A cold email is a commercial message sent to someone you have had no prior relationship with, usually to open a business conversation. The cold email that earns a reply differs from the rest because it says something true and specific about the recipient’s company, something that person recognises at once as being about them.
TL;DR. Cold email reply rates depend far less on writing technique than on the observation in the first line. We build that observation for each company from public data the recipient can check, pass it through a human approval gate, then read the email that actually left, because the template never shows what arrived. Published benchmarks range from 0.45% to 3.43%, and most of that gap comes from what gets measured.
What does cold email mean?
Cold email means the first written message you send to someone who does not know you and has asked you for nothing. It differs from warm email, sent to someone who met you, subscribed or filled in a form, and it differs from spam through its target, because it is written to one specific person, about their company.
The term borrows from cold calling, the phone call placed to a stranger. In B2B practice it now covers a whole discipline, from choosing whom to write to, through finding the right address, to the rules that decide whether the message lands in the inbox at all.
The distinction that matters in practice sits between a message you could send unchanged to five hundred companies and a message that makes sense for exactly one. The first stays noise, however polite it sounds. The second has a real chance of being read to the end, because the reader finds something in it about their own company.
Why do we run cold email at all, when we keep saying that positioning comes before channels? Because the founder of an industrial company has no time to discover us on their own, and we believe we owe them, before asking for anything, one true observation about their company that nobody outside it has made.
That is our WHY for this channel, and the rest of the method follows from it. Everything below, from address research to the approval gate, exists so that the one observation is accurate, arrives intact and leads somewhere the founder can verify.
Why start from the WHY rather than from a template?
Most cold email guides start from form, meaning subject line, length, send time and call to action. All of those matter, but they matter only after you have solved the thing that decides whether the reader continues, which is the reason you are writing to them in particular.
The test we run before any send is the same test we apply to brand messaging, the competitor test. If the first sentence could be sent unchanged by any other supplier in the category, it says nothing about the recipient, however many merge fields it contains. A field auto-filled with the company name remains a filled field.
Will Allred, co-founder of Lavender, a platform that helps sales teams write emails, described exactly this failure in a 2023 interview on personalising outreach at scale. Talking about personalisation that goes nowhere, he put it bluntly, “So what? That’s not personalization. I mean, technically it is personalisation, but it’s not good.”
In the same interview, Allred describes what the research is for. “I’m taking this information that I found in my research and I’m tying it back to a challenge that this person likely has that I can help them solve.” His sentence describes almost exactly what we try to do in the first line of every email.
How do we write a cold email, step by step?
Our method has five steps, and each step exists because without it something broke in a way we would not have seen in time. We describe it here without campaign figures, which belong to our relationship with the people we wrote to, but the steps are the same for every company.
The first step is choosing the company. We write only to companies that fit the ideal client profile described in our guide to buyer personas and ICPs, meaning manufacturers and B2B companies that have already built something real.
The reason sits in the observation itself, because a company without a track record has no change in its figures worth naming yet.
The second step is the address of the person who decides. We establish the founder’s or managing director’s address from verified data. We start from the pattern the company’s domain already uses, confirmed on real addresses the company publishes, and only then construct the person’s address.
An address guessed by permuting a name against a domain produces technical bounces, and bounces accumulate in the sending domain’s reputation.
The third step is the observation. We write it for each company separately, from public data the recipient can check alone, such as filed accounts, the company website, and how the company shows up in search results and in AI model answers.
A good observation reads like a change the founder already knows about and that nobody outside the company has mentioned.
To see what such an observation looks like when it is carried all the way through, look at the positioning analysis we published for Cablero Steel Group, a Romanian wire rope manufacturer (the document is in Romanian).
The company states that it produces one linear metre of cable every 1.6 seconds and supplies Groupe Renault through Dacia, the UAMT plant in Oradea and the shelves of the Dedeman retail chain.
None of that evidence sits on the website’s first page, where the top line talks about free delivery.
The pattern has a name in our methodology, the proof already exists, it simply is not communicated. The observation in a good cold email does the same thing on a small scale, naming something true that the company has and does not use.
The fourth step is the approval gate. Every observation lands in a file that I read and correct, sentence by sentence, before anything goes out.
An AI model helps us gather the public data and propose a first draft, but the decision to send a sentence in our name stays with a person. Anything that does not come from the data we gave the model gets cut, because an invented detail in a first line destroys the one thing the email has going for it.
The fifth step is checking the email that left. Sending platforms do not throw an error when a template variable is missing from the data. They quietly replace it with an empty string, so the recipient gets a greeting with no name or a sentence broken in the middle.
That is why we first verify, in code, that every variable has a value for every recipient.
After that check we read the rendered messages for several real recipients, and one copy reaches our own inbox, through the real infrastructure, before the rest leave.
