How to Test Cold Email Copy: The 18-Test Model

Picture a team that rewrites its cold email every Monday. The first version got 3 replies. The second got 5, so they kept it and changed the subject line and the list too. The third got 2. Nobody can say whether the copy improved, got worse, or whether one list was just friendlier than another.

That is what testing without a model looks like. Lots of changes, no learning.

This article explains the 18-test model we use: where the 18 tests come from, why you change one variable at a time, how many sends you need before a result means anything, which metrics to watch, when to kill an angle and how to lay out a test plan.

What the 18-Test Model Is

The 18-test model is a way to structure cold email testing so every result teaches you something. It comes straight from the ICP matrix.

• 1 ICP. Your ideal customer profile, the type of company you sell to best.
• 3 segments. That ICP split by situation: company size, business model or the trigger that makes them buy.
• 2 personas. The 2 roles that make or influence the decision.
• 6 lists. 3 segments by 2 personas.
• 3 angles per list. An angle is the core reason you give the buyer to care: a cost, a risk, a missed opportunity, a trigger.
• 18 tests. 6 lists by 3 angles.

Fig. 1How 1 ICP becomes 18 tests

1. 011 ICP

   The type of company you sell to best, written down.

2. 023 segments

   Split by situation, such as size, model or trigger.

3. 032 personas

   The 2 roles that buy, such as founder and sales leader.

4. 046 lists

   3 segments by 2 personas, each with its own reason to care.

5. 0518 tests

   3 angles per list, each sent and measured on its own.

Each test is one list paired with one angle. Because each list holds similar people in a similar situation, the difference between angles on the same list tells you which reason to care lands with that group. We cover how to build the matrix in How to Build an ICP Matrix.

Change One Variable at a Time

A classic testing mistake is changing several things at once. If a new email goes to a new list, with a new angle and a new subject line, and it does better, you have no idea why. You cannot repeat the win because you do not know what caused it.

The rule is simple: inside a test round, change only one thing.

• Testing angles? Same list, same length, same structure, same call to action. Only the reason to care changes.
• Testing lists? Same angle, same copy. Only the audience changes.
• Testing a smaller element, such as the first line or the call to action? Only after the angle is settled for that list.

Angle first is the right order. The reason you give a buyer to care moves results far more than a subject line or a sign-off. Polishing the wording of an angle that does not matter to the buyer is wasted effort.

Writing 3 angles that are genuinely different

A test only teaches you something if the angles are actually different. 3 rewordings of the same point will produce 3 similar results and no learning.

Good angles for the same list usually come from different reasons to care:

• A cost. Something the buyer is paying for now, in money or time, that they could stop paying.
• A risk. Something that could go wrong if they do nothing, such as a competitor moving first or a key person leaving.
• A trigger. Something that just happened to them, such as a new hire, a funding round or an expansion, that makes the problem urgent now.

Write each angle as one sentence before writing any email. If 2 of the sentences say the same thing, replace one. The signal activation play shows how to turn a fresh trigger into an angle the buyer recognizes. You can see angles built on real triggers in 9 B2B Cold Email Examples.

How Many Sends Before You Read a Result

This is where teams fool themselves, so it is worth reasoning through carefully instead of reaching for a magic number.

Positive replies in cold email are rare events. When you count rare events, small numbers swing a lot from pure chance. If you flip a coin 10 times, getting 7 heads is not surprising. Flip it 1,000 times and 700 heads would be.

Here is a labeled example to show why. Say two angles are sent to the same list:

• Angle A: 100 sends, 2 positive replies.
• Angle B: 100 sends, 4 positive replies.

It is tempting to say B is twice as good. But with counts this small, the difference of 2 replies could easily be luck: one person happened to be at their desk, another was between vendors. A rough rule of thumb for counts of rare events is that the natural wobble is about the square root of the count. For 4 replies, that wobble is about 2, which is the whole gap between the angles.

Now say each angle reaches 400 sends, and A has 8 positive replies while B has 20. The gap is now larger than the wobble on either count, and it is much harder to explain away as chance.

