There is a line that appears in cold emails so frequently it has become its own genre of failure. "I noticed your company recently raised a Series A — congratulations on the milestone!" Or: "I came across your LinkedIn post about scaling your sales team — really resonated with me."
These lines feel personal. They reference something real. They are technically specific. And they fool nobody. Every experienced B2B buyer has read hundreds of versions of these lines. The personalisation is real in the narrow sense. The interest is not.
"I noticed your company recently raised a Series A — congratulations on the milestone!"
"I came across your LinkedIn post about scaling your sales team — really resonated with me."
"As a fellow founder building in the [industry] space, I thought this might be relevant."
Data fields populated from a spreadsheet. Designed to create the impression of research without doing any. People who spend their careers reading email can feel the difference.
This is the fundamental problem with personalisation at scale: the tools make it easy to appear interested in someone without actually being interested in them. Here is what I have learned about personalisation after 10 years of outbound — what actually works, what is sophisticated theatre, and how to build genuine personalisation that scales without losing what makes it genuine.
The three levels of personalisation — and where most people stop
Most outbound personalisation operates at Level 1. The best outbound personalisation reaches Level 3. Very few teams ever build to Level 3 systematically.
Company name, industry, title, funding stage, company size
The data you can pull from a database without visiting a single company's website. Necessary but completely insufficient. Every tool that claims to "personalise at scale" starts and ends here.
Insufficient aloneWhat this company is actually doing right now
Recent hires on LinkedIn, product launches, job postings that imply a specific pain, funding rounds that suggest growth priorities. Requires more research than Level 1 — but can be partially systematised using Clay, which monitors these signals at scale.
Better — still not enoughWhat it actually feels like to be in this person's situation
Not from research — from experience. The kind of observation that makes a prospect think "this person has actually been here." This level cannot be generated from data. It can only come from someone who has genuinely done the work.
The level that consistently drives repliesWhy Level 3 is the only level that consistently drives replies
The challenge with Level 3 is that it cannot be scaled the way Levels 1 and 2 can. You cannot build a database of lived experience. You cannot automate genuine insight.
What you can do is build a personal narrative repository — a structured library of real experiences, real client situations, and real observations — and use AI to select and embed the most relevant fragments into outbound copy that already has Levels 1 and 2 in place.
The three-layer system in practice
Clay handles firmographic enrichment (Level 1) and monitors for contextual signals — recent hires, funding rounds, job postings, news mentions (Level 2). The personalised opening line is generated from something true and specific about this account right now.
The repository contains real stories, real client situations, and real observations from the founder's professional life — organised by theme. This is the layer no competitor can replicate. Clay's personalised line is passed to Claude along with repository fragments. Claude selects the most relevant one and weaves it in naturally.
A human reviews 10–15% of emails to catch cases where the narrative fragment doesn't quite fit. AI selects well — but not always. Do not skip this step. One tone failure reaching a high-value prospect costs more than the time saved.
The practical framework — step by step
Build your ICP list with Clay
Define your ICP precisely. Use Clay to build the prospect database with firmographic and technographic enrichment. Set up Clay to monitor for Level 2 signals — recent hires, funding rounds, job postings, news mentions — automatically.
Build your narrative repository
This is the step most teams skip because it takes time and requires genuine reflection. Document real client situations, real problems you have solved, real moments of insight. Organise them by theme — outbound challenges, ICP mistakes, positioning failures, client retention lessons. The more specific and honest the stories, the more useful they are.
Build the prompt that connects them
Write a prompt for Claude that takes Clay's Level 2 personalisation for a specific prospect and selects the most relevant narrative fragment from the repository. Specify: select the fragment that most closely matches the prospect's apparent situation, embed it naturally in the first or second paragraph, maintain the tone of the surrounding copy.
Human review — do not skip this
Review 10–15% of emails before they send. AI selects well but occasionally selects poorly. A human review catches tone failures before they reach prospects. The review is also how you learn which narrative fragments are landing and which aren't.
Measure and iterate
Track which narrative fragments generate the highest reply rates. The data will tell you which lived experiences resonate most with your ICP. Update the repository based on what is working. The system learns — not automatically, but through human judgment applied consistently.
The goal is not to eliminate the human element to scale.
It is to protect it while scaling everything else.
The teams that win at cold email in 2026 are not the ones with the most sophisticated automation. They are the ones who have built the deepest libraries of genuine experience and the most effective systems for embedding that experience into outbound at scale. The experience is the moat. The system is just the delivery mechanism.
Key takeaways
- Most personalisation is Level 1 demographic data — necessary but completely insufficient
- Level 2 contextual research improves performance meaningfully — and can be partially automated with Clay
- Level 3 experiential personalisation produces the biggest reply rate jump — but cannot be automated
- The personal narrative repository enables Level 3 at scale by supplying genuine human experience as AI input
- Reply rate difference: 3–5% (two layers) vs. 8–12% (three layers) — not marginal, structural
- Protect the human element while scaling everything else around it
