Operations

How to find a winning campaign in the first two weeks

By Travis Heurtelou · 6 min read · September 14, 2026

The short version

An angle is a campaign, not an A/B variant. Run each on its own random slice of the list and compare them.
Measure positive replies per contact. Opens and raw reply counts are noise.
A reply rate under half a percent is a deliverability problem, not a copy problem. Check that before touching a word.
Day fourteen: feed the winner, kill the losers, replace them with new angles. Don't add steps.

Angles are campaigns

Most people test copy by writing variant B of the same email and splitting traffic. That tells you whether a word choice mattered. It can't tell you whether the angle was wrong, because both variants share it.

An angle is the reason the email exists: the signal it opens on, the offer it makes, who it's addressed to. Change any one of those and it's a different campaign. Run three to five at the same time, each on its own slice of the list, each with the same two-email shape. Then compare.

The slice has to be random. Give one angle the fresh half of the list and another the stale half and you're comparing the angle and the list at once. You can't separate them afterward.

What to measure

Positive replies per contact. Not opens, because tracking pixels misfire and Apple Mail inflates them. Not raw replies, because out-of-office, 'unsubscribe' and 'wrong person' are replies. Positive replies, meaning a human wrote back and wanted the thing, divided by the number of people you contacted.

The denominator is contacts, not emails sent. A two-step sequence touches most people twice, so a per-email rate reads about 2× worse than a per-contact rate for the same result. Pick one and never mix them in the same sentence.

For untested copy, plan on a fraction of a percent. ListKit's published account-wide baseline across 7,930 campaigns was 0.046% positive per prospect. Their top 200 scripts ran 0.156%. Those are ceilings from a survivor set, not forecasts, but they tell you the order of magnitude. A thousand contacts isn't a test.

The check before you touch the copy

If replies of any kind are under half a percent of contacts, stop reading the copy. That number means the email isn't landing in inboxes. A person who saw the email replies at some rate no matter how bad it is, if only to say no.

Check the sending side first: warm-up age on the inboxes, bounce rate, whether a blacklist picked up a domain, whether the tracking domain is set up. Rewriting the email while it's going to spam tells you nothing, and the rewrite gets blamed for the spam.

Two weeks, then decide

Cold email doesn't give you a read in two days. Replies to the first email keep arriving for a week, and the follow-ups add more. The first real comparison between angles is around day ten to fourteen, once each has a few thousand contacts through it.

At that point the leader gets more list. The losers get killed, not fixed. A losing angle with a better subject line is still a losing angle. Replace it with a new angle and run the next round.

Don't add steps to a loser. In ListKit's 200 scripts, the sequences with 7 or 8 steps were the worst bucket by median (0.145% per prospect) and 1 to 3 steps the best (0.380%). More touches on a weak angle is more spend on the same answer.

The list caps the test

Three angles at a few thousand contacts each is ten thousand contacts before you've learned anything. The list has to exist before the campaign does. Size the test from the buyers you can actually reach in the market, not from the market's headcount, and build the list first. Inboxes are the easy part.

Three to five angles as separate campaigns on a random split. Positive replies per contact. Deliverability check before any rewrite. Day fourteen: kill the losers, feed the winner, replace the rest.