One email is one advert. How to expand into many angles, find the heat, and compress onto the winner.
Test cold email copy like a portfolio of adverts. Write 8 to 12 genuinely different angles, run each at enough volume to separate signal from luck, judge them on positive replies and meetings rather than opens, then move volume onto the winners. Keep around 15% of volume testing new challengers, because every winning angle eventually decays.
Treat every distinct email angle as a separate advert competing for budget. Once you think about it that way, the process stops being creative writing and becomes portfolio management.
The usual test is two versions of the same email with a different subject line, sent to a few hundred people each. The difference between them is small, the sample is smaller, and the "winner" is mostly noise. Then the team ships it, results regress, and everyone concludes testing does not work.
Good testing fixes three things at once: the variants are genuinely different, each one gets enough volume, and the scoring uses the outcome you actually care about.
Start wide. Write eight to twelve genuinely different angles, not twelve rewrites of one idea. Different problem framings, different proof, different asks. Most will lose. That is the point, and you cannot know which in advance.
Ways to make angles genuinely different:
If you only run three or four angles, the best you can do is pick the least bad of a mediocre pool. The angle that performs two or three times better than the rest usually comes from somewhere you would not have predicted.
Run them at real volume against comparable segments. This is where small lists kill programmes: with 2,000 contacts you cannot separate a good angle from a lucky week. You need enough sends per variant for the difference to mean something. As a rough rule, do not call a winner until each variant has a few hundred sends and at least a handful of positive replies.
Judge on positive replies, not opens and not total replies. An angle that generates lots of "please remove me" is generating engagement and no pipeline. Better still, track all the way to meetings booked, because some angles pull warm replies that never convert.
Expect the average to get worse before it gets better. When you test ten angles, most of them lose, so your blended numbers dip during the testing phase. If you or your client are not ready for that, you will kill the test just before it pays off.
Kill the losers quickly and move the volume onto what is working. Most programmes are too slow here out of attachment to an angle someone liked in a meeting. Cutting to winners only is often where the biggest single jump in results comes from.
Never put 100% of volume on the current winner. Fatigue is real, markets move, and the angle that works this quarter will decay. We hold roughly 15% of volume permanently in exploration, so there is always a challenger being tested. The cost of that 15% is far lower than the cost of discovering your only angle has died.
Log what each email contains: the subject line style, the opening, the ask, the proof, the tone. Without tags you can only compare whole campaigns. With tags you can compare elements, and see, for example, that every email opening on a specific pain point beats every email opening on a compliment, regardless of the rest of the copy.
Personalisation earns its cost when it references something the prospect would recognise as specific to them: a funding round, a regulatory change, a job posting, a tool they clearly use. Inserting a first name and a company name is not personalisation, and every recipient knows it.
Be careful with proxy wins. Personalised subject lines often lift replies without lifting meetings. Anything that improves a vanity number but not the outcome is a trap, and it is exactly the kind of trap that tagging and outcome tracking catch.
As many genuinely different angles as your volume can support with a few hundred sends each. For most B2B programmes that is somewhere between four and twelve live variants.
Eventually. Test the ask and the angle first, because they move results far more. Plain, internal-sounding subject lines are a good default.
Yes. Every follow-up should add something new, such as a new angle, proof or mechanism, rather than "just bumping this". Test whether each one adds meetings, and drop the ones that do not.