Learn / Email copy
The best performing cold email framework, pulled from 1.5 million sends
Last month we sent more than 1.5 million cold emails. Then I put all of them through Claude Code and asked one question: which script structure actually wins, and why. Not which individual emails performed. The pattern sitting underneath them, across every client workspace, every vertical, every offer.
Reading that volume by hand is not something a person does. The analysis came back with one framework in front by a distance, a 1.57% reply rate across 63,000 sends, and a second finding about follow-ups that nobody selling sequence templates wants printed. Below: how it ran, the winning structure taken apart line by line, and that same structure rebuilt live for three different industries.
How the analysis actually ran
The tool matters, so start with why it is not ChatGPT. A chat window cannot reference an external data set that size consistently. It summarizes what you paste and quietly stops being accurate around the point you stop checking. Claude Code sits in your terminal, holds your API keys, and runs a structured pass across the whole set in one place. I am not a developer and did not write a line of code for this.
The setup is three steps:
- Open Claude Code in a terminal and point it at a folder.
- Put your sequencer key in the .env file in that folder. Ours reads EmailBison with a super admin key, so it can see every client workspace at once.
- Give it one prompt.
The prompt is the whole job, and it has four parts:
- Extract. Pull the last 1.5 million sends across all workspaces.
- Define the target. Best performing means reply rate and positive reply rate, not vibes. If you know the benchmarks for your industry, hand them over so it has a bar to measure against.
- Pattern analysis, winners against losers. The part most people skip and the only part that matters: linguistic patterns, offer patterns, vertical breakdowns. What makes copy work at the pattern level, not a list of thirty nice emails.
- Three files out. One with the findings, one with the top 30 emails, one with the full analysis.
One caveat on the output: it calculated positive reply rate as a percentage of sends, not of replies the way most sequencers report it. So 0.41% positive on the winner is not the number your dashboard would show you. Check which denominator you are reading before you compare it to anything, including our own published benchmarks.
The winner: straight into the offer
Framework A took it unanimously. 63,000 sends, a 1.57% reply rate, 0.41% positive on sends. The shape:
{first name}, if we could {concrete deliverable} completely free, we'd {do this, this, and this}. You just {one small thing}. Would that be worth hearing more?
That is the entire email. No greeting, no company introduction, no observation about their LinkedIn, no hypothetical about their pain. The offer is doing every bit of the work, which is exactly why it wins and exactly where it breaks.
The framework is portable. The offer inside it is not.
Why it works, line by line
Copying a script without knowing why it worked is how people end up back at square one the week it stops converting: two decent weeks, a dead campaign, then the hunt for the next template. So here is the anatomy, in the order the reader meets it.
Subject line: two to four words, lower case, one or two variables. Something like FYI plus a first name. This is never the thing that moves the number, so do not spend your week on it.
Body: 25 to 35 words, two to three sentences, straight to the offer. No pain, no observation, no preamble. Every pain line is a guess about a person you have never met, and if you miss, you have told them this email was not written for them. Credibility gone, attention gone, and the best offer in the world will not get it back.
The parenthetical scope disclosure. The micro-pattern that did the heavy lifting, and almost nobody copies it. After the offer, in an aside, you list the three or four things you handle and the one thing they do. Here is the good thing, and by the way you do not have to do any of the annoying parts. It hands over the offer and clears the objection in one breath.
The CTA: a soft permission close."Would that be worth hearing more?" rather than "worth a chat?". Low friction, and it moves both ways depending on the temperature of the prospect. A plain yes means you push for time on the calendar. A "yes, but send me more info" tells you there is an objection to clear first.
The optional PS with one proof claim. Aligned specifically to the offer you just made. Sometimes it reads inorganic, sometimes it is the thing that makes the offer feel real. Test with and without.
The analysis missed one thing, and it is the most important one: how conversational this is. Lower case, no salutation, peer to peer, founder to founder. It reads like a message to a person, not a broadcast. They do not know you, trust you, or like you yet, and short is how you compensate. More on the reader's side of that in the three no-pitch scripts.
The runners-up, and what separates them
Two other frameworks placed. Both are worth having in the rotation, and both lean on something different from the winner.
| Framework | The opening move | What carries it | Use it when |
|---|---|---|---|
| A. Initial offer | If we could do X free, we'd handle A, B, C. You just D. | The offer, alone | The benefit is obvious in one sentence |
| B. Hypothetical outcome | What if we could help you hit X? Happy to send over a free deliverable personalized to you. | Desire, then a low-friction way to see it | The outcome is wanted but the mechanism needs explaining |
| C. Relevant observation | Saw a specific observable signal, one line on what we do, then a free deliverable. | The signal earning the right to make the offer | You have a real signal, not a scraped one |
Reach for B when your offer needs a sentence of setup: it opens on the outcome they already want, then makes seeing it cost nothing. C only works when the signal is genuinely observable. The moment it becomes "saw your post," it is the fake personalization everyone else is sending.