The absence of an error proves nothing, and what the platform editor shows often differs from what the recipient gets. The proof remains the message that left, read back from the system that sent it.
A small detail belongs to the same step. The greeting carries a single first name, even when the person has two on the company register, because that is how the people who know them write to them.
A greeting with both names betrays at once that the address came from a database, and the reader carries that suspicion through the rest of the message.
What kinds of observations earn a reply?
The observations that work share one property, which is that the recipient can verify them in five minutes without us. We group them into three families by data source, and each family corresponds to a problem pattern we meet in the positioning analyses we run.
The first family comes from filed accounts, wherever they are public, as they are in Romania through the tax authority. A margin that fell from one year to the next, revenue that grew while profit stood still, or headcount that rose faster than sales are public facts.
The founder who lived through that year recognises immediately what the sentence refers to, even though nobody outside the company has raised it with them. That recognition is the whole mechanism, because it tells the reader in a single line that the sender has looked at their company rather than at a list.
The second family comes from the company’s website. This is where the pattern from the example above appears most often, with real proof buried on a secondary page while the home page repeats what the whole category says, meaning quality, fair prices and fast delivery.
The observation names the proof and where it sits, and the reader sees the distance between what they have and what they show.
The third family comes from search engines and AI models. We ask the same category question a buyer would ask and check whether the company’s name appears in the answer. Sometimes it is missing because nothing on the site answers the question, and sometimes because the file that tells crawlers what they may read blocks the AI crawlers by name.
All three families share one structure inside the email, a fact, its source and a short question about the cause. We never write the conclusion for the reader, because a founder accepts a question about their company more easily than a diagnosis from a stranger, and a conclusion they reach on their own persuades them far more.
The contrast shows best on a weak sentence. “I noticed you are one of the leaders in your field and I think we could help you grow” can be sent to any company on the continent.
A sentence that names a year, a figure from the filed accounts and the change from the previous year can be sent to exactly one.
What is a good reply rate for B2B cold email?
It depends on what gets divided by what. Belkins measured an average reply rate of 0.45% across 7,530,489 emails sent in 2025, while Instantly reports an average of 3.43% in its benchmark report for 2026. The near eightfold gap comes from the population measured and from the definition, long before it comes from the market.
Belkins states its method openly. It counts unique replies over emails sent, excludes auto-replies and bounce notifications, and draws the data from campaigns it ran for its own clients, with open tracking switched off. Instantly aggregates the behaviour of the platform’s users, meaning people who send in very different ways from one another.
| Source | Data year | What it measures | What it shows |
|---|---|---|---|
| Belkins | 2025 | unique replies over emails sent, across 7,530,489 emails | 0.45% average, 0.57% for founders and owners, 0.42% for C-level executives |
| Instantly | 2026 report | average across platform users | 3.43% average, 58% of replies come from the first email |
| Gong Labs | 2026 | email and subject line length | best emails run 50-100 words, subject lines of 1-4 words perform best |
The practical conclusion for a founder receiving a campaign report is to ask first what was divided by what. A reply percentage without a declared denominator says very little, and comparing it with a market average measured differently says even less.
More useful for a founder is how the figures split inside the same study. In the Belkins 2025 data, founders and owners replied at 0.57%, above C-level executives at 0.42%, and companies with 11 to 50 employees replied at 0.49%, against 0.22% for those with more than 10,000 employees.
Those figures match our own client profile. The person who owns the company reads differently from the manager who runs it, because the owner feels directly in the margin what the observation describes. That is why we write almost exclusively to founders and managing directors, even though their addresses are harder to establish.
The reply rate still has a limit that the percentage hides. A reply is not a contract. Between “agreed” and a signature lies a long road, travelled by several people inside the buyer’s company, and that road is the subject of the next section.
Is cold email still effective in 2026?
Yes, provided you ask it for what it can deliver. Forrester found in 2024 that, on average, 13 people within an organisation are involved in a B2B buying decision, that 89% of purchases involve two or more departments and that 86% of purchases stall during the process.
A cold email cannot persuade thirteen people, but it can open a conversation with one of them.
This is where one of the principles we apply with every client comes in, the sequencing that educates the market gradually. As our doctrine states it, every stage, whether of the funnel, of the client relationship or of market maturity, has its own objective, and skipping a stage means burning stages and losing.
The operational rule that follows is short. The first contact is a meeting, not a sale. For cold email, this changes the objective of the first message. Its job is to get the person to reply and to accept the concrete thing you offer, meaning the analysis, the full observation or a short conversation about their own figures.
An email that tries to sell a twelve-month engagement in its first line skips three stages at once and loses exactly the person it wanted to win. That person does not yet know who is writing, has seen none of the work, and has no reason to make a decision on the strength of a message from a stranger.
