How many cold emails to send before you know if it works

You sent 100 cold emails, got one reply, and now you're wondering whether cold email works for you. Or you got four replies and think you've struck gold. Neither conclusion is safe. How many cold emails to send before you know depends on one thing: how much a few replies can move your answer. This post is about test design, not volume planning (for that, see how many cold emails per day).

How many cold emails to send: the small-numbers problem

Replies to cold email are rare events. When you expect only a few replies, luck decides a big share of the outcome.

Here is a plain way to see it. Say you send to 100 people and get 2 positive replies. If two more people happened to read it on a good day, you'd have 4, and your "rate" would double. If two had been out of office, you'd have 0. Same list, same email, wildly different story. With 1,000 sends the same two-person swing barely matters.

So the rule is simple: send enough that being off by a couple of replies wouldn't change what you'd do next. There is no universal number. A rough check you can run yourself: imagine the result being two replies higher or two lower. If either version would make you decide differently, you haven't sent enough.

That also explains why checking after one day misleads. Replies keep arriving for days after the last send, and follow-ups bring more. Wait until the sequence has finished and the late replies are in.

Design the test

One audience

Pick one narrow group of people who plausibly have the same problem: one role, one kind of company. If you email founders, ops managers and marketers in one blast, a good result in one group is hidden by a bad result in another. Narrow audiences also let you write a specific message.

One offer

Test one clear offer: what you do for them and what you want them to do next. Not three offers in three emails. If the offer is vague, no amount of volume will rescue the test.

One sequence, sent evenly

Use one short sequence for everyone in the test, sent on the same schedule. Cold email follow-up covers how many touches to use. Keep the sending healthy too: if half your emails land in spam, you're testing your deliverability, not your offer. Check placement first with an email spam test, and verify the list so bounces don't pollute the sample. In SendingOS, email verification is free.

A list big enough to finish

Decide the number of contacts before you start, and don't stop early because the first week looked good or bad. If your list is too small to be meaningful, say so and treat the result as a hint, not an answer. Plan how many contacts you need with the funnel math in how many cold emails per day.

What to change between tests

Change one thing per test. If you rewrite the subject line, the offer and the audience together, a better result tells you nothing about which change caused it.

A sensible order, roughly from biggest effect to smallest:

1. The audience. Who you email usually matters more than how you word it.

2. The offer. What you're asking and promising.

3. The opening and subject line. See cold email subject lines and how to write a cold email.

4. Follow-ups and timing.

Keep notes: what you changed, the date, the audience, the number sent and the number of positive replies. A test you can't look back on is a test you'll repeat.

If you want to compare two versions of an email within one test, split the same audience randomly between them, so neither version gets the easier half of the list. Two versions each sent to a small group give you very little signal; don't over-read them.

What counts as working

Opens are not the answer. Open tracking depends on images loading, and it can't tell you whether anyone wanted what you offered. A high open rate with no replies means your subject line worked and your offer didn't. A low reported open rate with real replies means the campaign is doing better than the open rate says.

Count these, in order of importance:

1. Booked meetings from the people in the test.

2. Positive replies: people who asked a question, wanted to talk or asked for details.

3. Neutral and negative replies: these still tell you something. "Wrong person" means your targeting is off; "not now" can be a follow-up later.

Sort every reply into a bucket. In SendingOS, replies are labelled for you (Interested, Meeting, Out of office), and a reply on any channel stops the sequence everywhere, so sorting them into buckets takes minutes. The reply agent offers real open times from your Cal.com or Calendly, so interested replies turn into booked meetings quickly.

Decide before you start what result would count as a go. For example: "if at least this many meetings come from this many people, we scale it." Fill in your own numbers, and write them down first. It is easy to move the goalposts after the results arrive.

What to do with the result

• Clear positive: keep the audience and offer, and scale gradually, watching bounces and complaints as volume grows.

• Clear nothing, with healthy deliverability: change the audience or the offer, not the formatting, and run the next test.

• Murky: you're probably in the small-numbers zone. Send to more people from the same audience with the same email before you decide.

If replies are low and you suspect spam placement instead of the offer, email deliverability shows how to separate the two.