Personalization that isn't templated
Almost every cold email I get calls itself personalized. Almost none of them are.
The first line says "I noticed Craftt is in the CRM implementation space" or "saw you do Attio work, must be exciting given the recent funding round" or "love what you are building in the GTM tools category." The sender thinks this is personalization because the words came out different for each recipient. It is not personalization. It is a template with a firmographic slot in it, and the recipient can tell.
I know they can tell because I am the recipient. And I know I do it too, because I run cold outreach for Craftt and I have caught myself producing the exact same thing under deadline pressure.
This post is the rule I now run our outreach by, the format that makes it cheap enough to actually do, and what I learned about why "personalized at scale" almost always means "templated with extra steps."
The two kinds of personalization
There is firmographic personalization and there is personal personalization. They look similar in a sent email and they are completely different in how they land.
Firmographic personalization slots in things about the company: the funding round, the industry, the product category, the technology stack. It is cheap because the data sits in a database and the merge happens at send time. Apollo, Clay, ZoomInfo, every enrichment tool sells this. It is what most "AI personalized outreach" tools mean.
Personal personalization slots in things about the human: what they posted last week, what they said in a podcast, the project they built before this job, the niche of the company they care about (not the company they happen to work at). It is expensive because the data does not sit in a database. Someone has to read a LinkedIn profile and a few posts and write a sentence that the recipient could not have received from anyone else.
The first kind is a template in disguise. The second kind is a letter.
The recipient can always tell the difference. Not because the words give it away (sometimes they do, sometimes they do not). Because of what the email asks of them. A firmographic email asks them to imagine themselves as a representative of their category. A personal email talks to them as a person who happens to work somewhere. The first is uncomfortable to reply to. The second is not.
The shift I had to make
I did not arrive at this rule from first principles. I arrived at it from sending a few hundred firmographic emails and watching the reply rate stay flat regardless of how clever the firmographic line was.
The thing that finally moved the number was not a better template. It was opening the recipient's LinkedIn before writing anything and finding the one specific thing they had been thinking about. Sometimes it was a post from two days ago. Sometimes it was a job title transition they had just made. Sometimes it was a side project on their profile that had nothing to do with their day job but was clearly the thing they cared about.
The line that came out of that reading was always different from the firmographic line, in a way the recipient noticed. The reply rate went up by a multiple. Not a percentage. A multiple.
The cost was real. Going from a firmographic merge to a per-person research pass took the per-email work from about thirty seconds to about ten minutes. That is the trade. The only honest version of "personalization at scale" is not scale, it is per-person research with a tighter format.
The format
Once I committed to per-person research, the next problem was that the personalization line was bloating the email. Two sentences of someone's biography is not a cold email. It is a creepy email.
The format I landed on, after a lot of iteration, is one line, lowercase, in a specific shape:
>
Examples from our recent outreach:
infra-focused vc partnerships at a16zrecruitment ops at a 12-person dev agencybookkeeping for ecommerce founders at bench
Three constraints make this work.
Lowercase, because capital letters in the middle of a sentence make the line read as a job title, and a job title is firmographic. Lowercase makes it feel like the sender wrote it down in a notebook.
, not industry. The niche is the part of the work the person actually cares about. "AI infrastructure VC" is a niche. "Venture capital" is an industry. The first comes from reading their bio. The second comes from a dropdown.
, not role. What they do every day, not what HR calls them. "Partnerships" or "ops" or "bookkeeping" or "policy research" are activities. "Director of Strategic Alliances" is a role. The activity is what they think of themselves as doing. The role is what their offer letter says.
The line goes into the merge slot in the first sentence, lowercase, and the rest of the email is about why the sender is reaching out. The personalization carries the weight of the opening without being the opening. The reader reads it and registers "this person actually looked at me" before they read the pitch.
I keep this field as a People attribute in Attio. It is called personalization. It is the only attribute that lives on a person specifically for outreach, and I will not send an outreach email if that field is empty on the recipient.
Why this is not "AI personalization"
There is a category of tools right now that promise AI personalization. They scrape LinkedIn, summarize a profile, and produce a personalized line at the top of a cold email. I have tried most of them. None of them produce a line I would actually send.
The reason is not that the AI cannot write the sentence. The AI can write the sentence. The reason is that the AI does not know which thing to write about, because choosing which thing is the actual creative work. A profile has dozens of facts. The one that matters is the one that signals "I read your work, not your title." A model that takes the average of the profile and writes a sentence about the average produces firmographic personalization in a more elaborate dress.
The version that works is the model writes a candidate sentence after I tell it which post or which line in the bio to use. I do the picking. The model does the typing. The picking is ten times faster than the writing, and the picking is the part that has to be human.
In practice this means I open three browser tabs: the recipient's LinkedIn, their most recent post or article, and our outreach skill in Attio. I read for two minutes. I write the personalization line in the shape above. The skill takes that one field and writes the rest of the email around it. The model does not produce personalization. It produces the scaffolding around it.
What this rules out
This rule means we will never run an outreach list of 10,000 people. We will run lists of 200. We will not buy a list of 10,000 contacts with a personalization tool bolted on. We will not run a sequence where the personalization step happens at send time.
It also means our cost per email is higher than the cost per email of the average firmographic outbound shop. It is not even close. A firmographic shop can send 10,000 emails for the cost of about 100 of ours.
The reply rate makes the trade work. A list of 200 where every email got ten minutes of research has, in our experience, a reply rate that lands the same total number of conversations as a list of 10,000 firmographic emails. With one difference: the 200 conversations are with people who already feel respected by the sender, and the 10,000 firmographic sends include some unknown number of people who now think less of the brand.
The asymmetry is the whole reason I run it this way. A bad cold email has a real downside. The recipient now associates the brand with the discomfort of receiving it. Most outreach math ignores that downside because it is hard to measure. It is real, and it adds up over thousands of sends.
If you are running cold outreach
If you are running cold outreach right now and the reply rate is flat, the most reliable thing you can do is cut the list by 90% and spend the saved time on per-person research. Not a better template. Not a smarter AI tool. Cut the list, read the profile, write the line.
The line does not have to be clever. It has to be specific. It has to be a thing the recipient knows is true about themselves and could not have received from anyone else. The format above is the cheapest version of that I have found, and the rule that goes with it is the rule I would not run an outbound campaign without.
Personalization that isn't templated is the only kind worth sending. Everything else is a template asking the recipient to pretend otherwise.
Need help with your CRM?
Messy data, manual processes, or a CRM that doesn't fit? Let's talk.
Book a call