Quick answer
A cold email that works is four lines long. An observation that proves you looked, a question that makes the problem feel present, one sentence on what you do and for whom, and an ask small enough that saying yes costs nothing. Everything before those four lines is targeting and offer work, and that is where most campaigns are actually lost.
I recorded a video walking through the framework I use for cold email copy. This is the written version, with the parts that matter most pulled to the front. I'm Hlib Storchak. I build and run outbound for B2B founders and sales teams, and I have booked 2000+ meetings for B2B clients using roughly this structure.
Why the old approach stopped working
The approach that used to work was: export a database, write one generic email, send it to everyone. That is dead, and not because of any algorithm. It is dead because buyers have now received thousands of those emails and can identify one in about two seconds.
What replaced it is worse in an interesting way. Most teams now send generated copy that is grammatically perfect and completely hollow. Open your spam folder and you can pick them out instantly: fluent, polite, structurally correct, and obviously written by nobody about no one. The reader does not consciously analyse it. They just feel nothing and delete it.
Cold email still works. I run it every day. But the bar moved from "did you send it" to "does this read like a person who looked at my company wrote it".
The problem is almost never the copy
When a campaign underperforms, everyone rewrites the email. It is the visible part, so it gets the blame. In my experience the list is wrong far more often than the copy is.
Getting the targeting right does not require a fifty step qualification process. Doing the basics properly already puts you ahead of most senders. Two layers:
The company. Industry and sub-niche, revenue band, employee count, technology they run, funding status. Specific enough that you could argue why this company and not the one next to it.
The person. Not just a job title. Who approves the spend, who actually uses the thing, who is most hurt by the problem today, and what number that person is measured on. If you cannot describe how a particular buyer thinks about the purchase, you are not ready to write to them.
Here is the level of definition I am talking about. Not "marketing leaders at SaaS companies", but: marketing directors at B2B SaaS companies with 50 to 200 employees, spending over ten thousand a month on paid ads, running HubSpot, who recently hired a demand generation manager.
You will not find thousands of those. That is the point. A tight definition produces a small list, and small precise lists consistently beat large vague ones. If the definition feels uncomfortably narrow, it is probably about right. This is the same argument I make in building an ICP that converts.
One company, three different fears
Inside a single company, the people you might write to want completely different things. Same product, same problem, three incompatible emails.
The marketer has to prove return on every channel. They love data, case studies and quick wins, and they switch off at theory and long timelines.
The operations lead is drowning in manual work. They want something that plugs in. The thing that loses them is a four month implementation, and if they suspect that is coming, they stop reading.
The founder has usually been burned by an agency before. They want predictable pipeline and they are frightened of long commitments and costs that appear later. Guarantees, transparent pricing and a direct relationship are what move them.
Write two or three of these personas down before you write a single email. The persona decides which pain you lead with and which offer you make, which means it decides most of the email before you have written a word.
The offer decides the reply, not the copy
You can write a technically excellent email and still get nothing, because the thing you asked for was too expensive. "A thirty minute discovery call to discuss how we can help" costs the reader thirty minutes and they know exactly what happens in it. Nobody wants that.
The alternative is to give something before asking for anything. Not a brochure. Something with real work in it.
| Instead of | Offer this | Why it converts |
|---|---|---|
| A 30 minute discovery call | An audit of their current setup with three specific things to fix | They get something usable even if they never reply again |
| A demo | A free trial with their own data loaded in | Removes the imagination step entirely |
| "Let's explore a partnership" | Two weeks of the actual work, at no cost | Almost nobody says no to seeing the thing run |
| "Can I pick your brain" | A 45 minute strategy session plus a roadmap they keep | Valuable whether or not they hire you |
The mechanism is reciprocity, and it is unusually reliable. Give something real first and people become far more inclined to give you something back. The other half of it is risk: every one of those offers is easy to say yes to because saying yes costs the reader almost nothing.
Lead with the pain, not the benefit
Benefit led copy works. Pain led copy works better, and in the campaigns I have run the gap is not small.
The reason is old and well established: losing something hurts more than gaining the same thing pleases. "Book more meetings" is pleasant and easy to ignore. "Your reps are spending a third of their week on leads that were never going to buy" is uncomfortable, and uncomfortable gets read.
This only works if the pain is real and specific to that persona, which is why the persona work has to come first. A generic pain is just a different flavour of generic.
The four line framework
This is the structure itself. I picked it up from a former colleague, Xavier, and I have used it more than any other. Four lines, in this order.