There is one more reason the channel keeps its value. Most of the messages a founder receives are written for anyone, so a message written for them stands out precisely by contrast. If automatically generated generic messages multiply, the value of an observation that only someone who has actually looked at the company could write rises with them.
Which layers does a working cold email stand on?
A cold email looks like a channel problem, meaning L1, the performance marketing layer, yet in our methodology it works only when it stands on all four layers, from L1 to L4, with L2 for revenue and the sales process, L3 for AI orchestration and system memory, and L4 for positioning and category.
At L4 it gets decided whether the observation has somewhere to lead. An email that names a real problem correctly and sends the reader to a website that says “quality, competitive prices, fast delivery” produces a conversation that fades on its own.
The reader searches the company name after reading the message, and what they find there either confirms or cancels the promise in the email.
At L3 sits everything that makes the method repeatable. The public data gathered for each company, the approved observations, the rules learned from mistakes and the history of every conversation live in a system that the AI model reads before proposing the next sentence.
Without that memory, every batch of messages would start from zero and repeat the mistakes of the previous batch.
At L2 you see whether a reply turns into revenue. This is where the double funnel comes in, which we cover at length in our article on the B2B sales system. Our doctrine states that marketing materials must be matched to the concrete sales stage, and that this is the key to performance.
A cold email works in the first funnel, the differentiation funnel, where a person moves from suspect to prospect and then to analysis. Its objective is a single step, moving the person into the analysis stage, and any material that tries to jump straight to negotiation works against its own funnel.
At L1 remain the technical matters, meaning sending domains, authentication, sending pace and address verification. They are necessary and they outsource well, but they cannot compensate for anything missing on the other three layers.
The principle that binds the layers is infrastructure before marketing. It states that a client rarely has a marketing problem, because what they actually lack is infrastructure, meaning systems, channels, positioning and a brand foundation, and marketing on a poor base burns money.
For cold email, the principle becomes a question we ask before any campaign. What does the reader find when they search the company name after reading the message? If the answer is a website that does not support the observation in the email, the campaign waits, and the work starts with the architecture of the growth system.
Why does deliverability fail before the reply rate does?
Because a wrong address produces a technical bounce, and bounces accumulate in your domain reputation until even the good messages stop arriving. Deliverability degrades silently, and the signal shows up first in the bounce figures, well before it shows up in replies.
One of the lessons that cost us came from servers configured to accept any address. Such a server says yes to any mailbox you ask about, so the address verification service has nothing to confirm and marks the address as acceptable. The address passes verification, and the message bounces only after it has been sent.
The rule we wrote into our procedure after that episode says that such a response means “cannot be known”. Since then, first contact goes only to individually confirmed addresses, even when the list comes out shorter, because a short, clean list protects the domain for every campaign that follows.
On top of this sit the rules the large platforms impose. Google has required, since 1 February 2024, that all senders to personal Gmail accounts set up SPF or DKIM, keep valid forward and reverse DNS records, use TLS and keep reported spam rates below 0.3%, and above 5,000 messages a day it also requires one-click unsubscribe.
Those thresholds are written for personal Gmail accounts. A plant director often sits on the business infrastructure of the same providers, where filters follow other rules. Your domain reputation stays the same everywhere, though, so there is no sense in leaning on that distinction.
Deliverability also covers a decision we make on every campaign, which is that we do not track opens. The tracking pixel adds one more element to the message without producing a figure you can trust, since Apple hides through Mail Privacy Protection whether a message was opened. Belkins also disabled open tracking across every campaign in the study cited above.
What goes into the first email to get a reply?
One verifiable observation about the recipient’s company, followed by one clear proposal, in fewer than a hundred words. Gong Labs published data in 2026 showing that reply rates drop sharply once emails exceed 100 words, with the best-performing emails landing between 50 and 100 words.
The same dataset shows that subject lines of one to four words perform best. Gong also reports, in the same 2026 analysis, that C-level executives are 30.2% less likely to reply to cold emails than non-executives. For us, that figure raises the bar for the message, because a decision maker reads little and replies only to what concerns them directly.
Length follows from a principle we call reducing cognitive cost. It states that an unclear brand imposes a cognitive tax on every impression, and that reducing friction, so that people think less, understand faster and act, lowers the cost of every contact. The operational rule judges any material by the effort it asks of the reader before they understand it.
In a cold email, cognitive cost means every sentence the reader has to decode to find out why you wrote. The observation therefore sits in the first line, the proposal in the second, and our introduction appears only at the end, in a single sentence, because the reader has no reason yet to read about us.
Two mistakes we made and corrected deserve a mention. The first concerns order. Value gets delivered in the first message, because someone who receives something useful replies, while someone asked for an hour of their time weighs whether it is worth it.
