can send a one-page teardown of your current target list with ten accounts I would call first. Want it?

[Sender name]

The email runs under 80 words. Its first line is checkable, it implies one problem (new reps starting without a tested list), the offer is a deliverable, and the ask has a one-word answer.

Example 2: email two, three days later

Subject: re: sdr hiring at [Company]

Hi [First name],

One addition to my note on Monday. If the teardown is more than you need, I can send just the ten accounts, with no call attached.

Reply "list" and I will send it today.

[Sender name]

The follow-up cuts the offer to its smallest useful form, so the reader can say yes in one word.

Questions

How long should a cold email be?

Between 50 and 100 words works for most offers. Hunter's analysis of 34 million emails found only a small gap between emails under 100 words (3.9 percent average reply rate) and longer ones (3.7 percent), so clarity matters more than the exact count.

How many follow-ups should I send?

Plan three or four emails in total. Smartlead's data shows reply rate falling from 1.18 percent on email one to 0.33 percent on email seven, and Woodpecker found follow-ups still add replies in short sequences.

What reply rate should I expect from cold email?

Our planning benchmark is 2 to 3 percent of sends replying, with 15 to 50 percent of those replies positive. Woodpecker reports a median of 1.5 percent across 56,614 campaigns, so a new, untested list can land below our range.

Should I A/B test subject lines?

Test them last. At a 2 percent reply rate, detecting a lift to 3 percent needs about 3,800 emails per variant, so offer and segment tests give you answers much sooner.