The follow-up finding
This landed harder than the copy patterns. Follow-up steps were 32% of the eligible rows, the largest bucket in the data set by volume, and they produced 58% of every losing email. They are not free either: they eat sending capacity and spend sender reputation, the resource Google's sender guidelines actually price. Bottom-decile results paid for with your domain health.
The read is not "stop following up." It is that the second email in almost every sequence needs surgery. Most people write step one with care and then bolt on a bump. The data says the bump is where your campaign is losing.
Three industries, one framework
The framework is doing the work, not the specific words, so adapting it to a new offer changes four things and leaves the rest alone: the deliverable, the workload you absorb, the one action left for them, and the proof claim in the PS. I had Claude Code adapt the winner live to three offers that share nothing.
A free LinkedIn lead magnet funnel, sold to SaaS companies:
{first name}, if we could build your next LinkedIn lead magnet funnel completely free, we'd write the magnet, design the landing page, and queue the nurture sequence. You just post the launch. Would that be worth hearing more? PS - last one we built added {X} demo calls a month for a similar SaaS team.
An email marketing agency selling to e-commerce brands:
{first name}, if we could rebuild your highest revenue email flow completely free, we'd write the copy, design the templates, and stage them inside Klaviyo. All you have to do is hit send. Would that be worth hearing more? PS - last brand we did this for saw a flow revenue jump of {X}% in 60 days.
A freight forwarder offering a free rate check:
{first name}, if we could run a free rate check on your top five shipping lanes, we'd pull the current carrier pricing, compare against what you're paying today, and send a side-by-side breakdown. You just forward the last invoice. Would that be worth hearing more? PS - most importers we run this for find {X} in annual savings sitting in lane mismatches.
Same skeleton three times. The freight one also produced the only subject line worth arguing about: instead of FYI plus a first name, "your shipping rates." It reads like something an importer already receives from a customer or a carrier, so curiosity does the opening for you.
Where the framework breaks
If the offer is weak, vague, or does not match how the prospect sees their world, there is nothing left in the email to save it. That is the trade you made when you deleted the preamble. It also struggles on complicated offers: if the benefit needs more than one line, 25 words will not carry it. Keep simplifying the offer until it fits rather than padding the copy until it sounds fine.
A winning script is one input. It still needs the infrastructure, the list, the sending volume, and someone working the replies. The rest of the terminal workflow is in the six Claude Code use cases.
tl;dr: across 1.5 million sends, the winner was 25 to 35 words that go straight to a free, concrete deliverable, list the three or four things you handle, and close with "would that be worth hearing more?". Change the offer, keep the skeleton.
FAQ
What is the best performing cold email framework?
Across 1.5 million sends in 30 days, the winner was a direct offer email: first name, if we could do this concrete thing for you completely free, we would handle these three or four steps, you just do one small thing, would that be worth hearing more. It ran a 1.57% reply rate across 63,000 sends. No pain line, no observation, no preamble.
How do you analyze 1.5 million cold emails?
Point Claude Code at a folder holding your sequencer API key, then give it one prompt: pull every send, define best performing by reply rate and positive reply rate, then compare winners against losers at the pattern level rather than listing good emails. Ask for three files back: the insights, the top 30 emails, and the full analysis. No coding involved.
Should a cold email mention the prospect's pain point?
The winning framework does not, and that is the point. A pain line is a guess about a stranger. Guess wrong and you have just told them the email was not written for them, which costs the reply even when the offer behind it is strong. Skipping the guess puts all the weight on the offer, so the offer has to be genuinely good.
How long should a cold email be?
The winning emails ran 25 to 35 words across two or three sentences, under a subject line of two to four words in lower case carrying one or two variables. The reason is not a rule about attention spans. A stranger owes you nothing and is giving you no trust, so a short message is the honest way to ask for their time.
Do cold email follow-ups actually work?
Mostly not, in this data. Follow-up steps were 32% of the rows analyzed, the largest bucket by volume, and they accounted for 58% of every losing email. They still burn sending capacity and sender reputation while producing bottom-decile results. The takeaway is not to delete them, it is that the second email in a sequence almost always needs surgery.
Can the same cold email framework work in a different industry?
Yes, because the framework does the work, not the words. Rebuilding it live for a SaaS lead magnet offer, an e-commerce email retention offer, and a freight rate check changed four things: the deliverable, the steps you take off their plate, the one action left for them, and the proof claim in the PS. Everything else held.
If you would rather have the offer, copy, infrastructure, and replies run for you, that is what we do: 3,000+ sales calls booked for clients in the past year on this exact system. Every training, grouped by topic, is free on the trainings hub, and the full video is on YouTube.
PS - the framework was unanimous across 1.5 million emails, which is the least interesting thing about it. The interesting part is that the winning emails were the ones that tried the least.