The second concerns negation. A sentence like “I am not asking you for anything right now” plants in the reader’s mind exactly the idea you wanted to avoid, so we no longer announce what we are not doing. We say what we offer, how long it takes and what the reader gets, and we leave everything else out.
The subject line follows the same logic. It makes no promise, builds no intrigue and does not imitate an internal message. It names the topic of the observation briefly, so that the reader knows, before opening, that the message concerns their company.
What happens after someone replies “agreed”?
This is where most campaigns lose, and the loss hurts more because the hard part was already done. Someone who replied “agreed” said yes to the concrete thing you promised in the email, and if you send them an automated message with a booking link, you have cancelled the very reason they replied.
The rule we apply is that a promise made about a specific company gets honoured on that company, with the work done. If the email promised a positioning analysis, the first thing going back is the analysis, built on the company’s public data. It takes time and cannot be automated, which is precisely why it works.
In double funnel terms, the reply “agreed” moves the person from the prospect stage into the analysis stage. The right material for that stage is the analysis itself, and the calendar comes only after the person has read it and has questions they want to discuss with someone.
This is also where the complication described by the Forrester figures above appears. The person who replied is not necessarily the person who signs, and the analysis sent to them has to survive being carried into a meeting with the finance director or a co-owner, without us in the room.
That is why we write the analysis so that each decision maker finds what concerns them. The finance director looks for the order of magnitude of what is at stake, calculated from public figures, while the founder looks for the reason the market compares the company on price. The same analysis carries both answers, each in its own section.
Are cold emails illegal in Europe?
Here a point deserves stating plainly, because many articles treat Europe as one jurisdiction on this. The General Data Protection Regulation, adopted in 2016, allows processing on the basis of legitimate interest under Article 6(1)(f), but commercial email also falls under a separate, specific rule.
The ePrivacy Directive of 2002, in Article 13(5), applies its mandatory consent rule only to subscribers who are natural persons, and leaves member states to ensure that the legitimate interests of other subscribers are sufficiently protected. The European minimum therefore concerns individuals, and each member state decides how far it goes with companies.
That single provision is why the answer changes at the border, and why a playbook lifted from one European market does not automatically transfer to another. Before any campaign in a new country, the national transposition of the directive is the text to read, not a guide written for a different market.
Romania, where our own outreach runs, chose the strict route. Law 506/2004, in Article 12(1), prohibits commercial communications by electronic mail without the express prior consent of the subscriber or user, and paragraph 4 of the same article extends the rule, correspondingly, to subscribers who are legal persons.
The same directive, in Article 13(4), prohibits direct marketing email that disguises or conceals the sender’s identity, or that lacks a valid address to which the recipient can send a request to stop. That rule applies everywhere in the Union, whatever each country decided about companies.
What we do in practice is treat every message as though it had to be defended in front of the person who receives it.
The sender’s identity is visible, the reply address is real and read by a person, unsubscribe works from the first message and is honoured immediately, and the data used is public, taken from filed accounts and the company’s own website.
This article describes what we read in the legislation and does not stand in for legal advice. For your company’s specific situation, ask a lawyer who specialises in data protection, and follow the guidance published by the data protection authority in each country you write to.
Frequently asked questions
What is the 3 email rule?
There is no universal number, but the data shows where replies come from. Instantly reports in its benchmark for 2026 that 58% of replies come from the first email in a sequence, with the remaining 42% coming from follow-ups. Follow-ups are worth sending, provided each one adds new information about the recipient’s company.
Does open rate still mean anything?
Very little. Apple’s Mail Privacy Protection hides whether a message was opened, so the open figure has become an estimate for a share of recipients. We do not track opens at all and watch replies, bounces and unsubscribes instead, meaning the figures that people produce.
What should you do when the bounce rate climbs?
Stop the campaign and fix the source of the addresses before you change a comma in the copy. A rising bounce rate almost always means that some addresses were guessed or are out of date, and domain reputation damages far faster than it recovers.
Is cold email worth outsourcing?
That depends on what you outsource. The technical part, meaning sending domains, authentication and address verification, outsources well. The observation about the recipient’s company outsources badly, because it requires understanding their business, and that is where it gets decided whether you receive a reply.
Cold email or paid ads?
The two answer different questions. An ad buys the attention of someone already searching, while a cold email reaches someone who is not searching but has the problem. When you lack clarity on who the ideal client is, both cost a lot and produce little, which is why the right order starts with cognitive ownership of your category.
What you can build without an advertising budget is the subject of a separate piece, our article on growth without ads.
If you want to see what verifiable observation could be written about your company, meaning exactly the kind of sentence people reply to, ask for a diagnostic. We run it on your public data and show you how you look from the outside, before any campaign.
