Cold email personalization that is worth the time

Personalization in cold email has come to mean merge tags: a first name in the greeting, a company name in the first sentence. That is not personalization, it is mail merge, and it has been standard in every sending tool for long enough that recipients read straight past it.

Real personalization means the message could only have been written to this person. It is expensive, it does not scale cleanly, and it is the single largest lever on reply rate once your infrastructure is sound.

The three tiers

It helps to be honest about what each level of effort actually buys.

Merge tags are table stakes and buy nothing. The interesting question is how to split effort between the second and third tiers, because the third does not scale and the second does.

Segment-level is the best value

Writing genuinely different copy for a tightly defined segment gets most of the benefit of individual research at a fraction of the cost. The narrower the segment, the more specific the copy can be while still being reusable.

A message to heads of sales at fifty-to-two-hundred person B2B software companies can name problems that group actually has, in the words they use. That is not possible in a message written for everyone, and it does not require researching each recipient.

• Define segments by shared problem, not by firmographic convenience.
• Write the whole message per segment, not one swapped sentence.
• Aim for segments small enough that the copy would look wrong sent to the next segment over.
• If the same copy works for two segments, they are one segment.

Individual research: what is worth finding

When you do research individuals, most of what people find is useless. Congratulating someone on a funding round is now so common it reads as automated, because it usually is.

What works is an observation that implies a problem you can solve. The test is whether the observation leads naturally into your next sentence. If the personalised line could be deleted without changing the rest of the message, it was decoration.

• Job listings that imply a gap the role is meant to fill.
• A recent product or market change that creates the problem you address.
• Something they wrote or said publicly that states the problem in their own words.
• A visible technical detail, when your product concerns that detail.

Anything you could learn about a thousand companies in a single query is not research, it is a merge tag with extra steps.

AI personalization

Generated opening lines are now common, and recipients have started to recognise them. The characteristic failure is a line that is fluent, accurate and completely generic - it describes the company correctly while saying nothing that implies a reason to write.

Used well, generation is a research accelerator rather than a writing tool: summarise a prospect, surface the detail worth mentioning, then write the sentence yourself. Used badly, it produces individually plausible messages that are collectively identical, which is exactly the pattern receiving systems are built to detect.

The arithmetic of personalisation at volume

Individual research at five minutes each caps one person at roughly a hundred emails a day of pure research time and nothing else. That is the real constraint, and it should shape the campaign rather than being discovered halfway through.

The usual resolution is tiering the list: individual research for the accounts worth winning, segment-level copy for the long tail. Deciding this in advance is what stops a campaign becoming a hundred rushed, badly researched messages that perform worse than good segment copy would have.

Volume itself has a hard floor set by infrastructure rather than by writing speed. Since a Microsoft 365 mailbox safely sends around three cold emails a day, the number of mailboxes you run bounds how much personalisation you need to produce - the volume calculation is worth doing before you plan the copy.

Personalization does not fix deliverability

A deeply researched message to exactly the right person still lands in Junk if the sending domain is unauthenticated or has no history. Personalisation raises reply rate among people who see the message; it has no effect on whether they see it.

The same reasoning applies to subject lines, where a specific observation outperforms a generic hook for the same reason. One caveat in the other direction: merge fields are a common source of visible breakage. A message opening Hi {first_name} is worse than one with no personalisation at all, and broken tokens are worth checking for before every send.