What this means in practice:

• Decide the number of sends per test before you start, and do not judge before it is reached.
• Pick a number large enough to produce a meaningful count of positive replies.
• Treat small differences as ties. If the counts are close, the angles are close.

What volume this takes

Sending capacity is the real limit. A cold inbox sends about 40 emails a day after warmup, so volume comes from more inboxes. Our own setup runs 12 inboxes across 5 sending domains, about 480 emails a day at full warmup.

As a worked example: 18 tests at 200 sends each is 3,600 sends. At about 480 a day, that is roughly 8 sending days for one full round. At 400 sends per test it is about 15 sending days. How many inboxes and domains you need for your own plan is covered in How Many Domains and Inboxes Do You Need for Cold Email?.

Reply Rate vs Positive Reply Rate vs Meetings

Three numbers matter, and they are often confused.

Fig. 2What to measure, from send to meeting

1. Delivered300

   Sends that reached an inbox without bouncing.

2. Replies12 (example)

   Every reply, including unsubscribes and out of office.

3. Positive replies5 (example)

   Interested, or a real question.

4. Meetings2 (example)

   Calls booked from the positive replies.

Reply rate is the number people quote most and the least useful for judging copy. It includes "remove me," out of office messages and angry replies. An aggressive angle can raise reply rate while lowering positive replies.

Positive reply rate is the main score for an angle. Decide in advance what counts as positive, such as interested, asked a question or referred you to a colleague, and sort every reply the same way. Our guide to handling cold email replies covers the categories.

Meetings is the number that matters to the business. It also depends on how fast and how well replies are answered, so it measures more than the copy.

Complaints are the safety check. Gmail and Yahoo have required bulk senders to keep spam complaints below 0.3% since February 2024, and Google advises staying under 0.1%. An angle that triggers complaints gets stopped whatever its reply rate.

Open rate is missing from this list on purpose. Many email clients load or block tracking images automatically, so opens are unreliable.

A Sample Test Plan

Here is how one round of the 18-test model might be laid out for a single segment, as an example. The other segments follow the same shape.

Each row is a separate test with its own list and copy. Nothing changes during the round except replies being handled.

When to Kill an Angle

Stopping angles is as important as finding winners. Our rules:

• Stop at once if an angle draws spam complaints, contains a factual error or reads badly to a buyer in a way you missed.
• Stop at the planned send count if it produced clearly fewer positive replies than the other angles on the same list.
• Keep it in the next round if results are close. A tie is a tie.
• Retire it if the same objection keeps coming back. Repeated objections are the market telling you the angle misses.

Write down why each angle was stopped or kept, in one line, next to the results. After a few rounds this log becomes the most useful document in your outbound program: a record of what your market responds to, by segment and persona, that any new copywriter or seller can read on their first day.

When an angle wins, it becomes the control for the next round, and a new challenger angle takes the loser's slot. Over a few rounds, each list ends up with the reason to care that actually works for it. The pain-qualified segments play shows how to build lists around a specific pain so angles have something real to test against.

Mistakes That Ruin Tests

• Judging after 50 sends. Small numbers mostly measure luck.
• Changing the list mid-test. Any change resets the test.
• Using open rate. It is unreliable.
• Counting all replies as wins. Sort them first.
• Testing wording before angles. The reason to care matters more.
• Testing on a burned domain. If deliverability is broken, every angle looks bad. See Cold Email Deliverability in 2026.

FAQs

Run Your 18 Tests With Oriel

Oriel runs the 18-test model for B2B companies above $100k/month in revenue. We build your ICP matrix into 6 lists, write 3 angles per list, send at warmup-safe volumes from separate domains you own, sort every reply and report positive replies and meetings per test every week. You approve every list and every first message before it sends.

On a 30-minute call, tell us who you sell to and what you have tried in cold email. We will show you what your 6 lists and first angles could be and how long a full round would take. If you are not ready for this yet, we will tell you that too.