1. The observation. Something specific about them that proves you looked. A trigger, a signal, a change. Not "I see you are in SaaS". Something that could only be written to this company this month.
2. The question that pokes the bear. Based on that observation, ask something that makes the problem feel present rather than theoretical. This is the line most people skip, and it is the one that turns a statement into a conversation. A question is answerable. A pitch is not.
3. The elevator pitch. One or two sentences. What you do, who you do it for, what changes. Not your positioning statement, not your features. The shortest true version.
4. The ask. Either the offer itself, or a low commitment question. This is where the offer work pays off.
Here is the whole thing assembled:
Hey Sarah, you have just brought on three new sales development reps, congratulations on the growth. Most teams struggle to ramp new reps fast enough and end up burning ninety days of budget on unqualified pipeline. How are you planning to get them productive in their first month? I help B2B teams cut ramp time roughly in half with a structured 30 day playbook. I have put one together for your new hires, want to take a look?
That is four lines. It proves research, names a real cost, asks something answerable, explains the offer in one sentence and ends with an ask that costs nothing to accept.
What breaks it
The framework fails in predictable ways, and they are all failures of the work before the writing.
| Symptom | What actually went wrong |
|---|---|
| The observation could apply to any company | Your list is not segmented tightly enough to have anything specific to say |
| The question is rhetorical rather than answerable | You wrote a statement with a question mark on it |
| The pitch runs to four sentences | You have not decided what you actually do for this persona |
| The ask is a meeting | No offer work was done, so the meeting is the only thing left to ask for |
| Everything is right and replies are still flat | Look at deliverability before the copy, see the deliverability guide |
Where AI actually fits
AI can do genuine personalisation at scale. The problem is that almost nobody knows how, so what they generate is fluent and empty, which is worse than a plain template because it wastes the reader's time more convincingly.
The distinction that matters: AI is good at applying a pattern you already understand to a thousand rows. It is bad at working out what the pattern should be. If you have not written the email yourself, you have nothing to scale.
Write fifty by hand before you automate anything
This is the part of the video I would keep if I had to delete everything else, and it is the least popular advice I give.
Before you touch a sequencer, pick fifty real prospects and write fifty emails to them yourself. No generation, no templates, no variables. Find the trigger for each one, write the question, write the pitch, make the ask.
It is slow and it is the fastest way to learn this. By email fifteen you will know which triggers are findable and which are not. By thirty you will have stopped writing observations that mean nothing. By fifty you will have a pattern worth automating, and automating a pattern that works is easy. Automating before you have one just produces volume.
This is also the honest answer to why most outbound fails. Not tooling, not deliverability, not the algorithm. The team never did the fifty.
What to watch once it is running
Positive reply rate, split by segment, is the number that tells you whether the targeting and the offer are right. Raw reply rate includes people telling you to stop. A blended figure across segments hides the one pocket that is working.
Judge the copy only after the list and the offer are settled, because those two decide most of the result. I go deeper on this in the metrics that actually measure outbound.
Key takeaways
- Four lines: an observation that proves you looked, a question that makes the pain present, one sentence on what you do, and an ask that costs nothing to accept.
- The list is wrong more often than the copy is. Define the company and the person precisely enough that the definition feels too narrow.
- Inside one company, the marketer, the operations lead and the founder are frightened of completely different things. Write personas before emails.
- Replace the discovery call with something that has real work in it and is useful whether or not they ever buy.
- Pain led copy outperforms benefit led copy, provided the pain is specific to that persona.
- Write fifty by hand before automating. You cannot scale a pattern you have not found yet.
FAQ
What reply rate should I expect from this?
On the campaigns I run, a tight list with a real offer lands well above what a generic blast produces, and I have seen high single digit and low double digit positive reply rates on narrow segments. Treat any specific number, including mine, as a description of one situation rather than a forecast for yours. Your segment, offer and market decide it.
Is cold email still worth doing in 2026?
Yes, and the bar is higher than it was. What died is the generic blast. What still works is a specific message to a well chosen person with an offer that is easy to accept.
How long should a cold email be?
Four lines, as above. If it is longer, something in the targeting or the offer is unresolved and the extra words are trying to cover for it.
Can I just have AI write these?
Once you have written fifty yourself and know what a good observation looks like for your market, yes, and it scales well. Before that it produces fluent emails about nothing, which perform worse than a plain template.
What if I cannot find a trigger for a prospect?
That is usually the list telling you something. If there is nothing specific to say about a company, it probably does not belong on this campaign. Segment differently rather than writing a vaguer observation.